- 15 Jul, 2016 20 commits
-
-
Arnaldo Carvalho de Melo authored
So that it can find asm/bitsperlong.h to get the __BITS_PER_LONG definition. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-pr3pvskh65pey4po7t122z4j@git.kernel.orgSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Wang Nan authored
When working with overwritable ring buffer there's a inconvenience problem: if perf dumps data after a long period after it starts, non-sample events may lost, which makes following 'perf report' unable to identify proc name and mmap layout. For example: # perf record -m 4 -e raw_syscalls:* -g --overwrite --switch-output \ dd if=/dev/zero of=/dev/null send SIGUSR2 after dd runs long enough. The resuling perf.data lost correct comm and mmap events: # perf script -i perf.data.2016061522374354 perf 24478 [004] 2581325.601789: raw_syscalls:sys_exit: NR 0 = 512 ^^^^ Should be 'dd' 27b2e8 syscall_slow_exit_work+0xfe2000e3 (/lib/modules/4.6.0-rc3+/build/vmlinux) 203cc7 do_syscall_64+0xfe200117 (/lib/modules/4.6.0-rc3+/build/vmlinux) b18d83 return_from_SYSCALL_64+0xfe200000 (/lib/modules/4.6.0-rc3+/build/vmlinux) 7f47c417edf0 [unknown] ([unknown]) ^^^^^^^^^^^^ Fail to unwind This patch provides a '--tail-synthesize' option, allows perf to collect system status when finalizing output file. In resuling output file, the non-sample events reflect system status when dumping data. After this patch: # perf record -m 4 -e raw_syscalls:* -g --overwrite --switch-output --tail-synthesize \ dd if=/dev/zero of=/dev/null # perf script -i perf.data.2016061600544998 dd 27364 [004] 2583244.994464: raw_syscalls:sys_enter: NR 1 (1, ... ^^ Correct comm 203a18 syscall_trace_enter_phase2+0xfe2001a8 ([kernel.kallsyms]) 203aa5 syscall_trace_enter+0xfe200055 ([kernel.kallsyms]) 203caa do_syscall_64+0xfe2000fa ([kernel.kallsyms]) b18d83 return_from_SYSCALL_64+0xfe200000 ([kernel.kallsyms]) d8e50 __GI___libc_write+0xffff01d9639f4010 (/tmp/oxygen_root-w00229757/lib64/libc-2.18.so) ^^^^^ Correct unwind This option doesn't aim to solve this problem completely. If a process terminates before SIGUSR2, we still lost its COMM and MMAP events. For example, we can't unwind correctly from the final perf.data we get from the previous example, because when perf collects the final output file (when we press C-c), 'dd' has been terminated so its '/proc/<pid>/mmap' becomes empty. However, this is a cheaper choice. To completely solve this problem we need to continously output non-sample events. To satisify the requirement of daemonization, we need to merge them periodically. It is possible but requires much more code and cycles. Automatically select --tail-synthesize when --overwrite is provided. Signed-off-by:Wang Nan <wangnan0@huawei.com> Cc: He Kuang <hekuang@huawei.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Nilay Vaish <nilayvaish@gmail.com> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1468485287-33422-16-git-send-email-wangnan0@huawei.comSigned-off-by:
-
Wang Nan authored
If write_backward attribute is set, records are written into kernel ring buffer from end to beginning, but read from beginning to end. To avoid 'XX out of order events recorded' warning message (timestamps of records is in reverse order when using write_backward), suppress the warning message if write_backward is selected by at lease one event. Result: Before this patch: # perf record -m 1 -e raw_syscalls:sys_exit/overwrite/ \ -e raw_syscalls:sys_enter \ dd if=/dev/zero of=/dev/null count=300 300+0 records in 300+0 records out 153600 bytes (154 kB) copied, 0.000601617 s, 255 MB/s [ perf record: Woken up 5 times to write data ] Warning: 40 out of order events recorded. [ perf record: Captured and wrote 0.096 MB perf.data (696 samples) ] After this patch: # perf record -m 1 -e raw_syscalls:sys_exit/overwrite/ \ -e raw_syscalls:sys_enter \ dd if=/dev/zero of=/dev/null count=300 300+0 records in 300+0 records out 153600 bytes (154 kB) copied, 0.000644873 s, 238 MB/s [ perf record: Woken up 5 times to write data ] [ perf record: Captured and wrote 0.096 MB perf.data (696 samples) ] Signed-off-by:Jiri Olsa <jolsa@kernel.org> Tested-by:
He Kuang <hekuang@huawei.com> Signed-off-by:
-
Wang Nan authored
This patch allows following config terms and option: Globally setting events to overwrite; # perf record --overwrite ... Set specific events to be overwrite or no-overwrite. # perf record --event cycles/overwrite/ ... # perf record --event cycles/no-overwrite/ ... Add missing config terms and update the config term array size because the longest string length has changed. For overwritable events, it automatically selects attr.write_backward since perf requires it to be backward for reading. Test result: # perf record --overwrite -e syscalls:*enter_nanosleep* usleep 1 [ perf record: Woken up 2 times to write data ] [ perf record: Captured and wrote 0.011 MB perf.data (1 samples) ] # perf evlist -v syscalls:sys_enter_nanosleep: type: 2, size: 112, config: 0x134, { sample_period, sample_freq }: 1, sample_type: IP|TID|TIME|CPU|PERIOD|RAW, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, write_backward: 1 # Tip: use 'perf evlist --trace-fields' to show fields for tracepoint events Signed-off-by: -
Wang Nan authored
There's no user of these two function outside evlist.c. Remove them from public namespace. Signed-off-by:
-
Wang Nan authored
Drive the evlist->bkw_mmap_state state machine during draining and when SIGUSR2 is received. Read the backward ring buffer in record__mmap_read_all. Signed-off-by:
-
Wang Nan authored
Introduce a bkw_mmap_state state machine to evlist: .________________(forbid)_____________. | V NOTREADY --(0)--> RUNNING --(1)--> DATA_PENDING --(2)--> EMPTY ^ ^ | ^ | | |__(forbid)____/ |___(forbid)___/| | | \_________________(3)_______________/ NOTREADY : Backward ring buffers are not ready RUNNING : Backward ring buffers are recording DATA_PENDING : We are required to collect data from backward ring buffers EMPTY : We have collected data from backward ring buffers. (0): Setup backward ring buffer (1): Pause ring buffers for reading (2): Read from ring buffers (3): Resume ring buffers for recording We can't avoid this complexity. Since we deliberately drop records from overwritable ring buffer, there's no way for us to check remaining from ring buffer itself (by checking head and old pointers). Therefore, we need DATA_PENDING and EMPTY state to help us recording what we have done to the ring buffer. In record__mmap_read_evlist(), drive this state machine from DATA_PENDING to EMPTY. In perf_evlist__mmap_per_evsel(), drive this state machine from NOTREADY to RUNNING when creating backward mmap. Signed-off-by: -
Wang Nan authored
Now there's no real user of evlist->backward. Drop it. We are going to use evlist->backward_mmap as a container for backward ring buffer. Signed-off-by:
-
Wang Nan authored
In perf_evlist__mmap_per_evsel(), select backward_mmap for backward events. Utilize new perf_mmap APIs. Dynamically alloc backward_mmap. Remove useless functions. Signed-off-by:
-
Wang Nan authored
Add backward_mmap to evlist, free it together with normal mmap. Improve perf_evlist__pick_pc(), search backward_mmap if evlist->mmap is not available. This patch doesn't alloc this array. It will be allocated conditionally in the following commits. Signed-off-by:
-
Wang Nan authored
In perf_evlist__mmap_per_cpu() and perf_evlist__mmap_per_thread(), in case of mmap failure, successfully created maps should be cleared. Current code uses two loops calling __perf_evlist__munmap() for each function. This patch extracts common code to perf_evlist__munmap_nofree() and use previous introduced decoupled API perf_mmap__munmap(). Now __perf_evlist__munmap() can be removed because of no user. Signed-off-by:
-
Wang Nan authored
Insetad of saving a index into fdarray entries private field, save the corresponding 'struct perf_mmap' pointer, and release them directly using perf_mmap__put(). Following commits introduce multiple mmap arrays to evlist. Without this patch, perf_evlist__munmap_filtered() is unable to retrive correct 'struct perf_mmap' pointer. Signed-off-by:
-
Wang Nan authored
Perf evlist will have multiple mmap arrays. Update record__mmap_read(): it should read from 'struct perf_mmap' directly. Also, make record__mmap_read() ready to read from backward ring buffer. Signed-off-by:
-
Wang Nan authored
Currently, the evlist mmap related helpers and APIs accept evlist and idx, and dereference 'struct perf_mmap' by evlist->mmap[idx]. This is unnecessary, and force each evlist contains only one mmap array. Following commits are going to introduce multiple mmap arrays to a evlist. This patch refators these APIs and helpers, introduces functions accept perf_mmap pointer directly. New helpers and APIs are decoupled with perf_evlist, and become perf_mmap functions (so they have perf_mmap prefix). Old functions are reimplemented with new functions. Some of them will be removed in following commits. Signed-off-by:
-
Wang Nan authored
Add a 'ptr' field to fdarray->priv array. This feature will be used by following commits, which introduce muiltiple 'struct perf_mmap' arrays for different types of mapping. Because of this, during fdarray__filter(), a simple 'idx' is not enough. Add a pointer cookie that allows to directly associate a 'struct perf_mmap' pointer to an fdarray entry. Signed-off-by:
-
Peter Zijlstra authored
Do it using (__CHAR_BIT__ * __SIZEOF_LONG__), simpler, works everywhere, reduces the complexity by ditching CONFIG_64BIT, that was being synthesized from yet another set of defines, which proved fragile, breaking the build on linux-next for no obvious reasons. Committer Note: Except on: gcc version 4.1.2 20080704 (Red Hat 4.1.2-55) Fallback to __WORDSIZE in that case... Reported-by:
Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by:
Peter Zijlstra <peterz@infradead.org> Tested-by:
-
Arnaldo Carvalho de Melo authored
evsel->overwrite indicator means an event should be put into overwritable ring buffer. In current implementation, it equals to evsel->attr.write_backward. To reduce compliexity, remove evsel->overwrite, use evsel->attr.write_backward instead. In addition, in __perf_evsel__open(), if kernel doesn't support write_backward and user explicitly set it in evsel, don't fallback like other missing feature, since it is meaningless to fall back to a forward ring buffer in this case: we are unable to stably read from an forward overwritable ring buffer. Cc: He Kuang <hekuang@huawei.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Nilay Vaish <nilayvaish@gmail.com> Cc: Wang Nan <wangnan0@huawei.com> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1468485287-33422-2-git-send-email-wangnan0@huawei.comSigned-off-by:
-
Jiri Olsa authored
By using 0 for base, the strtoull() detects the base automatically (see 'man strtoull'). ATM we have just one user of this function, the cpu__get_max_freq function reading the "cpuinfo_max_freq" sysfs file. It should not get affected by this change. Committer note: This change seems motivated by this discussion: "[PATCH] [RFC V1]s390/perf: fix 'start' address of module's map" http://lkml.kernel.org/r/20160711120155.GA29929@krava I.e. this patches paves the way for filename__read_ull() to be used in a S/390 related fix. Signed-off-by:
-
Arnaldo Carvalho de Melo authored
There are cases where further work would be needed to overcome the fact that neither sysconf(_SC_LEVEL1_DCACHE_LINESIZE) nor /sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size are available in some systems (Android, for instance), so bail out when such a situation takes place. 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: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-ho8d8g8mh0o2dri7ckcccafi@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
So far the cacheline_size is only useful for the "dcacheline" --sort order, i.e. if that is not used, which is the norm, then the user shouldn't care that he is running this, say, on an Android system where sysconf(_SC_LEVEL1_DCACHE_LINESIZE) and the /sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size sysfs file isn't available. An upcoming patch will emit an warning only for "--sort ...,dcacheline,...". Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Chris Phlipot <cphlipot0@gmail.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-580cnkvftunyvt9n7unsholi@git.kernel.orgSigned-off-by:
-
- 14 Jul, 2016 20 commits
-
-
Arnaldo Carvalho de Melo authored
As the gcc there is producing tons of: "warning: always_inline function might not be inlinable" At least on android-ndk-r12/platforms/android-24/arch-arm, so, for the time being, use this big hammer. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Chris Phlipot <cphlipot0@gmail.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-97l3eg3fnk5shmo4rsyyvj2t@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
The Bionic libc has this definition, so don't duplicate it. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Chris Phlipot <cphlipot0@gmail.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-rmd19832zkt07e4crdzyen9z@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
We need to include netinet/in.h to get the in6_addr struct definition, needed to build it on the Android NDK: In file included from event-parse.c:36:0: /home/acme/android/android-ndk-r12/platforms/android-24/arch-arm/usr/include/netinet/ip6.h:82:18: error: field 'ip6_src' has incomplete type struct in6_addr ip6_src; /* source address */ And it is the canonical way of getting IPv6 definitions, as described, for instance, in Linux's 'man ipv6' Doing that uncovers another problem: this source file uses PRIu64 but doesn't include it, depending on it being included by chance via the now replaced header (netinet/ip6.h), fix it. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Chris Phlipot <cphlipot0@gmail.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-tilr31n3yaba1whsd47qlwa3@git.kernel.orgSigned-off-by: -
Ingo Molnar authored
Merge tag 'perf-core-for-mingo-20160713' 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: - Finish merging initial SDT (Statically Defined Traces) support, see cset comments for details about how it all works (Masami Hiramatsu) - Support attaching eBPF programs to tracepoints (Wang Nan) Infrastructure changes: - Fix up BITS_PER_LONG setting (Arnaldo Carvalho de Melo) - Add fallback from ELF_C_READ_MMAP to ELF_C_READ in objtool, fixing the build in libelf implementations lacking that elf_begin() cmd, such as Alpine Linux's (Arnaldo Carvalho de Melo) - Avoid checking code drift on busybox's diff in objtool (Arnaldo Carvalho de Melo) Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
Masami Hiramatsu authored
Add a basic test case for SDT event support. This test scans an SDT event in perftools and check whether the SDT event is correctly stored into the buildid cache. Here is an example: ---- $ perf test sdt -v 47: Test SDT event probing : --- start --- test child forked, pid 20732 Found 72 SDTs in /home/mhiramat/ksrc/linux/tools/perf/perf Writing cache: %sdt_perf:test_target=test_target Cache committed: 0 symbol:test_target file:(null) line:0 offset:0 return:0 lazy:(null) test child finished with 0 ---- end ---- Test SDT event probing: Ok ---- Signed-off-by:
Masami Hiramatsu <mhiramat@kernel.org> Tested-by:
-
Masami Hiramatsu authored
This checks whether sys/sdt.h is available or not, which is required for DTRACE_PROBE(). We can disable this feature by passing NO_SDT=1 when building. This flag will be used for SDT test case and further SDT events in perftools. Signed-off-by:
-
Masami Hiramatsu authored
Support a special SDT probe format which can omit the '%' prefix only if the SDT group name starts with "sdt_". So, for example both of "%sdt_libc:setjump" and "sdt_libc:setjump" are acceptable for perf probe --add. E.g. without this: # perf probe -a sdt_libc:setjmp Semantic error :There is non-digit char in line number. ... With this: # perf probe -a sdt_libc:setjmp Added new event: sdt_libc:setjmp (on %setjmp in /usr/lib64/libc-2.20.so) You can now use it in all perf tools, such as: perf record -e sdt_libc:setjmp -aR sleep 1 Suggested-by:Brendan Gregg <brendan.d.gregg@gmail.com> Signed-off-by:
-
Masami Hiramatsu authored
Support @BUILDID or @FILE suffix for SDT events. This allows perf to add probes on SDTs/pre-cached events on given FILE or the file which has given BUILDID (also, this complements BUILDID.) For example, both gcc and libstdc++ has same SDTs as below. If you would like to add a probe on sdt_libstdcxx:catch on gcc, you can do as below. ---- # perf list sdt | tail -n 6 sdt_libstdcxx:catch@/usr/bin/gcc(0cc207fc4b27) [SDT event] sdt_libstdcxx:catch@/usr/lib64/libstdc++.so.6.0.20(91c7a88fdf49) sdt_libstdcxx:rethrow@/usr/bin/gcc(0cc207fc4b27) [SDT event] sdt_libstdcxx:rethrow@/usr/lib64/libstdc++.so.6.0.20(91c7a88fdf49) sdt_libstdcxx:throw@/usr/bin/gcc(0cc207fc4b27) [SDT event] sdt_libstdcxx:throw@/usr/lib64/libstdc++.so.6.0.20(91c7a88fdf49) # perf probe -a %sdt_libstdcxx:catch@0cc Added new event: sdt_libstdcxx:catch (on %catch in /usr/bin/gcc) You can now use it in all perf tools, such as: perf record -e sdt_libstdcxx:catch -aR sleep 1 ---- Committer note: Doing the full sequence of steps to get the results above: With a clean build-id cache: [root@jouet ~]# rm -rf ~/.debug/ [root@jouet ~]# perf list sdt List of pre-defined events (to be used in -e): [root@jouet ~]# No events whatsoever, then, we can add all events in gcc to the build-id cache, doing a --add + --dry-run: [root@jouet ~]# perf probe --dry-run --cache -x /usr/bin/gcc --add %sdt_libstdcxx:\* Added new events: sdt_libstdcxx:throw (on %* in /usr/bin/gcc) sdt_libstdcxx:rethrow (on %* in /usr/bin/gcc) sdt_libstdcxx:catch (on %* in /usr/bin/gcc) You can now use it in all perf tools, such as: perf record -e sdt_libstdcxx:catch -aR sleep 1 [root@jouet ~]# It really didn't add any events, it just cached them: [root@jouet ~]# perf probe -l [root@jouet ~]# We can see that it was cached as: [root@jouet ~]# ls -la ~/.debug/usr/bin/gcc/9a0730e2bcc6d2a2003d21ac46807e8ee6bcb7c2/ total 976 drwxr-xr-x. 2 root root 4096 Jul 13 21:47 . drwxr-xr-x. 3 root root 4096 Jul 13 21:47 .. -rwxr-xr-x. 4 root root 985912 Jun 22 18:52 elf -rw-r--r--. 1 root root 303 Jul 13 21:47 probes [root@jouet ~]# file ~/.debug/usr/bin/gcc/9a0730e2bcc6d2a2003d21ac46807e8ee6bcb7c2/elf /root/.debug/usr/bin/gcc/9a0730e2bcc6d2a2003d21ac46807e8ee6bcb7c2/elf: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=9a0730e2bcc6d2a2003d21ac46807e8ee6bcb7c2, stripped [root@jouet ~]# cat ~/.debug/usr/bin/gcc/9a0730e2bcc6d2a2003d21ac46807e8ee6bcb7c2/probes %sdt_libstdcxx:throw=throw p:sdt_libstdcxx/throw /usr/bin/gcc:0x71ffd %sdt_libstdcxx:rethrow=rethrow p:sdt_libstdcxx/rethrow /usr/bin/gcc:0x720b8 %sdt_libstdcxx:catch=catch p:sdt_libstdcxx/catch /usr/bin/gcc:0x7307f %sdt_libgcc:unwind=unwind p:sdt_libgcc/unwind /usr/bin/gcc:0x7eec0 #sdt_libstdcxx:*=%* [root@jouet ~]# Ok, now we can use 'perf probe' to refer to those cached entries as: Humm, nope, doing as above we end up with: [root@jouet ~]# perf probe -a %sdt_libstdcxx:catch Semantic error :* is bad for event name -it must follow C symbol-naming rule. Error: Failed to add events. [root@jouet ~]# But it worked at some point, lets try not using --dry-run: Resetting everything: # rm -rf ~/.debug/ # perf probe -d *:* # perf probe -l # perf list sdt List of pre-defined events (to be used in -e): # Ok, now it cached everything, even things we haven't asked it to (sdt_libgcc:unwind): [root@jouet ~]# perf probe -x /usr/bin/gcc --add %sdt_libstdcxx:\* Added new events: sdt_libstdcxx:throw (on %* in /usr/bin/gcc) sdt_libstdcxx:rethrow (on %* in /usr/bin/gcc) sdt_libstdcxx:catch (on %* in /usr/bin/gcc) You can now use it in all perf tools, such as: perf record -e sdt_libstdcxx:catch -aR sleep 1 [root@jouet ~]# perf list sdt List of pre-defined events (to be used in -e): sdt_libgcc:unwind [SDT event] sdt_libstdcxx:catch [SDT event] sdt_libstdcxx:rethrow [SDT event] sdt_libstdcxx:throw [SDT event] [root@jouet ~]# And we have the events in place: [root@jouet ~]# perf probe -l sdt_libstdcxx:catch (on execute_cfa_program+1551@../../../libgcc/unwind-dw2.c in /usr/bin/gcc) sdt_libstdcxx:rethrow (on d_print_subexpr+280@libsupc++/cp-demangle.c in /usr/bin/gcc) sdt_libstdcxx:throw (on d_print_subexpr+93@libsupc++/cp-demangle.c in /usr/bin/gcc) [root@jouet ~]# And trying to use them at least has 'perf trace --event sdt*:*' working. Then, if we try to add the ones in libstdc++: [root@jouet ~]# perf probe -x /usr/lib64/libstdc++.so.6 -a %sdt_libstdcxx:\* Error: event "catch" already exists. Hint: Remove existing event by 'perf probe -d' or force duplicates by 'perf probe -f' or set 'force=yes' in BPF source. Error: Failed to add events. [root@jouet ~]# Doesn't work, dups, but at least this served to, unbeknownst to the user, add the SDT probes in /usr/lib64/libstdc++.so.6! [root@jouet ~]# perf list sdt List of pre-defined events (to be used in -e): sdt_libgcc:unwind [SDT event] sdt_libstdcxx:catch@/usr/bin/gcc(9a0730e2bcc6) [SDT event] sdt_libstdcxx:catch@/usr/lib64/libstdc++.so.6.0.22(ef2b7066559a) [SDT event] sdt_libstdcxx:rethrow@/usr/bin/gcc(9a0730e2bcc6) [SDT event] sdt_libstdcxx:rethrow@/usr/lib64/libstdc++.so.6.0.22(ef2b7066559a) [SDT event] sdt_libstdcxx:throw@/usr/bin/gcc(9a0730e2bcc6) [SDT event] sdt_libstdcxx:throw@/usr/lib64/libstdc++.so.6.0.22(ef2b7066559a) [SDT event] [root@jouet ~]# Now we should be able to get to the original cset comment, if we remove all SDTs events in place, not from the cache, from the kernel, where it was set up as: [root@jouet ~]# ls -la /sys/kernel/debug/tracing/events/sdt_libstdcxx/ total 0 drwxr-xr-x. 5 root root 0 Jul 13 22:00 . drwxr-xr-x. 80 root root 0 Jul 13 21:56 .. drwxr-xr-x. 2 root root 0 Jul 13 22:00 catch -rw-r--r--. 1 root root 0 Jul 13 22:00 enable -rw-r--r--. 1 root root 0 Jul 13 22:00 filter drwxr-xr-x. 2 root root 0 Jul 13 22:00 rethrow drwxr-xr-x. 2 root root 0 Jul 13 22:00 throw [root@jouet ~]# [root@jouet ~]# head -2 /sys/kernel/debug/tracing/events/sdt_libstdcxx/throw/format name: throw ID: 2059 [root@jouet ~]# Now to remove it: [root@jouet ~]# perf probe -d sdt_libstdc*:* Removed event: sdt_libstdcxx:catch Removed event: sdt_libstdcxx:rethrow Removed event: sdt_libstdcxx:throw [root@jouet ~]# Which caused: [root@jouet ~]# ls -la /sys/kernel/debug/tracing/events/sdt_libstdcxx/ ls: cannot access '/sys/kernel/debug/tracing/events/sdt_libstdcxx/': No such file or directory [root@jouet ~]# Ok, now we can do: [root@jouet ~]# perf list sdt_libstdcxx:catch List of pre-defined events (to be used in -e): sdt_libstdcxx:catch@/usr/bin/gcc(9a0730e2bcc6) [SDT event] sdt_libstdcxx:catch@/usr/lib64/libstdc++.so.6.0.22(ef2b7066559a) [SDT event] [root@jouet ~]# So, these are not really 'pre-defined events', i.e. we can't use them with 'perf record --event': [root@jouet ~]# perf record --event sdt_libstdcxx:catch* event syntax error: 'sdt_libstdcxx:catch*' \___ unknown tracepoint Error: File /sys/kernel/debug/tracing/events/sdt_libstdcxx/catch* not found. Hint: Perhaps this kernel misses some CONFIG_ setting to enable this feature?. <SNIP> [root@jouet ~]# To have it really pre-defined we must use perf probe to get its definition from the cache and set it up in the kernel, creating the tracepoint to _then_ use it with 'perf record --event': [root@jouet ~]# perf probe -a sdt_libstdcxx:catch Semantic error :There is non-digit char in line number. <SNIP> Oops, there is another gotcha here, we need that pesky '%' character: [root@jouet ~]# perf probe -a %sdt_libstdcxx:catch Added new events: sdt_libstdcxx:catch (on %catch in /usr/bin/gcc) sdt_libstdcxx:catch_1 (on %catch in /usr/lib64/libstdc++.so.6.0.22) You can now use it in all perf tools, such as: perf record -e sdt_libstdcxx:catch_1 -aR sleep 1 [root@jouet ~]# But then we added _two_ events, one with the name we expected, the other one with a _ added, when doing the analysis we need to pay attention to who maps to who. And here is where we get to the point of this patch, which is to be able to disambiguate those definitions for 'catch' in the build-id cache, but first we need remove those events we just added: [root@jouet ~]# perf probe -d %sdt_libstdcxx:catch Oops, that didn't remove anything, we need to _remove_ that % char in this case: [root@jouet ~]# perf probe -d sdt_libstdcxx:catch Removed event: sdt_libstdcxx:catch And we need to remove the other event added, i.e. I forgot to add a * at the end: [root@jouet ~]# perf probe -d sdt_libstdcxx:catch* Removed event: sdt_libstdcxx:catch_1 [root@jouet ~]# Ok, disambiguating it using what is in this patch: [root@jouet ~]# perf list sdt_libstdcxx:catch List of pre-defined events (to be used in -e): sdt_libstdcxx:catch@/usr/bin/gcc(9a0730e2bcc6) [SDT event] sdt_libstdcxx:catch@/usr/lib64/libstdc++.so.6.0.22(ef2b7066559a) [SDT event] [root@jouet ~]# [root@jouet ~]# perf probe -a %sdt_libstdcxx:catch@9a07 Added new event: sdt_libstdcxx:catch (on %catch in /usr/bin/gcc) You can now use it in all perf tools, such as: perf record -e sdt_libstdcxx:catch -aR sleep 1 [root@jouet ~]# perf probe -l sdt_libstdcxx:catch (on execute_cfa_program+1551@../../../libgcc/unwind-dw2.c in /usr/bin/gcc) [root@jouet ~]# Yeah, it works! But we need to try and simplify this :-) Update: Some aspects of this simplification take place in the following patches. Signed-off-by: -
Masami Hiramatsu authored
Show SDT and pre-cached events by perf-list with "sdt". This also shows the binary and build-id where the events are placed only when there are same name events on different binaries. e.g.: # perf list sdt List of pre-defined events (to be used in -e): sdt_libc:lll_futex_wake [SDT event] sdt_libc:lll_lock_wait_private [SDT event] sdt_libc:longjmp [SDT event] sdt_libc:longjmp_target [SDT event] ... sdt_libstdcxx:rethrow@/usr/bin/gcc(0cc207fc4b27) [SDT event] sdt_libstdcxx:rethrow@/usr/lib64/libstdc++.so.6.0.20(91c7a88fdf49) sdt_libstdcxx:throw@/usr/bin/gcc(0cc207fc4b27) [SDT event] sdt_libstdcxx:throw@/usr/lib64/libstdc++.so.6.0.20(91c7a88fdf49) The binary path and build-id are shown in below format; <GROUP>:<EVENT>@<PATH>(<BUILD-ID>) Signed-off-by:Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Tested-by:
-
Masami Hiramatsu authored
Search SDT/cached event from all probe caches if user doesn't pass any binary. With this, we don't have to specify target binary for SDT and named cached events (which start with %). E.g. without this, a target binary must be passed with -x. # perf probe -x /usr/lib64/libc-2.20.so -a %sdt_libc:\* With this change, we don't need it anymore. # perf probe -a %sdt_libc:\* Signed-off-by:
-
Masami Hiramatsu authored
Allo glob wildcard for reusing cached/SDT events. E.g. # perf probe -x /usr/lib64/libc-2.20.so -a %sdt_libc:\* This example adds probes for all SDT in libc. Note that the SDTs must have been scanned by perf buildid-cache. Committer note: Using it to check what of those SDT probes would take place when doing a cargo run (rust): # trace --no-sys --event sdt_libc:* cargo run 0.000 sdt_libc:setjmp:(7f326b69c4d1)) 28.423 sdt_libc:setjmp:(7f4b0a5364d1)) 29.000 sdt_libc:setjmp:(7f4b0a5364d1)) 88.597 sdt_libc:setjmp:(7fc01fd414d1)) 89.220 sdt_libc:setjmp:(7fc01fd414d1)) 95.501 sdt_libc:setjmp:(7f326b69c4d1)) Running `target/debug/hello_world` 97.110 sdt_libc:setjmp:(7f95e09234d1)) Hello, world! # Signed-off-by: -
Masami Hiramatsu authored
Add for_each_probe_cache_entry() wrapper macro for hiding list in probe_cache. Signed-off-by:
-
Masami Hiramatsu authored
Make "perf probe --cache --list" show only available cached events by checking build-id validity. E.g. without this patch: ---- $ ./perf probe --cache --add oldevent=cmd_probe $ make #(to update ./perf) $ ./perf probe --cache --add newevent=cmd_probe $ ./perf probe --cache --list /home/mhiramat/ksrc/linux/tools/perf/perf (061e90539bac69 probe_perf:newevent=cmd_probe /home/mhiramat/ksrc/linux/tools/perf/perf (c2e44d614e33e1 probe_perf:oldevent=cmd_probe ---- It shows both of old and new events but user can not use old one. With this; ---- $ ./perf probe --cache -l /home/mhiramat/ksrc/linux/tools/perf/perf (061e90539bac69 probe_perf:newevent=cmd_probe ---- This shows only new events which are on the existing binary. Signed-off-by:
-
Masami Hiramatsu authored
To improve usability, support %[PROVIDER:]SDTEVENT format to add new probes on SDT and cached events. e.g. ---- # perf probe -x /lib/libc-2.17.so %lll_lock_wait_private Added new event: sdt_libc:lll_lock_wait_private (on %lll_lock_wait_private in /usr/lib/libc-2.17.so) You can now use it in all perf tools, such as: perf record -e sdt_libc:lll_lock_wait_private -aR sleep 1 # perf probe -l | more sdt_libc:lll_lock_wait_private (on __lll_lock_wait_private+21 in /usr/lib/libc-2.17.so) ---- Note that this is not only for SDT events, but also normal events with event-name. e.g. define "myevent" on cache (-n doesn't add the real probe) ---- # perf probe -x ./perf --cache -n --add 'myevent=dso__load $params' ---- Reuse the "myevent" from cache as below. ---- # perf probe -x ./perf %myevent ---- Signed-off-by: -
Masami Hiramatsu authored
Fix to show correct error messages for $vars and $params because those special variables requires debug information to find the real variables or function parameters. E.g. without this fix; ---- # perf probe -x /lib64/libc-2.23.so getaddrinfo \$params Failed to write event: Invalid argument Please upgrade your kernel to at least 3.14 to have access to feature $params Error: Failed to add events. ---- Perf ends up with an error, but the message is not correct. With this fix, perf shows correct error message as below. ---- # perf probe -x /lib64/libc-2.23.so getaddrinfo \$params The /usr/lib64/libc-2.23.so file has no debug information. Rebuild with -g, or install an appropriate debuginfo package. Error: Failed to add events. ---- Reported-and-Tested-by:Arnaldo Carvalho de Melo <acme@kernel.org> Signed-off-by:
-
Wang Nan authored
To support 98b5c2c6 ("perf, bpf: allow bpf programs attach to tracepoints"), this patch allows BPF scripts to select tracepoints in their section name. Example: # cat test_tracepoint.c /*********************************************/ #include <uapi/linux/bpf.h> #define SEC(NAME) __attribute__((section(NAME), used)) SEC("raw_syscalls:sys_enter") int func(void *ctx) { /* * /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/format: * ... * field:long id; offset:8; size:8; signed:1; * ... * ctx + 8 select 'id' */ u64 id = *((u64 *)(ctx + 8)); if (id == 1) return 1; return 0; } SEC("_write=sys_write") int _write(void *ctx) { return 1; } char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /*********************************************/ # perf record -e ./test_tracepoint.c dd if=/dev/zero of=/dev/null count=5 5+0 records in 5+0 records out 2560 bytes (2.6 kB) copied, 6.2281e-05 s, 41.1 MB/s [ perf record: Woken up 1 times to write data ] # perf script dd 13436 [005] 1596.490869: raw_syscalls:sys_enter: NR 1 (1, 178d000, 200, 7ffe82470d60, ffffffffffffe020, fffff dd 13436 [005] 1596.490871: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490873: raw_syscalls:sys_enter: NR 1 (1, 178d000, 200, ffffffffffffe000, ffffffffffffe020, f dd 13436 [005] 1596.490874: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490876: raw_syscalls:sys_enter: NR 1 (1, 178d000, 200, ffffffffffffe000, ffffffffffffe020, f dd 13436 [005] 1596.490876: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490878: raw_syscalls:sys_enter: NR 1 (1, 178d000, 200, ffffffffffffe000, ffffffffffffe020, f dd 13436 [005] 1596.490879: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490881: raw_syscalls:sys_enter: NR 1 (1, 178d000, 200, ffffffffffffe000, ffffffffffffe020, f dd 13436 [005] 1596.490882: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490900: raw_syscalls:sys_enter: NR 1 (2, 7ffe8246e640, 1f, 40acb8, 7f44bac74700, 7f44baa4fba dd 13436 [005] 1596.490901: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490917: raw_syscalls:sys_enter: NR 1 (2, 7ffe8246e640, 1a, fffffffa, 7f44bac74700, 7f44baa4f dd 13436 [005] 1596.490918: perf_bpf_probe:_write: (ffffffff812351e0) dd 13436 [005] 1596.490932: raw_syscalls:sys_enter: NR 1 (2, 7ffe8246e640, 1a, fffffff9, 7f44bac74700, 7f44baa4f dd 13436 [005] 1596.490933: perf_bpf_probe:_write: (ffffffff812351e0) Committer note: Further testing: # trace --no-sys --event /home/acme/bpf/tracepoint.c cat /etc/passwd > /dev/null 0.000 raw_syscalls:sys_enter:NR 1 (1, 7f0490504000, c48, 7f0490503010, ffffffffffffffff, 0)) 0.006 perf_bpf_probe:_write:(ffffffff81241bc0)) # Signed-off-by:
-
Wang Nan authored
Following commit will allow BPF script attach to tracepoints. bpf__foreach_tev() will iterate over all events, not only kprobes. Rename it to bpf__foreach_event(). Since only group and event are used by caller, there's no need to pass full 'struct probe_trace_event' to bpf_prog_iter_callback_t. Pass only these two strings. After this patch bpf_prog_iter_callback_t natually support tracepoints. Signed-off-by:
-
Wang Nan authored
Add missing 'const' qualifiers so following commits are able to create tracepoints using const strings. Signed-off-by:
-
Wang Nan authored
Now libbpf support tracepoint program type. Report meaningful error when kernel version is less than 4.7. Signed-off-by:
-
Wang Nan authored
Add 4 new APIs to adjust and query the type of a BPF program. Load program according to type set by caller. Default is set to BPF_PROG_TYPE_KPROBE. Signed-off-by:
-
