On Thu, Apr 3, 2025 at 1:24 PM Ian Rogers wrote:
>
> DRM clients expose information through usage stats as documented in
> Documentation/gpu/drm-usage-stats.rst (available online at
> https://docs.kernel.org/gpu/drm-usage-stats.html). Add a tool like
> PMU, similar to the hwmon PM
Add perf_pmus__scan_for_event that only reads sysfs for pmus that
could contain a given event.
Signed-off-by: Ian Rogers
---
tools/perf/util/parse-events.c | 4 ++--
tools/perf/util/pmus.c | 35 ++
tools/perf/util/pmus.h | 1 +
3 files changed
(exit 2) doesn't trigger the
trap cleanup.
v2: Add support to only scan hwmon and drm PMUs if the event or PMU
wildcard can match. Add a test as requested by Namhyung. Add file
comments.
v1:
https://lore.kernel.org/lkml/20250211071727.364389-1-irog...@google.com/
Ian Rogers (4):
perf
ot;perf pmus:
Restructure pmu_read_sysfs to scan fewer PMUs") and later so that only
if full wild carding is being done, the PMU starts with "drm_" or the
event starts with "drm-" will /proc be scanned. That is there should
be little to no cost in this PMU unless DRM events are
The test opens any DRM devices so that the shell has fdinfo files
containing the DRM data. The test then uses perf stat to make sure the
events can be read.
Signed-off-by: Ian Rogers
---
tools/perf/tests/shell/drm_pmu.sh | 78 +++
1 file changed, 78 insertions
can't be a hwmon PMU wildcard, reading hwmon PMUs is
avoided. On my laptop this reduces 'perf stat -e inst_retired.any
true' from performing 253 openat system calls down to 234.
Signed-off-by: Ian Rogers
---
tools/perf/util/parse-events.c | 2
Add perf_pmus__scan_for_event that only reads sysfs for pmus that
could contain a given event.
Signed-off-by: Ian Rogers
---
tools/perf/util/parse-events.c | 4 ++--
tools/perf/util/pmus.c | 35 ++
tools/perf/util/pmus.h | 1 +
3 files changed
ot;perf pmus:
Restructure pmu_read_sysfs to scan fewer PMUs") and later so that only
if full wild carding is being done, the PMU starts with "drm_" or the
event starts with "drm-" will /proc be scanned. That is there should
be little to no cost in this PMU unless DRM events are
The test opens any DRM devices so that the shell has fdinfo files
containing the DRM data. The test then uses perf stat to make sure the
events can be read.
Signed-off-by: Ian Rogers
---
tools/perf/tests/shell/drm_pmu.sh | 77 +++
1 file changed, 77 insertions
PMUs if the event or PMU
wildcard can match. Add a test as requested by Namhyung. Add file
comments.
v1:
https://lore.kernel.org/lkml/20250211071727.364389-1-irog...@google.com/
Ian Rogers (4):
perf parse-events: Avoid scanning PMUs that can't contain events
perf parse-events: Avoid sca
can't be a hwmon PMU wildcard, reading hwmon PMUs is
avoided. On my laptop this reduces 'perf stat -e inst_retired.any
true' from performing 253 openat system calls down to 234.
Signed-off-by: Ian Rogers
---
tools/perf/util/parse-events.c | 2
ot;perf pmus:
Restructure pmu_read_sysfs to scan fewer PMUs") so that only if full
wild carding is being done, that the drm PMUs will be read.
Signed-off-by: Ian Rogers
---
tools/perf/util/Build | 1 +
tools/perf/util/drm_pmu.c | 688 ++
tools/perf/util/
Previously the pid was highlighted through being a tagged paragraph
but not mentioned in the description. Add italics to the path
emphasizing pid and then change the first sentence to include pid in
the definition.
Suggested-by: G. Branden Robinson
---
v5. Switch .IR pid to .I pid
---
man/man5/p
Make the sections about eventfd, epoll, signalfd, inotify, fanotify,
timerfd better separated with a clearer subsection header.
---
man/man5/proc_pid_fdinfo.5 | 6 ++
1 file changed, 6 insertions(+)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5/proc_pid_fdinfo.5
index 2797ae216..c4914f1f
When /proc/pid/fdinfo was part of proc.5 man page the indentation made
sense. As a standalone man page the indentation doesn't need to be so
far over to the right. Remove the initial tagged pragraph, move the
"since Linux 2.6.22" to a new history subsection.
Suggested-by: G. Branden Robinson
---
Add description of DRM fdinfo information based on the Linux kernel's
`Documentation/gpu/drm-usage-stats.rst`:
https://docs.kernel.org/gpu/drm-usage-stats.html
---
v5. Update style a suggested-by G. Branden Robinson
---
man/man5/proc_pid_fdinfo.5 | 107 +
1
On Sat, Nov 2, 2024 at 3:17 AM G. Branden Robinson
wrote:
>
> Hi Ian,
>
> At 2024-11-01T12:11:54-0700, Ian Rogers wrote:
> > diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5/proc_pid_fdinfo.5
> > index 87e6dbe56..935b54b4c 100644
> > --- a/man/man5/proc_p
randen Robinson
Signed-off-by: Ian Rogers
---
v4. Move since to history from Alejandro Colomar's review
comment.
---
man/man5/proc_pid_fdinfo.5 | 51 +++---
1 file changed, 25 insertions(+), 26 deletions(-)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/ma
Previously the pid was highlighted through being a tagged paragraph
but not mentioned in the description. Add italics to the path
emphasizing pid and then change the first sentence to include pid in
the definition.
Suggested-by: G. Branden Robinson
Signed-off-by: Ian Rogers
---
man/man5
Add description of DRM fdinfo information based on the Linux kernel's
`Documentation/gpu/drm-usage-stats.rst`:
https://docs.kernel.org/gpu/drm-usage-stats.html
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 94 ++
1 file changed, 94 inser
Make the sections about eventfd, epoll, signalfd, inotify, fanotify,
timerfd better separated with a clearer subsection header.
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 6 ++
1 file changed, 6 insertions(+)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5
When /proc/pid/fdinfo was part of proc.5 man page the indentation made
sense. As a standalone man page the indentation doesn't need to be so
far over to the right. Remove the initial tagged pragraph.
Suggested-by: G. Branden Robinson
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdi
Make the sections about eventfd, epoll, signalfd, inotify, fanotify,
timerfd better separated with a clearer subsection header.
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 6 ++
1 file changed, 6 insertions(+)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5
Add description of DRM fdinfo information based on the Linux kernel's
`Documentation/gpu/drm-usage-stats.rst`:
https://docs.kernel.org/gpu/drm-usage-stats.html
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 94 ++
1 file changed, 94 inser
Previously the pid was highlighted through being a tagged paragraph
but not mentioned in the description. Add italics to the path
emphasizing pid and then change the first sentence to include pid in
the definition.
Suggested-by: G. Branden Robinson
Signed-off-by: Ian Rogers
---
man/man5
On Fri, Nov 1, 2024 at 6:24 AM Alejandro Colomar wrote:
>
> On Tue, Oct 15, 2024 at 02:17:17PM -0700, Ian Rogers wrote:
> > When /proc/pid/fdinfo was part of proc.5 man page the indentation made
> > sense. As a standalone man page the indentation doesn't need to be so
&g
Make the sections about eventfd, epoll, signalfd, inotify, fanotify,
timerfd better separated with a clearer subsection header.
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 6 ++
1 file changed, 6 insertions(+)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5
igned-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 66 ++
1 file changed, 32 insertions(+), 34 deletions(-)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5/proc_pid_fdinfo.5
index 1e23bbe02..8678caf4a 100644
--- a/man/man5/proc_pid_fdinfo.5
+++ b/man
Add description of DRM fdinfo information based on the Linux kernel's
`Documentation/gpu/drm-usage-stats.rst`:
https://docs.kernel.org/gpu/drm-usage-stats.html
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 94 ++
1 file changed, 94 inser
On Tue, Oct 15, 2024 at 1:32 PM G. Branden Robinson
wrote:
>
> At 2024-10-15T11:38:22-0700, Ian Rogers wrote:
> > When /proc/pid/fdinfo was part of proc.5 man page the indentation made
> > sense. As a standalone man page the indentation doesn't need to be so
&g
Add description of DRM fdinfo information based on the Linux kernel's
`Documentation/gpu/drm-usage-stats.rst`:
https://docs.kernel.org/gpu/drm-usage-stats.html
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 94 ++
1 file changed, 94 inser
Make the sections about eventfd, epoll, signalfd, inotify, fanotify,
timerfd better separated with a clearer subsection header.
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 6 ++
1 file changed, 6 insertions(+)
diff --git a/man/man5/proc_pid_fdinfo.5 b/man/man5
When /proc/pid/fdinfo was part of proc.5 man page the indentation made
sense. As a standalone man page the indentation doesn't need to be so
far over to the right.
Signed-off-by: Ian Rogers
---
man/man5/proc_pid_fdinfo.5 | 50 +++---
1 file changed, 25 inser
On Tue, Oct 15, 2024 at 11:27 AM Ian Rogers wrote:
>
> The PCI slot address for drm-pdev should be a colon not a period. On a
> i915 GPU I see:
> ```
> drm-pdev: :00:02.0
> ```
>
> Signed-off-by: Ian Rogers
Possibly:
Fixes: 055634e4b62f ("drm/i915: Expose
The PCI slot address for drm-pdev should be a colon not a period. On a
i915 GPU I see:
```
drm-pdev: :00:02.0
```
Signed-off-by: Ian Rogers
---
Documentation/gpu/drm-usage-stats.rst | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/Documentation/gpu/drm-usage
On Mon, Jul 22, 2024 at 2:07 PM Lucas De Marchi
wrote:
>
> If a pmu is unregistered while there's an active event, perf will still
> access the pmu via event->pmu, even after the event is destroyed. This
> makes it difficult for drivers like i915 that take a reference on the
> device when the even
36 matches
Mail list logo
