- 26 Jul, 2022 1 commit
-
-
Yang Jihong authored
The 'perf kwork' tool is used to trace time properties of kernel work (such as irq, softirq, and workqueue), including runtime, latency, and timehist, using the infrastructure in the perf tools to allow tracing extra targets. This is the first commit to reuse the 'perf record' framework code to implement a simple record function, kwork is not supported currently. Test cases: # perf usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS] The most commonly used perf commands are: <SNIP> iostat Show I/O performance metrics kallsyms Searches running kernel for symbols kmem Tool to trace/measure kernel memory properties kvm Tool to trace/measure kvm guest os kwork Tool to trace/measure kernel work properties (latencies) list List all symbolic event types lock Analyze lock events mem Profile memory accesses record Run a command and record its profile into perf.data <SNIP> See 'perf help COMMAND' for more information on a specific command. # perf kwork Usage: perf kwork [<options>] {record} -D, --dump-raw-trace dump raw trace in ASCII -f, --force don't complain, do it -k, --kwork <kwork> list of kwork to profile -v, --verbose be more verbose (show symbol address, etc) # perf kwork record -- sleep 1 [ perf record: Woken up 0 times to write data ] [ perf record: Captured and wrote 1.787 MB perf.data ] Signed-off-by:Yang Jihong <yangjihong1@huawei.com> Tested-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20220709015033.38326-2-yangjihong1@huawei.com [ Add {} for multiline if blocks ] Signed-off-by:
-
- 25 Jul, 2022 9 commits
-
-
Arnaldo Carvalho de Melo authored
Noticed when processing 'perf kwork' that includes util/data.h without, by luck, having included unistd.h indirectly to get the pid_t typedef. Signed-off-by: -
Namhyung Kim authored
Like perf lock report, it can report lock contention stat of each task. $ perf lock contention -t contended total wait max wait avg wait pid comm 5 945.20 us 902.08 us 189.04 us 316167 EventManager_De 33 98.17 us 6.78 us 2.97 us 766063 kworker/0:1-get 7 92.47 us 61.26 us 13.21 us 316170 EventManager_De 14 76.31 us 12.87 us 5.45 us 12949 timedcall 24 76.15 us 12.27 us 3.17 us 767992 sched-pipe 15 75.62 us 11.93 us 5.04 us 15127 switchto-defaul 24 71.84 us 5.59 us 2.99 us 629168 kworker/u513:2- 17 67.41 us 7.94 us 3.96 us 13504 coroner- 1 59.56 us 59.56 us 59.56 us 316165 EventManager_De 14 56.21 us 6.89 us 4.01 us 0 swapper Signed-off-by:Namhyung Kim <namhyung@kernel.org> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Waiman Long <longman@redhat.com> Cc: Will Deacon <will@kernel.org> Link: https://lore.kernel.org/r/20220725183124.368304-6-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
Like perf lock report, add -k/--key and -F/--field options to control output formatting and sorting. Note that it has slightly different default options as some fields are not available and to optimize the screen space. Signed-off-by:
-
Namhyung Kim authored
The 'perf lock contention' processes the lock contention events and displays the result like perf lock report. Right now, there's not much difference between the two but the lock contention specific features will come soon. $ perf lock contention contended total wait max wait avg wait type caller 238 1.41 ms 29.20 us 5.94 us spinlock update_blocked_averages+0x4c 1 902.08 us 902.08 us 902.08 us rwsem:R do_user_addr_fault+0x1dd 81 330.30 us 17.24 us 4.08 us spinlock _nohz_idle_balance+0x172 2 89.54 us 61.26 us 44.77 us spinlock do_anonymous_page+0x16d 24 78.36 us 12.27 us 3.27 us mutex pipe_read+0x56 2 71.58 us 59.56 us 35.79 us spinlock __handle_mm_fault+0x6aa 6 25.68 us 6.89 us 4.28 us spinlock do_idle+0x28d 1 18.46 us 18.46 us 18.46 us rtmutex exec_fw_cmd+0x21b 3 15.25 us 6.26 us 5.08 us spinlock tick_do_update_jiffies64+0x2c Signed-off-by: -
Namhyung Kim authored
Introduce the aggr_mode variable to prepare a later code change. The default is LOCK_AGGR_ADDR which aggregates the result for the lock instances. When -t/--threads option is given, it'd be set to LOCK_AGGR_TASK. The LOCK_AGGR_CALLER is for the contention analysis and it'd aggregate the stat by comparing the callstacks. Signed-off-by:
-
Namhyung Kim authored
For lock contention tracepoint analysis, it needs to keep the flags. As nr_readlock and nr_trylock fields are not used for it, let's make it a union. Signed-off-by:
-
Ian Rogers authored
The value should be non-zero on Intel while zero on everything else. Reviewed-by:
Kan Liang <kan.liang@linux.intel.com> Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexandre Torgue <alexandre.torgue@foss.st.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Caleb Biggers <caleb.biggers@intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.garry@huawei.com> Cc: Kshipra Bopardikar <kshipra.bopardikar@intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Maxime Coquelin <mcoquelin.stm32@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Perry Taylor <perry.taylor@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com> Link: https://lore.kernel.org/r/20220718164312.3994191-4-irogers@google.comSigned-off-by:
-
Ian Rogers authored
The CPUID method of arch_get_tsc_freq fails for older Intel processors, such as Skylake. Compute using /proc/cpuinfo. Reviewed-by:
-
Kan Liang authored
The TSC frequency information is required for the event metrics with the literal, system_tsc_freq. For the newer Intel platform, the TSC frequency information can be retrieved from the CPUID leaf 0x15. If the TSC frequency information isn't present the /proc/cpuinfo approach is used. Refactor cpuid() for this use. Note, the previous stack pushing/popping approach was broken on x86-64 that has stack red zones that would be clobbered. Committer testing: Before: $ perf record sleep 0.0001 [ perf record: Woken up 1 times to write data ] $ perf report --header-only |& grep cpuid # cpuid : AuthenticAMD,25,33,0 $ After the patch: $ perf record sleep 0.0001 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.002 MB perf.data (8 samples) ] $ perf report --header-only |& grep cpuid # cpuid : AuthenticAMD,25,33,0 $ Signed-off-by:
-
- 21 Jul, 2022 2 commits
-
-
Namhyung Kim authored
It should be lock_text_end instead of _start. Fixes: 0d2997f7 ("perf lock: Look up callchain for the contended locks") Signed-off-by:
-
Arnaldo Carvalho de Melo authored
Synch with upstream, to pick up recent developments. Signed-off-by:
-
- 20 Jul, 2022 28 commits
-
-
Arnaldo Carvalho de Melo authored
Noticed after switching to python3 by default on some older fedora releases: 35 38.20 fedora:27 : FAIL clang version 5.0.2 (tags/RELEASE_502/final) clang-5.0: error: argument unused during compilation: '-specs=/usr/lib/rpm/redhat/redhat-hardened-cc1' [-Werror,-Wunused-command-line-argument] clang-5.0: error: argument unused during compilation: '-specs=/usr/lib/rpm/redhat/redhat-hardened-cc1' [-Werror,-Wunused-command-line-argument] error: command 'clang' failed with exit status 1 Signed-off-by: -
Linus Torvalds authored
When the pipe is closed, we mark the associated watchqueue defunct by calling watch_queue_clear(). However, while that is protected by the watchqueue lock, new watchqueue entries aren't actually added under that lock at all: they use the pipe->rd_wait.lock instead, and looking up that pipe happens without any locking. The watchqueue code uses the RCU read-side section to make sure that the wqueue entry itself hasn't disappeared, but that does not protect the pipe_info in any way. So make sure to actually hold the wqueue lock when posting watch events, properly serializing against the pipe being torn down. Reported-by:
Noam Rathaus <noamr@ssd-disclosure.com> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: David Howells <dhowells@redhat.com> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Eric Snowberg authored
The lockdown LSM is primarily used in conjunction with UEFI Secure Boot. This LSM may also be used on machines without UEFI. It can also be enabled when UEFI Secure Boot is disabled. One of lockdown's features is to prevent kexec from loading untrusted kernels. Lockdown can be enabled through a bootparam or after the kernel has booted through securityfs. If IMA appraisal is used with the "ima_appraise=log" boot param, lockdown can be defeated with kexec on any machine when Secure Boot is disabled or unavailable. IMA prevents setting "ima_appraise=log" from the boot param when Secure Boot is enabled, but this does not cover cases where lockdown is used without Secure Boot. To defeat lockdown, boot without Secure Boot and add ima_appraise=log to the kernel command line; then: $ echo "integrity" > /sys/kernel/security/lockdown $ echo "appraise func=KEXEC_KERNEL_CHECK appraise_type=imasig" > \ /sys/kernel/security/ima/policy $ kexec -ls unsigned-kernel Add a call to verify ima appraisal is set to "enforce" whenever lockdown is enabled. This fixes CVE-2022-21505. Cc: stable@vger.kernel.org Fixes: 29d3c1c8 ("kexec: Allow kexec_file() with appropriate IMA policy when locked down") Signed-off-by:Eric Snowberg <eric.snowberg@oracle.com> Acked-by:
Mimi Zohar <zohar@linux.ibm.com> Reviewed-by:
John Haxby <john.haxby@oracle.com> Signed-off-by:
-
Arnaldo Carvalho de Melo authored
One in perf's CFLAGS and the other in the distro python binding scripts. So if use the usual technique of first -D_FORTIFY_SOURCE then -D it. Noticed with: opensuse tumbleweed: gcc version 12.1.1 20220629 [revision 7811663964aa7e31c3939b859bbfa2e16919639f] (SUSE Linux) Signed-off-by: -
Thomas Richter authored
Perf test case 83: perf stat CSV output linter might fail on s390. The reason for this is the output of the command ./perf stat -x, -A -a --no-merge true which depends on a .config file setting. When CONFIG_SCHED_TOPOLOGY is set, the output of above perf command is CPU0,1.50,msec,cpu-clock,1502781,100.00,1.052,CPUs utilized When CONFIG_SCHED_TOPOLOGY is *NOT* set the output of above perf command is 0.95,msec,cpu-clock,949800,100.00,1.060,CPUs utilized Fix the test case to accept both output formats. Output before: # perf test 83 83: perf stat CSV output linter : FAILED! # Output after: # ./perf test 83 83: perf stat CSV output linter : Ok # Fixes: ec906102 ("perf test: Fix "perf stat CSV output linter" test on s390") Signed-off-by:
Thomas Richter <tmricht@linux.ibm.com> Acked-by:
Sumanth Korikkar <sumanthk@linux.ibm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Link: https://lore.kernel.org/r/20220720123419.220953-1-tmricht@linux.ibm.comSigned-off-by:
-
Jason Wang authored
The double `the' is duplicated in the comment, remove one. Signed-off-by:
Jason Wang <wangborong@cdjrlc.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.garry@huawei.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Will Deacon <will@kernel.org> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Link: http://lore.kernel.org/lkml/20220716044040.43123-1-wangborong@cdjrlc.comSigned-off-by:
-
Jason Wang authored
The double `the' is duplicated in the comment, remove one. Signed-off-by:
-
Arnaldo Carvalho de Melo authored
On gcc 12 we started seeing this: In file included from /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/perl.h:2999, from util/scripting-engines/trace-event-perl.c:35: /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/inline.h: In function 'Perl_is_utf8_valid_partial_char_flags': /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/handy.h:125:23: error: cast from function call of type 'STRLEN' {aka 'long unsigned int'} to non-matching type '_Bool' [-Werror=bad-function-cast] 125 | #define cBOOL(cbool) ((bool) (cbool)) | ^ /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/inline.h:2363:12: note: in expansion of macro 'cBOOL' 2363 | return cBOOL(is_utf8_char_helper_(s0, e, flags)); | ^~~~~ In file included from /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/perl.h:7242: /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/inline.h: In function 'Perl_cop_file_avn': /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/inline.h:3489:5: error: ISO C90 forbids mixed declarations and code [-Werror=declaration-after-statement] 3489 | const char *file = CopFILE(cop); | ^~~~~ In file included from /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/perl.h:7243: /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/sv_inline.h: In function 'Perl_newSV_type': /usr/lib/perl5/5.36.0/x86_64-linux-thread-multi/CORE/sv_inline.h:376:5: error: enumeration value 'SVt_LAST' not handled in switch [-Werror=switch-enum] 376 | switch (type) { | ^~~~~~ So disable those warnings to keep building with perl devel headers. Noticed, among other distros, on opensuse tumbleweed: gcc version 12.1.1 20220629 [revision 7811663964aa7e31c3939b859bbfa2e16919639f] (SUSE Linux) Signed-off-by: -
Ian Rogers authored
Fix the following DeprecationWarning: tools/perf/util/setup.py:31: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives Note: the setuptools module may need installing, for example: $ sudo apt install python-setuptools Reviewer comments: James said: Tested it with python 2.7 and 3.8 by running "make install-python_ext PYTHON=..." Committer notes: Tested with: $ make -k BUILD_BPF_SKEL=1 PYTHON=python3 O=/tmp/build/perf -C tools/perf install-bin ; perf test python $ make -k BUILD_BPF_SKEL=1 O=/tmp/build/perf -C tools/perf install-bin ; perf test python Reviewed-by:
James Clark <james.clark@arm.com> Signed-off-by:
-
Ian Rogers authored
If HAVE_GTK2_SUPPORT isn't defined then --gtk can't succeed, don't support it as a command line option in this case. v2. Is a rebase. Patch appears to have been missed in: https://lore.kernel.org/lkml/Ygu40djM1MqAfkcF@kernel.org/Signed-off-by:
-
Adrian Hunter authored
Now it is possible to decode a host Intel PT trace including guest machine user space, add documentation for the steps needed to do it. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Acked-by:
-
Adrian Hunter authored
When decoding with guest sideband information, for VMX non-root (NR) i.e. guest events, replace the host (hypervisor) pid/tid with guest values, and provide also the new machine_pid and vcpu values. Signed-off-by:
-
Adrian Hunter authored
When decoding with guest sideband information, for VMX non-root (NR) i.e. guest errors, replace the host (hypervisor) pid/tid with guest values, and provide also the new machine_pid and vcpu values. Signed-off-by:
-
Adrian Hunter authored
Prior to decoding, determine what guest thread, if any, is running. Signed-off-by:
-
Adrian Hunter authored
The sync_switch facility attempts to better synchronize context switches with the Intel PT trace, however it is not designed for guest machine context switches, so disable it when guest sideband is detected. Signed-off-by:
-
Adrian Hunter authored
Use guest context switch events to keep track of which guest thread is running on a particular guest machine and VCPU. Signed-off-by:
-
Adrian Hunter authored
To aid debugging, add some more logging to intel_pt_walk_next_insn(). Signed-off-by:
-
Adrian Hunter authored
Remove guest_machine_pid because it is not needed. Signed-off-by:
-
Adrian Hunter authored
Add a helper function to determine if an event is a guest event. Signed-off-by:
-
Adrian Hunter authored
If a kernel mmap event was recorded inside a guest and injected into a host perf.data file, then it will match a host mmap_name not a guest mmap_name, see machine__set_mmap_name(). So try matching a host mmap_name in that case. Signed-off-by:
-
Adrian Hunter authored
Prepare machine__set_current_tid() for use with guest machines that do not currently have a machine->env->nr_cpus_avail value by making use of realloc_array_as_needed(). Signed-off-by:
-
Adrian Hunter authored
Inject events from a perf.data file recorded in a virtual machine into a perf.data file recorded on the host at the same time. Only side band events (e.g. mmap, comm, fork, exit etc) and build IDs are injected. Additionally, the guest kcore_dir is copied as kcore_dir__ appended to the machine PID. This is non-trivial because: o It is not possible to process 2 sessions simultaneously so instead events are first written to a temporary file. o To avoid conflict, guest sample IDs are replaced with new unused sample IDs. o Guest event's CPU is changed to be the host CPU because it is more useful for reporting and analysis. o Sample ID is mapped to machine PID which is recorded with VCPU in the id index. This is important to allow guest events to be related to the guest machine and VCPU. o Timestamps must be converted. o Events are inserted to obey finished-round ordering. The anticipated use-case is: - start recording sideband events in a guest machine - start recording an AUX area trace on the host which can trace also the guest (e.g. Intel PT) - run test case on the guest - stop recording on the host - stop recording on the guest - copy the guest perf.data file to the host - inject the guest perf.data file sideband events into the host perf.data file using perf inject - the resulting perf.data file can now be used Subsequent patches provide Intel PT support for this. Signed-off-by:
-
Adrian Hunter authored
Add helper reallocarray_as_needed() to reallocate an array to a larger size and initialize the extra entries to an arbitrary value. Signed-off-by:
-
Adrian Hunter authored
When registering a guest machine using machine_pid from the id index, check perf.data for a matching kcore_dir subdirectory and set the kallsyms file name accordingly. If set, use it to find the machine's kernel symbols and object code (from kcore). Signed-off-by:
-
Adrian Hunter authored
Copies of /proc/kallsyms, /proc/modules and an extract of /proc/kcore can be stored in the perf.data output directory under the subdirectory named kcore_dir. Guest machines will have their files also under subdirectories beginning kcore_dir__ followed by the machine pid. Make has_kcore_dir() return true also if there is a guest machine kcore_dir. Signed-off-by:
-
Adrian Hunter authored
Copies of /proc/kallsyms, /proc/modules and an extract of /proc/kcore can be stored in the perf.data output directory under the subdirectory named kcore_dir. Guest machines will have their files also under subdirectories beginning kcore_dir__ followed by the machine pid. Remove these also when removing kcore_dir. Signed-off-by:
-
Adrian Hunter authored
Add machine_pid and vcpu to the intel-pt-events.py script. Signed-off-by:
-
Adrian Hunter authored
Add machine_pid and vcpu to python sample events and context switch events. Signed-off-by:
-
