1. 08 11月, 2011 1 次提交
  2. 15 6月, 2011 1 次提交
    • S
      rcu: Use softirq to address performance regression · 09223371
      Shaohua Li 提交于
      Commit a26ac245(rcu: move TREE_RCU from softirq to kthread)
      introduced performance regression. In an AIM7 test, this commit degraded
      performance by about 40%.
      
      The commit runs rcu callbacks in a kthread instead of softirq. We observed
      high rate of context switch which is caused by this. Out test system has
      64 CPUs and HZ is 1000, so we saw more than 64k context switch per second
      which is caused by RCU's per-CPU kthread.  A trace showed that most of
      the time the RCU per-CPU kthread doesn't actually handle any callbacks,
      but instead just does a very small amount of work handling grace periods.
      This means that RCU's per-CPU kthreads are making the scheduler do quite
      a bit of work in order to allow a very small amount of RCU-related
      processing to be done.
      
      Alex Shi's analysis determined that this slowdown is due to lock
      contention within the scheduler.  Unfortunately, as Peter Zijlstra points
      out, the scheduler's real-time semantics require global action, which
      means that this contention is inherent in real-time scheduling.  (Yes,
      perhaps someone will come up with a workaround -- otherwise, -rt is not
      going to do well on large SMP systems -- but this patch will work around
      this issue in the meantime.  And "the meantime" might well be forever.)
      
      This patch therefore re-introduces softirq processing to RCU, but only
      for core RCU work.  RCU callbacks are still executed in kthread context,
      so that only a small amount of RCU work runs in softirq context in the
      common case.  This should minimize ksoftirqd execution, allowing us to
      skip boosting of ksoftirqd for CONFIG_RCU_BOOST=y kernels.
      Signed-off-by: NShaohua Li <shaohua.li@intel.com>
      Tested-by: N"Alex,Shi" <alex.shi@intel.com>
      Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
      09223371
  3. 06 5月, 2011 1 次提交
  4. 15 3月, 2011 2 次提交
  5. 07 2月, 2011 1 次提交
    • K
      perf tool: Fix gcc 4.6.0 issues · fb7d0b3c
      Kyle McMartin 提交于
      GCC 4.6.0 in Fedora rawhide turned up some compile errors in tools/perf
      due to the -Werror=unused-but-set-variable flag.
      
      I've gone through and annotated some of the assignments that had side
      effects (ie: return value from a function) with the __used annotation,
      and in some cases, just removed unused code.
      
      In a few cases, we were assigning something useful, but not using it in
      later parts of the function.
      
      kyle@dreadnought:~/src% gcc --version
      gcc (GCC) 4.6.0 20110122 (Red Hat 4.6.0-0.3)
      
      Cc: Ingo Molnar <mingo@redhat.com>
      LKML-Reference: <20110124161304.GK27353@bombadil.infradead.org>
      Signed-off-by: NKyle McMartin <kyle@redhat.com>
      [ committer note: Fixed up the annotation fixes, as that code moved recently ]
      Signed-off-by: NArnaldo Carvalho de Melo <acme@redhat.com>
      fb7d0b3c
  6. 11 5月, 2010 1 次提交
  7. 01 5月, 2010 1 次提交
    • F
      perf: Fix warning while reading ring buffer headers · d00a47cc
      Frederic Weisbecker 提交于
      commit e9e94e3b
      "perf trace: Ignore "overwrite" field if present in
      /events/header_page" makes perf trace launching spurious warnings
      about unexpected tokens read:
      
      	Warning: Error: expected type 6 but read 4
      
      This change tries to handle the overcommit field in the header_page
      file whenever this field is present or not.
      
      The problem is that if this field is not present, we try to find it
      and give up in the middle of the line when we realize we are actually
      dealing with another field, which is the "data" one. And this failure
      abandons the file pointer in the middle of the "data" description
      line:
      
      	field: u64 timestamp;	offset:0;	size:8;	signed:0;
      	field: local_t commit;	offset:8;	size:8;	signed:1;
      	field: char data;	offset:16;	size:4080;	signed:1;
                            ^^^
                            Here
      
      What happens next is that we want to read this line to parse the data
      field, but we fail because the pointer is not in the beginning of the
      line.
      
      We could probably fix that by rewinding the pointer. But in fact we
      don't care much about these headers that only concern the ftrace
      ring-buffer. We don't use them from perf.
      
      Just skip this part of perf.data, but don't remove it from recording
      to stay compatible with olders perf.data
      Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
      Cc: Ingo Molnar <mingo@elte.hu>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Stephane Eranian <eranian@google.com>
      Cc: Steven Rostedt <rostedt@goodmis.org>
      d00a47cc
  8. 15 4月, 2010 1 次提交
    • T
      perf: Fix dynamic field detection · a1e2f60e
      Thomas Gleixner 提交于
      Checking if a tracing field is an array with a dynamic length
      requires to check the field type and seek the "__data_loc"
      string that prepends the actual type, as can be found in a trace
      event format file:
      
      	field:__data_loc char[] name;	offset:16;	size:4;	signed:1;
      
      But we actually use strcmp() to check if the field type fully
      matches "__data_loc", which may fail as we trip over the rest of
      the type.
      
      To fix this, use strncmp to only check if it starts with
      "__data_loc".
      Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
      Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Tom Zanussi <tzanussi@gmail.com>
      Cc: Steven Rostedt <rostedt@goodmis.org>
      LKML-Reference: <1271282283-23721-1-git-send-regression-fweisbec@gmail.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      a1e2f60e
  9. 14 4月, 2010 1 次提交
    • I
      perf: Fix endianness argument compatibility with OPT_BOOLEAN() and introduce OPT_INCR() · c0555642
      Ian Munsie 提交于
      Parsing an option from the command line with OPT_BOOLEAN on a
      bool data type would not work on a big-endian machine due to the
      manner in which the boolean was being cast into an int and
      incremented. For example, running 'perf probe --list' on a
      PowerPC machine would fail to properly set the list_events bool
      and would therefore print out the usage information and
      terminate.
      
      This patch makes OPT_BOOLEAN work as expected with a bool
      datatype. For cases where the original OPT_BOOLEAN was
      intentionally being used to increment an int each time it was
      passed in on the command line, this patch introduces OPT_INCR
      with the old behaviour of OPT_BOOLEAN (the verbose variable is
      currently the only such example of this).
      
      I have reviewed every use of OPT_BOOLEAN to verify that a true
      C99 bool was passed. Where integers were used, I verified that
      they were only being used for boolean logic and changed them to
      bools to ensure that they would not be mistakenly used as ints.
      The major exception was the verbose variable which now uses
      OPT_INCR instead of OPT_BOOLEAN.
      Signed-off-by: NIan Munsie <imunsie@au.ibm.com>
      Acked-by: NDavid S. Miller <davem@davemloft.net>
      Cc: <stable@kernel.org> # NOTE: wont apply to .3[34].x cleanly, please backport
      Cc: Git development list <git@vger.kernel.org>
      Cc: Ian Munsie <imunsie@au1.ibm.com>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
      Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
      Cc: Rusty Russell <rusty@rustcorp.com.au>
      Cc: Frederic Weisbecker <fweisbec@gmail.com>
      Cc: Eric B Munson <ebmunson@us.ibm.com>
      Cc: Valdis.Kletnieks@vt.edu
      Cc: WANG Cong <amwang@redhat.com>
      Cc: Thiago Farina <tfransosi@gmail.com>
      Cc: Masami Hiramatsu <mhiramat@redhat.com>
      Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
      Cc: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
      Cc: Arjan van de Ven <arjan@linux.intel.com>
      Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
      Cc: Mike Galbraith <efault@gmx.de>
      Cc: Tom Zanussi <tzanussi@gmail.com>
      Cc: Anton Blanchard <anton@samba.org>
      Cc: John Kacur <jkacur@redhat.com>
      Cc: Li Zefan <lizf@cn.fujitsu.com>
      Cc: Steven Rostedt <rostedt@goodmis.org>
      LKML-Reference: <1271147857-11604-1-git-send-email-imunsie@au.ibm.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      c0555642
  10. 08 4月, 2010 1 次提交
    • A
      perf trace: Ignore "overwrite" field if present in /events/header_page · e9e94e3b
      Arnaldo Carvalho de Melo 提交于
      That is not used in perf where we have the LOST events.
      
      Without this patch we get:
      
      [root@doppio ~]# perf lock report | head -3
        Warning: Error: expected 'data' but read 'overwrite'
      
      So, to make the same perf command work with kernels with and without
      this field, introduce variants for the parsing routines to not warn the
      user in such case.
      
      Discussed-with: Steven Rostedt <rostedt@goodmis.org>
      Cc: Frédéric Weisbecker <fweisbec@gmail.com>
      Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
      Cc: Mike Galbraith <efault@gmx.de>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      LKML-Reference: <new-submission>
      Signed-off-by: NArnaldo Carvalho de Melo <acme@redhat.com>
      e9e94e3b
  11. 24 2月, 2010 1 次提交
  12. 31 1月, 2010 1 次提交
    • H
      perf tools: Add __data_loc support · 86d8d296
      Hitoshi Mitake 提交于
      This patch is required to test the next patch for perf lock.
      
      At 064739bc ,
      support for the modifier "__data_loc" of format is added.
      
      But, when I wanted to parse format of lock_acquired (or some
      event else), raw_field_ptr() did not returned correct pointer.
      
      So I modified raw_field_ptr() like this patch. Then
      raw_field_ptr() works well.
      Signed-off-by: NHitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
      Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Tom Zanussi <tzanussi@gmail.com>
      Cc: Steven Rostedt <srostedt@redhat.com>
      LKML-Reference: <1264851813-8413-2-git-send-email-mitake@dcl.info.waseda.ac.jp>
      [ v3: fixed minor stylistic detail ]
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      86d8d296
  13. 10 12月, 2009 1 次提交
  14. 07 12月, 2009 1 次提交
    • O
      perf tools: Misc small fixes · 7691b1ec
      OGAWA Hirofumi 提交于
      - util/header.c
      	"len" is aligned to 64. So, it tries to write the out of
      	long_name buffer.
      
      	So, this use "zero_buf" to write aligned area.
      
      - util/trace-event-read.c
      	"size" is not including nul byte. So, this allocates it, and set '\0'.
      
      - util/trace-event-parse.c
      	It needs parens to calc correct size.
      Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: Frederic Weisbecker <fweisbec@gmail.com>
      LKML-Reference: <87d42s8iiu.fsf_-_@devron.myhome.or.jp>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      7691b1ec
  15. 28 11月, 2009 3 次提交
    • T
      perf trace: Add interface to access perf data from Perl handlers · d1b93772
      Tom Zanussi 提交于
      The Perl scripting support for perf trace allows most of a trace
      event's data to be accessed directly as handler arguments, but
      not all of it e.g. the less common fields aren't passed in.  To
      give scripts access to the other fields and/or any other data or
      metadata in the main perf executable that might be useful, a way
      to access the C data in perf from Perl is needed; this patch
      uses the Perl XS facility to do it for the common_xxx event
      fields not passed to handler functions.
      
      Context.pm exports three functions to Perl scripts that access
      fields for the current event by calling back into perf:
      common_pc(), common_flags() and common_lock_depth().  Support
      for common_flags() field values was added to Core.pm and a
      script used to sanity check these and other basic scripting
      features, check-perf-trace.pl, was also added.
      Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
      Cc: fweisbec@gmail.com
      Cc: rostedt@goodmis.org
      Cc: anton@samba.org
      Cc: hch@infradead.org
      LKML-Reference: <1259133352-23685-6-git-send-email-tzanussi@gmail.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      d1b93772
    • T
      perf trace: Add Perl scripting support · 16c632de
      Tom Zanussi 提交于
      Implement trace_scripting_ops to make Perl a supported perf
      trace scripting language.
      
      Additionally adds code that allows Perl trace scripts to access
      the 'flag' and 'symbolic' (__print_flags(), __print_symbolic())
      field information parsed from the trace format files.
      
      Also adds the Perl implementation of the generate_script()
      trace_scripting_op, which creates a ready-to-run perf trace Perl
      script based on existing trace data.  Scripts generated by this
      implementation print out all the fields for each event mentioned
      in perf.data (and will detect and generate the proper scripting
      code for 'flag' and 'symbolic' fields), and will additionally
      generate handlers for the special 'trace_unhandled',
      'trace_begin' and 'trace_end' handlers.  Script authors can
      simply remove the printing code to implement their own custom
      event handling.
      Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
      Cc: fweisbec@gmail.com
      Cc: rostedt@goodmis.org
      Cc: anton@samba.org
      Cc: hch@infradead.org
      LKML-Reference: <1259133352-23685-4-git-send-email-tzanussi@gmail.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      16c632de
    • T
      perf trace: Add flag/symbolic format_flags · eb9a42ca
      Tom Zanussi 提交于
      It's useful to know whether a field is a flag or symbolic field
      for e.g. when generating scripts - it allows us to translate
      those fields specially rather than literally as plain numeric
      values.
      Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
      Cc: fweisbec@gmail.com
      Cc: rostedt@goodmis.org
      Cc: anton@samba.org
      Cc: hch@infradead.org
      LKML-Reference: <1259133352-23685-3-git-send-email-tzanussi@gmail.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      eb9a42ca
  16. 25 11月, 2009 1 次提交
    • T
      perf tools: Add 'signed' flag setting back into trace-event-parse.c · 0d0bea5e
      Tom Zanussi 提交于
      Commit 13999e59 (perf tools:
      Handle the case with and without the "signed" trace field)
      removed code to set the FIELD_IS_SIGNED flag that was originally
      added by commit 26a50744
      (tracing/events: Add 'signed' field to format files).
      
      This adds it back.
      Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
      Cc: Steven Rostedt <rostedt@goodmis.org>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Mike Galbraith <efault@gmx.de>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: Frederic Weisbecker <fweisbec@gmail.com>
      LKML-Reference: <1259133299-23594-2-git-send-email-tzanussi@gmail.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      0d0bea5e
  17. 21 10月, 2009 1 次提交
    • S
      perf tools: Use strsep() over strtok_r() for parsing single line · 4e3b799d
      Steven Rostedt 提交于
      The second argument in the strtok_r() function is not to be used
      generically and can have different implementations. Currently
      the function parsing of the perf trace code uses the second
      argument to copy data from. This can crash the tool or just have
      unpredictable results.
      
      The correct solution is to use strsep() which has a defined
      result.
      
      I also added a check to see if the result was correct, and will
      break out of the loop in case it fails to parse as expected.
      Reported-by: NArnaldo Carvalho de Melo <acme@redhat.com>
      Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>
      Cc: Peter Zijlstra <peterz@infradead.org>
      Cc: Frederic Weisbecker <fweisbec@gmail.com>
      LKML-Reference: <20091020232034.237814877@goodmis.org>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      4e3b799d
  18. 15 10月, 2009 13 次提交
  19. 12 10月, 2009 1 次提交
    • R
      perf tools: Fix const char type propagation · cbef79a8
      Randy Dunlap 提交于
      The following perf build warnings/errors in function
      argument types:
      
        builtin-sched.c:1894: warning: passing argument 1 of 'sort_dimension__add' discards qualifiers from pointer target type
        util/trace-event-parse.c:685: warning: passing argument 2 of 'read_expected' discards qualifiers from pointer target type
        util/trace-event-parse.c:741: warning: passing argument 4 of 'test_type_token' discards qualifiers from pointer target type
        util/trace-event-parse.c:706: warning: passing argument 2 of 'read_expected_item' discards qualifiers from pointer target type
      
      ... trigger because older GCC is not able to prove that
      sort_dimension__add() does not change the string.
      
      Some goes for test_type_token().
      
      Fix this by improving type consistency.
      Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
      Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
      Acked-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Paul Mackerras <paulus@samba.org>
      LKML-Reference: <20091005131729.78444bfb.randy.dunlap@oracle.com>
      [ Also remove ugly type cast now unnecessary. ]
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      cbef79a8
  20. 06 10月, 2009 4 次提交
  21. 14 9月, 2009 1 次提交
    • I
      perf tools: Implement counter output multiplexing · ea57c4f5
      Ingo Molnar 提交于
      Finish the -M/--multiplex option implementation:
      
       - separate it out from group_fd
      
       - correctly set it via the ioctl and dont mmap counters that
         are multiplexed
      
       - modify the perf record event loop to deal with buffer-less
         counters.
      
       - remove the -g option from perf sched record
      
       - account for unordered events in perf sched latency
      
       - (add -f to perf sched record to ease measurements)
      
       - skip idle threads (pid==0) in latency output
      
      The result is better latency output by 'perf sched latency':
      
       -----------------------------------------------------------------------------------
        Task              |  Runtime ms | Switches | Average delay ms | Maximum delay ms |
       -----------------------------------------------------------------------------------
        ksoftirqd/8       |    0.071 ms |        2 | avg:    0.458 ms | max:    0.913 ms |
        at-spi-registry   |    0.609 ms |       19 | avg:    0.013 ms | max:    0.023 ms |
        perf              |    3.316 ms |       16 | avg:    0.013 ms | max:    0.054 ms |
        Xorg              |    0.392 ms |       19 | avg:    0.011 ms | max:    0.018 ms |
        sleep             |    0.537 ms |        2 | avg:    0.009 ms | max:    0.009 ms |
       -----------------------------------------------------------------------------------
        TOTAL:            |    4.925 ms |       58 |
       ---------------------------------------------
      
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Mike Galbraith <efault@gmx.de>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Cc: Frederic Weisbecker <fweisbec@gmail.com>
      LKML-Reference: <new-submission>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      ea57c4f5
  22. 13 9月, 2009 1 次提交
    • F
      perf sched: Fix bad event alignment · 46538818
      Frederic Weisbecker 提交于
      perf sched raises the following error when it meets a sched
      switch event:
      
      perf: builtin-sched.c:286: register_pid: Assertion `!(pid >= 65536)' failed.
      Abandon
      
      Currently in x86-64, the sched switch events have a hole in the
      middle of the structure:
      
      	u16 common_type;
      	u8 common_flags;
      	u8 common_preempt_count;
      	u32 common_pid;
      	u32 common_tgid;
      
      	char prev_comm[16];
      	u32 prev_pid;
      	u32 prev_prio;
      			<--- there
      	u64 prev_state;
      	char next_comm[16];
      	u32 next_pid;
      	u32 next_prio;
      
      Gcc inserts a 4 bytes hole there for prev_state to be u64
      aligned. And the events are exported to userspace with this
      hole.
      
      But in userspace, from perf sched, we fetch it using a
      structure that has a new field in the beginning: u32 size. This
      is because our trace is exported with its size as a field. But
      now that we have this new field, the hole in the middle
      disappears because it makes prev_state becoming well aligned.
      
      And since we are using a pointer to the raw trace using this
      struct, instead of reading prev_state, we are reading the hole.
      
      We could fix it by keeping the size seperate from the struct
      but actually there a lot of other potential problems: some
      fields may be saved as long in a 64 bits system and later read
      as long in a 32 bits system. Also this direct cast doesn't care
      about the endianness differences between the host traced
      machine and the machine in which we do the post processing.
      
      So instead of using such dangerous direct casts, fetch the
      values using the trace parsing API that already takes care of
      all these problems.
      Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
      Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
      Cc: Mike Galbraith <efault@gmx.de>
      Cc: Paul Mackerras <paulus@samba.org>
      Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
      Signed-off-by: NIngo Molnar <mingo@elte.hu>
      46538818