Re: [PATCH 6/6] coresight: perf: Retry failed HW_ID writes
James Clark <[email protected]>
| Newsgroups | gmane.linux.kernel,gmane.linux.ports.arm.kernel,gmane.linux.kernel.perf.user |
|---|---|
| Message-ID | <[email protected]> |
On 21/08/2026 10:50, James Clark wrote: > Don't set the aux_hwid_done bit when writes fail so that it can be > retried later if the ring buffer was full. > > Add a second attempt on stop in case this session only has one call > to start. Otherwise there is no chance to retry and it isn't fixed by > not setting aux_hwid_done. > > Signed-off-by: James Clark <[email protected]> > --- > drivers/hwtracing/coresight/coresight-etm-perf.c | 52 +++++++++++++++--------- > 1 file changed, 32 insertions(+), 20 deletions(-) > > diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c > index 09b21a711a87..0ffc5cd9c88f 100644 > --- a/drivers/hwtracing/coresight/coresight-etm-perf.c > +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c > @@ -536,6 +536,31 @@ static int etm_event_resume(struct coresight_path *path) > return ret; > } > > +static void etm_output_hw_id(struct perf_event *event, > + struct etm_event_data *event_data, > + struct coresight_path *path, > + struct coresight_device *sink) > +{ > + int cpu = smp_processor_id(); > + u64 hw_id; > + > + /* > + * output cpu / trace ID in perf record, once for the lifetime > + * of the event. > + */ > + if (!cpumask_test_cpu(cpu, &event_data->aux_hwid_done)) { > + hw_id = FIELD_PREP(CS_AUX_HW_ID_MAJOR_VERSION_MASK, > + CS_AUX_HW_ID_MAJOR_VERSION); > + hw_id |= FIELD_PREP(CS_AUX_HW_ID_MINOR_VERSION_MASK, > + CS_AUX_HW_ID_MINOR_VERSION); > + hw_id |= FIELD_PREP(CS_AUX_HW_ID_TRACE_ID_MASK, path->trace_id); > + hw_id |= FIELD_PREP(CS_AUX_HW_ID_SINK_ID_MASK, coresight_get_sink_id(sink)); > + > + if (!perf_report_aux_output_id(event, hw_id)) > + cpumask_set_cpu(cpu, &event_data->aux_hwid_done); > + } > +} > + > static void etm_event_start(struct perf_event *event, int flags) > { > int cpu = smp_processor_id(); > @@ -544,7 +569,6 @@ static void etm_event_start(struct perf_event *event, int flags) > struct perf_output_handle *handle = &ctxt->handle; > struct coresight_device *source, *sink; > struct coresight_path *path; > - u64 hw_id; > > if (flags & PERF_EF_RESUME) { > path = etm_event_get_ctxt_path(ctxt); > @@ -596,22 +620,7 @@ static void etm_event_start(struct perf_event *event, int flags) > if (coresight_enable_source(source, event, CS_MODE_PERF, path)) > goto fail_disable_path; > > - /* > - * output cpu / trace ID in perf record, once for the lifetime > - * of the event. > - */ > - if (!cpumask_test_cpu(cpu, &event_data->aux_hwid_done)) { > - cpumask_set_cpu(cpu, &event_data->aux_hwid_done); > - > - hw_id = FIELD_PREP(CS_AUX_HW_ID_MAJOR_VERSION_MASK, > - CS_AUX_HW_ID_MAJOR_VERSION); > - hw_id |= FIELD_PREP(CS_AUX_HW_ID_MINOR_VERSION_MASK, > - CS_AUX_HW_ID_MINOR_VERSION); > - hw_id |= FIELD_PREP(CS_AUX_HW_ID_TRACE_ID_MASK, path->trace_id); > - hw_id |= FIELD_PREP(CS_AUX_HW_ID_SINK_ID_MASK, coresight_get_sink_id(sink)); > - > - perf_report_aux_output_id(event, hw_id); > - } > + etm_output_hw_id(event, event_data, path, sink); > > out: > /* Tell the perf core the event is alive */ > @@ -630,7 +639,7 @@ static void etm_event_start(struct perf_event *event, int flags) > */ > if (READ_ONCE(handle->event)) { > perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED); > - perf_aux_output_end(handle, 0); > + perf_aux_output_end(path->handle, 0); Accidentally left this in from hacking. It doesn't do anything so I'll remove it.