[PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Graham Woodward]

Currently the --itrace=b will only show branch-misses but this change
allows perf to synthesize branches as well.

The change also incorporates the ability to display the target
addresses when specifying the addr field if the instruction is a branch.

Graham Woodward (4):
  perf arm-spe: Set sample.addr to target address for instruction sample
  perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
  perf arm-spe: Correctly set sample flags
  perf arm-spe: Update --itrace help text

 tools/perf/Documentation/itrace.txt       |  2 +-
 tools/perf/Documentation/perf-arm-spe.txt |  2 +-
 tools/perf/builtin-script.c               |  1 +
 tools/perf/util/arm-spe.c                 | 31 ++++++++++++++++++-----
 tools/perf/util/auxtrace.h                |  3 +--
 tools/perf/util/event.h                   |  1 +
 6 files changed, 29 insertions(+), 11 deletions(-)

-- 
2.40.1

[PATCH v1 1/4] perf arm-spe: Set sample.addr to target address for instruction sample

[Graham Woodward]

For an instruction sample, assign the target address to the field
'to_ip'.
If it is a non-branch record, to_ip will be 0, presenting a non-valid
target address.

Signed-off-by: Graham Woodward <graham.woodward@arm.com>
---
 tools/perf/util/arm-spe.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 856cc36f33d7..49b763807360 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -418,7 +418,7 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
 
 	sample.id = spe_events_id;
 	sample.stream_id = spe_events_id;
-	sample.addr = record->virt_addr;
+	sample.addr = record->to_ip;
 	sample.phys_addr = record->phys_addr;
 	sample.data_src = data_src;
 	sample.period = spe->instructions_sample_period;
-- 
2.40.1

[PATCH v1 2/4] perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches

[Graham Woodward]

Instead of checking the type for just branch misses, we can instead
check for the OP_BRANCH_ERET and synthesise branches as well as
branch misses.

Signed-off-by: Graham Woodward <graham.woodward@arm.com>
---
 tools/perf/util/arm-spe.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 49b763807360..e60e21d24735 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -68,7 +68,7 @@ struct arm_spe {
 	u64				llc_access_id;
 	u64				tlb_miss_id;
 	u64				tlb_access_id;
-	u64				branch_miss_id;
+	u64				branch_id;
 	u64				remote_access_id;
 	u64				memory_id;
 	u64				instructions_id;
@@ -672,8 +672,8 @@ static int arm_spe_sample(struct arm_spe_queue *speq)
 		}
 	}
 
-	if (spe->sample_branch && (record->type & ARM_SPE_BRANCH_MISS)) {
-		err = arm_spe__synth_branch_sample(speq, spe->branch_miss_id);
+	if (spe->sample_branch && (record->op & ARM_SPE_OP_BRANCH_ERET)) {
+		err = arm_spe__synth_branch_sample(speq, spe->branch_id);
 		if (err)
 			return err;
 	}
@@ -1385,12 +1385,12 @@ arm_spe_synth_events(struct arm_spe *spe, struct perf_session *session)
 	if (spe->synth_opts.branches) {
 		spe->sample_branch = true;
 
-		/* Branch miss */
+		/* Branch */
 		err = perf_session__deliver_synth_attr_event(session, &attr, id);
 		if (err)
 			return err;
-		spe->branch_miss_id = id;
-		arm_spe_set_event_name(evlist, id, "branch-miss");
+		spe->branch_id = id;
+		arm_spe_set_event_name(evlist, id, "branch");
 		id += 1;
 	}
 
-- 
2.40.1

[PATCH v1 3/4] perf arm-spe: Correctly set sample flags

[Graham Woodward]

Set flags on all synthesized instruction and branch samples.

Signed-off-by: Graham Woodward <graham.woodward@arm.com>
---
 tools/perf/builtin-script.c |  1 +
 tools/perf/util/arm-spe.c   | 17 +++++++++++++++++
 tools/perf/util/event.h     |  1 +
 3 files changed, 19 insertions(+)

diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 8c5d5cecfba4..6b6d4472db6e 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -1728,6 +1728,7 @@ static struct {
 	{PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_TRACE_END, "tr end"},
 	{PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_CALL | PERF_IP_FLAG_VMENTRY, "vmentry"},
 	{PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_CALL | PERF_IP_FLAG_VMEXIT, "vmexit"},
+	{PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_BRANCH_MISS, "br miss"},
 	{0, NULL}
 };
 
diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index e60e21d24735..a291a412f6a1 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -100,6 +100,7 @@ struct arm_spe_queue {
 	u64				timestamp;
 	struct thread			*thread;
 	u64				period_instructions;
+	u32				flags;
 };
 
 static void arm_spe_dump(struct arm_spe *spe __maybe_unused,
@@ -394,6 +395,7 @@ static int arm_spe__synth_branch_sample(struct arm_spe_queue *speq,
 	sample.stream_id = spe_events_id;
 	sample.addr = record->to_ip;
 	sample.weight = record->latency;
+	sample.flags = speq->flags;
 
 	return arm_spe_deliver_synth_event(spe, speq, event, &sample);
 }
@@ -423,6 +425,7 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
 	sample.data_src = data_src;
 	sample.period = spe->instructions_sample_period;
 	sample.weight = record->latency;
+	sample.flags = speq->flags;
 
 	return arm_spe_deliver_synth_event(spe, speq, event, &sample);
 }
@@ -440,6 +443,19 @@ static const struct midr_range common_ds_encoding_cpus[] = {
 	{},
 };
 
+static void arm_spe__sample_flags(struct arm_spe_queue *speq)
+{
+	const struct arm_spe_record *record = &speq->decoder->record;
+
+	speq->flags = 0;
+	if (record->op & ARM_SPE_OP_BRANCH_ERET) {
+		speq->flags = PERF_IP_FLAG_BRANCH;
+
+		if (record->type & ARM_SPE_BRANCH_MISS)
+			speq->flags |= PERF_IP_FLAG_BRANCH_MISS;
+	}
+}
+
 static void arm_spe__synth_data_source_common(const struct arm_spe_record *record,
 					      union perf_mem_data_src *data_src)
 {
@@ -622,6 +638,7 @@ static int arm_spe_sample(struct arm_spe_queue *speq)
 	u64 data_src;
 	int err;
 
+	arm_spe__sample_flags(speq);
 	data_src = arm_spe__synth_data_source(speq, record);
 
 	if (spe->sample_flc) {
diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index f8742e6230a5..2744c54f404e 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -66,6 +66,7 @@ enum {
 	PERF_IP_FLAG_VMEXIT		= 1ULL << 12,
 	PERF_IP_FLAG_INTR_DISABLE	= 1ULL << 13,
 	PERF_IP_FLAG_INTR_TOGGLE	= 1ULL << 14,
+	PERF_IP_FLAG_BRANCH_MISS	= 1ULL << 15,
 };
 
 #define PERF_IP_FLAG_CHARS "bcrosyiABExghDt"
-- 
2.40.1

[PATCH v1 4/4] perf arm-spe: Update --itrace help text

[Graham Woodward]

The --itrace help now needs updating to reflect that
the --itrace=b argument sythesises branches as well
as branch misses.

Signed-off-by: Graham Woodward <graham.woodward@arm.com>
---
 tools/perf/Documentation/itrace.txt       | 2 +-
 tools/perf/Documentation/perf-arm-spe.txt | 2 +-
 tools/perf/util/auxtrace.h                | 3 +--
 3 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/tools/perf/Documentation/itrace.txt b/tools/perf/Documentation/itrace.txt
index 19cc179be9a7..40476b227f8d 100644
--- a/tools/perf/Documentation/itrace.txt
+++ b/tools/perf/Documentation/itrace.txt
@@ -1,6 +1,6 @@
 		i	synthesize instructions events
 		y	synthesize cycles events
-		b	synthesize branches events (branch misses for Arm SPE)
+		b	synthesize branches events
 		c	synthesize branches events (calls only)
 		r	synthesize branches events (returns only)
 		x	synthesize transactions events
diff --git a/tools/perf/Documentation/perf-arm-spe.txt b/tools/perf/Documentation/perf-arm-spe.txt
index 0a3eda482307..de2b0b479249 100644
--- a/tools/perf/Documentation/perf-arm-spe.txt
+++ b/tools/perf/Documentation/perf-arm-spe.txt
@@ -187,7 +187,7 @@ groups:
   7 llc-access
   2 tlb-miss
   1K tlb-access
-  36 branch-miss
+  36 branch
   0 remote-access
   900 memory
 
diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
index a1895a4f530b..dddaf4f3ffed 100644
--- a/tools/perf/util/auxtrace.h
+++ b/tools/perf/util/auxtrace.h
@@ -75,7 +75,6 @@ enum itrace_period_type {
  *          (not fully accurate, since CYC packets are only emitted
  *          together with other events, such as branches)
  * @branches: whether to synthesize 'branches' events
- *            (branch misses only for Arm SPE)
  * @transactions: whether to synthesize events for transactions
  * @ptwrites: whether to synthesize events for ptwrites
  * @pwr_events: whether to synthesize power events
@@ -650,7 +649,7 @@ bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
 #define ITRACE_HELP \
 "				i[period]:    		synthesize instructions events\n" \
 "				y[period]:    		synthesize cycles events (same period as i)\n" \
-"				b:	    		synthesize branches events (branch misses for Arm SPE)\n" \
+"				b:	    		synthesize branches events\n" \
 "				c:	    		synthesize branches events (calls only)\n"	\
 "				r:	    		synthesize branches events (returns only)\n" \
 "				x:	    		synthesize transactions events\n"		\
-- 
2.40.1

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Leo Yan]

On Fri, Oct 25, 2024 at 03:30:05PM +0100, Graham Woodward wrote:
> 
> Currently the --itrace=b will only show branch-misses but this change
> allows perf to synthesize branches as well.
> 
> The change also incorporates the ability to display the target
> addresses when specifying the addr field if the instruction is a branch.

Tested for this series:

  # perf record -e arm_spe_0/branch_filter=1,load_filter=1/u \
      -- ./false_sharing.exe 1

  # perf script --itrace=i10ib  -F,+addr,+flags
    false_sharing.e  880532 [005] 1852579.389533:          1                                    branch:   jmp                       ffff91beb224     ffff91beb220 __GI___tunables_init+0x40 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389538:          1                                    branch:   jmp                       ffff91bec318     ffff91bec314 _dl_next_ld_env_entry+0x24 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389620:          1                                    branch:   jmp                       ffff91be0f14     ffff91be0f10 _dl_new_object+0x168 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389802:          1                                    branch:   jmp                       ffff91be2cf0     ffff91be2cec _dl_map_object_deps+0x3f4 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389802:         10                              instructions:   jmp                       ffff91be2cf0     ffff91be2cec _dl_map_object_deps+0x3f4 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389824:          1                                    branch:   br miss                   ffff91bee4e4     ffff91bee4e0 strcmp+0xa0 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389849:          1                                    branch:   jmp                       ffff91be1868     ffff91be1880 _dl_relocate_object+0x4a8 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389858:          1                                    branch:   jmp                       ffff91be1868     ffff91be1880 _dl_relocate_object+0x4a8 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389861:          1                                    branch:   jmp                       ffff91be1c20     ffff91be1bcc _dl_relocate_object+0x7f4 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389875:         10                              instructions:                                        0     ffff91bdfe38 _dl_lookup_symbol_x+0x58 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389876:          1                                    branch:   jmp                       ffff91bdf3a8     ffff91bdf434 do_lookup_x+0x114 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389879:          1                                    branch:   jmp                       ffff91be18ec     ffff91be18e8 _dl_relocate_object+0x510 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389886:          1                                    branch:   jmp                       ffff91bee440     ffff91bdf2dc check_match+0x154 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389890:          1                                    branch:   jmp                       ffff91bdfed4     ffff91bdfed0 _dl_lookup_symbol_x+0xf0 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389893:         10                              instructions:                                        0     ffff91be1974 _dl_relocate_object+0x59c (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389894:          1                                    branch:   jmp                       ffff91bdf3f4     ffff91bdf3f0 do_lookup_x+0xd0 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
    false_sharing.e  880532 [005] 1852579.389906:          1                                    branch:   jmp                       ffff91bdfea4     ffff91bdfe90 _dl_lookup_symbol_x+0xb0 (/usr/lib/aarch64-linux-gnu/ld-2.31.so)

  # perf test "Check Arm SPE"
  114: Check Arm SPE trace data recording and synthesized samples      : Ok
  115: Check Arm SPE doesn't hang when there are forks                 : Ok

Tested-by: Leo Yan <leo.yan@arm.com>

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[James Clark]

On 25/10/2024 3:30 pm, Graham Woodward wrote:
> Currently the --itrace=b will only show branch-misses but this change
> allows perf to synthesize branches as well.
> 
> The change also incorporates the ability to display the target
> addresses when specifying the addr field if the instruction is a branch.
> 
> Graham Woodward (4):
>    perf arm-spe: Set sample.addr to target address for instruction sample
>    perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
>    perf arm-spe: Correctly set sample flags
>    perf arm-spe: Update --itrace help text
> 
>   tools/perf/Documentation/itrace.txt       |  2 +-
>   tools/perf/Documentation/perf-arm-spe.txt |  2 +-
>   tools/perf/builtin-script.c               |  1 +
>   tools/perf/util/arm-spe.c                 | 31 ++++++++++++++++++-----
>   tools/perf/util/auxtrace.h                |  3 +--
>   tools/perf/util/event.h                   |  1 +
>   6 files changed, 29 insertions(+), 11 deletions(-)
> 

Don't forget to pickup the review tags from the previous versions. If 
you use the b4 tool it does it automatically:

Reviewed-by: James Clark <james.clark@linaro.org>

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Namhyung Kim]

Hello,

On Fri, Oct 25, 2024 at 03:30:05PM +0100, Graham Woodward wrote:
> Currently the --itrace=b will only show branch-misses but this change
> allows perf to synthesize branches as well.
> 
> The change also incorporates the ability to display the target
> addresses when specifying the addr field if the instruction is a branch.
> 
> Graham Woodward (4):
>   perf arm-spe: Set sample.addr to target address for instruction sample
>   perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
>   perf arm-spe: Correctly set sample flags
>   perf arm-spe: Update --itrace help text

It doesn't apply to perf-tools-next cleanly.  Can you please rebase?

Thanks,
Namhyung

> 
>  tools/perf/Documentation/itrace.txt       |  2 +-
>  tools/perf/Documentation/perf-arm-spe.txt |  2 +-
>  tools/perf/builtin-script.c               |  1 +
>  tools/perf/util/arm-spe.c                 | 31 ++++++++++++++++++-----
>  tools/perf/util/auxtrace.h                |  3 +--
>  tools/perf/util/event.h                   |  1 +
>  6 files changed, 29 insertions(+), 11 deletions(-)
> 
> -- 
> 2.40.1
> 

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Leo Yan]

Hi Namhyung,

On Mon, Oct 28, 2024 at 09:40:21AM -0700, Namhyung Kim wrote:
> 
> Hello,
> 
> On Fri, Oct 25, 2024 at 03:30:05PM +0100, Graham Woodward wrote:
> > Currently the --itrace=b will only show branch-misses but this change
> > allows perf to synthesize branches as well.
> >
> > The change also incorporates the ability to display the target
> > addresses when specifying the addr field if the instruction is a branch.
> >
> > Graham Woodward (4):
> >   perf arm-spe: Set sample.addr to target address for instruction sample
> >   perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
> >   perf arm-spe: Correctly set sample flags
> >   perf arm-spe: Update --itrace help text
> 
> It doesn't apply to perf-tools-next cleanly.  Can you please rebase?

I confirmed this series can apply cleanly on the branch [1] with the
latest commit 150dab31d560 ("perf disasm: Fix not cleaning up
disasm_line in symbol__disassemble_raw()"):

  [1] https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git
      branch: perf-tools-next

If you are suggesting for the branch:

  [2] https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools.git
      branch: perf-tools

You can see it misses some Arm SPE patches which have been picked up
in the repo [1].

Please kindly suggest what is right thing to do.

Thanks,
Leo

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Namhyung Kim]

Hi Leo,

On Tue, Oct 29, 2024 at 05:03:46PM +0000, Leo Yan wrote:
> Hi Namhyung,
> 
> On Mon, Oct 28, 2024 at 09:40:21AM -0700, Namhyung Kim wrote:
> > 
> > Hello,
> > 
> > On Fri, Oct 25, 2024 at 03:30:05PM +0100, Graham Woodward wrote:
> > > Currently the --itrace=b will only show branch-misses but this change
> > > allows perf to synthesize branches as well.
> > >
> > > The change also incorporates the ability to display the target
> > > addresses when specifying the addr field if the instruction is a branch.
> > >
> > > Graham Woodward (4):
> > >   perf arm-spe: Set sample.addr to target address for instruction sample
> > >   perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
> > >   perf arm-spe: Correctly set sample flags
> > >   perf arm-spe: Update --itrace help text
> > 
> > It doesn't apply to perf-tools-next cleanly.  Can you please rebase?
> 
> I confirmed this series can apply cleanly on the branch [1] with the
> latest commit 150dab31d560 ("perf disasm: Fix not cleaning up
> disasm_line in symbol__disassemble_raw()"):
> 
>   [1] https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git
>       branch: perf-tools-next
> 
> If you are suggesting for the branch:
> 
>   [2] https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools.git
>       branch: perf-tools
> 
> You can see it misses some Arm SPE patches which have been picked up
> in the repo [1].
> 
> Please kindly suggest what is right thing to do.

Sorry, my bad.  It works ok.  I'll add it to tmp.perf-tools-next branch
and run some tests.

Thanks,
Namhyung

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Leo Yan]

Hi Namhyung,

On Tue, Oct 29, 2024 at 04:09:46PM -0700, Namhyung Kim wrote:

[...]

> > Please kindly suggest what is right thing to do.
> 
> Sorry, my bad.  It works ok.  I'll add it to tmp.perf-tools-next branch
> and run some tests.

Thanks for confirmation!

Leo

Re: [PATCH v1 0/4] perf arm-spe: Allow synthesizing of branch

[Namhyung Kim]

On Fri, 25 Oct 2024 15:30:05 +0100, Graham Woodward wrote:

> Currently the --itrace=b will only show branch-misses but this change
> allows perf to synthesize branches as well.
> 
> The change also incorporates the ability to display the target
> addresses when specifying the addr field if the instruction is a branch.
> 
> Graham Woodward (4):
>   perf arm-spe: Set sample.addr to target address for instruction sample
>   perf arm-spe: Use ARM_SPE_OP_BRANCH_ERET when synthesizing branches
>   perf arm-spe: Correctly set sample flags
>   perf arm-spe: Update --itrace help text
> 
> [...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung