The returned pointer from .alloc_buffer() callback can be an error, if only checking NULL pointer the driver cannot capture errors. The driver will proceed even after failure and cause kernel panic.
Change to use IS_ERR_OR_NULL() check for capture error cases.
Fixes: 0bcbf2e30ff2 ("coresight: etm-perf: new PMU driver for ETM tracers") Reported-by: Tamas Zsoldos tamas.zsoldos@arm.com Signed-off-by: Leo Yan leo.yan@arm.com --- drivers/hwtracing/coresight/coresight-etm-perf.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c index f677c08233ba1a28b277674662c6e6db904873dd..440d967f5d0962df187a81b0dd69a7d82a8b62ba 100644 --- a/drivers/hwtracing/coresight/coresight-etm-perf.c +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c @@ -198,7 +198,7 @@ static void free_sink_buffer(struct etm_event_data *event_data) cpumask_t *mask = &event_data->mask; struct coresight_device *sink;
- if (!event_data->snk_config) + if (IS_ERR_OR_NULL(event_data->snk_config)) return;
if (WARN_ON(cpumask_empty(mask))) @@ -450,7 +450,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages, event_data->snk_config = sink_ops(sink)->alloc_buffer(sink, event, pages, nr_pages, overwrite); - if (!event_data->snk_config) + if (IS_ERR_OR_NULL(event_data->snk_config)) goto err;
out:
--- base-commit: fa71e9cb4cfa59abb196229667ec84929bdc18fe change-id: 20250904-cs_etm_auxsetup_fix_error_handling-cb7e07ed9adf
Best regards,
On 04/09/2025 11:53 am, Leo Yan wrote:
The returned pointer from .alloc_buffer() callback can be an error, if only checking NULL pointer the driver cannot capture errors. The driver will proceed even after failure and cause kernel panic.
Change to use IS_ERR_OR_NULL() check for capture error cases.
Fixes: 0bcbf2e30ff2 ("coresight: etm-perf: new PMU driver for ETM tracers") Reported-by: Tamas Zsoldos tamas.zsoldos@arm.com Signed-off-by: Leo Yan leo.yan@arm.com
drivers/hwtracing/coresight/coresight-etm-perf.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c index f677c08233ba1a28b277674662c6e6db904873dd..440d967f5d0962df187a81b0dd69a7d82a8b62ba 100644 --- a/drivers/hwtracing/coresight/coresight-etm-perf.c +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c @@ -198,7 +198,7 @@ static void free_sink_buffer(struct etm_event_data *event_data) cpumask_t *mask = &event_data->mask; struct coresight_device *sink;
- if (!event_data->snk_config)
- if (IS_ERR_OR_NULL(event_data->snk_config)) return;
if (WARN_ON(cpumask_empty(mask))) @@ -450,7 +450,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages, event_data->snk_config = sink_ops(sink)->alloc_buffer(sink, event, pages, nr_pages, overwrite);
- if (!event_data->snk_config)
- if (IS_ERR_OR_NULL(event_data->snk_config)) goto err;
I think the bug is in TRBE. It's the only one that returns an error pointer, but only for -ENOMEM which would normally be NULL for alloc type functions anyway.
Also it's assigned to event_data->snk_config which is later NULL checked rather than IS_ERR_OR_NULL in free_sink_buffer(). Maybe that path doesn't happen, but all instances should be updated anyway.
It's much easier to keep it as NULL and make the fix in TRBE. It also wouldn't need to be backported as far.
out:
base-commit: fa71e9cb4cfa59abb196229667ec84929bdc18fe change-id: 20250904-cs_etm_auxsetup_fix_error_handling-cb7e07ed9adf
Best regards,
On 04/09/2025 1:00 pm, James Clark wrote:
On 04/09/2025 11:53 am, Leo Yan wrote:
The returned pointer from .alloc_buffer() callback can be an error, if only checking NULL pointer the driver cannot capture errors. The driver will proceed even after failure and cause kernel panic.
Change to use IS_ERR_OR_NULL() check for capture error cases.
Fixes: 0bcbf2e30ff2 ("coresight: etm-perf: new PMU driver for ETM tracers") Reported-by: Tamas Zsoldos tamas.zsoldos@arm.com Signed-off-by: Leo Yan leo.yan@arm.com
drivers/hwtracing/coresight/coresight-etm-perf.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/ drivers/hwtracing/coresight/coresight-etm-perf.c index f677c08233ba1a28b277674662c6e6db904873dd..440d967f5d0962df187a81b0dd69a7d82a8b62ba 100644 --- a/drivers/hwtracing/coresight/coresight-etm-perf.c +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c @@ -198,7 +198,7 @@ static void free_sink_buffer(struct etm_event_data *event_data) cpumask_t *mask = &event_data->mask; struct coresight_device *sink; - if (!event_data->snk_config) + if (IS_ERR_OR_NULL(event_data->snk_config)) return; if (WARN_ON(cpumask_empty(mask))) @@ -450,7 +450,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages, event_data->snk_config = sink_ops(sink)->alloc_buffer(sink, event, pages, nr_pages, overwrite); - if (!event_data->snk_config) + if (IS_ERR_OR_NULL(event_data->snk_config)) goto err;
I think the bug is in TRBE. It's the only one that returns an error pointer, but only for -ENOMEM which would normally be NULL for alloc type functions anyway.
Also it's assigned to event_data->snk_config which is later NULL checked rather than IS_ERR_OR_NULL in free_sink_buffer(). Maybe that path doesn't happen, but all instances should be updated anyway.
Oh sorry I was only looking at the update to etm_setup_aux() so both functions are covered. But the other points still apply.
It's much easier to keep it as NULL and make the fix in TRBE. It also wouldn't need to be backported as far.
out:
base-commit: fa71e9cb4cfa59abb196229667ec84929bdc18fe change-id: 20250904-cs_etm_auxsetup_fix_error_handling-cb7e07ed9adf
Best regards,
On Thu, Sep 04, 2025 at 01:11:37PM +0100, James Clark wrote:
[...]
I think the bug is in TRBE. It's the only one that returns an error pointer, but only for -ENOMEM which would normally be NULL for alloc type functions anyway.
Thought again, you are right — it is better to fix this in TRBE, since we don't really get any benefit from an artificial "-ENOMEM".
Will respin a new one.
Thanks, Leo
-
James Clark -
Leo Yan