[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20201229114828.GG521329@kernel.org>
Date: Tue, 29 Dec 2020 08:48:28 -0300
From: Arnaldo Carvalho de Melo <acme@...nel.org>
To: Namhyung Kim <namhyung@...nel.org>
Cc: Song Liu <songliubraving@...com>,
linux-kernel <linux-kernel@...r.kernel.org>,
Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Mark Rutland <mark.rutland@....com>,
Jiri Olsa <jolsa@...hat.com>, kernel-team@...com
Subject: Re: [PATCH v6 2/4] perf: support build BPF skeletons with perf
Em Tue, Dec 29, 2020 at 04:01:41PM +0900, Namhyung Kim escreveu:
> On Tue, Dec 29, 2020 at 2:41 AM Song Liu <songliubraving@...com> wrote:
> > BPF programs are useful in perf to profile BPF programs. BPF skeleton is
> I'm having difficulties understanding the first sentence - looks like a
> recursion. :) So do you want to use two (or more) BPF programs?
Yeah, we use perf to perf perf, so we need to use bpf with perf to perf
bpf :-)
Look at tools/perf/util/bpf_skel/bpf_prog_profiler.bpf.c, the BPF
skeleton used to create the in-kernel scaffold to profile BPF programs.
It uses two BPF programs (fentry/XXX and fexit/XXX) and some a
PERF_EVENT_ARRAY map and an array to diff counters read at exit from
counters read at exit of the profiled BPF programs and then accumulate
those diffs in another PERCPU_ARRAY.
This all ends up composing a "BPF PMU" that is what the userspace perf
tooling will read (from "accum_readings" BPF map) and 'perf stat' will
consume as if reading from an "old style perf counter" :-)
Song, did I get it right? :-)
For convenience, it is below:
- Arnaldo
[acme@...e perf]$ cat tools/perf/util/bpf_skel/bpf_prog_profiler.bpf.c
// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
// Copyright (c) 2020 Facebook
#include <linux/bpf.h>
#include <bpf/bpf_helpers.h>
#include <bpf/bpf_tracing.h>
/* map of perf event fds, num_cpu * num_metric entries */
struct {
__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
__uint(key_size, sizeof(__u32));
__uint(value_size, sizeof(int));
} events SEC(".maps");
/* readings at fentry */
struct {
__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
__uint(key_size, sizeof(__u32));
__uint(value_size, sizeof(struct bpf_perf_event_value));
__uint(max_entries, 1);
} fentry_readings SEC(".maps");
/* accumulated readings */
struct {
__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
__uint(key_size, sizeof(__u32));
__uint(value_size, sizeof(struct bpf_perf_event_value));
__uint(max_entries, 1);
} accum_readings SEC(".maps");
const volatile __u32 num_cpu = 1;
SEC("fentry/XXX")
int BPF_PROG(fentry_XXX)
{
__u32 key = bpf_get_smp_processor_id();
struct bpf_perf_event_value *ptr;
__u32 zero = 0;
long err;
/* look up before reading, to reduce error */
ptr = bpf_map_lookup_elem(&fentry_readings, &zero);
if (!ptr)
return 0;
err = bpf_perf_event_read_value(&events, key, ptr, sizeof(*ptr));
if (err)
return 0;
return 0;
}
static inline void
fexit_update_maps(struct bpf_perf_event_value *after)
{
struct bpf_perf_event_value *before, diff, *accum;
__u32 zero = 0;
before = bpf_map_lookup_elem(&fentry_readings, &zero);
/* only account samples with a valid fentry_reading */
if (before && before->counter) {
struct bpf_perf_event_value *accum;
diff.counter = after->counter - before->counter;
diff.enabled = after->enabled - before->enabled;
diff.running = after->running - before->running;
accum = bpf_map_lookup_elem(&accum_readings, &zero);
if (accum) {
accum->counter += diff.counter;
accum->enabled += diff.enabled;
accum->running += diff.running;
}
}
}
SEC("fexit/XXX")
int BPF_PROG(fexit_XXX)
{
struct bpf_perf_event_value reading;
__u32 cpu = bpf_get_smp_processor_id();
__u32 one = 1, zero = 0;
int err;
/* read all events before updating the maps, to reduce error */
err = bpf_perf_event_read_value(&events, cpu, &reading, sizeof(reading));
if (err)
return 0;
fexit_update_maps(&reading);
return 0;
}
char LICENSE[] SEC("license") = "Dual BSD/GPL";
[acme@...e perf]$
Powered by blists - more mailing lists