- 20 Oct, 2015 6 commits
-
-
Stephane Eranian authored
This patch add a new branch type sampling filter to perf record. It is named 'call' and maps to PERF_SAMPLE_BRANCH_CALL. It samples direct call branches only, unlike 'any_call' which includes indirect calls as well. $ perf record -j call -e cycles ..... The man page is updated accordingly. Signed-off-by:
Stephane Eranian <eranian@google.com> Signed-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vince Weaver <vincent.weaver@maine.edu> Cc: khandual@linux.vnet.ibm.com Link: http://lkml.kernel.org/r/1444720151-10275-5-git-send-email-eranian@google.comSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Stephane Eranian authored
The patch catches PERF_SAMPLE_BRANCH_CALL because it is not clear whether this is actually supported by the hardware. Signed-off-by:
-
Stephane Eranian authored
This patch enables the suport for the PERF_SAMPLE_BRANCH_CALL for Intel x86 processors. When the processor support LBR filtering this the selection is done in hardware. Otherwise, the filter is applied by software. Note that we chose to include zero length calls because they also represent calls. Signed-off-by:
-
Stephane Eranian authored
Add a new branch sample type to cover only call branches (function calls). The current ANY_CALL included direct, indirect calls and far jumps. We want to be able to differentiate indirect from direct calls. Therefore we introduce PERF_SAMPLE_BRANCH_CALL. The implementation is up to each architecture. Signed-off-by:
-
Adrian Hunter authored
Commit: b20112ed ("perf/x86: Improve accuracy of perf/sched clock") allowed the time_shift value in perf_event_mmap_page to be as much as 32. Unfortunately the documented algorithms for using time_shift have it shifting an integer, whereas to work correctly with the value 32, the type must be u64. In the case of perf tools, Intel PT decodes correctly but the timestamps that are output (for example by perf script) have lost 32-bits of granularity so they look like they are not changing at all. Fix by limiting the shift to 31 and adjusting the multiplier accordingly. Also update the documentation of perf_event_mmap_page so that new code based on it will be more future-proof. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements and fixes: User visible changes: - 'perf bench mem' now prefaults unconditionally, no sense in providing modes where page faults are measured. (Ingo Molnar) - Harmonize -l/--nr_loops accross 'perf bench'. (Ingo Molnar) - Various 'perf bench' consistency improvements. (Ingo Molnar) - Suppress libtraceevent warnings in non-verbose 'perf test' mode. (Namhyung Kim) - Move some tracepoint event test error messages to the verbose mode of 'perf test'. (Namhyung Kim) - Make 'perf help' usage message consistent with other tools. (Yunlong Song) Build fixes: - Fix 'perf bench' build with gcc 4.4.7. (Arnaldo Carvalho de Melo) Infrastructure changes: - 'perf stat' prep work for the 'perf stat scripting' patchkit. (Jiri Olsa) Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by:
-
- 19 Oct, 2015 23 commits
-
-
Arnaldo Carvalho de Melo authored
To avoid this splat with gcc 4.4.7: cc1: warnings being treated as errors bench/mem-functions.c:273: error: missing initializer bench/mem-functions.c:273: error: (near initialization for ‘memcpy_functions[4].desc’) bench/mem-functions.c:366: error: missing initializer bench/mem-functions.c:366: error: (near initialization for ‘memset_functions[4].desc’) Cc: David Ahern <dsahern@gmail.com> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-0s8o6tgw1pdwvdv02llb9tkd@git.kernel.orgSigned-off-by:
-
Jiri Olsa authored
There's no need to check sampling output fields for events without perf_event_attr::sample_type field set. Signed-off-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
Kan Liang <kan.liang@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1444992092-17897-51-git-send-email-jolsa@kernel.orgSigned-off-by:
-
Jiri Olsa authored
Adding data arg to cpu_map__build_map callback, so we could pass data along to the callback. It'll be needed in following patches to retrieve topology info from perf.data. Signed-off-by:
-
Jiri Olsa authored
We'll need to call it from perf stat in the stat_script patchkit Signed-off-by:
-
Jiri Olsa authored
Adding AGGR_UNSET mode, so we could distinguish unset aggr_mode in following patches. Signed-off-by:
-
Jiri Olsa authored
It's used as the perf_evsel::priv data, so the name suits better. Also we'll need the perf_stat name free for more generic struct. Signed-off-by:
-
Yunlong Song authored
Capitalize 'usage' to make it consistent with all the other 'Usage' in the codes, e.g., usage_builtin. Signed-off-by:
Yunlong Song <yunlong.song@huawei.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Ramkumar Ramachandra <artagnon@gmail.com> Cc: Sriram Raghunathan <sriram.r@nokia.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/r/1444894792-2338-3-git-send-email-yunlong.song@huawei.comSigned-off-by:
-
Ingo Molnar authored
So right now we output this text: memcpy: Benchmark for memcpy() functions memset: Benchmark for memset() functions all: Test all memory access benchmarks But the right verb to use with benchmarks is to 'run' them, not 'test' them. So change this (and all similar texts) to: memcpy: Benchmark for memcpy() functions memset: Benchmark for memset() functions all: Run all memory access benchmarks Signed-off-by: -
Ingo Molnar authored
So right now there's a somewhat inconsistent mess of the benchmarking code and options sometimes calling benchmarked functions 'functions', sometimes calling them 'routines'. Name them 'functions' consistently. Signed-off-by:
-
Ingo Molnar authored
We have three benchmarking subsystems that specify some sort of 'number of loops' parameter - but all of them do it inconsistently: numa: -l/--nr_loops sched messaging: -l/--loops mem memset/memcpy: -i/--iterations Harmonize them to -l/--nr_loops by picking the numa variant - which is also the most likely one to have existing scripting which we don't want to break. Plus improve the parameter help texts to indicate the default value for the nr_loops variable to keep users from guessing ... Also propagate the naming to internal variables. Signed-off-by:
-
Ingo Molnar authored
Reorder functions a bit, so that we synchronize the layout of the memcpy() and memset() portions of the code. This improves the code, especially after we'll add an strlcpy() variant as well. Signed-off-by:
-
Ingo Molnar authored
- fix various typos in user visible output strings - make the output consistent (wrt. capitalization and spelling) - offer the list of routines to benchmark on '-r help'. Signed-off-by:
-
Ingo Molnar authored
So 'perf bench mem memcpy/memset' consistently uses 'len' and 'length' for buffer sizes - while it's really a memory buffer size. (strings have length.) Rename all affected variables. Signed-off-by:
-
Ingo Molnar authored
So bench/mem-functions.c has a 'routine' name for the routines parameter string, but a 'length_str' name for the length parameter string. We also have another entity named 'routine': 'struct routine'. This is inconsistent and confusing: rename 'routine' to 'routine_str'. Also fix typos in the --routine help text. Signed-off-by:
-
Ingo Molnar authored
So 'perf bench mem memset/memcpy' has a CPU cycles measurement method, but calls it 'cycle' (singular) throughout the code, which makes it harder to read. Rename all related functions, variables and options to a plural 'cycles' nomenclature. Signed-off-by:
-
Ingo Molnar authored
So 'perf bench -h' is not very helpful when printing the help line about the output formatting options: -f, --format <default> Specify format style There are two output format styles, 'default' and 'simple', so improve the help text to: -f, --format <default|simple> Specify the output formatting style Signed-off-by: -
Ingo Molnar authored
So 'perf bench mem memcpy/memset' has elaborate code to measure memcpy()/memset() performance both with freshly allocated buffers (which includes initial page fault overhead) and with preallocated buffers. But the thing is, the resulting bandwidth results are mostly meaningless, because page faults dominate so much of the cost. It might make sense to measure cache cold vs. cache hot performance, but the code does not do this. So remove this complication, and always prefault the ranges before using them. Signed-off-by:
-
Ingo Molnar authored
So mem-memcpy.c started out as a simple memcpy() benchmark, then it grew memset() functionality and now I plan to add string copy benchmarks as well. This makes the file name a misnomer: rename it to the more generic mem-functions.c name. Signed-off-by:
-
Ingo Molnar authored
Signed-off-by:
-
Ingo Molnar authored
So few people know that the --routine option to 'perf bench memcpy/memset' exists, and would not know that it's capable of testing the kernel's memcpy/memset implementations. Furthermore, 'perf bench mem all' will not run all routines: vega:~> perf bench mem all # Running mem/memcpy benchmark... Routine default (Default memcpy() provided by glibc) # Copying 1MB Bytes ... 894.454383 MB/Sec 3.844734 GB/Sec (with prefault) # Running mem/memset benchmark... Routine default (Default memset() provided by glibc) # Copying 1MB Bytes ... 1.220703 GB/Sec 9.042245 GB/Sec (with prefault) Because misleadingly the 'all' refers to 'all sub-benchmarks', not 'all sub-benchmarks and routines'. Fix all this by making the memcpy/memset routine to default to 'all', which results in all the benchmarks being run: triton:~> perf bench mem all # Running mem/memcpy benchmark... Routine default (Default memcpy() provided by glibc) # Copying 1MB Bytes ... 1.448906 GB/Sec 4.957170 GB/Sec (with prefault) Routine x86-64-unrolled (unrolled memcpy() in arch/x86/lib/memcpy_64.S) # Copying 1MB Bytes ... 1.614153 GB/Sec 4.379204 GB/Sec (with prefault) Routine x86-64-movsq (movsq-based memcpy() in arch/x86/lib/memcpy_64.S) # Copying 1MB Bytes ... 1.570036 GB/Sec 4.264465 GB/Sec (with prefault) Routine x86-64-movsb (movsb-based memcpy() in arch/x86/lib/memcpy_64.S) # Copying 1MB Bytes ... 1.788576 GB/Sec 6.554111 GB/Sec (with prefault) # Running mem/memset benchmark... Routine default (Default memset() provided by glibc) # Copying 1MB Bytes ... 2.082223 GB/Sec 9.126752 GB/Sec (with prefault) Routine x86-64-unrolled (unrolled memset() in arch/x86/lib/memset_64.S) # Copying 1MB Bytes ... 5.710892 GB/Sec 8.346688 GB/Sec (with prefault) Routine x86-64-stosq (movsq-based memset() in arch/x86/lib/memset_64.S) # Copying 1MB Bytes ... 9.765625 GB/Sec 12.520032 GB/Sec (with prefault) Routine x86-64-stosb (movsb-based memset() in arch/x86/lib/memset_64.S) # Copying 1MB Bytes ... 9.668936 GB/Sec 12.682630 GB/Sec (with prefault) Signed-off-by:
-
Ingo Molnar authored
- improve the readability of initializations - fix unnecessary double negations - fix ugly line breaks - fix other small details Signed-off-by:
-
Namhyung Kim authored
Currently libtraceevent emits warning on unsupported event formats. However it'd be better to see them only -v option is given. To do that, it needs to override the warning() function which is used in the libtracevent. Thus add set_warning_routine() same as set_die_routine() and check the verbose flag in our warning routine. Before: # perf test 5 5: parse events tests : Warning: [kvmmmu:kvm_mmu_get_page] bad op token { Warning: [kvmmmu:kvm_mmu_sync_page] bad op token { Warning: [kvmmmu:kvm_mmu_unsync_page] bad op token { Warning: [kvmmmu:kvm_mmu_prepare_zap_page] bad op token { Warning: [kvmmmu:fast_page_fault] function is_writable_pte not defined ... Ok After: # perf test 5 5: parse events tests : Ok Signed-off-by:Namhyung Kim <namhyung@kernel.org> Tested-by:
David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/r/1445268229-1601-2-git-send-email-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
Currently, when 'perf test' is run by a normal user, it'll fail to access tracepoint events. The output becomes somewhat messy because it tries to be nice with long error messages and hints. IMHO this is not needed for 'perf test' by default and AFAIK 'perf test' uses pr_debug() rather than pr_err() for such messages so that one can use -v option to see further details on failed testcases if needed. Before: $ perf test 1: vmlinux symtab matches kallsyms : FAILED! 2: detect openat syscall event :Error: No permissions to read /sys/kernel/debug/tracing/events/syscalls/sys_enter_openat Hint: Try 'sudo mount -o remount,mode=755 /sys/kernel/debug/tracing' FAILED! 3: detect openat syscall event on all cpus :Error: No permissions to read /sys/kernel/debug/tracing/events/syscalls/sys_enter_openat Hint: Try 'sudo mount -o remount,mode=755 /sys/kernel/debug/tracing' FAILED! ... After: $ perf test 1: vmlinux symtab matches kallsyms : FAILED! 2: detect openat syscall event : FAILED! 3: detect openat syscall event on all cpus : FAILED! ... $ perf test -v 2 2: detect openat syscall event : --- start --- test child forked, pid 30575 Error: No permissions to read /sys/kernel/debug/tracing/events/syscalls/sys_enter_openat Hint: Try 'sudo mount -o remount,mode=755 /sys/kernel/debug/tracing' test child finished with -1 ---- end ---- detect openat syscall event: FAILED! Signed-off-by:
-
- 14 Oct, 2015 1 commit
-
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo: User visible changes: - Use the alternative with the most descriptive filename containing a vmlinux file for a given build-id, providing a better title line for tools such as 'annotate'. (Arnaldo Carvalho de Melo) - Remove help messages about previous right and left arrow keybidings, that were repurposed for horizontal scrolling. (Arnaldo Carvalho de Melo) - Inform how to reset the symbol filter in the hists browser. (top & report) (Arnaldo Carvalho de Melo) - Add 'm' key for context menu display in the hists browser, that became inacessible with the repurposing of the right arrow key for horizontal scrolling. (Namhyung Kim) - Use debug_frame for callchains if eh_frame is unusable. (Rabin Vicent) Build fixes: - Fix strict-aliasing breakage with gcc 4.4 in the READ_ONCE/WRITE_ONCE code adopted from the kernel tree, that builds with -fno-strict-aliasing while tools/perf/ uses -Wstrict-aliasing=3. (Jiri Olsa) - Fix unw_word_t pointer casts in code using libunwind for callchains, fixing the build in at least 32-bit MIPS systems. (Rabin Vicent) - Work around cross compile build problems related to fixdep. (Jiri Olsa) Signed-off-by:
-
- 13 Oct, 2015 5 commits
-
-
Jiri Olsa authored
He Kuang the new fixdep tool breaks cross compiling. The reason is it wouldn't get compiled under host arch, but under cross arch and failed to run. We need to add support for host side tools build, meanwhile disabling fixdep usage for cross arch builds. Reported-by:
He Kuang <hekuang@huawei.com> Signed-off-by:
-
Jiri Olsa authored
Vinson reported build breakage with gcc 4.4 due to strict-aliasing. CC util/annotate.o cc1: warnings being treated as errors util/annotate.c: In function ‘disasm__purge’: linux-next/tools/include/linux/compiler.h:66: error: dereferencing pointer ‘res.41’ does break strict-aliasing rules The reason is READ_ONCE/WRITE_ONCE code we took from kernel sources. They intentionaly break aliasing rules. While this is ok for kernel because it's built with -fno-strict-aliasing, it breaks perf which is build with -Wstrict-aliasing=3. Using extra __may_alias__ type to allow aliasing in this case. Reported-and-tested-by:
Vinson Lee <vlee@twopensource.com> Signed-off-by:
-
Namhyung Kim authored
With horizontal scrolling, the left/right arrow keys are used to scroll columns and ENTER/ESC keys are used to enter/exit menu. However if callchain is recorded, the ENTER key is used to toggle callchain expansion so there's no way to display menu. Use 'm' key to display the menu for this case. Signed-off-by:
-
Rabin Vincent authored
unw_word_t is uint64_t even on 32-bit MIPS. Cast it to uintptr_t before the cast to void *p to get rid of the following errors: util/unwind-libunwind.c: In function 'access_mem': util/unwind-libunwind.c:464:4: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] util/unwind-libunwind.c:475:2: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] cc1: all warnings being treated as errors make[3]: *** [util/unwind-libunwind.o] Error 1 Signed-off-by:
Rabin Vincent <rabin.vincent@axis.com> Acked-by:
-
Rabin Vincent authored
When NO_LIBUNWIND_DEBUG_FRAME=0, use the .debug_frame if the .eh_frame doesn't contain the approprate unwind tables. Signed-off-by:
-
- 12 Oct, 2015 3 commits
-
-
Arnaldo Carvalho de Melo authored
When in the hists browser, i.e. in 'perf report' or in 'perf top', it is possible to press '/' and specify a substring to filter by symbol name. Clarify how to remove a filter by making the prompt be: Please enter the name of symbol you want to see. To remove the filter later, press / + ENTER Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-vbq2b0kyufwy6p0ctkfswcoe@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
They were repurposed for horizontal scrolling, so use just ENTER/ESC in the help messages. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Wang Nan <wangnan0@huawei.com> Fixes: c6c3c02d ("perf hists browser: Implement horizontal scrolling") Link: http://lkml.kernel.org/n/tip-n5ar4qg8fs12ax4vhr3rxhxj@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Not as the first attempt at finding a vmlinux for the running kernel, this way we get a more informative filename to present in tools, it will check that the build-id is the same as the one previously loaded in the DSO in dso->build_id, reading from /sys/kernel/notes, for instance. E.g. in the annotation TUI, going from 'perf top', for the scsi_sg_alloc kernel function, in the first line: Before: scsi_sg_alloc /root/.debug/.build-id/28/2777c262e6b3c0451375163c9a81c893218ab1 After: scsi_sg_alloc /lib/modules/4.3.0-rc1+/build/vmlinux And: # ls -la /root/.debug/.build-id/28/2777c262e6b3c0451375163c9a81c893218ab1 lrwxrwxrwx. 1 root root 81 Sep 22 16:11 /root/.debug/.build-id/28/2777c262e6b3c0451375163c9a81c893218ab1 -> ../../home/git/build/v4.3.0-rc1+/vmlinux/282777c262e6b3c0451375163c9a81c893218ab1 # file ~/.debug/home/git/build/v4.3.0-rc1+/vmlinux/282777c262e6b3c0451375163c9a81c893218ab1 /root/.debug/home/git/build/v4.3.0-rc1+/vmlinux/282777c262e6b3c0451375163c9a81c893218ab1: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, BuildID[sha1]=282777c262e6b3c0451375163c9a81c893218ab1, not stripped # The same as: # file /lib/modules/4.3.0-rc1+/build/vmlinux /lib/modules/4.3.0-rc1+/build/vmlinux: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, BuildID[sha1]=282777c262e6b3c0451375163c9a81c893218ab1, not stripped Furthermore: # sha256sum /lib/modules/4.3.0-rc1+/build/vmlinux e7a789bbdc61029ec09140c228e1dd651271f38ef0b8416c0b7d5ff727b98be2 /lib/modules/4.3.0-rc1+/build/vmlinux # sha256sum ~/.debug/home/git/build/v4.3.0-rc1+/vmlinux/282777c262e6b3c0451375163c9a81c893218ab1 e7a789bbdc61029ec09140c228e1dd651271f38ef0b8416c0b7d5ff727b98be2 /root/.debug/home/git/build/v4.3.0-rc1+/vmlinux/282777c262e6b3c0451375163c9a81c893218ab1 [root@zoo new]# Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9y42ikzq3jisiddoi6f07n8z@git.kernel.orgSigned-off-by:
-
- 08 Oct, 2015 2 commits
-
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo: User visible changes: - Adding a field via 'perf report -F' that already is enabled makes the tool get stuck in a loop, fix it. (Jiri Olsa) Infrastructure changes: - Support PERF_RECORD_SWITCH in the python binding. (Arnaldo Carvalho de Melo) - Fix handling read() result using a signed variable, found with Coccinelle. (Andrzej Hajda) Signed-off-by:
-
Ingo Molnar authored
Signed-off-by:
-
