Linux-ARM-Kernel Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: James Clark <james.clark@linaro.org>
To: Leo Yan <leo.yan@arm.com>
Cc: Ingo Molnar <mingo@kernel.org>,
	linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
	coresight@lists.linaro.org, linux-arm-kernel@lists.infradead.org,
	linux-doc@vger.kernel.org, Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Namhyung Kim <namhyung@kernel.org>,
	Ian Rogers <irogers@google.com>,
	Adrian Hunter <adrian.hunter@intel.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Mike Leach <mike.leach@arm.com>,
	Anshuman Khandual <anshuman.khandual@arm.com>,
	Yeoreum Yun <yeoreum.yun@arm.com>,
	Jonathan Corbet <corbet@lwn.net>, Suyash Mahar <smahar@meta.com>,
	Amir Ayupov <aaupov@fb.com>
Subject: Re: [PATCH 16/25] coresight: perf: Support AUX sampling with per-CPU sinks
Date: Tue, 20 Oct 2026 14:21:58 +0100	[thread overview]
Message-ID: <89e50eff-5ebc-428a-978c-82ff1fbd3f8f@linaro.org> (raw)
In-Reply-To: <20260915-arm_cs_support_aux_sample-v1-16-35cf6f297d20@arm.com>



On 15/09/2026 16:49, Leo Yan wrote:
> Add snapshot_aux() for per-CPU sinks with overwrite buffers. Shared
> sinks can mix trace from multiple CPUs, making them unsuitable for
> context-based profiling.
> 
> Pad missing trace with leading zeros, or zero-fill the requested AUX
> payload when a snapshot is unavailable. This ensures that the decoder
> can safely discard the zero-filled trace data.
> 
> Reuse the AUX action guard to defer concurrent stops until the snapshot
> finishes. Skip snapshots during sink interrupt handling. On copy or
> restart failure, mark the AUX output truncated so perf core disables
> the event.
> 
> Assisted-by: Codex:gpt-6
> Signed-off-by: Leo Yan <leo.yan@arm.com>

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

> ---
>   drivers/hwtracing/coresight/coresight-etm-perf.c | 123 +++++++++++++++++++++++
>   drivers/hwtracing/coresight/coresight-etm-perf.h |   2 +
>   2 files changed, 125 insertions(+)
> 
> diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c
> index 8020a4d8af18bb669441824cfdb687f4fa299eb4..5e90ed49562c117879780f573072e6b522d9d048 100644
> --- a/drivers/hwtracing/coresight/coresight-etm-perf.c
> +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c
> @@ -456,6 +456,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages,
>   	if (!event_data)
>   		return NULL;
>   	INIT_WORK(&event_data->work, free_event_data);
> +	event_data->overwrite = overwrite;
>   
>   	/* First get the selected sink from user space. */
>   	sink_hash = ATTR_CFG_GET_FLD(&event->attr, sinkid);
> @@ -863,6 +864,127 @@ static void etm_event_del(struct perf_event *event, int mode)
>   	etm_event_stop(event, PERF_EF_UPDATE);
>   }
>   
> +/* A zero trace size pads the sample without accessing the AUX handle. */
> +static long etm_event_snapshot_aux_copy_data(struct perf_output_handle *aux_handle,
> +					     struct perf_output_handle *handle,
> +					     unsigned long trace_size,
> +					     unsigned long size)
> +{
> +	unsigned long pad;
> +	long ret;
> +
> +	trace_size = min(size, trace_size);
> +
> +	/* Pad the front with zeros to keep the latest trace data at the end */
> +	pad = size - trace_size;
> +	while (pad) {
> +		unsigned long chunk = min(pad, PAGE_SIZE);
> +
> +		if (perf_output_copy(handle, page_address(ZERO_PAGE(0)), chunk))
> +			return -EFAULT;
> +		pad -= chunk;
> +	}
> +
> +	/*
> +	 * Skip an empty snapshot, as equal offsets would otherwise copy the
> +	 * entire AUX ring.
> +	 */
> +	if (trace_size) {
> +		unsigned long to = aux_handle->head;
> +
> +		ret = perf_output_copy_aux(aux_handle, handle, to - trace_size, to);
> +		if (ret < 0)
> +			return ret;
> +	}
> +
> +	return size;
> +}
> +
> +static long etm_event_snapshot_aux(struct perf_event *event,
> +				   struct perf_output_handle *handle,
> +				   unsigned long size)
> +{
> +	struct etm_ctxt *ctxt = this_cpu_ptr(&etm_ctxt);
> +	struct perf_output_handle *aux_handle = &ctxt->handle;
> +	struct etm_event_data *event_data;
> +	struct coresight_device *source, *sink;
> +	struct coresight_path *path;
> +	unsigned long trace_size;
> +	long ret = 0;
> +
> +	if (!etm_event_aux_begin(event, ctxt))
> +		goto padding;
> +
> +	if (READ_ONCE(event->hw.aux_paused))
> +		goto out;
> +
> +	event_data = perf_get_aux(aux_handle);
> +	if (!event_data || !event_data->overwrite)
> +		goto out;
> +
> +	path = etm_event_get_ctxt_path(ctxt);
> +	if (!path)
> +		goto out;
> +
> +	source = coresight_get_source(path);
> +	sink = coresight_get_sink(path);
> +	if (WARN_ON_ONCE(!source || !sink))
> +		goto out;
> +
> +	if (!coresight_is_percpu_sink(sink)) {
> +		dev_err_once(&sink->dev, "AUX sampling requires a per-CPU sink\n");
> +		goto out;
> +	}
> +
> +	/* Skip snapshots that preempt the sink's interrupt handler */
> +	if (READ_ONCE(sink->in_interrupt))
> +		goto out;
> +
> +	coresight_pause_source(source);
> +	trace_size = sink_ops(sink)->update_buffer(sink, aux_handle,
> +						   event_data->snk_config);
> +	coresight_disable_path(path);
> +
> +	/*
> +	 * If the AUX handle was closed or marked truncated, the sink is
> +	 * expected to have notified perf core to disable the AUX event.
> +	 */
> +	if (!perf_get_aux(aux_handle) ||
> +	    (aux_handle->aux_flags & PERF_AUX_FLAG_TRUNCATED))
> +		goto fail;
> +
> +	ret = etm_event_snapshot_aux_copy_data(aux_handle, handle, trace_size, size);
> +	if (ret < 0)
> +		goto fail;
> +
> +	if (coresight_enable_path(path, CS_MODE_PERF))
> +		goto fail;
> +
> +	if (coresight_resume_source(source)) {
> +		coresight_disable_path(path);
> +		goto fail;
> +	}
> +
> +	goto out;
> +
> +fail:
> +	/* Set the truncated AUX flag so perf core can disable the event */
> +	if (perf_get_aux(aux_handle)) {
> +		perf_aux_output_flag(aux_handle, PERF_AUX_FLAG_TRUNCATED);
> +		perf_aux_output_end(aux_handle, 0);
> +	}
> +
> +out:
> +	etm_event_aux_end(event, ctxt);
> +
> +padding:
> +	/* Pad the buffer with zeros on an early exit */
> +	if (!ret)
> +		ret = etm_event_snapshot_aux_copy_data(NULL, handle, 0, size);
> +
> +	return ret;
> +}
> +
>   static int etm_addr_filters_validate(struct list_head *filters)
>   {
>   	bool range = false, address = false;
> @@ -1109,6 +1231,7 @@ int __init etm_perf_init(void)
>   	etm_pmu.free_aux		= etm_free_aux;
>   	etm_pmu.start			= etm_event_start;
>   	etm_pmu.stop			= etm_event_stop;
> +	etm_pmu.snapshot_aux		= etm_event_snapshot_aux;
>   	etm_pmu.add			= etm_event_add;
>   	etm_pmu.del			= etm_event_del;
>   	etm_pmu.addr_filters_sync	= etm_addr_filters_sync;
> diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.h b/drivers/hwtracing/coresight/coresight-etm-perf.h
> index 24d929428633f56f24a114d5bb93b8ac4799ca11..a69eb04a597efcfe630342a410cbf8fea3c4501e 100644
> --- a/drivers/hwtracing/coresight/coresight-etm-perf.h
> +++ b/drivers/hwtracing/coresight/coresight-etm-perf.h
> @@ -89,6 +89,7 @@ struct etm_filters {
>    * @aux_hwid_done:	Whether a CPU has emitted the TraceID packet or not.
>    * @snk_config:		The sink configuration.
>    * @cfg_hash:		The hash id of any coresight config selected.
> + * @overwrite:		Whether the AUX buffer uses overwrite mode.
>    * @path:		An array of path, each slot for one CPU.
>    */
>   struct etm_event_data {
> @@ -97,6 +98,7 @@ struct etm_event_data {
>   	cpumask_t aux_hwid_done;
>   	void *snk_config;
>   	u32 cfg_hash;
> +	bool overwrite;
>   	struct coresight_path * __percpu *path;
>   };
>   
> 



  reply	other threads:[~2026-10-20 13:22 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-15 15:48 [PATCH 00/25] CoreSight / perf: Support AUX sampling Leo Yan
2026-09-15 15:48 ` [PATCH 01/25] perf/core: Reject non-positive AUX snapshot sizes Leo Yan
2026-10-16 13:14   ` James Clark
2026-09-15 15:48 ` [PATCH 02/25] perf/core: Prevent AUX sampling from racing with pause/resume Leo Yan
2026-10-16 13:29   ` James Clark
2026-09-15 15:48 ` [PATCH 03/25] perf/core: Export output copy helpers for CoreSight Leo Yan
2026-10-16 13:30   ` James Clark
2026-09-15 15:48 ` [PATCH 04/25] coresight: perf: End AUX output when an event is throttled Leo Yan
2026-10-16 13:46   ` James Clark
2026-09-15 15:48 ` [PATCH 05/25] coresight: perf: Extract AUX buffer update helper Leo Yan
2026-10-16 13:53   ` James Clark
2026-09-15 15:48 ` [PATCH 06/25] coresight: perf: Simplify flow for CPUs without a path Leo Yan
2026-10-16 13:59   ` James Clark
2026-09-15 15:48 ` [PATCH 07/25] coresight: perf: Validate the live context through its path Leo Yan
2026-10-19  9:25   ` James Clark
2026-09-15 15:48 ` [PATCH 08/25] coresight: perf: Validate context before resuming trace Leo Yan
2026-10-19  9:27   ` James Clark
2026-09-15 15:48 ` [PATCH 09/25] coresight: perf: Improve hardware state transitions Leo Yan
2026-10-19  9:33   ` James Clark
2026-10-19  9:35   ` James Clark
2026-09-15 15:48 ` [PATCH 10/25] coresight: trbe: Pad snapshot buffers Leo Yan
2026-10-19  9:45   ` James Clark
2026-09-15 15:48 ` [PATCH 11/25] coresight: trbe: Use circular buffer mode for snapshots Leo Yan
2026-10-19 10:13   ` James Clark
2026-09-15 15:48 ` [PATCH 12/25] coresight: trbe: Track per-CPU sink interrupt handling Leo Yan
2026-10-19 11:54   ` James Clark
2026-09-15 15:48 ` [PATCH 13/25] coresight: trbe: Handle an already disabled sink Leo Yan
2026-10-19 12:44   ` James Clark
2026-09-15 15:49 ` [PATCH 14/25] coresight: perf: Look up the trace path in etm_event_pause() Leo Yan
2026-10-19 12:46   ` James Clark
2026-09-15 15:49 ` [PATCH 15/25] coresight: perf: Serialize AUX pause and resume with event stops Leo Yan
2026-10-19 13:02   ` James Clark
2026-09-15 15:49 ` [PATCH 16/25] coresight: perf: Support AUX sampling with per-CPU sinks Leo Yan
2026-10-20 13:21   ` James Clark [this message]
2026-09-15 15:49 ` [PATCH 17/25] perf cs-etm: Allow history collection for samples with zero IP Leo Yan
2026-10-20 13:23   ` James Clark
2026-09-15 15:49 ` [PATCH 18/25] perf cs-etm: Unify timeless buffer decoding Leo Yan
2026-10-20 14:15   ` James Clark
2026-09-15 15:49 ` [PATCH 19/25] perf cs-etm: Propagate errors from trace queue flushing Leo Yan
2026-09-15 15:49 ` [PATCH 20/25] perf cs-etm: Drain pending packets before finishing trace blocks Leo Yan
2026-09-15 15:49 ` [PATCH 21/25] perf cs-etm: Complete packet draining with end of trace Leo Yan
2026-09-15 15:49 ` [PATCH 22/25] perf cs-etm: Centralize sample history collection Leo Yan
2026-09-15 15:49 ` [PATCH 23/25] perf cs-etm: Decode AUX samples into callchains and branch stacks Leo Yan
2026-10-20 14:13   ` James Clark
2026-09-15 15:49 ` [PATCH 24/25] perf test: Add CoreSight AUX sample decoding test Leo Yan
2026-10-20 14:35   ` James Clark
2026-09-15 15:49 ` [PATCH 25/25] Documentation: coresight: Document AUX sample decoding Leo Yan
2026-10-20 14:26   ` James Clark

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=89e50eff-5ebc-428a-978c-82ff1fbd3f8f@linaro.org \
    --to=james.clark@linaro.org \
    --cc=aaupov@fb.com \
    --cc=acme@kernel.org \
    --cc=adrian.hunter@intel.com \
    --cc=anshuman.khandual@arm.com \
    --cc=corbet@lwn.net \
    --cc=coresight@lists.linaro.org \
    --cc=irogers@google.com \
    --cc=leo.yan@arm.com \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-perf-users@vger.kernel.org \
    --cc=mike.leach@arm.com \
    --cc=mingo@kernel.org \
    --cc=mingo@redhat.com \
    --cc=namhyung@kernel.org \
    --cc=peterz@infradead.org \
    --cc=smahar@meta.com \
    --cc=suzuki.poulose@arm.com \
    --cc=yeoreum.yun@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox