[PATCH 5.4 197/340] perf intel-pt: Fix premature IPC
From: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Date: 2021-03-01 22:42:57
Also in:
lkml
Subsystem:
performance events subsystem, the rest · Maintainers:
Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo, Namhyung Kim, Linus Torvalds
From: Adrian Hunter <adrian.hunter@intel.com>
[ 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@intel.com>
Reviewed-by: Andi Kleen <redacted>
Cc: Jiri Olsa <redacted>
Cc: linux-kernel@vger.kernel.org
Link: https://lore.kernel.org/r/20210205175350.23817-3-adrian.hunter@intel.com (local)
Signed-off-by: Arnaldo Carvalho de Melo <redacted>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
.../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