Re: [PATCH v2 13/16] coresight: Use per-sink trace ID maps for Perf sessions

Hi James

On 04/06/2024 15:30, James Clark wrote:

This will allow sessions with more than CORESIGHT_TRACE_IDS_MAX ETMs as long as there are fewer than that many ETMs connected to each sink.

Each sink owns its own trace ID map, and any Perf session connecting to that sink will allocate from it, even if the sink is currently in use by other users. This is similar to the existing behavior where the dynamic trace IDs are constant as long as there is any concurrent Perf session active. It's not completely optimal because slightly more IDs will be used than necessary, but the optimal solution involves tracking the PIDs of each session and allocating ID maps based on the session owner. This is difficult to do with the combination of per-thread and per-cpu modes and some scheduling issues. The complexity of this isn't likely to worth it because even with multiple users they'd just see a difference in the ordering of ID allocations rather than hitting any limits (unless the hardware does have too many ETMs connected to one sink).

Please find my comments below.

Signed-off-by: James Clark james.clark@arm.com

drivers/hwtracing/coresight/coresight-core.c | 10 ++++++++++ drivers/hwtracing/coresight/coresight-dummy.c | 3 ++- drivers/hwtracing/coresight/coresight-etm-perf.c | 15 ++++++++++----- .../hwtracing/coresight/coresight-etm3x-core.c | 9 +++++---- .../hwtracing/coresight/coresight-etm4x-core.c | 9 +++++---- drivers/hwtracing/coresight/coresight-stm.c | 3 ++- drivers/hwtracing/coresight/coresight-sysfs.c | 3 ++- drivers/hwtracing/coresight/coresight-tpdm.c | 3 ++- include/linux/coresight.h | 3 ++- 9 files changed, 40 insertions(+), 18 deletions(-)

diff --git a/drivers/hwtracing/coresight/coresight-core.c b/drivers/hwtracing/coresight/coresight-core.c index 9fc6f6b863e0..d5aaeafe5c7d 100644 --- a/drivers/hwtracing/coresight/coresight-core.c +++ b/drivers/hwtracing/coresight/coresight-core.c @@ -902,6 +902,7 @@ static void coresight_device_release(struct device *dev) struct coresight_device *csdev = to_coresight_device(dev); fwnode_handle_put(csdev->dev.fwnode);

• free_percpu(csdev->perf_sink_id_map.cpu_map); kfree(csdev); }

@@ -1159,6 +1160,14 @@ struct coresight_device *coresight_register(struct coresight_desc *desc) csdev->dev.fwnode = fwnode_handle_get(dev_fwnode(desc->dev)); dev_set_name(&csdev->dev, "%s", desc->name);

• if (csdev->type == CORESIGHT_DEV_TYPE_SINK ||

•    csdev->type == CORESIGHT_DEV_TYPE_LINKSINK) {
  

• csdev->perf_sink_id_map.cpu_map = alloc_percpu(atomic_t);
  

• if (!csdev->perf_sink_id_map.cpu_map) {
  

• 	ret = -ENOMEM;
  

• 	goto err_out;
  

• }
  

• } /*
  • Make sure the device registration and the connection fixup
  • are synchronised, so that we don't see uninitialised devices

@@ -1216,6 +1225,7 @@ struct coresight_device *coresight_register(struct coresight_desc *desc) err_out: /* Cleanup the connection information */ coresight_release_platform_data(NULL, desc->dev, desc->pdata);

• kfree(csdev);

This hunk looked suspicious to me and is problematic. If we fail to register the device with "device_register()", we already free the csdev, via coresight_device_release() triggered via the device_put(), (See comments in that case) and we could trigger a double free of the csdev with this change.

I would recommend, you free the "csdev" in the above case, if we fail to allocate the percpu storage and fall through.

return ERR_PTR(ret); } EXPORT_SYMBOL_GPL(coresight_register); diff --git a/drivers/hwtracing/coresight/coresight-dummy.c b/drivers/hwtracing/coresight/coresight-dummy.c index ac70c0b491be..1f1b9ad160f6 100644 --- a/drivers/hwtracing/coresight/coresight-dummy.c +++ b/drivers/hwtracing/coresight/coresight-dummy.c @@ -21,7 +21,8 @@ DEFINE_CORESIGHT_DEVLIST(source_devs, "dummy_source"); DEFINE_CORESIGHT_DEVLIST(sink_devs, "dummy_sink"); static int dummy_source_enable(struct coresight_device *csdev,

• 	       struct perf_event *event, enum cs_mode mode)
  

• 	       struct perf_event *event, enum cs_mode mode,
  

• 	       __maybe_unused struct coresight_trace_id_map *id_map)
  

  { dev_dbg(csdev->dev.parent, "Dummy source enabled\n");

diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c index c0c60e6a1703..7fb55dafb639 100644 --- a/drivers/hwtracing/coresight/coresight-etm-perf.c +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c @@ -229,10 +229,13 @@ static void free_event_data(struct work_struct *work) struct list_head **ppath; ppath = etm_event_cpu_path_ptr(event_data, cpu);

• if (!(IS_ERR_OR_NULL(*ppath)))
  

• if (!(IS_ERR_OR_NULL(*ppath))) {
  

• 	struct coresight_device *sink = coresight_get_sink(*ppath);
  

• 	coresight_trace_id_put_cpu_id_map(cpu, &sink->perf_sink_id_map);
  coresight_release_path(*ppath);
  

• }
  

  *ppath = NULL;

• coresight_trace_id_put_cpu_id(cpu);
  

  }

/* mark perf event as done for trace id allocator */ @@ -401,7 +404,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages, } /* ensure we can allocate a trace ID for this CPU */

• trace_id = coresight_trace_id_get_cpu_id(cpu);
  

• trace_id = coresight_trace_id_get_cpu_id_map(cpu, &sink->perf_sink_id_map);
  

  if (!IS_VALID_CS_TRACE_ID(trace_id)) { cpumask_clear_cpu(cpu, mask); coresight_release_path(path);

@@ -495,7 +498,8 @@ static void etm_event_start(struct perf_event *event, int flags) goto fail_end_stop; /* Finally enable the tracer */

• if (source_ops(csdev)->enable(csdev, event, CS_MODE_PERF))

• if (source_ops(csdev)->enable(csdev, event, CS_MODE_PERF,

• 		      &sink->perf_sink_id_map))
  

  goto fail_disable_path;

/* @@ -507,7 +511,8 @@ static void etm_event_start(struct perf_event *event, int flags) hw_id = FIELD_PREP(CS_AUX_HW_ID_VERSION_MASK, CS_AUX_HW_ID_CURR_VERSION); hw_id |= FIELD_PREP(CS_AUX_HW_ID_TRACE_ID_MASK,

• 		    coresight_trace_id_read_cpu_id(cpu));
  

• 		    coresight_trace_id_read_cpu_id_map(cpu,
  

• 						       &sink->perf_sink_id_map));
  

  perf_report_aux_output_id(event, hw_id); }

diff --git a/drivers/hwtracing/coresight/coresight-etm3x-core.c b/drivers/hwtracing/coresight/coresight-etm3x-core.c index 9d5c1391ffb1..65cf7456426a 100644 --- a/drivers/hwtracing/coresight/coresight-etm3x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm3x-core.c @@ -481,7 +481,8 @@ void etm_release_trace_id(struct etm_drvdata *drvdata) } static int etm_enable_perf(struct coresight_device *csdev,

• 	   struct perf_event *event)
  

• 	   struct perf_event *event,
  

• 	   struct coresight_trace_id_map *id_map)
  

  { struct etm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); int trace_id;

@@ -500,7 +501,7 @@ static int etm_enable_perf(struct coresight_device *csdev, * with perf locks - we know the ID cannot change until perf shuts down * the session */

• trace_id = coresight_trace_id_read_cpu_id(drvdata->cpu);

• trace_id = coresight_trace_id_read_cpu_id_map(drvdata->cpu, id_map); if (!IS_VALID_CS_TRACE_ID(trace_id)) { dev_err(&drvdata->csdev->dev, "Failed to set trace ID for %s on CPU%d\n", dev_name(&drvdata->csdev->dev), drvdata->cpu);

@@ -553,7 +554,7 @@ static int etm_enable_sysfs(struct coresight_device *csdev) } static int etm_enable(struct coresight_device *csdev, struct perf_event *event,

•       enum cs_mode mode)
  

•       enum cs_mode mode, struct coresight_trace_id_map *id_map)
  

  { int ret; struct etm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);

@@ -568,7 +569,7 @@ static int etm_enable(struct coresight_device *csdev, struct perf_event *event, ret = etm_enable_sysfs(csdev); break; case CS_MODE_PERF:

• ret = etm_enable_perf(csdev, event);
  

• ret = etm_enable_perf(csdev, event, id_map);
  

  break; default: ret = -EINVAL;

diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index a0bdfabddbc6..fcd0f9ba562d 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -752,7 +752,8 @@ static int etm4_parse_event_config(struct coresight_device *csdev, } static int etm4_enable_perf(struct coresight_device *csdev,

• 	    struct perf_event *event)
  

• 	    struct perf_event *event,
  

• 	    struct coresight_trace_id_map *id_map)
  

  { int ret = 0, trace_id; struct etmv4_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);

@@ -775,7 +776,7 @@ static int etm4_enable_perf(struct coresight_device *csdev, * with perf locks - we know the ID cannot change until perf shuts down * the session */

• trace_id = coresight_trace_id_read_cpu_id(drvdata->cpu);

• trace_id = coresight_trace_id_read_cpu_id_map(drvdata->cpu, id_map); if (!IS_VALID_CS_TRACE_ID(trace_id)) { dev_err(&drvdata->csdev->dev, "Failed to set trace ID for %s on CPU%d\n", dev_name(&drvdata->csdev->dev), drvdata->cpu);

@@ -837,7 +838,7 @@ static int etm4_enable_sysfs(struct coresight_device *csdev) } static int etm4_enable(struct coresight_device *csdev, struct perf_event *event,

•        enum cs_mode mode)
  

•        enum cs_mode mode, struct coresight_trace_id_map *id_map)
  

  { int ret;

@@ -851,7 +852,7 @@ static int etm4_enable(struct coresight_device *csdev, struct perf_event *event, ret = etm4_enable_sysfs(csdev); break; case CS_MODE_PERF:

• ret = etm4_enable_perf(csdev, event);
  

• ret = etm4_enable_perf(csdev, event, id_map);
  

  break; default: ret = -EINVAL;

diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index e1c62820dfda..a80ad1de4c23 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -194,7 +194,8 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) } static int stm_enable(struct coresight_device *csdev, struct perf_event *event,

•       enum cs_mode mode)
  

•       enum cs_mode mode,
  

•       __maybe_unused struct coresight_trace_id_map *trace_id)
  

  { struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);

diff --git a/drivers/hwtracing/coresight/coresight-sysfs.c b/drivers/hwtracing/coresight/coresight-sysfs.c index 1e67cc7758d7..a01c9e54e2ed 100644 --- a/drivers/hwtracing/coresight/coresight-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-sysfs.c @@ -9,6 +9,7 @@ #include <linux/kernel.h> #include "coresight-priv.h" +#include "coresight-trace-id.h" /*

• Use IDR to map the hash of the source's device name

@@ -63,7 +64,7 @@ static int coresight_enable_source_sysfs(struct coresight_device *csdev, */ lockdep_assert_held(&coresight_mutex); if (coresight_get_mode(csdev) != CS_MODE_SYSFS) {

• ret = source_ops(csdev)->enable(csdev, data, mode);
  

• ret = source_ops(csdev)->enable(csdev, data, mode, NULL);
  

  if (ret) return ret; }

diff --git a/drivers/hwtracing/coresight/coresight-tpdm.c b/drivers/hwtracing/coresight/coresight-tpdm.c index a9708ab0d488..0376ad326a2f 100644 --- a/drivers/hwtracing/coresight/coresight-tpdm.c +++ b/drivers/hwtracing/coresight/coresight-tpdm.c @@ -439,7 +439,8 @@ static void __tpdm_enable(struct tpdm_drvdata *drvdata) } static int tpdm_enable(struct coresight_device *csdev, struct perf_event *event,

•        enum cs_mode mode)
  

•        enum cs_mode mode,
  

•        __maybe_unused struct coresight_trace_id_map *id_map)
  

  { struct tpdm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);

diff --git a/include/linux/coresight.h b/include/linux/coresight.h index 7d62b88bfb5c..9c3067e2e38b 100644 --- a/include/linux/coresight.h +++ b/include/linux/coresight.h @@ -290,6 +290,7 @@ struct coresight_device { bool sysfs_sink_activated; struct dev_ext_attribute *ea; struct coresight_device *def_sink;

• struct coresight_trace_id_map perf_sink_id_map; /* sysfs links between components */ int nr_links; bool has_conns_grp;

@@ -384,7 +385,7 @@ struct coresight_ops_link { struct coresight_ops_source { int (*cpu_id)(struct coresight_device *csdev); int (*enable)(struct coresight_device *csdev, struct perf_event *event,

•       enum cs_mode mode);
  

•       enum cs_mode mode, struct coresight_trace_id_map *id_map);
  

I am slightly concerned by this change. We already pass down "event" for a generic "enable" call back which serves both SYSFS & PERF. Now we are adding one more, just for the PERF usage. Ideally, it is possible to pass a single argument, e.g. perf_aux_output_handle which can fetch you all the required infor for "enable".

i.e, hanlde->event, perf_get_aux(handle) -> event_data. We could even add a helper to find the trace_id map from the handle, which could hide the implementation details from the backend drivers.

e.g.,

etm_perf_get_trace_id_map(handle) -> { event_data = perf_get_aux() path = per_cpu(event_data->path) sink = sink_from_path(path) return trace_id of the sink; }

That said, we if we intend to extend the usage of trace_id_map for sysfs, this may be fine as is.

Suzuki

void (*disable)(struct coresight_device *csdev, struct perf_event *event); };