[v7] perf record --off-cpu: Dump off-cpu samples directly

[PATCH v7 04/10] perf record --off-cpu: Preparation of off-cpu BPF program

Posted by Howard Chu 1 year, 3 months ago

Set the perf_event map in BPF for dumping off-cpu samples.

Set the offcpu_thresh to specify the threshold.

Signed-off-by: Howard Chu <howardchu95@gmail.com>
---
 tools/perf/util/bpf_off_cpu.c          | 20 ++++++++++++++++++++
 tools/perf/util/bpf_skel/off_cpu.bpf.c |  2 ++
 2 files changed, 22 insertions(+)

diff --git a/tools/perf/util/bpf_off_cpu.c b/tools/perf/util/bpf_off_cpu.c
index 558c5e5c2dc3..cfe5b17393e9 100644
--- a/tools/perf/util/bpf_off_cpu.c
+++ b/tools/perf/util/bpf_off_cpu.c
@@ -13,6 +13,7 @@
 #include "util/cgroup.h"
 #include "util/strlist.h"
 #include <bpf/bpf.h>
+#include <internal/xyarray.h>
 
 #include "bpf_skel/off_cpu.skel.h"
 
@@ -73,6 +74,23 @@ static void off_cpu_start(void *arg)
 		bpf_map_update_elem(fd, &pid, &val, BPF_ANY);
 	}
 
+	/* update BPF perf_event map */
+	evsel = evlist__find_evsel_by_str(evlist, OFFCPU_EVENT);
+	if (evsel == NULL) {
+		pr_err("%s evsel not found\n", OFFCPU_EVENT);
+		return;
+	}
+
+	perf_cpu_map__for_each_cpu(pcpu, i, evsel->core.cpus) {
+		err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
+					   xyarray__entry(evsel->core.fd, i, 0),
+					   sizeof(__u32), BPF_ANY);
+		if (err) {
+			pr_err("Failed to update perf event map for direct off-cpu dumping\n");
+			return;
+		}
+	}
+
 	skel->bss->enabled = 1;
 }
 
@@ -261,6 +279,8 @@ int off_cpu_prepare(struct evlist *evlist, struct target *target,
 		}
 	}
 
+	skel->bss->offcpu_thresh = opts->off_cpu_thresh * 1000;
+
 	err = off_cpu_bpf__attach(skel);
 	if (err) {
 		pr_err("Failed to attach off-cpu BPF skeleton\n");
diff --git a/tools/perf/util/bpf_skel/off_cpu.bpf.c b/tools/perf/util/bpf_skel/off_cpu.bpf.c
index c152116df72f..dc6acafb9353 100644
--- a/tools/perf/util/bpf_skel/off_cpu.bpf.c
+++ b/tools/perf/util/bpf_skel/off_cpu.bpf.c
@@ -97,6 +97,8 @@ const volatile bool uses_cgroup_v1 = false;
 
 int perf_subsys_id = -1;
 
+__u64 offcpu_thresh;
+
 /*
  * Old kernel used to call it task_struct->state and now it's '__state'.
  * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
-- 
2.43.0

Re: [PATCH v7 04/10] perf record --off-cpu: Preparation of off-cpu BPF program

Posted by Ian Rogers 1 year, 2 months ago

On Fri, Nov 8, 2024 at 12:41 PM Howard Chu <howardchu95@gmail.com> wrote:
>
> Set the perf_event map in BPF for dumping off-cpu samples.
>
> Set the offcpu_thresh to specify the threshold.
>
> Signed-off-by: Howard Chu <howardchu95@gmail.com>
> ---
>  tools/perf/util/bpf_off_cpu.c          | 20 ++++++++++++++++++++
>  tools/perf/util/bpf_skel/off_cpu.bpf.c |  2 ++
>  2 files changed, 22 insertions(+)
>
> diff --git a/tools/perf/util/bpf_off_cpu.c b/tools/perf/util/bpf_off_cpu.c
> index 558c5e5c2dc3..cfe5b17393e9 100644
> --- a/tools/perf/util/bpf_off_cpu.c
> +++ b/tools/perf/util/bpf_off_cpu.c
> @@ -13,6 +13,7 @@
>  #include "util/cgroup.h"
>  #include "util/strlist.h"
>  #include <bpf/bpf.h>
> +#include <internal/xyarray.h>
>
>  #include "bpf_skel/off_cpu.skel.h"
>
> @@ -73,6 +74,23 @@ static void off_cpu_start(void *arg)
>                 bpf_map_update_elem(fd, &pid, &val, BPF_ANY);
>         }
>
> +       /* update BPF perf_event map */
> +       evsel = evlist__find_evsel_by_str(evlist, OFFCPU_EVENT);
> +       if (evsel == NULL) {
> +               pr_err("%s evsel not found\n", OFFCPU_EVENT);
> +               return;
> +       }
> +
> +       perf_cpu_map__for_each_cpu(pcpu, i, evsel->core.cpus) {
> +               err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
> +                                          xyarray__entry(evsel->core.fd, i, 0),
> +                                          sizeof(__u32), BPF_ANY);
> +               if (err) {
> +                       pr_err("Failed to update perf event map for direct off-cpu dumping\n");
> +                       return;
> +               }
> +       }
> +
>         skel->bss->enabled = 1;
>  }
>
> @@ -261,6 +279,8 @@ int off_cpu_prepare(struct evlist *evlist, struct target *target,
>                 }
>         }
>
> +       skel->bss->offcpu_thresh = opts->off_cpu_thresh * 1000;
> +
>         err = off_cpu_bpf__attach(skel);
>         if (err) {
>                 pr_err("Failed to attach off-cpu BPF skeleton\n");
> diff --git a/tools/perf/util/bpf_skel/off_cpu.bpf.c b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> index c152116df72f..dc6acafb9353 100644
> --- a/tools/perf/util/bpf_skel/off_cpu.bpf.c
> +++ b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> @@ -97,6 +97,8 @@ const volatile bool uses_cgroup_v1 = false;
>
>  int perf_subsys_id = -1;
>
> +__u64 offcpu_thresh;

nit: would be nice to include the unit in the variable name, ie
offcpu_thresh_us.

Reviewed-by: Ian Rogers <irogers@google.com>

Thanks,
Ian

> +
>  /*
>   * Old kernel used to call it task_struct->state and now it's '__state'.
>   * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
> --
> 2.43.0
>

Re: [PATCH v7 04/10] perf record --off-cpu: Preparation of off-cpu BPF program

Posted by Howard Chu 1 year, 2 months ago

Hi Ian,


On Mon, Nov 11, 2024 at 9:47 AM Ian Rogers <irogers@google.com> wrote:
>
> On Fri, Nov 8, 2024 at 12:41 PM Howard Chu <howardchu95@gmail.com> wrote:
> >
> > Set the perf_event map in BPF for dumping off-cpu samples.
> >
> > Set the offcpu_thresh to specify the threshold.
> >
> > Signed-off-by: Howard Chu <howardchu95@gmail.com>
> > ---
> >  tools/perf/util/bpf_off_cpu.c          | 20 ++++++++++++++++++++
> >  tools/perf/util/bpf_skel/off_cpu.bpf.c |  2 ++
> >  2 files changed, 22 insertions(+)
> >
> > diff --git a/tools/perf/util/bpf_off_cpu.c b/tools/perf/util/bpf_off_cpu.c
> > index 558c5e5c2dc3..cfe5b17393e9 100644
> > --- a/tools/perf/util/bpf_off_cpu.c
> > +++ b/tools/perf/util/bpf_off_cpu.c
> > @@ -13,6 +13,7 @@
> >  #include "util/cgroup.h"
> >  #include "util/strlist.h"
> >  #include <bpf/bpf.h>
> > +#include <internal/xyarray.h>
> >
> >  #include "bpf_skel/off_cpu.skel.h"
> >
> > @@ -73,6 +74,23 @@ static void off_cpu_start(void *arg)
> >                 bpf_map_update_elem(fd, &pid, &val, BPF_ANY);
> >         }
> >
> > +       /* update BPF perf_event map */
> > +       evsel = evlist__find_evsel_by_str(evlist, OFFCPU_EVENT);
> > +       if (evsel == NULL) {
> > +               pr_err("%s evsel not found\n", OFFCPU_EVENT);
> > +               return;
> > +       }
> > +
> > +       perf_cpu_map__for_each_cpu(pcpu, i, evsel->core.cpus) {
> > +               err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
> > +                                          xyarray__entry(evsel->core.fd, i, 0),
> > +                                          sizeof(__u32), BPF_ANY);
> > +               if (err) {
> > +                       pr_err("Failed to update perf event map for direct off-cpu dumping\n");
> > +                       return;
> > +               }
> > +       }
> > +
> >         skel->bss->enabled = 1;
> >  }
> >
> > @@ -261,6 +279,8 @@ int off_cpu_prepare(struct evlist *evlist, struct target *target,
> >                 }
> >         }
> >
> > +       skel->bss->offcpu_thresh = opts->off_cpu_thresh * 1000;
> > +
> >         err = off_cpu_bpf__attach(skel);
> >         if (err) {
> >                 pr_err("Failed to attach off-cpu BPF skeleton\n");
> > diff --git a/tools/perf/util/bpf_skel/off_cpu.bpf.c b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > index c152116df72f..dc6acafb9353 100644
> > --- a/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > +++ b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > @@ -97,6 +97,8 @@ const volatile bool uses_cgroup_v1 = false;
> >
> >  int perf_subsys_id = -1;
> >
> > +__u64 offcpu_thresh;
>
> nit: would be nice to include the unit in the variable name, ie
> offcpu_thresh_us.

I didn't see it yesterday sorry, but in BPF the unit should be
nanosecond, so I'll do offcpu_thresh_ns.

Thanks,
Howard

>
> Reviewed-by: Ian Rogers <irogers@google.com>
>
> Thanks,
> Ian
>
> > +
> >  /*
> >   * Old kernel used to call it task_struct->state and now it's '__state'.
> >   * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
> > --
> > 2.43.0
> >

Re: [PATCH v7 04/10] perf record --off-cpu: Preparation of off-cpu BPF program

Posted by Howard Chu 1 year, 2 months ago

Hello,

On Mon, Nov 11, 2024 at 9:47 AM Ian Rogers <irogers@google.com> wrote:
>
> On Fri, Nov 8, 2024 at 12:41 PM Howard Chu <howardchu95@gmail.com> wrote:
> >
> > Set the perf_event map in BPF for dumping off-cpu samples.
> >
> > Set the offcpu_thresh to specify the threshold.
> >
> > Signed-off-by: Howard Chu <howardchu95@gmail.com>
> > ---
> >  tools/perf/util/bpf_off_cpu.c          | 20 ++++++++++++++++++++
> >  tools/perf/util/bpf_skel/off_cpu.bpf.c |  2 ++
> >  2 files changed, 22 insertions(+)
> >
> > diff --git a/tools/perf/util/bpf_off_cpu.c b/tools/perf/util/bpf_off_cpu.c
> > index 558c5e5c2dc3..cfe5b17393e9 100644
> > --- a/tools/perf/util/bpf_off_cpu.c
> > +++ b/tools/perf/util/bpf_off_cpu.c
> > @@ -13,6 +13,7 @@
> >  #include "util/cgroup.h"
> >  #include "util/strlist.h"
> >  #include <bpf/bpf.h>
> > +#include <internal/xyarray.h>
> >
> >  #include "bpf_skel/off_cpu.skel.h"
> >
> > @@ -73,6 +74,23 @@ static void off_cpu_start(void *arg)
> >                 bpf_map_update_elem(fd, &pid, &val, BPF_ANY);
> >         }
> >
> > +       /* update BPF perf_event map */
> > +       evsel = evlist__find_evsel_by_str(evlist, OFFCPU_EVENT);
> > +       if (evsel == NULL) {
> > +               pr_err("%s evsel not found\n", OFFCPU_EVENT);
> > +               return;
> > +       }
> > +
> > +       perf_cpu_map__for_each_cpu(pcpu, i, evsel->core.cpus) {
> > +               err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
> > +                                          xyarray__entry(evsel->core.fd, i, 0),
> > +                                          sizeof(__u32), BPF_ANY);
> > +               if (err) {
> > +                       pr_err("Failed to update perf event map for direct off-cpu dumping\n");
> > +                       return;
> > +               }
> > +       }
> > +
> >         skel->bss->enabled = 1;
> >  }
> >
> > @@ -261,6 +279,8 @@ int off_cpu_prepare(struct evlist *evlist, struct target *target,
> >                 }
> >         }
> >
> > +       skel->bss->offcpu_thresh = opts->off_cpu_thresh * 1000;
> > +
> >         err = off_cpu_bpf__attach(skel);
> >         if (err) {
> >                 pr_err("Failed to attach off-cpu BPF skeleton\n");
> > diff --git a/tools/perf/util/bpf_skel/off_cpu.bpf.c b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > index c152116df72f..dc6acafb9353 100644
> > --- a/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > +++ b/tools/perf/util/bpf_skel/off_cpu.bpf.c
> > @@ -97,6 +97,8 @@ const volatile bool uses_cgroup_v1 = false;
> >
> >  int perf_subsys_id = -1;
> >
> > +__u64 offcpu_thresh;
>
> nit: would be nice to include the unit in the variable name, ie
> offcpu_thresh_us.

Sure I will, thanks.

>
> Reviewed-by: Ian Rogers <irogers@google.com>
>
> Thanks,
> Ian
>
> > +
> >  /*
> >   * Old kernel used to call it task_struct->state and now it's '__state'.
> >   * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
> > --
> > 2.43.0
> >

Thanks,
Howard