Skip to content

L
linux
Switch branch/tag
  1. 21 Jul, 2010 4 commits
    • KOSAKI Motohiro's avatar
      tracing: Shrink max latency ringbuffer if unnecessary · ef710e10
      KOSAKI Motohiro authored 
      Documentation/trace/ftrace.txt says

  buffer_size_kb:

        This sets or displays the number of kilobytes each CPU
        buffer can hold. The tracer buffers are the same size
        for each CPU. The displayed number is the size of the
        CPU buffer and not total size of all buffers. The
        trace buffers are allocated in pages (blocks of memory
        that the kernel uses for allocation, usually 4 KB in size).
        If the last page allocated has room for more bytes
        than requested, the rest of the page will be used,
        making the actual allocation bigger than requested.
        ( Note, the size may not be a multiple of the page size
          due to buffer management overhead. )

        This can only be updated when the current_tracer
        is set to "nop".

But it's incorrect. currently total memory consumption is
'buffer_size_kb x CPUs x 2'.

Why two times difference is there? because ftrace implicitly allocate
the buffer for max latency too.

That makes sad result when admin want to use large buffer. (If admin
want full logging and makes detail analysis). example, If admin
have 24 CPUs machine and write 200MB to buffer_size_kb, the system
consume ~10GB memory (200MB x 24 x 2). umm.. 5GB memory waste is
usually unacceptable.

Fortunatelly, almost all users don't use max latency feature.
The max latency buffer can be disabled easily.

This patch shrink buffer size of the max latency buffer if
unnecessary.
Signed-off-by: default avatarKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
LKML-Reference: <20100701104554.DA2D.A69D9226@jp.fujitsu.com>
Signed-off-by: default avatarSteven Rostedt <rostedt@goodmis.org>
      ef710e10
    • Lai Jiangshan's avatar
      tracing: Reduce latency and remove percpu trace_seq · bc289ae9
      Lai Jiangshan authored 
      __print_flags() and __print_symbolic() use percpu trace_seq:

1) Its memory is allocated at compile time, it wastes memory if we don't use tracing.
2) It is percpu data and it wastes more memory for multi-cpus system.
3) It disables preemption when it executes its core routine
   "trace_seq_printf(s, "%s: ", #call);" and introduces latency.

So we move this trace_seq to struct trace_iterator.
Signed-off-by: default avatarLai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <4C078350.7090106@cn.fujitsu.com>
Signed-off-by: default avatarSteven Rostedt <rostedt@goodmis.org>
      bc289ae9
    • Richard Kennedy's avatar
      trace: Reorder struct ring_buffer_per_cpu to remove padding on 64bit · 985023de
      Richard Kennedy authored 
      Reorder structure to remove 8 bytes of padding on 64 bit builds.
This shrinks the size to 128 bytes so allowing allocation from a smaller
slab & needed one fewer cache lines.
Signed-off-by: default avatarRichard Kennedy <richard@rsk.demon.co.uk>
LKML-Reference: <1269516456.2054.8.camel@localhost>
Signed-off-by: default avatarSteven Rostedt <rostedt@goodmis.org>
      985023de
    • Li Zefan's avatar
      tracing: Allow to disable cmdline recording · e870e9a1
      Li Zefan authored 
      We found that even enabling a single trace event that will rarely be
triggered can add big overhead to context switch.

(lmbench context switch test)
 -------------------------------------------------
 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
 ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw   ctxsw
------ ------ ------ ------ ------ ------- -------
  2.19   2.3   2.21   2.56   2.13     2.54    2.07
  2.39   2.51  2.35   2.75   2.27     2.81    2.24

The overhead is 6% ~ 11%.

It's because when a trace event is enabled 3 tracepoints (sched_switch,
sched_wakeup, sched_wakeup_new) will be activated to map pid to cmdname.

We'd like to avoid this overhead, so add a trace option '(no)record-cmd'
to allow to disable cmdline recording.
Signed-off-by: default avatarLi Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4C2D57F4.2050204@cn.fujitsu.com>
Signed-off-by: default avatarSteven Rostedt <rostedt@goodmis.org>
      e870e9a1
  3. 17 Jul, 2010 4 commits
  5. 16 Jul, 2010 3 commits
  7. 15 Jul, 2010 3 commits
  9. 12 Jul, 2010 1 commit
  11. 06 Jul, 2010 1 commit
  13. 05 Jul, 2010 14 commits
  15. 04 Jul, 2010 1 commit
    • Will Deacon's avatar
      ARM: 6205/1: perf: ensure counter delta is treated as unsigned · 446a5a8b
      Will Deacon authored 
      Hardware performance counters on ARM are 32-bits wide but atomic64_t
variables are used to represent counter data in the hw_perf_event structure.

The armpmu_event_update function right-shifts a signed 64-bit delta variable
and adds the result to the event count. This can lead to shifting in sign-bits
if the MSB of the 32-bit counter value is set. This results in perf output
such as:

 Performance counter stats for 'sleep 20':

 18446744073460670464  cycles             <-- 0xFFFFFFFFF12A6000
        7783773  instructions             #      0.000 IPC
            465  context-switches
            161  page-faults
        1172393  branches

   20.154242147  seconds time elapsed

This patch ensures that the delta value is treated as unsigned so that the
right shift sets the upper bits to zero.

Cc: <stable@kernel.org>
Acked-by: default avatarJamie Iles <jamie.iles@picochip.com>
Signed-off-by: default avatarWill Deacon <will.deacon@arm.com>
Signed-off-by: default avatarRussell King <rmk+kernel@arm.linux.org.uk>
      446a5a8b
  17. 03 Jul, 2010 1 commit
  19. 02 Jul, 2010 8 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7