linux.git/tools/perf, branch v5.6-rc5

perf annotate: Fix segfault with source toggle

2020-02-27T14:47:23+00:00

While rendering annotate browser from perf report tui, we keep track
of total number of lines(asm + source) in annotation->nr_entries and
total number of asm lines in annotation->nr_asm_entries. But we don't
reset them before starting. Thus if user annotates same function
multiple times, we restart incrementing these fields with old values.

This causes a segfault when user tries to toggle source code after
annotating same function multiple times. Fix it.

Signed-off-by: Ravi Bangoria 
Tested-by: Arnaldo Carvalho de Melo 
Acked-by: Jiri Olsa 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Namhyung Kim 
Cc: Song Liu 
Link: http://lore.kernel.org/lkml/20200204045233.474937-5-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Align struct annotate_args

2020-02-27T14:47:23+00:00

Align fields of struct annotate_args.

Signed-off-by: Ravi Bangoria 
Acked-by: Jiri Olsa 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Namhyung Kim 
Cc: Song Liu 
Link: http://lore.kernel.org/lkml/20200204045233.474937-4-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Simplify disasm_line allocation and freeing code

2020-02-27T14:07:13+00:00

We are allocating disasm_line object in annotation_line__new() instead
of disasm_line__new(). Similarly annotation_line__delete() is actually
freeing disasm_line object as well. This complexity is because of
privsize.  But we don't need privsize anymore so get rid of privsize and
simplify disasm_line allocation and freeing code.

Signed-off-by: Ravi Bangoria 
Acked-by: Jiri Olsa 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Namhyung Kim 
Cc: Song Liu 
Link: http://lore.kernel.org/lkml/20200204045233.474937-3-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Remove privsize from symbol__annotate() args

2020-02-27T14:06:14+00:00

privsize is passed as 0 from all the symbol__annotate() callers.
Remove it from argument list.

Signed-off-by: Ravi Bangoria 
Acked-by: Jiri Olsa 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Namhyung Kim 
Cc: Song Liu 
Link: http://lore.kernel.org/lkml/20200204045233.474937-2-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf probe: Check return value of strlist__add() for -ENOMEM

2020-02-27T14:03:13+00:00

strlist__add() may fail with -ENOMEM. Check it and give debugging hint
in advance.

Signed-off-by: He Zhe 
Acked-by: Jiri Olsa 
Cc: Alexander Shishkin 
Cc: Kate Stewart 
Cc: Mark Rutland 
Cc: Masami Hiramatsu 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Link: http://lore.kernel.org/lkml/1582727404-180095-1-git-send-email-zhe.he@windriver.com
Signed-off-by: Arnaldo Carvalho de Melo

perf config: Document missing config options

2020-02-27T13:45:19+00:00

While documenting annotate.show_nr_samples config option, I found many
other config options missing in perf-config documentation. Add them.

Signed-off-by: Ravi Bangoria 
Cc: Adrian Hunter 
Cc: Alexey Budankov 
Cc: Changbin Du 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Namhyung Kim 
Cc: Song Liu 
Cc: Taeung Song 
Cc: Thomas Richter 
Cc: Yisheng Xie 
Link: http://lore.kernel.org/lkml/20200213064306.160480-9-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Fix perf config option description

2020-02-27T13:45:13+00:00

perf config annotate options says it works only with TUI, which is wrong.
Most of the TUI options are applicable to stdio2 as well. So remove that
generic line and add individual line with each option stating which
browsers supports that option. Also, annotate.show_nr_samples config is
missing in Documentation. Describe it.

Signed-off-by: Ravi Bangoria 
Cc: Adrian Hunter 
Cc: Alexey Budankov 
Cc: Changbin Du 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Namhyung Kim 
Cc: Song Liu 
Cc: Taeung Song 
Cc: Thomas Richter 
Cc: Yisheng Xie 
Link: http://lore.kernel.org/lkml/20200213064306.160480-8-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Prefer cmdline option over default config

2020-02-27T13:45:08+00:00

For all the perf-config options that can also be set from command line
option, the preference is given to command line version in case of any
conflict. But that's opposite in case of perf annotate. i.e. the more
preference is given to default option rather than command line option.
Fix it.

Before:

  $ ./perf config
  annotate.show_nr_samples=false

  $ ./perf annotate shash --show-nr-samples
  Percent│
         │24:   mov    -0xc(%rbp),%eax
   49.19 │      imul   $0x1003f,%eax,%ecx
         │      mov    -0x18(%rbp),%rax

After:

  Samples│
         │24:   mov    -0xc(%rbp),%eax
       1 │      imul   $0x1003f,%eax,%ecx
         │      mov    -0x18(%rbp),%rax

Signed-off-by: Ravi Bangoria 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Adrian Hunter 
Cc: Alexey Budankov 
Cc: Changbin Du 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Namhyung Kim 
Cc: Song Liu 
Cc: Taeung Song 
Cc: Thomas Richter 
Cc: Yisheng Xie 
Link: http://lore.kernel.org/lkml/20200213064306.160480-7-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Make perf config effective

2020-02-27T13:44:59+00:00

perf default config set by user in [annotate] section is totally ignored
by annotate code. Fix it.

Before:

  $ ./perf config
  annotate.hide_src_code=true
  annotate.show_nr_jumps=true
  annotate.show_nr_samples=true

  $ ./perf annotate shash
         │    unsigned h = 0;
         │      movl   $0x0,-0xc(%rbp)
         │    while (*s)
         │    ↓ jmp    44
         │    h = 65599 * h + *s++;
   11.33 │24:   mov    -0xc(%rbp),%eax
   43.50 │      imul   $0x1003f,%eax,%ecx
         │      mov    -0x18(%rbp),%rax

After:

         │        movl   $0x0,-0xc(%rbp)
         │      ↓ jmp    44
       1 │1 24:   mov    -0xc(%rbp),%eax
       4 │        imul   $0x1003f,%eax,%ecx
         │        mov    -0x18(%rbp),%rax

Note that we have removed show_nr_samples and show_total_period from
annotation_options because they are not used. Instead of them we use
symbol_conf.show_nr_samples and symbol_conf.show_total_period.

Committer testing:

Using 'perf annotate --stdio2' to use the TUI rendering but emitting the output to stdio:

  # perf config
  #
  # perf config annotate.hide_src_code=true
  # perf config
  annotate.hide_src_code=true
  #
  # perf config annotate.show_nr_jumps=true
  # perf config annotate.show_nr_samples=true
  # perf config
  annotate.hide_src_code=true
  annotate.show_nr_jumps=true
  annotate.show_nr_samples=true
  #
  #

Before:

  # perf annotate --stdio2 ObjectInstance::weak_pointer_was_finalized
  Samples: 1  of event 'cycles', 4000 Hz, Event count (approx.): 830873, [percent: local period]
  ObjectInstance::weak_pointer_was_finalized() /usr/lib64/libgjs.so.0.0.0
  Percent
              00000000000609f0 :
                endbr64
                cmpq    $0x0,0x20(%rdi)
              ↓ je      10
                xor     %eax,%eax
              ← retq
                xchg    %ax,%ax
  100.00  10:   push    %rbp
                cmpq    $0x0,0x18(%rdi)
                mov     %rdi,%rbp
              ↓ jne     20
          1b:   xor     %eax,%eax
                pop     %rbp
              ← retq
                nop
          20:   lea     0x18(%rdi),%rdi
              → callq   JS_UpdateWeakPointerAfterGC(JS::Heap /dev/null
  Samples: 1  of event 'cycles', 4000 Hz, Event count (approx.): 830873, [percent: local period]
  ObjectInstance::weak_pointer_was_finalized() /usr/lib64/libgjs.so.0.0.0
  Samples       endbr64
                cmpq    $0x0,0x20(%rdi)
              ↓ je      10
                xor     %eax,%eax
              ← retq
                xchg    %ax,%ax
     1  1 10:   push    %rbp
                cmpq    $0x0,0x18(%rdi)
                mov     %rdi,%rbp
              ↓ jne     20
        1 1b:   xor     %eax,%eax
                pop     %rbp
              ← retq
                nop
        1 20:   lea     0x18(%rdi),%rdi
              → callq   JS_UpdateWeakPointerAfterGC(JS::Heap /dev/null
  Samples: 1  of event 'cycles', 4000 Hz, Event count (approx.): 830873, [percent: local period]
  ObjectInstance::weak_pointer_was_finalized() /usr/lib64/libgjs.so.0.0.0
  Samples       endbr64
                cmpq    $0x0,0x20(%rdi)
              ↓ je      10
                xor     %eax,%eax
              ← retq
                xchg    %ax,%ax
       1  10:   push    %rbp
                cmpq    $0x0,0x18(%rdi)
                mov     %rdi,%rbp
              ↓ jne     20
          1b:   xor     %eax,%eax
                pop     %rbp
              ← retq
                nop
          20:   lea     0x18(%rdi),%rdi
              → callq   JS_UpdateWeakPointerAfterGC(JS::Heap
Tested-by: Arnaldo Carvalho de Melo 
Cc: Adrian Hunter 
Cc: Alexey Budankov 
Cc: Changbin Du 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Namhyung Kim 
Cc: Song Liu 
Cc: Taeung Song 
Cc: Thomas Richter 
Cc: Yisheng Xie 
Link: http://lore.kernel.org/lkml/20200213064306.160480-6-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf config: Introduce perf_config_u8()

2020-02-27T13:44:54+00:00

Introduce perf_config_u8() utility function to convert char * input into
u8 destination. We will utilize it in followup patch.

Signed-off-by: Ravi Bangoria 
Cc: Adrian Hunter 
Cc: Alexey Budankov 
Cc: Changbin Du 
Cc: Ian Rogers 
Cc: Jin Yao 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Namhyung Kim 
Cc: Song Liu 
Cc: Taeung Song 
Cc: Thomas Richter 
Cc: Yisheng Xie 
Link: http://lore.kernel.org/lkml/20200213064306.160480-5-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo