Commit f0ff2447b861 ("selftests/sgx: Add a new kselftest: Unclobbered_vdso_oversubscribed") depends on __cpuid() without providing the dependency and thus introduces a build error:
$ make gcc -Wall -Werror -g -I../../../../tools/include -fPIC -z noexecstack -c main.c -o ../linux/tools/testing/selftests/sgx/main.o main.c: In function ‘get_total_epc_mem’: main.c:296:3: error: implicit declaration of function ‘__cpuid’ [-Werror=implicit-function-declaration] 296 | __cpuid(&eax, &ebx, &ecx, &edx); | ^~~~~~~ cc1: all warnings being treated as errors make: *** [Makefile:33: ../linux/tools/testing/selftests/sgx/main.o] Error 1 $
Clone kernel's __cpuid() implementation to the self-test in order to make it available to the EPC enumeration code.
Fixes: f0ff2447b861 ("selftests/sgx: Add a new kselftest: Unclobbered_vdso_oversubscribed") Reported-by: Jarkko Sakkinen jarkko@kernel.org Signed-off-by: Reinette Chatre reinette.chatre@intel.com ---
The commit introducing the issue can be found on the x86/sgx branch of tip.git.
Changes since V1: - V1: https://lore.kernel.org/linux-sgx/797ff1331cfe540fc378fcc4a4a7b00ff5099fbe.1... - Improve commit message. (Jarkko)
tools/testing/selftests/sgx/main.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/tools/testing/selftests/sgx/main.c b/tools/testing/selftests/sgx/main.c index 7e912db4c6c5..6dead57a3121 100644 --- a/tools/testing/selftests/sgx/main.c +++ b/tools/testing/selftests/sgx/main.c @@ -73,6 +73,18 @@ static bool vdso_get_symtab(void *addr, struct vdso_symtab *symtab) return true; }
+static inline void __cpuid(unsigned int *eax, unsigned int *ebx, + unsigned int *ecx, unsigned int *edx) +{ + asm volatile("cpuid" + : "=a" (*eax), + "=b" (*ebx), + "=c" (*ecx), + "=d" (*edx) + : "0" (*eax), "2" (*ecx) + : "memory"); +} + static unsigned long elf_sym_hash(const char *name) { unsigned long h = 0, high;
On 12/14/21 12:26 PM, Reinette Chatre wrote:
Commit f0ff2447b861 ("selftests/sgx: Add a new kselftest: Unclobbered_vdso_oversubscribed") depends on __cpuid() without providing the dependency and thus introduces a build error:
$ make gcc -Wall -Werror -g -I../../../../tools/include -fPIC -z noexecstack -c main.c -o ../linux/tools/testing/selftests/sgx/main.o main.c: In function ‘get_total_epc_mem’: main.c:296:3: error: implicit declaration of function ‘__cpuid’ [-Werror=implicit-function-declaration] 296 | __cpuid(&eax, &ebx, &ecx, &edx); | ^~~~~~~ cc1: all warnings being treated as errors make: *** [Makefile:33: ../linux/tools/testing/selftests/sgx/main.o] Error 1 $
Clone kernel's __cpuid() implementation to the self-test in order to make it available to the EPC enumeration code.
Fixes: f0ff2447b861 ("selftests/sgx: Add a new kselftest: Unclobbered_vdso_oversubscribed") Reported-by: Jarkko Sakkinen jarkko@kernel.org Signed-off-by: Reinette Chatre reinette.chatre@intel.com
The commit introducing the issue can be found on the x86/sgx branch of tip.git.
Changes since V1:
V1: https://lore.kernel.org/linux-sgx/797ff1331cfe540fc378fcc4a4a7b00ff5099fbe.1...
Improve commit message. (Jarkko)
tools/testing/selftests/sgx/main.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/tools/testing/selftests/sgx/main.c b/tools/testing/selftests/sgx/main.c index 7e912db4c6c5..6dead57a3121 100644 --- a/tools/testing/selftests/sgx/main.c +++ b/tools/testing/selftests/sgx/main.c @@ -73,6 +73,18 @@ static bool vdso_get_symtab(void *addr, struct vdso_symtab *symtab) return true; } +static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
unsigned int *ecx, unsigned int *edx)+{
- asm volatile("cpuid"
: "=a" (*eax),"=b" (*ebx),"=c" (*ecx),"=d" (*edx): "0" (*eax), "2" (*ecx): "memory");+}
- static unsigned long elf_sym_hash(const char *name) { unsigned long h = 0, high;
Let's not add one more __cpuid() define to the individual tests. We so far have:
tools/testing/selftests/vm/pkey-x86.h selftests/x86/corrupt_xstate_header.c
Let's move the defines to kselftest.h and remove all these duplicate defines.
For now you could include vm/pkey-x86.h just to fix the build error and do the proper cleanup.
There has been a proliferation of defines in tools and selftests. I just fixed/remove 25+ ARRAY_SIZE duplicate define in selftests.
thanks, -- Shuah
Hi Shuah,
On 12/14/2021 12:10 PM, Shuah Khan wrote:
Let's not add one more __cpuid() define to the individual tests. We so far have:
tools/testing/selftests/vm/pkey-x86.h selftests/x86/corrupt_xstate_header.c
Let's move the defines to kselftest.h and remove all these duplicate defines.
For now you could include vm/pkey-x86.h just to fix the build error and do the proper cleanup.
Thank you so much for taking a look. We actually do have an alternative fix that could be considered for the other users of __cpuid(). Instead of another clone of the kernel's __cpuid() the fix includes cpuid.h and uses the existing __cpuid_count() from it.
Please see: https://lore.kernel.org/linux-sgx/20211204202355.23005-1-jarkko@kernel.org/
We decided against the above fix using __cpuid_count() because we could not explain why all the existing users of __cpuid() implement their own and decided to follow the custom instead ...
Do you see any problem with including cpuid.h into a selftest? If not, then we can go back to our original fix of this issue and I could also submit a change to remove all the __cpuid() clones and replace them with the library's __cpuid_count().
Reinette
On 12/14/21 2:28 PM, Reinette Chatre wrote:
Hi Shuah,
On 12/14/2021 12:10 PM, Shuah Khan wrote:
Let's not add one more __cpuid() define to the individual tests. We so far have:
tools/testing/selftests/vm/pkey-x86.h selftests/x86/corrupt_xstate_header.c
Let's move the defines to kselftest.h and remove all these duplicate defines.
For now you could include vm/pkey-x86.h just to fix the build error and do the proper cleanup.
Thank you so much for taking a look. We actually do have an alternative fix that could be considered for the other users of __cpuid(). Instead of another clone of the kernel's __cpuid() the fix includes cpuid.h and uses the existing __cpuid_count() from it.
Please see: https://lore.kernel.org/linux-sgx/20211204202355.23005-1-jarkko@kernel.org/
This looks good to me.
We decided against the above fix using __cpuid_count() because we could not explain why all the existing users of __cpuid() implement their own and decided to follow the custom instead ...
Most likely, copy and paste and taking the easy route. I looked at a couple of defines and I dont't see a difference. If there are differences, that would be a bigger concern. The whole idea of not having duplicates is that we don't have bugs in these duplicates.
Do you see any problem with including cpuid.h into a selftest? If not, then we can go back to our original fix of this issue and I could also submit a change to remove all the __cpuid() clones and replace them with the library's __cpuid_count().
There are no problems including cpuid.h - where is this though. I couldn't find it in my cscope search?
thanks, -- Shuah
Hi Shuah,
On 12/14/2021 1:55 PM, Shuah Khan wrote:
On 12/14/21 2:28 PM, Reinette Chatre wrote:
Hi Shuah,
On 12/14/2021 12:10 PM, Shuah Khan wrote:
Let's not add one more __cpuid() define to the individual tests. We so far have:
tools/testing/selftests/vm/pkey-x86.h selftests/x86/corrupt_xstate_header.c
Let's move the defines to kselftest.h and remove all these duplicate defines.
For now you could include vm/pkey-x86.h just to fix the build error and do the proper cleanup.
Thank you so much for taking a look. We actually do have an alternative fix that could be considered for the other users of __cpuid(). Instead of another clone of the kernel's __cpuid() the fix includes cpuid.h and uses the existing __cpuid_count() from it.
Please see: https://lore.kernel.org/linux-sgx/20211204202355.23005-1-jarkko@kernel.org/
This looks good to me.
Thank you very much for taking a look.
We decided against the above fix using __cpuid_count() because we could not explain why all the existing users of __cpuid() implement their own and decided to follow the custom instead ...
Most likely, copy and paste and taking the easy route. I looked at a couple of defines and I dont't see a difference. If there are differences, that would be a bigger concern. The whole idea of not having duplicates is that we don't have bugs in these duplicates.
Do you see any problem with including cpuid.h into a selftest? If not, then we can go back to our original fix of this issue and I could also submit a change to remove all the __cpuid() clones and replace them with the library's __cpuid_count().
There are no problems including cpuid.h - where is this though. I couldn't find it in my cscope search?
On my system it arrived via user space's libgcc-dev package. This does not seem to be the first time including files from this source - I did a quick check and from what I can tell existing includes like stdarg.h, stdbool.h, stdatomic.h, unwind.h, x86intrin.h ... arrive via libgcc-dev.
Do you still find that cpuid.h is ok to include?
Reinette
On 12/14/21 3:27 PM, Reinette Chatre wrote:
Hi Shuah,
On 12/14/2021 1:55 PM, Shuah Khan wrote:
On 12/14/21 2:28 PM, Reinette Chatre wrote:
Hi Shuah,
On 12/14/2021 12:10 PM, Shuah Khan wrote:
Let's not add one more __cpuid() define to the individual tests. We so far have:
tools/testing/selftests/vm/pkey-x86.h selftests/x86/corrupt_xstate_header.c
Let's move the defines to kselftest.h and remove all these duplicate defines.
For now you could include vm/pkey-x86.h just to fix the build error and do the proper cleanup.
Thank you so much for taking a look. We actually do have an alternative fix that could be considered for the other users of __cpuid(). Instead of another clone of the kernel's __cpuid() the fix includes cpuid.h and uses the existing __cpuid_count() from it.
Please see: https://lore.kernel.org/linux-sgx/20211204202355.23005-1-jarkko@kernel.org/
This looks good to me.
Thank you very much for taking a look.
We decided against the above fix using __cpuid_count() because we could not explain why all the existing users of __cpuid() implement their own and decided to follow the custom instead ...
Most likely, copy and paste and taking the easy route. I looked at a couple of defines and I dont't see a difference. If there are differences, that would be a bigger concern. The whole idea of not having duplicates is that we don't have bugs in these duplicates.
Do you see any problem with including cpuid.h into a selftest? If not, then we can go back to our original fix of this issue and I could also submit a change to remove all the __cpuid() clones and replace them with the library's __cpuid_count().
There are no problems including cpuid.h - where is this though. I couldn't find it in my cscope search?
On my system it arrived via user space's libgcc-dev package. This does not seem to be the first time including files from this source - I did a quick check and from what I can tell existing includes like stdarg.h, stdbool.h, stdatomic.h, unwind.h, x86intrin.h ... arrive via libgcc-dev.
Do you still find that cpuid.h is ok to include?
Reinette
Yes. I found it on my system as well.
/usr/lib/llvm-13/lib/clang/13.0.0/include/cpuid.h /usr/lib/llvm-12/lib/clang/12.0.1/include/cpuid.h /usr/lib/gcc/x86_64-linux-gnu/11/include/cpuid.h /usr/lib/gcc/x86_64-linux-gnu/10/include/cpuid.h
No problems using it.
thanks, -- Shuah
linux-kselftest-mirror@lists.linaro.org
-
Reinette Chatre -
Shuah Khan