Namhyung Kim [Thu, 8 Feb 2024 18:10:25 +0000 (10:10 -0800)]
perf tools: Remove misleading comments on map functions
When it converts sample IP to or from objdump-capable one, there's a
comment saying that kernel modules have DSO_SPACE__USER. But commit 02213cec64bb ("perf maps: Mark module DSOs with kernel type") changed
it and makes the comment confusing. Let's get rid of it.
Yang Jihong [Tue, 6 Feb 2024 08:32:27 +0000 (08:32 +0000)]
perf sched: Move curr_pid and cpu_last_switched initialization to perf_sched__{lat|map|replay}()
The curr_pid and cpu_last_switched are used only for the
'perf sched replay/latency/map'. Put their initialization in
perf_sched__{lat|map|replay () to reduce unnecessary actions in other
commands.
Simple functional testing:
# perf sched record perf bench sched messaging
# Running 'sched/messaging' benchmark:
# 20 sender and receiver processes per group
# 10 groups == 400 processes run
Total time: 0.209 [sec]
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 16.456 MB perf.data (147907 samples) ]
# perf sched lat
-------------------------------------------------------------------------------------------------------------------------------------------
Task | Runtime ms | Switches | Avg delay ms | Max delay ms | Max delay start | Max delay end |
-------------------------------------------------------------------------------------------------------------------------------------------
sched-messaging:(401) | 2990.699 ms | 38705 | avg: 0.661 ms | max: 67.046 ms | max start: 456532.624830 s | max end: 456532.691876 s
qemu-system-x86:(7) | 179.764 ms | 2191 | avg: 0.152 ms | max: 21.857 ms | max start: 456532.576434 s | max end: 456532.598291 s
sshd:48125 | 0.522 ms | 2 | avg: 0.037 ms | max: 0.046 ms | max start: 456532.514610 s | max end: 456532.514656 s
<SNIP>
ksoftirqd/11:82 | 0.063 ms | 1 | avg: 0.005 ms | max: 0.005 ms | max start: 456532.769366 s | max end: 456532.769371 s
kworker/9:0-mm_:34624 | 0.233 ms | 20 | avg: 0.004 ms | max: 0.007 ms | max start: 456532.690804 s | max end: 456532.690812 s
migration/13:93 | 0.000 ms | 1 | avg: 0.004 ms | max: 0.004 ms | max start: 456532.512669 s | max end: 456532.512674 s
-----------------------------------------------------------------------------------------------------------------
TOTAL: | 3180.750 ms | 41368 |
---------------------------------------------------
Yang Jihong [Tue, 6 Feb 2024 08:32:26 +0000 (08:32 +0000)]
perf sched: Move curr_thread initialization to perf_sched__map()
The curr_thread is used only for the 'perf sched map'. Put initialization
in perf_sched__map() to reduce unnecessary actions in other commands.
Simple functional testing:
# perf sched record perf bench sched messaging
# Running 'sched/messaging' benchmark:
# 20 sender and receiver processes per group
# 10 groups == 400 processes run
Total time: 0.197 [sec]
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 15.526 MB perf.data (140095 samples) ]
Yang Jihong [Tue, 6 Feb 2024 08:32:24 +0000 (08:32 +0000)]
perf sched: Move start_work_mutex and work_done_wait_mutex initialization to perf_sched__replay()
The start_work_mutex and work_done_wait_mutex are used only for the
'perf sched replay'. Put their initialization in perf_sched__replay () to
reduce unnecessary actions in other commands.
Simple functional testing:
# perf sched record perf bench sched messaging
# Running 'sched/messaging' benchmark:
# 20 sender and receiver processes per group
# 10 groups == 400 processes run
Total time: 0.197 [sec]
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 14.952 MB perf.data (134165 samples) ]
Yicong Yang [Wed, 7 Feb 2024 09:12:22 +0000 (17:12 +0800)]
perf test: Skip metric w/o event name on arm64 in stat STD output linter
stat+std_output.sh test fails on my arm64 machine:
[root@localhost shell]# ./stat+std_output.sh
Checking STD output: no args Unknown event name in TopDownL1 # 0.18 retiring
[root@localhost shell]# ./stat+std_output.sh
Checking STD output: no args [Success]
Checking STD output: system wide [Success]
Checking STD output: interval [Success]
Checking STD output: per thread Unknown event name in tmux: server-1114960 # 0.41 frontend_bound
When no args specified `perf stat` will add TopdownL1 metric group
and the output will be like:
[root@localhost shell]# perf stat -- stress-ng --vm 1 --timeout 1
stress-ng: info: [3351733] setting to a 1 second run per stressor
stress-ng: info: [3351733] dispatching hogs: 1 vm
stress-ng: info: [3351733] successful run completed in 1.02s
Performance counter stats for 'stress-ng --vm 1 --timeout 1':
Metrics in group TopDownL1 don't have event name on arm64 but are not
listed in the $skip_metric list which they should be listed. Add them
to the skip list as what does for x86 platforms in [1].
[1] commit 4d60e83dfcee ("perf test: Skip metrics w/o event name in stat STD output linter")
Currently perf does not record module section addresses except for
the .text section. In general that means perf cannot get module section
mappings correct (except for .text) when loading symbols from a kernel
module file. (Note using --kcore does not have this issue)
Improve that situation slightly by identifying executable sections that
use the same mapping as the .text section. That happens when an
executable section comes directly after the .text section, both in memory
and on file, something that can be determined by following the same layout
rules used by the kernel, refer kernel layout_sections(). Note whether
that happens is somewhat arbitrary, so this is not a final solution.
Namhyung Kim [Fri, 12 Jan 2024 23:13:40 +0000 (15:13 -0800)]
perf record: Display data size on pipe mode
Currently pipe mode doesn't set the file size and it results in a
misleading message of 0 data size at the end. Although it might miss
some accounting for pipe header or more, just displaying the data size
would reduce the possible confusion.
Before:
$ perf record -o- perf test -w noploop | perf report -i- -q --percent-limit=1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB - ] <====== (here)
99.58% perf perf [.] noploop
After:
$ perf record -o- perf test -w noploop | perf report -i- -q --percent-limit=1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.229 MB - ]
99.46% perf perf [.] noploop
Kan Liang [Mon, 5 Feb 2024 14:58:19 +0000 (06:58 -0800)]
perf script: Print source line for each jump in brstackinsn
With the srcline option, the perf script only prints a source line at
the beginning of a sample with call/ret from functions, but not for
each jump in brstackinsn. It's useful to print a source line for each
jump in brstackinsn when the end user analyze the full assembler
sequences of branch sequences for the sample.
The srccode option can also be used to locate the source code line.
However, it's printed almost for every line and makes the output less
readable.
Ian Rogers [Tue, 6 Feb 2024 23:59:02 +0000 (15:59 -0800)]
perf kvm powerpc: Fix build
Updates to struct parse_events_error needed to be carried through to
PowerPC specific event parsing.
Fixes: fd7b8e8fb20f ("perf parse-events: Print all errors") Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Ian Rogers <irogers@google.com> Acked-by: Namhyung <namhyung@kernel.org> Cc: James Clark <james.clark@arm.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Kan Liang <kan.liang@linux.intel.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240206235902.2917395-1-irogers@google.com
Ben Gainey [Tue, 23 Jan 2024 10:31:37 +0000 (10:31 +0000)]
tools: perf: Expose sample ID / stream ID to python scripts
perf script exposes the evsel_name to python scripts as part of the data
passed to the sample or tracepoint handler function, and it passes the id and
stream_id to the throttled/unthrottled handler functions. This makes matching
throttle events and samples difficult.
To make this possible, this change exposes the sample id and stream_id values
to the script.
Signed-off-by: Ben Gainey <ben.gainey@arm.com> Reviewed-by: Adrian Hunter <adrian.hunter@intel.com> Cc: will@kernel.org Cc: linux-arm-kernel@lists.infradead.org Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240123103137.1890779-2-ben.gainey@arm.com
Arnaldo Carvalho de Melo [Fri, 2 Feb 2024 14:32:20 +0000 (11:32 -0300)]
perf bpf: Clean up the generated/copied vmlinux.h
When building perf with BPF skels we either copy the minimalistic
tools/perf/util/bpf_skel/vmlinux/vmlinux.h or use bpftool to generate a
vmlinux from BTF, storing the result in $(SKEL_OUT)/vmlinux.h.
We need to remove that when doing a 'make -C tools/perf clean', fix it.
Fixes: b7a2d774c9c5a9a3 ("perf build: Add ability to build with a generated vmlinux.h") Reviewed-by: Ian Rogers <irogers@google.com> Cc: Andrii Nakryiko <andrii@kernel.org> Cc: James Clark <james.clark@arm.com> Cc: Tiezhu Yang <yangtiezhu@loongson.cn> Cc: Yang Jihong <yangjihong1@huawei.com> Cc: bpf@vger.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/Zbz89KK5wHfZ82jv@x1
Ian Rogers [Wed, 31 Jan 2024 20:14:29 +0000 (12:14 -0800)]
perf jevents: Drop or simplify small integer values
Prior to this patch '0' would be dropped as the config values default
to 0. Some json values are hex and the string '0' wouldn't match '0x0'
as zero. Add a more robust is_zero test to drop these event terms.
When encoding numbers as hex, if the number is between 0 and 9
inclusive then don't add a 0x prefix.
Update test expectations for these changes.
On x86 this reduces the event/metric C string by 58,411 bytes.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: Kan Liang <kan.liang@linux.intel.com> Cc: Edward Baker <edward.baker@intel.com> Cc: Perry Taylor <perry.taylor@intel.com> Cc: Weilin Wang <weilin.wang@intel.com> Cc: John Garry <john.g.garry@oracle.com> Cc: Jing Zhang <renyu.zj@linux.alibaba.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Michael Petlan <mpetlan@redhat.com> Cc: Veronika Molnarova <vmolnaro@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240131201429.792138-1-irogers@google.com
Ian Rogers [Wed, 31 Jan 2024 13:49:40 +0000 (05:49 -0800)]
perf parse-events: Print all errors
Prior to this patch the first and the last error encountered during
parsing are printed. To see other errors verbose needs
enabling. Unfortunately this can drop useful errors, in particular on
terms. This patch changes the errors so that instead of the first and
last all errors are recorded and printed, the underlying data
structure is changed to a list.
Before:
```
$ perf stat -e 'slots/edge=2/' true
event syntax error: 'slots/edge=2/'
\___ Bad event or PMU
Unable to find PMU or event on a PMU of 'slots'
Initial error:
event syntax error: 'slots/edge=2/'
\___ Cannot find PMU `slots'. Missing kernel support?
Run 'perf list' for a list of valid events
Usage: perf stat [<options>] [<command>]
-e, --event <event> event selector. use 'perf list' to list available events
```
After:
```
$ perf stat -e 'slots/edge=2/' true
event syntax error: 'slots/edge=2/'
\___ Bad event or PMU
Unable to find PMU or event on a PMU of 'slots'
event syntax error: 'slots/edge=2/'
\___ value too big for format (edge), maximum is 1
event syntax error: 'slots/edge=2/'
\___ Cannot find PMU `slots'. Missing kernel support?
Run 'perf list' for a list of valid events
Usage: perf stat [<options>] [<command>]
-e, --event <event> event selector. use 'perf list' to list available events
```
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@arm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: tchen168@asu.edu Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Michael Petlan <mpetlan@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240131134940.593788-3-irogers@google.com
Ian Rogers [Wed, 31 Jan 2024 13:49:39 +0000 (05:49 -0800)]
perf parse-events: Improve error location of terms cloned from an event
A PMU event/alias will have a set of format terms that replace it when
an event is parsed. The location of the terms is their position when
parsed for the event/alias either from sysfs or json. This location is
of little use when an event fails to parse as the error will be given
in terms of the location in the string of events parsed not the json
or sysfs string. Fix this by making the cloned terms location that of
the event/alias.
If a cloned term from an event/alias is invalid the bad format is hard
to determine from the error string. Add the name of the bad format
into the error string.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@arm.com> Cc: tchen168@asu.edu Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Michael Petlan <mpetlan@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240131134940.593788-2-irogers@google.com
Ian Rogers [Wed, 31 Jan 2024 13:49:38 +0000 (05:49 -0800)]
perf tsc: Add missing newlines to debug statements
It is assumed that debug statements always print a newline, fix two
missing ones.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@arm.com> Cc: tchen168@asu.edu Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Michael Petlan <mpetlan@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240131134940.593788-1-irogers@google.com
Weilin Wang [Tue, 30 Jan 2024 18:09:07 +0000 (10:09 -0800)]
perf test: Simplify metric value validation test final report
The original test report was too complicated to read with information
that not really useful. This new update simplify the report which should
largely improve the readibility.
Signed-off-by: Weilin Wang <weilin.wang@intel.com> Reviewed-by: Ian Rogers <irogers@google.com> Cc: Caleb Biggers <caleb.biggers@intel.com> Cc: Perry Taylor <perry.taylor@intel.com> Cc: Samantha Alt <samantha.alt@intel.com> Cc: Kan Liang <kan.liang@linux.intel.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240130180907.639729-1-weilin.wang@intel.com
Ze Gao [Tue, 23 Jan 2024 07:02:11 +0000 (02:02 -0500)]
perf evsel: Rename get_states() to parse_task_states() and make it public
Since get_states() assumes the existence of libtraceevent, so move
to where it should belong, i.e, util/trace-event-parse.c, and also
rename it to parse_task_states().
Leave evsel_getstate() untouched as it fits well in the evsel
category.
Also make some necessary tweaks for python support, and get it
verified with: perf test python.
Ian Rogers [Wed, 24 Jan 2024 23:42:00 +0000 (15:42 -0800)]
perf pmu: Treat the msr pmu as software
The msr PMU is a software one, meaning msr events may be grouped
with events in a hardware context. As the msr PMU isn't marked as a
software PMU by perf_pmu__is_software, groups with the msr PMU in
are broken and the msr events placed in a different group. This
may lead to multiplexing errors where a hardware event isn't
counted while the msr event, such as tsc, is. Fix all of this by
marking the msr PMU as software, which agrees with the driver.
Before:
```
$ perf stat -e '{slots,tsc}' -a true
WARNING: events were regrouped to match PMUs
Fixes: 251aa040244a ("perf parse-events: Wildcard most "numeric" events") Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: Kan Liang <kan.liang@linux.intel.com> Cc: James Clark <james.clark@arm.com> Cc: Caleb Biggers <caleb.biggers@intel.com> Cc: Edward Baker <edward.baker@intel.com> Cc: Perry Taylor <perry.taylor@intel.com> Cc: Samantha Alt <samantha.alt@intel.com> Cc: Weilin Wang <weilin.wang@intel.com> Link: https://lore.kernel.org/r/20240124234200.1510417-1-irogers@google.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
James Clark [Tue, 23 Jan 2024 16:39:02 +0000 (16:39 +0000)]
perf test: Skip test_arm_callgraph_fp.sh if unwinding isn't built in
Even though this is a frame pointer unwind test, it's testing that a
frame pointer stack can be augmented correctly with a partial
Dwarf unwind. So add a feature check so that this test skips instead of
fails if Dwarf unwinding isn't present.
Signed-off-by: James Clark <james.clark@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Spoorthy S <spoorts2@in.ibm.com> Cc: Kajol Jain <kjain@linux.ibm.com> Link: https://lore.kernel.org/r/20240123163903.350306-3-james.clark@arm.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
James Clark [Tue, 23 Jan 2024 16:39:01 +0000 (16:39 +0000)]
perf version: Display availability of HAVE_DWARF_UNWIND_SUPPORT
Even though unwinding depends on either HAVE_DWARF_SUPPORT or
HAVE_LIBUNWIND, scripts testing unwinding can't just look for the
existence of either of those flags. This is because
NO_LIBDW_DWARF_UNWIND=1 can disable unwinding with libdw, but libdw will
still be linked leaving HAVE_DWARF_SUPPORT turned on. Presumably because
it is used for things other than unwinding, so I don't think this needs
to be fixed.
HAVE_DWARF_UNWIND_SUPPORT already takes the combination of all those
things into account, and is used to gate the built in tests like "Test
dwarf unwind", so add it to the feature list output so that it can be
used by the script tests too.
Signed-off-by: James Clark <james.clark@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Spoorthy S <spoorts2@in.ibm.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Link: https://lore.kernel.org/r/20240123163903.350306-2-james.clark@arm.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
James Clark [Wed, 24 Jan 2024 09:43:57 +0000 (09:43 +0000)]
perf evlist: Fix evlist__new_default() for > 1 core PMU
The 'Session topology' test currently fails with this message when
evlist__new_default() opens more than one event:
32: Session topology :
--- start ---
templ file: /tmp/perf-test-vv5YzZ
Using CPUID 0x00000000410fd070
Opening: unknown-hardware:HG
------------------------------------------------------------
perf_event_attr:
type 0 (PERF_TYPE_HARDWARE)
config 0xb00000000
disabled 1
------------------------------------------------------------
sys_perf_event_open: pid 0 cpu -1 group_fd -1 flags 0x8 = 4
Opening: unknown-hardware:HG
------------------------------------------------------------
perf_event_attr:
type 0 (PERF_TYPE_HARDWARE)
config 0xa00000000
disabled 1
------------------------------------------------------------
sys_perf_event_open: pid 0 cpu -1 group_fd -1 flags 0x8 = 5
non matching sample_type
FAILED tests/topology.c:73 can't get session
---- end ----
Session topology: FAILED!
This is because when re-opening the file and parsing the header, Perf
expects that any file that has more than one event has the sample ID
flag set. Perf record already sets the flag in a similar way when there
is more than one event, so add the same logic to evlist__new_default().
evlist__new_default() is only currently used in tests, so I don't
expect this change to have any other side effects. The other tests that
use it don't save and re-open the file so don't hit this issue.
The session topology test has been failing on Arm big.LITTLE platforms
since commit 251aa040244a ("perf parse-events: Wildcard most
"numeric" events") when evlist__new_default() started opening multiple
events for 'cycles'.
Fixes: 251aa040244a ("perf parse-events: Wildcard most "numeric" events") Closes: https://lore.kernel.org/lkml/CAP-5=fWVQ-7ijjK3-w1q+k2WYVNHbAcejb-xY0ptbjRw476VKA@mail.gmail.com/ Tested-by: Ian Rogers <irogers@google.com> Reviewed-by: Ian Rogers <irogers@google.com> Tested-by: Kan Liang <kan.liang@linux.intel.com> Signed-off-by: James Clark <james.clark@arm.com> Cc: Changbin Du <changbin.du@huawei.com> Cc: Yang Jihong <yangjihong1@huawei.com> Link: https://lore.kernel.org/r/20240124094358.489372-1-james.clark@arm.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Kan Liang [Tue, 23 Jan 2024 18:50:35 +0000 (10:50 -0800)]
perf mem: Clean up perf_mem_events__record_args()
The current code iterates all memory PMUs. It doesn't matter if the
system has only one memory PMU or multiple PMUs. The check of
perf_pmus__num_mem_pmus() is not required anymore.
The rec_tmp is not used in c2c and mem. Removing them as well.
Suggested-by: Leo Yan <leo.yan@linaro.org> Tested-by: Leo Yan <leo.yan@linaro.org> Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Reviewed-by: Ian Rogers <irogers@google.com> Cc: ravi.bangoria@amd.com Cc: james.clark@arm.com Cc: will@kernel.org Cc: mike.leach@linaro.org Cc: renyu.zj@linux.alibaba.com Cc: yuhaixin.yhx@linux.alibaba.com Cc: tmricht@linux.ibm.com Cc: atrajeev@linux.vnet.ibm.com Cc: linux-arm-kernel@lists.infradead.org Cc: john.g.garry@oracle.com Link: https://lore.kernel.org/r/20240123185036.3461837-7-kan.liang@linux.intel.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Kan Liang [Tue, 23 Jan 2024 18:50:33 +0000 (10:50 -0800)]
perf mem: Clean up perf_mem_event__supported()
For some ARCHs, e.g., ARM and AMD, to get the availability of the
mem-events, perf checks the existence of a specific PMU. For the other
ARCHs, e.g., Intel and Power, perf has to check the existence of some
specific events.
The current perf only iterates the mem-events-supported PMUs. It's not
required to check the existence of a specific PMU anymore.
Rename sysfs_name to event_name, which stores the specific mem-events.
Perf only needs to check those events for the availability of the
mem-events.
Rename perf_mem_event__supported to perf_pmu__mem_events_supported.
Reviewed-by: Ian Rogers <irogers@google.com> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com> Tested-by: Leo Yan <leo.yan@linaro.org> Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: james.clark@arm.com Cc: will@kernel.org Cc: mike.leach@linaro.org Cc: renyu.zj@linux.alibaba.com Cc: yuhaixin.yhx@linux.alibaba.com Cc: tmricht@linux.ibm.com Cc: atrajeev@linux.vnet.ibm.com Cc: linux-arm-kernel@lists.infradead.org Cc: john.g.garry@oracle.com Link: https://lore.kernel.org/r/20240123185036.3461837-5-kan.liang@linux.intel.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Kan Liang [Tue, 23 Jan 2024 18:50:31 +0000 (10:50 -0800)]
perf mem: Clean up perf_mem_events__ptr()
The mem_events can be retrieved from the struct perf_pmu now. An ARCH
specific perf_mem_events__ptr() is not required anymore. Remove all of
them.
The Intel hybrid has multiple mem-events-supported PMUs. But they share
the same mem_events. Other ARCHs only support one mem-events-supported
PMU. In the configuration, it's good enough to only configure the
mem_events for one PMU. Add perf_mem_events_find_pmu() which returns the
first mem-events-supported PMU.
In the perf_mem_events__init(), the perf_pmus__scan() is not required
anymore. It avoids checking the sysfs for every PMU on the system.
Make the perf_mem_events__record_args() more generic. Remove the
perf_mem_events__print_unsupport_hybrid().
Since pmu is added as a new parameter, rename perf_mem_events__ptr() to
perf_pmu__mem_events_ptr(). Several other functions also do a similar
rename.
Kan Liang [Tue, 23 Jan 2024 18:50:30 +0000 (10:50 -0800)]
perf mem: Add mem_events into the supported perf_pmu
With the mem_events, perf doesn't need to read sysfs for each PMU to
find the mem-events-supported PMU. The patch also makes it possible to
clean up the related __weak functions later.
The patch is only to add the mem_events into the perf_pmu for all ARCHs.
It will be used in the later cleanup patches.
Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: Kajol Jain <kjain@linux.ibm.com> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com> Tested-by: Leo Yan <leo.yan@linaro.org> Tested-by: Kajol Jain <kjain@linux.ibm.com> Suggested-by: Leo Yan <leo.yan@linaro.org> Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: will@kernel.org Cc: mike.leach@linaro.org Cc: renyu.zj@linux.alibaba.com Cc: yuhaixin.yhx@linux.alibaba.com Cc: tmricht@linux.ibm.com Cc: atrajeev@linux.vnet.ibm.com Cc: linux-arm-kernel@lists.infradead.org Cc: john.g.garry@oracle.com Link: https://lore.kernel.org/r/20240123185036.3461837-2-kan.liang@linux.intel.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ze Gao [Tue, 23 Jan 2024 02:24:25 +0000 (21:24 -0500)]
perf sched: Commit to evsel__taskstate() to parse task state info
Now that we have evsel__taskstate() which no longer relies on the
hardcoded task state string and has good backward compatibility,
we have a good reason to use it.
Note TASK_STATE_TO_CHAR_STR and task bitmasks are useless now so
we remove them for good. And now we pass the state info back and
forth in a symbolic char which explains itself well instead.
Ze Gao [Mon, 22 Jan 2024 07:08:57 +0000 (02:08 -0500)]
perf util: Add helpers to parse task state string from libtraceevent
Perf uses a hard coded string "RSDTtXZPI" to index the sched_switch
prev_state field raw bitmask value. This works well except for when
the kernel changes this string, in which case this will break again.
Instead we add a new way to parse task state string from tracepoint
print format already recorded by perf, which eliminates the further
dependencies with this hardcode and unmaintainable macro, and this
is exactly what libtraceevent[1] does for now.
So we borrow the print flags parsing logic from libtraceevent[1].
And in get_states(), we walk the print arguments until the
__print_flags() for the target state field is found, and use that to
build the states string for future parsing.
Ze Gao [Mon, 22 Jan 2024 07:08:56 +0000 (02:08 -0500)]
perf sched: Sync state char array with the kernel
Update state char array to match the latest kernel definitions and
remove unused state mapping macros.
Note this is the preparing patch for get rid of the way to parse
process state from raw bitmask value. Instead we are going to
parse it from the recorded tracepoint print format, and this change
marks why we're doing it.
Yang Jihong [Fri, 19 Jan 2024 04:03:03 +0000 (04:03 +0000)]
perf record: Check conflict between '--timestamp-filename' option and pipe mode before recording
In pipe mode, no need to switch perf data output, therefore,
'--timestamp-filename' option should not take effect.
Check the conflict before recording and output WARNING.
In this case, the check pipe mode in perf_data__switch() can be removed.
Before:
# perf record --timestamp-filename -o- perf test -w noploop | perf report -i- --percent-limit=1
# To display the perf.data header info, please use --header/--header-only options.
#
[ perf record: Woken up 1 times to write data ]
[ perf record: Dump -.2024011812110182 ]
#
# Total Lost Samples: 0
#
# Samples: 4K of event 'cycles:P'
# Event count (approx.): 2176784359
#
# Overhead Command Shared Object Symbol
# ........ ....... .................... ......................................
#
97.83% perf perf [.] noploop
#
# (Tip: Print event counts in CSV format with: perf stat -x,)
#
After:
# perf record --timestamp-filename -o- perf test -w noploop | perf report -i- --percent-limit=1
WARNING: --timestamp-filename option is not available in pipe mode.
# To display the perf.data header info, please use --header/--header-only options.
#
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB - ]
#
# Total Lost Samples: 0
#
# Samples: 4K of event 'cycles:P'
# Event count (approx.): 2185575421
#
# Overhead Command Shared Object Symbol
# ........ ....... ..................... .............................................
#
97.75% perf perf [.] noploop
Yang Jihong [Fri, 19 Jan 2024 04:03:02 +0000 (04:03 +0000)]
perf record: Fix possible incorrect free in record__switch_output()
perf_data__switch() may not assign a legal value to 'new_filename'.
In this case, 'new_filename' uses the on-stack value, which may cause a
incorrect free and unexpected result.
Fixes: 03724b2e9c45 ("perf record: Allow to limit number of reported perf.data files") Signed-off-by: Yang Jihong <yangjihong1@huawei.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240119040304.3708522-2-yangjihong1@huawei.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung Kim [Wed, 17 Jan 2024 06:26:57 +0000 (22:26 -0800)]
perf dwarf-aux: Check allowed DWARF Ops
The DWARF location expression can be fairly complex and it'd be hard
to match it with the condition correctly. So let's be conservative
and only allow simple expressions. For now it just checks the first
operation in the list. The following operations looks ok:
To refuse complex (and unsupported) location expressions, add
check_allowed_ops() to compare the rest of the list. It seems earlier
result contained those unsupported expressions. For example, I found
some local struct variable is placed like below.
Namhyung Kim [Wed, 17 Jan 2024 06:26:56 +0000 (22:26 -0800)]
perf annotate-data: Support stack variables
Local variables are allocated in the stack and the location list
should look like base register(s) and an offset. Extend the
die_find_variable_by_reg() to handle the following expressions
* DW_OP_breg{0..31}
* DW_OP_bregx
* DW_OP_fbreg
Ususally DWARF subprogram entries have frame base information and
use it to locate stack variable like below:
I found some differences on saving the frame base between gcc and clang.
The gcc uses the CFA to get the base so it needs to check the current
frame's CFI info. In this case, stack offset needs to be adjusted from
the start of the CFA.
Also it needs to update the offset after finding the type like global
variables since the offset was from the frame base. Factor out
match_var_offset() to check global and local variables in the same way.
The type stats are improved too:
Annotate data type stats:
total 294, ok 160 (54.4%), bad 134 (45.6%)
-----------------------------------------------------------
30 : no_sym
32 : no_mem_ops
51 : no_var
14 : no_typeinfo
7 : bad_offset
Namhyung Kim [Wed, 17 Jan 2024 06:26:55 +0000 (22:26 -0800)]
perf dwarf-aux: Add die_get_cfa()
The die_get_cfa() is to get frame base register and offset at the given
instruction address (pc). This info will be used to locate stack
variables which have location expression using DW_OP_fbreg.
Namhyung Kim [Wed, 17 Jan 2024 06:26:54 +0000 (22:26 -0800)]
perf annotate-data: Support global variables
Global variables are accessed using PC-relative address so it needs to
be handled separately. The PC-rel addressing is detected by using
DWARF_REG_PC. On x86, %rip register would be used.
The address can be calculated using the ip and offset in the
instruction. But it should start from the next instruction so add
calculate_pcrel_addr() to do it properly.
But global variables defined in a different file would only have a
declaration which doesn't include a location list. So it first tries
to get the type info using the address, and then looks up the variable
declarations using name. The name of global variables should be get
from the symbol table. The declaration would have the type info.
So extend find_var_type() to take both address and name for global
variables.
The stat is now looks like:
Annotate data type stats:
total 294, ok 153 (52.0%), bad 141 (48.0%)
-----------------------------------------------------------
30 : no_sym
32 : no_mem_ops
61 : no_var
10 : no_typeinfo
8 : bad_offset
Namhyung Kim [Wed, 17 Jan 2024 06:26:53 +0000 (22:26 -0800)]
perf annotate-data: Handle PC-relative addressing
Extend find_data_type_die() to find data type from PC-relative address
using die_find_variable_by_addr(). Users need to pass the address for
the (global) variable.
The offset for the variable should be updated after finding the type
because the offset in the instruction is just to calcuate the address
for the variable. So it changed to pass a pointer to offset and renamed
it to 'poffset'.
First it searches variables in the CU DIE as it's likely that the global
variables are defined in the file level. And then it iterates the scope
DIEs to find a local (static) variable.
Namhyung Kim [Wed, 17 Jan 2024 06:26:52 +0000 (22:26 -0800)]
perf annotate-data: Add stack operation pseudo type
A typical function prologue and epilogue include multiple stack
operations to save and restore the current value of registers.
On x86, it looks like below:
push r15
push r14
push r13
push r12
...
pop r12
pop r13
pop r14
pop r15
ret
As these all touches the stack memory region, chances are high that they
appear in a memory profile data. But these are not used for any real
purpose yet so it'd return no types.
One of my profile type shows that non neglible portion of data came from
the stack operations. It also seems GCC generates more stack operations
than clang.
Annotate Instruction stats
total 264, ok 169 (64.0%), bad 95 (36.0%)
Name : Good Bad
-----------------------------------------------------------
movq : 49 27
movl : 24 9
popq : 0 19 <-- here
cmpl : 17 2
addq : 14 1
cmpq : 12 2
cmpxchgl : 3 7
Instead of dealing them as unknown, let's create a seperate pseudo type
to represent those stack operations separately.
Namhyung Kim [Wed, 17 Jan 2024 06:26:51 +0000 (22:26 -0800)]
perf annotate-data: Handle array style accesses
On x86, instructions for array access often looks like below.
mov 0x1234(%rax,%rbx,8), %rcx
Usually the first register holds the type information and the second one
has the index. And the current code only looks up a variable for the
first register. But it's possible to be in the other way around so it
needs to check the second register if the first one failed.
The stat changed like this.
Annotate data type stats:
total 294, ok 148 (50.3%), bad 146 (49.7%)
-----------------------------------------------------------
30 : no_sym
32 : no_mem_ops
66 : no_var
10 : no_typeinfo
8 : bad_offset
Namhyung Kim [Wed, 17 Jan 2024 06:26:50 +0000 (22:26 -0800)]
perf annotate-data: Handle macro fusion on x86
When a sample was come from a conditional branch without a memory
operand, it could be due to a macro fusion with a previous instruction.
So it needs to check the memory operand in the previous one.
This improves the stat like below:
Annotate data type stats:
total 294, ok 147 (50.0%), bad 147 (50.0%)
-----------------------------------------------------------
30 : no_sym
32 : no_mem_ops
71 : no_var
6 : no_typeinfo
8 : bad_offset
Yang Jihong [Fri, 12 Jan 2024 03:40:19 +0000 (03:40 +0000)]
perf build: Check whether pkg-config is installed when libtraceevent is linked
If pkg-config is not installed when libtraceevent is linked, the build fails.
The error information is as follows:
$ make
<SNIP>
In file included from /home/yjh/projects_linux/perf-tool-next/linux/tools/perf/util/evsel.c:43:
/home/yjh/projects_linux/perf-tool-next/linux/tools/perf/util/trace-event.h:149:62: error: operator '&&' has no right operand
149 | #if defined(LIBTRACEEVENT_VERSION) && LIBTRACEEVENT_VERSION >= MAKE_LIBTRACEEVENT_VERSION(1, 5, 0)
| ^~
error: command '/usr/bin/gcc' failed with exit code 1
cp: cannot stat 'python_ext_build/lib/perf*.so': No such file or directory
make[2]: *** [Makefile.perf:668: python/perf.cpython-310-x86_64-linux-gnu.so] Error 1
make[2]: *** Waiting for unfinished jobs....
Because pkg-config is not installed, fail to get libtraceevent version in
Makefile.config file. As a result, LIBTRACEEVENT_VERSION is empty.
However, the preceding error information is not user-friendly.
Identify errors in advance by checking that pkg-config is installed at
compile time.
The build results of various scenarios are as follows:
1. build successful when libtraceevent is not linked and pkg-config is not installed
$ pkg-config --version
-bash: /usr/bin/pkg-config: No such file or directory
$ make clean >/dev/null
$ make NO_LIBTRACEEVENT=1 >/dev/null
Makefile.config:1133: No alternatives command found, you need to set JDIR= to point to the root of your Java directory
PERF_VERSION = 6.7.rc6.gd988c9f511af
$ echo $?
0
2. dummy pkg-config is missing when libtraceevent is linked
$ pkg-config --version
-bash: /usr/bin/pkg-config: No such file or directory
$ make clean >/dev/null
$ make >/dev/null
Makefile.config:221: *** Error: pkg-config needed by libtraceevent is missing on this system, please install it. Stop.
make[1]: *** [Makefile.perf:251: sub-make] Error 2
make: *** [Makefile:70: all] Error 2
$ echo $?
2
3. build successful when libtraceevent is linked and pkg-config is installed
$ pkg-config --version
0.29.2
$ make clean >/dev/null
$ make >/dev/null
Makefile.config:1133: No alternatives command found, you need to set JDIR= to point to the root of your Java directory
PERF_VERSION = 6.7.rc6.gd988c9f511af
$ echo $?
0
Thomas Richter [Mon, 8 Jan 2024 08:40:09 +0000 (09:40 +0100)]
perf test: raise limit to 20 percent for perf_stat_--bpf-counters_test
This test case often fails on s390 (about 2 out of 10) because the
10% percent limit on the difference between --bpf-counters event counting
and s390 hardware counting is more than 10% in all failure cases.
Raise the limit to 20% on s390 and the test case succeeds.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: gor@linux.ibm.com Cc: hca@linux.ibm.com Cc: sumanthk@linux.ibm.com Cc: svens@linux.ibm.com Link: https://lore.kernel.org/r/20240108084009.3959211-1-tmricht@linux.ibm.com Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Linus Torvalds [Sun, 21 Jan 2024 22:01:12 +0000 (14:01 -0800)]
Merge tag 'bcachefs-2024-01-21' of https://evilpiepirate.org/git/bcachefs
Pull more bcachefs updates from Kent Overstreet:
"Some fixes, Some refactoring, some minor features:
- Assorted prep work for disk space accounting rewrite
- BTREE_TRIGGER_ATOMIC: after combining our trigger callbacks, this
makes our trigger context more explicit
- A few fixes to avoid excessive transaction restarts on
multithreaded workloads: fstests (in addition to ktest tests) are
now checking slowpath counters, and that's shaking out a few bugs
- Assorted tracepoint improvements
- Starting to break up bcachefs_format.h and move on disk types so
they're with the code they belong to; this will make room to start
documenting the on disk format better.
- A few minor fixes"
* tag 'bcachefs-2024-01-21' of https://evilpiepirate.org/git/bcachefs: (46 commits)
bcachefs: Improve inode_to_text()
bcachefs: logged_ops_format.h
bcachefs: reflink_format.h
bcachefs; extents_format.h
bcachefs: ec_format.h
bcachefs: subvolume_format.h
bcachefs: snapshot_format.h
bcachefs: alloc_background_format.h
bcachefs: xattr_format.h
bcachefs: dirent_format.h
bcachefs: inode_format.h
bcachefs; quota_format.h
bcachefs: sb-counters_format.h
bcachefs: counters.c -> sb-counters.c
bcachefs: comment bch_subvolume
bcachefs: bch_snapshot::btime
bcachefs: add missing __GFP_NOWARN
bcachefs: opts->compression can now also be applied in the background
bcachefs: Prep work for variable size btree node buffers
bcachefs: grab s_umount only if snapshotting
...
Linus Torvalds [Sun, 21 Jan 2024 19:14:40 +0000 (11:14 -0800)]
Merge tag 'timers-core-2024-01-21' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull timer updates from Thomas Gleixner:
"Updates for time and clocksources:
- A fix for the idle and iowait time accounting vs CPU hotplug.
The time is reset on CPU hotplug which makes the accumulated
systemwide time jump backwards.
- Assorted fixes and improvements for clocksource/event drivers"
* tag 'timers-core-2024-01-21' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
tick-sched: Fix idle and iowait sleeptime accounting vs CPU hotplug
clocksource/drivers/ep93xx: Fix error handling during probe
clocksource/drivers/cadence-ttc: Fix some kernel-doc warnings
clocksource/drivers/timer-ti-dm: Fix make W=n kerneldoc warnings
clocksource/timer-riscv: Add riscv_clock_shutdown callback
dt-bindings: timer: Add StarFive JH8100 clint
dt-bindings: timer: thead,c900-aclint-mtimer: separate mtime and mtimecmp regs
Kent Overstreet [Tue, 16 Jan 2024 21:20:21 +0000 (16:20 -0500)]
bcachefs: opts->compression can now also be applied in the background
The "apply this compression method in the background" paths now use the
compression option if background_compression is not set; this means that
setting or changing the compression option will cause existing data to
be compressed accordingly in the background.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Tue, 16 Jan 2024 18:29:59 +0000 (13:29 -0500)]
bcachefs: Prep work for variable size btree node buffers
bcachefs btree nodes are big - typically 256k - and btree roots are
pinned in memory. As we're now up to 18 btrees, we now have significant
memory overhead in mostly empty btree roots.
And in the future we're going to start enforcing that certain btree node
boundaries exist, to solve lock contention issues - analagous to XFS's
AGIs.
Thus, we need to start allocating smaller btree node buffers when we
can. This patch changes code that refers to the filesystem constant
c->opts.btree_node_size to refer to the btree node buffer size -
btree_buf_bytes() - where appropriate.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
In __bch2_ioctl_subvolume_create(), we grab s_umount unconditionally
and unlock it at the end of the function. There is a comment
"why do we need this lock?" about the lock coming from
commit 42d237320e98 ("bcachefs: Snapshot creation, deletion")
The reason is that __bch2_ioctl_subvolume_create() calls
sync_inodes_sb() which enforce locked s_umount to writeback all dirty
nodes before doing snapshot works.
Fix it by read locking s_umount for snapshotting only and unlocking
s_umount after sync_inodes_sb().
Signed-off-by: Su Yue <glass.su@suse.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Colin Ian King [Tue, 16 Jan 2024 11:07:23 +0000 (11:07 +0000)]
bcachefs: remove redundant variable tmp
The variable tmp is being assigned a value but it isn't being
read afterwards. The assignment is redundant and so tmp can be
removed.
Cleans up clang scan build warning:
warning: Although the value stored to 'ret' is used in the enclosing
expression, the value is never actually read from 'ret'
[deadcode.DeadStores]
Signed-off-by: Colin Ian King <colin.i.king@gmail.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Tue, 16 Jan 2024 01:37:23 +0000 (20:37 -0500)]
bcachefs: Fix excess transaction restarts in __bchfs_fallocate()
drop_locks_do() should not be used in a fastpath without first trying
the do in nonblocking mode - the unlock and relock will cause excessive
transaction restarts and potentially livelocking with other threads that
are contending for the same locks.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 15 Jan 2024 22:59:51 +0000 (17:59 -0500)]
bcachefs: Better journal tracepoints
Factor out bch2_journal_bufs_to_text(), and use it in the
journal_entry_full() tracepoint; when we can't get a journal reservation
we need to know the outstanding journal entry sizes to know if the
problem is due to excessive flushing.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 15 Jan 2024 22:56:22 +0000 (17:56 -0500)]
bcachefs: Avoid flushing the journal in the discard path
When issuing discards, we may need to flush the journal if there's too
many buckets that can't be discarded until a journal flush.
But the heuristic was bad; we should be comparing the number of buckets
that need to flushes against the number of free buckets, not the number
of buckets we saw.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 15 Jan 2024 19:15:26 +0000 (14:15 -0500)]
bcachefs: bch2_kthread_io_clock_wait() no longer sleeps until full amount
Drop t he loop in bch2_kthread_io_clock_wait(): this allows the code
that uses it to be woken up for other reasons, and fixes a bug where
rebalance wouldn't wake up when a scan was requested.
This raises the possibility of spurious wakeups, but callers should
always be able to handle that reasonably well.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Thu, 11 Jan 2024 04:47:04 +0000 (23:47 -0500)]
bcachefs: Reduce would_deadlock restarts
We don't have to take locks in any particular ordering - we'll make
forward progress just fine - but if we try to stick to an ordering, it
can help to avoid excessive would_deadlock transaction restarts.
This tweaks the reflink path to take extents btree locks in the right
order.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
projects / users / dwmw2 / linux.git / log