libperf: CPU map question

libperf: CPU map question

[Tzvetomir Stoyanov]

Hello,
I'm trying to use libperf as an interface to perf from an application,
but I face some difficulties.  I cannot understand how perf_cpu_map_
APIs should be used. My use case is:
 I want to iterate over the CPUs from the map. I'm using
perf_evsel__cpus() API to get the map and perf_cpu_map__for_each_cpu()
macros to iterate over the map. However, using that logic I get a
struct perf_cpu for each CPU from the map. That structure seems to be
private and I cannot access the actual .cpu id. Am I missing something
?
Will be glad for any help and clarification.

Thanks!
-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center

Re: libperf: CPU map question

[Arnaldo Carvalho de Melo]

Em Mon, Feb 14, 2022 at 06:54:38PM +0200, Tzvetomir Stoyanov escreveu:
> Hello,
> I'm trying to use libperf as an interface to perf from an application,
> but I face some difficulties.  I cannot understand how perf_cpu_map_
> APIs should be used. My use case is:
>  I want to iterate over the CPUs from the map. I'm using
> perf_evsel__cpus() API to get the map and perf_cpu_map__for_each_cpu()
> macros to iterate over the map. However, using that logic I get a
> struct perf_cpu for each CPU from the map. That structure seems to be
> private and I cannot access the actual .cpu id. Am I missing something
> ?
> Will be glad for any help and clarification.

Jiri, can you help?

- Arnaldo

Re: libperf: CPU map question

[Jiri Olsa]

On Mon, Feb 14, 2022 at 06:54:38PM +0200, Tzvetomir Stoyanov wrote:
> Hello,
> I'm trying to use libperf as an interface to perf from an application,
> but I face some difficulties.  I cannot understand how perf_cpu_map_
> APIs should be used. My use case is:
>  I want to iterate over the CPUs from the map. I'm using
> perf_evsel__cpus() API to get the map and perf_cpu_map__for_each_cpu()
> macros to iterate over the map. However, using that logic I get a
> struct perf_cpu for each CPU from the map. That structure seems to be
> private and I cannot access the actual .cpu id. Am I missing something
> ?
> Will be glad for any help and clarification.

yea it's a bug ;-) there was a change recently that wrapped cpu
in struct perf_cpu to distinguish it from cpu indexes

  6d18804b963b perf cpumap: Give CPUs their own type

not sure there's workaround for that.. but I think change below should fix it

Ian, would this be ok? we need to add some test for this

jirka


---
diff --git a/tools/lib/perf/include/internal/cpumap.h b/tools/lib/perf/include/internal/cpumap.h
index 581f9ffb4237..1973a18c096b 100644
--- a/tools/lib/perf/include/internal/cpumap.h
+++ b/tools/lib/perf/include/internal/cpumap.h
@@ -3,11 +3,7 @@
 #define __LIBPERF_INTERNAL_CPUMAP_H
 
 #include <linux/refcount.h>
-
-/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
-struct perf_cpu {
-	int cpu;
-};
+#include <perf/cpumap.h>
 
 /**
  * A sized, reference counted, sorted array of integers representing CPU
diff --git a/tools/lib/perf/include/perf/cpumap.h b/tools/lib/perf/include/perf/cpumap.h
index 15b8faafd615..4a2edbdb5e2b 100644
--- a/tools/lib/perf/include/perf/cpumap.h
+++ b/tools/lib/perf/include/perf/cpumap.h
@@ -7,6 +7,11 @@
 #include <stdio.h>
 #include <stdbool.h>
 
+/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
+struct perf_cpu {
+	int cpu;
+};
+
 LIBPERF_API struct perf_cpu_map *perf_cpu_map__dummy_new(void);
 LIBPERF_API struct perf_cpu_map *perf_cpu_map__default_new(void);
 LIBPERF_API struct perf_cpu_map *perf_cpu_map__new(const char *cpu_list);

Re: libperf: CPU map question

[Ian Rogers]

On Mon, Feb 14, 2022 at 1:22 PM Jiri Olsa <olsajiri@gmail.com> wrote:
>
> On Mon, Feb 14, 2022 at 06:54:38PM +0200, Tzvetomir Stoyanov wrote:
> > Hello,
> > I'm trying to use libperf as an interface to perf from an application,
> > but I face some difficulties.  I cannot understand how perf_cpu_map_
> > APIs should be used. My use case is:
> >  I want to iterate over the CPUs from the map. I'm using
> > perf_evsel__cpus() API to get the map and perf_cpu_map__for_each_cpu()
> > macros to iterate over the map. However, using that logic I get a
> > struct perf_cpu for each CPU from the map. That structure seems to be
> > private and I cannot access the actual .cpu id. Am I missing something
> > ?
> > Will be glad for any help and clarification.
>
> yea it's a bug ;-) there was a change recently that wrapped cpu
> in struct perf_cpu to distinguish it from cpu indexes
>
>   6d18804b963b perf cpumap: Give CPUs their own type
>
> not sure there's workaround for that.. but I think change below should fix it
>
> Ian, would this be ok? we need to add some test for this

This looks good to me, thanks! Agreed on the test.

Thanks,
Ian

> jirka
>
>
> ---
> diff --git a/tools/lib/perf/include/internal/cpumap.h b/tools/lib/perf/include/internal/cpumap.h
> index 581f9ffb4237..1973a18c096b 100644
> --- a/tools/lib/perf/include/internal/cpumap.h
> +++ b/tools/lib/perf/include/internal/cpumap.h
> @@ -3,11 +3,7 @@
>  #define __LIBPERF_INTERNAL_CPUMAP_H
>
>  #include <linux/refcount.h>
> -
> -/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
> -struct perf_cpu {
> -       int cpu;
> -};
> +#include <perf/cpumap.h>
>
>  /**
>   * A sized, reference counted, sorted array of integers representing CPU
> diff --git a/tools/lib/perf/include/perf/cpumap.h b/tools/lib/perf/include/perf/cpumap.h
> index 15b8faafd615..4a2edbdb5e2b 100644
> --- a/tools/lib/perf/include/perf/cpumap.h
> +++ b/tools/lib/perf/include/perf/cpumap.h
> @@ -7,6 +7,11 @@
>  #include <stdio.h>
>  #include <stdbool.h>
>
> +/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
> +struct perf_cpu {
> +       int cpu;
> +};
> +
>  LIBPERF_API struct perf_cpu_map *perf_cpu_map__dummy_new(void);
>  LIBPERF_API struct perf_cpu_map *perf_cpu_map__default_new(void);
>  LIBPERF_API struct perf_cpu_map *perf_cpu_map__new(const char *cpu_list);

Re: libperf: CPU map question

[Tzvetomir Stoyanov]

On Tue, Feb 15, 2022 at 12:18 AM Ian Rogers <irogers@google.com> wrote:
>
> On Mon, Feb 14, 2022 at 1:22 PM Jiri Olsa <olsajiri@gmail.com> wrote:
> >
> > On Mon, Feb 14, 2022 at 06:54:38PM +0200, Tzvetomir Stoyanov wrote:
> > > Hello,
> > > I'm trying to use libperf as an interface to perf from an application,
> > > but I face some difficulties.  I cannot understand how perf_cpu_map_
> > > APIs should be used. My use case is:
> > >  I want to iterate over the CPUs from the map. I'm using
> > > perf_evsel__cpus() API to get the map and perf_cpu_map__for_each_cpu()
> > > macros to iterate over the map. However, using that logic I get a
> > > struct perf_cpu for each CPU from the map. That structure seems to be
> > > private and I cannot access the actual .cpu id. Am I missing something
> > > ?
> > > Will be glad for any help and clarification.
> >
> > yea it's a bug ;-) there was a change recently that wrapped cpu
> > in struct perf_cpu to distinguish it from cpu indexes
> >
> >   6d18804b963b perf cpumap: Give CPUs their own type
> >
> > not sure there's workaround for that.. but I think change below should fix it
> >
> > Ian, would this be ok? we need to add some test for this
>
> This looks good to me, thanks! Agreed on the test.
>
> Thanks,
> Ian
>
> > jirka
> >

I applied the patch, it solves my problem. Thanks!

I have one more problem, related to thread map allocation. I submitted
a patch a few days ago. Do not know if you have time to look at it,
I'll ping you on the other thread. Will be glad for any help or
comments on the patch.

> >
> > ---
> > diff --git a/tools/lib/perf/include/internal/cpumap.h b/tools/lib/perf/include/internal/cpumap.h
> > index 581f9ffb4237..1973a18c096b 100644
> > --- a/tools/lib/perf/include/internal/cpumap.h
> > +++ b/tools/lib/perf/include/internal/cpumap.h
> > @@ -3,11 +3,7 @@
> >  #define __LIBPERF_INTERNAL_CPUMAP_H
> >
> >  #include <linux/refcount.h>
> > -
> > -/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
> > -struct perf_cpu {
> > -       int cpu;
> > -};
> > +#include <perf/cpumap.h>
> >
> >  /**
> >   * A sized, reference counted, sorted array of integers representing CPU
> > diff --git a/tools/lib/perf/include/perf/cpumap.h b/tools/lib/perf/include/perf/cpumap.h
> > index 15b8faafd615..4a2edbdb5e2b 100644
> > --- a/tools/lib/perf/include/perf/cpumap.h
> > +++ b/tools/lib/perf/include/perf/cpumap.h
> > @@ -7,6 +7,11 @@
> >  #include <stdio.h>
> >  #include <stdbool.h>
> >
> > +/** A wrapper around a CPU to avoid confusion with the perf_cpu_map's map's indices. */
> > +struct perf_cpu {
> > +       int cpu;
> > +};
> > +
> >  LIBPERF_API struct perf_cpu_map *perf_cpu_map__dummy_new(void);
> >  LIBPERF_API struct perf_cpu_map *perf_cpu_map__default_new(void);
> >  LIBPERF_API struct perf_cpu_map *perf_cpu_map__new(const char *cpu_list);



-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center