perf pmu: Move pmu__find_core_pmu() to pmus.c
authorJames Clark <james.clark@arm.com>
Wed, 13 Sep 2023 15:33:48 +0000 (16:33 +0100)
committerNamhyung Kim <namhyung@kernel.org>
Fri, 15 Sep 2023 23:46:40 +0000 (16:46 -0700)
pmu__find_core_pmu() more logically belongs in pmus.c because it
iterates over all PMUs, so move it to pmus.c

At the same time rename it to perf_pmus__find_core_pmu() to match the
naming convention in this file.

list_prepare_entry() can't be used in perf_pmus__scan_core() anymore now
that it's called from the same compilation unit. This is with -O2
(specifically -O1 -ftree-vrp -finline-functions
-finline-small-functions) which allow the bounds of the array
access to be determined at compile time. list_prepare_entry() subtracts
the offset of the 'list' member in struct perf_pmu from &core_pmus,
which isn't a struct perf_pmu. The compiler sees that pmu results in
&core_pmus - 8 and refuses to compile. At runtime this works because
list_for_each_entry_continue() always adds the offset back again before
dereferencing ->next, but it's technically undefined behavior. With
-fsanitize=undefined an additional warning is generated.

Using list_first_entry_or_null() to get the first entry here avoids
doing &core_pmus - 8 but has the same result and fixes both the compile
warning and the undefined behavior warning. There are other uses of
list_prepare_entry() in pmus.c, but the compiler doesn't seem to be
able to see that they can also be called with &core_pmus, so I won't
change any at this time.

Signed-off-by: James Clark <james.clark@arm.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Haixin Yu <yuhaixin.yhx@linux.alibaba.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: https://lore.kernel.org/r/20230913153355.138331-2-james.clark@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
tools/perf/arch/arm64/util/pmu.c
tools/perf/tests/expr.c
tools/perf/util/expr.c
tools/perf/util/pmu.c
tools/perf/util/pmu.h
tools/perf/util/pmus.c

index 615084eb88d8c9535e7f9f02db2eb82f05bea7a5..3d9330feebd28fcc9d495ee6f24672784f5287d0 100644 (file)
@@ -10,7 +10,7 @@
 
 const struct pmu_metrics_table *pmu_metrics_table__find(void)
 {
-       struct perf_pmu *pmu = pmu__find_core_pmu();
+       struct perf_pmu *pmu = perf_pmus__find_core_pmu();
 
        if (pmu)
                return perf_pmu__find_metrics_table(pmu);
@@ -20,7 +20,7 @@ const struct pmu_metrics_table *pmu_metrics_table__find(void)
 
 const struct pmu_events_table *pmu_events_table__find(void)
 {
-       struct perf_pmu *pmu = pmu__find_core_pmu();
+       struct perf_pmu *pmu = perf_pmus__find_core_pmu();
 
        if (pmu)
                return perf_pmu__find_events_table(pmu);
@@ -32,7 +32,7 @@ double perf_pmu__cpu_slots_per_cycle(void)
 {
        char path[PATH_MAX];
        unsigned long long slots = 0;
-       struct perf_pmu *pmu = pmu__find_core_pmu();
+       struct perf_pmu *pmu = perf_pmus__find_core_pmu();
 
        if (pmu) {
                perf_pmu__pathname_scnprintf(path, sizeof(path),
index 78544092ef0c4ab91f85c43229bd52cecde0c977..e3aa9d4fcf3a56c67acb8f084079cc418ec65fd5 100644 (file)
@@ -76,7 +76,7 @@ static int test__expr(struct test_suite *t __maybe_unused, int subtest __maybe_u
        struct expr_parse_ctx *ctx;
        bool is_intel = false;
        char strcmp_cpuid_buf[256];
-       struct perf_pmu *pmu = pmu__find_core_pmu();
+       struct perf_pmu *pmu = perf_pmus__find_core_pmu();
        char *cpuid = perf_pmu__getcpuid(pmu);
        char *escaped_cpuid1, *escaped_cpuid2;
 
index 4488f306de7853daaf145fbe5a52c8ad271de6f1..7be23b3ac082178aed0debbbc842be83f14f7a00 100644 (file)
@@ -509,7 +509,7 @@ double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx __maybe_unused,
                       bool compute_ids __maybe_unused, const char *test_id)
 {
        double ret;
-       struct perf_pmu *pmu = pmu__find_core_pmu();
+       struct perf_pmu *pmu = perf_pmus__find_core_pmu();
        char *cpuid = perf_pmu__getcpuid(pmu);
 
        if (!cpuid)
index e2159854ab26c239019cb8a0248407ce74e5deff..f50a5636633ff7b0e88660af8461e50ab383f033 100644 (file)
@@ -2050,20 +2050,3 @@ void perf_pmu__delete(struct perf_pmu *pmu)
        zfree(&pmu->id);
        free(pmu);
 }
-
-struct perf_pmu *pmu__find_core_pmu(void)
-{
-       struct perf_pmu *pmu = NULL;
-
-       while ((pmu = perf_pmus__scan_core(pmu))) {
-               /*
-                * The cpumap should cover all CPUs. Otherwise, some CPUs may
-                * not support some events or have different event IDs.
-                */
-               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
-                       return NULL;
-
-               return pmu;
-       }
-       return NULL;
-}
index bd5d804a6736c5c94b8cbc8ba12faf3862d2d282..d7b46085642d9ad04fea16005f2fe14dfc132a3b 100644 (file)
@@ -264,6 +264,6 @@ int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename,
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *lookup_name);
 struct perf_pmu *perf_pmu__create_placeholder_core_pmu(struct list_head *core_pmus);
 void perf_pmu__delete(struct perf_pmu *pmu);
-struct perf_pmu *pmu__find_core_pmu(void);
+struct perf_pmu *perf_pmus__find_core_pmu(void);
 
 #endif /* __PMU_H */
index 6631367c756fddb46b99f22d5e31e8929442d9c0..cec869cbe163a0ca12990171adbd802f40be14f6 100644 (file)
@@ -10,6 +10,7 @@
 #include <pthread.h>
 #include <string.h>
 #include <unistd.h>
+#include "cpumap.h"
 #include "debug.h"
 #include "evsel.h"
 #include "pmus.h"
@@ -268,7 +269,7 @@ struct perf_pmu *perf_pmus__scan_core(struct perf_pmu *pmu)
 {
        if (!pmu) {
                pmu_read_sysfs(/*core_only=*/true);
-               pmu = list_prepare_entry(pmu, &core_pmus, list);
+               return list_first_entry_or_null(&core_pmus, typeof(*pmu), list);
        }
        list_for_each_entry_continue(pmu, &core_pmus, list)
                return pmu;
@@ -592,3 +593,20 @@ struct perf_pmu *evsel__find_pmu(const struct evsel *evsel)
        }
        return pmu;
 }
+
+struct perf_pmu *perf_pmus__find_core_pmu(void)
+{
+       struct perf_pmu *pmu = NULL;
+
+       while ((pmu = perf_pmus__scan_core(pmu))) {
+               /*
+                * The cpumap should cover all CPUs. Otherwise, some CPUs may
+                * not support some events or have different event IDs.
+                */
+               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
+                       return NULL;
+
+               return pmu;
+       }
+       return NULL;
+}