[PATCH] perf top: populate PMU capabilities data in perf

All of lore.kernel.org
 help / color / mirror / Atom feed

* [PATCH] perf top: populate PMU capabilities data in perf_env
@ 2025-05-01 18:41 Thomas Falcon
  2025-05-01 20:00 ` Ian Rogers
  0 siblings, 1 reply; 6+ messages in thread
From: Thomas Falcon @ 2025-05-01 18:41 UTC (permalink / raw)
  To: Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
	Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
	Ian Rogers, Adrian Hunter, Kan Liang
  Cc: linux-kernel, linux-perf-users, Thomas Falcon

Calling perf top with brach filters enabled on Intel hybrid CPU's
with branch counter event logging support results in a segfault.

$ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
	frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
perf: Segmentation fault
-------- backtrace --------
./perf() [0x55f460]
/lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
./perf() [0x57b4a7]
./perf() [0x561e5a]
./perf() [0x604a81]
./perf() [0x4395b5]
./perf() [0x601732]
./perf() [0x439bc1]
./perf() [0x5d35b3]
./perf() [0x43936c]
/lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
/lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]

The cause is that perf_env__find_br_cntr_info tries to access a
null pointer pmu_caps in the perf_env struct. Presumably this would
also be an issue when using the cpu_pmu_caps structure available for
homogeneous core CPU's.

Fix this by populating cpu_pmu_caps and pmu_caps structures with
values from sysfs when calling perf top with branch stack sampling
enabled.

Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
---
 tools/perf/builtin-top.c |   8 +++
 tools/perf/util/env.c    | 114 +++++++++++++++++++++++++++++++++++++++
 tools/perf/util/env.h    |   1 +
 3 files changed, 123 insertions(+)

diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 1061f4eebc3f..c2688e4ef3c4 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -1729,6 +1729,14 @@ int cmd_top(int argc, const char **argv)
 	if (opts->branch_stack && callchain_param.enabled)
 		symbol_conf.show_branchflag_count = true;
 
+	if (opts->branch_stack) {
+		status = perf_env__read_core_pmu_caps(&perf_env);
+		if (status) {
+			pr_err("PMU capability data is not available\n");
+			goto out_delete_evlist;
+		}
+	}
+
 	sort__mode = SORT_MODE__TOP;
 	/* display thread wants entries to be collapsed in a different tree */
 	perf_hpp_list.need_collapse = 1;
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 36411749e007..37ed6dc52cf3 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -416,6 +416,120 @@ static int perf_env__read_nr_cpus_avail(struct perf_env *env)
 	return env->nr_cpus_avail ? 0 : -ENOENT;
 }
 
+static int __perf_env__read_core_pmu_caps(struct perf_pmu *pmu,
+					  int *nr_caps, char ***caps,
+					  unsigned int *max_branches,
+					  unsigned int *br_cntr_nr,
+					  unsigned int *br_cntr_width)
+{
+	struct perf_pmu_caps *pcaps = NULL;
+	char *ptr, **tmp;
+	int ret = 0;
+
+	*nr_caps = 0;
+	*caps = NULL;
+
+	if (!pmu->nr_caps)
+		return 0;
+
+	*caps = zalloc(sizeof(char *) * pmu->nr_caps);
+	if (!*caps)
+		return -ENOMEM;
+
+	tmp = *caps;
+	list_for_each_entry(pcaps, &pmu->caps, list) {
+
+		if (asprintf(&ptr, "%s=%s", pcaps->name, pcaps->value) < 0) {
+			ret = -ENOMEM;
+			goto error;
+		}
+
+		*tmp++ = ptr;
+
+		if (!strcmp(pcaps->name, "branches"))
+			*max_branches = atoi(pcaps->value);
+
+		if (!strcmp(pcaps->name, "branch_counter_nr"))
+			*br_cntr_nr = atoi(pcaps->value);
+
+		if (!strcmp(pcaps->name, "branch_counter_width"))
+			*br_cntr_width = atoi(pcaps->value);
+	}
+	*nr_caps = pmu->nr_caps;
+	return 0;
+error:
+	while (tmp-- != *caps)
+		free(*tmp);
+	free(*caps);
+	*caps = NULL;
+	*nr_caps = 0;
+	return ret;
+}
+
+int perf_env__read_core_pmu_caps(struct perf_env *env)
+{
+	struct perf_pmu *pmu = NULL;
+	struct pmu_caps *pmu_caps;
+	int nr_pmu = 0, i = 0, j;
+	int ret;
+
+	nr_pmu = perf_pmus__num_core_pmus();
+
+	if (!nr_pmu)
+		return -ENODEV;
+
+	if (nr_pmu == 1) {
+		pmu = perf_pmus__scan_core(NULL);
+		if (!pmu)
+			return -ENODEV;
+		ret = perf_pmu__caps_parse(pmu);
+		if (ret < 0)
+			return ret;
+		return __perf_env__read_core_pmu_caps(pmu, &env->nr_cpu_pmu_caps,
+						      &env->cpu_pmu_caps,
+						      &env->max_branches,
+						      &env->br_cntr_nr,
+						      &env->br_cntr_width);
+	}
+
+	pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
+	if (!pmu_caps)
+		return -ENOMEM;
+
+	while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
+		if (perf_pmu__caps_parse(pmu) <= 0)
+			continue;
+		ret = __perf_env__read_core_pmu_caps(pmu, &pmu_caps[i].nr_caps,
+						     &pmu_caps[i].caps,
+						     &pmu_caps[i].max_branches,
+						     &pmu_caps[i].br_cntr_nr,
+						     &pmu_caps[i].br_cntr_width);
+		if (ret)
+			goto error;
+
+		pmu_caps[i].pmu_name = strdup(pmu->name);
+		if (!pmu_caps[i].pmu_name) {
+			ret = -ENOMEM;
+			goto error;
+		}
+		i++;
+	}
+
+	env->nr_pmus_with_caps = nr_pmu;
+	env->pmu_caps = pmu_caps;
+
+	return 0;
+error:
+	for (i = 0; i < nr_pmu; i++) {
+		for (j = 0; j < pmu_caps[i].nr_caps; j++)
+			free(pmu_caps[i].caps[j]);
+		free(pmu_caps[i].caps);
+		free(pmu_caps[i].pmu_name);
+	}
+	free(pmu_caps);
+	return ret;
+}
+
 const char *perf_env__raw_arch(struct perf_env *env)
 {
 	return env && !perf_env__read_arch(env) ? env->arch : "unknown";
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index d90e343cf1fa..135a1f714905 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -152,6 +152,7 @@ struct btf_node;
 
 extern struct perf_env perf_env;
 
+int perf_env__read_core_pmu_caps(struct perf_env *env);
 void perf_env__exit(struct perf_env *env);
 
 int perf_env__kernel_is_64_bit(struct perf_env *env);
-- 
2.49.0


^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH] perf top: populate PMU capabilities data in perf_env
  2025-05-01 18:41 [PATCH] perf top: populate PMU capabilities data in perf_env Thomas Falcon
@ 2025-05-01 20:00 ` Ian Rogers
  2025-05-01 20:46   ` Falcon, Thomas
  0 siblings, 1 reply; 6+ messages in thread
From: Ian Rogers @ 2025-05-01 20:00 UTC (permalink / raw)
  To: Thomas Falcon
  Cc: Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
	Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
	Adrian Hunter, Kan Liang, linux-kernel, linux-perf-users

On Thu, May 1, 2025 at 11:42 AM Thomas Falcon <thomas.falcon@intel.com> wrote:
>
> Calling perf top with brach filters enabled on Intel hybrid CPU's
> with branch counter event logging support results in a segfault.
>
> $ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
>         frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
> perf: Segmentation fault
> -------- backtrace --------
> ./perf() [0x55f460]
> /lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
> ./perf() [0x57b4a7]
> ./perf() [0x561e5a]
> ./perf() [0x604a81]
> ./perf() [0x4395b5]
> ./perf() [0x601732]
> ./perf() [0x439bc1]
> ./perf() [0x5d35b3]
> ./perf() [0x43936c]
> /lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
> /lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]

Thanks Thomas. Could you generate this backtrace in GDB? I did write a
patch to symbolize backtraces like this:
https://lore.kernel.org/lkml/20250313052952.871958-2-irogers@google.com/
Sadly without any reviewed tags and unmerged - the code calls routines
that malloc so it isn't strictly sound if say the backtrace was needed
from a SEGV in the malloc implementation, it is nicely
self-referencing the perf APIs, ..

> The cause is that perf_env__find_br_cntr_info tries to access a
> null pointer pmu_caps in the perf_env struct. Presumably this would
> also be an issue when using the cpu_pmu_caps structure available for
> homogeneous core CPU's.

I'm a little confused in the top code, we have the global perf_env
being used and one in the session:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/builtin-top.c?h=perf-tools-next#n649
```
ret = evlist__tui_browse_hists(top->evlist, help, &hbt, top->min_percent,
      &top->session->header.env, !top->record_opts.overwrite);
```
and the global perf_env:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/builtin-top.c?h=perf-tools-next#n1641
```
status = perf_env__read_cpuid(&perf_env);
```

I kind of wish we didn't have the global one as what's the deal with
ownership with it.

> Fix this by populating cpu_pmu_caps and pmu_caps structures with
> values from sysfs when calling perf top with branch stack sampling
> enabled.

I wonder if we could encounter similar problems from say a perf script
handling live data and so some kind of lazy initialization should be
employed. It is hard to say without seeing the backtrace.

Thanks,
Ian

> Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
> ---
>  tools/perf/builtin-top.c |   8 +++
>  tools/perf/util/env.c    | 114 +++++++++++++++++++++++++++++++++++++++
>  tools/perf/util/env.h    |   1 +
>  3 files changed, 123 insertions(+)
>
> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> index 1061f4eebc3f..c2688e4ef3c4 100644
> --- a/tools/perf/builtin-top.c
> +++ b/tools/perf/builtin-top.c
> @@ -1729,6 +1729,14 @@ int cmd_top(int argc, const char **argv)
>         if (opts->branch_stack && callchain_param.enabled)
>                 symbol_conf.show_branchflag_count = true;
>
> +       if (opts->branch_stack) {
> +               status = perf_env__read_core_pmu_caps(&perf_env);
> +               if (status) {
> +                       pr_err("PMU capability data is not available\n");
> +                       goto out_delete_evlist;
> +               }
> +       }
> +
>         sort__mode = SORT_MODE__TOP;
>         /* display thread wants entries to be collapsed in a different tree */
>         perf_hpp_list.need_collapse = 1;
> diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
> index 36411749e007..37ed6dc52cf3 100644
> --- a/tools/perf/util/env.c
> +++ b/tools/perf/util/env.c
> @@ -416,6 +416,120 @@ static int perf_env__read_nr_cpus_avail(struct perf_env *env)
>         return env->nr_cpus_avail ? 0 : -ENOENT;
>  }
>
> +static int __perf_env__read_core_pmu_caps(struct perf_pmu *pmu,
> +                                         int *nr_caps, char ***caps,
> +                                         unsigned int *max_branches,
> +                                         unsigned int *br_cntr_nr,
> +                                         unsigned int *br_cntr_width)
> +{
> +       struct perf_pmu_caps *pcaps = NULL;
> +       char *ptr, **tmp;
> +       int ret = 0;
> +
> +       *nr_caps = 0;
> +       *caps = NULL;
> +
> +       if (!pmu->nr_caps)
> +               return 0;
> +
> +       *caps = zalloc(sizeof(char *) * pmu->nr_caps);
> +       if (!*caps)
> +               return -ENOMEM;
> +
> +       tmp = *caps;
> +       list_for_each_entry(pcaps, &pmu->caps, list) {
> +
> +               if (asprintf(&ptr, "%s=%s", pcaps->name, pcaps->value) < 0) {
> +                       ret = -ENOMEM;
> +                       goto error;
> +               }
> +
> +               *tmp++ = ptr;
> +
> +               if (!strcmp(pcaps->name, "branches"))
> +                       *max_branches = atoi(pcaps->value);
> +
> +               if (!strcmp(pcaps->name, "branch_counter_nr"))
> +                       *br_cntr_nr = atoi(pcaps->value);
> +
> +               if (!strcmp(pcaps->name, "branch_counter_width"))
> +                       *br_cntr_width = atoi(pcaps->value);
> +       }
> +       *nr_caps = pmu->nr_caps;
> +       return 0;
> +error:
> +       while (tmp-- != *caps)
> +               free(*tmp);
> +       free(*caps);
> +       *caps = NULL;
> +       *nr_caps = 0;
> +       return ret;
> +}
> +
> +int perf_env__read_core_pmu_caps(struct perf_env *env)
> +{
> +       struct perf_pmu *pmu = NULL;
> +       struct pmu_caps *pmu_caps;
> +       int nr_pmu = 0, i = 0, j;
> +       int ret;
> +
> +       nr_pmu = perf_pmus__num_core_pmus();
> +
> +       if (!nr_pmu)
> +               return -ENODEV;
> +
> +       if (nr_pmu == 1) {
> +               pmu = perf_pmus__scan_core(NULL);
> +               if (!pmu)
> +                       return -ENODEV;
> +               ret = perf_pmu__caps_parse(pmu);
> +               if (ret < 0)
> +                       return ret;
> +               return __perf_env__read_core_pmu_caps(pmu, &env->nr_cpu_pmu_caps,
> +                                                     &env->cpu_pmu_caps,
> +                                                     &env->max_branches,
> +                                                     &env->br_cntr_nr,
> +                                                     &env->br_cntr_width);
> +       }
> +
> +       pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
> +       if (!pmu_caps)
> +               return -ENOMEM;
> +
> +       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> +               if (perf_pmu__caps_parse(pmu) <= 0)
> +                       continue;
> +               ret = __perf_env__read_core_pmu_caps(pmu, &pmu_caps[i].nr_caps,
> +                                                    &pmu_caps[i].caps,
> +                                                    &pmu_caps[i].max_branches,
> +                                                    &pmu_caps[i].br_cntr_nr,
> +                                                    &pmu_caps[i].br_cntr_width);
> +               if (ret)
> +                       goto error;
> +
> +               pmu_caps[i].pmu_name = strdup(pmu->name);
> +               if (!pmu_caps[i].pmu_name) {
> +                       ret = -ENOMEM;
> +                       goto error;
> +               }
> +               i++;
> +       }
> +
> +       env->nr_pmus_with_caps = nr_pmu;
> +       env->pmu_caps = pmu_caps;
> +
> +       return 0;
> +error:
> +       for (i = 0; i < nr_pmu; i++) {
> +               for (j = 0; j < pmu_caps[i].nr_caps; j++)
> +                       free(pmu_caps[i].caps[j]);
> +               free(pmu_caps[i].caps);
> +               free(pmu_caps[i].pmu_name);
> +       }
> +       free(pmu_caps);
> +       return ret;
> +}
> +
>  const char *perf_env__raw_arch(struct perf_env *env)
>  {
>         return env && !perf_env__read_arch(env) ? env->arch : "unknown";
> diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
> index d90e343cf1fa..135a1f714905 100644
> --- a/tools/perf/util/env.h
> +++ b/tools/perf/util/env.h
> @@ -152,6 +152,7 @@ struct btf_node;
>
>  extern struct perf_env perf_env;
>
> +int perf_env__read_core_pmu_caps(struct perf_env *env);
>  void perf_env__exit(struct perf_env *env);
>
>  int perf_env__kernel_is_64_bit(struct perf_env *env);
> --
> 2.49.0
>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] perf top: populate PMU capabilities data in perf_env
  2025-05-01 20:00 ` Ian Rogers
@ 2025-05-01 20:46   ` Falcon, Thomas
  2025-05-06 16:09     ` Falcon, Thomas
  0 siblings, 1 reply; 6+ messages in thread
From: Falcon, Thomas @ 2025-05-01 20:46 UTC (permalink / raw)
  To: irogers@google.com
  Cc: alexander.shishkin@linux.intel.com,
	linux-perf-users@vger.kernel.org, peterz@infradead.org,
	acme@kernel.org, mingo@redhat.com, Hunter, Adrian,
	namhyung@kernel.org, jolsa@kernel.org,
	linux-kernel@vger.kernel.org, kan.liang@linux.intel.com,
	mark.rutland@arm.com

On Thu, 2025-05-01 at 13:00 -0700, Ian Rogers wrote:
> On Thu, May 1, 2025 at 11:42 AM Thomas Falcon <thomas.falcon@intel.com> wrote:
> > 
> > Calling perf top with brach filters enabled on Intel hybrid CPU's
> > with branch counter event logging support results in a segfault.
> > 
> > $ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
> >         frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
> > perf: Segmentation fault
> > -------- backtrace --------
> > ./perf() [0x55f460]
> > /lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
> > ./perf() [0x57b4a7]
> > ./perf() [0x561e5a]
> > ./perf() [0x604a81]
> > ./perf() [0x4395b5]
> > ./perf() [0x601732]
> > ./perf() [0x439bc1]
> > ./perf() [0x5d35b3]
> > ./perf() [0x43936c]
> > /lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
> > /lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]
> 

Hi Ian, thanks for reviewing.

> Thanks Thomas. Could you generate this backtrace in GDB? I did write a
> patch to symbolize backtraces like this:
> https://lore.kernel.org/lkml/20250313052952.871958-2-irogers@google.com/
> Sadly without any reviewed tags and unmerged - the code calls routines
> that malloc so it isn't strictly sound if say the backtrace was needed
> from a SEGV in the malloc implementation, it is nicely
> self-referencing the perf APIs, ..

Sorry about that, here is the backtrace I'm seeing when running the perf top command in gdb:

Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
                                                                                                   
[Switching to Thread 0x7fffcb7fe6c0 (LWP 812169)]
0x000000000057b4a7 in perf_env.find_br_cntr_info ()
(gdb) backtrace
#0  0x000000000057b4a7 in perf_env.find_br_cntr_info ()
#1  0x0000000000561e5a in addr_map_symbol.account_cycles ()
#2  0x0000000000604a81 in hist.account_cycles ()
#3  0x00000000004395b5 in hist_iter.top_callback ()
#4  0x0000000000601732 in hist_entry_iter.add ()
#5  0x0000000000439bc1 in deliver_event ()
#6  0x00000000005d35b3 in __ordered_events__flush ()
#7  0x000000000043936c in process_thread ()
#8  0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
#9  0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78

Thanks,
Tom

> ```
> status = perf_env__read_cpuid(&perf_env);
> ```
> 
> I kind of wish we didn't have the global one as what's the deal with
> ownership with it.
> 
> > Fix this by populating cpu_pmu_caps and pmu_caps structures with
> > values from sysfs when calling perf top with branch stack sampling
> > enabled.
> 
> I wonder if we could encounter similar problems from say a perf script
> handling live data and so some kind of lazy initialization should be
> employed. It is hard to say without seeing the backtrace.
> 
> Thanks,
> Ian
> 
> > Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
> > ---
> >  tools/perf/builtin-top.c |   8 +++
> >  tools/perf/util/env.c    | 114 +++++++++++++++++++++++++++++++++++++++
> >  tools/perf/util/env.h    |   1 +
> >  3 files changed, 123 insertions(+)
> > 
> > diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> > index 1061f4eebc3f..c2688e4ef3c4 100644
> > --- a/tools/perf/builtin-top.c
> > +++ b/tools/perf/builtin-top.c
> > @@ -1729,6 +1729,14 @@ int cmd_top(int argc, const char **argv)
> >         if (opts->branch_stack && callchain_param.enabled)
> >                 symbol_conf.show_branchflag_count = true;
> > 
> > +       if (opts->branch_stack) {
> > +               status = perf_env__read_core_pmu_caps(&perf_env);
> > +               if (status) {
> > +                       pr_err("PMU capability data is not available\n");
> > +                       goto out_delete_evlist;
> > +               }
> > +       }
> > +
> >         sort__mode = SORT_MODE__TOP;
> >         /* display thread wants entries to be collapsed in a different tree */
> >         perf_hpp_list.need_collapse = 1;
> > diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
> > index 36411749e007..37ed6dc52cf3 100644
> > --- a/tools/perf/util/env.c
> > +++ b/tools/perf/util/env.c
> > @@ -416,6 +416,120 @@ static int perf_env__read_nr_cpus_avail(struct perf_env *env)
> >         return env->nr_cpus_avail ? 0 : -ENOENT;
> >  }
> > 
> > +static int __perf_env__read_core_pmu_caps(struct perf_pmu *pmu,
> > +                                         int *nr_caps, char ***caps,
> > +                                         unsigned int *max_branches,
> > +                                         unsigned int *br_cntr_nr,
> > +                                         unsigned int *br_cntr_width)
> > +{
> > +       struct perf_pmu_caps *pcaps = NULL;
> > +       char *ptr, **tmp;
> > +       int ret = 0;
> > +
> > +       *nr_caps = 0;
> > +       *caps = NULL;
> > +
> > +       if (!pmu->nr_caps)
> > +               return 0;
> > +
> > +       *caps = zalloc(sizeof(char *) * pmu->nr_caps);
> > +       if (!*caps)
> > +               return -ENOMEM;
> > +
> > +       tmp = *caps;
> > +       list_for_each_entry(pcaps, &pmu->caps, list) {
> > +
> > +               if (asprintf(&ptr, "%s=%s", pcaps->name, pcaps->value) < 0) {
> > +                       ret = -ENOMEM;
> > +                       goto error;
> > +               }
> > +
> > +               *tmp++ = ptr;
> > +
> > +               if (!strcmp(pcaps->name, "branches"))
> > +                       *max_branches = atoi(pcaps->value);
> > +
> > +               if (!strcmp(pcaps->name, "branch_counter_nr"))
> > +                       *br_cntr_nr = atoi(pcaps->value);
> > +
> > +               if (!strcmp(pcaps->name, "branch_counter_width"))
> > +                       *br_cntr_width = atoi(pcaps->value);
> > +       }
> > +       *nr_caps = pmu->nr_caps;
> > +       return 0;
> > +error:
> > +       while (tmp-- != *caps)
> > +               free(*tmp);
> > +       free(*caps);
> > +       *caps = NULL;
> > +       *nr_caps = 0;
> > +       return ret;
> > +}
> > +
> > +int perf_env__read_core_pmu_caps(struct perf_env *env)
> > +{
> > +       struct perf_pmu *pmu = NULL;
> > +       struct pmu_caps *pmu_caps;
> > +       int nr_pmu = 0, i = 0, j;
> > +       int ret;
> > +
> > +       nr_pmu = perf_pmus__num_core_pmus();
> > +
> > +       if (!nr_pmu)
> > +               return -ENODEV;
> > +
> > +       if (nr_pmu == 1) {
> > +               pmu = perf_pmus__scan_core(NULL);
> > +               if (!pmu)
> > +                       return -ENODEV;
> > +               ret = perf_pmu__caps_parse(pmu);
> > +               if (ret < 0)
> > +                       return ret;
> > +               return __perf_env__read_core_pmu_caps(pmu, &env->nr_cpu_pmu_caps,
> > +                                                     &env->cpu_pmu_caps,
> > +                                                     &env->max_branches,
> > +                                                     &env->br_cntr_nr,
> > +                                                     &env->br_cntr_width);
> > +       }
> > +
> > +       pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
> > +       if (!pmu_caps)
> > +               return -ENOMEM;
> > +
> > +       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> > +               if (perf_pmu__caps_parse(pmu) <= 0)
> > +                       continue;
> > +               ret = __perf_env__read_core_pmu_caps(pmu, &pmu_caps[i].nr_caps,
> > +                                                    &pmu_caps[i].caps,
> > +                                                    &pmu_caps[i].max_branches,
> > +                                                    &pmu_caps[i].br_cntr_nr,
> > +                                                    &pmu_caps[i].br_cntr_width);
> > +               if (ret)
> > +                       goto error;
> > +
> > +               pmu_caps[i].pmu_name = strdup(pmu->name);
> > +               if (!pmu_caps[i].pmu_name) {
> > +                       ret = -ENOMEM;
> > +                       goto error;
> > +               }
> > +               i++;
> > +       }
> > +
> > +       env->nr_pmus_with_caps = nr_pmu;
> > +       env->pmu_caps = pmu_caps;
> > +
> > +       return 0;
> > +error:
> > +       for (i = 0; i < nr_pmu; i++) {
> > +               for (j = 0; j < pmu_caps[i].nr_caps; j++)
> > +                       free(pmu_caps[i].caps[j]);
> > +               free(pmu_caps[i].caps);
> > +               free(pmu_caps[i].pmu_name);
> > +       }
> > +       free(pmu_caps);
> > +       return ret;
> > +}
> > +
> >  const char *perf_env__raw_arch(struct perf_env *env)
> >  {
> >         return env && !perf_env__read_arch(env) ? env->arch : "unknown";
> > diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
> > index d90e343cf1fa..135a1f714905 100644
> > --- a/tools/perf/util/env.h
> > +++ b/tools/perf/util/env.h
> > @@ -152,6 +152,7 @@ struct btf_node;
> > 
> >  extern struct perf_env perf_env;
> > 
> > +int perf_env__read_core_pmu_caps(struct perf_env *env);
> >  void perf_env__exit(struct perf_env *env);
> > 
> >  int perf_env__kernel_is_64_bit(struct perf_env *env);
> > --
> > 2.49.0
> > 


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] perf top: populate PMU capabilities data in perf_env
  2025-05-01 20:46   ` Falcon, Thomas
@ 2025-05-06 16:09     ` Falcon, Thomas
  2025-05-06 16:56       ` Arnaldo Carvalho de Melo
  0 siblings, 1 reply; 6+ messages in thread
From: Falcon, Thomas @ 2025-05-06 16:09 UTC (permalink / raw)
  To: irogers@google.com
  Cc: alexander.shishkin@linux.intel.com, peterz@infradead.org,
	acme@kernel.org, mingo@redhat.com,
	linux-perf-users@vger.kernel.org, Hunter, Adrian,
	namhyung@kernel.org, jolsa@kernel.org, kan.liang@linux.intel.com,
	linux-kernel@vger.kernel.org, mark.rutland@arm.com

On Thu, 2025-05-01 at 20:46 +0000, Falcon, Thomas wrote:
> On Thu, 2025-05-01 at 13:00 -0700, Ian Rogers wrote:
> > On Thu, May 1, 2025 at 11:42 AM Thomas Falcon <thomas.falcon@intel.com> wrote:
> > > 
> > > Calling perf top with brach filters enabled on Intel hybrid CPU's
> > > with branch counter event logging support results in a segfault.
> > > 
> > > $ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
> > >         frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
> > > perf: Segmentation fault
> > > -------- backtrace --------
> > > ./perf() [0x55f460]
> > > /lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
> > > ./perf() [0x57b4a7]
> > > ./perf() [0x561e5a]
> > > ./perf() [0x604a81]
> > > ./perf() [0x4395b5]
> > > ./perf() [0x601732]
> > > ./perf() [0x439bc1]
> > > ./perf() [0x5d35b3]
> > > ./perf() [0x43936c]
> > > /lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
> > > /lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]
> > 
> 
> Hi Ian, thanks for reviewing.
> 
> > Thanks Thomas. Could you generate this backtrace in GDB? I did write a
> > patch to symbolize backtraces like this:
> > https://lore.kernel.org/lkml/20250313052952.871958-2-irogers@google.com/
> > Sadly without any reviewed tags and unmerged - the code calls routines
> > that malloc so it isn't strictly sound if say the backtrace was needed
> > from a SEGV in the malloc implementation, it is nicely
> > self-referencing the perf APIs, ..
> 
> Sorry about that, here is the backtrace I'm seeing when running the perf top command in gdb:
> 
> Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
>                                                                                                    
> [Switching to Thread 0x7fffcb7fe6c0 (LWP 812169)]
> 0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> (gdb) backtrace
> #0  0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> #1  0x0000000000561e5a in addr_map_symbol.account_cycles ()
> #2  0x0000000000604a81 in hist.account_cycles ()
> #3  0x00000000004395b5 in hist_iter.top_callback ()
> #4  0x0000000000601732 in hist_entry_iter.add ()
> #5  0x0000000000439bc1 in deliver_event ()
> #6  0x00000000005d35b3 in __ordered_events__flush ()
> #7  0x000000000043936c in process_thread ()
> #8  0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
> #9  0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
> 
> Thanks,
> Tom
> 

Sorry, let me try this again...

Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffcf7fe6c0 (LWP 940046)]
perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
653			*width = env->cpu_pmu_caps ? env->br_cntr_width :
(gdb) bt
#0  perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
#1  0x00000000005ad829 in symbol__account_br_cntr (branch=0x7fffd11f9c00, evsel=0xfae480, offset=20, br_cntr=4) at util/annotate.c:345
#2  0x00000000005ada8b in symbol__account_cycles (addr=5580436, start=5580433, sym=0x7fffd00d3010, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:389
#3  0x00000000005adc06 in addr_map_symbol__account_cycles (ams=0x7fffd17b1e20, start=0x7fffd17b1f00, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:422
#4  0x0000000000688ab4 in hist__account_cycles (bs=0x10cbaa8, al=0x7fffcf7fd540, sample=0x7fffcf7fd760, nonany_branch_mode=false, total_cycles=0x0, evsel=0xfae480) at util/hist.c:2774
#5  0x0000000000446004 in hist_iter__top_callback (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, single=true, arg=0x7fffffff9de0) at builtin-top.c:737
#6  0x0000000000684d2a in hist_entry_iter__add (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, max_stack_depth=127, arg=0x7fffffff9de0) at util/hist.c:1291
#7  0x00000000004464fe in perf_event__process_sample (tool=0x7fffffff9de0, event=0x10cba70, evsel=0xfae480, sample=0x7fffcf7fd760, machine=0x105ec68) at builtin-top.c:845
#8  0x0000000000447523 in deliver_event (qe=0x7fffffffa0f8, qevent=0x10cdd60) at builtin-top.c:1211
#9  0x0000000000648aff in do_flush (oe=0x7fffffffa0f8, show_progress=false) at util/ordered-events.c:245
#10 0x0000000000648e56 in __ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP, timestamp=0) at util/ordered-events.c:324
#11 0x0000000000648f40 in ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP) at util/ordered-events.c:342
#12 0x0000000000447097 in process_thread (arg=0x7fffffff9de0) at builtin-top.c:1120
#13 0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
#14 0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78


> > ```
> > status = perf_env__read_cpuid(&perf_env);
> > ```
> > 
> > I kind of wish we didn't have the global one as what's the deal with
> > ownership with it.
> > 
> > > Fix this by populating cpu_pmu_caps and pmu_caps structures with
> > > values from sysfs when calling perf top with branch stack sampling
> > > enabled.
> > 
> > I wonder if we could encounter similar problems from say a perf script
> > handling live data and so some kind of lazy initialization should be
> > employed. It is hard to say without seeing the backtrace.
> > 
> > Thanks,
> > Ian
> > 
> > > Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
> > > ---
> > >  tools/perf/builtin-top.c |   8 +++
> > >  tools/perf/util/env.c    | 114 +++++++++++++++++++++++++++++++++++++++
> > >  tools/perf/util/env.h    |   1 +
> > >  3 files changed, 123 insertions(+)
> > > 
> > > diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> > > index 1061f4eebc3f..c2688e4ef3c4 100644
> > > --- a/tools/perf/builtin-top.c
> > > +++ b/tools/perf/builtin-top.c
> > > @@ -1729,6 +1729,14 @@ int cmd_top(int argc, const char **argv)
> > >         if (opts->branch_stack && callchain_param.enabled)
> > >                 symbol_conf.show_branchflag_count = true;
> > > 
> > > +       if (opts->branch_stack) {
> > > +               status = perf_env__read_core_pmu_caps(&perf_env);
> > > +               if (status) {
> > > +                       pr_err("PMU capability data is not available\n");
> > > +                       goto out_delete_evlist;
> > > +               }
> > > +       }
> > > +
> > >         sort__mode = SORT_MODE__TOP;
> > >         /* display thread wants entries to be collapsed in a different tree */
> > >         perf_hpp_list.need_collapse = 1;
> > > diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
> > > index 36411749e007..37ed6dc52cf3 100644
> > > --- a/tools/perf/util/env.c
> > > +++ b/tools/perf/util/env.c
> > > @@ -416,6 +416,120 @@ static int perf_env__read_nr_cpus_avail(struct perf_env *env)
> > >         return env->nr_cpus_avail ? 0 : -ENOENT;
> > >  }
> > > 
> > > +static int __perf_env__read_core_pmu_caps(struct perf_pmu *pmu,
> > > +                                         int *nr_caps, char ***caps,
> > > +                                         unsigned int *max_branches,
> > > +                                         unsigned int *br_cntr_nr,
> > > +                                         unsigned int *br_cntr_width)
> > > +{
> > > +       struct perf_pmu_caps *pcaps = NULL;
> > > +       char *ptr, **tmp;
> > > +       int ret = 0;
> > > +
> > > +       *nr_caps = 0;
> > > +       *caps = NULL;
> > > +
> > > +       if (!pmu->nr_caps)
> > > +               return 0;
> > > +
> > > +       *caps = zalloc(sizeof(char *) * pmu->nr_caps);
> > > +       if (!*caps)
> > > +               return -ENOMEM;
> > > +
> > > +       tmp = *caps;
> > > +       list_for_each_entry(pcaps, &pmu->caps, list) {
> > > +
> > > +               if (asprintf(&ptr, "%s=%s", pcaps->name, pcaps->value) < 0) {
> > > +                       ret = -ENOMEM;
> > > +                       goto error;
> > > +               }
> > > +
> > > +               *tmp++ = ptr;
> > > +
> > > +               if (!strcmp(pcaps->name, "branches"))
> > > +                       *max_branches = atoi(pcaps->value);
> > > +
> > > +               if (!strcmp(pcaps->name, "branch_counter_nr"))
> > > +                       *br_cntr_nr = atoi(pcaps->value);
> > > +
> > > +               if (!strcmp(pcaps->name, "branch_counter_width"))
> > > +                       *br_cntr_width = atoi(pcaps->value);
> > > +       }
> > > +       *nr_caps = pmu->nr_caps;
> > > +       return 0;
> > > +error:
> > > +       while (tmp-- != *caps)
> > > +               free(*tmp);
> > > +       free(*caps);
> > > +       *caps = NULL;
> > > +       *nr_caps = 0;
> > > +       return ret;
> > > +}
> > > +
> > > +int perf_env__read_core_pmu_caps(struct perf_env *env)
> > > +{
> > > +       struct perf_pmu *pmu = NULL;
> > > +       struct pmu_caps *pmu_caps;
> > > +       int nr_pmu = 0, i = 0, j;
> > > +       int ret;
> > > +
> > > +       nr_pmu = perf_pmus__num_core_pmus();
> > > +
> > > +       if (!nr_pmu)
> > > +               return -ENODEV;
> > > +
> > > +       if (nr_pmu == 1) {
> > > +               pmu = perf_pmus__scan_core(NULL);
> > > +               if (!pmu)
> > > +                       return -ENODEV;
> > > +               ret = perf_pmu__caps_parse(pmu);
> > > +               if (ret < 0)
> > > +                       return ret;
> > > +               return __perf_env__read_core_pmu_caps(pmu, &env->nr_cpu_pmu_caps,
> > > +                                                     &env->cpu_pmu_caps,
> > > +                                                     &env->max_branches,
> > > +                                                     &env->br_cntr_nr,
> > > +                                                     &env->br_cntr_width);
> > > +       }
> > > +
> > > +       pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
> > > +       if (!pmu_caps)
> > > +               return -ENOMEM;
> > > +
> > > +       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> > > +               if (perf_pmu__caps_parse(pmu) <= 0)
> > > +                       continue;
> > > +               ret = __perf_env__read_core_pmu_caps(pmu, &pmu_caps[i].nr_caps,
> > > +                                                    &pmu_caps[i].caps,
> > > +                                                    &pmu_caps[i].max_branches,
> > > +                                                    &pmu_caps[i].br_cntr_nr,
> > > +                                                    &pmu_caps[i].br_cntr_width);
> > > +               if (ret)
> > > +                       goto error;
> > > +
> > > +               pmu_caps[i].pmu_name = strdup(pmu->name);
> > > +               if (!pmu_caps[i].pmu_name) {
> > > +                       ret = -ENOMEM;
> > > +                       goto error;
> > > +               }
> > > +               i++;
> > > +       }
> > > +
> > > +       env->nr_pmus_with_caps = nr_pmu;
> > > +       env->pmu_caps = pmu_caps;
> > > +
> > > +       return 0;
> > > +error:
> > > +       for (i = 0; i < nr_pmu; i++) {
> > > +               for (j = 0; j < pmu_caps[i].nr_caps; j++)
> > > +                       free(pmu_caps[i].caps[j]);
> > > +               free(pmu_caps[i].caps);
> > > +               free(pmu_caps[i].pmu_name);
> > > +       }
> > > +       free(pmu_caps);
> > > +       return ret;
> > > +}
> > > +
> > >  const char *perf_env__raw_arch(struct perf_env *env)
> > >  {
> > >         return env && !perf_env__read_arch(env) ? env->arch : "unknown";
> > > diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
> > > index d90e343cf1fa..135a1f714905 100644
> > > --- a/tools/perf/util/env.h
> > > +++ b/tools/perf/util/env.h
> > > @@ -152,6 +152,7 @@ struct btf_node;
> > > 
> > >  extern struct perf_env perf_env;
> > > 
> > > +int perf_env__read_core_pmu_caps(struct perf_env *env);
> > >  void perf_env__exit(struct perf_env *env);
> > > 
> > >  int perf_env__kernel_is_64_bit(struct perf_env *env);
> > > --
> > > 2.49.0
> > > 
> 


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] perf top: populate PMU capabilities data in perf_env
  2025-05-06 16:09     ` Falcon, Thomas
@ 2025-05-06 16:56       ` Arnaldo Carvalho de Melo
  2025-05-06 18:52         ` Falcon, Thomas
  0 siblings, 1 reply; 6+ messages in thread
From: Arnaldo Carvalho de Melo @ 2025-05-06 16:56 UTC (permalink / raw)
  To: Falcon, Thomas
  Cc: irogers@google.com, alexander.shishkin@linux.intel.com,
	peterz@infradead.org, mingo@redhat.com,
	linux-perf-users@vger.kernel.org, Hunter, Adrian,
	namhyung@kernel.org, jolsa@kernel.org, kan.liang@linux.intel.com,
	linux-kernel@vger.kernel.org, mark.rutland@arm.com

On Tue, May 06, 2025 at 04:09:37PM +0000, Falcon, Thomas wrote:
> On Thu, 2025-05-01 at 20:46 +0000, Falcon, Thomas wrote:
> > On Thu, 2025-05-01 at 13:00 -0700, Ian Rogers wrote:
> > > On Thu, May 1, 2025 at 11:42 AM Thomas Falcon <thomas.falcon@intel.com> wrote:
> > > > Calling perf top with brach filters enabled on Intel hybrid CPU's
> > > > with branch counter event logging support results in a segfault.

> > > > $ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
> > > >         frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
> > > > perf: Segmentation fault
> > > > -------- backtrace --------
> > > > ./perf() [0x55f460]
> > > > /lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
> > > > ./perf() [0x57b4a7]
> > > > ./perf() [0x561e5a]
> > > > ./perf() [0x604a81]
> > > > ./perf() [0x4395b5]
> > > > ./perf() [0x601732]
> > > > ./perf() [0x439bc1]
> > > > ./perf() [0x5d35b3]
> > > > ./perf() [0x43936c]
> > > > /lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
> > > > /lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]

> > Hi Ian, thanks for reviewing.

> > > Thanks Thomas. Could you generate this backtrace in GDB? I did write a
> > > patch to symbolize backtraces like this:
> > > https://lore.kernel.org/lkml/20250313052952.871958-2-irogers@google.com/
> > > Sadly without any reviewed tags and unmerged - the code calls routines
> > > that malloc so it isn't strictly sound if say the backtrace was needed
> > > from a SEGV in the malloc implementation, it is nicely
> > > self-referencing the perf APIs, ..

> > Sorry about that, here is the backtrace I'm seeing when running the perf top command in gdb:

> > Thread 27 "perf" received signal SIGSEGV, Segmentation fault.

> > [Switching to Thread 0x7fffcb7fe6c0 (LWP 812169)]
> > 0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> > (gdb) backtrace
> > #0  0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> > #1  0x0000000000561e5a in addr_map_symbol.account_cycles ()
> > #2  0x0000000000604a81 in hist.account_cycles ()
> > #3  0x00000000004395b5 in hist_iter.top_callback ()
> > #4  0x0000000000601732 in hist_entry_iter.add ()
> > #5  0x0000000000439bc1 in deliver_event ()
> > #6  0x00000000005d35b3 in __ordered_events__flush ()
> > #7  0x000000000043936c in process_thread ()
> > #8  0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
> > #9  0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
 
> Sorry, let me try this again...
 
> Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
> [Switching to Thread 0x7fffcf7fe6c0 (LWP 940046)]
> perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
> 653			*width = env->cpu_pmu_caps ? env->br_cntr_width :
> (gdb) bt
> #0  perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
> #1  0x00000000005ad829 in symbol__account_br_cntr (branch=0x7fffd11f9c00, evsel=0xfae480, offset=20, br_cntr=4) at util/annotate.c:345
> #2  0x00000000005ada8b in symbol__account_cycles (addr=5580436, start=5580433, sym=0x7fffd00d3010, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:389
> #3  0x00000000005adc06 in addr_map_symbol__account_cycles (ams=0x7fffd17b1e20, start=0x7fffd17b1f00, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:422
> #4  0x0000000000688ab4 in hist__account_cycles (bs=0x10cbaa8, al=0x7fffcf7fd540, sample=0x7fffcf7fd760, nonany_branch_mode=false, total_cycles=0x0, evsel=0xfae480) at util/hist.c:2774
> #5  0x0000000000446004 in hist_iter__top_callback (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, single=true, arg=0x7fffffff9de0) at builtin-top.c:737
> #6  0x0000000000684d2a in hist_entry_iter__add (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, max_stack_depth=127, arg=0x7fffffff9de0) at util/hist.c:1291
> #7  0x00000000004464fe in perf_event__process_sample (tool=0x7fffffff9de0, event=0x10cba70, evsel=0xfae480, sample=0x7fffcf7fd760, machine=0x105ec68) at builtin-top.c:845
> #8  0x0000000000447523 in deliver_event (qe=0x7fffffffa0f8, qevent=0x10cdd60) at builtin-top.c:1211
> #9  0x0000000000648aff in do_flush (oe=0x7fffffffa0f8, show_progress=false) at util/ordered-events.c:245
> #10 0x0000000000648e56 in __ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP, timestamp=0) at util/ordered-events.c:324
> #11 0x0000000000648f40 in ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP) at util/ordered-events.c:342
> #12 0x0000000000447097 in process_thread (arg=0x7fffffff9de0) at builtin-top.c:1120
> #13 0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
> #14 0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78

I'll test on a 14700 later today, but on this one it is working:

root@x1:~# grep -m1 "model name" /proc/cpuinfo 
model name	: 13th Gen Intel(R) Core(TM) i7-1365U
root@x1:~# perf -vv
perf version 6.15.rc2.g8feafba59c51
                   aio: [ on  ]  # HAVE_AIO_SUPPORT
                   bpf: [ on  ]  # HAVE_LIBBPF_SUPPORT
         bpf_skeletons: [ on  ]  # HAVE_BPF_SKEL
            debuginfod: [ on  ]  # HAVE_DEBUGINFOD_SUPPORT
                 dwarf: [ on  ]  # HAVE_LIBDW_SUPPORT
    dwarf_getlocations: [ on  ]  # HAVE_LIBDW_SUPPORT
          dwarf-unwind: [ on  ]  # HAVE_DWARF_UNWIND_SUPPORT
              auxtrace: [ on  ]  # HAVE_AUXTRACE_SUPPORT
                libbfd: [ OFF ]  # HAVE_LIBBFD_SUPPORT ( tip: Deprecated, license incompatibility, use BUILD_NONDISTRO=1 and install binutils-dev[el] )
           libcapstone: [ on  ]  # HAVE_LIBCAPSTONE_SUPPORT
             libcrypto: [ on  ]  # HAVE_LIBCRYPTO_SUPPORT
    libdw-dwarf-unwind: [ on  ]  # HAVE_LIBDW_SUPPORT
                libelf: [ on  ]  # HAVE_LIBELF_SUPPORT
               libnuma: [ on  ]  # HAVE_LIBNUMA_SUPPORT
            libopencsd: [ on  ]  # HAVE_CSTRACE_SUPPORT
               libperl: [ on  ]  # HAVE_LIBPERL_SUPPORT
               libpfm4: [ on  ]  # HAVE_LIBPFM
             libpython: [ on  ]  # HAVE_LIBPYTHON_SUPPORT
              libslang: [ on  ]  # HAVE_SLANG_SUPPORT
         libtraceevent: [ on  ]  # HAVE_LIBTRACEEVENT
             libunwind: [ OFF ]  # HAVE_LIBUNWIND_SUPPORT ( tip: Deprecated, use LIBUNWIND=1 and install libunwind-dev[el] to build with it )
                  lzma: [ on  ]  # HAVE_LZMA_SUPPORT
numa_num_possible_cpus: [ on  ]  # HAVE_LIBNUMA_SUPPORT
                  zlib: [ on  ]  # HAVE_ZLIB_SUPPORT
                  zstd: [ on  ]  # HAVE_ZSTD_SUPPORT
root@x1:~# perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counte

With what is in perf-tools-next/perf-tools-next:

⬢ [acme@toolbx perf-tools-next]$ git log --oneline -10
8feafba59c510be3 (HEAD -> perf-tools-next, x1/perf-tools-next, x1/HEAD) perf test: Add direct off-cpu tests
9557c000768741bb perf record --off-cpu: Add --off-cpu-thresh option
74069a01609ef0f4 perf record --off-cpu: Dump the remaining PERF_SAMPLE_ in sample_type from BPF's stack trace map
8ae7a5769b0a3ac2 perf script: Display off-cpu samples correctly
7de1a87f1ee75743 perf record --off-cpu: Disable perf_event's callchain collection
7f8f56475d585117 perf evsel: Assemble off-cpu samples
d6948f2af24e04ea perf record --off-cpu: Dump off-cpu samples in BPF
282c195906c76ddf perf record --off-cpu: Preparation of off-cpu BPF program
0f72027bb9fb77a2 perf record --off-cpu: Parse off-cpu event
671e943452b18001 perf evsel: Expose evsel__is_offcpu_event() for future use
⬢ [acme@toolbx perf-tools-next]$ 

- Arnaldo

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] perf top: populate PMU capabilities data in perf_env
  2025-05-06 16:56       ` Arnaldo Carvalho de Melo
@ 2025-05-06 18:52         ` Falcon, Thomas
  0 siblings, 0 replies; 6+ messages in thread
From: Falcon, Thomas @ 2025-05-06 18:52 UTC (permalink / raw)
  To: acme@kernel.org
  Cc: alexander.shishkin@linux.intel.com, peterz@infradead.org,
	kan.liang@linux.intel.com, linux-perf-users@vger.kernel.org,
	mingo@redhat.com, Hunter, Adrian, namhyung@kernel.org,
	jolsa@kernel.org, linux-kernel@vger.kernel.org,
	irogers@google.com, mark.rutland@arm.com

On Tue, 2025-05-06 at 13:56 -0300, Arnaldo Carvalho de Melo wrote:
> On Tue, May 06, 2025 at 04:09:37PM +0000, Falcon, Thomas wrote:
> > On Thu, 2025-05-01 at 20:46 +0000, Falcon, Thomas wrote:
> > > On Thu, 2025-05-01 at 13:00 -0700, Ian Rogers wrote:
> > > > On Thu, May 1, 2025 at 11:42 AM Thomas Falcon <thomas.falcon@intel.com> wrote:
> > > > > Calling perf top with brach filters enabled on Intel hybrid CPU's
> > > > > with branch counter event logging support results in a segfault.
> 
> > > > > $ ./perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,\
> > > > >         frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
> > > > > perf: Segmentation fault
> > > > > -------- backtrace --------
> > > > > ./perf() [0x55f460]
> > > > > /lib64/libc.so.6(+0x1a050) [0x7fd8be227050]
> > > > > ./perf() [0x57b4a7]
> > > > > ./perf() [0x561e5a]
> > > > > ./perf() [0x604a81]
> > > > > ./perf() [0x4395b5]
> > > > > ./perf() [0x601732]
> > > > > ./perf() [0x439bc1]
> > > > > ./perf() [0x5d35b3]
> > > > > ./perf() [0x43936c]
> > > > > /lib64/libc.so.6(+0x70ba8) [0x7fd8be27dba8]
> > > > > /lib64/libc.so.6(+0xf4b8c) [0x7fd8be301b8c]
> 
> > > Hi Ian, thanks for reviewing.
> 
> > > > Thanks Thomas. Could you generate this backtrace in GDB? I did write a
> > > > patch to symbolize backtraces like this:
> > > > https://lore.kernel.org/lkml/20250313052952.871958-2-irogers@google.com/
> > > > Sadly without any reviewed tags and unmerged - the code calls routines
> > > > that malloc so it isn't strictly sound if say the backtrace was needed
> > > > from a SEGV in the malloc implementation, it is nicely
> > > > self-referencing the perf APIs, ..
> 
> > > Sorry about that, here is the backtrace I'm seeing when running the perf top command in gdb:
> 
> > > Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
> 
> > > [Switching to Thread 0x7fffcb7fe6c0 (LWP 812169)]
> > > 0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> > > (gdb) backtrace
> > > #0  0x000000000057b4a7 in perf_env.find_br_cntr_info ()
> > > #1  0x0000000000561e5a in addr_map_symbol.account_cycles ()
> > > #2  0x0000000000604a81 in hist.account_cycles ()
> > > #3  0x00000000004395b5 in hist_iter.top_callback ()
> > > #4  0x0000000000601732 in hist_entry_iter.add ()
> > > #5  0x0000000000439bc1 in deliver_event ()
> > > #6  0x00000000005d35b3 in __ordered_events__flush ()
> > > #7  0x000000000043936c in process_thread ()
> > > #8  0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
> > > #9  0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
>  
> > Sorry, let me try this again...
>  
> > Thread 27 "perf" received signal SIGSEGV, Segmentation fault.
> > [Switching to Thread 0x7fffcf7fe6c0 (LWP 940046)]
> > perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
> > 653			*width = env->cpu_pmu_caps ? env->br_cntr_width :
> > (gdb) bt
> > #0  perf_env__find_br_cntr_info (env=0xf328c0 <perf_env>, nr=0x0, width=0x7fffcf7fd2c0) at util/env.c:653
> > #1  0x00000000005ad829 in symbol__account_br_cntr (branch=0x7fffd11f9c00, evsel=0xfae480, offset=20, br_cntr=4) at util/annotate.c:345
> > #2  0x00000000005ada8b in symbol__account_cycles (addr=5580436, start=5580433, sym=0x7fffd00d3010, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:389
> > #3  0x00000000005adc06 in addr_map_symbol__account_cycles (ams=0x7fffd17b1e20, start=0x7fffd17b1f00, cycles=1, evsel=0xfae480, br_cntr=4) at util/annotate.c:422
> > #4  0x0000000000688ab4 in hist__account_cycles (bs=0x10cbaa8, al=0x7fffcf7fd540, sample=0x7fffcf7fd760, nonany_branch_mode=false, total_cycles=0x0, evsel=0xfae480) at util/hist.c:2774
> > #5  0x0000000000446004 in hist_iter__top_callback (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, single=true, arg=0x7fffffff9de0) at builtin-top.c:737
> > #6  0x0000000000684d2a in hist_entry_iter__add (iter=0x7fffcf7fd590, al=0x7fffcf7fd540, max_stack_depth=127, arg=0x7fffffff9de0) at util/hist.c:1291
> > #7  0x00000000004464fe in perf_event__process_sample (tool=0x7fffffff9de0, event=0x10cba70, evsel=0xfae480, sample=0x7fffcf7fd760, machine=0x105ec68) at builtin-top.c:845
> > #8  0x0000000000447523 in deliver_event (qe=0x7fffffffa0f8, qevent=0x10cdd60) at builtin-top.c:1211
> > #9  0x0000000000648aff in do_flush (oe=0x7fffffffa0f8, show_progress=false) at util/ordered-events.c:245
> > #10 0x0000000000648e56 in __ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP, timestamp=0) at util/ordered-events.c:324
> > #11 0x0000000000648f40 in ordered_events__flush (oe=0x7fffffffa0f8, how=OE_FLUSH__TOP) at util/ordered-events.c:342
> > #12 0x0000000000447097 in process_thread (arg=0x7fffffff9de0) at builtin-top.c:1120
> > #13 0x00007ffff6e7dba8 in start_thread (arg=<optimized out>) at pthread_create.c:448
> > #14 0x00007ffff6f01b8c in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
> 
> I'll test on a 14700 later today, but on this one it is working:
> 
> root@x1:~# grep -m1 "model name" /proc/cpuinfo 
> model name	: 13th Gen Intel(R) Core(TM) i7-1365U
> root@x1:~# perf -vv
> perf version 6.15.rc2.g8feafba59c51
>                    aio: [ on  ]  # HAVE_AIO_SUPPORT
>                    bpf: [ on  ]  # HAVE_LIBBPF_SUPPORT
>          bpf_skeletons: [ on  ]  # HAVE_BPF_SKEL
>             debuginfod: [ on  ]  # HAVE_DEBUGINFOD_SUPPORT
>                  dwarf: [ on  ]  # HAVE_LIBDW_SUPPORT
>     dwarf_getlocations: [ on  ]  # HAVE_LIBDW_SUPPORT
>           dwarf-unwind: [ on  ]  # HAVE_DWARF_UNWIND_SUPPORT
>               auxtrace: [ on  ]  # HAVE_AUXTRACE_SUPPORT
>                 libbfd: [ OFF ]  # HAVE_LIBBFD_SUPPORT ( tip: Deprecated, license incompatibility, use BUILD_NONDISTRO=1 and install binutils-dev[el] )
>            libcapstone: [ on  ]  # HAVE_LIBCAPSTONE_SUPPORT
>              libcrypto: [ on  ]  # HAVE_LIBCRYPTO_SUPPORT
>     libdw-dwarf-unwind: [ on  ]  # HAVE_LIBDW_SUPPORT
>                 libelf: [ on  ]  # HAVE_LIBELF_SUPPORT
>                libnuma: [ on  ]  # HAVE_LIBNUMA_SUPPORT
>             libopencsd: [ on  ]  # HAVE_CSTRACE_SUPPORT
>                libperl: [ on  ]  # HAVE_LIBPERL_SUPPORT
>                libpfm4: [ on  ]  # HAVE_LIBPFM
>              libpython: [ on  ]  # HAVE_LIBPYTHON_SUPPORT
>               libslang: [ on  ]  # HAVE_SLANG_SUPPORT
>          libtraceevent: [ on  ]  # HAVE_LIBTRACEEVENT
>              libunwind: [ OFF ]  # HAVE_LIBUNWIND_SUPPORT ( tip: Deprecated, use LIBUNWIND=1 and install libunwind-dev[el] to build with it )
>                   lzma: [ on  ]  # HAVE_LZMA_SUPPORT
> numa_num_possible_cpus: [ on  ]  # HAVE_LIBNUMA_SUPPORT
>                   zlib: [ on  ]  # HAVE_ZLIB_SUPPORT
>                   zstd: [ on  ]  # HAVE_ZSTD_SUPPORT
> root@x1:~# perf top -e '{cpu_core/cpu-cycles/,cpu_core/event=0xc6,umask=0x3,frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counte
> 

I should have been clearer in my commit message but it requires PMU's with "branch_counter_nr" and
"branch_counter_width" capabilities to recreate.

$ ls /sys/devices/cpu_core/caps
branch_counter_nr  branch_counter_width  branches  max_precise  pmu_name

$ sudo perf top -e '{cpu_core/cpu-
cycles/,cpu_core/event=0xc6,umask=0x3,frontend=0x11,name=frontend_retired_dsb_miss/}' -j any,counter
perf: Segmentation fault
-------- backtrace --------
perf(+0xfa734) [0x561b5b19f734]
/lib64/libc.so.6(+0x1a050) [0x7faa77a27050]
perf(perf_env__find_br_cntr_info+0x77) [0x561b5b1b8377]
perf(addr_map_symbol__account_cycles+0x22e) [0x561b5b1a27de]
perf(hist__account_cycles+0x241) [0x561b5b23dae1]
perf(+0x2c459) [0x561b5b0d1459]
perf(hist_entry_iter__add+0xe2) [0x561b5b239a32]
perf(+0x2ca08) [0x561b5b0d1a08]
perf(+0x168373) [0x561b5b20d373]
perf(+0x2c214) [0x561b5b0d1214]
/lib64/libc.so.6(+0x71168) [0x7faa77a7e168]
/lib64/libc.so.6(+0xf514c) [0x7faa77b0214c]

Thanks,
Tom

> Add --off-cpu-thresh option
> 74069a01609ef0f4 perf record --off-cpu: Dump the remaining PERF_SAMPLE_ in sample_type from BPF's stack trace map
> 8ae7a5769b0a3ac2 perf script: Display off-cpu samples correctly
> 7de1a87f1ee75743 perf record --off-cpu: Disable perf_event's callchain collection
> 7f8f56475d585117 perf evsel: Assemble off-cpu samples
> d6948f2af24e04ea perf record --off-cpu: Dump off-cpu samples in BPF
> 282c195906c76ddf perf record --off-cpu: Preparation of off-cpu BPF program
> 0f72027bb9fb77a2 perf record --off-cpu: Parse off-cpu event
> 671e943452b18001 perf evsel: Expose evsel__is_offcpu_event() for future use
> ⬢ [acme@toolbx perf-tools-next]$ 
> 
> - Arnaldo


^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2025-05-06 18:52 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-05-01 18:41 [PATCH] perf top: populate PMU capabilities data in perf_env Thomas Falcon
2025-05-01 20:00 ` Ian Rogers
2025-05-01 20:46   ` Falcon, Thomas
2025-05-06 16:09     ` Falcon, Thomas
2025-05-06 16:56       ` Arnaldo Carvalho de Melo
2025-05-06 18:52         ` Falcon, Thomas

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.