2021-09-29 10:44:43

by Bayduraev, Alexey V

Subject: [PATCH] perf report: Output non-zero offset for decompressed records

Print offset of PERF_RECORD_COMPRESSED record instead of zero for
decompressed records in raw trace dump (-D option of perf-report):

0x17cf08 [0x28]: event: 9

instead of:

0 [0x28]: event: 9

The fix is not critical, because currently file_pos for compressed
events is used in perf_session__process_event only to show offsets
in the raw dump.

This patch was separated from patchset:
https://lore.kernel.org/lkml/[email protected]/
and was already rewieved.

Acked-by: Namhyung Kim <[email protected]>
Acked-by: Andi Kleen <[email protected]>
Reviewed-by: Riccardo Mancini <[email protected]>
Tested-by: Riccardo Mancini <[email protected]>
Signed-off-by: Alexey Bayduraev <[email protected]>
---
tools/perf/util/session.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index 069c2cfdd3be..352f16076e01 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -2116,7 +2116,7 @@ fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
static int __perf_session__process_decomp_events(struct perf_session *session)
{
s64 skip;
- u64 size, file_pos = 0;
+ u64 size;
struct decomp *decomp = session->decomp_last;

if (!decomp)
@@ -2132,7 +2132,7 @@ static int __perf_session__process_decomp_events(struct perf_session *session)
size = event->header.size;

if (size < sizeof(struct perf_event_header) ||
- (skip = perf_session__process_event(session, event, file_pos)) < 0) {
+ (skip = perf_session__process_event(session, event, decomp->file_pos)) < 0) {
pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
decomp->file_pos + decomp->head, event->header.size, event->header.type);
return -EINVAL;
--
2.19.0

2021-10-04 06:43:33

by Jiri Olsa

Subject: Re: [PATCH] perf report: Output non-zero offset for decompressed records

On Wed, Sep 29, 2021 at 12:14:45PM +0300, Alexey Bayduraev wrote:
> Print offset of PERF_RECORD_COMPRESSED record instead of zero for
> decompressed records in raw trace dump (-D option of perf-report):
>
> 0x17cf08 [0x28]: event: 9
>
> instead of:
>
> 0 [0x28]: event: 9
>
> The fix is not critical, because currently file_pos for compressed
> events is used in perf_session__process_event only to show offsets
> in the raw dump.

hi,
I don't mind the change just curious, because I see also:

perf_session__process_event
perf_session__process_user_event
lseek(fd, file_offset, ...

which is not raw dump as the comment suggests

thanks,
jirka

>
> This patch was separated from patchset:
> https://lore.kernel.org/lkml/[email protected]/
> and was already rewieved.
>
> Acked-by: Namhyung Kim <[email protected]>
> Acked-by: Andi Kleen <[email protected]>
> Reviewed-by: Riccardo Mancini <[email protected]>
> Tested-by: Riccardo Mancini <[email protected]>
> Signed-off-by: Alexey Bayduraev <[email protected]>
> ---
> tools/perf/util/session.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
> index 069c2cfdd3be..352f16076e01 100644
> --- a/tools/perf/util/session.c
> +++ b/tools/perf/util/session.c
> @@ -2116,7 +2116,7 @@ fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
> static int __perf_session__process_decomp_events(struct perf_session *session)
> {
> s64 skip;
> - u64 size, file_pos = 0;
> + u64 size;
> struct decomp *decomp = session->decomp_last;
>
> if (!decomp)
> @@ -2132,7 +2132,7 @@ static int __perf_session__process_decomp_events(struct perf_session *session)
> size = event->header.size;
>
> if (size < sizeof(struct perf_event_header) ||
> - (skip = perf_session__process_event(session, event, file_pos)) < 0) {
> + (skip = perf_session__process_event(session, event, decomp->file_pos)) < 0) {
> pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
> decomp->file_pos + decomp->head, event->header.size, event->header.type);
> return -EINVAL;
> --
> 2.19.0
>

2021-10-04 07:03:12

by Bayduraev, Alexey V

Subject: Re: [PATCH] perf report: Output non-zero offset for decompressed records

On 04.10.2021 9:42, Jiri Olsa wrote:
> On Wed, Sep 29, 2021 at 12:14:45PM +0300, Alexey Bayduraev wrote:
>> Print offset of PERF_RECORD_COMPRESSED record instead of zero for
>> decompressed records in raw trace dump (-D option of perf-report):
>>
>> 0x17cf08 [0x28]: event: 9
>>
>> instead of:
>>
>> 0 [0x28]: event: 9
>>
>> The fix is not critical, because currently file_pos for compressed
>> events is used in perf_session__process_event only to show offsets
>> in the raw dump.
>
> hi,
> I don't mind the change just curious, because I see also:
>
> perf_session__process_event
> perf_session__process_user_event
> lseek(fd, file_offset, ...
>
> which is not raw dump as the comment suggests

Hi,

Yes, but this "lseek" only works for user events, whereas the
PERF_RECORD_COMPRESSED record shouln't contain such events.
Currently, the PERF_RECORD_COMPRESSED container can only pack
kernel events.

Regards,
Alexey

>
> thanks,
> jirka
>
>>
>> This patch was separated from patchset:
>> https://lore.kernel.org/lkml/[email protected]/
>> and was already rewieved.
>>
>> Acked-by: Namhyung Kim <[email protected]>
>> Acked-by: Andi Kleen <[email protected]>
>> Reviewed-by: Riccardo Mancini <[email protected]>
>> Tested-by: Riccardo Mancini <[email protected]>
>> Signed-off-by: Alexey Bayduraev <[email protected]>
>> ---
>> tools/perf/util/session.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
>> index 069c2cfdd3be..352f16076e01 100644
>> --- a/tools/perf/util/session.c
>> +++ b/tools/perf/util/session.c
>> @@ -2116,7 +2116,7 @@ fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
>> static int __perf_session__process_decomp_events(struct perf_session *session)
>> {
>> s64 skip;
>> - u64 size, file_pos = 0;
>> + u64 size;
>> struct decomp *decomp = session->decomp_last;
>>
>> if (!decomp)
>> @@ -2132,7 +2132,7 @@ static int __perf_session__process_decomp_events(struct perf_session *session)
>> size = event->header.size;
>>
>> if (size < sizeof(struct perf_event_header) ||
>> - (skip = perf_session__process_event(session, event, file_pos)) < 0) {
>> + (skip = perf_session__process_event(session, event, decomp->file_pos)) < 0) {
>> pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
>> decomp->file_pos + decomp->head, event->header.size, event->header.type);
>> return -EINVAL;
>> --
>> 2.19.0
>>
>

2021-10-04 07:08:34

by Jiri Olsa

Subject: Re: [PATCH] perf report: Output non-zero offset for decompressed records

On Mon, Oct 04, 2021 at 10:00:52AM +0300, Bayduraev, Alexey V wrote:
>
> On 04.10.2021 9:42, Jiri Olsa wrote:
> > On Wed, Sep 29, 2021 at 12:14:45PM +0300, Alexey Bayduraev wrote:
> >> Print offset of PERF_RECORD_COMPRESSED record instead of zero for
> >> decompressed records in raw trace dump (-D option of perf-report):
> >>
> >> 0x17cf08 [0x28]: event: 9
> >>
> >> instead of:
> >>
> >> 0 [0x28]: event: 9
> >>
> >> The fix is not critical, because currently file_pos for compressed
> >> events is used in perf_session__process_event only to show offsets
> >> in the raw dump.
> >
> > hi,
> > I don't mind the change just curious, because I see also:
> >
> > perf_session__process_event
> > perf_session__process_user_event
> > lseek(fd, file_offset, ...
> >
> > which is not raw dump as the comment suggests
>
> Hi,
>
> Yes, but this "lseek" only works for user events, whereas the
> PERF_RECORD_COMPRESSED record shouln't contain such events.
> Currently, the PERF_RECORD_COMPRESSED container can only pack
> kernel events.

Acked-by: Jiri Olsa <[email protected]>

thanks,
jirka

>
> Regards,
> Alexey
>
> >
> > thanks,
> > jirka
> >
> >>
> >> This patch was separated from patchset:
> >> https://lore.kernel.org/lkml/[email protected]/
> >> and was already rewieved.
> >>
> >> Acked-by: Namhyung Kim <[email protected]>
> >> Acked-by: Andi Kleen <[email protected]>
> >> Reviewed-by: Riccardo Mancini <[email protected]>
> >> Tested-by: Riccardo Mancini <[email protected]>
> >> Signed-off-by: Alexey Bayduraev <[email protected]>
> >> ---
> >> tools/perf/util/session.c | 4 ++--
> >> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
> >> index 069c2cfdd3be..352f16076e01 100644
> >> --- a/tools/perf/util/session.c
> >> +++ b/tools/perf/util/session.c
> >> @@ -2116,7 +2116,7 @@ fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
> >> static int __perf_session__process_decomp_events(struct perf_session *session)
> >> {
> >> s64 skip;
> >> - u64 size, file_pos = 0;
> >> + u64 size;
> >> struct decomp *decomp = session->decomp_last;
> >>
> >> if (!decomp)
> >> @@ -2132,7 +2132,7 @@ static int __perf_session__process_decomp_events(struct perf_session *session)
> >> size = event->header.size;
> >>
> >> if (size < sizeof(struct perf_event_header) ||
> >> - (skip = perf_session__process_event(session, event, file_pos)) < 0) {
> >> + (skip = perf_session__process_event(session, event, decomp->file_pos)) < 0) {
> >> pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
> >> decomp->file_pos + decomp->head, event->header.size, event->header.type);
> >> return -EINVAL;
> >> --
> >> 2.19.0
> >>
> >
>

2021-10-08 18:46:18

by Arnaldo Carvalho de Melo

Subject: Re: [PATCH] perf report: Output non-zero offset for decompressed records

Em Mon, Oct 04, 2021 at 09:06:36AM +0200, Jiri Olsa escreveu:
> On Mon, Oct 04, 2021 at 10:00:52AM +0300, Bayduraev, Alexey V wrote:
> > On 04.10.2021 9:42, Jiri Olsa wrote:
> > > On Wed, Sep 29, 2021 at 12:14:45PM +0300, Alexey Bayduraev wrote:
> > >> Print offset of PERF_RECORD_COMPRESSED record instead of zero for
> > >> decompressed records in raw trace dump (-D option of perf-report):

> > >> 0x17cf08 [0x28]: event: 9

> > >> instead of:

> > >> 0 [0x28]: event: 9

> > >> The fix is not critical, because currently file_pos for compressed
> > >> events is used in perf_session__process_event only to show offsets
> > >> in the raw dump.

> > > I don't mind the change just curious, because I see also:
> > >
> > > perf_session__process_event
> > > perf_session__process_user_event
> > > lseek(fd, file_offset, ...

> > > which is not raw dump as the comment suggests

> > Yes, but this "lseek" only works for user events, whereas the
> > PERF_RECORD_COMPRESSED record shouln't contain such events.
> > Currently, the PERF_RECORD_COMPRESSED container can only pack
> > kernel events.

> Acked-by: Jiri Olsa <[email protected]>

Thanks, applied.

- Arnaldo