linux-stable.git/tools/perf/util/sort.h, branch linux-6.3.y

perf c2c: Add report option to show false sharing in adjacent cachelines

2023-02-16T12:33:45+00:00

Many platforms have feature of adjacent cachelines prefetch, when it is
enabled, for data in RAM of 2 cachelines (2N and 2N+1) granularity, if
one is fetched to cache, the other one could likely be fetched too,
which sort of extends the cacheline size to double, thus the false
sharing could happens in adjacent cachelines.

0Day has captured performance changed related with this [1], and some
commercial software explicitly makes its hot global variables 128 bytes
aligned (2 cache lines) to avoid this kind of extended false sharing.

So add an option "--double-cl" for 'perf c2c report' to show false
sharing in double cache line granularity, which acts just like the
cacheline size is doubled. There is no change to c2c record. The
hardware events of shared cacheline are still per cacheline, and this
option just changes the granularity of how events are grouped and
displayed.

In the 'perf c2c report' output below (will-it-scale's 'pagefault2' case
on old kernel):

  ----------------------------------------------------------------------
     26       31        2        0        0        0  0xffff888103ec6000
  ----------------------------------------------------------------------
   35.48%   50.00%    0.00%    0.00%    0.00%   0x10     0       1  0xffffffff8133148b   1153   66    971   3748   74  [k] get_mem_cgroup_from_mm
    6.45%    0.00%    0.00%    0.00%    0.00%   0x10     0       1  0xffffffff813396e4    570    0   1531    879   75  [k] mem_cgroup_charge
   25.81%   50.00%    0.00%    0.00%    0.00%   0x54     0       1  0xffffffff81331472    949   70    593   3359   74  [k] get_mem_cgroup_from_mm
   19.35%    0.00%    0.00%    0.00%    0.00%   0x54     0       1  0xffffffff81339686   1352    0   1073   1022   74  [k] mem_cgroup_charge
    9.68%    0.00%    0.00%    0.00%    0.00%   0x54     0       1  0xffffffff813396d6   1401    0    863    768   74  [k] mem_cgroup_charge
    3.23%    0.00%    0.00%    0.00%    0.00%   0x54     0       1  0xffffffff81333106    618    0    804     11    9  [k] uncharge_batch

The offset 0x10 and 0x54 used to displayed in 2 groups, and now they are
listed together to give users a hint of extended false sharing.

[1]. https://lore.kernel.org/lkml/20201102091543.GM31092@shao2-debian/

Committer notes:

Link: https://lore.kernel.org/r/Y+wvVNWqXb70l4uy@feng-clx

Removed -a, leaving just as --double-cl, as this probably is not used so
frequently and perhaps will be even auto-detected if we manage to record
the MSR where this is configured.

Reviewed-by: Andi Kleen 
Reviewed-by: Leo Yan 
Signed-off-by: Feng Tang 
Tested-by: Leo Yan 
Acked-by: Joe Mario 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Tim Chen 
Cc: Xing Zhengjun 
Link: https://lore.kernel.org/r/20230214075823.246414-1-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf report: Support Retire Latency

2023-02-03T20:24:02+00:00

The Retire Latency field is added in the var3_w of the
PERF_SAMPLE_WEIGHT_STRUCT. The Retire Latency reports pipeline stall of
this instruction compared to the previous instruction in cycles.  That's
quite useful to display the information with perf mem report.

The p_stage_cyc for Power is also from the var3_w. Union the p_stage_cyc
and retire_lat to share the code.

Implement X86 specific codes to display the X86 specific header.

Add a new sort key retire_lat for the Retire Latency.

Reviewed-by: Andi Kleen 
Signed-off-by: Kan Liang 
Cc: Ian Rogers 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: http://lore.kernel.org/lkml/20230104201349.1451191-8-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf hist: Add perf_hpp_fmt->init() callback

2022-12-21T17:52:40+00:00

In __hists__insert_output_entry(), it calls fmt->sort() for dynamic
entries with NULL to update column width for tracepoint fields.
But it's a hacky abuse of the sort callback, better to have a proper
callback for that.  I'll add more use cases later.

Signed-off-by: Namhyung Kim 
Acked-by: Ian Rogers 
Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Leo Yan 
Cc: Milian Wolff 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/r/20221215192817.2734573-7-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf tools: Add 'addr' sort key

2022-10-04T11:55:22+00:00

Sometimes users want to see actual (virtual) address of sampled instructions.
Add a new 'addr' sort key to display the raw addresses.

  $ perf record -o- true | perf report -i- -s addr
  # To display the perf.data header info, please use --header/--header-only options.
  #
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.000 MB - ]
  #
  # Total Lost Samples: 0
  #
  # Samples: 12  of event 'cycles:u'
  # Event count (approx.): 252512
  #
  # Overhead  Address
  # ........  ..................
  #
      42.96%  0x7f96f08443d7
      29.55%  0x7f96f0859b50
      14.76%  0x7f96f0852e02
       8.30%  0x7f96f0855028
       4.43%  0xffffffff8de01087

Note that it just compares and displays the sample ip.  Each process can
have a different memory layout and the ip will be different even if they run
the same binary.  So this sort key is mostly meaningful for per-process
profile data.

Signed-off-by: Namhyung Kim 
Acked-by: Ian Rogers 
Cc: Adrian Hunter 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220923173142.805896-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf sort: Remove hist_entry__sort_list() and sort__first_dimension() leftover declarations

2022-10-04T11:55:22+00:00

The hist_entry__sort_list and sort__first_dimension functions have been
removed in commit cfaa154b2335d4c8 ("perf tools: Get rid of obsolete
hist_entry__sort_list"), remove them.

Signed-off-by: Gaosheng Cui 
Cc: Alexander Shishkin 
Cc: Athira Jajeev 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220909044542.1087870-2-cuigaosheng1@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo

perf report: Add "addr_from" and "addr_to" sort dimensions

2022-02-16T14:21:22+00:00

With the existing symbol_from/symbol_to, branches captured in the same
function would be collapsed into a single function if the latencies
associated with the each branch (cycles) were all the same.  That is the
case on Intel Broadwell, for instance. Since Intel Skylake, the latency
is captured by hardware and therefore is used to disambiguate branches.

Add addr_from/addr_to sort dimensions to sort branches based on their
addresses and not the function there are in. The output is still the
function name but the offset within the function is provided to uniquely
identify each branch.  These new sort dimensions also help with annotate
because they create different entries in the histogram which, in turn,
generates proper branch annotations.

Here is an example using AMD's branch sampling:

  $ perf record -a -b -c 1000037 -e cpu/branch-brs/ test_prg

  $ perf report
  Samples: 6M of event 'cpu/branch-brs/', Event count (approx.): 6901276
  Overhead  Command          Source Shared Object  Source Symbol                                   Target Symbol                                   Basic Block Cycle
    99.65%  test_prg	   test_prg              [.] test_thread                                 [.] test_thread                                 -
     0.02%  test_prg         [kernel.vmlinux]      [k] asm_sysvec_apic_timer_interrupt             [k] error_entry                                 -

  $ perf report -F overhead,comm,dso,addr_from,addr_to
  Samples: 6M of event 'cpu/branch-brs/', Event count (approx.): 6901276
  Overhead  Command          Shared Object     Source Address          Target Address
     4.22%  test_prg         test_prg          [.] test_thread+0x3c    [.] test_thread+0x4
     4.13%  test_prg         test_prg          [.] test_thread+0x4     [.] test_thread+0x3a
     4.09%  test_prg         test_prg          [.] test_thread+0x3a    [.] test_thread+0x6
     4.08%  test_prg         test_prg          [.] test_thread+0x2     [.] test_thread+0x3c
     4.06%  test_prg         test_prg          [.] test_thread+0x3e    [.] test_thread+0x2
     3.87%  test_prg         test_prg          [.] test_thread+0x6     [.] test_thread+0x38
     3.84%  test_prg         test_prg          [.] test_thread         [.] test_thread+0x3e
     3.76%  test_prg         test_prg          [.] test_thread+0x1e    [.] test_thread
     3.76%  test_prg         test_prg          [.] test_thread+0x38    [.] test_thread+0x8
     3.56%  test_prg         test_prg          [.] test_thread+0x22    [.] test_thread+0x1e
     3.54%  test_prg         test_prg          [.] test_thread+0x8     [.] test_thread+0x36
     3.47%  test_prg         test_prg          [.] test_thread+0x1c    [.] test_thread+0x22
     3.45%  test_prg         test_prg          [.] test_thread+0x36    [.] test_thread+0xa
     3.28%  test_prg         test_prg          [.] test_thread+0x24    [.] test_thread+0x1c
     3.25%  test_prg         test_prg          [.] test_thread+0xa     [.] test_thread+0x34
     3.24%  test_prg         test_prg          [.] test_thread+0x1a    [.] test_thread+0x24
     3.20%  test_prg         test_prg          [.] test_thread+0x34    [.] test_thread+0xc
     3.04%  test_prg         test_prg          [.] test_thread+0x26    [.] test_thread+0x1a
     3.01%  test_prg         test_prg          [.] test_thread+0xc     [.] test_thread+0x32
     2.98%  test_prg         test_prg          [.] test_thread+0x18    [.] test_thread+0x26
     2.94%  test_prg         test_prg          [.] test_thread+0x32    [.] test_thread+0xe
     2.76%  test_prg         test_prg          [.] test_thread+0x28    [.] test_thread+0x18
     2.73%  test_prg         test_prg          [.] test_thread+0xe     [.] test_thread+0x30
     2.67%  test_prg         test_prg          [.] test_thread+0x30    [.] test_thread+0x10
     2.67%  test_prg         test_prg          [.] test_thread+0x16    [.] test_thread+0x28
     2.46%  test_prg         test_prg          [.] test_thread+0x10    [.] test_thread+0x2e
     2.44%  test_prg         test_prg          [.] test_thread+0x2a    [.] test_thread+0x16
     2.38%  test_prg         test_prg          [.] test_thread+0x14    [.] test_thread+0x2a
     2.32%  test_prg         test_prg          [.] test_thread+0x2e    [.] test_thread+0x12
     2.28%  test_prg         test_prg          [.] test_thread+0x12    [.] test_thread+0x2c
     2.16%  test_prg         test_prg          [.] test_thread+0x2c    [.] test_thread+0x14
     0.02%  test_prg         [kernel.vmlinux]  [k] asm_sysvec_apic_ti+0x5  [k] error_entry

Signed-off-by: Stephane Eranian 
Cc: Jiri Olsa 
Cc: Kim Phillips 
Cc: Peter Zijlstra 
Cc: Song Liu 
Link: http://lore.kernel.org/lkml/20220208211637.2221872-13-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf sort: Include global and local variants for p_stage_cyc sort key

2022-01-10T18:39:00+00:00

Sort key 'p_stage_cyc' is used to present the latency cycles spent in
pipeline stages.

perf has local 'p_stage_cyc' sort key to display this info. There is no
global variant available for this sort key. The local variant shows
latency in a single sample, whereas the global value will be useful to
present the total latency (sum of latencies) in the hist entry. It
represents the latency number multiplied by the number of samples.

Add global ('p_stage_cyc') and local variant ('local_p_stage_cyc') for
this sort key. Use 'local_p_stage_cyc' as default option for "mem" sort
mode.

Also add this to the list of dynamic sort keys and made the
"dynamic_headers" and "arch_specific_sort_keys" as static.

Reported-by: Namhyung Kim 
Signed-off-by: Athira Jajeev 
Tested-by: Nageswara R Sastry 
Cc: Jiri Olsa 
Cc: Kajol Jain 
Cc: Madhavan Srinivasan 
Cc: Michael Ellerman 
Cc: linuxppc-dev@lists.ozlabs.org
Link: https://lore.kernel.org/r/20211203022038.48240-1-atrajeev@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf sort: Fix the 'p_stage_cyc' sort key behavior

2021-11-18T13:08:07+00:00

andle 'p_stage_cyc' (for pipeline stage cycles) sort key with the same
rationale as for the 'weight' and 'local_weight', see the fix in this
series for a full explanation.

Not sure it also needs the local and global variants.

But I couldn't test it actually because I don't have the machine.

Reviewed-by: Athira Jajeev 
Signed-off-by: Namhyung Kim 
Tested-by: Athira Jajeev 
Cc: Andi Kleen 
Cc: Athira Jajeev 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20211105225617.151364-3-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf sort: Fix the 'ins_lat' sort key behavior

2021-11-18T13:08:07+00:00

Handle 'ins_lat' (for instruction latency) and 'local_ins_lat' sort keys
with the same rationale as for the 'weight' and 'local_weight', see the
previous fix in this series for a full explanation.

But I couldn't test it actually, so only build tested.

Reviewed-by: Athira Jajeev 
Signed-off-by: Namhyung Kim 
Tested-by: Athira Jajeev 
Cc: Andi Kleen 
Cc: Athira Jajeev 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20211105225617.151364-2-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf sort: Fix the 'weight' sort key behavior

2021-11-18T13:08:07+00:00

Currently, the 'weight' field in the perf sample has latency information
for some instructions like in memory accesses.  And perf tool has 'weight'
and 'local_weight' sort keys to display the info.

But it's somewhat confusing what it shows exactly.  In my understanding,
'local_weight' shows a weight in a single sample, and (global) 'weight'
shows a sum of the weights in the hist_entry.

For example:

  $ perf mem record -t load dd if=/dev/zero of=/dev/null bs=4k count=1M

  $ perf report --stdio -n -s +local_weight
  ...
  #
  # Overhead  Samples  Command  Shared Object     Symbol                     Local Weight
  # ........  .......  .......  ................  .........................  ............
  #
      21.23%      313  dd       [kernel.vmlinux]  [k] lockref_get_not_zero   32
      12.43%      183  dd       [kernel.vmlinux]  [k] lockref_get_not_zero   35
      11.97%      159  dd       [kernel.vmlinux]  [k] lockref_get_not_zero   36
      10.40%      141  dd       [kernel.vmlinux]  [k] lockref_put_return     32
       7.63%      113  dd       [kernel.vmlinux]  [k] lockref_get_not_zero   33
       6.37%       92  dd       [kernel.vmlinux]  [k] lockref_get_not_zero   34
       6.15%       90  dd       [kernel.vmlinux]  [k] lockref_put_return     33
  ...

So let's look at the 'lockref_get_not_zero' symbols.  The top entry
shows that 313 samples were captured with 'local_weight' 32, so the
total weight should be 313 x 32 = 10016.  But it's not the case:

  $ perf report --stdio -n -s +local_weight,weight -S lockref_get_not_zero
  ...
  #
  # Overhead  Samples  Command  Shared Object     Local Weight  Weight
  # ........  .......  .......  ................  ............  ......
  #
       1.36%        4  dd       [kernel.vmlinux]  36            144
       0.47%        4  dd       [kernel.vmlinux]  37            148
       0.42%        4  dd       [kernel.vmlinux]  32            128
       0.40%        4  dd       [kernel.vmlinux]  34            136
       0.35%        4  dd       [kernel.vmlinux]  36            144
       0.34%        4  dd       [kernel.vmlinux]  35            140
       0.30%        4  dd       [kernel.vmlinux]  36            144
       0.30%        4  dd       [kernel.vmlinux]  34            136
       0.30%        4  dd       [kernel.vmlinux]  32            128
       0.30%        4  dd       [kernel.vmlinux]  32            128
  ...

With the 'weight' sort key, it's divided to 4 samples even with the same
info ('comm', 'dso', 'sym' and 'local_weight').  I don't think this is
what we want.

I found this because of the way it aggregates the 'weight' value.  Since
it's not a period, we should not add them in the he->stat.  Otherwise,
two 32 'weight' entries will create a 64 'weight' entry.

After that, new 32 'weight' samples don't have a matching entry so it'd
create a new entry and make it a 64 'weight' entry again and again.
Later, they will be merged into 128 'weight' entries during the
hists__collapse_resort() with 4 samples, multiple times like above.

Let's keep the weight and display it differently.  For 'local_weight',
it can show the weight as is, and for (global) 'weight' it can display
the number multiplied by the number of samples.

With this change, I can see the expected numbers.

  $ perf report --stdio -n -s +local_weight,weight -S lockref_get_not_zero
  ...
  #
  # Overhead  Samples  Command  Shared Object     Local Weight  Weight
  # ........  .......  .......  ................  ............  .....
  #
      21.23%      313  dd       [kernel.vmlinux]  32            10016
      12.43%      183  dd       [kernel.vmlinux]  35            6405
      11.97%      159  dd       [kernel.vmlinux]  36            5724
       7.63%      113  dd       [kernel.vmlinux]  33            3729
       6.37%       92  dd       [kernel.vmlinux]  34            3128
       4.17%       59  dd       [kernel.vmlinux]  37            2183
       0.08%        1  dd       [kernel.vmlinux]  269           269
       0.08%        1  dd       [kernel.vmlinux]  38            38

Reviewed-by: Athira Jajeev 
Signed-off-by: Namhyung Kim 
Tested-by: Athira Jajeev 
Cc: Andi Kleen 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20211105225617.151364-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo