| Message ID | 20230912-kselftest-param_test-c-v1-1-80a6cffc7374@google.com (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | selftests/rseq: fix kselftest Clang build warnings | expand |
Ping. Looking to get this patch and [1] slated for 6.7 which fixes some kselftest builds on older kernels. On Wed, Sep 13, 2023 at 6:03 AM Justin Stitt <justinstitt@google.com> wrote: > > When building with Clang, I am getting many warnings from the selftests/rseq tree. > > Here's one such example from rseq tree: > | param_test.c:1234:10: error: address argument to atomic operation must be a pointer to _Atomic type ('intptr_t *' (aka 'long *') invalid) > | 1234 | while (!atomic_load(&args->percpu_list_ptr)) {} > | | ^ ~~~~~~~~~~~~~~~~~~~~~~ > | /usr/local/google/home/justinstitt/repos/tc-build/build/llvm/final/lib/clang/18/include/stdatomic.h:140:29: note: expanded from macro 'atomic_load' > | 140 | #define atomic_load(object) __c11_atomic_load(object, __ATOMIC_SEQ_CST) > | | ^ ~~~~~~ > > Use compiler builtins `__atomic_load_n()` and `__atomic_store_n()` with > accompanying __ATOMIC_ACQUIRE and __ATOMIC_RELEASE, respectively. This > will fix the warnings because the compiler builtins do not expect their > arguments to have _Atomic type. This should also make TSAN happier. > > Link: https://github.com/ClangBuiltLinux/linux/issues/1698 > Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/61 > Suggested-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> > Signed-off-by: Justin Stitt <justinstitt@google.com> > --- > Note: Previous RFC https://lore.kernel.org/r/20230908-kselftest-param_test-c-v1-1-e35bd9052d61@google.com > --- > tools/testing/selftests/rseq/param_test.c | 20 ++++++++++---------- > 1 file changed, 10 insertions(+), 10 deletions(-) > > diff --git a/tools/testing/selftests/rseq/param_test.c b/tools/testing/selftests/rseq/param_test.c > index bf951a490bb4..20403d58345c 100644 > --- a/tools/testing/selftests/rseq/param_test.c > +++ b/tools/testing/selftests/rseq/param_test.c > @@ -1231,7 +1231,7 @@ void *test_membarrier_worker_thread(void *arg) > } > > /* Wait for initialization. */ > - while (!atomic_load(&args->percpu_list_ptr)) {} > + while (!__atomic_load_n(&args->percpu_list_ptr, __ATOMIC_ACQUIRE)) {} > > for (i = 0; i < iters; ++i) { > int ret; > @@ -1299,22 +1299,22 @@ void *test_membarrier_manager_thread(void *arg) > test_membarrier_init_percpu_list(&list_a); > test_membarrier_init_percpu_list(&list_b); > > - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); > + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); > > - while (!atomic_load(&args->stop)) { > + while (!__atomic_load_n(&args->stop, __ATOMIC_ACQUIRE)) { > /* list_a is "active". */ > cpu_a = rand() % CPU_SETSIZE; > /* > * As list_b is "inactive", we should never see changes > * to list_b. > */ > - if (expect_b != atomic_load(&list_b.c[cpu_b].head->data)) { > + if (expect_b != __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE)) { > fprintf(stderr, "Membarrier test failed\n"); > abort(); > } > > /* Make list_b "active". */ > - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_b); > + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_b, __ATOMIC_RELEASE); > if (rseq_membarrier_expedited(cpu_a) && > errno != ENXIO /* missing CPU */) { > perror("sys_membarrier"); > @@ -1324,27 +1324,27 @@ void *test_membarrier_manager_thread(void *arg) > * Cpu A should now only modify list_b, so the values > * in list_a should be stable. > */ > - expect_a = atomic_load(&list_a.c[cpu_a].head->data); > + expect_a = __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE); > > cpu_b = rand() % CPU_SETSIZE; > /* > * As list_a is "inactive", we should never see changes > * to list_a. > */ > - if (expect_a != atomic_load(&list_a.c[cpu_a].head->data)) { > + if (expect_a != __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE)) { > fprintf(stderr, "Membarrier test failed\n"); > abort(); > } > > /* Make list_a "active". */ > - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); > + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); > if (rseq_membarrier_expedited(cpu_b) && > errno != ENXIO /* missing CPU*/) { > perror("sys_membarrier"); > abort(); > } > /* Remember a value from list_b. */ > - expect_b = atomic_load(&list_b.c[cpu_b].head->data); > + expect_b = __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE); > } > > test_membarrier_free_percpu_list(&list_a); > @@ -1401,7 +1401,7 @@ void test_membarrier(void) > } > } > > - atomic_store(&thread_args.stop, 1); > + __atomic_store_n(&thread_args.stop, 1, __ATOMIC_RELEASE); > ret = pthread_join(manager_thread, NULL); > if (ret) { > errno = ret; > > --- > base-commit: 2dde18cd1d8fac735875f2e4987f11817cc0bc2c > change-id: 20230908-kselftest-param_test-c-1763b62e762f > > Best regards, > -- > Justin Stitt <justinstitt@google.com> > [1]: https://lore.kernel.org/all/20230908-kselftest-09-08-v2-0-0def978a4c1b@google.com/ Thanks Justin
On 9/26/23 08:20, Justin Stitt wrote: > Ping. > > Looking to get this patch and [1] slated for 6.7 which fixes some > kselftest builds on older kernels. > > On Wed, Sep 13, 2023 at 6:03 AM Justin Stitt <justinstitt@google.com> wrote: >> >> When building with Clang, I am getting many warnings from the selftests/rseq tree. >> >> Here's one such example from rseq tree: >> | param_test.c:1234:10: error: address argument to atomic operation must be a pointer to _Atomic type ('intptr_t *' (aka 'long *') invalid) >> | 1234 | while (!atomic_load(&args->percpu_list_ptr)) {} >> | | ^ ~~~~~~~~~~~~~~~~~~~~~~ >> | /usr/local/google/home/justinstitt/repos/tc-build/build/llvm/final/lib/clang/18/include/stdatomic.h:140:29: note: expanded from macro 'atomic_load' >> | 140 | #define atomic_load(object) __c11_atomic_load(object, __ATOMIC_SEQ_CST) >> | | ^ ~~~~~~ >> >> Use compiler builtins `__atomic_load_n()` and `__atomic_store_n()` with >> accompanying __ATOMIC_ACQUIRE and __ATOMIC_RELEASE, respectively. This >> will fix the warnings because the compiler builtins do not expect their >> arguments to have _Atomic type. This should also make TSAN happier. >> >> Link: https://github.com/ClangBuiltLinux/linux/issues/1698 >> Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/61 >> Suggested-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> >> Signed-off-by: Justin Stitt <justinstitt@google.com> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Peter, should this go through tip ? Shuah, should to go through selftests ? Thanks, Mathieu >> --- >> Note: Previous RFC https://lore.kernel.org/r/20230908-kselftest-param_test-c-v1-1-e35bd9052d61@google.com >> --- >> tools/testing/selftests/rseq/param_test.c | 20 ++++++++++---------- >> 1 file changed, 10 insertions(+), 10 deletions(-) >> >> diff --git a/tools/testing/selftests/rseq/param_test.c b/tools/testing/selftests/rseq/param_test.c >> index bf951a490bb4..20403d58345c 100644 >> --- a/tools/testing/selftests/rseq/param_test.c >> +++ b/tools/testing/selftests/rseq/param_test.c >> @@ -1231,7 +1231,7 @@ void *test_membarrier_worker_thread(void *arg) >> } >> >> /* Wait for initialization. */ >> - while (!atomic_load(&args->percpu_list_ptr)) {} >> + while (!__atomic_load_n(&args->percpu_list_ptr, __ATOMIC_ACQUIRE)) {} >> >> for (i = 0; i < iters; ++i) { >> int ret; >> @@ -1299,22 +1299,22 @@ void *test_membarrier_manager_thread(void *arg) >> test_membarrier_init_percpu_list(&list_a); >> test_membarrier_init_percpu_list(&list_b); >> >> - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); >> + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); >> >> - while (!atomic_load(&args->stop)) { >> + while (!__atomic_load_n(&args->stop, __ATOMIC_ACQUIRE)) { >> /* list_a is "active". */ >> cpu_a = rand() % CPU_SETSIZE; >> /* >> * As list_b is "inactive", we should never see changes >> * to list_b. >> */ >> - if (expect_b != atomic_load(&list_b.c[cpu_b].head->data)) { >> + if (expect_b != __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE)) { >> fprintf(stderr, "Membarrier test failed\n"); >> abort(); >> } >> >> /* Make list_b "active". */ >> - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_b); >> + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_b, __ATOMIC_RELEASE); >> if (rseq_membarrier_expedited(cpu_a) && >> errno != ENXIO /* missing CPU */) { >> perror("sys_membarrier"); >> @@ -1324,27 +1324,27 @@ void *test_membarrier_manager_thread(void *arg) >> * Cpu A should now only modify list_b, so the values >> * in list_a should be stable. >> */ >> - expect_a = atomic_load(&list_a.c[cpu_a].head->data); >> + expect_a = __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE); >> >> cpu_b = rand() % CPU_SETSIZE; >> /* >> * As list_a is "inactive", we should never see changes >> * to list_a. >> */ >> - if (expect_a != atomic_load(&list_a.c[cpu_a].head->data)) { >> + if (expect_a != __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE)) { >> fprintf(stderr, "Membarrier test failed\n"); >> abort(); >> } >> >> /* Make list_a "active". */ >> - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); >> + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); >> if (rseq_membarrier_expedited(cpu_b) && >> errno != ENXIO /* missing CPU*/) { >> perror("sys_membarrier"); >> abort(); >> } >> /* Remember a value from list_b. */ >> - expect_b = atomic_load(&list_b.c[cpu_b].head->data); >> + expect_b = __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE); >> } >> >> test_membarrier_free_percpu_list(&list_a); >> @@ -1401,7 +1401,7 @@ void test_membarrier(void) >> } >> } >> >> - atomic_store(&thread_args.stop, 1); >> + __atomic_store_n(&thread_args.stop, 1, __ATOMIC_RELEASE); >> ret = pthread_join(manager_thread, NULL); >> if (ret) { >> errno = ret; >> >> --- >> base-commit: 2dde18cd1d8fac735875f2e4987f11817cc0bc2c >> change-id: 20230908-kselftest-param_test-c-1763b62e762f >> >> Best regards, >> -- >> Justin Stitt <justinstitt@google.com> >> > > [1]: https://lore.kernel.org/all/20230908-kselftest-09-08-v2-0-0def978a4c1b@google.com/ > > Thanks > Justin
On 9/26/23 13:02, Mathieu Desnoyers wrote: > On 9/26/23 08:20, Justin Stitt wrote: >> Ping. >> >> Looking to get this patch and [1] slated for 6.7 which fixes some >> kselftest builds on older kernels. >> >> On Wed, Sep 13, 2023 at 6:03 AM Justin Stitt <justinstitt@google.com> wrote: >>> >>> When building with Clang, I am getting many warnings from the selftests/rseq tree. >>> >>> Here's one such example from rseq tree: >>> | param_test.c:1234:10: error: address argument to atomic operation must be a pointer to _Atomic type ('intptr_t *' (aka 'long *') invalid) >>> | 1234 | while (!atomic_load(&args->percpu_list_ptr)) {} >>> | | ^ ~~~~~~~~~~~~~~~~~~~~~~ >>> | /usr/local/google/home/justinstitt/repos/tc-build/build/llvm/final/lib/clang/18/include/stdatomic.h:140:29: note: expanded from macro 'atomic_load' >>> | 140 | #define atomic_load(object) __c11_atomic_load(object, __ATOMIC_SEQ_CST) >>> | | ^ ~~~~~~ >>> >>> Use compiler builtins `__atomic_load_n()` and `__atomic_store_n()` with >>> accompanying __ATOMIC_ACQUIRE and __ATOMIC_RELEASE, respectively. This >>> will fix the warnings because the compiler builtins do not expect their >>> arguments to have _Atomic type. This should also make TSAN happier. >>> >>> Link: https://github.com/ClangBuiltLinux/linux/issues/1698 >>> Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/61 >>> Suggested-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> >>> Signed-off-by: Justin Stitt <justinstitt@google.com> > > Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> > > Peter, should this go through tip ? > > Shuah, should to go through selftests ? > I can take this through selftests and apply it - hoping there are no conflicts with what's in Peter's tree. thanks, -- Shuah
On 9/26/23 14:39, Shuah Khan wrote: > On 9/26/23 13:02, Mathieu Desnoyers wrote: >> On 9/26/23 08:20, Justin Stitt wrote: >>> Ping. >>> >>> Looking to get this patch and [1] slated for 6.7 which fixes some >>> kselftest builds on older kernels. >>> >>> On Wed, Sep 13, 2023 at 6:03 AM Justin Stitt <justinstitt@google.com> wrote: >>>> >>>> When building with Clang, I am getting many warnings from the selftests/rseq tree. >>>> >>>> Here's one such example from rseq tree: >>>> | param_test.c:1234:10: error: address argument to atomic operation must be a pointer to _Atomic type ('intptr_t *' (aka 'long *') invalid) >>>> | 1234 | while (!atomic_load(&args->percpu_list_ptr)) {} >>>> | | ^ ~~~~~~~~~~~~~~~~~~~~~~ >>>> | /usr/local/google/home/justinstitt/repos/tc-build/build/llvm/final/lib/clang/18/include/stdatomic.h:140:29: note: expanded from macro 'atomic_load' >>>> | 140 | #define atomic_load(object) __c11_atomic_load(object, __ATOMIC_SEQ_CST) >>>> | | ^ ~~~~~~ >>>> >>>> Use compiler builtins `__atomic_load_n()` and `__atomic_store_n()` with >>>> accompanying __ATOMIC_ACQUIRE and __ATOMIC_RELEASE, respectively. This >>>> will fix the warnings because the compiler builtins do not expect their >>>> arguments to have _Atomic type. This should also make TSAN happier. >>>> >>>> Link: https://github.com/ClangBuiltLinux/linux/issues/1698 >>>> Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/61 >>>> Suggested-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> >>>> Signed-off-by: Justin Stitt <justinstitt@google.com> >> >> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> >> >> Peter, should this go through tip ? >> >> Shuah, should to go through selftests ? >> > > I can take this through selftests and apply it - hoping there > are no conflicts with what's in Peter's tree. > Applied to linux-kselftest next for Linux 6.7-rc1 thanks, -- Shuah
diff --git a/tools/testing/selftests/rseq/param_test.c b/tools/testing/selftests/rseq/param_test.c index bf951a490bb4..20403d58345c 100644 --- a/tools/testing/selftests/rseq/param_test.c +++ b/tools/testing/selftests/rseq/param_test.c @@ -1231,7 +1231,7 @@ void *test_membarrier_worker_thread(void *arg) } /* Wait for initialization. */ - while (!atomic_load(&args->percpu_list_ptr)) {} + while (!__atomic_load_n(&args->percpu_list_ptr, __ATOMIC_ACQUIRE)) {} for (i = 0; i < iters; ++i) { int ret; @@ -1299,22 +1299,22 @@ void *test_membarrier_manager_thread(void *arg) test_membarrier_init_percpu_list(&list_a); test_membarrier_init_percpu_list(&list_b); - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); - while (!atomic_load(&args->stop)) { + while (!__atomic_load_n(&args->stop, __ATOMIC_ACQUIRE)) { /* list_a is "active". */ cpu_a = rand() % CPU_SETSIZE; /* * As list_b is "inactive", we should never see changes * to list_b. */ - if (expect_b != atomic_load(&list_b.c[cpu_b].head->data)) { + if (expect_b != __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE)) { fprintf(stderr, "Membarrier test failed\n"); abort(); } /* Make list_b "active". */ - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_b); + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_b, __ATOMIC_RELEASE); if (rseq_membarrier_expedited(cpu_a) && errno != ENXIO /* missing CPU */) { perror("sys_membarrier"); @@ -1324,27 +1324,27 @@ void *test_membarrier_manager_thread(void *arg) * Cpu A should now only modify list_b, so the values * in list_a should be stable. */ - expect_a = atomic_load(&list_a.c[cpu_a].head->data); + expect_a = __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE); cpu_b = rand() % CPU_SETSIZE; /* * As list_a is "inactive", we should never see changes * to list_a. */ - if (expect_a != atomic_load(&list_a.c[cpu_a].head->data)) { + if (expect_a != __atomic_load_n(&list_a.c[cpu_a].head->data, __ATOMIC_ACQUIRE)) { fprintf(stderr, "Membarrier test failed\n"); abort(); } /* Make list_a "active". */ - atomic_store(&args->percpu_list_ptr, (intptr_t)&list_a); + __atomic_store_n(&args->percpu_list_ptr, (intptr_t)&list_a, __ATOMIC_RELEASE); if (rseq_membarrier_expedited(cpu_b) && errno != ENXIO /* missing CPU*/) { perror("sys_membarrier"); abort(); } /* Remember a value from list_b. */ - expect_b = atomic_load(&list_b.c[cpu_b].head->data); + expect_b = __atomic_load_n(&list_b.c[cpu_b].head->data, __ATOMIC_ACQUIRE); } test_membarrier_free_percpu_list(&list_a); @@ -1401,7 +1401,7 @@ void test_membarrier(void) } } - atomic_store(&thread_args.stop, 1); + __atomic_store_n(&thread_args.stop, 1, __ATOMIC_RELEASE); ret = pthread_join(manager_thread, NULL); if (ret) { errno = ret;
When building with Clang, I am getting many warnings from the selftests/rseq tree. Here's one such example from rseq tree: | param_test.c:1234:10: error: address argument to atomic operation must be a pointer to _Atomic type ('intptr_t *' (aka 'long *') invalid) | 1234 | while (!atomic_load(&args->percpu_list_ptr)) {} | | ^ ~~~~~~~~~~~~~~~~~~~~~~ | /usr/local/google/home/justinstitt/repos/tc-build/build/llvm/final/lib/clang/18/include/stdatomic.h:140:29: note: expanded from macro 'atomic_load' | 140 | #define atomic_load(object) __c11_atomic_load(object, __ATOMIC_SEQ_CST) | | ^ ~~~~~~ Use compiler builtins `__atomic_load_n()` and `__atomic_store_n()` with accompanying __ATOMIC_ACQUIRE and __ATOMIC_RELEASE, respectively. This will fix the warnings because the compiler builtins do not expect their arguments to have _Atomic type. This should also make TSAN happier. Link: https://github.com/ClangBuiltLinux/linux/issues/1698 Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/61 Suggested-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Signed-off-by: Justin Stitt <justinstitt@google.com> --- Note: Previous RFC https://lore.kernel.org/r/20230908-kselftest-param_test-c-v1-1-e35bd9052d61@google.com --- tools/testing/selftests/rseq/param_test.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) --- base-commit: 2dde18cd1d8fac735875f2e4987f11817cc0bc2c change-id: 20230908-kselftest-param_test-c-1763b62e762f Best regards, -- Justin Stitt <justinstitt@google.com>