Pull perf updates from Ingo Molnar:
"Kernel side changes:
- x86 Intel uncore driver cleanups and enhancements (Kan Liang)
- group scheduling and other fixes (Song Liu
- store frame pointer in the sample traces for better profiling
(Alexey Budankov)
- compat fixes/enhancements (Eugene Syromiatnikov)
Tooling side changes, which you can build and install in a single step
via:
make -C tools/perf clean install
perf annotate:
- Support 'perf annotate --group' for non-explicit recorded event
"groups", showing multiple columns, one for each event, just like
when dealing with explicit event groups (those enclosed with {})
(Jin Yao)
- Record min/max LBR cycles (>= Skylake) and add 'perf annotate' TUI
hotkey to show it (c) (Jin Yao)
perf bpf:
- Add infrastructure to help in writing eBPF C programs to be used
with '-e name.c' type events in tools such as 'record' and 'trace',
with headers for common constructs and an examples directory that
will get populated as we add more such helpers and the 'perf bpf'
(Arnaldo Carvalho de Melo)
perf stat:
- Display time in precision based on std deviation (Jiri Olsa)
- Add --table option to display time of each run (Jiri Olsa)
- Display length strings of each run for --table option (Jiri Olsa)
perf buildid-cache:
- Add --list and --purge-all options (Ravi Bangoria)
perf test:
- Let 'perf test list' display subtests (Hendrik Brueckner)
perf pti:
- Create extra kernel maps to help in decoding samples in x86 PTI
entry trampolines (Adrian Hunter)
- Copy x86 PTI entry trampoline sections in the kcore copy used for
annotation and intel_pt CPU traces decoding (Adrian Hunter)
... and a lot of other fixes, enhancements and cleanups I did not
list, see the shortlog and git log for details"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (111 commits)
perf/x86/intel/uncore: Clean up client IMC uncore
perf/x86/intel/uncore: Expose uncore_pmu_event*() functions
perf/x86/intel/uncore: Support IIO free-running counters on SKX
perf/x86/intel/uncore: Add infrastructure for free running counters
perf/x86/intel/uncore: Add new data structures for free running counters
perf/x86/intel/uncore: Correct fixed counter index check in generic code
perf/x86/intel/uncore: Correct fixed counter index check for NHM
perf/x86/intel/uncore: Introduce customized event_read() for client IMC uncore
perf/x86: Store user space frame-pointer value on a sample
perf/core: Wire up compat PERF_EVENT_IOC_QUERY_BPF, PERF_EVENT_IOC_MODIFY_ATTRIBUTES
perf/core: Fix bad use of igrab()
perf/core: Fix group scheduling with mixed hw and sw events
perf kcore_copy: Amend the offset of sections that remap kernel text
perf kcore_copy: Copy x86 PTI entry trampoline sections
perf kcore_copy: Get rid of kernel_map
perf kcore_copy: Iterate phdrs
perf kcore_copy: Layout sections
perf kcore_copy: Calculate offset from phnum
perf kcore_copy: Keep a count of phdrs
perf kcore_copy: Keep phdr data in a list
...
Perf stat doesn't count the uncore event aliases from the same uncore
block in a group, for example:
perf stat -e '{unc_m_cas_count.all,unc_m_clockticks}' -a -I 1000
# time counts unit events
1.000447342 <not counted> unc_m_cas_count.all
1.000447342 <not counted> unc_m_clockticks
2.000740654 <not counted> unc_m_cas_count.all
2.000740654 <not counted> unc_m_clockticks
The output is very misleading. It gives a wrong impression that the
uncore event doesn't work.
An uncore block could be composed by several PMUs. An uncore event alias
is a joint name which means the same event runs on all PMUs of a block.
Perf doesn't support mixed events from different PMUs in the same group.
It is wrong to put uncore event aliases in a big group.
The right way is to split the big group into multiple small groups which
only include the events from the same PMU.
Only uncore event aliases from the same uncore block should be specially
handled here. It doesn't make sense to mix the uncore events with other
uncore events from different blocks or even core events in a group.
With the patch:
# time counts unit events
1.001557653 140,833 unc_m_cas_count.all
1.001557653 1,330,231,332 unc_m_clockticks
2.002709483 85,007 unc_m_cas_count.all
2.002709483 1,429,494,563 unc_m_clockticks
Reported-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Agustin Vega-Frias <agustinv@codeaurora.org>
Cc: Ganapatrao Kulkarni <ganapatrao.kulkarni@cavium.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Shaokun Zhang <zhangshaokun@hisilicon.com>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1525727623-19768-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When perf data is recorded with the call-graph option enabled, the
callchain shown by perf script shows the binary offsets of the symbols
as the ip. This is incorrect for kernel symbols as the ip values are
always off by a fixed offset depending on the architecture. If the
offsets from the start of the symbols are printed, they are also
incorrect for both kernel and userspace symbols.
Without the call-graph option, the callchain shows the virtual addresses
of the symbols rather than their binary offsets. The offsets printed in
this case are also correct.
This fixes the inconsistency in perf script's output.
This can be verified on a powerpc64le system running Fedora 27 as
follows:
# cat /proc/kallsyms | grep sys_write
...
c0000000004025a0 T sys_write
c0000000004025a0 T __se_sys_write
...
# perf probe -a sys_write
Before applying this patch:
# perf record -e probe:sys_write -g ~/test
# perf script -F ip,sym,symoff
4125b0 sys_write+0x8000000000008010
1b9e0 system_call+0x8000000000008058
118234 __GI___libc_write+0xffff0000f52c0024
92c74 _IO_file_write@@GLIBC_2.17+0xffff0000f52c0044
5afbfd8a [unknown]
91a60 new_do_write+0xffff0000f52c0090
94638 _IO_do_write@@GLIBC_2.17+0xffff0000f52c0038
94bbc _IO_file_overflow@@GLIBC_2.17+0xffff0000f52c014c
95a24 __overflow+0xffff0000f52c0064
84548 _IO_puts+0xffff0000f52c0218
440 main+0xffffffffe0000020
236a0 generic_start_main.isra.0+0xffff0000f52c0140
23898 __libc_start_main+0xffff0000f52c00b8
0 [unknown]
...
# perf record -e probe:sys_write ~/test
# perf script -F ip,sym,symoff
c0000000004025b0 sys_write+0x10
...
After applying this patch:
# perf record -e probe:sys_write -g ~/test
# perf script -F ip,sym,symoff
c0000000004025b0 sys_write+0x10
c00000000000b9e0 system_call+0x58
7fffb70d8234 __GI___libc_write+0x24
7fffb7052c74 _IO_file_write@@GLIBC_2.17+0x44
5afc1818 [unknown]
7fffb7051a60 new_do_write+0x90
7fffb7054638 _IO_do_write@@GLIBC_2.17+0x38
7fffb7054bbc _IO_file_overflow@@GLIBC_2.17+0x14c
7fffb7055a24 __overflow+0x64
7fffb7044548 _IO_puts+0x218
10000440 main+0x20
7fffb6fe36a0 generic_start_main.isra.0+0x140
7fffb6fe3898 __libc_start_main+0xb8
0 [unknown]
...
# perf record -e probe:sys_write ~/test
# perf script -F ip,sym,symoff
c0000000004025b0 sys_write+0x10
...
Signed-off-by: Sandipan Das <sandipan@linux.vnet.ibm.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Link: http://lkml.kernel.org/r/20180517063326.6319-1-sandipan@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf stat doesn't count the uncore event aliases from the same uncore
block in a group, for example:
perf stat -e '{unc_m_cas_count.all,unc_m_clockticks}' -a -I 1000
# time counts unit events
1.000447342 <not counted> unc_m_cas_count.all
1.000447342 <not counted> unc_m_clockticks
2.000740654 <not counted> unc_m_cas_count.all
2.000740654 <not counted> unc_m_clockticks
The output is very misleading. It gives a wrong impression that the
uncore event doesn't work.
An uncore block could be composed by several PMUs. An uncore event alias
is a joint name which means the same event runs on all PMUs of a block.
Perf doesn't support mixed events from different PMUs in the same group.
It is wrong to put uncore event aliases in a big group.
The right way is to split the big group into multiple small groups which
only include the events from the same PMU.
Only uncore event aliases from the same uncore block should be specially
handled here. It doesn't make sense to mix the uncore events with other
uncore events from different blocks or even core events in a group.
With the patch:
# time counts unit events
1.001557653 140,833 unc_m_cas_count.all
1.001557653 1,330,231,332 unc_m_clockticks
2.002709483 85,007 unc_m_cas_count.all
2.002709483 1,429,494,563 unc_m_clockticks
Reported-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Agustin Vega-Frias <agustinv@codeaurora.org>
Cc: Ganapatrao Kulkarni <ganapatrao.kulkarni@cavium.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Shaokun Zhang <zhangshaokun@hisilicon.com>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1525727623-19768-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This is not specific to BPF but was found when parsing a .c BPF proggie
that while valid, had no events attached to tracepoints, kprobes, etc:
Very minimal file that perf's BPF code can compile:
# cat empty.c
char _license[] __attribute__((section("license"), used)) = "GPL";
int _version __attribute__((section("version"), used)) = LINUX_VERSION_CODE;
#
Before this patch:
# perf trace -e empty.c
WARNING: event parser found nothinginvalid or unsupported event: 'empty.c'
Run 'perf list' for a list of valid events
Usage: perf trace [<options>] [<command>]
or: perf trace [<options>] -- <command> [<options>]
or: perf trace record [<options>] [<command>]
or: perf trace record [<options>] -- <command> [<options>]
-e, --event <event> event/syscall selector. use 'perf list' to list available events
#
After:
# perf trace -e empty.c
WARNING: event parser found nothing
invalid or unsupported event: 'empty.c'
Run 'perf list' for a list of valid events
Usage: perf trace [<options>] [<command>]
or: perf trace [<options>] -- <command> [<options>]
or: perf trace record [<options>] [<command>]
or: perf trace record [<options>] -- <command> [<options>]
-e, --event <event> event/syscall selector. use 'perf list' to list available events
#
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: https://lkml.kernel.org/n/tip-8ysughiz00h6mjpcot04qyjj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
CoreSight doesn't allocate thread structure for unknown_thread in ETM
auxtrace, so unknown_thread is NULL pointer. If the perf data doesn't
contain valid tid and then cs_etm__mem_access() uses unknown_thread
instead as thread handler, this results in a segmentation fault when
thread__find_addr_map() accesses the thread handler.
This commit creates a new thread data which is used by unknown_thread, so
CoreSight tracing can roll back to use unknown_thread if perf data
doesn't include valid thread info. This commit also releases thread
data for initialization failure case and for normal auxtrace free flow.
Signed-off-by: Leo Yan <leo.yan@linaro.org>
Acked-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1525924920-4381-1-git-send-email-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When we perform the following command lines:
$ perf record -e "{cycles,branches}" ./div
$ perf annotate main --stdio
The output shows only the first event, "cycles" and the displaying
format is not correct.
Percent | Source code & Disassembly of div for cycles (44550 samples)
-----------------------------------------------------------------------------------
:
:
:
: Disassembly of section .text:
:
: 00000000004004b0 <main>:
: main():
:
: return i;
: }
:
: int main(void)
: {
0.00 : 4004b0: push %rbx
: int i;
: int flag;
: volatile double x = 1212121212, y = 121212;
:
: s_randseed = time(0);
0.00 : 4004b1: xor %edi,%edi
: srand(s_randseed);
0.00 : 4004b3: mov $0x77359400,%ebx
:
: return i;
: }
The issue is that the value of the 'nr_percent' variable is hardcoded to
1. This patch fixes it.
With this patch, the output is:
Percent | Source code & Disassembly of div for cycles (44550 samples)
-----------------------------------------------------------------------------------
:
:
:
: Disassembly of section .text:
:
: 00000000004004b0 <main>:
: main():
:
: return i;
: }
:
: int main(void)
: {
0.00 0.00 : 4004b0: push %rbx
: int i;
: int flag;
: volatile double x = 1212121212, y = 121212;
:
: s_randseed = time(0);
0.00 0.00 : 4004b1: xor %edi,%edi
: srand(s_randseed);
0.00 0.00 : 4004b3: mov $0x77359400,%ebx
:
: return i;
: }
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: f681d593d1 ("perf annotate: Remove disasm__calc_percent() from disasm_line__print()")
Link: http://lkml.kernel.org/r/1525881435-4092-1-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
