Pull perf/core improvements from Arnaldo Carvalho de Melo:
User visible changes:
- Automagically create a 'bpf-output' event, easing the setup of BPF
C "scripts" that produce output via the perf ring buffer. Now it is
just a matter of calling any perf tool, such as 'trace', with a C
source file that references the __bpf_stdout__ output channel and
that channel will be created and connected to the script:
# trace -e nanosleep --event test_bpf_stdout.c usleep 1
0.013 ( 0.013 ms): usleep/2818 nanosleep(rqtp: 0x7ffcead45f40 ) ...
0.013 ( ): __bpf_stdout__:Raise a BPF event!..)
0.015 ( ): perf_bpf_probe:func_begin:(ffffffff81112460))
0.261 ( ): __bpf_stdout__:Raise a BPF event!..)
0.262 ( ): perf_bpf_probe:func_end:(ffffffff81112460 <- ffffffff81003d92))
0.264 ( 0.264 ms): usleep/2818 ... [continued]: nanosleep()) = 0
#
Further work is needed to reduce the number of lines in a perf bpf C source
file, this being the part where we greatly reduce the command line setup (Wang Nan)
- 'perf trace' now supports callchains, with 'trace --call-graph dwarf' using
libunwind, just like 'perf top', to ask the kernel for stack dumps for CFI
processing. This reduces the overhead by asking just for userspace callchains
and also only for the syscall exit tracepoint (raw_syscalls:sys_exit)
(Milian Wolff, Arnaldo Carvalho de Melo)
Try it with, for instance:
# perf trace --call dwarf ping 127.0.0.1
An excerpt of a system wide 'perf trace --call dwarf" session is at:
https://fedorapeople.org/~acme/perf/perf-trace--call-graph-dwarf--all-cpus.txt
You may need to bump the number of mmap pages, using -m/--mmap-pages,
but on a Broadwell machine the defaults allowed system wide tracing to
work without losing that many records, experiment with just some
syscalls, like:
# perf trace --call dwarf -e nanosleep,futex
All the targets available for 'perf record', 'perf top' (--pid, --tid, --cpu,
etc) should work. Also --duration may be interesting to try.
To get filenames from in various syscalls pointer args (open, ettc), add this
to the mix:
# perf probe 'vfs_getname=getname_flags:72 pathname=filename:string'
Making this work is next in line:
# trace --call dwarf --ev sched:sched_switch/call-graph=fp/ usleep 1
I.e. honouring per-tracepoint callchains in 'perf trace' in addition to
in raw_syscalls:sys_exit.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- Beautify more syscall arguments in 'perf trace', using the type column in
tracepoint /format fields to attach, for instance, a pid_t resolver to the
thread COMM, also attach a mode_t beautifier in the same fashion
(Arnaldo Carvalho de Melo)
- Build the syscall table id <-> name resolver using the same .tbl file
used in the kernel to generate headers, to avoid the delay in getting
new syscalls supported in the audit-libs external dependency, done so
far only for x86_64 (Arnaldo Carvalho de Melo)
- Improve the documentation of event specifications (Andi Kleen)
- Process update events in 'perf script', fixing up this use case:
# perf stat -a -I 1000 -e cycles record | perf script -s script.py
- Shared object symbol adjustment fixes, fixing symbol resolution in
Android (Wang Nan)
Infrastructure changes:
- Add dedicated unwind addr_space member into thread struct, to allow
tools to use thread->priv, noticed while working on having callchains
in 'perf trace' (Jiri Olsa)
Build fixes:
- Fix the build in Ubuntu 12.04 (Arnaldo Carvalho de Melo, Vinson Lee)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The fprintf_sym() and fprintf_callchain() methods now allow users to
change the existing behaviour of showing "[unknown]" as the name of
unresolved symbols to instead show "[0x123456]", i.e. its address.
The current patch doesn't change tools to use this facility, the results
from 'perf trace' and 'perf script' cotinue like:
70.109 ( 0.001 ms): qemu-system-x8/10153 poll(ufds: 0x7f2d93ffe870, nfds: 1) = 0 Timeout
[unknown] (/usr/lib64/libc-2.22.so)
[unknown] (/usr/lib64/libspice-server.so.1.10.0)
[unknown] (/usr/lib64/libspice-server.so.1.10.0)
[unknown] (/usr/lib64/libspice-server.so.1.10.0)
start_thread+0xca (/usr/lib64/libpthread-2.22.so)
__clone+0x6d (/usr/lib64/libc-2.22.so)
The next patch will make 'perf trace' use the new formatting.
Suggested-by: Milian Wolff <milian.wolff@kdab.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: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-fja1ods5vqpg42mdz09xcz3r@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch allows cloning bpf-output event configuration among multiple
bpf scripts. If there exist a map named '__bpf_output__' and not
configured using 'map:__bpf_output__.event=', this patch clones the
configuration of another '__bpf_stdout__' map. For example, following
command:
# perf trace --ev bpf-output/no-inherit,name=evt/ \
--ev ./test_bpf_trace.c/map:__bpf_stdout__.event=evt/ \
--ev ./test_bpf_trace2.c usleep 100000
equals to:
# perf trace --ev bpf-output/no-inherit,name=evt/ \
--ev ./test_bpf_trace.c/map:__bpf_stdout__.event=evt/ \
--ev ./test_bpf_trace2.c/map:__bpf_stdout__.event=evt/ \
usleep 100000
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1460128045-97310-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The readdir() function is thread safe as long as just one thread uses a
DIR, which is the case when parsing tracepoint event definitions, to
avoid breaking the build with glibc-2.23.90 (upcoming 2.24), use it
instead of readdir_r().
See: http://man7.org/linux/man-pages/man3/readdir.3.html
"However, in modern implementations (including the glibc implementation),
concurrent calls to readdir() that specify different directory streams
are thread-safe. In cases where multiple threads must read from the
same directory stream, using readdir() with external synchronization is
still preferable to the use of the deprecated readdir_r(3) function."
Noticed while building on a Fedora Rawhide docker container.
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-wddn49r6bz6wq4ee3dxbl7lo@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The readdir() function is thread safe as long as just one thread uses a
DIR, which is the case when synthesizing events for pre-existing threads
by traversing /proc, so, to avoid breaking the build with glibc-2.23.90
(upcoming 2.24), use it instead of readdir_r().
See: http://man7.org/linux/man-pages/man3/readdir.3.html
"However, in modern implementations (including the glibc implementation),
concurrent calls to readdir() that specify different directory streams
are thread-safe. In cases where multiple threads must read from the
same directory stream, using readdir() with external synchronization is
still preferable to the use of the deprecated readdir_r(3) function."
Noticed while building on a Fedora Rawhide docker container.
CC /tmp/build/perf/util/event.o
util/event.c: In function '__event__synthesize_thread':
util/event.c:466:2: error: 'readdir_r' is deprecated [-Werror=deprecated-declarations]
while (!readdir_r(tasks, &dirent, &next) && next) {
^~~~~
In file included from /usr/include/features.h:368:0,
from /usr/include/stdint.h:25,
from /usr/lib/gcc/x86_64-redhat-linux/6.0.0/include/stdint.h:9,
from /git/linux/tools/include/linux/types.h:6,
from util/event.c:1:
/usr/include/dirent.h:189:12: note: declared here
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-i1vj7nyjp2p750rirxgrfd3c@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The readdir() function is thread safe as long as just one thread uses a
DIR, which is the case in thread_map, so, to avoid breaking the build
with glibc-2.23.90 (upcoming 2.24), use it instead of readdir_r().
See: http://man7.org/linux/man-pages/man3/readdir.3.html
"However, in modern implementations (including the glibc implementation),
concurrent calls to readdir() that specify different directory streams
are thread-safe. In cases where multiple threads must read from the
same directory stream, using readdir() with external synchronization is
still preferable to the use of the deprecated readdir_r(3) function."
Noticed while building on a Fedora Rawhide docker container.
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-del8h2a0f40z75j4r42l96l0@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
He Kuang reported a problem that perf fails to get correct symbol on
Android platform in [1]. The problem can be reproduced on normal x86_64
platform. I will describe the reproducing steps in detail at the end of
commit message.
The reason of this problem is the missing of symbol adjustment for normal
shared objects. In most of the cases skipping adjustment is okay. However,
when '.text' section have different 'address' and 'offset' the result is wrong.
I checked all shared objects in my working platform, only wine dll objects and
debug objects (in .debug) have this problem. However, it is common on Android.
For example:
$ readelf -S ./libsurfaceflinger.so | grep \.text
[10] .text PROGBITS 0000000000029030 00012030
This patch enables symbol adjustment for dynamic objects so the symbol
address got from elfutils would be adjusted correctly.
Now nearly all types of ELF files should adjust symbols. Makes
ss->adjust_symbols default to true.
Steps to reproduce the problem:
$ cat ./Makefile
PWD := $(shell pwd)
LDFLAGS += "-Wl,-rpath=$(PWD)"
CFLAGS += -g
main: main.c libbuggy.so
libbuggy.so: buggy.c
gcc -g -shared -fPIC -Wl,-Ttext-segment=0x200000 $< -o $@
clean:
rm -rf main libbuggy.so *.o
$ cat ./buggy.c
int fib(int x)
{
return (x == 0) ? 1 : (x == 1) ? 1 : fib(x - 1) + fib(x - 2);
}
$ cat ./main.c
#include <stdio.h>
extern int fib(int x);
int main()
{
int i;
for (i = 0; i < 40; i++)
printf("%d\n", fib(i));
return 0;
}
$ make
$ perf record ./main
...
$ perf report --stdio
# Overhead Command Shared Object Symbol
# ........ ....... ................. ...............................
#
14.97% main libbuggy.so [.] 0x000000000000066c
8.68% main libbuggy.so [.] 0x00000000000006aa
8.52% main libbuggy.so [.] fib@plt
7.95% main libbuggy.so [.] 0x0000000000000664
5.94% main libbuggy.so [.] 0x00000000000006a9
5.35% main libbuggy.so [.] 0x0000000000000678
...
The correct result should be (after this patch):
# Overhead Command Shared Object Symbol
# ........ ....... ................. ...............................
#
91.47% main libbuggy.so [.] fib
8.52% main libbuggy.so [.] fib@plt
0.00% main [kernel.kallsyms] [k] kmem_cache_free
[1] http://lkml.kernel.org/g/1452567507-54013-1-git-send-email-hekuang@huawei.com
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Cody P Schafer <dev@codyps.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kirill Smelkov <kirr@nexedi.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1460024671-64774-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In this patch, the offset of '.text' section is stored into dso
and used here to re-calculate address to objdump.
In most of the cases, executable code is in '.text' section, so the
adjustment made to a symbol in dso__load_sym (using
sym.st_value -= shdr.sh_addr - shdr.sh_offset) should equal to
'sym.st_value -= dso->text_offset'. Therefore, adding text_offset back
get objdump address from symbol address (rip). However, it is not true
for kernel and kernel module since there could be multiple executable
sections with different offset. Exclude kernel for this reason.
After this patch, even dso->adjust_symbols is set to true for shared
objects, map__rip_2objdump() and map__objdump_2mem() would return
correct result, so perf behavior of annotate won't be changed.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Cody P Schafer <dev@codyps.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kirill Smelkov <kirr@nexedi.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1460024671-64774-2-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We used libaudit to map ids to syscall names and vice-versa, but that
imposes a delay in supporting new syscalls, having to wait for libaudit
to get those new syscalls on its tables.
To remove that delay, for x86_64 initially, grab a copy of
arch/x86/entry/syscalls/syscall_64.tbl and use it to generate those
tables.
Syscalls currently not available in audit-libs:
# trace -e copy_file_range,membarrier,mlock2,pread64,pwrite64,timerfd_create,userfaultfd
Error: Invalid syscall copy_file_range, membarrier, mlock2, pread64, pwrite64, timerfd_create, userfaultfd
Hint: try 'perf list syscalls:sys_enter_*'
Hint: and: 'man syscalls'
#
With this patch:
# trace -e copy_file_range,membarrier,mlock2,pread64,pwrite64,timerfd_create,userfaultfd
8505.733 ( 0.010 ms): gnome-shell/2519 timerfd_create(flags: 524288) = 36
8506.688 ( 0.005 ms): gnome-shell/2519 timerfd_create(flags: 524288) = 40
30023.097 ( 0.025 ms): qemu-system-x8/24629 pwrite64(fd: 18, buf: 0x7f63ae382000, count: 4096, pos: 529592320) = 4096
31268.712 ( 0.028 ms): qemu-system-x8/24629 pwrite64(fd: 18, buf: 0x7f63afd8b000, count: 4096, pos: 2314133504) = 4096
31268.854 ( 0.016 ms): qemu-system-x8/24629 pwrite64(fd: 18, buf: 0x7f63afda2000, count: 4096, pos: 2314137600) = 4096
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-51xfjbxevdsucmnbc4ka5r88@git.kernel.org
[ Added make dep for 'prepare' in 'LIBPERF_IN', fix by Wang Nan to fix parallell build ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tools should use a mechanism similar to arch/x86/entry/syscalls/ to
generate a header file with the definitions for two variables:
static const char *syscalltbl_x86_64[] = {
[0] = "read",
[1] = "write",
<SNIP>
[324] = "membarrier",
[325] = "mlock2",
[326] = "copy_file_range",
};
static const int syscalltbl_x86_64_max_id = 326;
In a per arch file that should then be included in
tools/perf/util/syscalltbl.c.
First one will be for x86_64.
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-02uuamkxgccczdth8komspgp@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We're using libaudit for doing name to id and id to syscall name
translations, but that makes 'perf trace' to have to wait for newer
libaudit versions supporting recently added syscalls, such as
"userfaultfd" at the time of this changeset.
We have all the information right there, in the kernel sources, so move
this code to a separate place, wrapped behind functions that will
progressively use the kernel source files to extract the syscall table
for use in 'perf trace'.
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-i38opd09ow25mmyrvfwnbvkj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This ended up triggering these warnings when building on Ubuntu 12.04.5:
util/scripting-engines/trace-event-perl.c: In function 'perl_process_callchain':
util/scripting-engines/trace-event-perl.c:293:4: error: value computed is not used [-Werror=unused-value]
util/scripting-engines/trace-event-perl.c:294:4: error: value computed is not used [-Werror=unused-value]
util/scripting-engines/trace-event-perl.c:295:4: error: value computed is not used [-Werror=unused-value]
util/scripting-engines/trace-event-perl.c:297:4: error: value computed is not used [-Werror=unused-value]
util/scripting-engines/trace-event-perl.c:309:4: error: value computed is not used [-Werror=unused-value]
cc1: all warnings being treated as errors
mv: cannot stat `/tmp/build/perf/util/scripting-engines/.trace-event-perl.o.tmp': No such file or directory
make[4]: *** [/tmp/build/perf/util/scripting-engines/trace-event-perl.o] Error 1
Fix it by doing error checking when building the perl data structures
related to callchains.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Dima Kogan <dima@secretsauce.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@gmail.com>
Fixes: f7380c12ec ("perf script perl: Perl scripts now get a backtrace, like the python ones")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently the max value of format is calculated by the bits number. It
relies on the continuity of the format.
However, uncore event format is not continuous. E.g. uncore qpi event
format can be 0-7,21.
If bit 21 is set, there is parsing issues as below.
$ perf stat -a -e uncore_qpi_0/event=0x200002,umask=0x8/
event syntax error: '..pi_0/event=0x200002,umask=0x8/'
\___ value too big for format, maximum is 511
This patch return the real max value by setting all possible bits to 1.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1459365375-14285-1-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Intel PT uses the time members from the perf_event_mmap_page to convert
between TSC and perf time.
Due to a lack of foresight when Intel PT was implemented, those time
members were recorded in the (implementation dependent) AUXTRACE_INFO
event, the structure of which is generally inaccessible outside of the
Intel PT decoder. However now the conversion between TSC and perf time
is needed when processing a jitdump file when Intel PT has been used for
tracing.
So add a user event to record the time members. 'perf record' will
synthesize the event if the information is available. And session
processing will put a copy of the event on the session so that tools
like 'perf inject' can easily access it.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1457426324-30158-1-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes:
User visible changes:
- Add support for skipping itrace instructions, useful to fast forward
processor trace (Intel PT, BTS) to right after initialization code at the start
of a workload (Andi Kleen)
- Add support for backtraces in perl 'perf script's (Dima Kogan)
- Add -U/-K (--all-user/--all-kernel) options to 'perf mem' (Jiri Olsa)
- Make -f/--force option documentation consistent across tools (Jiri Olsa)
Infrastructure changes:
- Add 'perf test' to check for event times (Jiri Olsa)
- 'perf config' cleanups (Taeung Song)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When using 'perf script' to look at PT traces it is often useful to
ignore the initialization code at the beginning.
On larger traces which may have many millions of instructions in
initialization code doing that in a pipeline can be very slow, with perf
script spending a lot of CPU time calling printf and writing data.
This patch adds an extension to the --itrace argument that skips 'n'
events (instructions, branches or transactions) at the beginning. This
is much more efficient.
v2:
Add support for BTS (Adrian Hunter)
Document in itrace.txt
Fix branch check
Check transactions and instructions too
Committer note:
To test intel_pt one needs to make sure VT-x isn't active, i.e.
stopping KVM guests on the test machine, as described by Andi Kleen
at http://lkml.kernel.org/r/20160301234953.GD23621@tassilo.jf.intel.com
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1459187142-20035-1-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We have some infrastructure to use perl or python to analyze logs
generated by perf. Prior to this patch, only the python tools had
access to backtrace information. This patch makes this information
available to perl scripts as well. Example:
Let's look at malloc() calls made by the seq utility. First we
create a probe point:
$ perf probe -x /lib/x86_64-linux-gnu/libc.so.6 malloc
Added new events:
...
Now we run seq, while monitoring malloc() calls with perf
$ perf record --call-graph=dwarf -e probe_libc:malloc seq 5
1
2
3
4
5
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.064 MB perf.data (6 samples) ]
We can use perf to look at its log to see the malloc calls and the backtrace
$ perf script
seq 14195 [000] 1927993.748254: probe_libc:malloc: (7f9ff8edd320) bytes=0x22
7f9ff8edd320 malloc (/lib/x86_64-linux-gnu/libc-2.22.so)
7f9ff8e8eab0 set_binding_values.part.0 (/lib/x86_64-linux-gnu/libc-2.22.so)
7f9ff8e8eda1 __bindtextdomain (/lib/x86_64-linux-gnu/libc-2.22.so)
401b22 main (/usr/bin/seq)
7f9ff8e82610 __libc_start_main (/lib/x86_64-linux-gnu/libc-2.22.so)
402799 _start (/usr/bin/seq)
...
We can also use the scripting facilities. We create a skeleton perl
script that simply prints out the events
$ perf script -g perl
generated Perl script: perf-script.pl
We can then use this script to see the malloc() calls with a
backtrace. Prior to this patch, the backtrace was not available to
the perl scripts.
$ perf script -s perf-script.pl
probe_libc::malloc 0 1927993.748254260 14195 seq __probe_ip=140325052863264, bytes=34
[7f9ff8edd320] malloc
[7f9ff8e8eab0] set_binding_values.part.0
[7f9ff8e8eda1] __bindtextdomain
[401b22] main
[7f9ff8e82610] __libc_start_main
[402799] _start
...
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/87mvphzld0.fsf@secretsauce.net
Signed-off-by: Dima Kogan <dima@secretsauce.net>
In 473398a21d ("perf tools: Add cpumode to struct perf_sample"), I
missed some places where perf_sample fields are directly initialized in
addition to what is done in perf_evsel__parse_sample(), namely when
synthesizing PERF_RECORD_{MMAP*,COMM,FORK,EXIT} for pre-existing threads
and also in intel_pt and intel_bts when synthesizing events from
processor trace, the jitdump code also was affected, fix it.
The problem was noticed with running:
# perf record -e intel_pt//u true
# perf script
Where the samples wouldn't get resolved because perf_sample.cpumode
would be left as zero, i.e. PERF_RECORD_MISC_CPUMODE_UNKNOWN, not
resolving as kernel, hypervisor or user cpu modes.
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 473398a21d ("perf tools: Add cpumode to struct perf_sample")
Link: http://lkml.kernel.org/n/tip-n5sdauxgk24d5nun8kuuu2mh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf fixes from Ingo Molnar:
"This tree contains various perf fixes on the kernel side, plus three
hw/event-enablement late additions:
- Intel Memory Bandwidth Monitoring events and handling
- the AMD Accumulated Power Mechanism reporting facility
- more IOMMU events
... and a final round of perf tooling updates/fixes"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (44 commits)
perf llvm: Use strerror_r instead of the thread unsafe strerror one
perf llvm: Use realpath to canonicalize paths
perf tools: Unexport some methods unused outside strbuf.c
perf probe: No need to use formatting strbuf method
perf help: Use asprintf instead of adhoc equivalents
perf tools: Remove unused perf_pathdup, xstrdup functions
perf tools: Do not include stringify.h from the kernel sources
tools include: Copy linux/stringify.h from the kernel
tools lib traceevent: Remove redundant CPU output
perf tools: Remove needless 'extern' from function prototypes
perf tools: Simplify die() mechanism
perf tools: Remove unused DIE_IF macro
perf script: Remove lots of unused arguments
perf thread: Rename perf_event__preprocess_sample_addr to thread__resolve
perf machine: Rename perf_event__preprocess_sample to machine__resolve
perf tools: Add cpumode to struct perf_sample
perf tests: Forward the perf_sample in the dwarf unwind test
perf tools: Remove misplaced __maybe_unused
perf list: Fix documentation of :ppp
perf bench numa: Fix assertion for nodes bitfield
...
