linux-stable.git/tools/perf/util, branch v5.4.48

perf symbols: Fix kernel maps for kcore and eBPF

2020-06-22T07:31:27+00:00

commit 0affd0e5262b6d40f5f63466d88933e99698e240 upstream.

Adjust 'map->pgoff' also when moving a map's start address.

Example with v5.4.34 based kernel:

  Before:

    $ sudo tools/perf/perf record -a --kcore -e intel_pt//k sleep 1
    [ perf record: Woken up 1 times to write data ]
    [ perf record: Captured and wrote 1.958 MB perf.data ]
    $ sudo tools/perf/perf script --itrace=e >/dev/null
    Warning:
    961 instruction trace errors

  After:

    $ sudo tools/perf/perf script --itrace=e >/dev/null
    $

Committer testing:

  # uname -a
  Linux seventh 5.6.10-100.fc30.x86_64 #1 SMP Mon May 4 15:36:44 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
  #

Before:

  # perf record -a --kcore -e intel_pt//k sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.923 MB perf.data ]
  # perf script --itrace=e >/dev/null
  Warning:
  295 instruction trace errors
  #

After:

  # perf record -a --kcore -e intel_pt//k sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.919 MB perf.data ]
  # perf script --itrace=e >/dev/null
  #

Fixes: fb5a88d4131a ("perf tools: Preserve eBPF maps when loading kcore")
Signed-off-by: Adrian Hunter 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/20200602112505.1406-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf symbols: Fix debuginfo search for Ubuntu

2020-06-22T07:31:27+00:00

commit 85afd35575a3c1a3a905722dde5ee70b49282e70 upstream.

Reportedly, from 19.10 Ubuntu has begun mixing up the location of some
debug symbol files, putting files expected to be in
/usr/lib/debug/usr/lib into /usr/lib/debug/lib instead. Fix by adding
another dso_binary_type.

Example on Ubuntu 20.04

  Before:

    $ perf record -e intel_pt//u uname
    Linux
    [ perf record: Woken up 1 times to write data ]
    [ perf record: Captured and wrote 0.030 MB perf.data ]
    $ perf script --call-trace | head -5
           uname 14003 [005] 15321.764958566:  cbr: 42 freq: 4219 MHz (156%)
           uname 14003 [005] 15321.764958566: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )          7f1e71cc4100
           uname 14003 [005] 15321.764961566: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )              7f1e71cc4df0
           uname 14003 [005] 15321.764961900: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )              7f1e71cc4e18
           uname 14003 [005] 15321.764963233: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )              7f1e71cc5128

  After:

    $ perf script --call-trace | head -5
           uname 14003 [005] 15321.764958566:  cbr: 42 freq: 4219 MHz (156%)
           uname 14003 [005] 15321.764958566: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )      _start
           uname 14003 [005] 15321.764961566: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )          _dl_start
           uname 14003 [005] 15321.764961900: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )          _dl_start
           uname 14003 [005] 15321.764963233: (/usr/lib/x86_64-linux-gnu/ld-2.31.so              )          _dl_start

Reported-by: Travis Downs 
Signed-off-by: Adrian Hunter 
Cc: Jiri Olsa 
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/20200526155207.9172-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf probe: Check address correctness by map instead of _etext

2020-06-22T07:31:27+00:00

commit 2ae5d0d7d8868df7c05c2013c0b9cddd4d40610e upstream.

Since commit 03db8b583d1c ("perf tools: Fix
maps__find_symbol_by_name()") introduced map address range check in
maps__find_symbol_by_name(), we can not get "_etext" from kernel map
because _etext is placed on the edge of the kernel .text section (=
kernel map in perf.)

To fix this issue, this checks the address correctness by map address
range information (map->start and map->end) instead of using _etext
address.

This can cause an error if the target inlined function is embedded in
both __init function and normal function.

For exaample, request_resource() is a normal function but also embedded
in __init reserve_setup(). In this case, the probe point in
reserve_setup() must be skipped.

However, without this fix, it failes to setup all probe points:

  # ./perf probe -v request_resource
  probe-definition(0): request_resource
  symbol:request_resource file:(null) line:0 offset:0 return:0 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /usr/lib/debug/lib/modules/5.5.17-200.fc31.x86_64/vmlinux for symbols
  Open Debuginfo file: /usr/lib/debug/lib/modules/5.5.17-200.fc31.x86_64/vmlinux
  Try to find probe point from debuginfo.
  Matched function: request_resource [15e29ad]
  found inline addr: 0xffffffff82fbf892
  Probe point found: reserve_setup+204
  found inline addr: 0xffffffff810e9790
  Probe point found: request_resource+0
  Found 2 probe_trace_events.
  Opening /sys/kernel/debug/tracing//kprobe_events write=1
  Opening /sys/kernel/debug/tracing//README write=0
  Writing event: p:probe/request_resource _text+33290386
  Failed to write event: Invalid argument
    Error: Failed to add events. Reason: Invalid argument (Code: -22)
  #

With this fix,

  # ./perf probe request_resource
  reserve_setup is out of .text, skip it.
  Added new events:
    (null):(null)        (on request_resource)
    probe:request_resource (on request_resource)

  You can now use it in all perf tools, such as:

  	perf record -e probe:request_resource -aR sleep 1

  #

Fixes: 03db8b583d1c ("perf tools: Fix maps__find_symbol_by_name()")
Reported-by: Arnaldo Carvalho de Melo 
Signed-off-by: Masami Hiramatsu 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/158763967332.30755.4922496724365529088.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf probe: Fix to check blacklist address correctly

2020-06-22T07:31:27+00:00

commit 80526491c2ca6abc028c0f0dbb0707a1f35fb18a upstream.

Fix to check kprobe blacklist address correctly with relocated address
by adjusting debuginfo address.

Since the address in the debuginfo is same as objdump, it is different
from relocated kernel address with KASLR.  Thus, 'perf probe' always
misses to catch the blacklisted addresses.

Without this patch, 'perf probe' can not detect the blacklist addresses
on a KASLR enabled kernel.

  # perf probe kprobe_dispatcher
  Failed to write event: Invalid argument
    Error: Failed to add events.
  #

With this patch, it correctly shows the error message.

  # perf probe kprobe_dispatcher
  kprobe_dispatcher is blacklisted function, skip it.
  Probe point 'kprobe_dispatcher' not found.
    Error: Failed to add events.
  #

Fixes: 9aaf5a5f479b ("perf probe: Check kprobes blacklist when adding new events")
Signed-off-by: Masami Hiramatsu 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/158763966411.30755.5882376357738273695.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf probe: Accept the instance number of kretprobe event

2020-06-17T14:40:23+00:00

[ Upstream commit c6aab66a728b6518772c74bd9dff66e1a1c652fd ]

Since the commit 6a13a0d7b4d1 ("ftrace/kprobe: Show the maxactive number
on kprobe_events") introduced to show the instance number of kretprobe
events, the length of the 1st format of the kprobe event will not 1, but
it can be longer.  This caused a parser error in perf-probe.

Skip the length check the 1st format of the kprobe event to accept this
instance number.

Without this fix:

  # perf probe -a vfs_read%return
  Added new event:
    probe:vfs_read__return (on vfs_read%return)

  You can now use it in all perf tools, such as:

  	perf record -e probe:vfs_read__return -aR sleep 1

  # perf probe -l
  Semantic error :Failed to parse event name: r16:probe/vfs_read__return
    Error: Failed to show event list.

And with this fixes:

  # perf probe -a vfs_read%return
  ...
  # perf probe -l
    probe:vfs_read__return (on vfs_read%return)

Fixes: 6a13a0d7b4d1 ("ftrace/kprobe: Show the maxactive number on kprobe_events")
Reported-by: Yuxuan Shui 
Signed-off-by: Masami Hiramatsu 
Tested-by: Yuxuan Shui 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: stable@vger.kernel.org
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=207587
Link: http://lore.kernel.org/lkml/158877535215.26469.1113127926699134067.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Sasha Levin

perf: Make perf able to build with latest libbfd

2020-06-03T06:21:39+00:00

commit 0ada120c883d4f1f6aafd01cf0fbb10d8bbba015 upstream.

libbfd has changed the bfd_section_* macros to inline functions
bfd_section_ since 2019-09-18. See below two commits:
  o http://www.sourceware.org/ml/gdb-cvs/2019-09/msg00064.html
  o https://www.sourceware.org/ml/gdb-cvs/2019-09/msg00072.html

This fix make perf able to build with both old and new libbfd.

Signed-off-by: Changbin Du 
Acked-by: Jiri Olsa 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20200128152938.31413-1-changbin.du@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo 
Cc: Marek Vasut 
Signed-off-by: Greg Kroah-Hartman

perf map: Fix off by one in strncpy() size argument

2020-04-02T13:11:02+00:00

commit db2c549407d4a76563c579e4768f7d6d32afefba upstream.

This patch fixes an off-by-one error in strncpy size argument in
tools/perf/util/map.c. The issue is that in:

        strncmp(filename, "/system/lib/", 11)

the passed string literal: "/system/lib/" has 12 bytes (without the NULL
byte) and the passed size argument is 11. As a result, the logic won't
match the ending "/" byte and will pass filepaths that are stored in
other directories e.g. "/system/libmalicious/bin" or just
"/system/libmalicious".

This functionality seems to be present only on Android. I assume the
/system/ directory is only writable by the root user, so I don't think
this bug has much (or any) security impact.

Fixes: eca818369996 ("perf tools: Add automatic remapping of Android libraries")
Signed-off-by: disconnect3d 
Cc: Alexander Shishkin 
Cc: Changbin Du 
Cc: Jiri Olsa 
Cc: John Keeping 
Cc: Mark Rutland 
Cc: Michael Lentine 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Song Liu 
Cc: Stephane Eranian 
Link: http://lore.kernel.org/lkml/20200309104855.3775-1-dominik.b.czarnota@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf probe: Do not depend on dwfl_module_addrsym()

2020-04-01T09:02:00+00:00

commit 1efde2754275dbd9d11c6e0132a4f09facf297ab upstream.

Do not depend on dwfl_module_addrsym() because it can fail on user-space
shared libraries.

Actually, same bug was fixed by commit 664fee3dc379 ("perf probe: Do not
use dwfl_module_addrsym if dwarf_diename finds symbol name"), but commit
07d369857808 ("perf probe: Fix wrong address verification) reverted to
get actual symbol address from symtab.

This fixes it again by getting symbol address from DIE, and only if the
DIE has only address range, it uses dwfl_module_addrsym().

Fixes: 07d369857808 ("perf probe: Fix wrong address verification)
Reported-by: Alexandre Ghiti 
Signed-off-by: Masami Hiramatsu 
Tested-by: Alexandre Ghiti 
Cc: Alexander Shishkin 
Cc: Greg Kroah-Hartman 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Sasha Levin 
Link: http://lore.kernel.org/lkml/158281812176.476.14164573830975116234.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf probe: Fix to delete multiple probe event

2020-04-01T09:01:59+00:00

commit 6b8d68f1ce9266b05a55e93c62923ff51daae4c1 upstream.

When we put an event with multiple probes, perf-probe fails to delete
with filters. This comes from a failure to list up the event name
because of overwrapping its name.

To fix this issue, skip to list up the event which has same name.

Without this patch:

  # perf probe -l \*
    probe_perf:map__map_ip (on perf_sample__fprintf_brstackoff:21@
    probe_perf:map__map_ip (on perf_sample__fprintf_brstackoff:25@
    probe_perf:map__map_ip (on append_inlines:12@util/machine.c in
    probe_perf:map__map_ip (on unwind_entry:19@util/machine.c in /
    probe_perf:map__map_ip (on map__map_ip@util/map.h in /home/mhi
    probe_perf:map__map_ip (on map__map_ip@util/map.h in /home/mhi
  # perf probe -d \*
  "*" does not hit any event.
    Error: Failed to delete events. Reason: No such file or directory (Code: -2)

With it:

  # perf probe -d \*
  Removed event: probe_perf:map__map_ip
  #

Fixes: 72363540c009 ("perf probe: Support multiprobe event")
Reported-by: Arnaldo Carvalho de Melo 
Reported-by: He Zhe 
Signed-off-by: Masami Hiramatsu 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Adrian Hunter 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/158287666197.16697.7514373548551863562.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf stat: Don't report a null stalled cycles per insn metric

2020-02-19T18:53:08+00:00

commit 80cc7bb6c104d733bff60ddda09f19139c61507c upstream.

For data collected on machines with front end stalled cycles supported,
such as found on modern AMD CPU families, commit 146540fb545b ("perf
stat: Always separate stalled cycles per insn") introduces a new line in
CSV output with a leading comma that upsets some automated scripts.
Scripts have to use "-e ex_ret_instr" to work around this issue, after
upgrading to a version of perf with that commit.

We could add "if (have_frontend_stalled && !config->csv_sep)" to the not
(total && avg) else clause, to emphasize that CSV users are usually
scripts, and are written to do only what is needed, i.e., they wouldn't
typically invoke "perf stat" without specifying an explicit event list.

But - let alone CSV output - why should users now tolerate a constant
0-reporting extra line in regular terminal output?:

BEFORE:

$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1

 Performance counter stats for 'system wide':

       181,110,981      instructions              #    0.58  insn per cycle
                                                  #    0.00  stalled cycles per insn
       309,876,469      cycles

       1.002202582 seconds time elapsed

The user would not like to see the now permanent:

  "0.00  stalled cycles per insn"

line fixture, as it gives no useful information.

So this patch removes the printing of the zeroed stalled cycles line
altogether, almost reverting the very original commit fb4605ba47e7
("perf stat: Check for frontend stalled for metrics"), which seems like
it was written to normalize --metric-only column output of common Intel
machines at the time: modern Intel machines have ceased to support the
genericised frontend stalled metrics AFAICT.

AFTER:

$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1

 Performance counter stats for 'system wide':

       244,071,432      instructions              #    0.69  insn per cycle
       355,353,490      cycles

       1.001862516 seconds time elapsed

Output behaviour when stalled cycles is indeed measured is not affected
(BEFORE == AFTER):

$ sudo perf stat --all-cpus -einstructions,cycles,stalled-cycles-frontend -- sleep 1

 Performance counter stats for 'system wide':

       247,227,799      instructions              #    0.63  insn per cycle
                                                  #    0.26  stalled cycles per insn
       394,745,636      cycles
        63,194,485      stalled-cycles-frontend   #   16.01% frontend cycles idle

       1.002079770 seconds time elapsed

Fixes: 146540fb545b ("perf stat: Always separate stalled cycles per insn")
Signed-off-by: Kim Phillips 
Acked-by: Andi Kleen 
Acked-by: Jiri Olsa 
Acked-by: Song Liu 
Cc: Alexander Shishkin 
Cc: Cong Wang 
Cc: Davidlohr Bueso 
Cc: Jin Yao 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20200207230613.26709-1-kim.phillips@amd.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman