[PATCH] perf/core: Skip empty AUX records with only format flags

Leo Yan posted 1 patch 1 month ago
There is a newer version of this series
kernel/events/ring_buffer.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
[PATCH] perf/core: Skip empty AUX records with only format flags
Posted by Leo Yan 1 month ago
perf_aux_output_end() emits a PERF_RECORD_AUX when the recorded size is
nonzero or when any flag other than PERF_AUX_FLAG_OVERWRITE is set.

PMU format flags describe how an AUX payload is encoded. TRBE driver
sets PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW for raw trace buffers, causing
an AUX record to be emitted even when no trace data.

This is noticeable when tracing a task with strace. Ptrace stops
repeatedly end empty AUX transactions, producing many zero-sized
PERF_RECORD_AUX records. For example:

  perf record -e cs_etm//u -m,128M -- strace ls

  perf script -D 2>&1 |
     awk '/PERF_RECORD_AUX offset/ {
        for (i = 1; i <= NF; i++)
                if ($i == "size:" && $(i + 1) == "0")
                        count++
     }
     END { print count }'
  165

This recording contains 165 zero-sized AUX records which provide no
useful information to userspace.

Ignore PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK, together with
PERF_AUX_FLAG_OVERWRITE, when deciding whether an empty AUX record is
useful. Zero-sized records carrying TRUNCATED, PARTIAL or COLLISION
are still emitted.

Fixes: 547b60988e63 ("perf: aux: Add flags for the buffer format")
Reported-by: Tamas Petz <tamas.petz@arm.com>
Signed-off-by: Leo Yan <leo.yan@arm.com>
---
 kernel/events/ring_buffer.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 9fe92161715e0987520cddb873dd162fd2a77d5c..1b1ffe0533e58a9d976d0eb0fc23e09d5936da2b 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -509,7 +509,10 @@ void perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
 	/*
 	 * Only send RECORD_AUX if we have something useful to communicate
 	 *
-	 * Note: the OVERWRITE records by themselves are not considered
+	 * PMU_FORMAT bits identify the PMU type rather than an AUX event
+	 * has occurred, so ignore them for zero-sized records.
+	 *
+	 * The OVERWRITE records by themselves are not considered
 	 * useful, as they don't communicate any *new* information,
 	 * aside from the short-lived offset, that becomes history at
 	 * the next event sched-in and therefore isn't useful.
@@ -518,7 +521,9 @@ void perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
 	 * offset. So, from now on we don't output AUX records that
 	 * have *only* OVERWRITE flag set.
 	 */
-	if (size || (handle->aux_flags & ~(u64)PERF_AUX_FLAG_OVERWRITE))
+	if (size ||
+	    (handle->aux_flags & ~(u64)(PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK |
+					PERF_AUX_FLAG_OVERWRITE)))
 		perf_event_aux_event(handle->event, aux_head, size,
 				     handle->aux_flags);
 

---
base-commit: 66498c75b4f8017f62d720d9b59675bdf3abce91
change-id: 20260825-perf_core_fix_zero_aux_records-8873172d2596

Best regards,
-- 
Leo Yan <leo.yan@arm.com>
Re: [PATCH] perf/core: Skip empty AUX records with only format flags
Posted by James Clark 1 month ago

On 25/08/2026 11:37, Leo Yan wrote:
> perf_aux_output_end() emits a PERF_RECORD_AUX when the recorded size is
> nonzero or when any flag other than PERF_AUX_FLAG_OVERWRITE is set.
> 
> PMU format flags describe how an AUX payload is encoded. TRBE driver
> sets PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW for raw trace buffers, causing
> an AUX record to be emitted even when no trace data.
> 
> This is noticeable when tracing a task with strace. Ptrace stops
> repeatedly end empty AUX transactions, producing many zero-sized
> PERF_RECORD_AUX records. For example:
> 
>    perf record -e cs_etm//u -m,128M -- strace ls
> 
>    perf script -D 2>&1 |
>       awk '/PERF_RECORD_AUX offset/ {
>          for (i = 1; i <= NF; i++)
>                  if ($i == "size:" && $(i + 1) == "0")
>                          count++
>       }
>       END { print count }'
>    165
> 
> This recording contains 165 zero-sized AUX records which provide no
> useful information to userspace.
> 
> Ignore PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK, together with
> PERF_AUX_FLAG_OVERWRITE, when deciding whether an empty AUX record is
> useful. Zero-sized records carrying TRUNCATED, PARTIAL or COLLISION
> are still emitted.
> 
> Fixes: 547b60988e63 ("perf: aux: Add flags for the buffer format")
> Reported-by: Tamas Petz <tamas.petz@arm.com>
> Signed-off-by: Leo Yan <leo.yan@arm.com>
> ---
>   kernel/events/ring_buffer.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
> index 9fe92161715e0987520cddb873dd162fd2a77d5c..1b1ffe0533e58a9d976d0eb0fc23e09d5936da2b 100644
> --- a/kernel/events/ring_buffer.c
> +++ b/kernel/events/ring_buffer.c
> @@ -509,7 +509,10 @@ void perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
>   	/*
>   	 * Only send RECORD_AUX if we have something useful to communicate
>   	 *
> -	 * Note: the OVERWRITE records by themselves are not considered
> +	 * PMU_FORMAT bits identify the PMU type rather than an AUX event
> +	 * has occurred, so ignore them for zero-sized records.
> +	 *
> +	 * The OVERWRITE records by themselves are not considered
>   	 * useful, as they don't communicate any *new* information,
>   	 * aside from the short-lived offset, that becomes history at
>   	 * the next event sched-in and therefore isn't useful.
> @@ -518,7 +521,9 @@ void perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
>   	 * offset. So, from now on we don't output AUX records that
>   	 * have *only* OVERWRITE flag set.
>   	 */
> -	if (size || (handle->aux_flags & ~(u64)PERF_AUX_FLAG_OVERWRITE))
> +	if (size ||
> +	    (handle->aux_flags & ~(u64)(PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK |
> +					PERF_AUX_FLAG_OVERWRITE)))

Almost seems like it would be better to define the union of what's 
considered useful where they are defined in perf_event.h. Or invert it 
here and do (TRUNCATED | PARTIAL etc).  Otherwise the same thing will 
happen when the next thing is added.

Other than that:

Reviewed-by: James Clark <james.clark@linaro.org>

>   		perf_event_aux_event(handle->event, aux_head, size,
>   				     handle->aux_flags);
>   
> 
> ---
> base-commit: 66498c75b4f8017f62d720d9b59675bdf3abce91
> change-id: 20260825-perf_core_fix_zero_aux_records-8873172d2596
> 
> Best regards,