With this patch, we can use the 'percore' event qualifier in perf-stat.
root@skl:/tmp# perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/ -a -A -I1000
1.000773050 S0-C0 98,352,832 cpu/event=0,umask=0x3,percore=1/ (50.01%)
1.000773050 S0-C1 103,763,057 cpu/event=0,umask=0x3,percore=1/ (50.02%)
1.000773050 S0-C2 196,776,995 cpu/event=0,umask=0x3,percore=1/ (50.02%)
1.000773050 S0-C3 176,493,779 cpu/event=0,umask=0x3,percore=1/ (50.02%)
1.000773050 CPU0 47,699,641 cpu/event=0,umask=0x3/ (50.02%)
1.000773050 CPU1 49,052,451 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU2 102,771,422 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU3 100,784,662 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU4 43,171,342 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU5 54,152,158 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU6 93,618,410 cpu/event=0,umask=0x3/ (49.98%)
1.000773050 CPU7 74,477,589 cpu/event=0,umask=0x3/ (49.99%)
In this example, we count the event 'ref-cycles' per-core and per-CPU in
one perf stat command-line. From the output, we can see:
S0-C0 = CPU0 + CPU4
S0-C1 = CPU1 + CPU5
S0-C2 = CPU2 + CPU6
S0-C3 = CPU3 + CPU7
So the result is expected (tiny difference is ignored).
Note that, the 'percore' event qualifier needs to use with option '-A'.
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-4-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Move the aggregate counts printing to a new function
print_counter_aggrdata, which will be used in following patches.
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-3-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a 'percore' event qualifier, like cpu/event=0,umask=0x3,percore=1/,
that sums up the event counts for both hardware threads in a core.
We can already do this with --per-core, but it's often useful to do
this together with other metrics that are collected per hardware thread.
So we need to support this per-core counting on a event level.
This can be implemented in only the user tool, no kernel support needed.
v4:
---
1. Add Arnaldo's patch which updates the documentation for
this new qualifier.
2. Rebase to latest perf/core branch
v3:
---
Simplify the code according to Jiri's comments.
Before:
"return term->val.percore ? true : false;"
Now:
"return term->val.percore;"
v2:
---
Change the qualifier name from 'coresum' to 'percore' according to
comments from Jiri and Andi.
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-2-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The sample timestamp is updated to ensure that the timestamp represents
the time of the sample and not a branch that the decoder is still
walking towards. The sample timestamp is updated when the decoder
returns, but the decoder does not return for non-taken branches. Update
the sample timestamp then also.
Note that commit 3f04d98e97 ("perf intel-pt: Improve sample
timestamp") was also a stable fix and appears, for example, in v4.4
stable tree as commit a4ebb58fd124 ("perf intel-pt: Improve sample
timestamp").
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org # v4.4+
Fixes: 3f04d98e97 ("perf intel-pt: Improve sample timestamp")
Link: http://lkml.kernel.org/r/20190510124143.27054-4-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The decoder uses its current timestamp in samples. Usually that is a
timestamp that has already passed, but in some cases it is a timestamp
for a branch that the decoder is walking towards, and consequently
hasn't reached.
The intel_pt_sample_time() function decides which is which, but was not
handling TNT packets exactly correctly.
In the case of TNT, the timestamp applies to the first branch, so the
decoder must first walk to that branch.
That means intel_pt_sample_time() should return true for TNT, and this
patch makes that change. However, if the first branch is a non-taken
branch (i.e. a 'N'), then intel_pt_sample_time() needs to return false
for subsequent taken branches in the same TNT packet.
To handle that, introduce a new state INTEL_PT_STATE_TNT_CONT to
distinguish the cases.
Note that commit 3f04d98e97 ("perf intel-pt: Improve sample
timestamp") was also a stable fix and appears, for example, in v4.4
stable tree as commit a4ebb58fd124 ("perf intel-pt: Improve sample
timestamp").
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org # v4.4+
Fixes: 3f04d98e97 ("perf intel-pt: Improve sample timestamp")
Link: http://lkml.kernel.org/r/20190510124143.27054-3-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The timestamp used to determine if an instruction sample is made, is an
estimate based on the number of instructions since the last known
timestamp. A consequence is that it might go backwards, which results in
extra samples. Change it so that a sample is only made when the
timestamp goes forwards.
Note this does not affect a sampling period of 0 or sampling periods
specified as a count of instructions.
Example:
Before:
$ perf script --itrace=i10us
ls 13812 [003] 2167315.222583: 3270 instructions:u: 7fac71e2e494 __GI___tunables_init+0xf4 (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 30902 instructions:u: 7fac71e2da0f _dl_cache_libcmp+0x2f (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 10 instructions:u: 7fac71e2d9ff _dl_cache_libcmp+0x1f (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 8 instructions:u: 7fac71e2d9ea _dl_cache_libcmp+0xa (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 14 instructions:u: 7fac71e2d9ea _dl_cache_libcmp+0xa (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 6 instructions:u: 7fac71e2d9ff _dl_cache_libcmp+0x1f (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 14 instructions:u: 7fac71e2d9ff _dl_cache_libcmp+0x1f (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 4 instructions:u: 7fac71e2dab2 _dl_cache_libcmp+0xd2 (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222728: 16423 instructions:u: 7fac71e2477a _dl_map_object_deps+0x1ba (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222734: 12731 instructions:u: 7fac71e27938 _dl_name_match_p+0x68 (/lib/x86_64-linux-gnu/ld-2.28.so)
...
After:
$ perf script --itrace=i10us
ls 13812 [003] 2167315.222583: 3270 instructions:u: 7fac71e2e494 __GI___tunables_init+0xf4 (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222667: 30902 instructions:u: 7fac71e2da0f _dl_cache_libcmp+0x2f (/lib/x86_64-linux-gnu/ld-2.28.so)
ls 13812 [003] 2167315.222728: 16479 instructions:u: 7fac71e2477a _dl_map_object_deps+0x1ba (/lib/x86_64-linux-gnu/ld-2.28.so)
...
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Fixes: f4aa081949 ("perf tools: Add Intel PT decoder")
Link: http://lkml.kernel.org/r/20190510124143.27054-2-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
There may be different register mask for use with intr or user on some
platforms, e.g. Icelake.
Add weak functions arch__intr_reg_mask() and arch__user_reg_mask() to
return intr and user register mask respectively.
Check mask before printing or comparing the register name.
Generic code always return PERF_REGS_MASK. No functional change.
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1557865174-56264-2-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The available registers for --int-regs and --user-regs may be different,
e.g. XMM registers.
Split parse_regs into two dedicated functions for --int-regs and
--user-regs respectively.
Modify the warning message. "--user-regs=?" should be applied to show
the available registers for --user-regs.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1557865174-56264-1-git-send-email-kan.liang@linux.intel.com
[ Changed docs as suggested by Ravi and agreed by Kan ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
zstd_init(, comp_level = 0) initializes decompression part of API only
hat now consists of zstd_decompress_stream() function.
The perf.data PERF_RECORD_COMPRESSED records are decompressed using
zstd_decompress_stream() function into a linked list of mmaped memory
regions of mmap_comp_len size (struct decomp).
After decompression of one COMPRESSED record its content is iterated and
fetched for usual processing. The mmaped memory regions with
decompressed events are kept in the linked list till the tool process
termination.
When dumping raw records (e.g., perf report -D --header) file offsets of
events from compressed records are printed as zero.
Committer notes:
Since now we have support for processing PERF_RECORD_COMPRESSED, we see
none, in raw form, like we saw in the previous patch commiter notes,
they were decompressed into the usual PERF_RECORD_{FORK,MMAP,COMM,etc}
records, we only see the stats for those PERF_RECORD_COMPRESSED events,
and since I used the file generated in the commiter notes for the
previous patch, there they are, 2 compressed records:
$ perf report --header-only | grep cmdline
# cmdline : /home/acme/bin/perf record -z2 sleep 1
$ perf report -D | grep COMPRESS
COMPRESSED events: 2
COMPRESSED events: 0
$ perf report --stdio
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 15 of event 'cycles:u'
# Event count (approx.): 962227
#
# Overhead Command Shared Object Symbol
# ........ ....... ................ ...........................
#
46.99% sleep libc-2.28.so [.] _dl_addr
29.24% sleep [unknown] [k] 0xffffffffaea00a67
16.45% sleep libc-2.28.so [.] __GI__IO_un_link.part.1
5.92% sleep ld-2.28.so [.] _dl_setup_hash
1.40% sleep libc-2.28.so [.] __nanosleep
0.00% sleep [unknown] [k] 0xffffffffaea00163
#
# (Tip: To see callchains in a more compact form: perf report -g folded)
#
$
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
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: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/304b0a59-942c-3fe1-da02-aa749f87108b@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Committer note:
Split from a larger patch, this only dumps PERF_RECORD_COMPRESSED as
unhandled, so that when we introduce the record part in the next patch,
we don't see unhandled events when using 'perf record -D'.
Changed it so that we dump the event if the handler is just a stub, i.e.
for the case where we don't have ZSTD linked but we're processing a
perf.data file generated by a tool with that linked.
Also when failing to decompress we can't just dump the uncompressed
event and return 0, we have to propagate the error.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/304b0a59-942c-3fe1-da02-aa749f87108b@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Compression is implemented using the functions from zstd.c. As the memory
to operate on the compression uses mmap->aio.data[] buffers. If Zstd
streaming compression API fails for some reason the data to be compressed
are just copied into the memory buffers using plain memcpy().
Compressed trace frame consists of an array of PERF_RECORD_COMPRESSED
records. Each element of the array is not longer that PERF_SAMPLE_MAX_SIZE
and consists of perf_event_header followed by the compressed chunk
that is decompressed on the loading stage.
perf_mmap__aio_push() is replaced by perf_mmap__push() which is now used
in the both serial and AIO streaming cases. perf_mmap__push() is extended
with positive return values to signify absence of data ready for
processing.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/77db2b2c-5d03-dbb0-aeac-c4dd92129ab9@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Compression is implemented using the functions from zstd.c. As the
memory to operate on the compression uses mmap->data buffer.
If Zstd streaming compression API fails for some reason the data to be
compressed are just copied into the memory buffers using plain memcpy().
Compressed trace frame consists of an array of PERF_RECORD_COMPRESSED
records. Each element of the array is not longer that
PERF_SAMPLE_MAX_SIZE and consists of perf_event_header followed by the
compressed chunk that is decompressed on the loading stage.
Comitter notes:
Undo some unnecessary line breaks, remove some unnecessary () around
zstd_data to then just get its address, and fix conflicts with
BPF_PROG_INFO/BPF_BTF patchkits.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/744df43f-3932-2594-ddef-1e99a3cad03a@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implemented functions are based on Zstd streaming compression API.
The functions are used in runtime to compress data that come from mmaped
kernel buffer. zstd_init(), zstd_fini() are used for initialization and
finalization to allocate and deallocate internal zstd objects.
zstd_compress_stream_to_records() is used to convert parts of mmaped
kernel buffer into an array of PERF_RECORD_COMPRESSED records.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/18bf36f3-b85a-1fe2-dd83-10e0c6069568@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implemented mmap data buffer that is used as the memory to operate
on when compressing data in case of serial trace streaming.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/49b31321-0f70-392b-9a4f-649d3affe090@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implemented PERF_RECORD_COMPRESSED event, related data types, header
feature and functions to write, read and print feature attributes from
the trace header section.
comp_mmap_len preserves the size of mmaped kernel buffer that was used
during collection. comp_mmap_len size is used on loading stage as the
size of decomp buffer for decompression of COMPRESSED events content.
Committer notes:
Fixed up conflict with BPF_PROG_INFO and BTF_BTF header features.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/ebbaf031-8dda-3864-ebc6-7922d43ee515@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Define 'bytes_transferred' and 'bytes_compressed' metrics to calculate
ratio in the end of the data collection:
compression ratio = bytes_transferred / bytes_compressed
The 'bytes_transferred' metric accumulates the amount of bytes that was
extracted from the mmaped kernel buffers for compression, while
'bytes_compressed' accumulates the amount of bytes that was received
after applying compression.
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1d4bf499-cb03-26dc-6fc6-f14fec7622ce@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If fgets() fails due to any other error besides end-of-file, the version
char array may not even be null-terminated.
Signed-off-by: Donald Yandt <donald.yandt@gmail.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Avi Kivity <avi@scylladb.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yanmin Zhang <yanmin_zhang@linux.intel.com>
Fixes: a1645ce12a ("perf: 'perf kvm' tool for monitoring guest performance from host")
Link: http://lkml.kernel.org/r/20190514110100.22019-1-donald.yandt@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Icelake and later platforms support collecting XMM registers with PEBS
event.
Add support for 'perf script' to dump them, and support for the register
parser in 'perf record -I=' ... to configure them.
For now they are just printed in hex, we could potentially later add
other formats too.
Committer testing:
Before:
# perf record -IXMM0
Warning:
unknown register XMM0, check man page or run 'perf record -I?'
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
#
# perf record -I?
available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
#
After:
# perf record -IXMM0
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (cycles).
/bin/dmesg | grep -i perf may provide additional information.
#
# perf record -I?
available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 XMM9 XMM10 XMM11 XMM12 XMM13 XMM14 XMM15
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-I, --intr-regs[=<any register>]
sample selected machine registers on interrupt, use -I ? to list register names
#
More work is needed to, when faced with such error, warn the user that
that register is not available on the running platform.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190506141926.13659-1-kan.liang@linux.intel.com
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add quotes around the register name and suggest using 'perf record -I?'
to get the list of available registers.
Before:
# perf record -Idi,xmm20,xmm1
Warning:
unknown register xmm20, check man page
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-I, --intr-regs[=<any register>]
sample selected machine registers on interrupt, use -I ? to list register names
#
# perf record -Idi,xmm20,xmm1
Warning:
unknown register "xmm20", check man page or run "perf record -I?"
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-I, --intr-regs[=<any register>]
sample selected machine registers on interrupt, use -I ? to list register names
#
Cc: Adrian Hunter <adrian.hunter@intel.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: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: https://lkml.kernel.org/n/tip-9a9hyuum8c0oggg86xd3sxc5@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When compiled with libunwind, perf does some preparatory work when
processing side-band events. This is not needed when report actually
don't unwind dwarf callchains, so it's disabled with
dwarf_callchain_users bool.
However we could move that check to higher level and shield more
unwanted code for normal report processing, giving us following speed up
on kernel build profile:
Before:
$ perf record make -j40
...
$ ll ../../perf.data
-rw-------. 1 jolsa jolsa 461783932 Apr 26 09:11 perf.data
$ perf stat -e cycles:u,instructions:u perf report -i perf.data > out
Performance counter stats for 'perf report -i perf.data':
78,669,920,155 cycles:u
99,076,431,951 instructions:u # 1.26 insn per cycle
55.382823668 seconds time elapsed
27.512341000 seconds user
27.712871000 seconds sys
After:
$ perf stat -e cycles:u,instructions:u perf report -i perf.data > out
Performance counter stats for 'perf report -i perf.data':
59,626,798,904 cycles:u
88,583,575,849 instructions:u # 1.49 insn per cycle
21.296935559 seconds time elapsed
20.010191000 seconds user
1.202935000 seconds sys
The speed is higher with profile having many side-band events,
because these trigger libunwind preparatory code.
This does not apply for perf compiled with libdw for dwarf unwind,
only for build with libunwind.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190426073804.17238-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The hist__account_cycles() function is executed when the
hist_iter__branch_callback() is called.
But it looks it's not necessary. In hist__account_cycles, it already
walks on all branch entries.
This patch moves the hist__account_cycles out of callback, now the data
processing is much faster than before.
Previous code has an issue that the ch[offset].num++ (in
__symbol__account_cycles) is executed repeatedly since
hist__account_cycles is called in each hist_iter__branch_callback, so
the counting of ch[offset].num is not correct (too big).
With this patch, the issue is fixed. And we don't need the code of
"ch->reset >= ch->num / 2" to check if there are too many overlaps (in
annotation__count_and_fill), otherwise some data would be hidden.
Now, we can try, for example:
perf record -b ...
perf annotate or perf report -s symbol
The before/after output should be no change.
v3:
---
Fix the crash in stdio mode.
Like previous code, it needs the checking of ui__has_annotation()
before hist__account_cycles()
v2:
---
1. Cover the similar perf report
2. Remove the checking code "ch->reset >= ch->num / 2"
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: Jin Yao <yao.jin@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1552684577-29041-1-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf updates from Ingo Molnar:
"The main kernel changes were:
- add support for Intel's "adaptive PEBS v4" - which embedds LBS data
in PEBS records and can thus batch up and reduce the IRQ (NMI) rate
significantly - reducing overhead and making call-graph profiling
less intrusive.
- add Intel CPU core and uncore support updates for Tremont, Icelake,
- extend the x86 PMU constraints scheduler with 'constraint ranges'
to better support Icelake hw constraints,
- make x86 call-chain support work better with CONFIG_FRAME_POINTER=y
- misc other changes
Tooling changes:
- updates to the main tools: 'perf record', 'perf trace', 'perf
stat'
- updated Intel and S/390 vendor events
- libtraceevent updates
- misc other updates and fixes"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (69 commits)
perf/x86: Make perf callchains work without CONFIG_FRAME_POINTER
watchdog: Fix typo in comment
perf/x86/intel: Add Tremont core PMU support
perf/x86/intel/uncore: Add Intel Icelake uncore support
perf/x86/msr: Add Icelake support
perf/x86/intel/rapl: Add Icelake support
perf/x86/intel/cstate: Add Icelake support
perf/x86/intel: Add Icelake support
perf/x86: Support constraint ranges
perf/x86/lbr: Avoid reading the LBRs when adaptive PEBS handles them
perf/x86/intel: Support adaptive PEBS v4
perf/x86/intel/ds: Extract code of event update in short period
perf/x86/intel: Extract memory code PEBS parser for reuse
perf/x86: Support outputting XMM registers
perf/x86/intel: Force resched when TFA sysctl is modified
perf/core: Add perf_pmu_resched() as global function
perf/headers: Fix stale comment for struct perf_addr_filter
perf/core: Make perf_swevent_init_cpu() static
perf/x86: Add sanity checks to x86_schedule_events()
perf/x86: Optimize x86_schedule_events()
...
We were including sys/syscall.h and asm/unistd.h, since sys/syscall.h
includes asm/unistd.h, sometimes this leads to the redefinition of
defines, breaking the build.
Noticed on ARC with uCLibc.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Vineet Gupta <Vineet.Gupta1@synopsys.com>
Link: https://lkml.kernel.org/n/tip-xjpf80o64i2ko74aj2jih0qg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Robert Walker reported a segmentation fault is observed when process
CoreSight trace data; this issue can be easily reproduced by the command
'perf report --itrace=i1000i' for decoding tracing data.
If neither the 'b' flag (synthesize branches events) nor 'l' flag
(synthesize last branch entries) are specified to option '--itrace',
cs_etm_queue::prev_packet will not been initialised. After merging the
code to support exception packets and sample flags, there introduced a
number of uses of cs_etm_queue::prev_packet without checking whether it
is valid, for these cases any accessing to uninitialised prev_packet
will cause crash.
As cs_etm_queue::prev_packet is used more widely now and it's already
hard to follow which functions have been called in a context where the
validity of cs_etm_queue::prev_packet has been checked, this patch
always allocates memory for cs_etm_queue::prev_packet.
Reported-by: Robert Walker <robert.walker@arm.com>
Suggested-by: Robert Walker <robert.walker@arm.com>
Signed-off-by: Leo Yan <leo.yan@linaro.org>
Tested-by: Robert Walker <robert.walker@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Suzuki K Poulouse <suzuki.poulose@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Fixes: 7100b12cf4 ("perf cs-etm: Generate branch sample for exception packet")
Fixes: 24fff5eb2b ("perf cs-etm: Avoid stale branch samples when flush packet")
Link: http://lkml.kernel.org/r/20190428083228.20246-1-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Since cs_etm_queue::prev_packet is allocated for all cases, it will
never be NULL pointer; now validity checking prev_packet is pointless,
remove all of them.
Signed-off-by: Leo Yan <leo.yan@linaro.org>
Tested-by: Robert Walker <robert.walker@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Suzuki K Poulouse <suzuki.poulose@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/20190428083228.20246-2-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
An -ENOMEM error is not reported in the GTK GUI. Instead this error
message pops up on the screen:
[root@m35lp76 perf]# ./perf report -i perf.data.error68-1
Processing events... [974K/3M]
Error:failed to process sample
0xf4198 [0x8]: failed to process type: 68
However when I use the same perf.data file with --stdio it works:
[root@m35lp76 perf]# ./perf report -i perf.data.error68-1 --stdio \
| head -12
# Total Lost Samples: 0
#
# Samples: 76K of event 'cycles'
# Event count (approx.): 99056160000
#
# Overhead Command Shared Object Symbol
# ........ ............... ................. .........
#
8.81% find [kernel.kallsyms] [k] ftrace_likely_update
8.74% swapper [kernel.kallsyms] [k] ftrace_likely_update
8.34% sshd [kernel.kallsyms] [k] ftrace_likely_update
2.19% kworker/u512:1- [kernel.kallsyms] [k] ftrace_likely_update
The sample precentage is a bit low.....
The GUI always fails in the FINISHED_ROUND event (68) and does not
indicate the reason why.
When happened is the following. Perf report calls a lot of functions and
down deep when a FINISHED_ROUND event is processed, these functions are
called:
perf_session__process_event()
+ perf_session__process_user_event()
+ process_finished_round()
+ ordered_events__flush()
+ __ordered_events__flush()
+ do_flush()
+ ordered_events__deliver_event()
+ perf_session__deliver_event()
+ machine__deliver_event()
+ perf_evlist__deliver_event()
+ process_sample_event()
+ hist_entry_iter_add() --> only called in GUI case!!!
+ hist_iter__report__callback()
+ symbol__inc_addr_sample()
Now this functions runs out of memory and
returns -ENOMEM. This is reported all the way up
until function
perf_session__process_event() returns to its caller, where -ENOMEM is
changed to -EINVAL and processing stops:
if ((skip = perf_session__process_event(session, event, head)) < 0) {
pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
head, event->header.size, event->header.type);
err = -EINVAL;
goto out_err;
}
This occurred in the FINISHED_ROUND event when it has to process some
10000 entries and ran out of memory.
This patch indicates the root cause and displays it in the status line
of ther perf report GUI.
Output before (on GUI status line):
0xf4198 [0x8]: failed to process type: 68
Output after:
0xf4198 [0x8]: failed to process type: 68 [not enough memory]
Committer notes:
the 'skip' variable needs to be initialized to -EINVAL, so that when the
size is less than sizeof(struct perf_event_attr) we avoid this valid
compiler warning:
util/session.c: In function ‘perf_session__process_events’:
util/session.c:1936:7: error: ‘skip’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
err = skip;
~~~~^~~~~~
util/session.c:1874:6: note: ‘skip’ was declared here
s64 skip;
^~~~
cc1: all warnings being treated as errors
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20190423105303.61683-1-tmricht@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Commit 6987561c9e ("perf annotate: Enable annotation of BPF programs") adds
support for BPF programs annotations but the new code does not build on 32-bit.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Acked-by: Song Liu <songliubraving@fb.com>
Fixes: 6987561c9e ("perf annotate: Enable annotation of BPF programs")
Link: http://lkml.kernel.org/r/20190403194452.10845-1-cascardo@canonical.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In perf_env__find_btf(), we're returning without unlocking
"env->bpf_progs.lock". There may be cause lockdep issue.
Detected by CoversityScan, CID# 1444762:(program hangs(LOCK))
Signed-off-by: Bo YU <tsu.yubo@gmail.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Cc: bpf@vger.kernel.org
Cc: netdev@vger.kernel.org
Fixes: 2db7b1e0bd: (perf bpf: Return NULL when RB tree lookup fails in perf_env__find_btf())
Link: http://lkml.kernel.org/r/20190422080138.10088-1-tsu.yubo@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We don't return NULL when we don't find the bpf_prog_info_node, fix
that.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Reported-by: Song Liu <songliubraving@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: 3792cb2ff4 ("perf bpf: Save BTF in a rbtree in perf_env")
Link: http://lkml.kernel.org/r/20190417145539.11669-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
By calling maps__insert() we assume to get 2 references on the map,
which we relese within maps__remove call.
However if there's already same map name, we currently don't bump the
reference and can crash, like:
Program received signal SIGABRT, Aborted.
0x00007ffff75e60f5 in raise () from /lib64/libc.so.6
(gdb) bt
#0 0x00007ffff75e60f5 in raise () from /lib64/libc.so.6
#1 0x00007ffff75d0895 in abort () from /lib64/libc.so.6
#2 0x00007ffff75d0769 in __assert_fail_base.cold () from /lib64/libc.so.6
#3 0x00007ffff75de596 in __assert_fail () from /lib64/libc.so.6
#4 0x00000000004fc006 in refcount_sub_and_test (i=1, r=0x1224e88) at tools/include/linux/refcount.h:131
#5 refcount_dec_and_test (r=0x1224e88) at tools/include/linux/refcount.h:148
#6 map__put (map=0x1224df0) at util/map.c:299
#7 0x00000000004fdb95 in __maps__remove (map=0x1224df0, maps=0xb17d80) at util/map.c:953
#8 maps__remove (maps=0xb17d80, map=0x1224df0) at util/map.c:959
#9 0x00000000004f7d8a in map_groups__remove (map=<optimized out>, mg=<optimized out>) at util/map_groups.h:65
#10 machine__process_ksymbol_unregister (sample=<optimized out>, event=0x7ffff7279670, machine=<optimized out>) at util/machine.c:728
#11 machine__process_ksymbol (machine=<optimized out>, event=0x7ffff7279670, sample=<optimized out>) at util/machine.c:741
#12 0x00000000004fffbb in perf_session__deliver_event (session=0xb11390, event=0x7ffff7279670, tool=0x7fffffffc7b0, file_offset=13936) at util/session.c:1362
#13 0x00000000005039bb in do_flush (show_progress=false, oe=0xb17e80) at util/ordered-events.c:243
#14 __ordered_events__flush (oe=0xb17e80, how=OE_FLUSH__ROUND, timestamp=<optimized out>) at util/ordered-events.c:322
#15 0x00000000005005e4 in perf_session__process_user_event (session=session@entry=0xb11390, event=event@entry=0x7ffff72a4af8,
...
Add the map to the list and getting the reference event if we find the
map with same name.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Eric Saint-Etienne <eric.saint.etienne@oracle.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Fixes: 1e6285699b ("perf symbols: Fix slowness due to -ffunction-section")
Link: http://lkml.kernel.org/r/20190416160127.30203-10-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Current perf_evlist__poll_thread() code could finish without draining
the data. Adding the logic that makes sure we won't finish before the
drain.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Fixes: 657ee55319 ("perf evlist: Introduce side band thread")
Link: http://lkml.kernel.org/r/20190416160127.30203-9-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As reported by Jiri Olsa in:
"[BUG] perf: intel_pt won't display kernel function"
https://lore.kernel.org/lkml/20190403143738.GB32001@krava
Recent changes to support PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT
broke --kallsyms option. This is because it broke test __map__is_kmodule.
This patch fixes this by adding check for bpf program, so that these maps
are not mistaken as kernel modules.
Signed-off-by: Song Liu <songliubraving@fb.com>
Reported-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yonghong Song <yhs@fb.com>
Link: http://lkml.kernel.org/r/20190416160127.30203-8-jolsa@kernel.org
Fixes: 76193a9452 ("perf, bpf: Introduce PERF_RECORD_KSYMBOL")
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We currently don't return NULL in case we don't find the
bpf_prog_info_node, fixing that.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: e4378f0cb9 ("perf bpf: Save bpf_prog_info in a rbtree in perf_env")
Link: http://lkml.kernel.org/r/20190416134151.15282-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
On 32-bits platform with more than 32 registers, the 64 bits mask is
truncate to the lower 32 bits and the return value of hweight_long will
always smaller than 32. When kernel outputs more than 32 registers, but
the user perf program only counts 32, there will be a data mismatch
result to overflow check fail.
Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
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: Stephane Eranian <eranian@google.com>
Fixes: 6a21c0b5c2 ("perf tools: Add core support for sampling intr machine state regs")
Fixes: d03f217054 ("perf tools: Expand perf_event__synthesize_sample()")
Fixes: 0f6a30150c ("perf tools: Support user regs and stack in sample parsing")
Link: http://lkml.kernel.org/r/29ad7947dc8fd1ff0abd2093a72cc27a2446be9f.1554883878.git.han_mao@c-sky.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fix lock/unlock imbalances by refactoring the code a bit and adding
calls to up_write() before return.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Acked-by: Song Liu <songliubraving@fb.com>
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>
Addresses-Coverity-ID: 1444315 ("Missing unlock")
Addresses-Coverity-ID: 1444316 ("Missing unlock")
Fixes: a70a112317 ("perf bpf: Save BTF information as headers to perf.data")
Fixes: 606f972b13 ("perf bpf: Save bpf_prog_info information as headers to perf.data")
Link: http://lkml.kernel.org/r/20190408173355.GA10501@embeddedor
[ Simplified the exit path to have just one up_write() + return ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implement a --mmap-flush option that specifies minimal number of bytes
that is extracted from mmaped kernel buffer to store into a trace. The
default option value is 1 byte what means every time trace writing
thread finds some new data in the mmaped buffer the data is extracted,
possibly compressed and written to a trace.
$ tools/perf/perf record --mmap-flush 1024 -e cycles -- matrix.gcc
$ tools/perf/perf record --aio --mmap-flush 1K -e cycles -- matrix.gcc
The option is independent from -z setting, doesn't vary with compression
level and can serve two purposes.
The first purpose is to increase the compression ratio of a trace data.
Larger data chunks are compressed more effectively so the implemented
option allows specifying data chunk size to compress. Also at some cases
executing more write syscalls with smaller data size can take longer
than executing less write syscalls with bigger data size due to syscall
overhead so extracting bigger data chunks specified by the option value
could additionally decrease runtime overhead.
The second purpose is to avoid self monitoring live-lock issue in system
wide (-a) profiling mode. Profiling in system wide mode with compression
(-a -z) can additionally induce data into the kernel buffers along with
the data from monitored processes. If performance data rate and volume
from the monitored processes is high then trace streaming and
compression activity in the tool is also high. High tool process
activity can lead to subtle live-lock effect when compression of single
new byte from some of mmaped kernel buffer leads to generation of the
next single byte at some mmaped buffer. So perf tool process ends up in
endless self monitoring.
Implemented synch parameter is the mean to force data move independently
from the specified flush threshold value. Despite the provided flush
value the tool needs capability to unconditionally drain memory buffers,
at least in the end of the collection.
Committer testing:
Running with the default value, i.e. as soon as there is something to
read go on consuming, we first write the synthesized events, small
chunks of about 128 bytes:
# perf trace -m 2048 --call-graph dwarf -e write -- perf record
<SNIP>
101.142 ( 0.004 ms): perf/25821 write(fd: 3</root/perf.data>, buf: 0x210db60, count: 120) = 120
__libc_write (/usr/lib64/libpthread-2.28.so)
ion (/home/acme/bin/perf)
record__write (inlined)
process_synthesized_event (/home/acme/bin/perf)
perf_tool__process_synth_event (inlined)
perf_event__synthesize_mmap_events (/home/acme/bin/perf)
Then we move to reading the mmap buffers consuming the events put there
by the kernel perf infrastructure:
107.561 ( 0.005 ms): perf/25821 write(fd: 3</root/perf.data>, buf: 0x7f1befc02000, count: 336) = 336
__libc_write (/usr/lib64/libpthread-2.28.so)
ion (/home/acme/bin/perf)
record__write (inlined)
record__pushfn (/home/acme/bin/perf)
perf_mmap__push (/home/acme/bin/perf)
record__mmap_read_evlist (inlined)
record__mmap_read_all (inlined)
__cmd_record (inlined)
cmd_record (/home/acme/bin/perf)
12919.953 ( 0.136 ms): perf/25821 write(fd: 3</root/perf.data>, buf: 0x7f1befc83150, count: 184984) = 184984
<SNIP same backtrace as in the 107.561 timestamp>
12920.094 ( 0.155 ms): perf/25821 write(fd: 3</root/perf.data>, buf: 0x7f1befc02150, count: 261816) = 261816
<SNIP same backtrace as in the 107.561 timestamp>
12920.253 ( 0.093 ms): perf/25821 write(fd: 3</root/perf.data>, buf: 0x7f1befb81120, count: 170832) = 170832
<SNIP same backtrace as in the 107.561 timestamp>
If we limit it to write only when more than 16MB are available for
reading, it throttles that to a quarter of the --mmap-pages set for
'perf record', which by default get to 528384 bytes, found out using
'record -v':
mmap flush: 132096
mmap size 528384B
With that in place all the writes coming from
record__mmap_read_evlist(), i.e. from the mmap buffers setup by the
kernel perf infrastructure were at least 132096 bytes long.
Trying with a bigger mmap size:
perf trace -e write perf record -v -m 2048 --mmap-flush 16M
74982.928 ( 2.471 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff94a6cc000, count: 3580888) = 3580888
74985.406 ( 2.353 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff949ecb000, count: 3453256) = 3453256
74987.764 ( 2.629 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff9496ca000, count: 3859232) = 3859232
74990.399 ( 2.341 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff948ec9000, count: 3769032) = 3769032
74992.744 ( 2.064 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff9486c8000, count: 3310520) = 3310520
74994.814 ( 2.619 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff947ec7000, count: 4194688) = 4194688
74997.439 ( 2.787 ms): perf/26500 write(fd: 3</root/perf.data>, buf: 0x7ff9476c6000, count: 4029760) = 4029760
Was again limited to a quarter of the mmap size:
mmap flush: 2098176
mmap size 8392704B
A warning about that would be good to have but can be added later,
something like:
"max flush is a quarter of the mmap size, if wanting to bump the mmap
flush further, bump the mmap size as well using -m/--mmap-pages"
Also rename the 'sync' parameters to 'synch' to keep tools/perf building
with older glibcs:
cc1: warnings being treated as errors
builtin-record.c: In function 'record__mmap_read_evlist':
builtin-record.c:775: warning: declaration of 'sync' shadows a global declaration
/usr/include/unistd.h:933: warning: shadowed declaration is here
builtin-record.c: In function 'record__mmap_read_all':
builtin-record.c:856: warning: declaration of 'sync' shadows a global declaration
/usr/include/unistd.h:933: warning: shadowed declaration is here
Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
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: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/f6600d72-ecfa-2eb7-7e51-f6954547d500@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The member "pevent" of the struct tep_event is renamed to "tep". This
makes the struct consistent with the chosen naming convention:
tep (trace event parser), instead of the old pevent.
Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lore.kernel.org/linux-trace-devel/20190401132111.13727-3-tstoyanov@vmware.com
Link: http://lkml.kernel.org/r/20190401164344.627724996@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Rename some traceevent APIs for consistency:
tep_pid_is_registered() to tep_is_pid_registered()
tep_file_bigendian() to tep_is_file_bigendian()
to make the names and return values consistent with other tep_is_... APIs
tep_data_lat_fmt() to tep_data_latency_format()
to make the name more descriptive
tep_host_bigendian() to tep_is_bigendian()
tep_set_host_bigendian() to tep_set_local_bigendian()
tep_is_host_bigendian() to tep_is_local_bigendian()
"host" can be confused with VMs, and "local" is about the local
machine. All tep_is_..._bigendian(struct tep_handle *tep) APIs return
the saved data in the tep handle, while tep_is_bigendian() returns
the running machine's endianness.
All tep_is_... functions are modified to return bool value, instead of int.
Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20190327141946.4353-2-tstoyanov@vmware.com
Link: http://lkml.kernel.org/r/20190401164344.288624897@goodmis.org
[ Removed some extra parenthesis around return statements ]
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add support in 'perf list' to output tool internal events, currently
only 'duration_time'.
Committer testing:
$ perf list dur*
List of pre-defined events (to be used in -e):
duration_time [Tool event]
Metric Groups:
$ perf list sw
List of pre-defined events (to be used in -e):
alignment-faults [Software event]
bpf-output [Software event]
context-switches OR cs [Software event]
cpu-clock [Software event]
cpu-migrations OR migrations [Software event]
dummy [Software event]
emulation-faults [Software event]
major-faults [Software event]
minor-faults [Software event]
page-faults OR faults [Software event]
task-clock [Software event]
duration_time [Tool event]
$ perf list | grep duration
duration_time [Tool event]
[L1D miss outstandings duration in cycles]
page walk duration are excluded in Skylake]
load. EPT page walk duration are excluded in Skylake]
page walk duration are excluded in Skylake]
store. EPT page walk duration are excluded in Skylake]
(instruction fetch) request. EPT page walk duration are excluded in
instruction fetch request. EPT page walk duration are excluded in
$
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190326221823.11518-5-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implement printing the correct name for duration_time
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190326221823.11518-4-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf metric expression use 'duration_time' internally to normalize
events. Normal 'perf stat' without -x also prints the duration time.
But when using -x, the interval is not output anywhere, which is
inconvenient for any post processing which often wants to normalize
values to time.
So implement 'duration_time' as a proper perf event that can be
specified explicitely with -e.
The previous implementation of 'duration_time' only worked for metric
processing. This adds the concept of a tool event that is handled by the
tool. On the kernel level it is still mapped to the dummy software
event, but the values are not read anymore, but instead computed by the
tool.
Add proper plumbing to handle this in the event parser, and display it
in 'perf stat'. We don't want 'duration_time' to be added up, so it's
only printed for the first CPU.
% perf stat -e duration_time,cycles true
Performance counter stats for 'true':
555,476 ns duration_time
771,958 cycles
0.000555476 seconds time elapsed
0.000644000 seconds user
0.000000000 seconds sys
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190326221823.11518-3-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This reverts e864c5ca14 ("perf stat: Hide internal duration_time
counter") but doing it manually since the code has now moved to a
different file.
The next patch will properly implement duration_time as a full event, so
no need to hide it anymore.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190326221823.11518-2-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When adding the 'struct namespaces_event' to event.h, referencing the
'struct perf_ns_link_info' type, we forgot to add the header where it is
defined, getting that definition only by sheer luck.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: f3b3614a28 ("perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info")
Link: https://lkml.kernel.org/n/tip-qkrld0v7boc9uabjbd8csxux@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf fails to parse uncore event alias, for example:
# perf stat -e unc_m_clockticks -a --no-merge sleep 1
event syntax error: 'unc_m_clockticks'
\___ parser error
Current code assumes that the event alias is from one specific PMU.
To find the PMU, perf strcmps the PMU name of event alias with the real
PMU name on the system.
However, the uncore event alias may be from multiple PMUs with common
prefix. The PMU name of uncore event alias is the common prefix.
For example, UNC_M_CLOCKTICKS is clock event for iMC, which include 6
PMUs with the same prefix "uncore_imc" on a skylake server.
The real PMU names on the system for iMC are uncore_imc_0 ...
uncore_imc_5.
The strncmp is used to only check the common prefix for uncore event
alias.
With the patch:
# perf stat -e unc_m_clockticks -a --no-merge sleep 1
Performance counter stats for 'system wide':
723,594,722 unc_m_clockticks [uncore_imc_5]
724,001,954 unc_m_clockticks [uncore_imc_3]
724,042,655 unc_m_clockticks [uncore_imc_1]
724,161,001 unc_m_clockticks [uncore_imc_4]
724,293,713 unc_m_clockticks [uncore_imc_2]
724,340,901 unc_m_clockticks [uncore_imc_0]
1.002090060 seconds time elapsed
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: stable@vger.kernel.org
Fixes: ea1fa48c05 ("perf stat: Handle different PMU names with common prefix")
Link: http://lkml.kernel.org/r/1552672814-156173-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Since commit 1fb87b8e95 ("perf machine: Don't search for active kernel
start in __machine__create_kernel_maps"), the __machine__create_kernel_maps()
just create a map what start and end are both zero. Though the address will be
updated later, the order of map in the rbtree may be incorrect.
The commit ee05d21791 ("perf machine: Set main kernel end address properly")
fixed the logic in machine__create_kernel_maps(), but it's still wrong in
function machine__process_kernel_mmap_event().
To reproduce this issue, we need an environment which the module address
is before the kernel text segment. I tested it on an aarch64 machine with
kernel 4.19.25:
[root@localhost hulk]# grep _stext /proc/kallsyms
ffff000008081000 T _stext
[root@localhost hulk]# grep _etext /proc/kallsyms
ffff000009780000 R _etext
[root@localhost hulk]# tail /proc/modules
hisi_sas_v2_hw 77824 0 - Live 0xffff00000191d000
nvme_core 126976 7 nvme, Live 0xffff0000018b6000
mdio 20480 1 ixgbe, Live 0xffff0000018ab000
hisi_sas_main 106496 1 hisi_sas_v2_hw, Live 0xffff000001861000
hns_mdio 20480 2 - Live 0xffff000001822000
hnae 28672 3 hns_dsaf,hns_enet_drv, Live 0xffff000001815000
dm_mirror 40960 0 - Live 0xffff000001804000
dm_region_hash 32768 1 dm_mirror, Live 0xffff0000017f5000
dm_log 32768 2 dm_mirror,dm_region_hash, Live 0xffff0000017e7000
dm_mod 315392 17 dm_mirror,dm_log, Live 0xffff000001780000
[root@localhost hulk]#
Before fix:
[root@localhost bin]# perf record sleep 3
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.011 MB perf.data (9 samples) ]
[root@localhost bin]# perf buildid-list -i perf.data
4c4e46c971ca935f781e603a09b52a92e8bdfee8 [vdso]
[root@localhost bin]# perf buildid-list -i perf.data -H
0000000000000000000000000000000000000000 /proc/kcore
[root@localhost bin]#
After fix:
[root@localhost tools]# ./perf/perf record sleep 3
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.011 MB perf.data (9 samples) ]
[root@localhost tools]# ./perf/perf buildid-list -i perf.data
28a6c690262896dbd1b5e1011ed81623e6db0610 [kernel.kallsyms]
106c14ce6e4acea3453e484dc604d66666f08a2f [vdso]
[root@localhost tools]# ./perf/perf buildid-list -i perf.data -H
28a6c690262896dbd1b5e1011ed81623e6db0610 /proc/kcore
Signed-off-by: Wei Li <liwei391@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Li Bin <huawei.libin@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190228092003.34071-1-liwei391@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
After a discussion with Andi, move the perf_event_attr.precise_ip
detection for maximum precise config (via :P modifier or for default
cycles event) to perf_evsel__open().
The current detection in perf_event_attr__set_max_precise_ip() is
tricky, because precise_ip config is specific for given event and it
currently checks only hw cycles.
We now check for valid precise_ip value right after failing
sys_perf_event_open() for specific event, before any of the
perf_event_attr fallback code gets executed.
This way we get the proper config in perf_event_attr together with
allowed precise_ip settings.
We can see that code activity with -vv, like:
$ perf record -vv ls
...
------------------------------------------------------------
perf_event_attr:
size 112
{ sample_period, sample_freq } 4000
...
precise_ip 3
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
------------------------------------------------------------
sys_perf_event_open: pid 9926 cpu 0 group_fd -1 flags 0x8
sys_perf_event_open failed, error -95
decreasing precise_ip by one (2)
------------------------------------------------------------
perf_event_attr:
size 112
{ sample_period, sample_freq } 4000
...
precise_ip 2
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
------------------------------------------------------------
sys_perf_event_open: pid 9926 cpu 0 group_fd -1 flags 0x8 = 4
...
Suggested-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/n/tip-dkvxxbeg7lu74155d4jhlmc9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
A TSC packet can slip past MTC packets so that the timestamp appears to
go backwards. One estimate is that can be up to about 40 CPU cycles,
which is certainly less than 0x1000 TSC ticks, but accept slippage an
order of magnitude more to be on the safe side.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Fixes: 79b58424b8 ("perf tools: Add Intel PT support for decoding MTC packets")
Link: http://lkml.kernel.org/r/20190325135135.18348-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The following error was thrown when compiling `tools/perf` using OpenCSD
v0.11.1. This patch fixes said error.
CC util/intel-pt-decoder/intel-pt-log.o
CC util/cs-etm-decoder/cs-etm-decoder.o
util/cs-etm-decoder/cs-etm-decoder.c: In function
‘cs_etm_decoder__buffer_range’:
util/cs-etm-decoder/cs-etm-decoder.c:370:2: error: enumeration value
‘OCSD_INSTR_WFI_WFE’ not handled in switch [-Werror=switch-enum]
switch (elem->last_i_type) {
^~~~~~
CC util/intel-pt-decoder/intel-pt-decoder.o
cc1: all warnings being treated as errors
Because `OCSD_INSTR_WFI_WFE` case was added only in v0.11.0, the minimum
required OpenCSD library version for this patch is no longer v0.10.0.
Signed-off-by: Solomon Tan <solomonbobstoner@gmail.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Walker <robert.walker@arm.com>
Cc: Suzuki K Poulouse <suzuki.poulose@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/20190322052255.GA4809@w-OptiPlex-7050
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch enables showing bpf program name, address, and size in the
header.
Before the patch:
perf report --header-only
...
# bpf_prog_info of id 9
# bpf_prog_info of id 10
# bpf_prog_info of id 13
After the patch:
# bpf_prog_info 9: bpf_prog_7be49e3934a125ba addr 0xffffffffa0024947 size 229
# bpf_prog_info 10: bpf_prog_2a142ef67aaad174 addr 0xffffffffa007c94d size 229
# bpf_prog_info 13: bpf_prog_47368425825d7384_task__task_newt addr 0xffffffffa0251137 size 369
Committer notes:
Fix the fallback definition when HAVE_LIBBPF_SUPPORT is not defined,
i.e. add the missing 'static inline' and add the __maybe_unused to the
args. Also add stdio.h since we now use FILE * in bpf-event.h.
Signed-off-by: Song Liu <songliubraving@fb.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stanislav Fomichev <sdf@google.com>
Link: http://lkml.kernel.org/r/20190319165454.1298742-3-songliubraving@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Extract logic to create program names to synthesize_bpf_prog_name(), so
that it can be reused in header.c:print_bpf_prog_info().
This commit doesn't change the behavior.
Signed-off-by: Song Liu <songliubraving@fb.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stanislav Fomichev <sdf@google.com>
Link: http://lkml.kernel.org/r/20190319165454.1298742-2-songliubraving@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
