On 15/09/2026 16:49, Leo Yan wrote:
Perf clears the IP of a user-only PMU sample when its interrupt skids into the kernel. The timestamp and TID remain valid, but the zero-IP check prevents --itrace=L from attaching decoded branch history.
Remove the IP check and keep matching samples to trace by time and thread.
Assisted-by: Codex:gpt-6 Signed-off-by: Leo Yan leo.yan@arm.com
tools/perf/util/cs-etm.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c index 00407a80933e1c0b76260aed5894dece50b9aecd..baa962085dba5226d9729a0ee8039374d5fcb1a0 100644 --- a/tools/perf/util/cs-etm.c +++ b/tools/perf/util/cs-etm.c @@ -3097,6 +3097,8 @@ static int cs_etm__br_stack_init(struct cs_etm_auxtrace *etm, /*
- Add decoded branch history to an existing sample. The sample keeps its own
- ip, callchain and event identity; only an absent branch stack is filled in.
- A user-only PMU sample can have a zero IP when the interrupt skids into
*/ static int cs_etm__process_sample(struct cs_etm_auxtrace *etm, struct perf_session *session,
- the kernel; its thread and trace window still identify valid history.
@@ -3107,7 +3109,7 @@ static int cs_etm__process_sample(struct cs_etm_auxtrace *etm, int err; if (!etm->synth_opts.add_last_branch || sample->branch_stack ||
!sample->ip || !sample->time || sample->time == (u64)-1)
!sample->time || sample->time == (u64)-1)/* Adding branch history to existing samples supports the host only */
Looks like we can fix this up in Amir's patch that introduced it when I post it as part of the JIT stuff. Then this commit can be dropped.