From: Carsten Haitzler carsten.haitzler@arm.com
Perf test's shell runner will just run everything in the tests directory (as long as it's not another directory or does not begin with a dot), but sometimes you find files in there that are not shell scripts - perf.data output for example if you do some testing and then the next time you run perf test it tries to run these. Check the files are executable so they are actually intended to be test scripts and not just some "random junk" files there.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com --- tools/perf/tests/builtin-test.c | 4 +++- tools/perf/util/path.c | 14 +++++++++++++- tools/perf/util/path.h | 1 + 3 files changed, 17 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index fac3717d9ba1..3c34cb766724 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,7 +296,9 @@ static const char *shell_test__description(char *description, size_t size,
#define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \ - if (!is_directory(base, ent) && ent->d_name[0] != '.') + if (!is_directory(base, ent) && \ + is_executable_file(base, ent) && \ + ent->d_name[0] != '.')
static const char *shell_tests__dir(char *path, size_t size) { diff --git a/tools/perf/util/path.c b/tools/perf/util/path.c index caed0336429f..ce80b79be103 100644 --- a/tools/perf/util/path.c +++ b/tools/perf/util/path.c @@ -86,9 +86,21 @@ bool is_directory(const char *base_path, const struct dirent *dent) char path[PATH_MAX]; struct stat st;
- sprintf(path, "%s/%s", base_path, dent->d_name); + snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name); if (stat(path, &st)) return false;
return S_ISDIR(st.st_mode); } + +bool is_executable_file(const char *base_path, const struct dirent *dent) +{ + char path[PATH_MAX]; + struct stat st; + + snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name); + if (stat(path, &st)) + return false; + + return !S_ISDIR(st.st_mode) && (st.st_mode & S_IXUSR); +} diff --git a/tools/perf/util/path.h b/tools/perf/util/path.h index 083429b7efa3..d94902c22222 100644 --- a/tools/perf/util/path.h +++ b/tools/perf/util/path.h @@ -12,5 +12,6 @@ int path__join3(char *bf, size_t size, const char *path1, const char *path2, con
bool is_regular_file(const char *file); bool is_directory(const char *base_path, const struct dirent *dent); +bool is_executable_file(const char *base_path, const struct dirent *dent);
#endif /* _PERF_PATH_H */
From: Carsten Haitzler carsten.haitzler@arm.com
You edit your scripts in the tests and end up with your usual shell backup files with ~ or .bak or something else at the end, but then your next perf test run wants to run the backups too. You might also have perf .data files in the directory or something else undesireable as well. You end up chasing which test is the one you edited and the backup and have to keep removing all the backup files, so automatically skip any files that are not plain *.sh scripts to limit the time wasted in chasing ghosts.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com --- tools/perf/tests/builtin-test.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index 3c34cb766724..3a02ba7a7a89 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,9 +296,22 @@ static const char *shell_test__description(char *description, size_t size,
#define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \ - if (!is_directory(base, ent) && \ + if (ent->d_name[0] != '.' && \ + !is_directory(base, ent) && \ is_executable_file(base, ent) && \ - ent->d_name[0] != '.') + is_shell_script(ent->d_name)) + +static bool is_shell_script(const char *file) +{ + const char *ext; + + ext = strrchr(file, '.'); + if (!ext) + return false; + if (!strcmp(ext, ".sh")) + return true; + return false; +}
static const char *shell_tests__dir(char *path, size_t size) {
On Wed, Mar 09, 2022 at 12:28:58PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
You edit your scripts in the tests and end up with your usual shell backup files with ~ or .bak or something else at the end, but then your next perf test run wants to run the backups too. You might also have perf .data files in the directory or something else undesireable as well. You end up chasing which test is the one you edited and the backup and have to keep removing all the backup files, so automatically skip any files that are not plain *.sh scripts to limit the time wasted in chasing ghosts.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
tools/perf/tests/builtin-test.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index 3c34cb766724..3a02ba7a7a89 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,9 +296,22 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && \
if (ent->d_name[0] != '.' && \!is_directory(base, ent) && \ is_executable_file(base, ent) && \
ent->d_name[0] != '.')
is_shell_script(ent->d_name))
Just nitpick: since multiple conditions are added, seems to me it's good to use a single function is_executable_shell_script() to make decision if a file is an executable shell script.
And the condition checking 'ent->d_name[0] != '.'' would be redundant after we have checked the file suffix '.sh'.
Thanks, Leo
+static bool is_shell_script(const char *file) +{
- const char *ext;
- ext = strrchr(file, '.');
- if (!ext)
return false;- if (!strcmp(ext, ".sh"))
return true;- return false;
+} static const char *shell_tests__dir(char *path, size_t size) { -- 2.32.0
On 4/10/22 03:28, Leo Yan wrote:
On Wed, Mar 09, 2022 at 12:28:58PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
You edit your scripts in the tests and end up with your usual shell backup files with ~ or .bak or something else at the end, but then your next perf test run wants to run the backups too. You might also have perf .data files in the directory or something else undesireable as well. You end up chasing which test is the one you edited and the backup and have to keep removing all the backup files, so automatically skip any files that are not plain *.sh scripts to limit the time wasted in chasing ghosts.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
tools/perf/tests/builtin-test.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index 3c34cb766724..3a02ba7a7a89 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,9 +296,22 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && \
if (ent->d_name[0] != '.' && \!is_directory(base, ent) && \ is_executable_file(base, ent) && \
ent->d_name[0] != '.')
is_shell_script(ent->d_name))Just nitpick: since multiple conditions are added, seems to me it's good to use a single function is_executable_shell_script() to make decision if a file is an executable shell script.
I'd certainly make a function if this was being re-used, but as the "coding pattern" was to do all the tests already inside the if() in only one place, I kept with the style there and didn't change the code that didn't need changing. I can rewrite this code and basically make a function that is just an if ...:
bool is_exe_shell_script(const char *base, struct dirent *ent) { return ent->d_name[0] != '.' && !is_directory(base, ent) && is_executable_file(base, ent) && is_shell_script(ent->d_name); }
And macro becomes:
#define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \ if (is_shell(base, ent))
But one catch... it really should be is_non_hidden_exe_shell_script() as it's checking that it's not a hidden file AND is a shell script. Or do I keep the hidden file test outside of the function in the if? If we're nit picking then I need to know exactly what you want here as your suggested name is actually incorrect.
And the condition checking 'ent->d_name[0] != '.'' would be redundant after we have checked the file suffix '.sh'.
This isn't actually redundant. You can have .something.sh :) If the idea is we skip anything with a . at the start first always... then the if (to me) is obvious.
Thanks, Leo
+static bool is_shell_script(const char *file) +{
- const char *ext;
- ext = strrchr(file, '.');
- if (!ext)
return false;- if (!strcmp(ext, ".sh"))
return true;- return false;
+} static const char *shell_tests__dir(char *path, size_t size) { -- 2.32.0
On Thu, Apr 21, 2022 at 05:21:27PM +0100, Carsten Haitzler wrote:
On 4/10/22 03:28, Leo Yan wrote:
On Wed, Mar 09, 2022 at 12:28:58PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
You edit your scripts in the tests and end up with your usual shell backup files with ~ or .bak or something else at the end, but then your next perf test run wants to run the backups too. You might also have perf .data files in the directory or something else undesireable as well. You end up chasing which test is the one you edited and the backup and have to keep removing all the backup files, so automatically skip any files that are not plain *.sh scripts to limit the time wasted in chasing ghosts.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
tools/perf/tests/builtin-test.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index 3c34cb766724..3a02ba7a7a89 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,9 +296,22 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && \
if (ent->d_name[0] != '.' && \!is_directory(base, ent) && \ is_executable_file(base, ent) && \
ent->d_name[0] != '.')
is_shell_script(ent->d_name))Just nitpick: since multiple conditions are added, seems to me it's good to use a single function is_executable_shell_script() to make decision if a file is an executable shell script.
I'd certainly make a function if this was being re-used, but as the "coding pattern" was to do all the tests already inside the if() in only one place, I kept with the style there and didn't change the code that didn't need changing. I can rewrite this code and basically make a function that is just an if ...:
bool is_exe_shell_script(const char *base, struct dirent *ent) { return ent->d_name[0] != '.' && !is_directory(base, ent) && is_executable_file(base, ent) && is_shell_script(ent->d_name); }
And macro becomes:
#define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \ if (is_shell(base, ent))
Sorry for long latency.
If the condition checking gets complex, seems to me it is reasonable to use a static function (or a macro?) to encapsulate the logics.
But one catch... it really should be is_non_hidden_exe_shell_script() as it's checking that it's not a hidden file AND is a shell script. Or do I keep the hidden file test outside of the function in the if? If we're nit picking then I need to know exactly what you want here as your suggested name is actually incorrect.
I personally prefer to use the condition:
if (is_exe_shell_script() && ent->d_name[0] != '.') do_something...
The reason is the function is_exe_shell_script() is more common and we use it easily in wider scope.
And the condition checking 'ent->d_name[0] != '.'' would be redundant after we have checked the file suffix '.sh'.
This isn't actually redundant. You can have .something.sh :) If the idea is we skip anything with a . at the start first always... then the if (to me) is obvious.
Yeah, I agree the checking the start char '.' is the right thing to do.
Thanks, Leo
On 5/26/22 11:14, Leo Yan wrote:
On Thu, Apr 21, 2022 at 05:21:27PM +0100, Carsten Haitzler wrote:
On 4/10/22 03:28, Leo Yan wrote:
On Wed, Mar 09, 2022 at 12:28:58PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
You edit your scripts in the tests and end up with your usual shell backup files with ~ or .bak or something else at the end, but then your next perf test run wants to run the backups too. You might also have perf .data files in the directory or something else undesireable as well. You end up chasing which test is the one you edited and the backup and have to keep removing all the backup files, so automatically skip any files that are not plain *.sh scripts to limit the time wasted in chasing ghosts.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
tools/perf/tests/builtin-test.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index 3c34cb766724..3a02ba7a7a89 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,9 +296,22 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && \
if (ent->d_name[0] != '.' && \!is_directory(base, ent) && \ is_executable_file(base, ent) && \
ent->d_name[0] != '.')
is_shell_script(ent->d_name))Just nitpick: since multiple conditions are added, seems to me it's good to use a single function is_executable_shell_script() to make decision if a file is an executable shell script.
I'd certainly make a function if this was being re-used, but as the "coding pattern" was to do all the tests already inside the if() in only one place, I kept with the style there and didn't change the code that didn't need changing. I can rewrite this code and basically make a function that is just an if ...:
bool is_exe_shell_script(const char *base, struct dirent *ent) { return ent->d_name[0] != '.' && !is_directory(base, ent) && is_executable_file(base, ent) && is_shell_script(ent->d_name); }
And macro becomes:
#define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \ if (is_shell(base, ent))
Sorry for long latency.
No problem.
If the condition checking gets complex, seems to me it is reasonable to use a static function (or a macro?) to encapsulate the logics.
Well normally my rule i s - if it gets re-used then do it, otherwise it just involves more indirection to follow. :) But regardless of that, given some other things you ask for that kind of makes this discussion moot as it requires much bigger wholesale changes to the test infra which will make these patches a lot more work. I'll get to that later in mails.
But one catch... it really should be is_non_hidden_exe_shell_script() as it's checking that it's not a hidden file AND is a shell script. Or do I keep the hidden file test outside of the function in the if? If we're nit picking then I need to know exactly what you want here as your suggested name is actually incorrect.
I personally prefer to use the condition:
if (is_exe_shell_script() && ent->d_name[0] != '.') do_something...
The reason is the function is_exe_shell_script() is more common and we use it easily in wider scope.
As above - will probably have to redo a lot of the test infra involving the shell tests to handle some of your other requests, but if we don't go that way, I have got where you want to go and I can do this.
And the condition checking 'ent->d_name[0] != '.'' would be redundant after we have checked the file suffix '.sh'.
This isn't actually redundant. You can have .something.sh :) If the idea is we skip anything with a . at the start first always... then the if (to me) is obvious.
Yeah, I agree the checking the start char '.' is the right thing to do.
:)
On Mon, Jun 13, 2022 at 02:08:30PM +0100, Carsten Haitzler wrote:
[...]
If the condition checking gets complex, seems to me it is reasonable to use a static function (or a macro?) to encapsulate the logics.
Well normally my rule i s - if it gets re-used then do it, otherwise it just involves more indirection to follow. :) But regardless of that, given some other things you ask for that kind of makes this discussion moot as it requires much bigger wholesale changes to the test infra which will make these patches a lot more work. I'll get to that later in mails.
Your mentioned rule makes sense to me.
But one catch... it really should be is_non_hidden_exe_shell_script() as it's checking that it's not a hidden file AND is a shell script. Or do I keep the hidden file test outside of the function in the if? If we're nit picking then I need to know exactly what you want here as your suggested name is actually incorrect.
I personally prefer to use the condition:
if (is_exe_shell_script() && ent->d_name[0] != '.') do_something...
The reason is the function is_exe_shell_script() is more common and we use it easily in wider scope.
As above - will probably have to redo a lot of the test infra involving the shell tests to handle some of your other requests, but if we don't go that way, I have got where you want to go and I can do this.
To be honest, I am not sure if this patch is related with refactoring test infrastructure or not. You could reconsider when you spin for next patch set (as you said, might refactor test infra).
In case you still want to keep this patch as it is, it would be fine for me and you could add my reviewed tag:
Reviewed-by: Leo Yan leo.yan@linaro.org
Thanks, Leo
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com --- MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++ 24 files changed, 823 insertions(+), 4 deletions(-) create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.*
diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt @@ -0,0 +1,140 @@ +Arm Coresight Support +===================== + +Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream. Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like: + + perf record -e cs_etm//u testbinary + +This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like: + + perf report --stdio --dump -i perf.data + +You should find some sections of this file have AUX data blocks like: + + 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1 + + . ... CoreSight ETM Trace data: size 73168 bytes + Idx:0; ID:10; I_ASYNC : Alignment Synchronisation. + Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 } + Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000; + Idx:26; ID:10; I_TRACE_ON : Trace On. + Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS; + Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEE + Idx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEE + Idx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEE + Idx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN + ... + +If you see these above, then your system is tracing coresight data +correctly. + +To compile perf with coresight support in the perf directory do + + make CORESIGHT=1 + +This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight. + +You will also want coresight support enabled in your kernel config. +Ensure it is enabled with: + + CONFIG_CORESIGHT=y + +There are various other coresight options you probably also want +enabled like: + + CONFIG_CORESIGHT_LINKS_AND_SINKS=y + CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y + CONFIG_CORESIGHT_CATU=y + CONFIG_CORESIGHT_SINK_TPIU=y + CONFIG_CORESIGHT_SINK_ETBV10=y + CONFIG_CORESIGHT_SOURCE_ETM4X=y + CONFIG_CORESIGHT_STM=y + CONFIG_CORESIGHT_CPU_DEBUG=y + CONFIG_CORESIGHT_CTI=y + CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y + +Please refer to the kernel configuration help for more information. + +Perf test - Verify kernel and userspace perf coresight work +=========================================================== + +When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like: + + Check Arm CoreSight trace data recording and synthesized samples + +Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like: + + Coresight / Memcpy 1M 25 Threads + Coresight / Unroll Loop Thread 2 + ... + +These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead. + +These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like: + + stats-asm_pure_loop-out.csv + stats-bubble_sort-random.csv + ... + +These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better. + +Be aware that amny of these tests take quite a while to run, specifically +in processing the perf data file and dumping contents to then examine what +is inside. + +You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like: + + export PERF_TEST_CORESIGHT_STATDIR=/var/tmp + perf test + +They will also store resulting perf output data in the current +directory for later inspection like: + + perf-memcpy-1m.data + perf-thread_loop-2th.data + ... + +You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as: + + PERF_TEST_CORESIGHT_DATADIR=/var/tmp + perf test + +You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@
-all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight + +tests-coresight-targets: FORCE + $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) + +tests-coresight-targets-clean: + $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean + +all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets
# Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib' + $(Q)$(MAKE) -C tests/shell/coresight install-tests
install-bin: install-tools install-tests install-traceevent-plugins
@@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS)
-clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS)
endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak + +SUBDIRS = \ + asm_pure_loop \ + thread_loop \ + memcpy_thread \ + unroll_loop_thread + +all: $(SUBDIRS) +$(SUBDIRS): + $(Q)$(MAKE) -C $@ + +INSTALLDIRS = $(SUBDIRS:%=install-%) + +install-tests: $(INSTALLDIRS) +$(INSTALLDIRS): + $(Q)$(MAKE) -C $(@:install-%=%) install-tests + +CLEANDIRS = $(SUBDIRS:%=clean-%) + +clean: $(CLEANDIRS) +$(CLEANDIRS): + $(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null + +.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS) + diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +ifndef DESTDIR +prefix ?= $(HOME) +endif + +DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir) + +ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif + +perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install + +include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +include ../Makefile.miniconfig + +BIN=asm_pure_loop +LIB= + +all: $(BIN) + +$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB) +endif +endif + +install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(call QUIET_INSTALL, tests) \ + $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \ + $(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)' +endif +endif + +clean: + $(Q)$(RM) -f $(BIN) + +.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */ + +.globl _start +_start: + mov x0, 0x0000ffff + mov x1, xzr +loop: + nop + nop + cbnz x1, noskip + nop + nop + adrp x2, skip + add x2, x2, :lo12:skip + br x2 + nop + nop +noskip: + nop + nop +skip: + sub x0, x0, 1 + cbnz x0, loop + + mov x0, #0 + mov x8, #93 // __NR_exit syscall + svc #0 diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig + +BIN=memcpy_thread +LIB=-pthread + +all: $(BIN) + +$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB) +endif +endif + +install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(call QUIET_INSTALL, tests) \ + $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \ + $(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)' +endif +endif + +clean: + $(Q)$(RM) -f $(BIN) + +.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> + +struct args { + unsigned long loops; + unsigned long size; + pthread_t th; + void *ret; +}; + +static void *thrfn(void *arg) +{ + struct args *a = arg; + unsigned long i, len = a->loops; + unsigned char *src, *dst; + + src = malloc(a->size * 1024); + dst = malloc(a->size * 1024); + if ((!src) || (!dst)) { + printf("ERR: Can't allocate memory\n"); + exit(1); + } + for (i = 0; i < len; i++) + memcpy(dst, src, a->size * 1024); +} + +static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{ + pthread_t t; + pthread_attr_t attr; + + pthread_attr_init(&attr); + pthread_create(&t, &attr, fn, arg); + return t; +} + +int main(int argc, char **argv) +{ + unsigned long i, len, size, thr; + pthread_t threads[256]; + struct args args[256]; + long long v; + + if (argc < 4) { + printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]); + exit(1); + } + + v = atoll(argv[1]); + if ((v < 1) || (v > (1024 * 1024))) { + printf("ERR: max memory 1GB (1048576 KB)\n"); + exit(1); + } + size = v; + thr = atol(argv[2]); + if ((thr < 1) || (thr > 256)) { + printf("ERR: threads 1-256\n"); + exit(1); + } + v = atoll(argv[3]); + if ((v < 1) || (v > 40000000000ll)) { + printf("ERR: loops 1-40000000000 (hundreds)\n"); + exit(1); + } + len = v * 100; + for (i = 0; i < thr; i++) { + args[i].loops = len; + args[i].size = size; + args[i].th = new_thr(thrfn, &(args[i])); + } + for (i = 0; i < thr; i++) + pthread_join(args[i].th, &(args[i].ret)); + return 0; +} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig + +BIN=thread_loop +LIB=-pthread + +all: $(BIN) + +$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB) +endif +endif + +install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(call QUIET_INSTALL, tests) \ + $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \ + $(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)' +endif +endif + +clean: + $(Q)$(RM) -f $(BIN) + +.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 + +// define this for gettid() +#define _GNU_SOURCE + +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid) + +struct args { + unsigned int loops; + pthread_t th; + void *ret; +}; + +static void *thrfn(void *arg) +{ + struct args *a = arg; + int i = 0, len = a->loops; + + if (getenv("SHOW_TID")) { + unsigned long long tid = gettid(); + + printf("%llu\n", tid); + } + asm volatile( + "loop:\n" + "add %[i], %[i], #1\n" + "cmp %[i], %[len]\n" + "blt loop\n" + : /* out */ + : /* in */ [i] "r" (i), [len] "r" (len) + : /* clobber */ + ); + return (void *)(long)i; +} + +static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{ + pthread_t t; + pthread_attr_t attr; + + pthread_attr_init(&attr); + pthread_create(&t, &attr, fn, arg); + return t; +} + +int main(int argc, char **argv) +{ + unsigned int i, len, thr; + pthread_t threads[256]; + struct args args[256]; + + if (argc < 3) { + printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]); + exit(1); + } + + thr = atoi(argv[1]); + if ((thr < 1) || (thr > 256)) { + printf("ERR: threads 1-256\n"); + exit(1); + } + len = atoi(argv[2]); + if ((len < 1) || (len > 4000)) { + printf("ERR: max loops 4000 (millions)\n"); + exit(1); + } + len *= 1000000; + for (i = 0; i < thr; i++) { + args[i].loops = len; + args[i].th = new_thr(thrfn, &(args[i])); + } + for (i = 0; i < thr; i++) + pthread_join(args[i].th, &(args[i].ret)); + return 0; +} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig + +BIN=unroll_loop_thread +LIB=-pthread + +all: $(BIN) + +$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB) +endif +endif + +install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64) + $(call QUIET_INSTALL, tests) \ + $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \ + $(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)' +endif +endif + +clean: + $(Q)$(RM) -f $(BIN) + +.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> + +struct args { + pthread_t th; + unsigned int in, out; + void *ret; +}; + +static void *thrfn(void *arg) +{ + struct args *a = arg; + unsigned int i, in = a->in; + + for (i = 0; i < 10000; i++) { + asm volatile ( +// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000 + SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 + : /* out */ + : /* in */ [in] "r" (in) + : /* clobber */ + ); + } +} + +static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{ + pthread_t t; + pthread_attr_t attr; + + pthread_attr_init(&attr); + pthread_create(&t, &attr, fn, arg); + return t; +} + +int main(int argc, char **argv) +{ + unsigned int i, thr; + pthread_t threads[256]; + struct args args[256]; + + if (argc < 2) { + printf("ERR: %s [numthreads]\n", argv[0]); + exit(1); + } + + thr = atoi(argv[1]); + if ((thr > 256) || (thr < 1)) { + printf("ERR: threads 1-256\n"); + exit(1); + } + for (i = 0; i < thr; i++) { + args[i].in = rand(); + args[i].th = new_thr(thrfn, &(args[i])); + } + for (i = 0; i < thr; i++) + pthread_join(args[i].th, &(args[i].ret)); + return 0; +} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop + +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data" + +perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS + +perf_dump_aux_verify "$DATA" 10 10 10 + +err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads + +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data" + +perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS + +perf_dump_aux_verify "$DATA" 10 10 10 + +err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID + +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout" + +SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO + +perf_dump_aux_tid_verify "$DATA" "$STDO" + +err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID + +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout" + +SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO + +perf_dump_aux_tid_verify "$DATA" "$STDO" + +err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10 + +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data" + +perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS + +perf_dump_aux_verify "$DATA" 10 10 10 + +err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 + +# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is. + +# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u" + +# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then + echo "Not running as root... skip" + exit 2 +fi + +TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then + DATD="$PERF_TEST_CORESIGHT_DATADIR"; +fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then + STATD="$PERF_TEST_CORESIGHT_STATDIR"; +fi + +# Called if the test fails - error code 2 +err() { + echo "$1" + exit 1 +} + +# Check that some statistics from our perf +check_val_min() { + STATF="$4" + if test "$2" -lt "$3"; then + echo ", FAILED" >> "$STATF" + err "Sanity check number of $1 is too low ($2 < $3)" + fi +} + +perf_dump_aux_verify() { + # Some basic checking that the AUX chunk contains some sensible data + # to see that we are recording something and at least a minimum + # amount of it. We should almost always see F3 atoms in just about + # anything but certainly we will see some trace info and async atom + # chunks. + DUMP="$DATD/perf-tmp-aux-dump.txt" + perf report --stdio --dump -i "$1" | \ + grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP" + # Simply count how many of these atoms we find to see that we are + # producing a reasonable amount of data - exact checks are not sane + # as this is a lossy process where we may lose some blocks and the + # compiler may produce different code depending on the compiler and + # optimization options, so this is rough just to see if we're + # either missing almost all the data or all of it + ATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l` + ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l` + ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l` + rm -f "$DUMP" + + # Arguments provide minimums for a pass + CHECK_F3_MIN="$2" + CHECK_ASYNC_MIN="$3" + CHECK_TRACE_INFO_MIN="$4" + + # Write out statistics, so over time you can track results to see if + # there is a pattern - for example we have less "noisy" results that + # produce more consistent amounts of data each run, to see if over + # time any techinques to minimize data loss are having an effect or + # not + STATF="$STATD/stats-$TEST-$DATV.csv" + if ! test -f "$STATF"; then + echo "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF" + fi + echo -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF" + + # Actually check to see if we passed or failed. + check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF" + check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF" + check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF" + echo ", Ok" >> "$STATF" +} + +perf_dump_aux_tid_verify() { + # Specifically crafted test will produce a list of Tread ID's to + # stdout that need to be checked to see that they have had trace + # info collected in AUX blocks in the perf data. This will go + # through all the TID's that are listed as CID=0xabcdef and see + # that all the Thread IDs the test tool reports are in the perf + # data AUX chunks + + # The TID test tools will print a TID per stdout line that are being + # tested + TIDS=`cat "$2"` + # Scan the perf report to find the TIDs that are actually CID in hex + # and build a list of the ones found + FOUND_TIDS=`perf report --stdio --dump -i "$1" | \ + grep -o "CID=0x[0-9a-z]+" | sed 's/CID=//g' | \ + uniq | sort | uniq` + + # Iterate over the list of TIDs that the test says it has and find + # them in the TIDs found in the perf report + MISSING="" + for TID2 in $TIDS; do + FOUND="" + for TIDHEX in $FOUND_TIDS; do + TID=`printf "%i" $TIDHEX` + if test "$TID" -eq "$TID2"; then + FOUND="y" + break + fi + done + if test -z "$FOUND"; then + MISSING="$MISSING $TID" + fi + done + if test -n "$MISSING"; then + err "Thread IDs $MISSING not found in perf AUX data" + fi +}
Hi Carsten,
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++
Very big change... Why squash all patches form previous verion to this single one big patch? Usually the format with small patches is much better for reviewing.
And I cannot apply cleanly on perf core branch: https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git branch: perf/core.
24 files changed, 823 insertions(+), 4 deletions(-) create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.* diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt @@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream.
Maybe here need to sync a bit for the terminology in Documentation/trace/coresight/coresight.rst.
Something like:
"Coresight is a feature of some Arm based processors that allows for debugging. One of the things it can do is trace instruction path and expose that information in a hardware compressed stream for either debugger or HW assisted tracing locally.
See Documentation/trace/coresight/coresight.rst for details."
Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
It is inaccurate that if we don't mention openCSD lib.
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
For build perf tool, I think above paragraphs are duplicate with the document Documentation/trace/coresight/coresight.rst. Can we simply say:
"The details for building perf tool with support Arm Coresight can be found in the "HOWTO.md" file of the openCSD gitHub repository: https://github.com/Linaro/opencsd.
And "HOWTO.md" file gives the information and examples for how to use perf tool to record and report Coresight trace data. It's the prerequisite for this perf Coresight test."
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
I prefer to remove these kernel configuration since they are not inconsistent on different platforms (e.g. ETBV10, ETM4X, etc), and some configurations might not necessary (e.g. CPU_DEBUG).
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
Please update based on the latest test case names, at my side, I can see the testing case like:
Coresight / ASM Pure Loop Coresight / Memcpy 16k 10 Threads Coresight / Thread Loop 10 Threads - Check TID Coresight / Thread Loop 2 Threads - Check TID Coresight / Unroll Loop Thread 10
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
I am confused by this narrative. Does it try to remind that the final testing result (pass or fail) is not stable? Or should we run for multiple times so have more chance to capture issues?
+Be aware that amny of these tests take quite a while to run, specifically
s/amny/many
+in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@ -all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)
+tests-coresight-targets-clean:
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean
+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets # Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
- $(Q)$(MAKE) -C tests/shell/coresight install-tests
install-bin: install-tools install-tests install-traceevent-plugins @@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS) -clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS)
I don't find other places using TESTS_CORESIGHT_TARGETS. Is this redundant?
endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
- asm_pure_loop \
- thread_loop \
- memcpy_thread \
- unroll_loop_thread
+all: $(SUBDIRS) +$(SUBDIRS):
- $(Q)$(MAKE) -C $@
+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
- $(Q)$(MAKE) -C $(@:install-%=%) install-tests
+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
- $(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null
+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop
Do we really need there '.gitignore' files under the folder 'tools/perf/tests/shell/coresight/'.
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
Remove the unused variable 'LIB='.
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests
There have four sub folders under tools/perf/tests/shell/coresight:
asm_pure_loop memcpy_thread thread_loop unroll_loop_thread
And every folder has its own Makefile and every Makefile is quite close to each other. I am just wandering if it's possible to remove the 4 Makefiles in these four sub folders, and simply use tools/perf/tests/shell/coresight/Makefile as the central place to build these assistant programs.
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
I tested the case "ASM Pure Loop" on my Juno board, and it complaints:
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9063 failed to mmap with 12 (Cannot allocate memory) test child finished with -1 ---- end ---- Coresight / ASM Pure Loop: FAILED!
Since I only setup the 1GB memory for the Linux kernel, it fails to allocate AUX ring buffer with the size 256MB. So I manully change the buffer size to 8MB in tools/perf/tests/shell/lib/coresight.sh:
PERFRECMEM="-m ,8M"
So finally I can see the test case is passed:
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9481 -m ,8M -e cs_etm//u [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.681 MB ./perf-asm_pure_loop-out.data ] test child finished with 0 ---- end ---- Coresight / ASM Pure Loop: Ok
Do you think we really need to use 256MiB as the AUX buffer size? IIRC, it means we allocate 256MiB per CPU for this case, on the other hand, you could see the final perf data file size is small (0.681 MiB).
Seems to me, it's not necessary to allocate so big buffer for the test, and I tried to run below 4 cases with 8MiB, all of them can pass the testing :)
diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- unsigned long loops;
- unsigned long size;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned long i, len = a->loops;
- unsigned char *src, *dst;
- src = malloc(a->size * 1024);
- dst = malloc(a->size * 1024);
- if ((!src) || (!dst)) {
printf("ERR: Can't allocate memory\n");exit(1);- }
- for (i = 0; i < len; i++)
memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned long i, len, size, thr;
- pthread_t threads[256];
- struct args args[256];
- long long v;
- if (argc < 4) {
printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);- }
- v = atoll(argv[1]);
- if ((v < 1) || (v > (1024 * 1024))) {
printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);- }
- size = v;
- thr = atol(argv[2]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- v = atoll(argv[3]);
- if ((v < 1) || (v > 40000000000ll)) {
printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);- }
- len = v * 100;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
- unsigned int loops;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- int i = 0, len = a->loops;
- if (getenv("SHOW_TID")) {
unsigned long long tid = gettid();printf("%llu\n", tid);- }
- asm volatile(
"loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */- );
- return (void *)(long)i;
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, len, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 3) {
printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- len = atoi(argv[2]);
- if ((len < 1) || (len > 4000)) {
printf("ERR: max loops 4000 (millions)\n");exit(1);- }
- len *= 1000000;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- pthread_t th;
- unsigned int in, out;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned int i, in = a->in;
- for (i = 0; i < 10000; i++) {
asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);- }
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 2) {
printf("ERR: %s [numthreads]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr > 256) || (thr < 1)) {
printf("ERR: threads 1-256\n");exit(1);- }
- for (i = 0; i < thr; i++) {
args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err
Can we organize the shell scripts by moving them into the folder tools/perf/tests/shell/coresight?
coresight_asm_pure_loop.sh coresight_memcpy_thread_16k_10.sh coresight_thread_loop_check_tid_10.sh coresight_thread_loop_check_tid_2.sh coresight_unroll_loop_thread_10.sh
And we even can consider to move script test_arm_coresight.sh into the folder tools/perf/tests/shell/coresight and change its name as 'coresight_smoke_test.sh'.
Thanks, Leo
diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
- echo "Not running as root... skip"
- exit 2
+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
- DATD="$PERF_TEST_CORESIGHT_DATADIR";
+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
- STATD="$PERF_TEST_CORESIGHT_STATDIR";
+fi
+# Called if the test fails - error code 2 +err() {
- echo "$1"
- exit 1
+}
+# Check that some statistics from our perf +check_val_min() {
- STATF="$4"
- if test "$2" -lt "$3"; then
echo ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"- fi
+}
+perf_dump_aux_verify() {
- # Some basic checking that the AUX chunk contains some sensible data
- # to see that we are recording something and at least a minimum
- # amount of it. We should almost always see F3 atoms in just about
- # anything but certainly we will see some trace info and async atom
- # chunks.
- DUMP="$DATD/perf-tmp-aux-dump.txt"
- perf report --stdio --dump -i "$1" | \
grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"- # Simply count how many of these atoms we find to see that we are
- # producing a reasonable amount of data - exact checks are not sane
- # as this is a lossy process where we may lose some blocks and the
- # compiler may produce different code depending on the compiler and
- # optimization options, so this is rough just to see if we're
- # either missing almost all the data or all of it
- ATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`
- ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`
- ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`
- rm -f "$DUMP"
- # Arguments provide minimums for a pass
- CHECK_F3_MIN="$2"
- CHECK_ASYNC_MIN="$3"
- CHECK_TRACE_INFO_MIN="$4"
- # Write out statistics, so over time you can track results to see if
- # there is a pattern - for example we have less "noisy" results that
- # produce more consistent amounts of data each run, to see if over
- # time any techinques to minimize data loss are having an effect or
- # not
- STATF="$STATD/stats-$TEST-$DATV.csv"
- if ! test -f "$STATF"; then
echo "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"- fi
- echo -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"
- # Actually check to see if we passed or failed.
- check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"
- check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"
- check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"
- echo ", Ok" >> "$STATF"
+}
+perf_dump_aux_tid_verify() {
- # Specifically crafted test will produce a list of Tread ID's to
- # stdout that need to be checked to see that they have had trace
- # info collected in AUX blocks in the perf data. This will go
- # through all the TID's that are listed as CID=0xabcdef and see
- # that all the Thread IDs the test tool reports are in the perf
- # data AUX chunks
- # The TID test tools will print a TID per stdout line that are being
- # tested
- TIDS=`cat "$2"`
- # Scan the perf report to find the TIDs that are actually CID in hex
- # and build a list of the ones found
- FOUND_TIDS=`perf report --stdio --dump -i "$1" | \
grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`- # Iterate over the list of TIDs that the test says it has and find
- # them in the TIDs found in the perf report
- MISSING=""
- for TID2 in $TIDS; do
FOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fi- done
- if test -n "$MISSING"; then
err "Thread IDs $MISSING not found in perf AUX data"- fi
+}
2.32.0
On 4/10/22 09:30, Leo Yan wrote:
Hi Carsten,
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++
Very big change... Why squash all patches form previous verion to this single one big patch? Usually the format with small patches is much better for reviewing.
I was asked to re-jig the tree and in doing so I also ended up cutting down the size a lot so this just makes more sense together as a "here are the tests" as adding infra without any tests makes no sense and the tests themelves are self-contained in their own directories and source files and "drivign scripts" thus it's essentially patch 1 appended to patch 2 to patch 3 etc. and still broken up in the patch file by file.
And I cannot apply cleanly on perf core branch: https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git branch: perf/core.
I generated this based on tip. I can re-do it based on the above.
24 files changed, 823 insertions(+), 4 deletions(-) create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.* diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt @@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream.
Maybe here need to sync a bit for the terminology in Documentation/trace/coresight/coresight.rst.
Something like:
"Coresight is a feature of some Arm based processors that allows for debugging. One of the things it can do is trace instruction path and expose that information in a hardware compressed stream for either debugger or HW assisted tracing locally.
See Documentation/trace/coresight/coresight.rst for details."
Sure. Can do that.
Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
It is inaccurate that if we don't mention openCSD lib.
Do you mean I need to mention that you need the opencsd library installed too?
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
For build perf tool, I think above paragraphs are duplicate with the document Documentation/trace/coresight/coresight.rst. Can we simply say:
"The details for building perf tool with support Arm Coresight can be found in the "HOWTO.md" file of the openCSD gitHub repository: https://github.com/Linaro/opencsd.
I can. I put this here as I didn't go clone OpencCSD first but used my distro OpenCSD packages and thus of course didn't have the documentation in front of me. I spent some time wondering why it wasn't building with coresight support even though it detected OpenCSD when I compiled... I didn't expect to have to go to some separate project git repository and read docs there on how to build the perf tool here in the kernel. I wrote this because it was an actual problem I hit and it's a lot less frustrating to "end users" to give them the information they need in the relevant place they need it instead of sending them around to other project trees. Building perf with coresight support is handled by the perf tree int he kernel, not OpenCSD, thus IMHO that is where the documentation belongs - alongside the thing that determines how to build something.
And "HOWTO.md" file gives the information and examples for how to use perf tool to record and report Coresight trace data. It's the prerequisite for this perf Coresight test."
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
I prefer to remove these kernel configuration since they are not inconsistent on different platforms (e.g. ETBV10, ETM4X, etc), and some configurations might not necessary (e.g. CPU_DEBUG).
Certainly there should be some documentation on which kernel configs you might want to turn on then? Imagine someone new comes along and doesn't have any idea what to possible enable at all and manages to build perf with coresight support (as above) then finds it doesn't work because they didn't enable enough config in the kernel? Sure - could probably trim these down a bit but the point here is to alert the user to there being a range of coresight config options that you need to turn on that you likely will find are not turned on. They certainly are not turned on on distro kernels and a lot of the time when you have a platform that already boots/works you start with your distro kernel config file because you want everything enabled so it actually boots. I've learned the hard way to do this as you manage to forget to turn on some MMC driver or some other feature and your boot hangs or doesn't find rootfs etc.
What would you recommend then as a "turn these on and coresight will almost certainly work for you on your given hardware " then?
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
Please update based on the latest test case names, at my side, I can see the testing case like:
Coresight / ASM Pure Loop Coresight / Memcpy 16k 10 Threads Coresight / Thread Loop 10 Threads - Check TID Coresight / Thread Loop 2 Threads - Check TID Coresight / Unroll Loop Thread 10
Oh sorry - yeah. I wrote the docs based on the earlier tests. Will fix.
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
I am confused by this narrative. Does it try to remind that the final testing result (pass or fail) is not stable? Or should we run for multiple times so have more chance to capture issues?
That is correct. I thought I was clear that it's lossy. That is actually the case. I have tests here that actually fail because there is no data collected from some threads at all (missing CID blocks for some of the threads that run in the test). The point is to have tests that may be failing now but in future will improve. I lowered the minimum bar to pass for most tests to have "at least just a little data" but most tests show highly variable amount of captured data. the csv files are there to over-time give you a good idea of the stability of the captured data.
+Be aware that amny of these tests take quite a while to run, specifically
s/amny/many
Indeed. will fix.
+in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@ -all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)
+tests-coresight-targets-clean:
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean
+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets # Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
- $(Q)$(MAKE) -C tests/shell/coresight install-tests
install-bin: install-tools install-tests install-traceevent-plugins @@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS) -clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS)
I don't find other places using TESTS_CORESIGHT_TARGETS. Is this redundant?
I'll check - it may have been left over from my previous patch set.
endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
- asm_pure_loop \
- thread_loop \
- memcpy_thread \
- unroll_loop_thread
+all: $(SUBDIRS) +$(SUBDIRS):
- $(Q)$(MAKE) -C $@
+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
- $(Q)$(MAKE) -C $(@:install-%=%) install-tests
+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
- $(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null
+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop
Do we really need there '.gitignore' files under the folder 'tools/perf/tests/shell/coresight/'.
Where would you rather have them to ignore the generated binary tools?
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
Remove the unused variable 'LIB='.
I have this because I wanted to have a simple template to be able to re-use for more tests over time. It's so much easier to maintain and extend if every makefile and tool follow a similar pattern and you can almost copy & paste between them as they don't have "exceptions". You really want me to remove this?
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests
There have four sub folders under tools/perf/tests/shell/coresight:
asm_pure_loop memcpy_thread thread_loop unroll_loop_thread
And every folder has its own Makefile and every Makefile is quite close to each other. I am just wandering if it's possible to remove the 4 Makefiles in these four sub folders, and simply use tools/perf/tests/shell/coresight/Makefile as the central place to build these assistant programs.
I did this so it's easier to etxent over time. having a single parent makefile that over time accumulates little ugly "if's" and exceptions makes longer-term maintenance and extending harder. I did it this way to make this easy - make a copy of a dir - add that dir to a parent makefile then modify the makefile as needed (but only as needed).
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
I tested the case "ASM Pure Loop" on my Juno board, and it complaints:
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9063 failed to mmap with 12 (Cannot allocate memory) test child finished with -1 ---- end ---- Coresight / ASM Pure Loop: FAILED!
Since I only setup the 1GB memory for the Linux kernel, it fails to allocate AUX ring buffer with the size 256MB. So I manully change the buffer size to 8MB in tools/perf/tests/shell/lib/coresight.sh:
PERFRECMEM="-m ,8M"
So finally I can see the test case is passed:
This is artificial isn't it? limiting to 1GB. You certainly have far more memory than that available. My testse were on a system with 4GB and I had no issues.
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9481 -m ,8M -e cs_etm//u [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.681 MB ./perf-asm_pure_loop-out.data ] test child finished with 0 ---- end ---- Coresight / ASM Pure Loop: Ok
Do you think we really need to use 256MiB as the AUX buffer size? IIRC, it means we allocate 256MiB per CPU for this case, on the other hand, you could see the final perf data file size is small (0.681 MiB).
Seems to me, it's not necessary to allocate so big buffer for the test, and I tried to run below 4 cases with 8MiB, all of them can pass the testing :)
I didn't think anyone with a system with coresight support that would be running perf record locally would only have 1GB of ram... I knew junos had 8GB and my dragonboard has 4GB ... so I know I was on the smaller side. I thought a larger buffer == safer results (less chance of needing to write out the buffer during capture). Admittdly I used 256Mb when my tests ran for much longer and collected more data. I can try drop to 8 or 16gb and see.
diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- unsigned long loops;
- unsigned long size;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned long i, len = a->loops;
- unsigned char *src, *dst;
- src = malloc(a->size * 1024);
- dst = malloc(a->size * 1024);
- if ((!src) || (!dst)) {
printf("ERR: Can't allocate memory\n");exit(1);- }
- for (i = 0; i < len; i++)
memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned long i, len, size, thr;
- pthread_t threads[256];
- struct args args[256];
- long long v;
- if (argc < 4) {
printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);- }
- v = atoll(argv[1]);
- if ((v < 1) || (v > (1024 * 1024))) {
printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);- }
- size = v;
- thr = atol(argv[2]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- v = atoll(argv[3]);
- if ((v < 1) || (v > 40000000000ll)) {
printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);- }
- len = v * 100;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
- unsigned int loops;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- int i = 0, len = a->loops;
- if (getenv("SHOW_TID")) {
unsigned long long tid = gettid();printf("%llu\n", tid);- }
- asm volatile(
"loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */- );
- return (void *)(long)i;
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, len, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 3) {
printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- len = atoi(argv[2]);
- if ((len < 1) || (len > 4000)) {
printf("ERR: max loops 4000 (millions)\n");exit(1);- }
- len *= 1000000;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- pthread_t th;
- unsigned int in, out;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned int i, in = a->in;
- for (i = 0; i < 10000; i++) {
asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);- }
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 2) {
printf("ERR: %s [numthreads]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr > 256) || (thr < 1)) {
printf("ERR: threads 1-256\n");exit(1);- }
- for (i = 0; i < thr; i++) {
args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err
Can we organize the shell scripts by moving them into the folder tools/perf/tests/shell/coresight?
We can - but it comes with a fair few more changes.
coresight_asm_pure_loop.sh coresight_memcpy_thread_16k_10.sh coresight_thread_loop_check_tid_10.sh coresight_thread_loop_check_tid_2.sh coresight_unroll_loop_thread_10.sh
And we even can consider to move script test_arm_coresight.sh into the folder tools/perf/tests/shell/coresight and change its name as 'coresight_smoke_test.sh'.
Indeed these other tests I left alone for now and had not thought about how to marry these together yet - leaving this for another day and another patch set rather than this patch set itself. That was my thoguht. I was trying to make an "Easier to extend by just dropping a test into a dir" setup here to make maintenance and expansion easier over time (and thus encourage testing by having a simple repeatable test infra to duplicate). I ended up with a dir per test tool you need to build and a driver script in the tests/shell dir. I think this is certainly worth considering but perhaps as a separate set of work to marry these?
I piggybacked on the existing shell test infra but added a fair few more scripts. To do what you suggest I'd need to modify the core shell test code to walk subdirs recursively then looking for child scripts. The problem is how does perf test's shell handling know about the coresight subdir vs the lib subdir? Both contain *.sh shell scripts - the difference is the ones in lib are not executable. Is this sufficiently different? I could also open them to check the have #!/bin/... as the first line. Hardcoding just a single coresight subdir just feels wrong and hacky to me, thus the generic recursion solution I suggest here.
I can definitely see how extending to subdirs would make supporting testing cleaner and divide things into their own domains (dirs).
Thanks, Leo
diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
- echo "Not running as root... skip"
- exit 2
+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
- DATD="$PERF_TEST_CORESIGHT_DATADIR";
+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
- STATD="$PERF_TEST_CORESIGHT_STATDIR";
+fi
+# Called if the test fails - error code 2 +err() {
- echo "$1"
- exit 1
+}
+# Check that some statistics from our perf +check_val_min() {
- STATF="$4"
- if test "$2" -lt "$3"; then
echo ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"- fi
+}
+perf_dump_aux_verify() {
- # Some basic checking that the AUX chunk contains some sensible data
- # to see that we are recording something and at least a minimum
- # amount of it. We should almost always see F3 atoms in just about
- # anything but certainly we will see some trace info and async atom
- # chunks.
- DUMP="$DATD/perf-tmp-aux-dump.txt"
- perf report --stdio --dump -i "$1" | \
grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"- # Simply count how many of these atoms we find to see that we are
- # producing a reasonable amount of data - exact checks are not sane
- # as this is a lossy process where we may lose some blocks and the
- # compiler may produce different code depending on the compiler and
- # optimization options, so this is rough just to see if we're
- # either missing almost all the data or all of it
- ATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`
- ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`
- ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`
- rm -f "$DUMP"
- # Arguments provide minimums for a pass
- CHECK_F3_MIN="$2"
- CHECK_ASYNC_MIN="$3"
- CHECK_TRACE_INFO_MIN="$4"
- # Write out statistics, so over time you can track results to see if
- # there is a pattern - for example we have less "noisy" results that
- # produce more consistent amounts of data each run, to see if over
- # time any techinques to minimize data loss are having an effect or
- # not
- STATF="$STATD/stats-$TEST-$DATV.csv"
- if ! test -f "$STATF"; then
echo "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"- fi
- echo -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"
- # Actually check to see if we passed or failed.
- check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"
- check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"
- check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"
- echo ", Ok" >> "$STATF"
+}
+perf_dump_aux_tid_verify() {
- # Specifically crafted test will produce a list of Tread ID's to
- # stdout that need to be checked to see that they have had trace
- # info collected in AUX blocks in the perf data. This will go
- # through all the TID's that are listed as CID=0xabcdef and see
- # that all the Thread IDs the test tool reports are in the perf
- # data AUX chunks
- # The TID test tools will print a TID per stdout line that are being
- # tested
- TIDS=`cat "$2"`
- # Scan the perf report to find the TIDs that are actually CID in hex
- # and build a list of the ones found
- FOUND_TIDS=`perf report --stdio --dump -i "$1" | \
grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`- # Iterate over the list of TIDs that the test says it has and find
- # them in the TIDs found in the perf report
- MISSING=""
- for TID2 in $TIDS; do
FOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fi- done
- if test -n "$MISSING"; then
err "Thread IDs $MISSING not found in perf AUX data"- fi
+}
2.32.0
Hi Carsten,
Sorry for late response.
On Thu, Apr 21, 2022 at 06:38:33PM +0100, Carsten Haitzler wrote:
[...]
Very big change... Why squash all patches form previous verion to this single one big patch? Usually the format with small patches is much better for reviewing.
I was asked to re-jig the tree and in doing so I also ended up cutting down the size a lot so this just makes more sense together as a "here are the tests" as adding infra without any tests makes no sense and the tests themelves are self-contained in their own directories and source files and "drivign scripts" thus it's essentially patch 1 appended to patch 2 to patch 3 etc. and still broken up in the patch file by file.
I am not sure if I understand the meaning, seems to me you could organize the patch series like:
- Patch for common files (e.g. script lib/coresight.sh or some Makefile changes); - Patches for enabling test cases, E.g.: patch for asm_pure_loop; patch for thread loop (include unroll loop); patch for memcpy; - Patch for documentation.
If this is not comfortable for you, at least we can use three patches:
- Patch for common file (e.g. script lib/coresight.sh); - Patch for test cases; - Patch for documentation.
[...]
+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
It is inaccurate that if we don't mention openCSD lib.
Do you mean I need to mention that you need the opencsd library installed too?
Yes, otherwise, users might directly build perf without opencsd lib, then finally they cannot use perf with Arm CoreSight.
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
For build perf tool, I think above paragraphs are duplicate with the document Documentation/trace/coresight/coresight.rst. Can we simply say:
"The details for building perf tool with support Arm Coresight can be found in the "HOWTO.md" file of the openCSD gitHub repository: https://github.com/Linaro/opencsd.
I can. I put this here as I didn't go clone OpencCSD first but used my distro OpenCSD packages and thus of course didn't have the documentation in front of me. I spent some time wondering why it wasn't building with coresight support even though it detected OpenCSD when I compiled... I didn't expect to have to go to some separate project git repository and read docs there on how to build the perf tool here in the kernel. I wrote this because it was an actual problem I hit and it's a lot less frustrating to "end users" to give them the information they need in the relevant place they need it instead of sending them around to other project trees. Building perf with coresight support is handled by the perf tree int he kernel, not OpenCSD, thus IMHO that is where the documentation belongs - alongside the thing that determines how to build something.
Understand.
And "HOWTO.md" file gives the information and examples for how to use perf tool to record and report Coresight trace data. It's the prerequisite for this perf Coresight test."
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
I prefer to remove these kernel configuration since they are not inconsistent on different platforms (e.g. ETBV10, ETM4X, etc), and some configurations might not necessary (e.g. CPU_DEBUG).
Certainly there should be some documentation on which kernel configs you might want to turn on then? Imagine someone new comes along and doesn't have any idea what to possible enable at all and manages to build perf with coresight support (as above) then finds it doesn't work because they didn't enable enough config in the kernel? Sure - could probably trim these down a bit but the point here is to alert the user to there being a range of coresight config options that you need to turn on that you likely will find are not turned on. They certainly are not turned on on distro kernels and a lot of the time when you have a platform that already boots/works you start with your distro kernel config file because you want everything enabled so it actually boots. I've learned the hard way to do this as you manage to forget to turn on some MMC driver or some other feature and your boot hangs or doesn't find rootfs etc.
So far, we will have two documents in Linux kernel:
- Documentation/trace/coresight/coresight.rst; - tools/perf/Documentation/arm-coresight.txt.
We need to avoid overlap between these two files. I think we could use the file Documentation/trace/coresight/coresight.rst to focus on CoreSight driver module relates stuffs and tools/perf/Documentation/arm-coresight.txt is more about the perf usages.
But, the file Documentation/trace/coresight/coresight.rst doesn't give any info for kernel configs, I think which would be a better place to give information for building kernel modules.
What would you recommend then as a "turn these on and coresight will almost certainly work for you on your given hardware " then?
This would be fine. Alternatively, we could add a section in the file Documentation/trace/coresight/coresight.rst to describe how to build CoreSight modules.
How you think for this? I also would like to get suggestions from CoreSight maintainers Suzuki/Mathieu/Mike.
[...]
Please update based on the latest test case names, at my side, I can see the testing case like:
Coresight / ASM Pure Loop Coresight / Memcpy 16k 10 Threads Coresight / Thread Loop 10 Threads - Check TID Coresight / Thread Loop 2 Threads - Check TID Coresight / Unroll Loop Thread 10Oh sorry - yeah. I wrote the docs based on the earlier tests. Will fix.
Thanks.
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
I am confused by this narrative. Does it try to remind that the final testing result (pass or fail) is not stable? Or should we run for multiple times so have more chance to capture issues?
That is correct. I thought I was clear that it's lossy. That is actually the case. I have tests here that actually fail because there is no data collected from some threads at all (missing CID blocks for some of the threads that run in the test). The point is to have tests that may be failing now but in future will improve. I lowered the minimum bar to pass for most tests to have "at least just a little data" but most tests show highly variable amount of captured data. the csv files are there to over-time give you a good idea of the stability of the captured data.
Okay, this would be fine for me. Though I am a bit worry that later if users report a failure, then how we can tell them this is a bug or it's just tracing quality issue?
[...]
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop
Do we really need there '.gitignore' files under the folder 'tools/perf/tests/shell/coresight/'.
Where would you rather have them to ignore the generated binary tools?
It's interesting that I wanted to find a case to object you, so I tried to check the folder linux/samples/bpf, but it does use .gitignore file to ignore built binaries :)
Adding .gitignore is in practice and this would be fine for me.
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
Remove the unused variable 'LIB='.
I have this because I wanted to have a simple template to be able to re-use for more tests over time. It's so much easier to maintain and extend if every makefile and tool follow a similar pattern and you can almost copy & paste between them as they don't have "exceptions". You really want me to remove this?
It's fine to keep it. Could you add a comment for this?
To be honest, I am not experienced for bash shell script, so I have no idea why write like this way. If you think this is very common usage in shell, then you could keep it and don't need to add comment.
[...]
There have four sub folders under tools/perf/tests/shell/coresight:
asm_pure_loop memcpy_thread thread_loop unroll_loop_thread
And every folder has its own Makefile and every Makefile is quite close to each other. I am just wandering if it's possible to remove the 4 Makefiles in these four sub folders, and simply use tools/perf/tests/shell/coresight/Makefile as the central place to build these assistant programs.
I did this so it's easier to etxent over time. having a single parent makefile that over time accumulates little ugly "if's" and exceptions makes longer-term maintenance and extending harder. I did it this way to make this easy - make a copy of a dir - add that dir to a parent makefile then modify the makefile as needed (but only as needed).
Okay, let's keep the saperate makefiles.
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
I tested the case "ASM Pure Loop" on my Juno board, and it complaints:
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9063 failed to mmap with 12 (Cannot allocate memory) test child finished with -1 ---- end ---- Coresight / ASM Pure Loop: FAILED!
Since I only setup the 1GB memory for the Linux kernel, it fails to allocate AUX ring buffer with the size 256MB. So I manully change the buffer size to 8MB in tools/perf/tests/shell/lib/coresight.sh:
PERFRECMEM="-m ,8M"
So finally I can see the test case is passed:
This is artificial isn't it? limiting to 1GB. You certainly have far more memory than that available. My testse were on a system with 4GB and I had no issues.
Please see below comment.
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9481 -m ,8M -e cs_etm//u [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.681 MB ./perf-asm_pure_loop-out.data ] test child finished with 0 ---- end ---- Coresight / ASM Pure Loop: Ok
Do you think we really need to use 256MiB as the AUX buffer size? IIRC, it means we allocate 256MiB per CPU for this case, on the other hand, you could see the final perf data file size is small (0.681 MiB).
Seems to me, it's not necessary to allocate so big buffer for the test, and I tried to run below 4 cases with 8MiB, all of them can pass the testing :)
I didn't think anyone with a system with coresight support that would be running perf record locally would only have 1GB of ram... I knew junos had 8GB and my dragonboard has 4GB ... so I know I was on the smaller side. I thought a larger buffer == safer results (less chance of needing to write out the buffer during capture). Admittdly I used 256Mb when my tests ran for much longer and collected more data. I can try drop to 8 or 16gb and see.
Yes, my Juno board has 8GB but I also have DB410c with 1GB with quad coes [1]. I am still concern for 256MB buffer size, it's not friendly for embedded system, and even not good for server. For example, if we run this testing on Arm server with 96 cores (like Hisilicon D06 board), then we need the buffer size is:
256MiB * 96 = 16GiB
I agree usually 16GiB is not a problem for server, but seems to me it doesn't make much sense to consume huge memory resource for the testing.
In other words, if set 8MiB (or 16MiB, 32MiB) buffer size and doesn't see testing result regression, I think this would be good to decrease the buffer size.
[1] https://www.96boards.org/product/dragonboard410c/
[...]
diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err
Can we organize the shell scripts by moving them into the folder tools/perf/tests/shell/coresight?
We can - but it comes with a fair few more changes.
coresight_asm_pure_loop.sh coresight_memcpy_thread_16k_10.sh coresight_thread_loop_check_tid_10.sh coresight_thread_loop_check_tid_2.sh coresight_unroll_loop_thread_10.sh
And we even can consider to move script test_arm_coresight.sh into the folder tools/perf/tests/shell/coresight and change its name as 'coresight_smoke_test.sh'.
Indeed these other tests I left alone for now and had not thought about how to marry these together yet - leaving this for another day and another patch set rather than this patch set itself. That was my thoguht. I was trying to make an "Easier to extend by just dropping a test into a dir" setup here to make maintenance and expansion easier over time (and thus encourage testing by having a simple repeatable test infra to duplicate). I ended up with a dir per test tool you need to build and a driver script in the tests/shell dir. I think this is certainly worth considering but perhaps as a separate set of work to marry these?
Okay, it would be fine to use separate set for moving the script test_arm_coresight.sh, which is a simple case.
I piggybacked on the existing shell test infra but added a fair few more scripts. To do what you suggest I'd need to modify the core shell test code to walk subdirs recursively then looking for child scripts. The problem is how does perf test's shell handling know about the coresight subdir vs the lib subdir?
Yeah, now I understand your point. How about file layout like below?
tools/perf/tests/shell/coresight_test_hub.sh tools/perf/tests/shell/coresight/coresight_asm_pure_loop.sh tools/perf/tests/shell/coresight/coresight_memcpy_thread_16k_10.sh tools/perf/tests/shell/coresight/coresight_thread_loop_check_tid_10.sh tools/perf/tests/shell/coresight/coresight_thread_loop_check_tid_2.sh tools/perf/tests/shell/coresight/coresight_unroll_loop_thread_10.sh
So we use tools/perf/tests/shell/coresight_test_hub.sh as an interface to hook with Perf test infrastructure, and then hub.sh file calls testing scripts under the sub folder. Seems to me, this is also friendly for later's extension.
Both contain *.sh shell scripts - the difference is the ones in lib are not executable. Is this sufficiently different? I could also open them to check the have #!/bin/... as the first line. Hardcoding just a single coresight subdir just feels wrong and hacky to me, thus the generic recursion solution I suggest here.
Agreed. I also don't prefer this way.
I can definitely see how extending to subdirs would make supporting testing cleaner and divide things into their own domains (dirs).
Thanks a lot for the work! The test cases are good for me (but I would say Mike is the best person for reviewing testing trace data quaility), I just want to make sure it's not hard for later maintenance.
Leo
On Thu, May 26, 2022 at 04:20:39PM +0800, Leo Yan wrote:
[...]
Yes, my Juno board has 8GB but I also have DB410c with 1GB with quad coes [1]. I am still concern for 256MB buffer size, it's not friendly for embedded system, and even not good for server. For example, if we run this testing on Arm server with 96 cores (like Hisilicon D06 board), then we need the buffer size is:
256MiB * 96 = 16GiB
Correct for my bad math... here should be 256MiB * 96 = 24GiB.
On 5/26/22 09:20, Leo Yan wrote:
Hi Carsten,
Sorry for late response.
On Thu, Apr 21, 2022 at 06:38:33PM +0100, Carsten Haitzler wrote:
[...]
Very big change... Why squash all patches form previous verion to this single one big patch? Usually the format with small patches is much better for reviewing.
I was asked to re-jig the tree and in doing so I also ended up cutting down the size a lot so this just makes more sense together as a "here are the tests" as adding infra without any tests makes no sense and the tests themelves are self-contained in their own directories and source files and "drivign scripts" thus it's essentially patch 1 appended to patch 2 to patch 3 etc. and still broken up in the patch file by file.
I am not sure if I understand the meaning, seems to me you could organize the patch series like:
- Patch for common files (e.g. script lib/coresight.sh or some Makefile changes);
- Patches for enabling test cases, E.g.: patch for asm_pure_loop; patch for thread loop (include unroll loop); patch for memcpy;
- Patch for documentation.
Each of those last patches are simple a patch with added "fluff" to add a line to a makefile each time then entire new files. A single patch set already divides each of these into their own files in the same patch are already divided - the only difference is - do you get the added fluff fiddling to add files to the makefile or just have them all added at once and the rest is already broken up. If you scroll through a patch with 3 files, or have 3 patches with 1 file in each - it's the same in the end. It's divided either way, but creating a patch set where one patch after the other adds a new test and each adds a line to the same makefle is a more work as I can't just git add/commit the test files to each patch. I have to first remove all tests. Then add the parent makefile for the tools with 1 line in it for 1 tool and 1 set of patch scripts. I then have to add the next one and edit this parent makefile to add one line and so on - where it actually makes the total size of all the patches bigger as this makefile that lists all the child dirs keeps being modified only to just add a single line added to it and the rest is all stand-alone new files anyway. As all the changes to this parent makefile all modify the same context, they can't be separated out and cherry-picked as they then get conflicts in the context anyways so the patches can't be chosen separately without resolving conflicts anyway, thus to me it makes no sense to do all the extra work for a bigger total patch set that can't be split out without conflicts anyway...
But I kind of give up. It seems there is a view (that I don't understand as over-all the data is more complex and larger to review when broken out) that more data in N patches is preferable to less data in a single patch (even though it's all broken out into different file blocks in the same patch so you scroll down and basically in essence "cat *.patch" and "cat onefile.patch" produce the same thing other than the "add one more line to this makefile"). I'll break it up - it just is more busy-work to create a bigger overall patch set with more changes to look at overall.
If this is not comfortable for you, at least we can use three patches:
- Patch for common file (e.g. script lib/coresight.sh);
- Patch for test cases;
- Patch for documentation.
The documentation I can see breaking out - though that wants to be moved somewhere else in the tree now and not alongside the perf tool. I generally go for the view of documentation belongs with the thing it documents and if there is new information to document, it belongs with the patch introducing what it documents. :)
[...]
+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
It is inaccurate that if we don't mention openCSD lib.
Do you mean I need to mention that you need the opencsd library installed too?
Yes, otherwise, users might directly build perf without opencsd lib, then finally they cannot use perf with Arm CoreSight.
OK - sure. Will document that OpenCSD is needed and where to possibly get it (git, packages etc.).
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
For build perf tool, I think above paragraphs are duplicate with the document Documentation/trace/coresight/coresight.rst. Can we simply say:
"The details for building perf tool with support Arm Coresight can be found in the "HOWTO.md" file of the openCSD gitHub repository: https://github.com/Linaro/opencsd.
I can. I put this here as I didn't go clone OpencCSD first but used my distro OpenCSD packages and thus of course didn't have the documentation in front of me. I spent some time wondering why it wasn't building with coresight support even though it detected OpenCSD when I compiled... I didn't expect to have to go to some separate project git repository and read docs there on how to build the perf tool here in the kernel. I wrote this because it was an actual problem I hit and it's a lot less frustrating to "end users" to give them the information they need in the relevant place they need it instead of sending them around to other project trees. Building perf with coresight support is handled by the perf tree int he kernel, not OpenCSD, thus IMHO that is where the documentation belongs - alongside the thing that determines how to build something.
Understand.
I'll move this over to the place that Matthieu and you mention (Documentation/trace/coresight), but I do think this kind of information is necessary as above, so thanks. :)
And "HOWTO.md" file gives the information and examples for how to use perf tool to record and report Coresight trace data. It's the prerequisite for this perf Coresight test."
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
I prefer to remove these kernel configuration since they are not inconsistent on different platforms (e.g. ETBV10, ETM4X, etc), and some configurations might not necessary (e.g. CPU_DEBUG).
Certainly there should be some documentation on which kernel configs you might want to turn on then? Imagine someone new comes along and doesn't have any idea what to possible enable at all and manages to build perf with coresight support (as above) then finds it doesn't work because they didn't enable enough config in the kernel? Sure - could probably trim these down a bit but the point here is to alert the user to there being a range of coresight config options that you need to turn on that you likely will find are not turned on. They certainly are not turned on on distro kernels and a lot of the time when you have a platform that already boots/works you start with your distro kernel config file because you want everything enabled so it actually boots. I've learned the hard way to do this as you manage to forget to turn on some MMC driver or some other feature and your boot hangs or doesn't find rootfs etc.
So far, we will have two documents in Linux kernel:
- Documentation/trace/coresight/coresight.rst;
- tools/perf/Documentation/arm-coresight.txt.
We need to avoid overlap between these two files. I think we could use the file Documentation/trace/coresight/coresight.rst to focus on CoreSight driver module relates stuffs and tools/perf/Documentation/arm-coresight.txt is more about the perf usages.
That's the case here with this patch set. To me - I would look for the perf docs where I put them - in the perf documentation directory, not somewhere else.
But, the file Documentation/trace/coresight/coresight.rst doesn't give any info for kernel configs, I think which would be a better place to give information for building kernel modules.
Correct. I do think you're right. I think that the "core docs" should give this information and the perf tool docs should at least reference these so you know to look there. I think I should at least put a stub doc in the perf tool doc tree to tell someone to go look at the "core docs" for more information.
What would you recommend then as a "turn these on and coresight will almost certainly work for you on your given hardware " then?
This would be fine. Alternatively, we could add a section in the file Documentation/trace/coresight/coresight.rst to describe how to build CoreSight modules.
I think that is good/best and then "link" the docs so someone knows where to look for more information.
How you think for this? I also would like to get suggestions from CoreSight maintainers Suzuki/Mathieu/Mike.
[...]
Please update based on the latest test case names, at my side, I can see the testing case like:
Coresight / ASM Pure Loop Coresight / Memcpy 16k 10 Threads Coresight / Thread Loop 10 Threads - Check TID Coresight / Thread Loop 2 Threads - Check TID Coresight / Unroll Loop Thread 10Oh sorry - yeah. I wrote the docs based on the earlier tests. Will fix.
Thanks.
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
I am confused by this narrative. Does it try to remind that the final testing result (pass or fail) is not stable? Or should we run for multiple times so have more chance to capture issues?
That is correct. I thought I was clear that it's lossy. That is actually the case. I have tests here that actually fail because there is no data collected from some threads at all (missing CID blocks for some of the threads that run in the test). The point is to have tests that may be failing now but in future will improve. I lowered the minimum bar to pass for most tests to have "at least just a little data" but most tests show highly variable amount of captured data. the csv files are there to over-time give you a good idea of the stability of the captured data.
Okay, this would be fine for me. Though I am a bit worry that later if users report a failure, then how we can tell them this is a bug or it's just tracing quality issue?
Well this is kind of both a bug and a quality issue. Reality is I have tests that literally get no information in the perf trace for some threads. Reality is we have what I'd call "Quality of trace" bugs and I was doing these test as a way of beginning to explore that and quantify it as that is not something the existing tests were doing. We want to have tests that in future with new designs show the quality bug to be fixed. It's kind of test driven development. Expose the quality bug, document it as such and then in future show it to be fixed when that happens (v9 should fix this...). I probably should document this as such (as a quality bug and why such failures are to be expected) so I'll add that to my TODO to make sure it's clear that these tests might fail and it's a quality issue (I mention the CSV files that are generated as a way to over-time track that quality).
[...]
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop
Do we really need there '.gitignore' files under the folder 'tools/perf/tests/shell/coresight/'.
Where would you rather have them to ignore the generated binary tools?
It's interesting that I wanted to find a case to object you, so I tried to check the folder linux/samples/bpf, but it does use .gitignore file to ignore built binaries :)
Adding .gitignore is in practice and this would be fine for me.
:)
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
Remove the unused variable 'LIB='.
I have this because I wanted to have a simple template to be able to re-use for more tests over time. It's so much easier to maintain and extend if every makefile and tool follow a similar pattern and you can almost copy & paste between them as they don't have "exceptions". You really want me to remove this?
It's fine to keep it. Could you add a comment for this?
No problems! Can do.
To be honest, I am not experienced for bash shell script, so I have no idea why write like this way. If you think this is very common usage in shell, then you could keep it and don't need to add comment.
Well this is a makefile above (not shell), but having done lots of makefile and other higher level make tooling (autotools, cmake, meson) and having to maintain larger trees with "We build the same kind of thing N times but each module/tool needs slightly different linking/includes/tooling" ... I have learned the value of using templates that are the same but sometimes you have empty fields or fields that change. When you have to fix/change a "design pattern" issue it's easier to do when the files share more in common and it's easier for someone to start a new tool/module/subdir by taking an existing template and just modifying it as needed and everything they need is there. It really makes maintenance so much easier in the long-run as you expand the set of things something builds. Design patterns in your build system really help. :)
[...]
There have four sub folders under tools/perf/tests/shell/coresight:
asm_pure_loop memcpy_thread thread_loop unroll_loop_threadAnd every folder has its own Makefile and every Makefile is quite close to each other. I am just wandering if it's possible to remove the 4 Makefiles in these four sub folders, and simply use tools/perf/tests/shell/coresight/Makefile as the central place to build these assistant programs.
I did this so it's easier to etxent over time. having a single parent makefile that over time accumulates little ugly "if's" and exceptions makes longer-term maintenance and extending harder. I did it this way to make this easy - make a copy of a dir - add that dir to a parent makefile then modify the makefile as needed (but only as needed).
Okay, let's keep the saperate makefiles.
:)
diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
I tested the case "ASM Pure Loop" on my Juno board, and it complaints:
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9063 failed to mmap with 12 (Cannot allocate memory) test child finished with -1 ---- end ---- Coresight / ASM Pure Loop: FAILED!
Since I only setup the 1GB memory for the Linux kernel, it fails to allocate AUX ring buffer with the size 256MB. So I manully change the buffer size to 8MB in tools/perf/tests/shell/lib/coresight.sh:
PERFRECMEM="-m ,8M"So finally I can see the test case is passed:
This is artificial isn't it? limiting to 1GB. You certainly have far more memory than that available. My testse were on a system with 4GB and I had no issues.
Please see below comment.
root@debian:/mnt/export/arm-linux-kernel/tools/perf# ./perf test -v 76 76: Coresight / ASM Pure Loop : --- start --- test child forked, pid 9481 -m ,8M -e cs_etm//u [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.681 MB ./perf-asm_pure_loop-out.data ] test child finished with 0 ---- end ---- Coresight / ASM Pure Loop: Ok
Do you think we really need to use 256MiB as the AUX buffer size? IIRC, it means we allocate 256MiB per CPU for this case, on the other hand, you could see the final perf data file size is small (0.681 MiB).
Seems to me, it's not necessary to allocate so big buffer for the test, and I tried to run below 4 cases with 8MiB, all of them can pass the testing :)
I didn't think anyone with a system with coresight support that would be running perf record locally would only have 1GB of ram... I knew junos had 8GB and my dragonboard has 4GB ... so I know I was on the smaller side. I thought a larger buffer == safer results (less chance of needing to write out the buffer during capture). Admittdly I used 256Mb when my tests ran for much longer and collected more data. I can try drop to 8 or 16gb and see.
Yes, my Juno board has 8GB but I also have DB410c with 1GB with quad coes [1]. I am still concern for 256MB buffer size, it's not friendly for embedded system, and even not good for server. For example, if we run this testing on Arm server with 96 cores (like Hisilicon D06 board), then we need the buffer size is:
256MiB * 96 = 16GiB
A bit more actually 24M - but I see you correct this already in a following mail, but your point remains the same. :)
I agree usually 16GiB is not a problem for server, but seems to me it doesn't make much sense to consume huge memory resource for the testing.
Even 24GB is not a problem IMHO if you have 96 cores, but ... I see your point.
In other words, if set 8MiB (or 16MiB, 32MiB) buffer size and doesn't see testing result regression, I think this would be good to decrease the buffer size.
I'll drop it down to something less - sure. Maybe between 16-64M and that should mean everything you have there is able to do this without issues. 1GB with 8 cores would only need 512M for buffers - you have a lot left over. So 64M or less would be just fine I think (and with 4 cores... even more headroom).
[1] https://www.96boards.org/product/dragonboard410c/
[...]
diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err
Can we organize the shell scripts by moving them into the folder tools/perf/tests/shell/coresight?
We can - but it comes with a fair few more changes.
coresight_asm_pure_loop.sh coresight_memcpy_thread_16k_10.sh coresight_thread_loop_check_tid_10.sh coresight_thread_loop_check_tid_2.sh coresight_unroll_loop_thread_10.shAnd we even can consider to move script test_arm_coresight.sh into the folder tools/perf/tests/shell/coresight and change its name as 'coresight_smoke_test.sh'.
Indeed these other tests I left alone for now and had not thought about how to marry these together yet - leaving this for another day and another patch set rather than this patch set itself. That was my thoguht. I was trying to make an "Easier to extend by just dropping a test into a dir" setup here to make maintenance and expansion easier over time (and thus encourage testing by having a simple repeatable test infra to duplicate). I ended up with a dir per test tool you need to build and a driver script in the tests/shell dir. I think this is certainly worth considering but perhaps as a separate set of work to marry these?
Okay, it would be fine to use separate set for moving the script test_arm_coresight.sh, which is a simple case.
I think that breaking out test_arm_coresight.sh into sub-tests that are stand-alone is better. I've had to deal with test suites that have one test case and that test fails/succeeds BUT this test case actually tests 30+ API calls or something like that and one of those fail. Which one? You don't immediately know. You have to now go digging through log files to find out. Breaking this out into less efficient but easier to see "one test per thing you test" is much better and long-run really saves time and headaches, but this is "let's do this another day" but it is why I am over-engineering the tests with the view that these will expand in number and move over such tests.
I piggybacked on the existing shell test infra but added a fair few more scripts. To do what you suggest I'd need to modify the core shell test code to walk subdirs recursively then looking for child scripts. The problem is how does perf test's shell handling know about the coresight subdir vs the lib subdir?
Yeah, now I understand your point. How about file layout like below?
tools/perf/tests/shell/coresight_test_hub.sh tools/perf/tests/shell/coresight/coresight_asm_pure_loop.sh tools/perf/tests/shell/coresight/coresight_memcpy_thread_16k_10.sh tools/perf/tests/shell/coresight/coresight_thread_loop_check_tid_10.sh tools/perf/tests/shell/coresight/coresight_thread_loop_check_tid_2.sh tools/perf/tests/shell/coresight/coresight_unroll_loop_thread_10.sh
So we use tools/perf/tests/shell/coresight_test_hub.sh as an interface to hook with Perf test infrastructure, and then hub.sh file calls testing scripts under the sub folder. Seems to me, this is also friendly for later's extension.
Aaaaah here I disagree. The current test infra will only see "coresight_test_hub.sh" and then that all passes, fails or skips as a single test (see above). ub-tests are hidden when you do "perf test". I've had to deal with this before and it is (IMHO) a horrible way to go as you spend time then re-running the tests by hand and digging through other logs to see what failed rather than the toplevel tests indicating right there pass, fail or skip. Over time it becomes a time-sink to hunt these and mystifies new people who don't know where to dig for the specific failure.
To break them up like you want into subdirs then requires me to go to perf test (builtin-script.c) and change how the walking of test subdirs works with shells. It seems to have some legacy code in there that walks some "lang" dirs with "bin" dirs in them that seemingly do not exist today in the kernel tree. The right way to do this if you want to break this out into a coresight dir would be to modify how perf test walks dirs and builds the list of tests. I'd have to walk all subdirs then specifically filter for tests it knows will exist (executable, end in .sh, maybe be named like parentdir/parentdir_XXX.sh like your above suggestion of coresight/coresight_XXX.sh etc.). This now involves re-jigging a lot more code and abstracting the walking into a single function that walks once then stores the data in an array of struct pef_shell_test or such which now is filtered (and sorted which we should do which we don't now), and thus clean this all up, remove the (seemingly) legacy lang/bin dir walk code (I need to dig through history to find out where this came in and why it was there) and so on.
This certainly then makes this set of patches bigger and I need to break out this "revamp" of the walking of test shell scripts a preparation patch, but just saying: It certainly raises the bar for work and I don't think having a single parent .sh that then walks all the children and produces a single test output is right. I could popen this test and read the output to generate multiple output tests but, IMHO, that is worse and more complex than fixing the tree walk above to build an index of all shell tests once etc. like:
struct script_file { char *dir; char *file; char *label; }
struct script_file *list_script_files(void);
... so we run this whenever we want a list of shell tests and just iterate over it until a NULL member (dir, file, label are all NULL). Simple and does the job but moves all the walking into a single function instead of there being like 3 of them now.
Both contain *.sh shell scripts - the difference is the ones in lib are not executable. Is this sufficiently different? I could also open them to check the have #!/bin/... as the first line. Hardcoding just a single coresight subdir just feels wrong and hacky to me, thus the generic recursion solution I suggest here.
Agreed. I also don't prefer this way.
I can definitely see how extending to subdirs would make supporting testing cleaner and divide things into their own domains (dirs).
Thanks a lot for the work! The test cases are good for me (but I would say Mike is the best person for reviewing testing trace data quaility), I just want to make sure it's not hard for later maintenance.
Sure. Let me at least make a start on the shell-test walk code and cleaning it up. I've just been staring at this and trying to find a "better way" but I'm pretty much at "Do it my current way with no subdir and keep the patch simple" or "Make this a more complex a patch to try streamline the code and abstract the shell tests better and include sub-dir walking" as I describe above. Long term I think this is actually a very good idea to clean this up and allow tests then in general to go into domain-specific subdirs. It just brings forward this kind of work ahead of getting extra tests in. I can't really see another sensible option.
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++ 24 files changed, 823 insertions(+), 4 deletions(-)
As Leo pointed out this is a big patch and hard to digest intellectually.
create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.* diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt
I think it would be best to keep all the coresight documentation under the current coresight documentation repository[1]. That way all the information on coresight can be found in a central place.
Some part of what is added by this patch is redundant with what is currently available in [1]. Other parts are tests specific and should be added under something like "coresight-perf-test.rst".
Thanks, Mathieu
[1]. Documentation/trace/coresight/
@@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream. Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
+Be aware that amny of these tests take quite a while to run, specifically +in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@ -all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)
+tests-coresight-targets-clean:
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean
+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets # Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
- $(Q)$(MAKE) -C tests/shell/coresight install-tests
install-bin: install-tools install-tests install-traceevent-plugins @@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS) -clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS) endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
- asm_pure_loop \
- thread_loop \
- memcpy_thread \
- unroll_loop_thread
+all: $(SUBDIRS) +$(SUBDIRS):
- $(Q)$(MAKE) -C $@
+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
- $(Q)$(MAKE) -C $(@:install-%=%) install-tests
+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
- $(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null
+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- unsigned long loops;
- unsigned long size;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned long i, len = a->loops;
- unsigned char *src, *dst;
- src = malloc(a->size * 1024);
- dst = malloc(a->size * 1024);
- if ((!src) || (!dst)) {
printf("ERR: Can't allocate memory\n");exit(1);- }
- for (i = 0; i < len; i++)
memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned long i, len, size, thr;
- pthread_t threads[256];
- struct args args[256];
- long long v;
- if (argc < 4) {
printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);- }
- v = atoll(argv[1]);
- if ((v < 1) || (v > (1024 * 1024))) {
printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);- }
- size = v;
- thr = atol(argv[2]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- v = atoll(argv[3]);
- if ((v < 1) || (v > 40000000000ll)) {
printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);- }
- len = v * 100;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
- unsigned int loops;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- int i = 0, len = a->loops;
- if (getenv("SHOW_TID")) {
unsigned long long tid = gettid();printf("%llu\n", tid);- }
- asm volatile(
"loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */- );
- return (void *)(long)i;
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, len, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 3) {
printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- len = atoi(argv[2]);
- if ((len < 1) || (len > 4000)) {
printf("ERR: max loops 4000 (millions)\n");exit(1);- }
- len *= 1000000;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- pthread_t th;
- unsigned int in, out;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned int i, in = a->in;
- for (i = 0; i < 10000; i++) {
asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);- }
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 2) {
printf("ERR: %s [numthreads]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr > 256) || (thr < 1)) {
printf("ERR: threads 1-256\n");exit(1);- }
- for (i = 0; i < thr; i++) {
args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
- echo "Not running as root... skip"
- exit 2
+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
- DATD="$PERF_TEST_CORESIGHT_DATADIR";
+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
- STATD="$PERF_TEST_CORESIGHT_STATDIR";
+fi
+# Called if the test fails - error code 2 +err() {
- echo "$1"
- exit 1
+}
+# Check that some statistics from our perf +check_val_min() {
- STATF="$4"
- if test "$2" -lt "$3"; then
echo ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"- fi
+}
+perf_dump_aux_verify() {
- # Some basic checking that the AUX chunk contains some sensible data
- # to see that we are recording something and at least a minimum
- # amount of it. We should almost always see F3 atoms in just about
- # anything but certainly we will see some trace info and async atom
- # chunks.
- DUMP="$DATD/perf-tmp-aux-dump.txt"
- perf report --stdio --dump -i "$1" | \
grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"- # Simply count how many of these atoms we find to see that we are
- # producing a reasonable amount of data - exact checks are not sane
- # as this is a lossy process where we may lose some blocks and the
- # compiler may produce different code depending on the compiler and
- # optimization options, so this is rough just to see if we're
- # either missing almost all the data or all of it
- ATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`
- ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`
- ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`
- rm -f "$DUMP"
- # Arguments provide minimums for a pass
- CHECK_F3_MIN="$2"
- CHECK_ASYNC_MIN="$3"
- CHECK_TRACE_INFO_MIN="$4"
- # Write out statistics, so over time you can track results to see if
- # there is a pattern - for example we have less "noisy" results that
- # produce more consistent amounts of data each run, to see if over
- # time any techinques to minimize data loss are having an effect or
- # not
- STATF="$STATD/stats-$TEST-$DATV.csv"
- if ! test -f "$STATF"; then
echo "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"- fi
- echo -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"
- # Actually check to see if we passed or failed.
- check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"
- check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"
- check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"
- echo ", Ok" >> "$STATF"
+}
+perf_dump_aux_tid_verify() {
- # Specifically crafted test will produce a list of Tread ID's to
- # stdout that need to be checked to see that they have had trace
- # info collected in AUX blocks in the perf data. This will go
- # through all the TID's that are listed as CID=0xabcdef and see
- # that all the Thread IDs the test tool reports are in the perf
- # data AUX chunks
- # The TID test tools will print a TID per stdout line that are being
- # tested
- TIDS=`cat "$2"`
- # Scan the perf report to find the TIDs that are actually CID in hex
- # and build a list of the ones found
- FOUND_TIDS=`perf report --stdio --dump -i "$1" | \
grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`- # Iterate over the list of TIDs that the test says it has and find
- # them in the TIDs found in the perf report
- MISSING=""
- for TID2 in $TIDS; do
FOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fi- done
- if test -n "$MISSING"; then
err "Thread IDs $MISSING not found in perf AUX data"- fi
+}
2.32.0
On Mon, 30 May 2022 at 10:27, Mathieu Poirier mathieu.poirier@linaro.org wrote:
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++ 24 files changed, 823 insertions(+), 4 deletions(-)
As Leo pointed out this is a big patch and hard to digest intellectually.
create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.*
diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt
I think it would be best to keep all the coresight documentation under the current coresight documentation repository[1]. That way all the information on coresight can be found in a central place.
Some part of what is added by this patch is redundant with what is currently available in [1]. Other parts are tests specific and should be added under something like "coresight-perf-test.rst".
Thanks, Mathieu
[1]. Documentation/trace/coresight/
I forgot... Please add a proper cover letter for this patchset.
@@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream. Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
+Be aware that amny of these tests take quite a while to run, specifically +in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@
-all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
$(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)+tests-coresight-targets-clean:
$(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets
# Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
$(Q)$(MAKE) -C tests/shell/coresight install-testsinstall-bin: install-tools install-tests install-traceevent-plugins
@@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS)
-clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS)
endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
asm_pure_loop \thread_loop \memcpy_thread \unroll_loop_thread+all: $(SUBDIRS) +$(SUBDIRS):
$(Q)$(MAKE) -C $@+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
$(Q)$(MAKE) -C $(@:install-%=%) install-tests+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
$(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
mov x0, 0x0000ffffmov x1, xzr+loop:
nopnopcbnz x1, noskipnopnopadrp x2, skipadd x2, x2, :lo12:skipbr x2nopnop+noskip:
nopnop+skip:
sub x0, x0, 1cbnz x0, loopmov x0, #0mov x8, #93 // __NR_exit syscallsvc #0diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
unsigned long loops;unsigned long size;pthread_t th;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;unsigned long i, len = a->loops;unsigned char *src, *dst;src = malloc(a->size * 1024);dst = malloc(a->size * 1024);if ((!src) || (!dst)) {printf("ERR: Can't allocate memory\n");exit(1);}for (i = 0; i < len; i++)memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned long i, len, size, thr;pthread_t threads[256];struct args args[256];long long v;if (argc < 4) {printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);}v = atoll(argv[1]);if ((v < 1) || (v > (1024 * 1024))) {printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);}size = v;thr = atol(argv[2]);if ((thr < 1) || (thr > 256)) {printf("ERR: threads 1-256\n");exit(1);}v = atoll(argv[3]);if ((v < 1) || (v > 40000000000ll)) {printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);}len = v * 100;for (i = 0; i < thr; i++) {args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
unsigned int loops;pthread_t th;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;int i = 0, len = a->loops;if (getenv("SHOW_TID")) {unsigned long long tid = gettid();printf("%llu\n", tid);}asm volatile("loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */);return (void *)(long)i;+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned int i, len, thr;pthread_t threads[256];struct args args[256];if (argc < 3) {printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);}thr = atoi(argv[1]);if ((thr < 1) || (thr > 256)) {printf("ERR: threads 1-256\n");exit(1);}len = atoi(argv[2]);if ((len < 1) || (len > 4000)) {printf("ERR: max loops 4000 (millions)\n");exit(1);}len *= 1000000;for (i = 0; i < thr; i++) {args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
pthread_t th;unsigned int in, out;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;unsigned int i, in = a->in;for (i = 0; i < 10000; i++) {asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);}+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned int i, thr;pthread_t threads[256];struct args args[256];if (argc < 2) {printf("ERR: %s [numthreads]\n", argv[0]);exit(1);}thr = atoi(argv[1]);if ((thr > 256) || (thr < 1)) {printf("ERR: threads 1-256\n");exit(1);}for (i = 0; i < thr; i++) {args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
echo "Not running as root... skip"exit 2+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
DATD="$PERF_TEST_CORESIGHT_DATADIR";+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
STATD="$PERF_TEST_CORESIGHT_STATDIR";+fi
+# Called if the test fails - error code 2 +err() {
echo "$1"exit 1+}
+# Check that some statistics from our perf +check_val_min() {
STATF="$4"if test "$2" -lt "$3"; thenecho ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"fi+}
+perf_dump_aux_verify() {
# Some basic checking that the AUX chunk contains some sensible data# to see that we are recording something and at least a minimum# amount of it. We should almost always see F3 atoms in just about# anything but certainly we will see some trace info and async atom# chunks.DUMP="$DATD/perf-tmp-aux-dump.txt"perf report --stdio --dump -i "$1" | \grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"# Simply count how many of these atoms we find to see that we are# producing a reasonable amount of data - exact checks are not sane# as this is a lossy process where we may lose some blocks and the# compiler may produce different code depending on the compiler and# optimization options, so this is rough just to see if we're# either missing almost all the data or all of itATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`rm -f "$DUMP"# Arguments provide minimums for a passCHECK_F3_MIN="$2"CHECK_ASYNC_MIN="$3"CHECK_TRACE_INFO_MIN="$4"# Write out statistics, so over time you can track results to see if# there is a pattern - for example we have less "noisy" results that# produce more consistent amounts of data each run, to see if over# time any techinques to minimize data loss are having an effect or# notSTATF="$STATD/stats-$TEST-$DATV.csv"if ! test -f "$STATF"; thenecho "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"fiecho -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"# Actually check to see if we passed or failed.check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"echo ", Ok" >> "$STATF"+}
+perf_dump_aux_tid_verify() {
# Specifically crafted test will produce a list of Tread ID's to# stdout that need to be checked to see that they have had trace# info collected in AUX blocks in the perf data. This will go# through all the TID's that are listed as CID=0xabcdef and see# that all the Thread IDs the test tool reports are in the perf# data AUX chunks# The TID test tools will print a TID per stdout line that are being# testedTIDS=`cat "$2"`# Scan the perf report to find the TIDs that are actually CID in hex# and build a list of the ones foundFOUND_TIDS=`perf report --stdio --dump -i "$1" | \grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`# Iterate over the list of TIDs that the test says it has and find# them in the TIDs found in the perf reportMISSING=""for TID2 in $TIDS; doFOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fidoneif test -n "$MISSING"; thenerr "Thread IDs $MISSING not found in perf AUX data"fi+}
2.32.0
On 5/30/22 17:47, Mathieu Poirier wrote:
On Mon, 30 May 2022 at 10:27, Mathieu Poirier mathieu.poirier@linaro.org wrote:
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++ 24 files changed, 823 insertions(+), 4 deletions(-)
As Leo pointed out this is a big patch and hard to digest intellectually.
create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.*
diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt
I think it would be best to keep all the coresight documentation under the current coresight documentation repository[1]. That way all the information on coresight can be found in a central place.
Some part of what is added by this patch is redundant with what is currently available in [1]. Other parts are tests specific and should be added under something like "coresight-perf-test.rst".
Thanks, Mathieu
[1]. Documentation/trace/coresight/
I forgot... Please add a proper cover letter for this patchset.
ok - sure. next round.
@@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream. Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
+Be aware that amny of these tests take quite a while to run, specifically +in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@
-all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
$(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)+tests-coresight-targets-clean:
$(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets
# Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
$(Q)$(MAKE) -C tests/shell/coresight install-testsinstall-bin: install-tools install-tests install-traceevent-plugins
@@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS)
-clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS)
endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
asm_pure_loop \thread_loop \memcpy_thread \unroll_loop_thread+all: $(SUBDIRS) +$(SUBDIRS):
$(Q)$(MAKE) -C $@+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
$(Q)$(MAKE) -C $(@:install-%=%) install-tests+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
$(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
mov x0, 0x0000ffffmov x1, xzr+loop:
nopnopcbnz x1, noskipnopnopadrp x2, skipadd x2, x2, :lo12:skipbr x2nopnop+noskip:
nopnop+skip:
sub x0, x0, 1cbnz x0, loopmov x0, #0mov x8, #93 // __NR_exit syscallsvc #0diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
unsigned long loops;unsigned long size;pthread_t th;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;unsigned long i, len = a->loops;unsigned char *src, *dst;src = malloc(a->size * 1024);dst = malloc(a->size * 1024);if ((!src) || (!dst)) {printf("ERR: Can't allocate memory\n");exit(1);}for (i = 0; i < len; i++)memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned long i, len, size, thr;pthread_t threads[256];struct args args[256];long long v;if (argc < 4) {printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);}v = atoll(argv[1]);if ((v < 1) || (v > (1024 * 1024))) {printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);}size = v;thr = atol(argv[2]);if ((thr < 1) || (thr > 256)) {printf("ERR: threads 1-256\n");exit(1);}v = atoll(argv[3]);if ((v < 1) || (v > 40000000000ll)) {printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);}len = v * 100;for (i = 0; i < thr; i++) {args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
unsigned int loops;pthread_t th;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;int i = 0, len = a->loops;if (getenv("SHOW_TID")) {unsigned long long tid = gettid();printf("%llu\n", tid);}asm volatile("loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */);return (void *)(long)i;+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned int i, len, thr;pthread_t threads[256];struct args args[256];if (argc < 3) {printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);}thr = atoi(argv[1]);if ((thr < 1) || (thr > 256)) {printf("ERR: threads 1-256\n");exit(1);}len = atoi(argv[2]);if ((len < 1) || (len > 4000)) {printf("ERR: max loops 4000 (millions)\n");exit(1);}len *= 1000000;for (i = 0; i < thr; i++) {args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
$(call QUIET_INSTALL, tests) \$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
$(Q)$(RM) -f $(BIN)+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
pthread_t th;unsigned int in, out;void *ret;+};
+static void *thrfn(void *arg) +{
struct args *a = arg;unsigned int i, in = a->in;for (i = 0; i < 10000; i++) {asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);}+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
pthread_t t;pthread_attr_t attr;pthread_attr_init(&attr);pthread_create(&t, &attr, fn, arg);return t;+}
+int main(int argc, char **argv) +{
unsigned int i, thr;pthread_t threads[256];struct args args[256];if (argc < 2) {printf("ERR: %s [numthreads]\n", argv[0]);exit(1);}thr = atoi(argv[1]);if ((thr > 256) || (thr < 1)) {printf("ERR: threads 1-256\n");exit(1);}for (i = 0; i < thr; i++) {args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));}for (i = 0; i < thr; i++)pthread_join(args[i].th, &(args[i].ret));return 0;+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
echo "Not running as root... skip"exit 2+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
DATD="$PERF_TEST_CORESIGHT_DATADIR";+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
STATD="$PERF_TEST_CORESIGHT_STATDIR";+fi
+# Called if the test fails - error code 2 +err() {
echo "$1"exit 1+}
+# Check that some statistics from our perf +check_val_min() {
STATF="$4"if test "$2" -lt "$3"; thenecho ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"fi+}
+perf_dump_aux_verify() {
# Some basic checking that the AUX chunk contains some sensible data# to see that we are recording something and at least a minimum# amount of it. We should almost always see F3 atoms in just about# anything but certainly we will see some trace info and async atom# chunks.DUMP="$DATD/perf-tmp-aux-dump.txt"perf report --stdio --dump -i "$1" | \grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"# Simply count how many of these atoms we find to see that we are# producing a reasonable amount of data - exact checks are not sane# as this is a lossy process where we may lose some blocks and the# compiler may produce different code depending on the compiler and# optimization options, so this is rough just to see if we're# either missing almost all the data or all of itATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`rm -f "$DUMP"# Arguments provide minimums for a passCHECK_F3_MIN="$2"CHECK_ASYNC_MIN="$3"CHECK_TRACE_INFO_MIN="$4"# Write out statistics, so over time you can track results to see if# there is a pattern - for example we have less "noisy" results that# produce more consistent amounts of data each run, to see if over# time any techinques to minimize data loss are having an effect or# notSTATF="$STATD/stats-$TEST-$DATV.csv"if ! test -f "$STATF"; thenecho "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"fiecho -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"# Actually check to see if we passed or failed.check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"echo ", Ok" >> "$STATF"+}
+perf_dump_aux_tid_verify() {
# Specifically crafted test will produce a list of Tread ID's to# stdout that need to be checked to see that they have had trace# info collected in AUX blocks in the perf data. This will go# through all the TID's that are listed as CID=0xabcdef and see# that all the Thread IDs the test tool reports are in the perf# data AUX chunks# The TID test tools will print a TID per stdout line that are being# testedTIDS=`cat "$2"`# Scan the perf report to find the TIDs that are actually CID in hex# and build a list of the ones foundFOUND_TIDS=`perf report --stdio --dump -i "$1" | \grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`# Iterate over the list of TIDs that the test says it has and find# them in the TIDs found in the perf reportMISSING=""for TID2 in $TIDS; doFOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fidoneif test -n "$MISSING"; thenerr "Thread IDs $MISSING not found in perf AUX data"fi+}
2.32.0
On 5/30/22 17:27, Mathieu Poirier wrote:
On Wed, Mar 09, 2022 at 12:28:59PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
This adds a test harness and tests to run perf record and examine the resuling output when coresight is enabled on arm64 and check the resulting quality of the output as part of perf test. These tests use various tools to produce output from perf record then measure some key specific aspects of that data to see if the data exists at all and contains key aspects such as measuring some data for every thread of a test or produces sufficient data for large exeuction runs of a large executable. etc.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
MAINTAINERS | 4 + tools/perf/.gitignore | 6 +- tools/perf/Documentation/arm-coresight.txt | 140 ++++++++++++++++++ tools/perf/Makefile.perf | 14 +- tools/perf/tests/shell/coresight/Makefile | 30 ++++ .../tests/shell/coresight/Makefile.miniconfig | 23 +++ .../shell/coresight/asm_pure_loop/.gitignore | 1 + .../shell/coresight/asm_pure_loop/Makefile | 30 ++++ .../coresight/asm_pure_loop/asm_pure_loop.S | 28 ++++ .../shell/coresight/memcpy_thread/.gitignore | 1 + .../shell/coresight/memcpy_thread/Makefile | 29 ++++ .../coresight/memcpy_thread/memcpy_thread.c | 79 ++++++++++ .../shell/coresight/thread_loop/.gitignore | 1 + .../shell/coresight/thread_loop/Makefile | 29 ++++ .../shell/coresight/thread_loop/thread_loop.c | 86 +++++++++++ .../coresight/unroll_loop_thread/.gitignore | 1 + .../coresight/unroll_loop_thread/Makefile | 29 ++++ .../unroll_loop_thread/unroll_loop_thread.c | 74 +++++++++ .../tests/shell/coresight_asm_pure_loop.sh | 18 +++ .../shell/coresight_memcpy_thread_16k_10.sh | 18 +++ .../coresight_thread_loop_check_tid_10.sh | 19 +++ .../coresight_thread_loop_check_tid_2.sh | 19 +++ .../shell/coresight_unroll_loop_thread_10.sh | 18 +++ tools/perf/tests/shell/lib/coresight.sh | 130 ++++++++++++++++ 24 files changed, 823 insertions(+), 4 deletions(-)
As Leo pointed out this is a big patch and hard to digest intellectually.
create mode 100644 tools/perf/Documentation/arm-coresight.txt create mode 100644 tools/perf/tests/shell/coresight/Makefile create mode 100644 tools/perf/tests/shell/coresight/Makefile.miniconfig create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c create mode 100644 tools/perf/tests/shell/coresight/thread_loop/.gitignore create mode 100644 tools/perf/tests/shell/coresight/thread_loop/Makefile create mode 100644 tools/perf/tests/shell/coresight/thread_loop/thread_loop.c create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile create mode 100644 tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c create mode 100755 tools/perf/tests/shell/coresight_asm_pure_loop.sh create mode 100755 tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh create mode 100755 tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh create mode 100755 tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh create mode 100644 tools/perf/tests/shell/lib/coresight.sh
diff --git a/MAINTAINERS b/MAINTAINERS index 673c7124ca82..18cc20609f2e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1918,10 +1918,14 @@ F: drivers/hwtracing/coresight/* F: include/dt-bindings/arm/coresight-cti-dt.h F: include/linux/coresight* F: samples/coresight/* +F: tools/perf/Documentation/arm-coresight.txt F: tools/perf/arch/arm/util/auxtrace.c F: tools/perf/arch/arm/util/cs-etm.c F: tools/perf/arch/arm/util/cs-etm.h F: tools/perf/arch/arm/util/pmu.c +F: tools/perf/tests/shell/coresight_* +F: tools/perf/tests/shell/tools/Makefile +F: tools/perf/tests/shell/tools/coresight/* F: tools/perf/util/cs-etm-decoder/* F: tools/perf/util/cs-etm.* diff --git a/tools/perf/.gitignore b/tools/perf/.gitignore index 20b8ab984d5f..138c679ecacd 100644 --- a/tools/perf/.gitignore +++ b/tools/perf/.gitignore @@ -15,8 +15,9 @@ perf*.1 perf*.xml perf*.html common-cmds.h -perf.data -perf.data.old +perf*.data +perf*.data.old +stats-*.csv output.svg perf-archive perf-with-kcore @@ -30,6 +31,7 @@ config.mak.autogen *-flex.* *.pyc *.pyo +*.stdout .config-detected util/intel-pt-decoder/inat-tables.c arch/*/include/generated/ diff --git a/tools/perf/Documentation/arm-coresight.txt b/tools/perf/Documentation/arm-coresight.txt new file mode 100644 index 000000000000..3a9e6c573c58 --- /dev/null +++ b/tools/perf/Documentation/arm-coresight.txt
I think it would be best to keep all the coresight documentation under the current coresight documentation repository[1]. That way all the information on coresight can be found in a central place.
OK. I just added this here because this was relevant "get you going" documentation for perf + coresight and testing and other relevant docs were here including architecture specific ones like the intel docs. I'll find a way of merging it with the above mentioned docs.
Some part of what is added by this patch is redundant with what is currently available in [1]. Other parts are tests specific and should be added under something like "coresight-perf-test.rst".
Sure - some is, but most is not mentioned there (the vast majority). I'd argue essentially all of it is relevant to someone getting started and the documentation in the above location is not very friendly or useful in that regard (which is why I wrote the documentation here to begin with). As I can't really find anything it duplicates other than the initial paragraph which I can remove, I'll merge the rest in and link it in as above, unless you can be specific on what it really duplicates?
Thanks, Mathieu
[1]. Documentation/trace/coresight/
@@ -0,0 +1,140 @@ +Arm Coresight Support +=====================
+Coresight is a feature of some Arm based processors that allows for +debugging. One of the things it can do is trace every instruction +executed and remotely expose that information in a hardware compressed +stream. Perf is able to locally access that stream and store it to the +output perf data files. This stream can then be later decoded to give the +instructions that were traced for debugging or profiling purposes. You +can log such data with a perf record command like:
- perf record -e cs_etm//u testbinary
+This would run some test binary (testbinary) until it exits and record +a perf.data trace file. That file would have AUX sections if coresight +is working correctly. You can dump the content of this file as +readable text with a command like:
- perf report --stdio --dump -i perf.data
+You should find some sections of this file have AUX data blocks like:
- 0x1e78 [0x30]: PERF_RECORD_AUXTRACE size: 0x11dd0 offset: 0 ref: 0x1b614fc1061b0ad1 idx: 0 tid: 531230 cpu: -1
- . ... CoreSight ETM Trace data: size 73168 bytes
Idx:0; ID:10; I_ASYNC : Alignment Synchronisation.Idx:12; ID:10; I_TRACE_INFO : Trace Info.; INFO=0x0 { CC.0 }Idx:17; ID:10; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; Addr=0x0000000000000000;Idx:26; ID:10; I_TRACE_ON : Trace On.Idx:27; ID:10; I_ADDR_CTXT_L_64IS0 : Address & Context, Long, 64 bit, IS0.; Addr=0x0000FFFFB6069140; Ctxt: AArch64,EL0, NS;Idx:38; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:39; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:40; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEEEEEEEEEEEEEEIdx:41; ID:10; I_ATOM_F6 : Atom format 6.; EEEEEEEEEEEN...+If you see these above, then your system is tracing coresight data +correctly.
+To compile perf with coresight support in the perf directory do
- make CORESIGHT=1
+This will compile the perf tool with coresight support as well as +build some small test binaries for perf test. This requires you also +be compiling for 64bit Arm (ARM64/aarch64). The tools run as part of +perf coresight tracing are in tests/shell/tools/coresight.
+You will also want coresight support enabled in your kernel config. +Ensure it is enabled with:
- CONFIG_CORESIGHT=y
+There are various other coresight options you probably also want +enabled like:
- CONFIG_CORESIGHT_LINKS_AND_SINKS=y
- CONFIG_CORESIGHT_LINK_AND_SINK_TMC=y
- CONFIG_CORESIGHT_CATU=y
- CONFIG_CORESIGHT_SINK_TPIU=y
- CONFIG_CORESIGHT_SINK_ETBV10=y
- CONFIG_CORESIGHT_SOURCE_ETM4X=y
- CONFIG_CORESIGHT_STM=y
- CONFIG_CORESIGHT_CPU_DEBUG=y
- CONFIG_CORESIGHT_CTI=y
- CONFIG_CORESIGHT_CTI_INTEGRATION_REGS=y
+Please refer to the kernel configuration help for more information.
+Perf test - Verify kernel and userspace perf coresight work +===========================================================
+When you run perf test, it will do a lot of self tests. Some of those +tests will cover Coresight (only if enabled and on ARM64). You +generally would run perf test from the tools/perf directory in the +kernel tree. Some tests will check some internal perf support like:
- Check Arm CoreSight trace data recording and synthesized samples
+Some others will actually use perf record and some test binaries that +are in tests/shell/tools/coresight and will collect traces to ensure a +minimum level of functionality is met. The scripts that launch these +tests are in tests/shell. These will all look like:
- Coresight / Memcpy 1M 25 Threads
- Coresight / Unroll Loop Thread 2
- ...
+These perf record tests will not run if the tool binaries do not exist +in tests/shell/tools/coresight/*/ and will be skipped. If you do not +have coresight support in hardware then either do not build perf with +coresight support or remove these binaries in order to not have these +tests fail and have them skip instead.
+These tests will log historical results in the current working +directory (e.g. tools/perf) and will be named stats-*.csv like:
- stats-asm_pure_loop-out.csv
- stats-bubble_sort-random.csv
- ...
+These statistic files log some aspects of the AUX data sections in +the perf data output counting some numbers of certain encodings (a +good way to know that it's working in a very simple way). One problem +with coresight is that given a large enough amount of data needing to +be logged, some of it can be lost due to the processor not waking up +in time to read out all the data from buffers etc.. You will notice +that the amount of data collected can vary a lot per run of perf test. +If you wish to see how this changes over time, simply run perf test +multiple times and all these csv files will have more and more data +appended to it that you can later examine, graph and otherwise use to +figure out if things have become worse or better.
+Be aware that amny of these tests take quite a while to run, specifically +in processing the perf data file and dumping contents to then examine what +is inside.
+You can change where these csv logs are stored by setting the +PERF_TEST_CORESIGHT_STATDIR environment variable before running perf +test like:
- export PERF_TEST_CORESIGHT_STATDIR=/var/tmp
- perf test
+They will also store resulting perf output data in the current +directory for later inspection like:
- perf-memcpy-1m.data
- perf-thread_loop-2th.data
- ...
+You can alter where the perf data files are stored by setting the +PERF_TEST_CORESIGHT_DATADIR environment variable such as:
- PERF_TEST_CORESIGHT_DATADIR=/var/tmp
- perf test
+You may wish to set these above environment variables if you which to +keep the output of tests outside of the current working directory for +longer term storage and examination. diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf index ac861e42c8f7..b97db83992e0 100644 --- a/tools/perf/Makefile.perf +++ b/tools/perf/Makefile.perf @@ -630,7 +630,15 @@ sync_file_range_tbls := $(srctree)/tools/perf/trace/beauty/sync_file_range.sh $(sync_file_range_arrays): $(linux_uapi_dir)/fs.h $(sync_file_range_tbls) $(Q)$(SHELL) '$(sync_file_range_tbls)' $(linux_uapi_dir) > $@ -all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) +TESTS_CORESIGHT_DIR := $(srctree)/tools/perf/tests/shell/coresight
+tests-coresight-targets: FORCE
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR)
+tests-coresight-targets-clean:
- $(Q)$(MAKE) -C $(TESTS_CORESIGHT_DIR) clean
+all: shell_compatibility_test $(ALL_PROGRAMS) $(LANG_BINDINGS) $(OTHER_PROGRAMS) tests-coresight-targets # Create python binding output directory if not already present _dummy := $(shell [ -d '$(OUTPUT)python' ] || mkdir -p '$(OUTPUT)python') @@ -1020,6 +1028,7 @@ install-tests: all install-gtk $(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \ $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \ $(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
- $(Q)$(MAKE) -C tests/shell/coresight install-tests
install-bin: install-tools install-tests install-traceevent-plugins @@ -1088,7 +1097,7 @@ endif # BUILD_BPF_SKEL bpf-skel-clean: $(call QUIET_CLEAN, bpf-skel) $(RM) -r $(SKEL_TMP_OUT) $(SKELETONS) -clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean +clean:: $(LIBTRACEEVENT)-clean $(LIBAPI)-clean $(LIBBPF)-clean $(LIBSUBCMD)-clean $(LIBPERF)-clean fixdep-clean python-clean bpf-skel-clean tests-coresight-targets-clean $(call QUIET_CLEAN, core-objs) $(RM) $(LIBPERF_A) $(OUTPUT)perf-archive $(OUTPUT)perf-with-kcore $(OUTPUT)perf-iostat $(LANG_BINDINGS) $(Q)find $(if $(OUTPUT),$(OUTPUT),.) -name '*.o' -delete -o -name '.*.cmd' -delete -o -name '.*.d' -delete $(Q)$(RM) $(OUTPUT).config-detected @@ -1155,5 +1164,6 @@ FORCE: .PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell .PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE prepare .PHONY: libtraceevent_plugins archheaders +.PHONY: $(TESTS_CORESIGHT_TARGETS) endif # force_fixdep diff --git a/tools/perf/tests/shell/coresight/Makefile b/tools/perf/tests/shell/coresight/Makefile new file mode 100644 index 000000000000..dda99aeac158 --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../../../../../tools/scripts/Makefile.include +include ../../../../../tools/scripts/Makefile.arch +include ../../../../../tools/scripts/utilities.mak
+SUBDIRS = \
- asm_pure_loop \
- thread_loop \
- memcpy_thread \
- unroll_loop_thread
+all: $(SUBDIRS) +$(SUBDIRS):
- $(Q)$(MAKE) -C $@
+INSTALLDIRS = $(SUBDIRS:%=install-%)
+install-tests: $(INSTALLDIRS) +$(INSTALLDIRS):
- $(Q)$(MAKE) -C $(@:install-%=%) install-tests
+CLEANDIRS = $(SUBDIRS:%=clean-%)
+clean: $(CLEANDIRS) +$(CLEANDIRS):
- $(Q)$(MAKE) -C $(@:clean-%=%) clean >/dev/null
+.PHONY: all clean $(SUBDIRS) $(CLEANDIRS) $(INSTALLDIRS)
diff --git a/tools/perf/tests/shell/coresight/Makefile.miniconfig b/tools/perf/tests/shell/coresight/Makefile.miniconfig new file mode 100644 index 000000000000..893c12685fed --- /dev/null +++ b/tools/perf/tests/shell/coresight/Makefile.miniconfig @@ -0,0 +1,23 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+ifndef DESTDIR +prefix ?= $(HOME) +endif
+DESTDIR_SQ = $(subst ',''',$(DESTDIR)) +perfexecdir = libexec/perf-core +perfexec_instdir = $(perfexecdir)
+ifneq ($(filter /%,$(firstword $(perfexecdir))),) +perfexec_instdir = $(perfexecdir) +else +perfexec_instdir = $(prefix)/$(perfexecdir) +endif
+perfexec_instdir_SQ = $(subst ',''',$(perfexec_instdir)) +INSTALL = install
+include ../../../../../scripts/Makefile.include +include ../../../../../scripts/Makefile.arch +include ../../../../../scripts/utilities.mak diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore new file mode 100644 index 000000000000..468673ac32e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/.gitignore @@ -0,0 +1 @@ +asm_pure_loop diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile new file mode 100644 index 000000000000..10c5a60cb71c --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/Makefile @@ -0,0 +1,30 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+include ../Makefile.miniconfig
+BIN=asm_pure_loop +LIB=
+all: $(BIN)
+$(BIN): $(BIN).S +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).S -nostdlib -static -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S new file mode 100644 index 000000000000..75cf084a927d --- /dev/null +++ b/tools/perf/tests/shell/coresight/asm_pure_loop/asm_pure_loop.S @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* Tamas Zsoldos tamas.zsoldos@arm.com, 2021 */
+.globl _start +_start:
- mov x0, 0x0000ffff
- mov x1, xzr
+loop:
- nop
- nop
- cbnz x1, noskip
- nop
- nop
- adrp x2, skip
- add x2, x2, :lo12:skip
- br x2
- nop
- nop
+noskip:
- nop
- nop
+skip:
- sub x0, x0, 1
- cbnz x0, loop
- mov x0, #0
- mov x8, #93 // __NR_exit syscall
- svc #0
diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore new file mode 100644 index 000000000000..f8217e56091e --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/.gitignore @@ -0,0 +1 @@ +memcpy_thread diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/Makefile b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile new file mode 100644 index 000000000000..e2604cfae74b --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=memcpy_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c new file mode 100644 index 000000000000..a7e169d1bf64 --- /dev/null +++ b/tools/perf/tests/shell/coresight/memcpy_thread/memcpy_thread.c @@ -0,0 +1,79 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- unsigned long loops;
- unsigned long size;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned long i, len = a->loops;
- unsigned char *src, *dst;
- src = malloc(a->size * 1024);
- dst = malloc(a->size * 1024);
- if ((!src) || (!dst)) {
printf("ERR: Can't allocate memory\n");exit(1);- }
- for (i = 0; i < len; i++)
memcpy(dst, src, a->size * 1024);+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned long i, len, size, thr;
- pthread_t threads[256];
- struct args args[256];
- long long v;
- if (argc < 4) {
printf("ERR: %s [copysize Kb] [numthreads] [numloops (hundreds)]\n", argv[0]);exit(1);- }
- v = atoll(argv[1]);
- if ((v < 1) || (v > (1024 * 1024))) {
printf("ERR: max memory 1GB (1048576 KB)\n");exit(1);- }
- size = v;
- thr = atol(argv[2]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- v = atoll(argv[3]);
- if ((v < 1) || (v > 40000000000ll)) {
printf("ERR: loops 1-40000000000 (hundreds)\n");exit(1);- }
- len = v * 100;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].size = size;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/thread_loop/.gitignore b/tools/perf/tests/shell/coresight/thread_loop/.gitignore new file mode 100644 index 000000000000..6d4c33eaa9e8 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/.gitignore @@ -0,0 +1 @@ +thread_loop diff --git a/tools/perf/tests/shell/coresight/thread_loop/Makefile b/tools/perf/tests/shell/coresight/thread_loop/Makefile new file mode 100644 index 000000000000..424df4e8b0e6 --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=thread_loop +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c new file mode 100644 index 000000000000..c0158fac7d0b --- /dev/null +++ b/tools/perf/tests/shell/coresight/thread_loop/thread_loop.c @@ -0,0 +1,86 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021
+// define this for gettid() +#define _GNU_SOURCE
+#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h> +#include <sys/syscall.h> +#ifndef SYS_gettid +// gettid is 178 on arm64 +# define SYS_gettid 178 +#endif +#define gettid() syscall(SYS_gettid)
+struct args {
- unsigned int loops;
- pthread_t th;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- int i = 0, len = a->loops;
- if (getenv("SHOW_TID")) {
unsigned long long tid = gettid();printf("%llu\n", tid);- }
- asm volatile(
"loop:\n""add %[i], %[i], #1\n""cmp %[i], %[len]\n""blt loop\n": /* out */: /* in */ [i] "r" (i), [len] "r" (len): /* clobber */- );
- return (void *)(long)i;
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, len, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 3) {
printf("ERR: %s [numthreads] [numloops (millions)]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr < 1) || (thr > 256)) {
printf("ERR: threads 1-256\n");exit(1);- }
- len = atoi(argv[2]);
- if ((len < 1) || (len > 4000)) {
printf("ERR: max loops 4000 (millions)\n");exit(1);- }
- len *= 1000000;
- for (i = 0; i < thr; i++) {
args[i].loops = len;args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore new file mode 100644 index 000000000000..2cb4e996dbf3 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/.gitignore @@ -0,0 +1 @@ +unroll_loop_thread diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile new file mode 100644 index 000000000000..45ab2be8be92 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/Makefile @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021 +include ../Makefile.miniconfig
+BIN=unroll_loop_thread +LIB=-pthread
+all: $(BIN)
+$(BIN): $(BIN).c +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(Q)$(CC) $(BIN).c -o $(BIN) $(LIB)
+endif +endif
+install-tests: all +ifdef CORESIGHT +ifeq ($(ARCH),arm64)
- $(call QUIET_INSTALL, tests) \
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)'; \$(INSTALL) $(BIN) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/tools/$(BIN)/$(BIN)'+endif +endif
+clean:
- $(Q)$(RM) -f $(BIN)
+.PHONY: all clean install-tests diff --git a/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c new file mode 100644 index 000000000000..cb9d22c7dfb9 --- /dev/null +++ b/tools/perf/tests/shell/coresight/unroll_loop_thread/unroll_loop_thread.c @@ -0,0 +1,74 @@ +// SPDX-License-Identifier: GPL-2.0 +// Carsten Haitzler carsten.haitzler@arm.com, 2021 +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <pthread.h>
+struct args {
- pthread_t th;
- unsigned int in, out;
- void *ret;
+};
+static void *thrfn(void *arg) +{
- struct args *a = arg;
- unsigned int i, in = a->in;
- for (i = 0; i < 10000; i++) {
asm volatile (+// force an unroll of thia add instruction so we can test long runs of code +#define SNIP1 "add %[in], %[in], #1\n" +// 10 +#define SNIP2 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 SNIP1 +// 100 +#define SNIP3 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 SNIP2 +// 1000 +#define SNIP4 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 SNIP3 +// 10000 +#define SNIP5 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 SNIP4 +// 100000
SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5 SNIP5: /* out */: /* in */ [in] "r" (in): /* clobber */);- }
+}
+static pthread_t new_thr(void *(*fn) (void *arg), void *arg) +{
- pthread_t t;
- pthread_attr_t attr;
- pthread_attr_init(&attr);
- pthread_create(&t, &attr, fn, arg);
- return t;
+}
+int main(int argc, char **argv) +{
- unsigned int i, thr;
- pthread_t threads[256];
- struct args args[256];
- if (argc < 2) {
printf("ERR: %s [numthreads]\n", argv[0]);exit(1);- }
- thr = atoi(argv[1]);
- if ((thr > 256) || (thr < 1)) {
printf("ERR: threads 1-256\n");exit(1);- }
- for (i = 0; i < thr; i++) {
args[i].in = rand();args[i].th = new_thr(thrfn, &(args[i]));- }
- for (i = 0; i < thr; i++)
pthread_join(args[i].th, &(args[i].ret));- return 0;
+} diff --git a/tools/perf/tests/shell/coresight_asm_pure_loop.sh b/tools/perf/tests/shell/coresight_asm_pure_loop.sh new file mode 100755 index 000000000000..3f0dbefcad50 --- /dev/null +++ b/tools/perf/tests/shell/coresight_asm_pure_loop.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / ASM Pure Loop
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="asm_pure_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="" +DATV="out" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh new file mode 100755 index 000000000000..8972af835016 --- /dev/null +++ b/tools/perf/tests/shell/coresight_memcpy_thread_16k_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Memcpy 16k 10 Threads
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="memcpy_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="16 10 1" +DATV="16k_10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh new file mode 100755 index 000000000000..5b468901f89b --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_10.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 10 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="10 1" +DATV="check-tid-10th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh new file mode 100755 index 000000000000..f8b7abd3aa03 --- /dev/null +++ b/tools/perf/tests/shell/coresight_thread_loop_check_tid_2.sh @@ -0,0 +1,19 @@ +#!/bin/sh -e +# Coresight / Thread Loop 2 Threads - Check TID
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="thread_loop" +. $(dirname $0)/lib/coresight.sh +ARGS="2 20" +DATV="check-tid-2th" +DATA="$DATD/perf-$TEST-$DATV.data" +STDO="$DATD/perf-$TEST-$DATV.stdout"
+SHOW_TID=1 perf record -s $PERFRECOPT -o "$DATA" "$BIN" $ARGS > $STDO
+perf_dump_aux_tid_verify "$DATA" "$STDO"
+err=$? +exit $err diff --git a/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh new file mode 100755 index 000000000000..c985dfb025c2 --- /dev/null +++ b/tools/perf/tests/shell/coresight_unroll_loop_thread_10.sh @@ -0,0 +1,18 @@ +#!/bin/sh -e +# Coresight / Unroll Loop Thread 10
+# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+TEST="unroll_loop_thread" +. $(dirname $0)/lib/coresight.sh +ARGS="10" +DATV="10" +DATA="$DATD/perf-$TEST-$DATV.data"
+perf record $PERFRECOPT -o "$DATA" "$BIN" $ARGS
+perf_dump_aux_verify "$DATA" 10 10 10
+err=$? +exit $err diff --git a/tools/perf/tests/shell/lib/coresight.sh b/tools/perf/tests/shell/lib/coresight.sh new file mode 100644 index 000000000000..6a611b073f02 --- /dev/null +++ b/tools/perf/tests/shell/lib/coresight.sh @@ -0,0 +1,130 @@ +# SPDX-License-Identifier: GPL-2.0 +# Carsten Haitzler carsten.haitzler@arm.com, 2021
+# This is sourced from a driver script so no need for #!/bin... etc. at the +# top - the assumption below is that it runs as part of sourcing after the +# test sets up some basic env vars to say what it is.
+# perf record options for the perf tests to use +PERFRECMEM="-m ,128M" +PERFRECOPT="$PERFRECMEM -e cs_etm//u"
+# These tests need to be run as root or coresight won't allow large buffers +# and will not collect proper data +UID=`id -u` +if test "$UID" -ne 0; then
- echo "Not running as root... skip"
- exit 2
+fi
+TOOLS=$(dirname $0) +DIR="$TOOLS/coresight/$TEST" +BIN="$DIR/$TEST" +# If the test tool/binary does not exist and is executable then skip the test +if ! test -x "$BIN"; then exit 2; fi +DATD="." +# If the data dir env is set then make the data dir use that instead of ./ +if test -n "$PERF_TEST_CORESIGHT_DATADIR"; then
- DATD="$PERF_TEST_CORESIGHT_DATADIR";
+fi +# If the stat dir env is set then make the data dir use that instead of ./ +STATD="." +if test -n "$PERF_TEST_CORESIGHT_STATDIR"; then
- STATD="$PERF_TEST_CORESIGHT_STATDIR";
+fi
+# Called if the test fails - error code 2 +err() {
- echo "$1"
- exit 1
+}
+# Check that some statistics from our perf +check_val_min() {
- STATF="$4"
- if test "$2" -lt "$3"; then
echo ", FAILED" >> "$STATF"err "Sanity check number of $1 is too low ($2 < $3)"- fi
+}
+perf_dump_aux_verify() {
- # Some basic checking that the AUX chunk contains some sensible data
- # to see that we are recording something and at least a minimum
- # amount of it. We should almost always see F3 atoms in just about
- # anything but certainly we will see some trace info and async atom
- # chunks.
- DUMP="$DATD/perf-tmp-aux-dump.txt"
- perf report --stdio --dump -i "$1" | \
grep -o -e I_ATOM_F3 -e I_ASYNC -e I_TRACE_INFO > "$DUMP"- # Simply count how many of these atoms we find to see that we are
- # producing a reasonable amount of data - exact checks are not sane
- # as this is a lossy process where we may lose some blocks and the
- # compiler may produce different code depending on the compiler and
- # optimization options, so this is rough just to see if we're
- # either missing almost all the data or all of it
- ATOM_F3_NUM=`grep I_ATOM_F3 "$DUMP" | wc -l`
- ATOM_ASYNC_NUM=`grep I_ASYNC "$DUMP" | wc -l`
- ATOM_TRACE_INFO_NUM=`grep I_TRACE_INFO "$DUMP" | wc -l`
- rm -f "$DUMP"
- # Arguments provide minimums for a pass
- CHECK_F3_MIN="$2"
- CHECK_ASYNC_MIN="$3"
- CHECK_TRACE_INFO_MIN="$4"
- # Write out statistics, so over time you can track results to see if
- # there is a pattern - for example we have less "noisy" results that
- # produce more consistent amounts of data each run, to see if over
- # time any techinques to minimize data loss are having an effect or
- # not
- STATF="$STATD/stats-$TEST-$DATV.csv"
- if ! test -f "$STATF"; then
echo "ATOM F3 Count, Minimum, ATOM ASYNC Count, Minimum, TRACE INFO Count, Minimum" > "$STATF"- fi
- echo -n "$ATOM_F3_NUM, $CHECK_F3_MIN, $ATOM_ASYNC_NUM, $CHECK_ASYNC_MIN, $ATOM_TRACE_INFO_NUM, $CHECK_TRACE_INFO_MIN" >> "$STATF"
- # Actually check to see if we passed or failed.
- check_val_min "ATOM_F3" "$ATOM_F3_NUM" "$CHECK_F3_MIN" "$STATF"
- check_val_min "ASYNC" "$ATOM_ASYNC_NUM" "$CHECK_ASYNC_MIN" "$STATF"
- check_val_min "TRACE_INFO" "$ATOM_TRACE_INFO_NUM" "$CHECK_TRACE_INFO_MIN" "$STATF"
- echo ", Ok" >> "$STATF"
+}
+perf_dump_aux_tid_verify() {
- # Specifically crafted test will produce a list of Tread ID's to
- # stdout that need to be checked to see that they have had trace
- # info collected in AUX blocks in the perf data. This will go
- # through all the TID's that are listed as CID=0xabcdef and see
- # that all the Thread IDs the test tool reports are in the perf
- # data AUX chunks
- # The TID test tools will print a TID per stdout line that are being
- # tested
- TIDS=`cat "$2"`
- # Scan the perf report to find the TIDs that are actually CID in hex
- # and build a list of the ones found
- FOUND_TIDS=`perf report --stdio --dump -i "$1" | \
grep -o "CID=0x[0-9a-z]\+" | sed 's/CID=//g' | \uniq | sort | uniq`- # Iterate over the list of TIDs that the test says it has and find
- # them in the TIDs found in the perf report
- MISSING=""
- for TID2 in $TIDS; do
FOUND=""for TIDHEX in $FOUND_TIDS; doTID=`printf "%i" $TIDHEX`if test "$TID" -eq "$TID2"; thenFOUND="y"breakfidoneif test -z "$FOUND"; thenMISSING="$MISSING $TID"fi- done
- if test -n "$MISSING"; then
err "Thread IDs $MISSING not found in perf AUX data"- fi
+}
2.32.0
On Wed, Mar 09, 2022 at 12:28:57PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
Perf test's shell runner will just run everything in the tests directory (as long as it's not another directory or does not begin with a dot), but sometimes you find files in there that are not shell scripts - perf.data output for example if you do some testing and then the next time you run perf test it tries to run these. Check the files are executable so they are actually intended to be test scripts and not just some "random junk" files there.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
Reviewed-by: Leo Yan leo.yan@linaro.org
tools/perf/tests/builtin-test.c | 4 +++- tools/perf/util/path.c | 14 +++++++++++++- tools/perf/util/path.h | 1 + 3 files changed, 17 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index fac3717d9ba1..3c34cb766724 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,7 +296,9 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && ent->d_name[0] != '.')
if (!is_directory(base, ent) && \is_executable_file(base, ent) && \ent->d_name[0] != '.')static const char *shell_tests__dir(char *path, size_t size) { diff --git a/tools/perf/util/path.c b/tools/perf/util/path.c index caed0336429f..ce80b79be103 100644 --- a/tools/perf/util/path.c +++ b/tools/perf/util/path.c @@ -86,9 +86,21 @@ bool is_directory(const char *base_path, const struct dirent *dent) char path[PATH_MAX]; struct stat st;
- sprintf(path, "%s/%s", base_path, dent->d_name);
- snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name); if (stat(path, &st)) return false;
return S_ISDIR(st.st_mode); }
+bool is_executable_file(const char *base_path, const struct dirent *dent) +{
- char path[PATH_MAX];
- struct stat st;
- snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name);
- if (stat(path, &st))
return false;- return !S_ISDIR(st.st_mode) && (st.st_mode & S_IXUSR);
+} diff --git a/tools/perf/util/path.h b/tools/perf/util/path.h index 083429b7efa3..d94902c22222 100644 --- a/tools/perf/util/path.h +++ b/tools/perf/util/path.h @@ -12,5 +12,6 @@ int path__join3(char *bf, size_t size, const char *path1, const char *path2, con bool is_regular_file(const char *file); bool is_directory(const char *base_path, const struct dirent *dent); +bool is_executable_file(const char *base_path, const struct dirent *dent);
#endif /* _PERF_PATH_H */
2.32.0
Em Sun, Apr 10, 2022 at 09:24:10AM +0800, Leo Yan escreveu:
On Wed, Mar 09, 2022 at 12:28:57PM +0000, carsten.haitzler@foss.arm.com wrote:
From: Carsten Haitzler carsten.haitzler@arm.com
Perf test's shell runner will just run everything in the tests directory (as long as it's not another directory or does not begin with a dot), but sometimes you find files in there that are not shell scripts - perf.data output for example if you do some testing and then the next time you run perf test it tries to run these. Check the files are executable so they are actually intended to be test scripts and not just some "random junk" files there.
Signed-off-by: Carsten Haitzler carsten.haitzler@arm.com
Reviewed-by: Leo Yan leo.yan@linaro.org
Thanks, applied.
- Arnaldo
tools/perf/tests/builtin-test.c | 4 +++- tools/perf/util/path.c | 14 +++++++++++++- tools/perf/util/path.h | 1 + 3 files changed, 17 insertions(+), 2 deletions(-)
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c index fac3717d9ba1..3c34cb766724 100644 --- a/tools/perf/tests/builtin-test.c +++ b/tools/perf/tests/builtin-test.c @@ -296,7 +296,9 @@ static const char *shell_test__description(char *description, size_t size, #define for_each_shell_test(entlist, nr, base, ent) \ for (int __i = 0; __i < nr && (ent = entlist[__i]); __i++) \
if (!is_directory(base, ent) && ent->d_name[0] != '.')
if (!is_directory(base, ent) && \is_executable_file(base, ent) && \ent->d_name[0] != '.')static const char *shell_tests__dir(char *path, size_t size) { diff --git a/tools/perf/util/path.c b/tools/perf/util/path.c index caed0336429f..ce80b79be103 100644 --- a/tools/perf/util/path.c +++ b/tools/perf/util/path.c @@ -86,9 +86,21 @@ bool is_directory(const char *base_path, const struct dirent *dent) char path[PATH_MAX]; struct stat st;
- sprintf(path, "%s/%s", base_path, dent->d_name);
- snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name); if (stat(path, &st)) return false;
return S_ISDIR(st.st_mode); }
+bool is_executable_file(const char *base_path, const struct dirent *dent) +{
- char path[PATH_MAX];
- struct stat st;
- snprintf(path, sizeof(path), "%s/%s", base_path, dent->d_name);
- if (stat(path, &st))
return false;- return !S_ISDIR(st.st_mode) && (st.st_mode & S_IXUSR);
+} diff --git a/tools/perf/util/path.h b/tools/perf/util/path.h index 083429b7efa3..d94902c22222 100644 --- a/tools/perf/util/path.h +++ b/tools/perf/util/path.h @@ -12,5 +12,6 @@ int path__join3(char *bf, size_t size, const char *path1, const char *path2, con bool is_regular_file(const char *file); bool is_directory(const char *base_path, const struct dirent *dent); +bool is_executable_file(const char *base_path, const struct dirent *dent);
#endif /* _PERF_PATH_H */
2.32.0
-
Arnaldo Carvalho de Melo -
Carsten Haitzler -
carsten.haitzler＠foss.arm.com
-
Leo Yan -
Mathieu Poirier