* [PATCH] perf, script: Fix brcntr output with --xed
@ 2026-02-18 1:40 Andi Kleen
2026-02-25 17:28 ` Ian Rogers
2026-02-26 18:41 ` Namhyung Kim
0 siblings, 2 replies; 4+ messages in thread
From: Andi Kleen @ 2026-02-18 1:40 UTC (permalink / raw)
To: linux-perf-users; +Cc: Andi Kleen
brcntr in perf script brstack insn currently outputs
$ perf record -j any,counter ...
$ perf script -F +brcntr,+brstackinsn
...
BC1s 3450809 5665912.127194: 100127
cpu_core/cycles/: 7f0475d6cc89 handle_intel.constprop.0+0x2b
(/usr/lib64/ld-linux-
x86-64.so.2)
intel_check_word.constprop.0+224:
00007f0475d6ca7e insn: 00 4b db br_cntr: # PRED 21 cycles [21]
...
This has two issues:
- The description says no event is a single dash, but that is not what is printed.
- The b in brcntr is ambigious with the hex numbers in insns, which
breaks with --xed. It parses the b as another instruction byte and
merges the instruction with a missing b and no space:
$ perf script -F +brstackinsn,+brcntr --xed
...
00005618c6d683b5 jnz 0x5618c6d683bdr_cntr: # PRED 5 cycles [1396] 8.60 IPC
This patches fixes these two problems. It moves the brcntr output into
the "#" comment which also looks nicer and also fixes the no event case.
$ perf script -F +brstackinsn,+brcntr --xed
...
00005618c6d6624f jnz 0x5618c6d65fb7 # br_cntr: - MISPRED 1 cycles [1398] 3.00 IPC
Since the old broken format has shipped for a few releases there is a
risk of breaking some existing parser, but since this is a obscure
feature I hope they're not too common and can adapt.
Signed-off-by: Andi Kleen <andi@firstfloor.org>
---
tools/perf/builtin-script.c | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 62e43d3c5ad7..ac361aa51094 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -1264,11 +1264,11 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
if (PRINT_FIELD(BRCNTR)) {
struct evsel *pos = evsel__leader(evsel);
- unsigned int i = 0, j, num, mask, width;
+ unsigned int i = 0, j, num, mask, width, numprinted = 0;
perf_env__find_br_cntr_info(evsel__env(evsel), NULL, &width);
mask = (1L << width) - 1;
- printed += fprintf(fp, "br_cntr: ");
+ printed += fprintf(fp, "\t# br_cntr: ");
evlist__for_each_entry_from(evsel->evlist, pos) {
if (!(pos->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
continue;
@@ -1276,16 +1276,20 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
break;
num = (br_cntr >> (i++ * width)) & mask;
+ numprinted += num;
if (!verbose) {
for (j = 0; j < num; j++)
printed += fprintf(fp, "%s", pos->abbr_name);
} else
printed += fprintf(fp, "%s %d ", pos->name, num);
}
- printed += fprintf(fp, "\t");
+ if (numprinted == 0 && !verbose)
+ fprintf(fp, "-");
+ printed += fprintf(fp, " ");
}
- printed += fprintf(fp, "#%s%s%s%s",
+ printed += fprintf(fp, "%s%s%s%s%s",
+ !PRINT_FIELD(BRCNTR) ? "#" : "",
en->flags.predicted ? " PRED" : "",
en->flags.mispred ? " MISPRED" : "",
en->flags.in_tx ? " INTX" : "",
--
2.52.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* Re: [PATCH] perf, script: Fix brcntr output with --xed
2026-02-18 1:40 [PATCH] perf, script: Fix brcntr output with --xed Andi Kleen
@ 2026-02-25 17:28 ` Ian Rogers
2026-02-26 1:05 ` Namhyung Kim
2026-02-26 18:41 ` Namhyung Kim
1 sibling, 1 reply; 4+ messages in thread
From: Ian Rogers @ 2026-02-25 17:28 UTC (permalink / raw)
To: Andi Kleen; +Cc: linux-perf-users
On Tue, Feb 17, 2026 at 5:48 PM Andi Kleen <andi@firstfloor.org> wrote:
>
> brcntr in perf script brstack insn currently outputs
>
> $ perf record -j any,counter ...
> $ perf script -F +brcntr,+brstackinsn
> ...
> BC1s 3450809 5665912.127194: 100127
> cpu_core/cycles/: 7f0475d6cc89 handle_intel.constprop.0+0x2b
> (/usr/lib64/ld-linux-
> x86-64.so.2)
> intel_check_word.constprop.0+224:
> 00007f0475d6ca7e insn: 00 4b db br_cntr: # PRED 21 cycles [21]
> ...
>
> This has two issues:
> - The description says no event is a single dash, but that is not what is printed.
> - The b in brcntr is ambigious with the hex numbers in insns, which
> breaks with --xed. It parses the b as another instruction byte and
> merges the instruction with a missing b and no space:
>
> $ perf script -F +brstackinsn,+brcntr --xed
> ...
> 00005618c6d683b5 jnz 0x5618c6d683bdr_cntr: # PRED 5 cycles [1396] 8.60 IPC
>
> This patches fixes these two problems. It moves the brcntr output into
> the "#" comment which also looks nicer and also fixes the no event case.
>
> $ perf script -F +brstackinsn,+brcntr --xed
> ...
> 00005618c6d6624f jnz 0x5618c6d65fb7 # br_cntr: - MISPRED 1 cycles [1398] 3.00 IPC
>
> Since the old broken format has shipped for a few releases there is a
> risk of breaking some existing parser, but since this is a obscure
> feature I hope they're not too common and can adapt.
>
> Signed-off-by: Andi Kleen <andi@firstfloor.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Thanks.
Ian
> ---
> tools/perf/builtin-script.c | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index 62e43d3c5ad7..ac361aa51094 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -1264,11 +1264,11 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
>
> if (PRINT_FIELD(BRCNTR)) {
> struct evsel *pos = evsel__leader(evsel);
> - unsigned int i = 0, j, num, mask, width;
> + unsigned int i = 0, j, num, mask, width, numprinted = 0;
>
> perf_env__find_br_cntr_info(evsel__env(evsel), NULL, &width);
> mask = (1L << width) - 1;
> - printed += fprintf(fp, "br_cntr: ");
> + printed += fprintf(fp, "\t# br_cntr: ");
> evlist__for_each_entry_from(evsel->evlist, pos) {
> if (!(pos->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
> continue;
> @@ -1276,16 +1276,20 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
> break;
>
> num = (br_cntr >> (i++ * width)) & mask;
> + numprinted += num;
> if (!verbose) {
> for (j = 0; j < num; j++)
> printed += fprintf(fp, "%s", pos->abbr_name);
> } else
> printed += fprintf(fp, "%s %d ", pos->name, num);
> }
> - printed += fprintf(fp, "\t");
> + if (numprinted == 0 && !verbose)
> + fprintf(fp, "-");
> + printed += fprintf(fp, " ");
> }
>
> - printed += fprintf(fp, "#%s%s%s%s",
> + printed += fprintf(fp, "%s%s%s%s%s",
> + !PRINT_FIELD(BRCNTR) ? "#" : "",
> en->flags.predicted ? " PRED" : "",
> en->flags.mispred ? " MISPRED" : "",
> en->flags.in_tx ? " INTX" : "",
> --
> 2.52.0
>
>
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [PATCH] perf, script: Fix brcntr output with --xed
2026-02-25 17:28 ` Ian Rogers
@ 2026-02-26 1:05 ` Namhyung Kim
0 siblings, 0 replies; 4+ messages in thread
From: Namhyung Kim @ 2026-02-26 1:05 UTC (permalink / raw)
To: Ian Rogers; +Cc: Andi Kleen, linux-perf-users
Hello,
On Wed, Feb 25, 2026 at 09:28:23AM -0800, Ian Rogers wrote:
> On Tue, Feb 17, 2026 at 5:48 PM Andi Kleen <andi@firstfloor.org> wrote:
> >
> > brcntr in perf script brstack insn currently outputs
> >
> > $ perf record -j any,counter ...
> > $ perf script -F +brcntr,+brstackinsn
> > ...
> > BC1s 3450809 5665912.127194: 100127
> > cpu_core/cycles/: 7f0475d6cc89 handle_intel.constprop.0+0x2b
> > (/usr/lib64/ld-linux-
> > x86-64.so.2)
> > intel_check_word.constprop.0+224:
> > 00007f0475d6ca7e insn: 00 4b db br_cntr: # PRED 21 cycles [21]
> > ...
> >
> > This has two issues:
> > - The description says no event is a single dash, but that is not what is printed.
> > - The b in brcntr is ambigious with the hex numbers in insns, which
> > breaks with --xed. It parses the b as another instruction byte and
> > merges the instruction with a missing b and no space:
> >
> > $ perf script -F +brstackinsn,+brcntr --xed
> > ...
> > 00005618c6d683b5 jnz 0x5618c6d683bdr_cntr: # PRED 5 cycles [1396] 8.60 IPC
> >
> > This patches fixes these two problems. It moves the brcntr output into
> > the "#" comment which also looks nicer and also fixes the no event case.
> >
> > $ perf script -F +brstackinsn,+brcntr --xed
> > ...
> > 00005618c6d6624f jnz 0x5618c6d65fb7 # br_cntr: - MISPRED 1 cycles [1398] 3.00 IPC
> >
> > Since the old broken format has shipped for a few releases there is a
> > risk of breaking some existing parser, but since this is a obscure
> > feature I hope they're not too common and can adapt.
:)
> >
> > Signed-off-by: Andi Kleen <andi@firstfloor.org>
>
> Reviewed-by: Ian Rogers <irogers@google.com>
>
> Thanks.
> Ian
>
> > ---
> > tools/perf/builtin-script.c | 12 ++++++++----
> > 1 file changed, 8 insertions(+), 4 deletions(-)
> >
> > diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> > index 62e43d3c5ad7..ac361aa51094 100644
> > --- a/tools/perf/builtin-script.c
> > +++ b/tools/perf/builtin-script.c
> > @@ -1264,11 +1264,11 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
> >
> > if (PRINT_FIELD(BRCNTR)) {
> > struct evsel *pos = evsel__leader(evsel);
> > - unsigned int i = 0, j, num, mask, width;
> > + unsigned int i = 0, j, num, mask, width, numprinted = 0;
> >
> > perf_env__find_br_cntr_info(evsel__env(evsel), NULL, &width);
> > mask = (1L << width) - 1;
> > - printed += fprintf(fp, "br_cntr: ");
> > + printed += fprintf(fp, "\t# br_cntr: ");
> > evlist__for_each_entry_from(evsel->evlist, pos) {
> > if (!(pos->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
> > continue;
> > @@ -1276,16 +1276,20 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
> > break;
> >
> > num = (br_cntr >> (i++ * width)) & mask;
> > + numprinted += num;
> > if (!verbose) {
> > for (j = 0; j < num; j++)
> > printed += fprintf(fp, "%s", pos->abbr_name);
> > } else
> > printed += fprintf(fp, "%s %d ", pos->name, num);
> > }
> > - printed += fprintf(fp, "\t");
> > + if (numprinted == 0 && !verbose)
> > + fprintf(fp, "-");
I think it should be: printed += fprintf(fp, "-");
Anyway, it's a nitpick. I can update it locally.
Thanks,
Namhyung
> > + printed += fprintf(fp, " ");
> > }
> >
> > - printed += fprintf(fp, "#%s%s%s%s",
> > + printed += fprintf(fp, "%s%s%s%s%s",
> > + !PRINT_FIELD(BRCNTR) ? "#" : "",
> > en->flags.predicted ? " PRED" : "",
> > en->flags.mispred ? " MISPRED" : "",
> > en->flags.in_tx ? " INTX" : "",
> > --
> > 2.52.0
> >
> >
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] perf, script: Fix brcntr output with --xed
2026-02-18 1:40 [PATCH] perf, script: Fix brcntr output with --xed Andi Kleen
2026-02-25 17:28 ` Ian Rogers
@ 2026-02-26 18:41 ` Namhyung Kim
1 sibling, 0 replies; 4+ messages in thread
From: Namhyung Kim @ 2026-02-26 18:41 UTC (permalink / raw)
To: linux-perf-users, Andi Kleen
On Tue, 17 Feb 2026 17:40:56 -0800, Andi Kleen wrote:
> brcntr in perf script brstack insn currently outputs
>
> $ perf record -j any,counter ...
> $ perf script -F +brcntr,+brstackinsn
> ...
> BC1s 3450809 5665912.127194: 100127
> cpu_core/cycles/: 7f0475d6cc89 handle_intel.constprop.0+0x2b
> (/usr/lib64/ld-linux-
> x86-64.so.2)
> intel_check_word.constprop.0+224:
> 00007f0475d6ca7e insn: 00 4b db br_cntr: # PRED 21 cycles [21]
> ...
>
> [...]
Applied to perf-tools-next, thanks!
Best regards,
Namhyung
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-02-26 18:41 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-02-18 1:40 [PATCH] perf, script: Fix brcntr output with --xed Andi Kleen
2026-02-25 17:28 ` Ian Rogers
2026-02-26 1:05 ` Namhyung Kim
2026-02-26 18:41 ` Namhyung Kim
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox