[PATCH bpf-next] libbpf: add API to consume the perf ring buffer content

[PATCH bpf-next] libbpf: add API to consume the perf ring buffer content

[Eelco Chaudron]

This new API, perf_buffer__consume, can be used as follows:
- When you have a perf ring where wakeup_events is higher than 1,
  and you have remaining data in the rings you would like to pull
  out on exit (or maybe based on a timeout).
- For low latency cases where you burn a CPU that constantly polls
  the queues.

Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
---
 tools/lib/bpf/libbpf.c   |   23 +++++++++++++++++++++++
 tools/lib/bpf/libbpf.h   |    1 +
 tools/lib/bpf/libbpf.map |    1 +
 3 files changed, 25 insertions(+)

diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index fa04cbe547ed..cbef3dac7507 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -8456,6 +8456,29 @@ int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
 	return cnt < 0 ? -errno : cnt;
 }
 
+int perf_buffer__consume(struct perf_buffer *pb)
+{
+	int i;
+
+	if (!pb)
+		return -EINVAL;
+
+	if (!pb->cpu_bufs)
+		return 0;
+
+	for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
+		int err;
+		struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
+
+		err = perf_buffer__process_records(pb, cpu_buf);
+		if (err) {
+			pr_warn("error while processing records: %d\n", err);
+			return err;
+		}
+	}
+	return 0;
+}
+
 struct bpf_prog_info_array_desc {
 	int	array_offset;	/* e.g. offset of jited_prog_insns */
 	int	count_offset;	/* e.g. offset of jited_prog_len */
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
index 8ea69558f0a8..1e2e399a5f2c 100644
--- a/tools/lib/bpf/libbpf.h
+++ b/tools/lib/bpf/libbpf.h
@@ -533,6 +533,7 @@ perf_buffer__new_raw(int map_fd, size_t page_cnt,
 
 LIBBPF_API void perf_buffer__free(struct perf_buffer *pb);
 LIBBPF_API int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms);
+LIBBPF_API int perf_buffer__consume(struct perf_buffer *pb);
 
 typedef enum bpf_perf_event_ret
 	(*bpf_perf_event_print_t)(struct perf_event_header *hdr,
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
index 0133d469d30b..381a7342ecfc 100644
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -262,4 +262,5 @@ LIBBPF_0.0.9 {
 		bpf_link_get_fd_by_id;
 		bpf_link_get_next_id;
 		bpf_program__attach_iter;
+		perf_buffer__consume;
 } LIBBPF_0.0.8;

Re: [PATCH bpf-next] libbpf: add API to consume the perf ring buffer content

[Andrii Nakryiko]

On Mon, May 25, 2020 at 2:01 PM Eelco Chaudron <echaudro@redhat.com> wrote:
>
> This new API, perf_buffer__consume, can be used as follows:

I wonder, was it inspired by yet-to-be committed
ring_buffer__consume() or it's just a coincidence?

> - When you have a perf ring where wakeup_events is higher than 1,
>   and you have remaining data in the rings you would like to pull
>   out on exit (or maybe based on a timeout).
> - For low latency cases where you burn a CPU that constantly polls
>   the queues.
>
> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
> ---
>  tools/lib/bpf/libbpf.c   |   23 +++++++++++++++++++++++
>  tools/lib/bpf/libbpf.h   |    1 +
>  tools/lib/bpf/libbpf.map |    1 +
>  3 files changed, 25 insertions(+)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index fa04cbe547ed..cbef3dac7507 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -8456,6 +8456,29 @@ int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
>         return cnt < 0 ? -errno : cnt;
>  }
>
> +int perf_buffer__consume(struct perf_buffer *pb)
> +{
> +       int i;
> +
> +       if (!pb)
> +               return -EINVAL;

we don't check this in perf_buffer__poll, IMO, checking this in every
"method" is an overkill.

> +
> +       if (!pb->cpu_bufs)
> +               return 0;

no need to check. It's either non-NULL for valid perf_buffer, or
calloc could return NULL if pb->cpu_cnt is zero (not sure it's
possible, but still), but then loop below will never access
pb->cpu_bufs[i].

> +
> +       for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {

I think pb->cpu_bufs[i] check is wrong, it will stop iteration
prematurely if cpu_bufs are sparsely populated. So move check inside
and continue loop if NULL.

> +               int err;

nit: declare it together with "i" above, similar to how
perf_buffer__poll does it

> +               struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
> +
> +               err = perf_buffer__process_records(pb, cpu_buf);
> +               if (err) {
> +                       pr_warn("error while processing records: %d\n", err);
> +                       return err;
> +               }
> +       }
> +       return 0;
> +}
> +
>  struct bpf_prog_info_array_desc {
>         int     array_offset;   /* e.g. offset of jited_prog_insns */
>         int     count_offset;   /* e.g. offset of jited_prog_len */
> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> index 8ea69558f0a8..1e2e399a5f2c 100644
> --- a/tools/lib/bpf/libbpf.h
> +++ b/tools/lib/bpf/libbpf.h
> @@ -533,6 +533,7 @@ perf_buffer__new_raw(int map_fd, size_t page_cnt,
>
>  LIBBPF_API void perf_buffer__free(struct perf_buffer *pb);
>  LIBBPF_API int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms);
> +LIBBPF_API int perf_buffer__consume(struct perf_buffer *pb);
>
>  typedef enum bpf_perf_event_ret
>         (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> index 0133d469d30b..381a7342ecfc 100644
> --- a/tools/lib/bpf/libbpf.map
> +++ b/tools/lib/bpf/libbpf.map
> @@ -262,4 +262,5 @@ LIBBPF_0.0.9 {
>                 bpf_link_get_fd_by_id;
>                 bpf_link_get_next_id;
>                 bpf_program__attach_iter;
> +               perf_buffer__consume;
>  } LIBBPF_0.0.8;
>

Re: [PATCH bpf-next] libbpf: add API to consume the perf ring buffer content

[Eelco Chaudron]

On 26 May 2020, at 7:29, Andrii Nakryiko wrote:

> On Mon, May 25, 2020 at 2:01 PM Eelco Chaudron <echaudro@redhat.com> 
> wrote:
>>
>> This new API, perf_buffer__consume, can be used as follows:
>
> I wonder, was it inspired by yet-to-be committed
> ring_buffer__consume() or it's just a coincidence?

Just coincidence, I was needing a function to flush the remaining ring 
entries, as I was using a larger wakeup_events value.
Initially, I called the function ring_buffer_flush(), but once I noticed 
your patch I renamed it :)

>> - When you have a perf ring where wakeup_events is higher than 1,
>>   and you have remaining data in the rings you would like to pull
>>   out on exit (or maybe based on a timeout).
>> - For low latency cases where you burn a CPU that constantly polls
>>   the queues.
>>
>> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
>> ---
>>  tools/lib/bpf/libbpf.c   |   23 +++++++++++++++++++++++
>>  tools/lib/bpf/libbpf.h   |    1 +
>>  tools/lib/bpf/libbpf.map |    1 +
>>  3 files changed, 25 insertions(+)
>>
>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
>> index fa04cbe547ed..cbef3dac7507 100644
>> --- a/tools/lib/bpf/libbpf.c
>> +++ b/tools/lib/bpf/libbpf.c
>> @@ -8456,6 +8456,29 @@ int perf_buffer__poll(struct perf_buffer *pb, 
>> int timeout_ms)
>>         return cnt < 0 ? -errno : cnt;
>>  }
>>
>> +int perf_buffer__consume(struct perf_buffer *pb)
>> +{
>> +       int i;
>> +
>> +       if (!pb)
>> +               return -EINVAL;
>
> we don't check this in perf_buffer__poll, IMO, checking this in every
> "method" is an overkill.

Ack, will fix in v2

>> +
>> +       if (!pb->cpu_bufs)
>> +               return 0;
>
> no need to check. It's either non-NULL for valid perf_buffer, or
> calloc could return NULL if pb->cpu_cnt is zero (not sure it's
> possible, but still), but then loop below will never access
> pb->cpu_bufs[i].

Agreed, was just adding some safety checks, but in the constantly poll 
mode this is a lot of overhead. Will remover in v2.

>> +
>> +       for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
>
> I think pb->cpu_bufs[i] check is wrong, it will stop iteration
> prematurely if cpu_bufs are sparsely populated. So move check inside
> and continue loop if NULL.

Mimicked the behavior from other functions, however just to be safe I 
split it up.

>> +               int err;
>
> nit: declare it together with "i" above, similar to how
> perf_buffer__poll does it

Put it down here as it’s only used in the context of the for loop, but 
will move it up in the v2.

>> +               struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
>> +
>> +               err = perf_buffer__process_records(pb, cpu_buf);
>> +               if (err) {
>> +                       pr_warn("error while processing records: 
>> %d\n", err);
>> +                       return err;
>> +               }
>> +       }
>> +       return 0;
>> +}
>> +
>>  struct bpf_prog_info_array_desc {
>>         int     array_offset;   /* e.g. offset of jited_prog_insns */
>>         int     count_offset;   /* e.g. offset of jited_prog_len */
>> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
>> index 8ea69558f0a8..1e2e399a5f2c 100644
>> --- a/tools/lib/bpf/libbpf.h
>> +++ b/tools/lib/bpf/libbpf.h
>> @@ -533,6 +533,7 @@ perf_buffer__new_raw(int map_fd, size_t page_cnt,
>>
>>  LIBBPF_API void perf_buffer__free(struct perf_buffer *pb);
>>  LIBBPF_API int perf_buffer__poll(struct perf_buffer *pb, int 
>> timeout_ms);
>> +LIBBPF_API int perf_buffer__consume(struct perf_buffer *pb);
>>
>>  typedef enum bpf_perf_event_ret
>>         (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
>> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
>> index 0133d469d30b..381a7342ecfc 100644
>> --- a/tools/lib/bpf/libbpf.map
>> +++ b/tools/lib/bpf/libbpf.map
>> @@ -262,4 +262,5 @@ LIBBPF_0.0.9 {
>>                 bpf_link_get_fd_by_id;
>>                 bpf_link_get_next_id;
>>                 bpf_program__attach_iter;
>> +               perf_buffer__consume;
>>  } LIBBPF_0.0.8;
>>

Thanks for the review, will send out a v2 soon.

Re: [PATCH bpf-next] libbpf: add API to consume the perf ring buffer content

[Andrii Nakryiko]

On Tue, May 26, 2020 at 1:07 AM Eelco Chaudron <echaudro@redhat.com> wrote:
>
>
>
> On 26 May 2020, at 7:29, Andrii Nakryiko wrote:
>
> > On Mon, May 25, 2020 at 2:01 PM Eelco Chaudron <echaudro@redhat.com>
> > wrote:
> >>
> >> This new API, perf_buffer__consume, can be used as follows:
> >
> > I wonder, was it inspired by yet-to-be committed
> > ring_buffer__consume() or it's just a coincidence?
>
> Just coincidence, I was needing a function to flush the remaining ring
> entries, as I was using a larger wakeup_events value.
> Initially, I called the function ring_buffer_flush(), but once I noticed
> your patch I renamed it :)

Nice, thanks, I love consistent naming :)

>
> >> - When you have a perf ring where wakeup_events is higher than 1,
> >>   and you have remaining data in the rings you would like to pull
> >>   out on exit (or maybe based on a timeout).
> >> - For low latency cases where you burn a CPU that constantly polls
> >>   the queues.
> >>
> >> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
> >> ---
> >>  tools/lib/bpf/libbpf.c   |   23 +++++++++++++++++++++++
> >>  tools/lib/bpf/libbpf.h   |    1 +
> >>  tools/lib/bpf/libbpf.map |    1 +
> >>  3 files changed, 25 insertions(+)
> >>
> >> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> >> index fa04cbe547ed..cbef3dac7507 100644
> >> --- a/tools/lib/bpf/libbpf.c
> >> +++ b/tools/lib/bpf/libbpf.c
> >> @@ -8456,6 +8456,29 @@ int perf_buffer__poll(struct perf_buffer *pb,
> >> int timeout_ms)
> >>         return cnt < 0 ? -errno : cnt;
> >>  }
> >>
> >> +int perf_buffer__consume(struct perf_buffer *pb)
> >> +{
> >> +       int i;
> >> +
> >> +       if (!pb)
> >> +               return -EINVAL;
> >
> > we don't check this in perf_buffer__poll, IMO, checking this in every
> > "method" is an overkill.
>
> Ack, will fix in v2
>
> >> +
> >> +       if (!pb->cpu_bufs)
> >> +               return 0;
> >
> > no need to check. It's either non-NULL for valid perf_buffer, or
> > calloc could return NULL if pb->cpu_cnt is zero (not sure it's
> > possible, but still), but then loop below will never access
> > pb->cpu_bufs[i].
>
> Agreed, was just adding some safety checks, but in the constantly poll
> mode this is a lot of overhead. Will remover in v2.
>
> >> +
> >> +       for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
> >
> > I think pb->cpu_bufs[i] check is wrong, it will stop iteration
> > prematurely if cpu_bufs are sparsely populated. So move check inside
> > and continue loop if NULL.
>
> Mimicked the behavior from other functions, however just to be safe I
> split it up.

You mean perf_buffer__poll() or perf_buffer__free() loop? In the
perf_buffer__poll() case, first N events will always correspond to
non-NULL buffers. It's very different from what you are doing here.
But I think perf_buffer__free() actually is buggy similarly to how I
pointed out in this case. We need to fix that.

>
> >> +               int err;
> >
> > nit: declare it together with "i" above, similar to how
> > perf_buffer__poll does it
>
> Put it down here as it’s only used in the context of the for loop, but
> will move it up in the v2.
>
> >> +               struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
> >> +
> >> +               err = perf_buffer__process_records(pb, cpu_buf);
> >> +               if (err) {
> >> +                       pr_warn("error while processing records:
> >> %d\n", err);
> >> +                       return err;
> >> +               }
> >> +       }
> >> +       return 0;
> >> +}
> >> +
> >>  struct bpf_prog_info_array_desc {
> >>         int     array_offset;   /* e.g. offset of jited_prog_insns */
> >>         int     count_offset;   /* e.g. offset of jited_prog_len */
> >> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> >> index 8ea69558f0a8..1e2e399a5f2c 100644
> >> --- a/tools/lib/bpf/libbpf.h
> >> +++ b/tools/lib/bpf/libbpf.h
> >> @@ -533,6 +533,7 @@ perf_buffer__new_raw(int map_fd, size_t page_cnt,
> >>
> >>  LIBBPF_API void perf_buffer__free(struct perf_buffer *pb);
> >>  LIBBPF_API int perf_buffer__poll(struct perf_buffer *pb, int
> >> timeout_ms);
> >> +LIBBPF_API int perf_buffer__consume(struct perf_buffer *pb);
> >>
> >>  typedef enum bpf_perf_event_ret
> >>         (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
> >> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> >> index 0133d469d30b..381a7342ecfc 100644
> >> --- a/tools/lib/bpf/libbpf.map
> >> +++ b/tools/lib/bpf/libbpf.map
> >> @@ -262,4 +262,5 @@ LIBBPF_0.0.9 {
> >>                 bpf_link_get_fd_by_id;
> >>                 bpf_link_get_next_id;
> >>                 bpf_program__attach_iter;
> >> +               perf_buffer__consume;
> >>  } LIBBPF_0.0.8;
> >>
>
> Thanks for the review, will send out a v2 soon.
>