Changes since v4:
* Rebase onto perf/core * Convert new perf_pmu__cpu_slots_per_cycle() function to use new helper functions
=========================== Changes since v3:
* Scale time estimates by INSTR_PER_NS, rather than assuming 1 instruction = 1ns * Add a new commit that fixes some issues around timestamps going backwards * Use nanoseconds inside cs-etm-decoder.c, rather than storing the raw time values and converting when a sample is synthesized. This simplifies some of the code like estimating the first timestamp.
=========================== Changes since v2:
* Remove const to non-const change and copy strings where needed instead. * Use sizeof() instead of PATH_MAX * Append "will not be set accurately." to new error message * Remove unneeded stat() call * Rebase on perf/core
==========================
Changes since v1:
* Add 3 refactor commits for sysfs reading around pmu.c as suggested by Arnaldo here [1] * The dependency on [2] has now reached mainline so is no longer blocking * Rebase on perf/core
[1]: https://lore.kernel.org/all/YnqVqq5QW%2Fb14oPZ@kernel.org/ [2]: https://lore.kernel.org/all/20220503123537.1003035-1-german.gomez@arm.com/
German Gomez (4): perf pmu: Add function to check if a pmu file exists perf cs_etm: Keep separate symbols for ETMv4 and ETE parameters perf cs_etm: Record ts_source in AUXTRACE_INFO for ETMv4 and ETE perf cs_etm: Set the time field in the synthetic samples
James Clark (4): perf: Remove duplication around EVENT_SOURCE_DEVICE_PATH perf: Use perf_pmu__open_file() and perf_pmu__scan_file() perf: Remove remaining duplication of bus/event_source/devices/... perf: cs-etm: Ensure that Coresight timestamps don't go backwards
tools/perf/arch/arm/util/auxtrace.c | 5 +- tools/perf/arch/arm/util/cs-etm.c | 91 ++++++++- tools/perf/arch/arm64/util/pmu.c | 4 +- tools/perf/arch/x86/util/pmu.c | 12 +- tools/perf/util/cputopo.c | 9 +- tools/perf/util/cs-etm-base.c | 34 +++- .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 68 +++++-- tools/perf/util/cs-etm.c | 95 +++++++++- tools/perf/util/cs-etm.h | 16 +- tools/perf/util/pmu-hybrid.c | 27 +-- tools/perf/util/pmu.c | 177 +++++++----------- tools/perf/util/pmu.h | 10 +- 12 files changed, 351 insertions(+), 197 deletions(-)
base-commit: 1962ab6f6e0b39e4216206205bda14aff87705f3 prerequisite-patch-id: 9722bf86e3e6d16d177ff9a1411992a795a7dcbd prerequisite-patch-id: b05dbef439c2ea8465f3321532257b0ca29f21f9 prerequisite-patch-id: 92680a4781cbcf010fcb007e6ea030f59e9eaefc prerequisite-patch-id: 8e3a73a04e4b89b503377b5fac1d89d551159393 prerequisite-patch-id: 09980d8fedcdaa70b220a7802428109f48448a58
The pattern for accessing EVENT_SOURCE_DEVICE_PATH is duplicated in a few places, so add two utility functions to cover it. Also just use perf_pmu__scan_file() instead of pmu_type() which already does the same thing.
No functional changes.
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com --- tools/perf/arch/arm/util/auxtrace.c | 5 +- tools/perf/arch/arm64/util/pmu.c | 4 +- tools/perf/arch/x86/util/pmu.c | 12 +-- tools/perf/util/pmu.c | 110 +++++++++++----------------- tools/perf/util/pmu.h | 5 +- 5 files changed, 51 insertions(+), 85 deletions(-)
diff --git a/tools/perf/arch/arm/util/auxtrace.c b/tools/perf/arch/arm/util/auxtrace.c index deeb163999ce..adec6c9ee11d 100644 --- a/tools/perf/arch/arm/util/auxtrace.c +++ b/tools/perf/arch/arm/util/auxtrace.c @@ -55,17 +55,16 @@ static struct perf_pmu **find_all_arm_spe_pmus(int *nr_spes, int *err)
static struct perf_pmu **find_all_hisi_ptt_pmus(int *nr_ptts, int *err) { - const char *sysfs = sysfs__mountpoint(); struct perf_pmu **hisi_ptt_pmus = NULL; struct dirent *dent; char path[PATH_MAX]; DIR *dir = NULL; int idx = 0;
- snprintf(path, PATH_MAX, "%s" EVENT_SOURCE_DEVICE_PATH, sysfs); + perf_pmu__event_source_devices_scnprintf(path, sizeof(path)); dir = opendir(path); if (!dir) { - pr_err("can't read directory '%s'\n", EVENT_SOURCE_DEVICE_PATH); + pr_err("can't read directory '%s'\n", path); *err = -EINVAL; return NULL; } diff --git a/tools/perf/arch/arm64/util/pmu.c b/tools/perf/arch/arm64/util/pmu.c index 9e674cac5a73..801bf52e2ea6 100644 --- a/tools/perf/arch/arm64/util/pmu.c +++ b/tools/perf/arch/arm64/util/pmu.c @@ -44,8 +44,8 @@ double perf_pmu__cpu_slots_per_cycle(void) struct perf_pmu *pmu = pmu__find_core_pmu();
if (pmu) { - scnprintf(path, PATH_MAX, - EVENT_SOURCE_DEVICE_PATH "%s/caps/slots", pmu->name); + perf_pmu__pathname_scnprintf(path, sizeof(path), + pmu->name, "caps/slots"); /* * The value of slots is not greater than 32 bits, but sysfs__read_int * can't read value with 0x prefix, so use sysfs__read_ull instead. diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c index 74d69db1ea99..358340b34243 100644 --- a/tools/perf/arch/x86/util/pmu.c +++ b/tools/perf/arch/x86/util/pmu.c @@ -15,8 +15,6 @@ #include "../../../util/pmu.h" #include "../../../util/fncache.h"
-#define TEMPLATE_ALIAS "%s/bus/event_source/devices/%s/alias" - struct pmu_alias { char *name; char *alias; @@ -72,18 +70,14 @@ static int setup_pmu_alias_list(void) char path[PATH_MAX]; DIR *dir; struct dirent *dent; - const char *sysfs = sysfs__mountpoint(); struct pmu_alias *pmu_alias; char buf[MAX_PMU_NAME_LEN]; FILE *file; int ret = -ENOMEM;
- if (!sysfs) + if (!perf_pmu__event_source_devices_scnprintf(path, sizeof(path))) return -1;
- snprintf(path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH, sysfs); - dir = opendir(path); if (!dir) return -errno; @@ -93,9 +87,7 @@ static int setup_pmu_alias_list(void) !strcmp(dent->d_name, "..")) continue;
- snprintf(path, PATH_MAX, - TEMPLATE_ALIAS, sysfs, dent->d_name); - + perf_pmu__pathname_scnprintf(path, sizeof(path), dent->d_name, "alias"); if (!file_available(path)) continue;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index cbb4fbf124bd..1edbb714ff32 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -108,14 +108,10 @@ int perf_pmu__format_parse(char *dir, struct list_head *head) static int pmu_format(const char *name, struct list_head *format) { char path[PATH_MAX]; - const char *sysfs = sysfs__mountpoint();
- if (!sysfs) + if (!perf_pmu__pathname_scnprintf(path, sizeof(path), name, "format")) return -1;
- snprintf(path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH "%s/format", sysfs, name); - if (!file_available(path)) return 0;
@@ -514,14 +510,10 @@ static int pmu_aliases_parse(char *dir, struct list_head *head) static int pmu_aliases(const char *name, struct list_head *head) { char path[PATH_MAX]; - const char *sysfs = sysfs__mountpoint();
- if (!sysfs) + if (!perf_pmu__pathname_scnprintf(path, sizeof(path), name, "events")) return -1;
- snprintf(path, PATH_MAX, - "%s/bus/event_source/devices/%s/events", sysfs, name); - if (!file_available(path)) return 0;
@@ -555,52 +547,16 @@ static int pmu_alias_terms(struct perf_pmu_alias *alias, return 0; }
-/* - * Reading/parsing the default pmu type value, which should be - * located at: - * /sys/bus/event_source/devices/<dev>/type as sysfs attribute. - */ -static int pmu_type(const char *name, __u32 *type) -{ - char path[PATH_MAX]; - FILE *file; - int ret = 0; - const char *sysfs = sysfs__mountpoint(); - - if (!sysfs) - return -1; - - snprintf(path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH "%s/type", sysfs, name); - - if (access(path, R_OK) < 0) - return -1; - - file = fopen(path, "r"); - if (!file) - return -EINVAL; - - if (1 != fscanf(file, "%u", type)) - ret = -1; - - fclose(file); - return ret; -} - /* Add all pmus in sysfs to pmu list: */ static void pmu_read_sysfs(void) { char path[PATH_MAX]; DIR *dir; struct dirent *dent; - const char *sysfs = sysfs__mountpoint();
- if (!sysfs) + if (!perf_pmu__event_source_devices_scnprintf(path, sizeof(path))) return;
- snprintf(path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH, sysfs); - dir = opendir(path); if (!dir) return; @@ -697,14 +653,9 @@ static char *pmu_id(const char *name) static int is_arm_pmu_core(const char *name) { char path[PATH_MAX]; - const char *sysfs = sysfs__mountpoint();
- if (!sysfs) + if (!perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpus")) return 0; - - /* Look for cpu sysfs (specific to arm) */ - scnprintf(path, PATH_MAX, "%s/bus/event_source/devices/%s/cpus", - sysfs, name); return file_available(path); }
@@ -970,11 +921,8 @@ static struct perf_pmu *pmu_lookup(const char *lookup_name) return NULL;
/* - * Check the type first to avoid unnecessary work. + * Check the aliases first to avoid unnecessary work. */ - if (pmu_type(name, &type)) - return NULL; - if (pmu_aliases(name, &aliases)) return NULL;
@@ -984,9 +932,14 @@ static struct perf_pmu *pmu_lookup(const char *lookup_name)
pmu->cpus = pmu_cpumask(name); pmu->name = strdup(name); + if (!pmu->name) goto err;
+ /* Read type, and ensure that type value is successfully assigned (return 1) */ + if (perf_pmu__scan_file(pmu, "type", "%u", &type) != 1) + goto err; + alias_name = pmu_find_alias_name(name); if (alias_name) { pmu->alias_name = strdup(alias_name); @@ -1787,16 +1740,11 @@ bool pmu_have_event(const char *pname, const char *name) static FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) { char path[PATH_MAX]; - const char *sysfs;
- sysfs = sysfs__mountpoint(); - if (!sysfs) + if (!perf_pmu__pathname_scnprintf(path, sizeof(path), pmu->name, name) || + !file_available(path)) return NULL;
- snprintf(path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH "%s/%s", sysfs, pmu->name, name); - if (!file_available(path)) - return NULL; return fopen(path, "r"); }
@@ -1850,7 +1798,6 @@ int perf_pmu__caps_parse(struct perf_pmu *pmu) { struct stat st; char caps_path[PATH_MAX]; - const char *sysfs = sysfs__mountpoint(); DIR *caps_dir; struct dirent *evt_ent;
@@ -1859,12 +1806,9 @@ int perf_pmu__caps_parse(struct perf_pmu *pmu)
pmu->nr_caps = 0;
- if (!sysfs) + if (!perf_pmu__pathname_scnprintf(caps_path, sizeof(caps_path), pmu->name, "caps")) return -1;
- snprintf(caps_path, PATH_MAX, - "%s" EVENT_SOURCE_DEVICE_PATH "%s/caps", sysfs, pmu->name); - if (stat(caps_path, &st) < 0) { pmu->caps_initialized = true; return 0; /* no error if caps does not exist */ @@ -1999,3 +1943,31 @@ double __weak perf_pmu__cpu_slots_per_cycle(void) { return NAN; } + +int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size) +{ + const char *sysfs = sysfs__mountpoint(); + + if (!sysfs) + return 0; + return scnprintf(pathname, size, "%s/bus/event_source/devices/", sysfs); +} + +/* + * Fill 'buf' with the path to a file or folder in 'pmu_name' in + * sysfs. For example if pmu_name = "cs_etm" and 'filename' = "format" + * then pathname will be filled with + * "/sys/bus/event_source/devices/cs_etm/format" + * + * Return 0 if the sysfs mountpoint couldn't be found or if no + * characters were written. + */ +int perf_pmu__pathname_scnprintf(char *buf, size_t size, + const char *pmu_name, const char *filename) +{ + char base_path[PATH_MAX]; + + if (!perf_pmu__event_source_devices_scnprintf(base_path, sizeof(base_path))) + return 0; + return scnprintf(buf, size, "%s%s/%s", base_path, pmu_name, filename); +} diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index fd414ba1d776..96d030c8b3b3 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -22,7 +22,6 @@ enum { };
#define PERF_PMU_FORMAT_BITS 64 -#define EVENT_SOURCE_DEVICE_PATH "/bus/event_source/devices/" #define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus" #define MAX_PMU_NAME_LEN 128
@@ -260,4 +259,8 @@ int perf_pmu__cpus_match(struct perf_pmu *pmu, struct perf_cpu_map *cpus, char *pmu_find_real_name(const char *name); char *pmu_find_alias_name(const char *name); double perf_pmu__cpu_slots_per_cycle(void); +int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size); +int perf_pmu__pathname_scnprintf(char *buf, size_t size, + const char *pmu_name, const char *filename); + #endif /* __PMU_H */
Remove some code that duplicates existing methods. Copy strings where const strings are required.
No functional changes.
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com --- tools/perf/util/cputopo.c | 9 +------- tools/perf/util/pmu-hybrid.c | 27 +++++------------------- tools/perf/util/pmu.c | 40 +++++++++++------------------------- tools/perf/util/pmu.h | 3 ++- 4 files changed, 20 insertions(+), 59 deletions(-)
diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c index 1a3ff6449158..e08797c3cdbc 100644 --- a/tools/perf/util/cputopo.c +++ b/tools/perf/util/cputopo.c @@ -422,8 +422,6 @@ void numa_topology__delete(struct numa_topology *tp) static int load_hybrid_node(struct hybrid_topology_node *node, struct perf_pmu *pmu) { - const char *sysfs; - char path[PATH_MAX]; char *buf = NULL, *p; FILE *fp; size_t len = 0; @@ -432,12 +430,7 @@ static int load_hybrid_node(struct hybrid_topology_node *node, if (!node->pmu_name) return -1;
- sysfs = sysfs__mountpoint(); - if (!sysfs) - goto err; - - snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, pmu->name); - fp = fopen(path, "r"); + fp = perf_pmu__open_file(pmu, "cpus"); if (!fp) goto err;
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c index f51ccaac60ee..38628805a952 100644 --- a/tools/perf/util/pmu-hybrid.c +++ b/tools/perf/util/pmu-hybrid.c @@ -20,32 +20,15 @@ LIST_HEAD(perf_pmu__hybrid_pmus);
bool perf_pmu__hybrid_mounted(const char *name) { - char path[PATH_MAX]; - const char *sysfs; - FILE *file; - int n, cpu; + int cpu; + char pmu_name[PATH_MAX]; + struct perf_pmu pmu = {.name = pmu_name};
if (strncmp(name, "cpu_", 4)) return false;
- sysfs = sysfs__mountpoint(); - if (!sysfs) - return false; - - snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name); - if (!file_available(path)) - return false; - - file = fopen(path, "r"); - if (!file) - return false; - - n = fscanf(file, "%u", &cpu); - fclose(file); - if (n <= 0) - return false; - - return true; + strlcpy(pmu_name, name, sizeof(pmu_name)); + return perf_pmu__scan_file(&pmu, "cpus", "%u", &cpu) > 0; }
struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name) diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index 1edbb714ff32..a771a5972fc5 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -571,45 +571,31 @@ static void pmu_read_sysfs(void) closedir(dir); }
-static struct perf_cpu_map *__pmu_cpumask(const char *path) -{ - FILE *file; - struct perf_cpu_map *cpus; - - file = fopen(path, "r"); - if (!file) - return NULL; - - cpus = perf_cpu_map__read(file); - fclose(file); - return cpus; -} - /* * Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64) * may have a "cpus" file. */ #define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier" -#define CPUS_TEMPLATE_UNCORE "%s/bus/event_source/devices/%s/cpumask"
static struct perf_cpu_map *pmu_cpumask(const char *name) { - char path[PATH_MAX]; struct perf_cpu_map *cpus; - const char *sysfs = sysfs__mountpoint(); const char *templates[] = { - CPUS_TEMPLATE_UNCORE, - CPUS_TEMPLATE_CPU, + "cpumask", + "cpus", NULL }; const char **template; + char pmu_name[PATH_MAX]; + struct perf_pmu pmu = {.name = pmu_name}; + FILE *file;
- if (!sysfs) - return NULL; - + strlcpy(pmu_name, name, sizeof(pmu_name)); for (template = templates; *template; template++) { - snprintf(path, PATH_MAX, *template, sysfs, name); - cpus = __pmu_cpumask(path); + file = perf_pmu__open_file(&pmu, *template); + if (!file) + continue; + cpus = perf_cpu_map__read(file); if (cpus) return cpus; } @@ -620,13 +606,11 @@ static struct perf_cpu_map *pmu_cpumask(const char *name) static bool pmu_is_uncore(const char *name) { char path[PATH_MAX]; - const char *sysfs;
if (perf_pmu__hybrid_mounted(name)) return false;
- sysfs = sysfs__mountpoint(); - snprintf(path, PATH_MAX, CPUS_TEMPLATE_UNCORE, sysfs, name); + perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpumask"); return file_available(path); }
@@ -1737,7 +1721,7 @@ bool pmu_have_event(const char *pname, const char *name) return false; }
-static FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) { char path[PATH_MAX];
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index 96d030c8b3b3..742d4db319a0 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -7,6 +7,7 @@ #include <linux/perf_event.h> #include <linux/list.h> #include <stdbool.h> +#include <stdio.h> #include "parse-events.h" #include "pmu-events/pmu-events.h"
@@ -22,7 +23,6 @@ enum { };
#define PERF_PMU_FORMAT_BITS 64 -#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus" #define MAX_PMU_NAME_LEN 128
struct perf_event_attr; @@ -262,5 +262,6 @@ double perf_pmu__cpu_slots_per_cycle(void); int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size); int perf_pmu__pathname_scnprintf(char *buf, size_t size, const char *pmu_name, const char *filename); +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
#endif /* __PMU_H */
Em Fri, Jan 20, 2023 at 02:36:55PM +0000, James Clark escreveu:
Remove some code that duplicates existing methods. Copy strings where const strings are required.
No functional changes.
Have you used 'perf test'?
[acme@quaco perf]$ perf test -v python Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc 19: 'import perf' in python : --- start --- test child forked, pid 232379 python usage test: "echo "import sys ; sys.path.append('/tmp/build/perf/python'); import perf" | '/usr/bin/python3' " Traceback (most recent call last): File "<stdin>", line 1, in <module> ImportError: /tmp/build/perf/python/perf.cpython-310-x86_64-linux-gnu.so: undefined symbol: perf_pmu__scan_file test child finished with -1 ---- end ---- 'import perf' in python: FAILED! [acme@quaco perf]$
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com
tools/perf/util/cputopo.c | 9 +------- tools/perf/util/pmu-hybrid.c | 27 +++++------------------- tools/perf/util/pmu.c | 40 +++++++++++------------------------- tools/perf/util/pmu.h | 3 ++- 4 files changed, 20 insertions(+), 59 deletions(-)
diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c index 1a3ff6449158..e08797c3cdbc 100644 --- a/tools/perf/util/cputopo.c +++ b/tools/perf/util/cputopo.c @@ -422,8 +422,6 @@ void numa_topology__delete(struct numa_topology *tp) static int load_hybrid_node(struct hybrid_topology_node *node, struct perf_pmu *pmu) {
- const char *sysfs;
- char path[PATH_MAX]; char *buf = NULL, *p; FILE *fp; size_t len = 0;
@@ -432,12 +430,7 @@ static int load_hybrid_node(struct hybrid_topology_node *node, if (!node->pmu_name) return -1;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
goto err;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, pmu->name);
- fp = fopen(path, "r");
- fp = perf_pmu__open_file(pmu, "cpus"); if (!fp) goto err;
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c index f51ccaac60ee..38628805a952 100644 --- a/tools/perf/util/pmu-hybrid.c +++ b/tools/perf/util/pmu-hybrid.c @@ -20,32 +20,15 @@ LIST_HEAD(perf_pmu__hybrid_pmus); bool perf_pmu__hybrid_mounted(const char *name) {
- char path[PATH_MAX];
- const char *sysfs;
- FILE *file;
- int n, cpu;
- int cpu;
- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
if (strncmp(name, "cpu_", 4)) return false;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
return false;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name);
- if (!file_available(path))
return false;- file = fopen(path, "r");
- if (!file)
return false;- n = fscanf(file, "%u", &cpu);
- fclose(file);
- if (n <= 0)
return false;- return true;
- strlcpy(pmu_name, name, sizeof(pmu_name));
- return perf_pmu__scan_file(&pmu, "cpus", "%u", &cpu) > 0;
} struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name) diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index 1edbb714ff32..a771a5972fc5 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -571,45 +571,31 @@ static void pmu_read_sysfs(void) closedir(dir); } -static struct perf_cpu_map *__pmu_cpumask(const char *path) -{
- FILE *file;
- struct perf_cpu_map *cpus;
- file = fopen(path, "r");
- if (!file)
return NULL;- cpus = perf_cpu_map__read(file);
- fclose(file);
- return cpus;
-}
/*
- Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64)
- may have a "cpus" file.
*/ #define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier" -#define CPUS_TEMPLATE_UNCORE "%s/bus/event_source/devices/%s/cpumask" static struct perf_cpu_map *pmu_cpumask(const char *name) {
- char path[PATH_MAX]; struct perf_cpu_map *cpus;
- const char *sysfs = sysfs__mountpoint(); const char *templates[] = {
CPUS_TEMPLATE_UNCORE,CPUS_TEMPLATE_CPU,
"cpumask","cpus",- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
- FILE *file;
- if (!sysfs)
return NULL;
- strlcpy(pmu_name, name, sizeof(pmu_name)); for (template = templates; *template; template++) {
snprintf(path, PATH_MAX, *template, sysfs, name);cpus = __pmu_cpumask(path);
file = perf_pmu__open_file(&pmu, *template);if (!file)continue;cpus = perf_cpu_map__read(file);@@ -620,13 +606,11 @@ static struct perf_cpu_map *pmu_cpumask(const char *name) static bool pmu_is_uncore(const char *name) { char path[PATH_MAX];
- const char *sysfs;
if (perf_pmu__hybrid_mounted(name)) return false;
- sysfs = sysfs__mountpoint();
- snprintf(path, PATH_MAX, CPUS_TEMPLATE_UNCORE, sysfs, name);
- perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpumask"); return file_available(path);
} @@ -1737,7 +1721,7 @@ bool pmu_have_event(const char *pname, const char *name) return false; } -static FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) { char path[PATH_MAX]; diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index 96d030c8b3b3..742d4db319a0 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -7,6 +7,7 @@ #include <linux/perf_event.h> #include <linux/list.h> #include <stdbool.h> +#include <stdio.h> #include "parse-events.h" #include "pmu-events/pmu-events.h" @@ -22,7 +23,6 @@ enum { }; #define PERF_PMU_FORMAT_BITS 64 -#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus" #define MAX_PMU_NAME_LEN 128 struct perf_event_attr; @@ -262,5 +262,6 @@ double perf_pmu__cpu_slots_per_cycle(void); int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size); int perf_pmu__pathname_scnprintf(char *buf, size_t size, const char *pmu_name, const char *filename); +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
#endif /* __PMU_H */
2.25.1
Em Fri, Jan 20, 2023 at 02:21:56PM -0300, Arnaldo Carvalho de Melo escreveu:
Em Fri, Jan 20, 2023 at 02:36:55PM +0000, James Clark escreveu:
Remove some code that duplicates existing methods. Copy strings where const strings are required.
No functional changes.
Have you used 'perf test'?
[acme@quaco perf]$ perf test -v python Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc 19: 'import perf' in python : --- start --- test child forked, pid 232379 python usage test: "echo "import sys ; sys.path.append('/tmp/build/perf/python'); import perf" | '/usr/bin/python3' " Traceback (most recent call last): File "<stdin>", line 1, in <module> ImportError: /tmp/build/perf/python/perf.cpython-310-x86_64-linux-gnu.so: undefined symbol: perf_pmu__scan_file test child finished with -1 ---- end ---- 'import perf' in python: FAILED! [acme@quaco perf]$
I added this to this cset, now it passes.
diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c index d948455e5ed43656..9e5d881b098767a3 100644 --- a/tools/perf/util/python.c +++ b/tools/perf/util/python.c @@ -20,6 +20,7 @@ #include "stat.h" #include "metricgroup.h" #include "util/env.h" +#include "util/pmu.h" #include <internal/lib.h> #include "util.h"
@@ -83,7 +84,7 @@ void perf_stat__collect_metric_expr(struct evlist *evsel_list) }
/* - * This one is needed not to drag the PMU bandwagon, jevents generated + * These ones are needed not to drag the PMU bandwagon, jevents generated * pmu_sys_event_tables, etc and evsel__find_pmu() is used so far just for * doing per PMU perf_event_attr.exclude_guest handling, not really needed, so * far, for the perf python binding known usecases, revisit if this become @@ -94,6 +95,11 @@ struct perf_pmu *evsel__find_pmu(struct evsel *evsel __maybe_unused) return NULL; }
+int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt, ...) +{ + return EOF; +} + /* * Add this one here not to drag util/metricgroup.c */
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com
tools/perf/util/cputopo.c | 9 +------- tools/perf/util/pmu-hybrid.c | 27 +++++------------------- tools/perf/util/pmu.c | 40 +++++++++++------------------------- tools/perf/util/pmu.h | 3 ++- 4 files changed, 20 insertions(+), 59 deletions(-)
diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c index 1a3ff6449158..e08797c3cdbc 100644 --- a/tools/perf/util/cputopo.c +++ b/tools/perf/util/cputopo.c @@ -422,8 +422,6 @@ void numa_topology__delete(struct numa_topology *tp) static int load_hybrid_node(struct hybrid_topology_node *node, struct perf_pmu *pmu) {
- const char *sysfs;
- char path[PATH_MAX]; char *buf = NULL, *p; FILE *fp; size_t len = 0;
@@ -432,12 +430,7 @@ static int load_hybrid_node(struct hybrid_topology_node *node, if (!node->pmu_name) return -1;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
goto err;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, pmu->name);
- fp = fopen(path, "r");
- fp = perf_pmu__open_file(pmu, "cpus"); if (!fp) goto err;
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c index f51ccaac60ee..38628805a952 100644 --- a/tools/perf/util/pmu-hybrid.c +++ b/tools/perf/util/pmu-hybrid.c @@ -20,32 +20,15 @@ LIST_HEAD(perf_pmu__hybrid_pmus); bool perf_pmu__hybrid_mounted(const char *name) {
- char path[PATH_MAX];
- const char *sysfs;
- FILE *file;
- int n, cpu;
- int cpu;
- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
if (strncmp(name, "cpu_", 4)) return false;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
return false;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name);
- if (!file_available(path))
return false;- file = fopen(path, "r");
- if (!file)
return false;- n = fscanf(file, "%u", &cpu);
- fclose(file);
- if (n <= 0)
return false;- return true;
- strlcpy(pmu_name, name, sizeof(pmu_name));
- return perf_pmu__scan_file(&pmu, "cpus", "%u", &cpu) > 0;
} struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name) diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index 1edbb714ff32..a771a5972fc5 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -571,45 +571,31 @@ static void pmu_read_sysfs(void) closedir(dir); } -static struct perf_cpu_map *__pmu_cpumask(const char *path) -{
- FILE *file;
- struct perf_cpu_map *cpus;
- file = fopen(path, "r");
- if (!file)
return NULL;- cpus = perf_cpu_map__read(file);
- fclose(file);
- return cpus;
-}
/*
- Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64)
- may have a "cpus" file.
*/ #define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier" -#define CPUS_TEMPLATE_UNCORE "%s/bus/event_source/devices/%s/cpumask" static struct perf_cpu_map *pmu_cpumask(const char *name) {
- char path[PATH_MAX]; struct perf_cpu_map *cpus;
- const char *sysfs = sysfs__mountpoint(); const char *templates[] = {
CPUS_TEMPLATE_UNCORE,CPUS_TEMPLATE_CPU,
"cpumask","cpus",- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
- FILE *file;
- if (!sysfs)
return NULL;
- strlcpy(pmu_name, name, sizeof(pmu_name)); for (template = templates; *template; template++) {
snprintf(path, PATH_MAX, *template, sysfs, name);cpus = __pmu_cpumask(path);
file = perf_pmu__open_file(&pmu, *template);if (!file)continue;cpus = perf_cpu_map__read(file);@@ -620,13 +606,11 @@ static struct perf_cpu_map *pmu_cpumask(const char *name) static bool pmu_is_uncore(const char *name) { char path[PATH_MAX];
- const char *sysfs;
if (perf_pmu__hybrid_mounted(name)) return false;
- sysfs = sysfs__mountpoint();
- snprintf(path, PATH_MAX, CPUS_TEMPLATE_UNCORE, sysfs, name);
- perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpumask"); return file_available(path);
} @@ -1737,7 +1721,7 @@ bool pmu_have_event(const char *pname, const char *name) return false; } -static FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) { char path[PATH_MAX]; diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index 96d030c8b3b3..742d4db319a0 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -7,6 +7,7 @@ #include <linux/perf_event.h> #include <linux/list.h> #include <stdbool.h> +#include <stdio.h> #include "parse-events.h" #include "pmu-events/pmu-events.h" @@ -22,7 +23,6 @@ enum { }; #define PERF_PMU_FORMAT_BITS 64 -#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus" #define MAX_PMU_NAME_LEN 128 struct perf_event_attr; @@ -262,5 +262,6 @@ double perf_pmu__cpu_slots_per_cycle(void); int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size); int perf_pmu__pathname_scnprintf(char *buf, size_t size, const char *pmu_name, const char *filename); +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
#endif /* __PMU_H */
2.25.1
--
- Arnaldo
Em Fri, Jan 20, 2023 at 02:32:30PM -0300, Arnaldo Carvalho de Melo escreveu:
Em Fri, Jan 20, 2023 at 02:21:56PM -0300, Arnaldo Carvalho de Melo escreveu:
Em Fri, Jan 20, 2023 at 02:36:55PM +0000, James Clark escreveu:
Remove some code that duplicates existing methods. Copy strings where const strings are required.
No functional changes.
Have you used 'perf test'?
[acme@quaco perf]$ perf test -v python Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc 19: 'import perf' in python : --- start --- test child forked, pid 232379 python usage test: "echo "import sys ; sys.path.append('/tmp/build/perf/python'); import perf" | '/usr/bin/python3' " Traceback (most recent call last): File "<stdin>", line 1, in <module> ImportError: /tmp/build/perf/python/perf.cpython-310-x86_64-linux-gnu.so: undefined symbol: perf_pmu__scan_file test child finished with -1 ---- end ---- 'import perf' in python: FAILED! [acme@quaco perf]$
I added this to this cset, now it passes.
So, what I have is now at my tmp.perf/core branch, pending container testing, later today probably will move to perf/core, so that it gets exposure on linux-next for v6.3.
For your convenience:
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/log/?h=tmp.pe...
Thanks,
- Arnaldo
diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c index d948455e5ed43656..9e5d881b098767a3 100644 --- a/tools/perf/util/python.c +++ b/tools/perf/util/python.c @@ -20,6 +20,7 @@ #include "stat.h" #include "metricgroup.h" #include "util/env.h" +#include "util/pmu.h" #include <internal/lib.h> #include "util.h" @@ -83,7 +84,7 @@ void perf_stat__collect_metric_expr(struct evlist *evsel_list) } /*
- This one is needed not to drag the PMU bandwagon, jevents generated
- These ones are needed not to drag the PMU bandwagon, jevents generated
- pmu_sys_event_tables, etc and evsel__find_pmu() is used so far just for
- doing per PMU perf_event_attr.exclude_guest handling, not really needed, so
- far, for the perf python binding known usecases, revisit if this become
@@ -94,6 +95,11 @@ struct perf_pmu *evsel__find_pmu(struct evsel *evsel __maybe_unused) return NULL; } +int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt, ...) +{
- return EOF;
+}
/*
- Add this one here not to drag util/metricgroup.c
*/
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com
tools/perf/util/cputopo.c | 9 +------- tools/perf/util/pmu-hybrid.c | 27 +++++------------------- tools/perf/util/pmu.c | 40 +++++++++++------------------------- tools/perf/util/pmu.h | 3 ++- 4 files changed, 20 insertions(+), 59 deletions(-)
diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c index 1a3ff6449158..e08797c3cdbc 100644 --- a/tools/perf/util/cputopo.c +++ b/tools/perf/util/cputopo.c @@ -422,8 +422,6 @@ void numa_topology__delete(struct numa_topology *tp) static int load_hybrid_node(struct hybrid_topology_node *node, struct perf_pmu *pmu) {
- const char *sysfs;
- char path[PATH_MAX]; char *buf = NULL, *p; FILE *fp; size_t len = 0;
@@ -432,12 +430,7 @@ static int load_hybrid_node(struct hybrid_topology_node *node, if (!node->pmu_name) return -1;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
goto err;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, pmu->name);
- fp = fopen(path, "r");
- fp = perf_pmu__open_file(pmu, "cpus"); if (!fp) goto err;
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c index f51ccaac60ee..38628805a952 100644 --- a/tools/perf/util/pmu-hybrid.c +++ b/tools/perf/util/pmu-hybrid.c @@ -20,32 +20,15 @@ LIST_HEAD(perf_pmu__hybrid_pmus); bool perf_pmu__hybrid_mounted(const char *name) {
- char path[PATH_MAX];
- const char *sysfs;
- FILE *file;
- int n, cpu;
- int cpu;
- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
if (strncmp(name, "cpu_", 4)) return false;
- sysfs = sysfs__mountpoint();
- if (!sysfs)
return false;- snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name);
- if (!file_available(path))
return false;- file = fopen(path, "r");
- if (!file)
return false;- n = fscanf(file, "%u", &cpu);
- fclose(file);
- if (n <= 0)
return false;- return true;
- strlcpy(pmu_name, name, sizeof(pmu_name));
- return perf_pmu__scan_file(&pmu, "cpus", "%u", &cpu) > 0;
} struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name) diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index 1edbb714ff32..a771a5972fc5 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -571,45 +571,31 @@ static void pmu_read_sysfs(void) closedir(dir); } -static struct perf_cpu_map *__pmu_cpumask(const char *path) -{
- FILE *file;
- struct perf_cpu_map *cpus;
- file = fopen(path, "r");
- if (!file)
return NULL;- cpus = perf_cpu_map__read(file);
- fclose(file);
- return cpus;
-}
/*
- Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64)
- may have a "cpus" file.
*/ #define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier" -#define CPUS_TEMPLATE_UNCORE "%s/bus/event_source/devices/%s/cpumask" static struct perf_cpu_map *pmu_cpumask(const char *name) {
- char path[PATH_MAX]; struct perf_cpu_map *cpus;
- const char *sysfs = sysfs__mountpoint(); const char *templates[] = {
CPUS_TEMPLATE_UNCORE,CPUS_TEMPLATE_CPU,
"cpumask","cpus",- char pmu_name[PATH_MAX];
- struct perf_pmu pmu = {.name = pmu_name};
- FILE *file;
- if (!sysfs)
return NULL;
- strlcpy(pmu_name, name, sizeof(pmu_name)); for (template = templates; *template; template++) {
snprintf(path, PATH_MAX, *template, sysfs, name);cpus = __pmu_cpumask(path);
file = perf_pmu__open_file(&pmu, *template);if (!file)continue;cpus = perf_cpu_map__read(file);@@ -620,13 +606,11 @@ static struct perf_cpu_map *pmu_cpumask(const char *name) static bool pmu_is_uncore(const char *name) { char path[PATH_MAX];
- const char *sysfs;
if (perf_pmu__hybrid_mounted(name)) return false;
- sysfs = sysfs__mountpoint();
- snprintf(path, PATH_MAX, CPUS_TEMPLATE_UNCORE, sysfs, name);
- perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpumask"); return file_available(path);
} @@ -1737,7 +1721,7 @@ bool pmu_have_event(const char *pname, const char *name) return false; } -static FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name) { char path[PATH_MAX]; diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index 96d030c8b3b3..742d4db319a0 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -7,6 +7,7 @@ #include <linux/perf_event.h> #include <linux/list.h> #include <stdbool.h> +#include <stdio.h> #include "parse-events.h" #include "pmu-events/pmu-events.h" @@ -22,7 +23,6 @@ enum { }; #define PERF_PMU_FORMAT_BITS 64 -#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus" #define MAX_PMU_NAME_LEN 128 struct perf_event_attr; @@ -262,5 +262,6 @@ double perf_pmu__cpu_slots_per_cycle(void); int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size); int perf_pmu__pathname_scnprintf(char *buf, size_t size, const char *pmu_name, const char *filename); +FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
#endif /* __PMU_H */
2.25.1
--
- Arnaldo
--
- Arnaldo
On 20/01/2023 17:43, Arnaldo Carvalho de Melo wrote:
Em Fri, Jan 20, 2023 at 02:32:30PM -0300, Arnaldo Carvalho de Melo escreveu:
Em Fri, Jan 20, 2023 at 02:21:56PM -0300, Arnaldo Carvalho de Melo escreveu:
Em Fri, Jan 20, 2023 at 02:36:55PM +0000, James Clark escreveu:
Remove some code that duplicates existing methods. Copy strings where const strings are required.
No functional changes.
Have you used 'perf test'?
[acme@quaco perf]$ perf test -v python Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc 19: 'import perf' in python : --- start --- test child forked, pid 232379 python usage test: "echo "import sys ; sys.path.append('/tmp/build/perf/python'); import perf" | '/usr/bin/python3' " Traceback (most recent call last): File "<stdin>", line 1, in <module> ImportError: /tmp/build/perf/python/perf.cpython-310-x86_64-linux-gnu.so: undefined symbol: perf_pmu__scan_file test child finished with -1 ---- end ---- 'import perf' in python: FAILED! [acme@quaco perf]$
I added this to this cset, now it passes.
So, what I have is now at my tmp.perf/core branch, pending container testing, later today probably will move to perf/core, so that it gets exposure on linux-next for v6.3.
Gah! Sorry, I must have only run the Coresight tests. I will make sure all the tests are passing on my setup so it's easier to spot next time.
Thanks for the fix.
Use the new perf_pmu__pathname_scnprintf() instead. No functional changes.
Reviewed-by: Leo Yan leo.yan@linaro.org Signed-off-by: James Clark james.clark@arm.com --- tools/perf/util/pmu.c | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-)
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index a771a5972fc5..23e1d70fa343 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -575,8 +575,6 @@ static void pmu_read_sysfs(void) * Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64) * may have a "cpus" file. */ -#define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier" - static struct perf_cpu_map *pmu_cpumask(const char *name) { struct perf_cpu_map *cpus; @@ -619,9 +617,9 @@ static char *pmu_id(const char *name) char path[PATH_MAX], *str; size_t len;
- snprintf(path, PATH_MAX, SYS_TEMPLATE_ID, name); + perf_pmu__pathname_scnprintf(path, sizeof(path), name, "identifier");
- if (sysfs__read_str(path, &str, &len) < 0) + if (filename__read_str(path, &str, &len) < 0) return NULL;
str[len - 1] = 0; /* remove line feed */ @@ -867,16 +865,11 @@ pmu_find_alias_name(const char *name __maybe_unused) return NULL; }
-static int pmu_max_precise(const char *name) +static int pmu_max_precise(struct perf_pmu *pmu) { - char path[PATH_MAX]; int max_precise = -1;
- scnprintf(path, PATH_MAX, - "bus/event_source/devices/%s/caps/max_precise", - name); - - sysfs__read_int(path, &max_precise); + perf_pmu__scan_file(pmu, "caps/max_precise", "%d", &max_precise); return max_precise; }
@@ -935,7 +928,7 @@ static struct perf_pmu *pmu_lookup(const char *lookup_name) pmu->is_uncore = pmu_is_uncore(name); if (pmu->is_uncore) pmu->id = pmu_id(name); - pmu->max_precise = pmu_max_precise(name); + pmu->max_precise = pmu_max_precise(pmu); pmu_add_cpu_aliases(&aliases, pmu); pmu_add_sys_aliases(&aliases, pmu);
From: German Gomez german.gomez@arm.com
Add a utility function perf_pmu__file_exists() to check if a given pmu file exists in the sysfs filesystem.
Signed-off-by: German Gomez german.gomez@arm.com Signed-off-by: James Clark james.clark@arm.com --- tools/perf/util/pmu.c | 10 ++++++++++ tools/perf/util/pmu.h | 2 ++ 2 files changed, 12 insertions(+)
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c index 23e1d70fa343..8abf5b8439a7 100644 --- a/tools/perf/util/pmu.c +++ b/tools/perf/util/pmu.c @@ -1742,6 +1742,16 @@ int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt, return ret; }
+bool perf_pmu__file_exists(struct perf_pmu *pmu, const char *name) +{ + char path[PATH_MAX]; + + if (!perf_pmu__pathname_scnprintf(path, sizeof(path), pmu->name, name)) + return false; + + return file_available(path); +} + static int perf_pmu__new_caps(struct list_head *list, char *name, char *value) { struct perf_pmu_caps *caps = zalloc(sizeof(*caps)); diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h index 742d4db319a0..2bdc560f19c7 100644 --- a/tools/perf/util/pmu.h +++ b/tools/perf/util/pmu.h @@ -230,6 +230,8 @@ bool pmu_have_event(const char *pname, const char *name);
int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt, ...) __scanf(3, 4);
+bool perf_pmu__file_exists(struct perf_pmu *pmu, const char *name); + int perf_pmu__test(void);
struct perf_event_attr *perf_pmu__get_default_config(struct perf_pmu *pmu);
From: German Gomez german.gomez@arm.com
Previously, adding a new parameter at the end of ETMv4 meant adding it somewhere in the middle of ETE, which is not supported by the current header version.
Reviewed-by: Mike Leach mike.leach@linaro.org Signed-off-by: German Gomez german.gomez@arm.com Signed-off-by: James Clark james.clark@arm.com --- tools/perf/arch/arm/util/cs-etm.c | 43 ++++++++++++++++++++++++++----- tools/perf/util/cs-etm-base.c | 32 +++++++++++++++++------ tools/perf/util/cs-etm.c | 12 ++++----- tools/perf/util/cs-etm.h | 11 +++++++- 4 files changed, 76 insertions(+), 22 deletions(-)
diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c index a346d5f3dafa..b526ffe550a5 100644 --- a/tools/perf/arch/arm/util/cs-etm.c +++ b/tools/perf/arch/arm/util/cs-etm.c @@ -53,7 +53,15 @@ static const char * const metadata_etmv4_ro[] = { [CS_ETMV4_TRCIDR2] = "trcidr/trcidr2", [CS_ETMV4_TRCIDR8] = "trcidr/trcidr8", [CS_ETMV4_TRCAUTHSTATUS] = "mgmt/trcauthstatus", - [CS_ETE_TRCDEVARCH] = "mgmt/trcdevarch" +}; + +static const char * const metadata_ete_ro[] = { + [CS_ETE_TRCIDR0] = "trcidr/trcidr0", + [CS_ETE_TRCIDR1] = "trcidr/trcidr1", + [CS_ETE_TRCIDR2] = "trcidr/trcidr2", + [CS_ETE_TRCIDR8] = "trcidr/trcidr8", + [CS_ETE_TRCAUTHSTATUS] = "mgmt/trcauthstatus", + [CS_ETE_TRCDEVARCH] = "mgmt/trcdevarch", };
static bool cs_etm_is_etmv4(struct auxtrace_record *itr, int cpu); @@ -617,7 +625,7 @@ static bool cs_etm_is_ete(struct auxtrace_record *itr, int cpu) { struct cs_etm_recording *ptr = container_of(itr, struct cs_etm_recording, itr); struct perf_pmu *cs_etm_pmu = ptr->cs_etm_pmu; - int trcdevarch = cs_etm_get_ro(cs_etm_pmu, cpu, metadata_etmv4_ro[CS_ETE_TRCDEVARCH]); + int trcdevarch = cs_etm_get_ro(cs_etm_pmu, cpu, metadata_ete_ro[CS_ETE_TRCDEVARCH]);
/* * ETE if ARCHVER is 5 (ARCHVER is 4 for ETM) and ARCHPART is 0xA13. @@ -648,6 +656,31 @@ static void cs_etm_save_etmv4_header(__u64 data[], struct auxtrace_record *itr, metadata_etmv4_ro[CS_ETMV4_TRCAUTHSTATUS]); }
+static void cs_etm_save_ete_header(__u64 data[], struct auxtrace_record *itr, int cpu) +{ + struct cs_etm_recording *ptr = container_of(itr, struct cs_etm_recording, itr); + struct perf_pmu *cs_etm_pmu = ptr->cs_etm_pmu; + + /* Get trace configuration register */ + data[CS_ETE_TRCCONFIGR] = cs_etmv4_get_config(itr); + /* Get traceID from the framework */ + data[CS_ETE_TRCTRACEIDR] = coresight_get_trace_id(cpu); + /* Get read-only information from sysFS */ + data[CS_ETE_TRCIDR0] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCIDR0]); + data[CS_ETE_TRCIDR1] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCIDR1]); + data[CS_ETE_TRCIDR2] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCIDR2]); + data[CS_ETE_TRCIDR8] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCIDR8]); + data[CS_ETE_TRCAUTHSTATUS] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCAUTHSTATUS]); + /* ETE uses the same registers as ETMv4 plus TRCDEVARCH */ + data[CS_ETE_TRCDEVARCH] = cs_etm_get_ro(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TRCDEVARCH]); +} + static void cs_etm_get_metadata(int cpu, u32 *offset, struct auxtrace_record *itr, struct perf_record_auxtrace_info *info) @@ -661,11 +694,7 @@ static void cs_etm_get_metadata(int cpu, u32 *offset, /* first see what kind of tracer this cpu is affined to */ if (cs_etm_is_ete(itr, cpu)) { magic = __perf_cs_ete_magic; - /* ETE uses the same registers as ETMv4 plus TRCDEVARCH */ - cs_etm_save_etmv4_header(&info->priv[*offset], itr, cpu); - info->priv[*offset + CS_ETE_TRCDEVARCH] = - cs_etm_get_ro(cs_etm_pmu, cpu, - metadata_etmv4_ro[CS_ETE_TRCDEVARCH]); + cs_etm_save_ete_header(&info->priv[*offset], itr, cpu);
/* How much space was used */ increment = CS_ETE_PRIV_MAX; diff --git a/tools/perf/util/cs-etm-base.c b/tools/perf/util/cs-etm-base.c index 597542410854..282042c6e944 100644 --- a/tools/perf/util/cs-etm-base.c +++ b/tools/perf/util/cs-etm-base.c @@ -36,7 +36,20 @@ static const char * const cs_etmv4_priv_fmts[] = { [CS_ETMV4_TRCIDR2] = " TRCIDR2 %llx\n", [CS_ETMV4_TRCIDR8] = " TRCIDR8 %llx\n", [CS_ETMV4_TRCAUTHSTATUS] = " TRCAUTHSTATUS %llx\n", - [CS_ETE_TRCDEVARCH] = " TRCDEVARCH %llx\n" +}; + +static const char * const cs_ete_priv_fmts[] = { + [CS_ETM_MAGIC] = " Magic number %llx\n", + [CS_ETM_CPU] = " CPU %lld\n", + [CS_ETM_NR_TRC_PARAMS] = " NR_TRC_PARAMS %llx\n", + [CS_ETE_TRCCONFIGR] = " TRCCONFIGR %llx\n", + [CS_ETE_TRCTRACEIDR] = " TRCTRACEIDR %llx\n", + [CS_ETE_TRCIDR0] = " TRCIDR0 %llx\n", + [CS_ETE_TRCIDR1] = " TRCIDR1 %llx\n", + [CS_ETE_TRCIDR2] = " TRCIDR2 %llx\n", + [CS_ETE_TRCIDR8] = " TRCIDR8 %llx\n", + [CS_ETE_TRCAUTHSTATUS] = " TRCAUTHSTATUS %llx\n", + [CS_ETE_TRCDEVARCH] = " TRCDEVARCH %llx\n", };
static const char * const param_unk_fmt = @@ -96,19 +109,22 @@ static int cs_etm__print_cpu_metadata_v1(u64 *val, int *offset) else fprintf(stdout, cs_etm_priv_fmts[j], val[i]); } - } else if (magic == __perf_cs_etmv4_magic || magic == __perf_cs_ete_magic) { - /* - * ETE and ETMv4 can be printed in the same block because the number of parameters - * is saved and they share the list of parameter names. ETE is also only supported - * in V1 files. - */ + } else if (magic == __perf_cs_etmv4_magic) { for (j = 0; j < total_params; j++, i++) { /* if newer record - could be excess params */ - if (j >= CS_ETE_PRIV_MAX) + if (j >= CS_ETMV4_PRIV_MAX) fprintf(stdout, param_unk_fmt, j, val[i]); else fprintf(stdout, cs_etmv4_priv_fmts[j], val[i]); } + } else if (magic == __perf_cs_ete_magic) { + for (j = 0; j < total_params; j++, i++) { + /* if newer record - could be excess params */ + if (j >= CS_ETE_PRIV_MAX) + fprintf(stdout, param_unk_fmt, j, val[i]); + else + fprintf(stdout, cs_ete_priv_fmts[j], val[i]); + } } else { /* failure - note bad magic value and error out */ fprintf(stdout, magic_unk_fmt, magic); diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c index 33303d03c2fa..879576d5f899 100644 --- a/tools/perf/util/cs-etm.c +++ b/tools/perf/util/cs-etm.c @@ -464,12 +464,12 @@ static void cs_etm__set_trace_param_ete(struct cs_etm_trace_params *t_params, u64 **metadata = etm->metadata;
t_params[idx].protocol = CS_ETM_PROTO_ETE; - t_params[idx].ete.reg_idr0 = metadata[idx][CS_ETMV4_TRCIDR0]; - t_params[idx].ete.reg_idr1 = metadata[idx][CS_ETMV4_TRCIDR1]; - t_params[idx].ete.reg_idr2 = metadata[idx][CS_ETMV4_TRCIDR2]; - t_params[idx].ete.reg_idr8 = metadata[idx][CS_ETMV4_TRCIDR8]; - t_params[idx].ete.reg_configr = metadata[idx][CS_ETMV4_TRCCONFIGR]; - t_params[idx].ete.reg_traceidr = metadata[idx][CS_ETMV4_TRCTRACEIDR]; + t_params[idx].ete.reg_idr0 = metadata[idx][CS_ETE_TRCIDR0]; + t_params[idx].ete.reg_idr1 = metadata[idx][CS_ETE_TRCIDR1]; + t_params[idx].ete.reg_idr2 = metadata[idx][CS_ETE_TRCIDR2]; + t_params[idx].ete.reg_idr8 = metadata[idx][CS_ETE_TRCIDR8]; + t_params[idx].ete.reg_configr = metadata[idx][CS_ETE_TRCCONFIGR]; + t_params[idx].ete.reg_traceidr = metadata[idx][CS_ETE_TRCTRACEIDR]; t_params[idx].ete.reg_devarch = metadata[idx][CS_ETE_TRCDEVARCH]; }
diff --git a/tools/perf/util/cs-etm.h b/tools/perf/util/cs-etm.h index 5da50d5dae6b..c5925428afa9 100644 --- a/tools/perf/util/cs-etm.h +++ b/tools/perf/util/cs-etm.h @@ -82,7 +82,16 @@ enum { * added in header V1 */ enum { - CS_ETE_TRCDEVARCH = CS_ETMV4_PRIV_MAX, + /* Dynamic, configurable parameters */ + CS_ETE_TRCCONFIGR = CS_ETM_COMMON_BLK_MAX_V1, + CS_ETE_TRCTRACEIDR, + /* RO, taken from sysFS */ + CS_ETE_TRCIDR0, + CS_ETE_TRCIDR1, + CS_ETE_TRCIDR2, + CS_ETE_TRCIDR8, + CS_ETE_TRCAUTHSTATUS, + CS_ETE_TRCDEVARCH, CS_ETE_PRIV_MAX };
From: German Gomez german.gomez@arm.com
Read the value of ts_source exposed by the driver and store it in the ETMv4 and ETE header. If the interface doesn't exist (such as in older Kernels), defaults to a safe value of -1.
Signed-off-by: German Gomez german.gomez@arm.com Signed-off-by: James Clark james.clark@arm.com --- tools/perf/arch/arm/util/cs-etm.c | 48 +++++++++++++++++++++++++++++++ tools/perf/util/cs-etm-base.c | 2 ++ tools/perf/util/cs-etm.h | 2 ++ 3 files changed, 52 insertions(+)
diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c index b526ffe550a5..481e170cd3f1 100644 --- a/tools/perf/arch/arm/util/cs-etm.c +++ b/tools/perf/arch/arm/util/cs-etm.c @@ -53,6 +53,7 @@ static const char * const metadata_etmv4_ro[] = { [CS_ETMV4_TRCIDR2] = "trcidr/trcidr2", [CS_ETMV4_TRCIDR8] = "trcidr/trcidr8", [CS_ETMV4_TRCAUTHSTATUS] = "mgmt/trcauthstatus", + [CS_ETMV4_TS_SOURCE] = "ts_source", };
static const char * const metadata_ete_ro[] = { @@ -62,6 +63,7 @@ static const char * const metadata_ete_ro[] = { [CS_ETE_TRCIDR8] = "trcidr/trcidr8", [CS_ETE_TRCAUTHSTATUS] = "mgmt/trcauthstatus", [CS_ETE_TRCDEVARCH] = "mgmt/trcdevarch", + [CS_ETE_TS_SOURCE] = "ts_source", };
static bool cs_etm_is_etmv4(struct auxtrace_record *itr, int cpu); @@ -613,6 +615,32 @@ static int cs_etm_get_ro(struct perf_pmu *pmu, int cpu, const char *path) return val; }
+static int cs_etm_get_ro_signed(struct perf_pmu *pmu, int cpu, const char *path) +{ + char pmu_path[PATH_MAX]; + int scan; + int val = 0; + + /* Get RO metadata from sysfs */ + snprintf(pmu_path, PATH_MAX, "cpu%d/%s", cpu, path); + + scan = perf_pmu__scan_file(pmu, pmu_path, "%d", &val); + if (scan != 1) + pr_err("%s: error reading: %s\n", __func__, pmu_path); + + return val; +} + +static bool cs_etm_pmu_path_exists(struct perf_pmu *pmu, int cpu, const char *path) +{ + char pmu_path[PATH_MAX]; + + /* Get RO metadata from sysfs */ + snprintf(pmu_path, PATH_MAX, "cpu%d/%s", cpu, path); + + return perf_pmu__file_exists(pmu, pmu_path); +} + #define TRCDEVARCH_ARCHPART_SHIFT 0 #define TRCDEVARCH_ARCHPART_MASK GENMASK(11, 0) #define TRCDEVARCH_ARCHPART(x) (((x) & TRCDEVARCH_ARCHPART_MASK) >> TRCDEVARCH_ARCHPART_SHIFT) @@ -654,6 +682,16 @@ static void cs_etm_save_etmv4_header(__u64 data[], struct auxtrace_record *itr, metadata_etmv4_ro[CS_ETMV4_TRCIDR8]); data[CS_ETMV4_TRCAUTHSTATUS] = cs_etm_get_ro(cs_etm_pmu, cpu, metadata_etmv4_ro[CS_ETMV4_TRCAUTHSTATUS]); + + /* Kernels older than 5.19 may not expose ts_source */ + if (cs_etm_pmu_path_exists(cs_etm_pmu, cpu, metadata_etmv4_ro[CS_ETMV4_TS_SOURCE])) + data[CS_ETMV4_TS_SOURCE] = (__u64) cs_etm_get_ro_signed(cs_etm_pmu, cpu, + metadata_etmv4_ro[CS_ETMV4_TS_SOURCE]); + else { + pr_warning("[%03d] pmu file 'ts_source' not found. Fallback to safe value (-1)\n", + cpu); + data[CS_ETMV4_TS_SOURCE] = (__u64) -1; + } }
static void cs_etm_save_ete_header(__u64 data[], struct auxtrace_record *itr, int cpu) @@ -679,6 +717,16 @@ static void cs_etm_save_ete_header(__u64 data[], struct auxtrace_record *itr, in /* ETE uses the same registers as ETMv4 plus TRCDEVARCH */ data[CS_ETE_TRCDEVARCH] = cs_etm_get_ro(cs_etm_pmu, cpu, metadata_ete_ro[CS_ETE_TRCDEVARCH]); + + /* Kernels older than 5.19 may not expose ts_source */ + if (cs_etm_pmu_path_exists(cs_etm_pmu, cpu, metadata_ete_ro[CS_ETE_TS_SOURCE])) + data[CS_ETE_TS_SOURCE] = (__u64) cs_etm_get_ro_signed(cs_etm_pmu, cpu, + metadata_ete_ro[CS_ETE_TS_SOURCE]); + else { + pr_warning("[%03d] pmu file 'ts_source' not found. Fallback to safe value (-1)\n", + cpu); + data[CS_ETE_TS_SOURCE] = (__u64) -1; + } }
static void cs_etm_get_metadata(int cpu, u32 *offset, diff --git a/tools/perf/util/cs-etm-base.c b/tools/perf/util/cs-etm-base.c index 282042c6e944..5f48b756c4cf 100644 --- a/tools/perf/util/cs-etm-base.c +++ b/tools/perf/util/cs-etm-base.c @@ -36,6 +36,7 @@ static const char * const cs_etmv4_priv_fmts[] = { [CS_ETMV4_TRCIDR2] = " TRCIDR2 %llx\n", [CS_ETMV4_TRCIDR8] = " TRCIDR8 %llx\n", [CS_ETMV4_TRCAUTHSTATUS] = " TRCAUTHSTATUS %llx\n", + [CS_ETMV4_TS_SOURCE] = " TS_SOURCE %lld\n", };
static const char * const cs_ete_priv_fmts[] = { @@ -50,6 +51,7 @@ static const char * const cs_ete_priv_fmts[] = { [CS_ETE_TRCIDR8] = " TRCIDR8 %llx\n", [CS_ETE_TRCAUTHSTATUS] = " TRCAUTHSTATUS %llx\n", [CS_ETE_TRCDEVARCH] = " TRCDEVARCH %llx\n", + [CS_ETE_TS_SOURCE] = " TS_SOURCE %lld\n", };
static const char * const param_unk_fmt = diff --git a/tools/perf/util/cs-etm.h b/tools/perf/util/cs-etm.h index c5925428afa9..ad790930bcbc 100644 --- a/tools/perf/util/cs-etm.h +++ b/tools/perf/util/cs-etm.h @@ -71,6 +71,7 @@ enum { CS_ETMV4_TRCIDR2, CS_ETMV4_TRCIDR8, CS_ETMV4_TRCAUTHSTATUS, + CS_ETMV4_TS_SOURCE, CS_ETMV4_PRIV_MAX, };
@@ -92,6 +93,7 @@ enum { CS_ETE_TRCIDR8, CS_ETE_TRCAUTHSTATUS, CS_ETE_TRCDEVARCH, + CS_ETE_TS_SOURCE, CS_ETE_PRIV_MAX };
From: German Gomez german.gomez@arm.com
If virtual timestamps are detected, set sample time field accordingly, otherwise warn the user that the samples will not include accurate time data.
| Test notes (FEAT_TRF platform) | | $ ./perf record -e cs_etm//u -a -- sleep 4 | $ ./perf script --fields +time | perf 422 [000] 163.375100: 1 branches:uH: 0 [unknown] ([unknown]) | perf 422 [000] 163.375100: 1 branches:uH: ffffb8009544 ioctl+0x14 (/lib/aarch64-linux-gnu/libc-2.27.so) | perf 422 [000] 163.375100: 1 branches:uH: aaaaab6bebf4 perf_evsel__run_ioctl+0x90 (/home/german/linux/tools/perf/perf) | [...] | perf 422 [000] 167.393100: 1 branches:uH: aaaaab6bda00 __xyarray__entry+0x74 (/home/german/linux/tools/perf/perf) | perf 422 [000] 167.393099: 1 branches:uH: aaaaab6bda0c __xyarray__entry+0x80 (/home/german/linux/tools/perf/perf) | perf 422 [000] 167.393099: 1 branches:uH: ffffb8009538 ioctl+0x8 (/lib/aarch64-linux-gnu/libc-2.27.so) | | The time from the first sample to the last sample is 4 seconds
Now that times are converted to nanoseconds, also try to estimate the timestamps more accurately be dividing by some fixed value for instructions per ns. This prevents long ranges from being estimated too far in the past than would be realistic.
Signed-off-by: German Gomez german.gomez@arm.com Signed-off-by: James Clark james.clark@arm.com --- .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 47 +++++++++-- tools/perf/util/cs-etm.c | 83 ++++++++++++++++++- tools/perf/util/cs-etm.h | 3 +- 3 files changed, 120 insertions(+), 13 deletions(-)
diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c index 31fa3b45134a..440fe844ed17 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c @@ -30,6 +30,15 @@ #endif #endif
+/* + * Assume a maximum of 0.1ns elapsed per instruction. This would be the + * case with a theoretical 10GHz core executing 1 instruction per cycle. + * Used to estimate the sample time for synthesized instructions because + * Coresight only emits a timestamp for a range of instructions rather + * than per instruction. + */ +const u32 INSTR_PER_NS = 10; + struct cs_etm_decoder { void *data; void (*packet_printer)(const char *msg); @@ -112,6 +121,20 @@ int cs_etm_decoder__get_packet(struct cs_etm_packet_queue *packet_queue, return 1; }
+/* + * Calculate the number of nanoseconds elapsed. + * + * instr_count is updated in place with the remainder of the instructions + * which didn't make up a whole nanosecond. + */ +static u32 cs_etm_decoder__dec_instr_count_to_ns(u32 *instr_count) +{ + const u32 instr_copy = *instr_count; + + *instr_count %= INSTR_PER_NS; + return instr_copy / INSTR_PER_NS; +} + static int cs_etm_decoder__gen_etmv3_config(struct cs_etm_trace_params *params, ocsd_etmv3_cfg *config) { @@ -267,8 +290,8 @@ cs_etm_decoder__do_soft_timestamp(struct cs_etm_queue *etmq, packet_queue->cs_timestamp = packet_queue->next_cs_timestamp;
/* Estimate the timestamp for the next range packet */ - packet_queue->next_cs_timestamp += packet_queue->instr_count; - packet_queue->instr_count = 0; + packet_queue->next_cs_timestamp += + cs_etm_decoder__dec_instr_count_to_ns(&packet_queue->instr_count);
/* Tell the front end which traceid_queue needs attention */ cs_etm__etmq_set_traceid_queue_timestamp(etmq, trace_chan_id); @@ -283,24 +306,31 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, const ocsd_trc_index_t indx) { struct cs_etm_packet_queue *packet_queue; + u64 converted_timestamp;
/* First get the packet queue for this traceID */ packet_queue = cs_etm__etmq_get_packet_queue(etmq, trace_chan_id); if (!packet_queue) return OCSD_RESP_FATAL_SYS_ERR;
+ /* + * Coresight timestamps are raw timer values which need to be scaled to ns. Assume + * 0 is a bad value so don't try to convert it. + */ + converted_timestamp = elem->timestamp ? + cs_etm__convert_sample_time(etmq, elem->timestamp) : 0; + /* * We've seen a timestamp packet before - simply record the new value. * Function do_soft_timestamp() will report the value to the front end, * hence asking the decoder to keep decoding rather than stopping. */ if (packet_queue->cs_timestamp) { - packet_queue->next_cs_timestamp = elem->timestamp; + packet_queue->next_cs_timestamp = converted_timestamp; return OCSD_RESP_CONT; }
- - if (!elem->timestamp) { + if (!converted_timestamp) { /* * Zero timestamps can be seen due to misconfiguration or hardware bugs. * Warn once, and don't try to subtract instr_count as it would result in an @@ -312,7 +342,7 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, ". Decoding may be improved by prepending 'Z' to your current --itrace arguments.\n", indx);
- } else if (packet_queue->instr_count > elem->timestamp) { + } else if (packet_queue->instr_count / INSTR_PER_NS > converted_timestamp) { /* * Sanity check that the elem->timestamp - packet_queue->instr_count would not * result in an underflow. Warn and clamp at 0 if it would. @@ -327,9 +357,10 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, * which instructions started by subtracting the number of instructions * executed to the timestamp. */ - packet_queue->cs_timestamp = elem->timestamp - packet_queue->instr_count; + packet_queue->cs_timestamp = converted_timestamp - + (packet_queue->instr_count / INSTR_PER_NS); } - packet_queue->next_cs_timestamp = elem->timestamp; + packet_queue->next_cs_timestamp = converted_timestamp; packet_queue->instr_count = 0;
/* Tell the front end which traceid_queue needs attention */ diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c index 879576d5f899..f65bac5ddbdb 100644 --- a/tools/perf/util/cs-etm.c +++ b/tools/perf/util/cs-etm.c @@ -35,6 +35,7 @@ #include "tool.h" #include "thread.h" #include "thread-stack.h" +#include "tsc.h" #include <tools/libc_compat.h> #include "util/synthetic-events.h"
@@ -46,10 +47,12 @@ struct cs_etm_auxtrace { struct perf_session *session; struct machine *machine; struct thread *unknown_thread; + struct perf_tsc_conversion tc;
u8 timeless_decoding; u8 snapshot_mode; u8 data_queued; + u8 has_virtual_ts; /* Virtual/Kernel timestamps in the trace. */
int num_cpu; u64 latest_kernel_timestamp; @@ -1161,6 +1164,30 @@ static void cs_etm__copy_insn(struct cs_etm_queue *etmq, sample->insn_len, (void *)sample->insn); }
+u64 cs_etm__convert_sample_time(struct cs_etm_queue *etmq, u64 cs_timestamp) +{ + struct cs_etm_auxtrace *etm = etmq->etm; + + if (etm->has_virtual_ts) + return tsc_to_perf_time(cs_timestamp, &etm->tc); + else + return cs_timestamp; +} + +static inline u64 cs_etm__resolve_sample_time(struct cs_etm_queue *etmq, + struct cs_etm_traceid_queue *tidq) +{ + struct cs_etm_auxtrace *etm = etmq->etm; + struct cs_etm_packet_queue *packet_queue = &tidq->packet_queue; + + if (etm->timeless_decoding) + return 0; + else if (etm->has_virtual_ts) + return packet_queue->cs_timestamp; + else + return etm->latest_kernel_timestamp; +} + static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq, struct cs_etm_traceid_queue *tidq, u64 addr, u64 period) @@ -1174,8 +1201,9 @@ static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq, event->sample.header.misc = cs_etm__cpu_mode(etmq, addr); event->sample.header.size = sizeof(struct perf_event_header);
- if (!etm->timeless_decoding) - sample.time = etm->latest_kernel_timestamp; + /* Set time field based on etm auxtrace config. */ + sample.time = cs_etm__resolve_sample_time(etmq, tidq); + sample.ip = addr; sample.pid = tidq->pid; sample.tid = tidq->tid; @@ -1232,8 +1260,9 @@ static int cs_etm__synth_branch_sample(struct cs_etm_queue *etmq, event->sample.header.misc = cs_etm__cpu_mode(etmq, ip); event->sample.header.size = sizeof(struct perf_event_header);
- if (!etm->timeless_decoding) - sample.time = etm->latest_kernel_timestamp; + /* Set time field based on etm auxtrace config. */ + sample.time = cs_etm__resolve_sample_time(etmq, tidq); + sample.ip = ip; sample.pid = tidq->pid; sample.tid = tidq->tid; @@ -2746,12 +2775,42 @@ static int cs_etm__queue_aux_records(struct perf_session *session) return 0; }
+#define HAS_PARAM(j, type, param) (metadata[(j)][CS_ETM_NR_TRC_PARAMS] <= \ + (CS_##type##_##param - CS_ETM_COMMON_BLK_MAX_V1)) + +/* + * Loop through the ETMs and complain if we find at least one where ts_source != 1 (virtual + * timestamps). + */ +static bool cs_etm__has_virtual_ts(u64 **metadata, int num_cpu) +{ + int j; + + for (j = 0; j < num_cpu; j++) { + switch (metadata[j][CS_ETM_MAGIC]) { + case __perf_cs_etmv4_magic: + if (HAS_PARAM(j, ETMV4, TS_SOURCE) || metadata[j][CS_ETMV4_TS_SOURCE] != 1) + return false; + break; + case __perf_cs_ete_magic: + if (HAS_PARAM(j, ETE, TS_SOURCE) || metadata[j][CS_ETE_TS_SOURCE] != 1) + return false; + break; + default: + /* Unknown / unsupported magic number. */ + return false; + } + } + return true; +} + int cs_etm__process_auxtrace_info_full(union perf_event *event, struct perf_session *session) { struct perf_record_auxtrace_info *auxtrace_info = &event->auxtrace_info; struct cs_etm_auxtrace *etm = NULL; struct int_node *inode; + struct perf_record_time_conv *tc = &session->time_conv; int event_header_size = sizeof(struct perf_event_header); int total_size = auxtrace_info->header.size; int priv_size = 0; @@ -2886,6 +2945,13 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event, etm->auxtrace_type = auxtrace_info->type; etm->timeless_decoding = cs_etm__is_timeless_decoding(etm);
+ /* Use virtual timestamps if all ETMs report ts_source = 1 */ + etm->has_virtual_ts = cs_etm__has_virtual_ts(metadata, num_cpu); + + if (!etm->has_virtual_ts) + ui__warning("Virtual timestamps are not enabled, or not supported by the traced system.\n" + "The time field of the samples will not be set accurately.\n\n"); + etm->auxtrace.process_event = cs_etm__process_event; etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event; etm->auxtrace.flush_events = cs_etm__flush_events; @@ -2915,6 +2981,15 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event, goto err_delete_thread; }
+ etm->tc.time_shift = tc->time_shift; + etm->tc.time_mult = tc->time_mult; + etm->tc.time_zero = tc->time_zero; + if (event_contains(*tc, time_cycles)) { + etm->tc.time_cycles = tc->time_cycles; + etm->tc.time_mask = tc->time_mask; + etm->tc.cap_user_time_zero = tc->cap_user_time_zero; + etm->tc.cap_user_time_short = tc->cap_user_time_short; + } err = cs_etm__synth_events(etm, session); if (err) goto err_delete_thread; diff --git a/tools/perf/util/cs-etm.h b/tools/perf/util/cs-etm.h index ad790930bcbc..98a4f7113d2f 100644 --- a/tools/perf/util/cs-etm.h +++ b/tools/perf/util/cs-etm.h @@ -192,7 +192,7 @@ struct cs_etm_packet_queue { u32 head; u32 tail; u32 instr_count; - u64 cs_timestamp; + u64 cs_timestamp; /* Timestamp from trace data, converted to ns if possible */ u64 next_cs_timestamp; struct cs_etm_packet packet_buffer[CS_ETM_PACKET_MAX_BUFFER]; }; @@ -231,6 +231,7 @@ struct cs_etm_packet_queue *cs_etm__etmq_get_packet_queue(struct cs_etm_queue *etmq, u8 trace_chan_id); int cs_etm__process_auxtrace_info_full(union perf_event *event __maybe_unused, struct perf_session *session __maybe_unused); +u64 cs_etm__convert_sample_time(struct cs_etm_queue *etmq, u64 cs_timestamp); #else static inline int cs_etm__process_auxtrace_info_full(union perf_event *event __maybe_unused,
There are some edge cases around estimated timestamps that can result in them going backwards.
One is that after a discontinuity, the last used timestamp is set to 0. The duration of the next range is then subtracted which could result in an earlier timestamp than the last instruction. Fix this by not resetting the last timestamp used on a discontinuity, and make sure that new estimated timestamps are clamped to be later than that.
Another case is that estimated timestamps could compound over time to end up being more than the next real timestamp in the trace. Fix this by clamping the estimates in cs_etm_decoder__do_soft_timestamp() to be no later than it.
cs_etm_decoder__do_soft_timestamp() also updated next_cs_timestamp, which meant that the next real timestamp was lost and not stored anywhere. Fix that by only updating cs_timestamp for estimates and keep next_cs_timestamp untouched.
Finally, use next_cs_timestamp to signify if a timestamp has been received previously. Because cs_timestamp has the first range subtracted, it could technically go to 0 which would break the logic.
Testing =======
It can be verified that timestamps don't go backwards when tracing on a single core with the following commands. Across multiple cores it's expected that timestamps are interleaved:
$ perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 $ perf script --itrace=i1ns --ns -Fcomm,tid,pid,time,cpu,event,ip,sym,addr,symoff,flags,callindent > itrace $ sed 's/://g' itrace | awk -F ' ' ' { print $4 } ' | awk '{ if ($1 < prev) { print "line:" NR " " $0 } {prev=$1}}'
Reported-by: Tanmay Jagdale tanmay@marvell.com Signed-off-by: James Clark james.clark@arm.com --- .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 29 ++++++++++++------- 1 file changed, 19 insertions(+), 10 deletions(-)
diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c index 440fe844ed17..63afa2d05b46 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c @@ -283,15 +283,17 @@ cs_etm_decoder__do_soft_timestamp(struct cs_etm_queue *etmq, struct cs_etm_packet_queue *packet_queue, const uint8_t trace_chan_id) { + u64 estimated_ts; + /* No timestamp packet has been received, nothing to do */ - if (!packet_queue->cs_timestamp) + if (!packet_queue->next_cs_timestamp) return OCSD_RESP_CONT;
- packet_queue->cs_timestamp = packet_queue->next_cs_timestamp; + estimated_ts = packet_queue->cs_timestamp + + cs_etm_decoder__dec_instr_count_to_ns(&packet_queue->instr_count);
- /* Estimate the timestamp for the next range packet */ - packet_queue->next_cs_timestamp += - cs_etm_decoder__dec_instr_count_to_ns(&packet_queue->instr_count); + /* Estimated TS can never be higher than the next real one in the trace */ + packet_queue->cs_timestamp = min(packet_queue->next_cs_timestamp, estimated_ts);
/* Tell the front end which traceid_queue needs attention */ cs_etm__etmq_set_traceid_queue_timestamp(etmq, trace_chan_id); @@ -307,6 +309,7 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, { struct cs_etm_packet_queue *packet_queue; u64 converted_timestamp; + u64 estimated_first_ts;
/* First get the packet queue for this traceID */ packet_queue = cs_etm__etmq_get_packet_queue(etmq, trace_chan_id); @@ -325,7 +328,12 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, * Function do_soft_timestamp() will report the value to the front end, * hence asking the decoder to keep decoding rather than stopping. */ - if (packet_queue->cs_timestamp) { + if (packet_queue->next_cs_timestamp) { + /* + * What was next is now where new ranges start from, overwriting + * any previous estimate in cs_timestamp + */ + packet_queue->cs_timestamp = packet_queue->next_cs_timestamp; packet_queue->next_cs_timestamp = converted_timestamp; return OCSD_RESP_CONT; } @@ -355,10 +363,12 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, * or a discontinuity. Since timestamps packets are generated *after* * range packets have been generated, we need to estimate the time at * which instructions started by subtracting the number of instructions - * executed to the timestamp. + * executed to the timestamp. Don't estimate earlier than the last used + * timestamp though. */ - packet_queue->cs_timestamp = converted_timestamp - - (packet_queue->instr_count / INSTR_PER_NS); + estimated_first_ts = converted_timestamp - + (packet_queue->instr_count / INSTR_PER_NS); + packet_queue->cs_timestamp = max(packet_queue->cs_timestamp, estimated_first_ts); } packet_queue->next_cs_timestamp = converted_timestamp; packet_queue->instr_count = 0; @@ -373,7 +383,6 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, static void cs_etm_decoder__reset_timestamp(struct cs_etm_packet_queue *packet_queue) { - packet_queue->cs_timestamp = 0; packet_queue->next_cs_timestamp = 0; packet_queue->instr_count = 0; }
Em Fri, Jan 20, 2023 at 02:37:01PM +0000, James Clark escreveu:
There are some edge cases around estimated timestamps that can result in them going backwards.
One is that after a discontinuity, the last used timestamp is set to 0. The duration of the next range is then subtracted which could result in an earlier timestamp than the last instruction. Fix this by not resetting the last timestamp used on a discontinuity, and make sure that new estimated timestamps are clamped to be later than that.
Another case is that estimated timestamps could compound over time to end up being more than the next real timestamp in the trace. Fix this by clamping the estimates in cs_etm_decoder__do_soft_timestamp() to be no later than it.
cs_etm_decoder__do_soft_timestamp() also updated next_cs_timestamp, which meant that the next real timestamp was lost and not stored anywhere. Fix that by only updating cs_timestamp for estimates and keep next_cs_timestamp untouched.
Finally, use next_cs_timestamp to signify if a timestamp has been received previously. Because cs_timestamp has the first range subtracted, it could technically go to 0 which would break the logic.
Testing
It can be verified that timestamps don't go backwards when tracing on a single core with the following commands. Across multiple cores it's expected that timestamps are interleaved:
$ perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 $ perf script --itrace=i1ns --ns -Fcomm,tid,pid,time,cpu,event,ip,sym,addr,symoff,flags,callindent > itrace $ sed 's/://g' itrace | awk -F ' ' ' { print $4 } ' | awk '{ if ($1 < prev) { print "line:" NR " " $0 } {prev=$1}}'
Trying:
root@roc-rk3399-pc:~# uname -a Linux roc-rk3399-pc 6.1.0-rc5-00123-g4dd7ff4a0311 #2 SMP PREEMPT Wed Nov 16 19:55:11 UTC 2022 aarch64 aarch64 aarch64 GNU/Linux root@roc-rk3399-pc:~# root@roc-rk3399-pc:~# perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 failed to set sink "tmc_etr0" on event cs_etm/@tmc_etr0/k with 2 (No such file or directory) root@roc-rk3399-pc:~#
We could have a better message at some point, right? :-)
Something like:
root@roc-rk3399-pc:~# perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 This system lacks the CoreSight component. root@roc-rk3399-pc:~#
Anyway, applied the patches and test building now.
- Arnaldo
Reported-by: Tanmay Jagdale tanmay@marvell.com Signed-off-by: James Clark james.clark@arm.com
.../perf/util/cs-etm-decoder/cs-etm-decoder.c | 29 ++++++++++++------- 1 file changed, 19 insertions(+), 10 deletions(-)
diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c index 440fe844ed17..63afa2d05b46 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c @@ -283,15 +283,17 @@ cs_etm_decoder__do_soft_timestamp(struct cs_etm_queue *etmq, struct cs_etm_packet_queue *packet_queue, const uint8_t trace_chan_id) {
- u64 estimated_ts;
- /* No timestamp packet has been received, nothing to do */
- if (!packet_queue->cs_timestamp)
- if (!packet_queue->next_cs_timestamp) return OCSD_RESP_CONT;
- packet_queue->cs_timestamp = packet_queue->next_cs_timestamp;
- estimated_ts = packet_queue->cs_timestamp +
cs_etm_decoder__dec_instr_count_to_ns(&packet_queue->instr_count);
- /* Estimate the timestamp for the next range packet */
- packet_queue->next_cs_timestamp +=
cs_etm_decoder__dec_instr_count_to_ns(&packet_queue->instr_count);
- /* Estimated TS can never be higher than the next real one in the trace */
- packet_queue->cs_timestamp = min(packet_queue->next_cs_timestamp, estimated_ts);
/* Tell the front end which traceid_queue needs attention */ cs_etm__etmq_set_traceid_queue_timestamp(etmq, trace_chan_id); @@ -307,6 +309,7 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, { struct cs_etm_packet_queue *packet_queue; u64 converted_timestamp;
- u64 estimated_first_ts;
/* First get the packet queue for this traceID */ packet_queue = cs_etm__etmq_get_packet_queue(etmq, trace_chan_id); @@ -325,7 +328,12 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, * Function do_soft_timestamp() will report the value to the front end, * hence asking the decoder to keep decoding rather than stopping. */
- if (packet_queue->cs_timestamp) {
- if (packet_queue->next_cs_timestamp) {
/** What was next is now where new ranges start from, overwriting* any previous estimate in cs_timestamp*/packet_queue->cs_timestamp = packet_queue->next_cs_timestamp;@@ -355,10 +363,12 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, * or a discontinuity. Since timestamps packets are generated *after* * range packets have been generated, we need to estimate the time at * which instructions started by subtracting the number of instructions
* executed to the timestamp.
* executed to the timestamp. Don't estimate earlier than the last used* timestamp though.
packet_queue->cs_timestamp = converted_timestamp -(packet_queue->instr_count / INSTR_PER_NS);
estimated_first_ts = converted_timestamp -(packet_queue->instr_count / INSTR_PER_NS);packet_queue->cs_timestamp = max(packet_queue->cs_timestamp, estimated_first_ts);@@ -373,7 +383,6 @@ cs_etm_decoder__do_hard_timestamp(struct cs_etm_queue *etmq, static void cs_etm_decoder__reset_timestamp(struct cs_etm_packet_queue *packet_queue) {
- packet_queue->cs_timestamp = 0; packet_queue->next_cs_timestamp = 0; packet_queue->instr_count = 0;
}
2.25.1
On 20/01/2023 17:12, Arnaldo Carvalho de Melo wrote:
Em Fri, Jan 20, 2023 at 02:37:01PM +0000, James Clark escreveu:
There are some edge cases around estimated timestamps that can result in them going backwards.
One is that after a discontinuity, the last used timestamp is set to 0. The duration of the next range is then subtracted which could result in an earlier timestamp than the last instruction. Fix this by not resetting the last timestamp used on a discontinuity, and make sure that new estimated timestamps are clamped to be later than that.
Another case is that estimated timestamps could compound over time to end up being more than the next real timestamp in the trace. Fix this by clamping the estimates in cs_etm_decoder__do_soft_timestamp() to be no later than it.
cs_etm_decoder__do_soft_timestamp() also updated next_cs_timestamp, which meant that the next real timestamp was lost and not stored anywhere. Fix that by only updating cs_timestamp for estimates and keep next_cs_timestamp untouched.
Finally, use next_cs_timestamp to signify if a timestamp has been received previously. Because cs_timestamp has the first range subtracted, it could technically go to 0 which would break the logic.
Testing
It can be verified that timestamps don't go backwards when tracing on a single core with the following commands. Across multiple cores it's expected that timestamps are interleaved:
$ perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 $ perf script --itrace=i1ns --ns -Fcomm,tid,pid,time,cpu,event,ip,sym,addr,symoff,flags,callindent > itrace $ sed 's/://g' itrace | awk -F ' ' ' { print $4 } ' | awk '{ if ($1 < prev) { print "line:" NR " " $0 } {prev=$1}}'
Trying:
root@roc-rk3399-pc:~# uname -a Linux roc-rk3399-pc 6.1.0-rc5-00123-g4dd7ff4a0311 #2 SMP PREEMPT Wed Nov 16 19:55:11 UTC 2022 aarch64 aarch64 aarch64 GNU/Linux root@roc-rk3399-pc:~# root@roc-rk3399-pc:~# perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 failed to set sink "tmc_etr0" on event cs_etm/@tmc_etr0/k with 2 (No such file or directory) root@roc-rk3399-pc:~#
We could have a better message at some point, right? :-)
Something like:
root@roc-rk3399-pc:~# perf record -e cs_etm/@tmc_etr0/k -C 4 taskset -c 4 sleep 1 This system lacks the CoreSight component. root@roc-rk3399-pc:~#
Should be possible, I'm having a look now
On 20/01/2023 14:36, James Clark wrote:
Changes since v4:
- Rebase onto perf/core
- Convert new perf_pmu__cpu_slots_per_cycle() function to use new helper functions
Changes since v3:
- Scale time estimates by INSTR_PER_NS, rather than assuming 1 instruction = 1ns
- Add a new commit that fixes some issues around timestamps going backwards
- Use nanoseconds inside cs-etm-decoder.c, rather than storing the raw time values and converting when a sample is synthesized. This simplifies some of the code like estimating the first timestamp.
Changes since v2:
- Remove const to non-const change and copy strings where needed instead.
- Use sizeof() instead of PATH_MAX
- Append "will not be set accurately." to new error message
- Remove unneeded stat() call
- Rebase on perf/core
Changes since v1:
- Add 3 refactor commits for sysfs reading around pmu.c as suggested by Arnaldo here [1]
- The dependency on [2] has now reached mainline so is no longer blocking
- Rebase on perf/core
[1]: https://lore.kernel.org/all/YnqVqq5QW%2Fb14oPZ@kernel.org/ [2]: https://lore.kernel.org/all/20220503123537.1003035-1-german.gomez@arm.com/
German Gomez (4): perf pmu: Add function to check if a pmu file exists perf cs_etm: Keep separate symbols for ETMv4 and ETE parameters perf cs_etm: Record ts_source in AUXTRACE_INFO for ETMv4 and ETE perf cs_etm: Set the time field in the synthetic samples
James Clark (4): perf: Remove duplication around EVENT_SOURCE_DEVICE_PATH perf: Use perf_pmu__open_file() and perf_pmu__scan_file() perf: Remove remaining duplication of bus/event_source/devices/... perf: cs-etm: Ensure that Coresight timestamps don't go backwards
tools/perf/arch/arm/util/auxtrace.c | 5 +- tools/perf/arch/arm/util/cs-etm.c | 91 ++++++++- tools/perf/arch/arm64/util/pmu.c | 4 +- tools/perf/arch/x86/util/pmu.c | 12 +- tools/perf/util/cputopo.c | 9 +- tools/perf/util/cs-etm-base.c | 34 +++- .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 68 +++++-- tools/perf/util/cs-etm.c | 95 +++++++++- tools/perf/util/cs-etm.h | 16 +- tools/perf/util/pmu-hybrid.c | 27 +-- tools/perf/util/pmu.c | 177 +++++++----------- tools/perf/util/pmu.h | 10 +- 12 files changed, 351 insertions(+), 197 deletions(-)
For the series,
Acked-by: Suzuki K Poulose suzuki.poulose@arm.com
base-commit: 1962ab6f6e0b39e4216206205bda14aff87705f3 prerequisite-patch-id: 9722bf86e3e6d16d177ff9a1411992a795a7dcbd prerequisite-patch-id: b05dbef439c2ea8465f3321532257b0ca29f21f9 prerequisite-patch-id: 92680a4781cbcf010fcb007e6ea030f59e9eaefc prerequisite-patch-id: 8e3a73a04e4b89b503377b5fac1d89d551159393 prerequisite-patch-id: 09980d8fedcdaa70b220a7802428109f48448a58
Hi James,
Changes since v4:
- Rebase onto perf/core
- Convert new perf_pmu__cpu_slots_per_cycle() function to use new helper functions
=========================== Changes since v3:
- Scale time estimates by INSTR_PER_NS, rather than assuming 1 instruction = 1ns
- Add a new commit that fixes some issues around timestamps going backwards
- Use nanoseconds inside cs-etm-decoder.c, rather than storing the raw time values and converting when a sample is synthesized. This simplifies some of the code like estimating the first timestamp.
=========================== Changes since v2:
- Remove const to non-const change and copy strings where needed instead.
- Use sizeof() instead of PATH_MAX
- Append "will not be set accurately." to new error message
- Remove unneeded stat() call
- Rebase on perf/core
==========================
Changes since v1:
- Add 3 refactor commits for sysfs reading around pmu.c as suggested by Arnaldo here [1]
- The dependency on [2] has now reached mainline so is no longer blocking
- Rebase on perf/core
German Gomez (4): perf pmu: Add function to check if a pmu file exists perf cs_etm: Keep separate symbols for ETMv4 and ETE parameters perf cs_etm: Record ts_source in AUXTRACE_INFO for ETMv4 and ETE perf cs_etm: Set the time field in the synthetic samples
James Clark (4): perf: Remove duplication around EVENT_SOURCE_DEVICE_PATH perf: Use perf_pmu__open_file() and perf_pmu__scan_file() perf: Remove remaining duplication of bus/event_source/devices/... perf: cs-etm: Ensure that Coresight timestamps don't go backwards
tools/perf/arch/arm/util/auxtrace.c | 5 +- tools/perf/arch/arm/util/cs-etm.c | 91 ++++++++- tools/perf/arch/arm64/util/pmu.c | 4 +- tools/perf/arch/x86/util/pmu.c | 12 +- tools/perf/util/cputopo.c | 9 +- tools/perf/util/cs-etm-base.c | 34 +++- .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 68 +++++-- tools/perf/util/cs-etm.c | 95 +++++++++- tools/perf/util/cs-etm.h | 16 +- tools/perf/util/pmu-hybrid.c | 27 +-- tools/perf/util/pmu.c | 177 +++++++----------- tools/perf/util/pmu.h | 10 +- 12 files changed, 351 insertions(+), 197 deletions(-)
base-commit: 1962ab6f6e0b39e4216206205bda14aff87705f3 prerequisite-patch-id: 9722bf86e3e6d16d177ff9a1411992a795a7dcbd prerequisite-patch-id: b05dbef439c2ea8465f3321532257b0ca29f21f9 prerequisite-patch-id: 92680a4781cbcf010fcb007e6ea030f59e9eaefc prerequisite-patch-id: 8e3a73a04e4b89b503377b5fac1d89d551159393 prerequisite-patch-id: 09980d8fedcdaa70b220a7802428109f48448a58
For the series,
Tested-by: Tanmay Jagdale tanmay@marvell.com
Thanks, Tanmay
-- 2.25.1
-
Arnaldo Carvalho de Melo -
James Clark -
Suzuki K Poulose -
Tanmay Jagdale