- 04 Oct, 2022 40 commits
-
-
Namhyung Kim authored
This is a preparation to display accurate lost sample counts for each evsel. Reviewed-by:
Adrian Hunter <adrian.hunter@intel.com> Signed-off-by:
Namhyung Kim <namhyung@kernel.org> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20220901195739.668604-5-namhyung@kernel.orgSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Namhyung Kim authored
When there are lost samples, it can read the number of PERF_FORMAT_LOST and convert it to PERF_RECORD_LOST_SAMPLES and write to the data file at the end. Reviewed-by:
-
Namhyung Kim authored
As we want to see the number of lost samples in the perf report, set the LOST format when it configs evsel. On old kernels, it'd fallback to disable it. Reviewed-by:
-
Namhyung Kim authored
So that we can see it with: $ perf record -vv pwd ... perf_event_attr: size 128 { sample_period, sample_freq } 4000 sample_type IP|TID|TIME|PERIOD read_format ID|LOST disabled 1 inherit 1 exclude_kernel 1 freq 1 enable_on_exec 1 precise_ip 3 sample_id_all 1 exclude_guest 1 Reviewed-by: -
Ian Rogers authored
Make the header guard consistent with others. Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Caleb Biggers <caleb.biggers@intel.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.garry@huawei.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kshipra Bopardikar <kshipra.bopardikar@intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Miaoqian Lin <linmq006@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: Thomas Richter <tmricht@linux.ibm.com> Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com> Cc: florian fischer <florian.fischer@muhq.space> Link: http://lore.kernel.org/lkml/20220830164846.401143-2-irogers@google.comSigned-off-by:
-
Anshuman Khandual authored
This updates the perf tool with arch specific branch type classification used for BRBE on arm64 platform as added in the kernel earlier. Signed-off-by:
Anshuman Khandual <anshuman.khandual@arm.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Robin Murphy <robin.murphy@arm.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will@kernel.org> Cc: linux-arm-kernel@lists.infradead.org Link: https://lore.kernel.org/r/20220824044822.70230-9-anshuman.khandual@arm.comSigned-off-by:
-
Anshuman Khandual authored
This updates the perf tools with branch privilege information request flag i.e PERF_SAMPLE_BRANCH_PRIV_SAVE that has been added earlier in the kernel. This also updates 'perf record' documentation, branch_modes[], and generic branch privilege level enumeration as added earlier in the kernel. Signed-off-by:
-
Anshuman Khandual authored
This updates the perf tool with generic branch type classification with new ABI extender place holder i.e PERF_BR_EXTEND_ABI, the new 4 bit branch type field i.e perf_branch_entry.new_type, new generic page fault related branch types and some arch specific branch types as added earlier in the kernel. Committer note: Add an extra entry to the branch_type_name array to cope with PERF_BR_EXTEND_ABI, to address build warnings on some compiler/systems, like: 75 8.89 ubuntu:20.04-x-powerpc64el : FAIL gcc version 10.3.0 (Ubuntu 10.3.0-1ubuntu1~20.04) inlined from 'branch_type_stat_display' at util/branch.c:152:4: /usr/powerpc64le-linux-gnu/include/bits/stdio2.h:100:10: error: '%8s' directive argument is null [-Werror=format-overflow=] 100 | return __fprintf_chk (__stream, __USE_FORTIFY_LEVEL - 1, __fmt, | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 101 | __va_arg_pack ()); | ~~~~~~~~~~~~~~~~~ Signed-off-by: -
Anshuman Khandual authored
This updates the perf tool with generic branch type classification with two new branch types i.e system error (PERF_BR_SERROR) and not in transaction (PERF_BR_NO_TX) which got updated earlier in the kernel. This also updates corresponding branch type strings in branch_type_name(). Committer notes: At perf tools merge time this is only on PeterZ's tree, at: git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git perf/core So for testing one has to build a kernel with that branch, then test the tooling side from: git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git perf/core Reviewed-by:
James Clark <james.clark@arm.com> Signed-off-by:
-
Ian Rogers authored
If building with clang then enable -Wthread-safety warnings. Signed-off-by:
-
Ian Rogers authored
Add annotations to describe lock behavior. Signed-off-by:
-
Ian Rogers authored
Add annotations to describe lock behavior. Add unlocks so that mutexes aren't conditionally held on exit from perf_sched__replay. Add an exit variable so that thread_func can terminate, rather than leaving the threads blocked on mutexes. Signed-off-by:
-
Ian Rogers authored
Add thread safety annotations to struct mutex so that when compiled with clang's -Wthread-safety warnings are generated for erroneous lock patterns. NO_THREAD_SAFETY_ANALYSIS is needed for mutex_lock/mutex_unlock as the analysis doesn't under pthread calls. Signed-off-by:
-
Ian Rogers authored
There may be threads racing to update dso->nsinfo: https://lore.kernel.org/linux-perf-users/CAP-5=fWZH20L4kv-BwVtGLwR=Em3AOOT+Q4QGivvQuYn5AsPRg@mail.gmail.com/ Holding the dso->lock avoids use-after-free, memory leaks and other such bugs. Apply the fix in: https://lore.kernel.org/linux-perf-users/20211118193714.2293728-1-irogers@google.com/ of there being a missing nsinfo__put now that the accesses are data race free. Fixes test "Lookup mmap thread" when compiled with address sanitizer. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Ian Rogers authored
The comment says it is for cpu_set_t which isn't used in the header. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Update cmd_sched so that we always explicitly destroy the mutexes. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking for synth_lock. Signed-off-by:
-
Ian Rogers authored
No pthread usage in builtin-lock.c. Signed-off-by:
-
Ian Rogers authored
No pthread usage in bpf-event.h. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Ian Rogers authored
pthread.h is being included for the side-effect of getting sched.h and macros like CPU_CLR. Switch to directly using sched.h, or if that is already present, just remove the pthread.h inclusion entirely. Signed-off-by:
-
Ian Rogers authored
Switch to the use of mutex wrappers that provide better error checking. Signed-off-by:
-
Pavithra Gurushankar authored
Added a new header file mutex.h that wraps the usage of pthread_mutex_t and pthread_cond_t. By abstracting these it is possible to introduce error checking. Signed-off-by:
Pavithra Gurushankar <gpavithrasha@gmail.com> Reviewed-by:
-
Adrian Hunter authored
AUX area traces can produce too much data to record successfully or analyze subsequently. Add another means to reduce data collection by allowing multiple recording time ranges. This is useful, for instance, in cases where a workload produces predictably reproducible events in specific time ranges. Today we only have perf record -D <msecs> to start at a specific region, or some complicated approach using snapshot mode and external scripts sending signals or using the fifos. But these approaches are difficult to set up compared with simply having perf do it. Extend perf record option -D/--delay option to specifying relative time stamps for start stop controlled by perf with the right time offset, for instance: perf record -e intel_pt// -D 10-20,30-40 to record 10ms to 20ms into the trace and 30ms to 40ms. Example: The example workload is: $ cat repeat-usleep.c int usleep(useconds_t usec); int usage(int ret, const char *msg) { if (msg) fprintf(stderr, "%s\n", msg); fprintf(stderr, "Usage is: repeat-usleep <microseconds>\n"); return ret; } int main(int argc, char *argv[]) { unsigned long usecs; char *end_ptr; if (argc != 2) return usage(1, "Error: Wrong number of arguments!"); errno = 0; usecs = strtoul(argv[1], &end_ptr, 0); if (errno || *end_ptr || usecs > UINT_MAX) return usage(1, "Error: Invalid argument!"); while (1) { int ret = usleep(usecs); if (ret & errno != EINTR) return usage(1, "Error: usleep() failed!"); } return 0; } $ perf record -e intel_pt//u --delay 10-20,40-70,110-160 -- ./repeat-usleep 500 Events disabled Events enabled Events disabled Events enabled Events disabled Events enabled Events disabled [ perf record: Woken up 5 times to write data ] [ perf record: Captured and wrote 0.204 MB perf.data ] Terminated A dlfilter is used to determine continuous data collection (timestamps less than 1ms apart): $ cat dlfilter-show-delays.c static __u64 start_time; static __u64 last_time; int start(void **data, void *ctx) { printf("%-17s\t%-9s\t%-6s\n", " Time", " Duration", " Delay"); return 0; } int filter_event_early(void *data, const struct perf_dlfilter_sample *sample, void *ctx) { __u64 delta; if (!sample->time) return 1; if (!last_time) goto out; delta = sample->time - last_time; if (delta < 1000000) goto out2;; printf("%17.9f\t%9.1f\t%6.1f\n", start_time / 1000000000.0, (last_time - start_time) / 1000000.0, delta / 1000000.0); out: start_time = sample->time; out2: last_time = sample->time; return 1; } int stop(void *data, void *ctx) { printf("%17.9f\t%9.1f\n", start_time / 1000000000.0, (last_time - start_time) / 1000000.0); return 0; } The result shows the times roughly match the --delay option: $ perf script --itrace=qb --dlfilter dlfilter-show-delays.so Time Duration Delay 39215.302317300 9.7 20.5 39215.332480217 30.4 40.9 39215.403837717 49.8 Signed-off-by: -
Adrian Hunter authored
Dummy events are used to provide sideband information like MMAP events that are always needed even when main events are disabled. Add functions that take that into account. Signed-off-by:
-
Adrian Hunter authored
Patch "perf record: Fix way of handling non-perf-event pollfds" added a generic way to handle non-perf-event file descriptors like evlist->ctl_fd. Use it instead of handling evlist->ctl_fd separately. Signed-off-by:
-
Adrian Hunter authored
evlist__add_wakeup_eventfd() calls perf_evlist__add_pollfd() to add a non-perf-event to the evlist pollfds. Since commit 415ccb58 ("perf record: Introduce thread specific data array") that doesn't work because evlist pollfs is not polled and done_fd is not duplicated into thread-data. Patch "perf record: Fix way of handling non-perf-event pollfds" added a new approach that ensures file descriptors like done_fd are handled correctly by flagging them as fdarray_flag__non_perf_event. Fix by flagging done_fd as fdarray_flag__non_perf_event. Example: Before: $ sleep 3 & perf record -vv -p $! ... thread_data[0x55f44bd34140]: pollfd[0] <- event_fd=5 thread_data[0x55f44bd34140]: pollfd[1] <- event_fd=6 thread_data[0x55f44bd34140]: pollfd[2] <- event_fd=7 thread_data[0x55f44bd34140]: pollfd[3] <- event_fd=8 thread_data[0x55f44bd34140]: pollfd[4] <- event_fd=9 thread_data[0x55f44bd34140]: pollfd[5] <- event_fd=10 thread_data[0x55f44bd34140]: pollfd[6] <- event_fd=11 thread_data[0x55f44bd34140]: pollfd[7] <- event_fd=12 ... After: $ sleep 3 & perf record -vv -p $! ... thread_data[0x55a8ded89140]: pollfd[0] <- event_fd=5 thread_data[0x55a8ded89140]: pollfd[1] <- event_fd=6 thread_data[0x55a8ded89140]: pollfd[2] <- event_fd=7 thread_data[0x55a8ded89140]: pollfd[3] <- event_fd=8 thread_data[0x55a8ded89140]: pollfd[4] <- event_fd=9 thread_data[0x55a8ded89140]: pollfd[5] <- event_fd=10 thread_data[0x55a8ded89140]: pollfd[6] <- event_fd=11 thread_data[0x55a8ded89140]: pollfd[7] <- event_fd=12 thread_data[0x55a8ded89140]: pollfd[8] <- non_perf_event fd=4 ... This patch depends on "perf record: Fix way of handling non-perf-event pollfds". Fixes: 415ccb58 ("perf record: Introduce thread specific data array") Signed-off-by:
-
Adrian Hunter authored
perf record __cmd_record() does not poll evlist pollfds. Instead it polls thread_data[0].pollfd. That happens whether or not threads are being used. perf record duplicates evlist mmap pollfds as needed for separate threads. The non-perf-event represented by evlist->ctl_fd has to handled separately, which is done explicitly, duplicating it into the thread_data[0] pollfds. That approach neglects any other non-perf-event file descriptors. Currently there is also done_fd which needs the same handling. Add a new generalized approach. Add fdarray_flag__non_perf_event to identify the file descriptors that need the special handling. For those cases, also keep a mapping of the evlist pollfd index and thread pollfd index, so that the evlist revents can be updated. Although this patch adds the new handling, it does not take it into use. There is no functional change, but it is the precursor to a fix, so is marked as a fix. Fixes: 415ccb58 ("perf record: Introduce thread specific data array") Signed-off-by:
-
Ian Rogers authored
When libbpf is present the build uses definitions in libbpf hashmap.c, however, libbpf's hashmap.h wasn't being used. Switch to using the correct hashmap.h dependent on the define HAVE_LIBBPF_SUPPORT. This was the original intent in: https://lore.kernel.org/lkml/20200515221732.44078-8-irogers@google.com/Signed-off-by:
-
Xin Gao authored
'unsigned int' should be clearer than 'unsigned'. Signed-off-by:
Xin Gao <gaoxin@cdjrlc.com> Cc: Ian Rogers <irogers@google.com> Cc: John Garry <john.garry@huawei.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lore.kernel.org/lkml/20220816173804.7539-1-gaoxin@cdjrlc.comSigned-off-by:
-
Xin Gao authored
'unsigned int' should be clearer than 'unsigned'. Signed-off-by:
-
Roberto Sassu authored
Sometimes, features are simply different flavors of another feature, to properly detect the exact dependencies needed by different Linux distributions. For example, libbfd has three flavors: libbfd if the distro does not require any additional dependency; libbfd-liberty if it requires libiberty; libbfd-liberty-z if it requires libiberty and libz. It might not be clear to the user whether a feature has been successfully detected or not, given that some of its flavors will be set to OFF, others to ON. Instead, display only the feature main flavor if not in verbose mode (VF != 1), and set it to ON if at least one of its flavors has been successfully detected (logical OR), OFF otherwise. Omit the other flavors. Accomplish that by declaring a FEATURE_GROUP_MEMBERS-<feature main flavor> variable, with the list of the other flavors as variable value. For now, do it just for libbfd. In verbose mode, of if no group is defined for a feature, show the feature detection result as before. Committer testing: Collecting the output from: $ make -C tools/bpf/bpftool/ clean $ make -C tools/bpf/bpftool/ |& grep "Auto-detecting system features" -A10 $ diff -u before after --- before 2022-08-18 10:06:40.422086966 -0300 +++ after 2022-08-18 10:07:59.202138282 -0300 @@ -1,6 +1,4 @@ Auto-detecting system features: ... libbfd: [ on ] -... libbfd-liberty: [ on ] -... libbfd-liberty-z: [ on ] ... libcap: [ on ] ... clang-bpf-co-re: [ on ] $ Signed-off-by:
Roberto Sassu <roberto.sassu@huawei.com> Tested-by:
-
Roberto Sassu authored
Since now there are features with a long name, increase the room for them, so that fields are correctly aligned. Signed-off-by:
-
Roberto Sassu authored
As the first eval expansion is used only to generate Makefile statements, messages should not be displayed at this stage, as for example conditional expressions are not evaluated. It can be seen for example in the output of feature detection for bpftool, where the number of detected features does not change, despite turning on the verbose mode (VF = 1) and there are additional features to display. Fix this issue by escaping the $ before $(info) statements, to ensure that messages are printed only when the function containing them is actually executed, and not when it is expanded. In addition, move the $(info) statement out of feature_print_status, due to the fact that is called both inside and outside an eval context, and place it to the caller so that the $ can be escaped when necessary. For symmetry, move the $(info) statement also out of feature_print_text, and place it to the caller. Force the TMP variable evaluation in verbose mode, to display the features in FEATURE_TESTS that are not in FEATURE_DISPLAY. Reorder perf feature detection messages (first non-verbose, then verbose ones) by moving the call to feature_display_entries earlier, before the VF environment variable check. Also, remove the newline from that function, as perf might display additional messages. Move the newline to perf Makefile, and display another one if displaying the detection result is not deferred as in the case of bpftool. Committer testing: Collecting the output from: $ make VF=1 -C tools/bpf/bpftool/ |& grep "Auto-detecting system features" -A20 $ diff -u before after --- before 2022-08-18 09:59:55.460529231 -0300 +++ after 2022-08-18 10:01:11.182517282 -0300 @@ -4,3 +4,5 @@ ... libbfd-liberty-z: [ on ] ... libcap: [ on ] ... clang-bpf-co-re: [ on ] +... disassembler-four-args: [ on ] +... disassembler-init-styled: [ OFF ] $ Fixes: 0afc5cad ("perf build: Separate feature make support into config/Makefile.feature") Signed-off-by:
-
Raul Silvera authored
This commit adds the option --known-build-ids to perf inject. It allows the user to explicitly specify the build id for a given path, instead of retrieving it from the current system. This is useful in cases where a perf.data file is processed on a different system from where it was collected, or if some of the binaries are no longer available. The build ids and paths are specified in pairs in the command line. Using the file:// specifier, build ids can be loaded from a file directly generated by perf buildid-list. This is convenient to copy build ids from one perf.data file to another. ** Example: In this example we use perf record to create two perf.data files, one with build ids and another without, and use perf buildid-list and perf inject to copy the build ids from the first file to the second. $ perf record ls /tmp $ perf record --no-buildid -o perf.data.no-buildid ls /tmp $ perf buildid-list > build-ids.txt $ perf inject -b --known-build-ids='file://build-ids.txt' \ -i perf.data.no-buildid -o perf.data.buildid Signed-off-by:
Raul Silvera <rsilvera@google.com> Acked-by:
-
Linus Torvalds authored
Merge tag 'statx-dioalign-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux Pull STATX_DIOALIGN support from Eric Biggers: "Make statx() support reporting direct I/O (DIO) alignment information. This provides a generic interface for userspace programs to determine whether a file supports DIO, and if so with what alignment restrictions. Specifically, STATX_DIOALIGN works on block devices, and on regular files when their containing filesystem has implemented support. An interface like this has been requested for years, since the conditions for when DIO is supported in Linux have gotten increasingly complex over time. Today, DIO support and alignment requirements can be affected by various filesystem features such as multi-device support, data journalling, inline data, encryption, verity, compression, checkpoint disabling, log-structured mode, etc. Further complicating things, Linux v6.0 relaxed the traditional rule of DIO needing to be aligned to the block device's logical block size; now user buffers (but not file offsets) only need to be aligned to the DMA alignment. The approach of uplifting the XFS specific ioctl XFS_IOC_DIOINFO was discarded in favor of creating a clean new interface with statx(). For more information, see the individual commits and the man page update[1]" Link: https://lore.kernel.org/r/20220722074229.148925-1-ebiggers@kernel.org [1] * tag 'statx-dioalign-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux: xfs: support STATX_DIOALIGN f2fs: support STATX_DIOALIGN f2fs: simplify f2fs_force_buffered_io() f2fs: move f2fs_force_buffered_io() into file.c ext4: support STATX_DIOALIGN fscrypt: change fscrypt_dio_supported() to prepare for STATX_DIOALIGN vfs: support STATX_DIOALIGN on block devices statx: add direct I/O alignment information
-
