Skip to content

L
linux
Switch branch/tag
  1. 14 Aug, 2019 15 commits
  3. 12 Aug, 2019 13 commits
    • Igor Lubashev's avatar
      tools build: Add capability-related feature detection · 74d5f3d0
      Igor Lubashev authored 
      Add utilities to help checking capabilities of the running procss.  Make
perf link with libcap, if it is available. If no libcap-dev[el], assume
no capabilities.

Committer testing:

  $ make O=/tmp/build/perf -C tools/perf install-bin
  make: Entering directory '/home/acme/git/perf/tools/perf'
    BUILD:   Doing 'make -j8' parallel build

  Auto-detecting system features:
  <SNIP>
  ...                        libbfd: [ on  ]
  ...                        libcap: [ OFF ]
  ...                        libelf: [ on  ]
  <SNIP>
  Makefile.config:833: No libcap found, disables capability support, please install libcap-devel/libcap-dev
  <SNIP>
  $ grep libcap /tmp/build/perf/FEATURE-DUMP
  feature-libcap=0
  $ cat /tmp/build/perf/feature/test-libcap.make.output
  test-libcap.c:2:10: fatal error: sys/capability.h: No such file or directory
      2 | #include <sys/capability.h>
        |          ^~~~~~~~~~~~~~~~~~
  compilation terminated.
  $

Now install libcap-devel and try again:

  $ make O=/tmp/build/perf -C tools/perf install-bin
  make: Entering directory '/home/acme/git/perf/tools/perf'
    BUILD:   Doing 'make -j8' parallel build
  Warning: Kernel ABI header at 'tools/include/linux/bits.h' differs from latest version at 'include/linux/bits.h'
  diff -u tools/include/linux/bits.h include/linux/bits.h
  Warning: Kernel ABI header at 'tools/arch/x86/include/asm/cpufeatures.h' differs from latest version at 'arch/x86/include/asm/cpufeatures.h'
  diff -u tools/arch/x86/include/asm/cpufeatures.h arch/x86/include/asm/cpufeatures.h

  Auto-detecting system features:
  <SNIP>
  ...                        libbfd: [ on  ]
  ...                        libcap: [ on  ]
  ...                        libelf: [ on  ]
  <SNIP>>
    CC       /tmp/build/perf/jvmti/libjvmti.o
  <SNIP>>
  $ grep libcap /tmp/build/perf/FEATURE-DUMP
  feature-libcap=1
  $ cat /tmp/build/perf/feature/test-libcap.make.output
  $ ldd /tmp/build/perf/feature/test-libcap.make.bin
  ldd: /tmp/build/perf/feature/test-libcap.make.bin: No such file or directory
  $ ldd /tmp/build/perf/feature/test-libcap.bin
  	linux-vdso.so.1 (0x00007ffc35bfe000)
  	libcap.so.2 => /lib64/libcap.so.2 (0x00007ff9c62ff000)
  	libc.so.6 => /lib64/libc.so.6 (0x00007ff9c6139000)
  	/lib64/ld-linux-x86-64.so.2 (0x00007ff9c6326000)
  $
Signed-off-by: default avatarIgor Lubashev <ilubashe@akamai.com>
Acked-by: default avatarJiri Olsa <jolsa@kernel.org>
Tested-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: James Morris <jmorris@namei.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
[ split from a larger patch ]
Link: http://lkml.kernel.org/r/8a1e76cf5c7c9796d0d4d240fbaa85305298aafa.1565188228.git.ilubashe@akamai.comSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      74d5f3d0
    • Arnaldo Carvalho de Melo's avatar
      perf top: Collapse and resort all evsels in a group · 40d81772
      Arnaldo Carvalho de Melo authored 
      And link them, i.e. find the hist entries in the non-leader events and
link them to the ones in the leader.

This should be the same thing already done for the 'perf report' case,
but now we do it periodically.

With this in place we get percentages in from the second overhead column
on, not just on the first (the leader).

Try it using:

  perf top --stdio -e '{cycles,instructions}'

You should see something like:

   PerfTop:   20776 irqs/sec  kernel:68.7%  exact:  0.0% lost: 0/0 drop: 0/0 [cycles],  (all, 8 CPUs)
  ---------------------------------------------------------------------------------------------------

     4.44%   0.44%  [kernel]                 [k] do_syscall_64
     2.27%   0.17%  [kernel]                 [k] entry_SYSCALL_64
     1.73%   0.27%  [kernel]                 [k] syscall_return_via_sysret
     1.60%   0.91%  [kernel]                 [k] _raw_spin_lock_irqsave
     1.45%   3.53%  libglib-2.0.so.0.6000.4  [.] g_string_insert_unichar
     1.39%   0.21%  [kernel]                 [k] copy_user_enhanced_fast_string
     1.26%   1.15%  [kernel]                 [k] psi_task_change
     1.16%   0.14%  libpixman-1.so.0.38.0    [.] 0x000000000006f403
     1.00%   0.32%  [kernel]                 [k] __sched_text_start
     0.97%   2.11%  [kernel]                 [k] n_tty_write
     0.96%   0.04%  [kernel]                 [k] queued_spin_lock_slowpath
     0.93%   0.88%  [kernel]                 [k] menu_select
     0.87%   0.14%  [kernel]                 [k] try_to_wake_up
     0.77%   0.10%  libpixman-1.so.0.38.0    [.] 0x000000000006f40b
     0.73%   0.09%  libpixman-1.so.0.38.0    [.] 0x000000000006f413
     0.69%   0.48%  libc-2.29.so             [.] __memmove_avx_unaligned_erms
     0.68%   0.29%  [kernel]                 [k] _raw_spin_lock_irq
     0.61%   0.04%  libpixman-1.so.0.38.0    [.] 0x000000000006f423
     0.60%   0.37%  [kernel]                 [k] native_sched_clock
     0.57%   0.23%  [kernel]                 [k] do_idle
     0.57%   0.23%  [kernel]                 [k] __fget
     0.56%   0.30%  [kernel]                 [k] __switch_to_asm
     0.56%   0.00%  libc-2.29.so             [.] __memset_avx2_erms
     0.52%   0.32%  [kernel]                 [k] _raw_spin_lock
     0.49%   0.24%  [kernel]                 [k] n_tty_poll
     0.49%   0.54%  libglib-2.0.so.0.6000.4  [.] g_mutex_lock
     0.48%   0.62%  [kernel]                 [k] _raw_spin_unlock_irqrestore
     0.47%   0.27%  [kernel]                 [k] __switch_to
     0.47%   0.25%  [kernel]                 [k] pick_next_task_fair
     0.45%   0.17%  [kernel]                 [k] filldir64
     0.40%   0.16%  [kernel]                 [k] update_rq_clock
     0.39%   0.19%  [kernel]                 [k] enqueue_task_fair
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-uw8cjeifxvjpkjp6x2iil0ar@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      40d81772
    • Arnaldo Carvalho de Melo's avatar
      perf hist: Remove dummy entries when finding real ones. · 5f8b4d5d
      Arnaldo Carvalho de Melo authored 
      When he have an event group we have multiple struct hist instances, one
per evsel, and in each of these hists we may have hist_entries that
point to the same thing being observed, say a symbol, i.e. if we're
looking at instructions and cycles, then we'll have one hist_entry in
the "instructions" evsel and another in the "cycles" evsel.

We need to link those to then show one column for each. When we're
looking at some other pair of events, say instructions and cache misses,
we may have just the "instructions" hist entry and not one for "cache
misses", as instructions not necessarily generate cache misses, as the
logic expects one hist_entry per evsel, we end up adding "dummy"
hist_entries.

This is enough for 'perf report', that does this matching operation
(hists__match()) just once after processing all events, but for 'perf
top', we do this at each refresh, so we may finally find events matching
and then we need to trow away the dummies and link with the real events.

So if we find a match, traverse the link of matches and trow away
dummies for that hists.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-dwvtjqqifsbsczeb35q6mqkk@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      5f8b4d5d
    • Leo Yan's avatar
      perf trace: Fix segmentation fault when access syscall info on arm64 · 3e70008a
      Leo Yan authored 
      'perf trace' reports the segmentation fault as below on Arm64:

  # perf trace -e string -e augmented_raw_syscalls.c
  LLVM: dumping tools/perf/examples/bpf/augmented_raw_syscalls.o
  perf: Segmentation fault
  Obtained 12 stack frames.
  perf(sighandler_dump_stack+0x47) [0xaaaaac96ac87]
  linux-vdso.so.1(+0x5b7) [0xffffadbeb5b7]
  /lib/aarch64-linux-gnu/libc.so.6(strlen+0x10) [0xfffface7d5d0]
  /lib/aarch64-linux-gnu/libc.so.6(_IO_vfprintf+0x1ac7) [0xfffface49f97]
  /lib/aarch64-linux-gnu/libc.so.6(__vsnprintf_chk+0xc7) [0xffffacedfbe7]
  perf(scnprintf+0x97) [0xaaaaac9ca3ff]
  perf(+0x997bb) [0xaaaaac8e37bb]
  perf(cmd_trace+0x28e7) [0xaaaaac8ec09f]
  perf(+0xd4a13) [0xaaaaac91ea13]
  perf(main+0x62f) [0xaaaaac8a147f]
  /lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0xe3) [0xfffface22d23]
  perf(+0x57723) [0xaaaaac8a1723]
  Segmentation fault

This issue is introduced by commit 30a910d7 ("perf trace:
Preallocate the syscall table"), it allocates trace->syscalls.table[]
array and the element count is 'trace->sctbl->syscalls.nr_entries'; but
on Arm64, the system call number is not continuously used; e.g. the
syscall maximum id is 436 but the real entries is only 281.

So the table is allocated with 'nr_entries' as the element count, but it
accesses the table with the syscall id, which might be out of the bound
of the array and cause the segmentation fault.

This patch allocates trace->syscalls.table[] with the element count is
'trace->sctbl->syscalls.max_id + 1', this allows any id to access the
table without out of the bound.
Signed-off-by: default avatarLeo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Fixes: 30a910d7 ("perf trace: Preallocate the syscall table")
Link: http://lkml.kernel.org/r/20190809104752.27338-1-leo.yan@linaro.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      3e70008a
    • Arnaldo Carvalho de Melo's avatar
      perf hists: Do not link a pair if already linked · 7d1a5efa
      Arnaldo Carvalho de Melo authored 
      When we have multiple events in a group we link hist_entries in the
non-leader evsel hists to the one in the leader that points to the same
sorting criteria, in hists__match().

For 'perf report' we do this just once and then print the results, but
for 'perf top' we need to look if this was already done in the previous
refresh of the screen, so check for that and don't try to link again.

This is part of having 'perf top' using the hists browser for showing
multiple events in multiple columns.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-iwvb37rgb7upswhruwpcdnhw@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      7d1a5efa
    • Arnaldo Carvalho de Melo's avatar
      perf top: Set display thread COMM to help with debugging · 1205a271
      Arnaldo Carvalho de Melo authored 
      When we want to attach just to the thread that updates the display it
helps having its COMM stand out, so change it from the default "perf" to
"perf-top-UI".

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-5w0hmlk3zfvysxvpsh763k9w@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      1205a271
    • Haiyan Song's avatar
      perf vendor events intel: Add Icelake V1.00 event file · b115df07
      Haiyan Song authored 
      Add a Intel event file for perf.
Signed-off-by: default avatarHaiyan Song <haiyanx.song@intel.com>
Reviewed-by: default avatarKan Liang <kan.liang@linux.intel.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: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/8859095e-5b02-d6b7-fbdc-3f42b714bae0@intel.comSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      b115df07
    • Luke Mujica's avatar
      perf tools: Fix paths in include statements · 2b75863b
      Luke Mujica authored 
      These paths point to the wrong location but still work because they get
picked up by a -I flag that happens to direct to the correct file. Fix
paths to lead to the actual file location without help from include
flags.
Signed-off-by: default avatarLuke Mujica <lukemujica@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20190719202253.220261-1-lukemujica@google.comSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      2b75863b
    • Arnaldo Carvalho de Melo's avatar
      perf test vfs_getname: Disable ~/.perfconfig to get default output · 4fe94ce1
      Arnaldo Carvalho de Melo authored 
      To get the expected output we have to ignore whatever changes the user
has in its ~/.perfconfig file, so set PERF_CONFIG to /dev/null to
achieve that.

Before:

  # egrep 'trace|show_' ~/.perfconfig
  [trace]
  	show_zeros = yes
  	show_duration = no
  	show_timestamp = no
  	show_arg_names = no
  	show_prefix = yes
  # echo $PERF_CONFIG

  # perf test "trace + vfs_getname"
  70: Check open filename arg using perf trace + vfs_getname: FAILED!
  # export PERF_CONFIG=/dev/null
  # perf test "trace + vfs_getname"
  70: Check open filename arg using perf trace + vfs_getname: Ok
  #

After:

  # egrep 'trace|show_' ~/.perfconfig
  [trace]
  	show_zeros = yes
  	show_duration = no
  	show_timestamp = no
  	show_arg_names = no
  	show_prefix = yes
  # echo $PERF_CONFIG

  # perf test "trace + vfs_getname"
  70: Check open filename arg using perf trace + vfs_getname: Ok
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Taeung Song <treeze.taeung@gmail.com>
Link: https://lkml.kernel.org/n/tip-3up27pexg5i3exuzqrvt4m8u@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      4fe94ce1
    • Arnaldo Carvalho de Melo's avatar
      perf config: Document the PERF_CONFIG environment variable · 5de9e5fd
      Arnaldo Carvalho de Melo authored 
      There was a provision for setting this variable, but not the
getenv("PERF_CONFIG") call to set it, as this was fixed in the previous
cset, document that it can be used to ask for using an alternative
.perfconfig file or to disable reading whatever file exists in the
system or home directory, i.e. using:

  export PERF_CONFIG=/dev/null

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Taeung Song <treeze.taeung@gmail.com>
Link: https://lkml.kernel.org/n/tip-0u4o967hsk7j0o50zp9ctn89@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      5de9e5fd
    • Arnaldo Carvalho de Melo's avatar
      perf config: Honour $PERF_CONFIG env var to specify alternate .perfconfig · 61a461fc
      Arnaldo Carvalho de Melo authored 
      We had this comment in Documentation/perf_counter/config.c, i.e. since
when we got this from the git sources, but never really did that
getenv("PERF_CONFIG"), do it now as I need to disable whatever
~/.perfconfig root has so that tests parsing tool output are done for
the expected default output or that we specify an alternate config file
that when read will make the tools produce expected output.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Taeung Song <treeze.taeung@gmail.com>
Fixes: 07800601 ("perf_counter tools: add in basic glue from Git")
Link: https://lkml.kernel.org/n/tip-jo209zac9rut0dz1rqvbdlgm@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      61a461fc
    • Arnaldo Carvalho de Melo's avatar
      perf session: Avoid infinite loop when seeing invalid header.size · 57fc032a
      Arnaldo Carvalho de Melo authored 
      Vince reported that when fuzzing the userland perf tool with a bogus
perf.data file he got into a infinite loop in 'perf report'.

Changing the return of fetch_mmaped_event() to ERR_PTR(-EINVAL) for that
case gets us out of that infinite loop.
Reported-by: default avatarVince Weaver <vincent.weaver@maine.edu>
Tested-by: default avatarVince Weaver <vincent.weaver@maine.edu>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20190726211415.GE24867@kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      57fc032a
    • Arnaldo Carvalho de Melo's avatar
      Merge remote-tracking branch 'torvalds/master' into perf/core · 272172bd
      Arnaldo Carvalho de Melo authored 
      To get closer to upstream and check if we need to sync more UAPI
headers, pick up fixes for libbpf that prevent perf's container tests
from completing successfuly, etc.
Signed-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      272172bd
  5. 11 Aug, 2019 3 commits
  7. 10 Aug, 2019 9 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7