From: Adrian Hunter <adrian.hunter@xxxxxxxxx> [ Upstream commit 20aa39708a5999b7921b27482a756766272286ac ] The code assumed a change in cycle count means accurate IPC. That is not correct, for example when sampling both branches and instructions, or at a FUP packet (which is not CYC-eligible) address. Fix by using an explicit flag to indicate when IPC can be sampled. Fixes: 5b1dc0fd1da06 ("perf intel-pt: Add support for samples to contain IPC ratio") Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx> Reviewed-by: Andi Kleen <ak@xxxxxxxxxxxxxxx> Cc: Jiri Olsa <jolsa@xxxxxxxxxx> Cc: linux-kernel@xxxxxxxxxxxxxxx Link: https://lore.kernel.org/r/20210205175350.23817-3-adrian.hunter@xxxxxxxxx Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx> --- .../util/intel-pt-decoder/intel-pt-decoder.c | 11 ++++++++++- .../util/intel-pt-decoder/intel-pt-decoder.h | 1 + tools/perf/util/intel-pt.c | 16 ++++++---------- 3 files changed, 17 insertions(+), 11 deletions(-) diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c index a3fdea49ad663..7f53b63088b2c 100644 --- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c +++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c @@ -2637,9 +2637,18 @@ const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder) } if (intel_pt_sample_time(decoder->pkt_state)) { intel_pt_update_sample_time(decoder); - if (decoder->sample_cyc) + if (decoder->sample_cyc) { decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt; + decoder->state.flags |= INTEL_PT_SAMPLE_IPC; + decoder->sample_cyc = false; + } } + /* + * When using only TSC/MTC to compute cycles, IPC can be + * sampled as soon as the cycle count changes. + */ + if (!decoder->have_cyc) + decoder->state.flags |= INTEL_PT_SAMPLE_IPC; } decoder->state.timestamp = decoder->sample_timestamp; diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h index e289e463d635e..7396da0fa3a7c 100644 --- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h +++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h @@ -17,6 +17,7 @@ #define INTEL_PT_ABORT_TX (1 << 1) #define INTEL_PT_ASYNC (1 << 2) #define INTEL_PT_FUP_IP (1 << 3) +#define INTEL_PT_SAMPLE_IPC (1 << 4) enum intel_pt_sample_type { INTEL_PT_BRANCH = 1 << 0, diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c index 8aeaeba48a41f..d0e0ce11faf58 100644 --- a/tools/perf/util/intel-pt.c +++ b/tools/perf/util/intel-pt.c @@ -1304,7 +1304,8 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq) sample.branch_stack = (struct branch_stack *)&dummy_bs; } - sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_br_cyc_cnt; + if (ptq->state->flags & INTEL_PT_SAMPLE_IPC) + sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_br_cyc_cnt; if (sample.cyc_cnt) { sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_br_insn_cnt; ptq->last_br_insn_cnt = ptq->ipc_insn_cnt; @@ -1366,7 +1367,8 @@ static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq) sample.stream_id = ptq->pt->instructions_id; sample.period = ptq->state->tot_insn_cnt - ptq->last_insn_cnt; - sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_in_cyc_cnt; + if (ptq->state->flags & INTEL_PT_SAMPLE_IPC) + sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_in_cyc_cnt; if (sample.cyc_cnt) { sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_in_insn_cnt; ptq->last_in_insn_cnt = ptq->ipc_insn_cnt; @@ -1901,14 +1903,8 @@ static int intel_pt_sample(struct intel_pt_queue *ptq) ptq->have_sample = false; - if (ptq->state->tot_cyc_cnt > ptq->ipc_cyc_cnt) { - /* - * Cycle count and instruction count only go together to create - * a valid IPC ratio when the cycle count changes. - */ - ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt; - ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt; - } + ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt; + ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt; /* * Do PEBS first to allow for the possibility that the PEBS timestamp -- 2.27.0