]> pilppa.com Git - linux-2.6-omap-h63xx.git/log
linux-2.6-omap-h63xx.git
15 years agoftrace: Add check of sched_stopped for probe_sched_wakeup
Zhaolei [Tue, 31 Mar 2009 07:24:51 +0000 (15:24 +0800)]
ftrace: Add check of sched_stopped for probe_sched_wakeup

The wakeup tracing in sched_switch does not stop when a user
disables tracing. This is because the probe_sched_wakeup() is missing
the check to prevent the wakeup from being traced.

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
LKML-Reference: <49D1C543.3010307@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agofunction-graph: add proper initialization for init task
Steven Rostedt [Thu, 26 Mar 2009 00:55:00 +0000 (20:55 -0400)]
function-graph: add proper initialization for init task

Impact: fix to crash going to kexec

The init task did not properly initialize the function graph pointers.
Altough these pointers are NULL, they can not be assumed to be NULL
for the init task, and must still be properly initialize.

This usually is not an issue since a problem only arises when a task
exits, and the init tasks do not usually exit. But when doing tests
with kexec, the init tasks do exit, and the bug appears.

This patch properly initializes the init tasks function graph data
structures.

Reported-and-Tested-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <alpine.DEB.2.00.0903252053080.5675@gandalf.stny.rr.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/ftrace: fix missing include string.h
Frederic Weisbecker [Fri, 27 Mar 2009 13:22:10 +0000 (14:22 +0100)]
tracing/ftrace: fix missing include string.h

Building a kernel with tracing can raise the following warning on
tip/master:

kernel/trace/trace.c:1249: error: implicit declaration of function 'vbin_printf'

We are missing an include to string.h

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1238160130-7437-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: fix incorrect return type of ns2usecs()
Lai Jiangshan [Mon, 30 Mar 2009 05:48:00 +0000 (13:48 +0800)]
tracing: fix incorrect return type of ns2usecs()

Impact: fix time output bug in 32bits system

ns2usecs() returns 'long', it's incorrect.

(In i386)
...
          <idle>-0     [000]   521.442100: _spin_lock <-tick_do_update_jiffies64
          <idle>-0     [000]   521.442101: do_timer <-tick_do_update_jiffies64
          <idle>-0     [000]   521.442102: update_wall_time <-do_timer
          <idle>-0     [000]   521.442102: update_xtime_cache <-update_wall_time
....
(It always print the time less than 2200 seconds besides ...)
Because 'long' is 32bits in i386. ( (1<<31) useconds is about 2200 seconds)

...
          <idle>-0     [001] 4154502640.134759: rcu_bh_qsctr_inc <-__do_softirq
          <idle>-0     [001] 4154502640.134760: _local_bh_enable <-__do_softirq
          <idle>-0     [001] 4154502640.134761: idle_cpu <-irq_exit
...
(very large value)
Because 'long' is a signed type and it is 32bits in i386.

Changes in v2:
return 'unsigned long long' instead of 'cycle_t'

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <49D05D10.4030009@cn.fujitsu.com>
Reported-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: remove CALLER_ADDR2 from wakeup tracer
Steven Rostedt [Fri, 3 Apr 2009 15:12:23 +0000 (11:12 -0400)]
tracing: remove CALLER_ADDR2 from wakeup tracer

Maneesh Soni was getting a crash when running the wakeup tracer.
We debugged it down to the recording of the function with the
CALLER_ADDR2 macro.  This is used to get the location of the caller
to schedule.

But the problem comes when schedule is called by assmebly. In the case
that Maneesh had, retint_careful would call schedule. But retint_careful
does not set up a proper frame pointer. CALLER_ADDR2 is defined as
__builtin_return_address(2). This produces the following assembly in
the wakeup tracer code.

   mov    0x0(%rbp),%rcx  <--- get the frame pointer of the caller
   mov    %r14d,%r8d
   mov    0xf2de8e(%rip),%rdi

   mov    0x8(%rcx),%rsi  <-- this is __builtin_return_address(1)
   mov    0x28(%rdi,%rax,8),%rbx

   mov    (%rcx),%rax  <-- get the frame pointer of the caller's caller
   mov    %r12,%rcx
   mov    0x8(%rax),%rdx <-- this is __builtin_return_address(2)

At the reading of 0x8(%rax) Maneesh's machine would take a fault.
The reason is that retint_careful did not set up the return address
and the content of %rax here was zero.

To verify this, I sent Maneesh a patch to create a frame pointer
in retint_careful. He ran the test again but this time he would take
the same type of fault from sysret_careful. The retint_careful was no
longer an issue, but there are other callers that still have issues.

Instead of adding frame pointers for all callers to schedule (in possibly
all archs), it is much safer to simply not use CALLER_ADDR2. This
loses out on knowing what called schedule, but the function tracer
will help there if needed.

Reported-by: Maneesh Soni <maneesh@in.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge branch 'tracing/blktrace-fixes' into tracing/urgent
Ingo Molnar [Tue, 7 Apr 2009 11:40:49 +0000 (13:40 +0200)]
Merge branch 'tracing/blktrace-fixes' into tracing/urgent

Merge reason: this used to be a tracing/blktrace-v2 devel topic still
              cooking during the merge window - has propagated to fixes

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix pdu_len when tracing packet command requests
Li Zefan [Thu, 2 Apr 2009 05:43:26 +0000 (13:43 +0800)]
blktrace: fix pdu_len when tracing packet command requests

Impact: output all of packet commands - not just the first 4 / 8 bytes

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add
large command support"), struct request->cmd has been changed from
unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

v1 -> v2: by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>

- make sure rq->cmd_len is always intialized, and then we can use
  rq->cmd_len instead of BLK_MAX_CDB.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
LKML-Reference: <49D4507E.2060602@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: small cleanup in blk_msg_write()
Li Zefan [Fri, 3 Apr 2009 07:31:34 +0000 (15:31 +0800)]
blktrace: small cleanup in blk_msg_write()

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "Alan D. Brunelle" <alan.brunelle@hp.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
LKML-Reference: <49D5BB56.7000807@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: NUL-terminate user space messages
Carl Henrik Lunde [Fri, 3 Apr 2009 12:27:15 +0000 (14:27 +0200)]
blktrace: NUL-terminate user space messages

Impact: fix corrupted blkparse output

Make sure messages from user space are NUL-terminated strings,
otherwise we could dump random memory to the block trace file.

Additionally, I've limited the message to BLK_TN_MAX_MSG-1
characters, because the last character would be stripped by
vscnprintf anyway.

Signed-off-by: Carl Henrik Lunde <chlunde@ping.uio.no>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "Alan D. Brunelle" <alan.brunelle@hp.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20090403122714.GT5178@kernel.dk>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: move scripts/trace/power.pl to scripts/tracing/power.pl
Lai Jiangshan [Mon, 30 Mar 2009 06:50:04 +0000 (14:50 +0800)]
tracing: move scripts/trace/power.pl to scripts/tracing/power.pl

Impact: Cleanup

We use scripts/tracing/ to contain tracing scripts.
Use one directory only instead of two.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Steven Rostedt <srostedt@redhat.com>
LKML-Reference: <49D06B9C.3070209@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge branches 'tracing/docs', 'tracing/filters', 'tracing/ftrace', 'tracing/kprobes...
Ingo Molnar [Tue, 31 Mar 2009 15:46:40 +0000 (17:46 +0200)]
Merge branches 'tracing/docs', 'tracing/filters', 'tracing/ftrace', 'tracing/kprobes', 'tracing/blktrace-v2' and 'tracing/textedit' into tracing/core-v2

15 years agotrace: make argument 'mem' of trace_seq_putmem() const
Li Zefan [Fri, 27 Mar 2009 02:21:00 +0000 (10:21 +0800)]
trace: make argument 'mem' of trace_seq_putmem() const

Impact: fix build warning

I passed a const value to trace_seq_putmem(), and I got compile warning.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add missing 'extern' keywords to trace_output.h
Eduard - Gabriel Munteanu [Mon, 23 Mar 2009 13:12:23 +0000 (15:12 +0200)]
tracing: add missing 'extern' keywords to trace_output.h

Impact: cleanup

Many declarations within trace_output.h are missing the 'extern' keyword
in an inconsistent manner. This adds 'extern' where it should be.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: provide trace_seq_reserve()
Eduard - Gabriel Munteanu [Mon, 23 Mar 2009 13:12:22 +0000 (15:12 +0200)]
tracing: provide trace_seq_reserve()

trace_seq_reserve() allows a caller to reserve space in a trace_seq and
write directly into it. This makes it easier to export binary data to
userspace via the tracing interface, by simply filling in a struct.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: print out BLK_TN_MESSAGE properly
Li Zefan [Fri, 27 Mar 2009 02:21:54 +0000 (10:21 +0800)]
blktrace: print out BLK_TN_MESSAGE properly

Impact: improve ftrace plugin output

Before this patch:

 # cat trace
         make-5383  [001]   741.240059:   8,7    P   N [make]
 __trace_note_message: cfq1074

 # echo 1 > options/blk_classic
 # cat trace
   8,7    1     0.692221252     0  C   W 130411392 + 1024 [0]
 Bad pc action 6361
 Bad pc action 283d

 # echo 0 > options/blk_classic
 # echo bin > trace_options
 # cat trace_pipe | blkparse -i -
 (can't parse messages generated by blk_add_trace_msg())

After this patch:
 # cat trace
      <idle>-0     [001]   187.600933:   8,7    C   W 145220224 + 8 [0]
      <idle>-0     [001]   187.600946:   8,7    m   N cfq1076 complete

 # echo 1 > options/blk_classic
 # cat trace
   8,7    1     0.256378996   238  I   W 113190728 + 8 [pdflush]
   8,7    1     0.256378998   238  m   N cfq1076 insert_request

 # echo 0 > options/blk_classic
 # echo bin > trace_options
 # cat trace_pipe | blkparse -i -
  8,7    1        0    22.973250293     0  C   W 102770576 + 8 [0]
  8,7    1        0    22.973259213     0  m   N cfq1076 complete

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: extract duplidate code
Li Zefan [Fri, 27 Mar 2009 02:21:23 +0000 (10:21 +0800)]
blktrace: extract duplidate code

Impact: cleanup

blk_trace_event_print() and blk_tracer_print_line() share most of the code.

   text    data     bss     dec     hex filename
   8605     393      12    9010    2332 kernel/trace/blktrace.o.orig
   text    data     bss     dec     hex filename
   8555     393      12    8960    2300 kernel/trace/blktrace.o

This patch also prepares for the next patch, that prints out BLK_TN_MESSAGE.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix memory leak when freeing struct blk_io_trace
Li Zefan [Fri, 27 Mar 2009 02:20:24 +0000 (10:20 +0800)]
blktrace: fix memory leak when freeing struct blk_io_trace

Impact: fix mixed ioctl and ftrace-plugin blktrace use memory leak

When mixing the use of ioctl-based blktrace and ftrace-based blktrace,
we can leak memory in this way:

  # btrace /dev/sda > /dev/null &
  # echo 0 > /sys/block/sda/sda1/trace/enable

now we leak bt->dropped_file, bt->msg_file, bt->rchan...

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix blk_probes_ref chaos
Li Zefan [Fri, 27 Mar 2009 02:20:09 +0000 (10:20 +0800)]
blktrace: fix blk_probes_ref chaos

Impact: fix mixed ioctl and ftrace-plugin blktrace use refcount bugs

ioctl-based blktrace allocates bt and registers tracepoints when
ioctl(BLKTRACESETUP), and do all cleanups when ioctl(BLKTRACETEARDOWN).

while ftrace-based blktrace allocates/frees bt when:
  # echo 1/0 > /sys/block/sda/sda1/trace/enable

and registers/unregisters tracepoints when:
  # echo blk/nop > /debugfs/tracing/current_tracer
or
  # echo 1/0 > /debugfs/tracing/tracing_enable

The separatation of allocation and registeration causes 2 problems:

  1. current user-space blktrace still calls ioctl(TEARDOWN) when
     ioctl(SETUP) failed:
       # echo 1 > /sys/block/sda/sda1/trace/enable
       # blktrace /dev/sda
         BLKTRACESETUP: Device or resource busy
         ^C
     and now blk_probes_ref == -1

  2. Another way to make blk_probes_ref == -1:
     # plugin sdb && mount sdb1
     # echo 1 > /sys/block/sdb/sdb1/trace/enable
     # remove sdb

This patch does the allocation and registeration when writing
sdaX/trace/enable.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: make classic output more classic
Li Zefan [Fri, 27 Mar 2009 02:19:46 +0000 (10:19 +0800)]
blktrace: make classic output more classic

Impact: fix ftrace plugin timestamp output

In the classic user-space blktrace, the output timestamp is sec.nsec
not sec.usec.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix off-by-one bug
Li Zefan [Tue, 24 Mar 2009 08:05:27 +0000 (16:05 +0800)]
blktrace: fix off-by-one bug

'what' is used as the index of array what2act, so it can't >= the array size.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix the original blktrace
Li Zefan [Wed, 25 Mar 2009 09:21:26 +0000 (17:21 +0800)]
blktrace: fix the original blktrace

Currently the original blktrace, which is using relay and is used via
ioctl, is broken. You can use ftrace to see the output of blktrace,
but user-space blktrace is unusable.

It's broken by "blktrace: add ftrace plugin"
(c71a896154119f4ca9e89d6078f5f63ad60ef199)

 - if (unlikely(bt->trace_state != Blktrace_running))
 + if (unlikely(bt->trace_state != Blktrace_running || !blk_tracer_enabled))
return;

With this patch, both ioctl and ftrace can be used, but of course you
can't use both of them at the same time.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix a race when creating blk_tree_root in debugfs
Li Zefan [Wed, 25 Mar 2009 09:19:33 +0000 (17:19 +0800)]
blktrace: fix a race when creating blk_tree_root in debugfs

t1                                t2
------                            ------
do_blk_trace_setup()              do_blk_trace_setup()
  if (!blk_tree_root) {
                                    if (!blk_tree_root)
    blk_tree_root = create_dir()
                                      blk_tree_root = create_dir();
                                      (now blk_tree_root == NULL)
  ...
  dir = create_dir(name, blk_tree_root);

Due to this race, t1 will create 'dir' in /debugfs but not /debugfs/block.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix timestamp in binary output
Li Zefan [Wed, 25 Mar 2009 09:18:56 +0000 (17:18 +0800)]
blktrace: fix timestamp in binary output

I found the timestamp is wrong:

 # echo bin > trace_option
 # echo blk > current_tracer
 # cat trace_pipe | blkparse -i -
 8,0    0        0     0.000000000   504  A   W ...
 ...
 8,7    1        0     0.008534097     0  C   R ...
            (should be 8.534097xxx)

user-space blkparse expects the timestamp to be nanosecond.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing, Text Edit Lock: cleanup
Ingo Molnar [Mon, 30 Mar 2009 22:25:23 +0000 (00:25 +0200)]
tracing, Text Edit Lock: cleanup

Remove incorrectly introduced headers.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: filter fix for TRACE_EVENT_FORMAT events
Tom Zanussi [Thu, 26 Mar 2009 06:24:34 +0000 (01:24 -0500)]
tracing: filter fix for TRACE_EVENT_FORMAT events

Impact: fix crash (hang) when using TRACE_EVENT_FORMAT filter files

filters are only hooked up to the tracepoint events defined using
TRACE_EVENT but not the tracers that use TRACE_EVENT_FORMAT, such
as ftrace.

Do not display the filter files at all for TRACE_EVENT_FORMAT events
for the time being.

Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878882.8339.61.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoftrace: Using FTRACE_WARN_ON() to check "freed record" in ftrace_release()
Zhaolei [Wed, 25 Mar 2009 04:06:05 +0000 (12:06 +0800)]
ftrace: Using FTRACE_WARN_ON() to check "freed record" in ftrace_release()

 "Because when we call ftrace_free_rec we change the rec->ip to point to the
  next record in the chain. Something is very wrong if rec->ip >= s &&
  rec->ip < e and the record is already free."

 "Note, use FTRACE_WARN_ON() macro. This way it shuts down ftrace if it is
  hit and helps to avoid further damage later."
                   -- Steven Rostedt <rostedt@goodmis.org>

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agox86: kretprobe-booster interrupt emulation code fix
Masami Hiramatsu [Mon, 23 Mar 2009 14:14:52 +0000 (10:14 -0400)]
x86: kretprobe-booster interrupt emulation code fix

Fix interrupt emulation code in kretprobe-booster according to
pt_regs update (es/ds change and gs adding).

This issue has been reported on systemtap-bugzilla:

  http://sources.redhat.com/bugzilla/show_bug.cgi?id=9965

  | On a -tip kernel on x86_32, kretprobe_example (from samples) triggers the
  | following backtrace when its retprobing a class of functions that cause a
  | copy_from/to_user().
  |
  | BUG: sleeping function called from invalid context at mm/memory.c:3196
  | in_atomic(): 0, irqs_disabled(): 1, pid: 2286, name: cat

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Tested-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Cc: systemtap-ml <systemtap@sources.redhat.com>
LKML-Reference: <49C7995C.2010601@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoinit,cpuset: fix initialize order
Lai Jiangshan [Wed, 25 Mar 2009 09:06:30 +0000 (17:06 +0800)]
init,cpuset: fix initialize order

Impact: cpuset_wq should be initialized after init_workqueues()

When I read /debugfs/tracing/trace_stat/workqueues,
I got this:

 # CPU  INSERTED  EXECUTED   NAME
 # |      |         |          |

   0      0          0       cpuset
   0    285        285       events/0
   0      2          2       work_on_cpu/0
   0   1115       1115       khelper
   0    325        325       kblockd/0
   0      0          0       kacpid
   0      0          0       kacpi_notify
   0      0          0       ata/0
   0      0          0       ata_aux
   0      0          0       ksuspend_usbd
   0      0          0       aio/0
   0      0          0       nfsiod
   0      0          0       kpsmoused
   0      0          0       kstriped
   0      0          0       kondemand/0
   0      1          1       hid_compat
   0      0          0       rpciod/0

   1     64         64       events/1
   1      2          2       work_on_cpu/1
   1      5          5       kblockd/1
   1      0          0       ata/1
   1      0          0       aio/1
   1      0          0       kondemand/1
   1      0          0       rpciod/1

I found "cpuset" is at the earliest.

I found a create_singlethread_workqueue() is earlier than
init_workqueues():

kernel_init()
->cpuset_init_smp()
  ->create_singlethread_workqueue()
->do_basic_setup()
  ->init_workqueues()

I think it's better that create_singlethread_workqueue() is called
after workqueue subsystem has been initialized.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Menage <menage@google.com>
Cc: miaoxie <miaox@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <49C9F416.1050707@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotrace_workqueues: fix empty line's output
Lai Jiangshan [Wed, 25 Mar 2009 08:59:18 +0000 (16:59 +0800)]
trace_workqueues: fix empty line's output

Empty lines separate cpus stat. After previous
fix(trace_stat: keep original order) applied, the empty lines
are displayed at incorrect position.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <49C9F266.2060706@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotrace_stat: keep original order
Lai Jiangshan [Wed, 25 Mar 2009 08:58:39 +0000 (16:58 +0800)]
trace_stat: keep original order

Impact: make trace_stat files show items with the original order

trace_stat tracer reverse the items, it makes the output
looks a little ugly.

Example, when we read trace_stat/workqueues, we get cpu#7's stat.
at first, and then cpu#6... cpu#0.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <49C9F23F.5040307@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotrace_stat: don't call seq_printf() in seq_operation->start()
Lai Jiangshan [Wed, 25 Mar 2009 08:27:17 +0000 (16:27 +0800)]
trace_stat: don't call seq_printf() in seq_operation->start()

Impact: Fix incorrect way using seq_file's API

Use SEQ_START_TOKEN instead of calling ->stat_headers()
int seq_operation->start().

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
LKML-Reference: <49C9EAE5.5070202@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: Documentation / sample code fixes for tracepoints
Jody McIntyre [Tue, 24 Mar 2009 20:00:28 +0000 (16:00 -0400)]
tracing: Documentation / sample code fixes for tracepoints

Fix the tracepoint documentation to refer to "tracepoint-sample"
instead of "tracepoint-example" to match what actually exists;
fix the directory, and clarify how to compile.

Change every instance of "example" in the sample tracepoint code
to "sample" for consistency.

Signed-off-by: Jody McIntyre <scjody@sun.com>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: torvalds@linux-foundation.org
LKML-Reference: <20090324200027.GH8294@clouds>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: use union for multi-usages field
Lai Jiangshan [Tue, 24 Mar 2009 05:38:06 +0000 (13:38 +0800)]
tracing: use union for multi-usages field

Impact: cleanup

struct dyn_ftrace::ip has different usages in his lifecycle,
we use union for it. And also for struct dyn_ftrace::flags.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <srostedt@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <49C871BE.3080405@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoftrace: show virtual PID
Lai Jiangshan [Tue, 24 Mar 2009 03:03:01 +0000 (11:03 +0800)]
ftrace: show virtual PID

Impact: fix PID output under namespaces

When current namespace is not the global namespace,
pid read from set_ftrace_pid is no correct.

 # ~/newpid_namespace_run bash
 # echo $$
 1
 # echo 1 > set_ftrace_pid
 # cat set_ftrace_pid
 3756

Since we write virtual PID to set_ftrace_pid, we need get
virtual PID when we read it.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <srostedt@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <49C84D65.9050606@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agofunction-graph: add option for include sleep times
Steven Rostedt [Tue, 24 Mar 2009 15:06:24 +0000 (11:06 -0400)]
function-graph: add option for include sleep times

Impact: give user a choice to show times spent while sleeping

The user may want to see the time a function spent sleeping.
This patch adds the trace option "sleep-time" to allow that.
The "sleep-time" option is default on.

 echo sleep-time > /debug/tracing/trace_options

produces:

 ------------------------------------------
 2)  avahi-d-3428  =>    <idle>-0
 ------------------------------------------

 2)               |      finish_task_switch() {
 2)   0.621 us    |        _spin_unlock_irq();
 2)   2.202 us    |      }
 2) ! 1002.197 us |    }
 2) ! 1003.521 us |  }

where as,

 echo nosleep-time > /debug/tracing/trace_options

produces:

 0)    <idle>-0    =>  yum-upd-3416
 ------------------------------------------

 0)               |              finish_task_switch() {
 0)   0.643 us    |                _spin_unlock_irq();
 0)   2.342 us    |              }
 0) + 41.302 us   |            }
 0) + 42.453 us   |          }

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agofunction-graph: ignore times across schedule
Steven Rostedt [Tue, 24 Mar 2009 05:10:15 +0000 (01:10 -0400)]
function-graph: ignore times across schedule

Impact: more accurate timings

The current method of function graph tracing does not take into
account the time spent when a task is not running. This shows functions
that call schedule have increased costs:

 3) + 18.664 us   |      }
 ------------------------------------------
 3)    <idle>-0    =>  kblockd-123
 ------------------------------------------

 3)               |      finish_task_switch() {
 3)   1.441 us    |        _spin_unlock_irq();
 3)   3.966 us    |      }
 3) ! 2959.433 us |    }
 3) ! 2961.465 us |  }

This patch uses the tracepoint in the scheduling context switch to
account for time that has elapsed while a task is scheduled out.
Now we see:

 ------------------------------------------
 3)    <idle>-0    =>  edac-po-1067
 ------------------------------------------

 3)               |      finish_task_switch() {
 3)   0.685 us    |        _spin_unlock_irq();
 3)   2.331 us    |      }
 3) + 41.439 us   |    }
 3) + 42.663 us   |  }

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agofunction-graph: prevent more than one tracer registering
Steven Rostedt [Tue, 24 Mar 2009 04:18:31 +0000 (00:18 -0400)]
function-graph: prevent more than one tracer registering

Impact: prevent crash due to multiple function graph tracers

The function graph tracer can currently only handle a single tracer
being registered. If another tracer registers with the function
graph tracer it can crash the system.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agofunction-graph: moved the timestamp from arch to generic code
Steven Rostedt [Tue, 24 Mar 2009 03:38:49 +0000 (23:38 -0400)]
function-graph: moved the timestamp from arch to generic code

This patch move the timestamp from happening in the arch specific
code into the general code. This allows for better control by the tracer
to time manipulation.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agotracing: fix memory leak in trace_stat
Steven Rostedt [Sat, 21 Mar 2009 06:44:50 +0000 (02:44 -0400)]
tracing: fix memory leak in trace_stat

If the function profiler does not have any items recorded and one were
to cat the function stat file, the kernel would take a BUG with a NULL
pointer dereference.

Looking further into this, I found that returning NULL from stat_start
did not stop the stat logic, and would later call stat_next. This breaks
from the way seq_file works, so I looked into fixing the stat code.

This is where I noticed that the last next_entry is never freed.
It is allocated, and if the stat_next returns NULL, the code breaks out
of the loop, unlocks the mutex and exits. We never link the next_entry
nor do we free it. Thus it is a real memory leak.

This patch rearranges the code a bit to not only fix the memory leak,
but also to act more like seq_file where nothing is printed if there
is nothing to print. That is, stat_start returns NULL.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agoblktrace: print human-readable act_mask
Li Zefan [Tue, 24 Mar 2009 09:43:30 +0000 (17:43 +0800)]
blktrace: print human-readable act_mask

Impact: new feature, allow symbolic values in /debug/tracing/act_mask

Print stringified act_mask instead of hex value:

 # cat act_mask
 read,write,barrier,sync,queue,requeue,issue,complete,fs,pc,ahead,meta,
 discard,drv_data
 # echo "meta,write" > act_mask
 # cat act_mask
 write,meta

Also:
 - make act_mask accept "ahead", "meta", "discard" and "drv_data"
 - use strsep() instead of strchr() to parse user input
 - return -EINVAL if a token is not found in the mask map
 - fix a bug that 'value' is unsigned, so it can < 0
 - propagate error value of blk_trace_mask2str() to userspace, but not
   always return -ENXIO.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <49C8AB42.1000802@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix t_error()
Li Zefan [Tue, 24 Mar 2009 08:05:51 +0000 (16:05 +0800)]
blktrace: fix t_error()

Impact: fix error flag output

t_error() should return t->error but not t->sector.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <49C8945F.5020802@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix wrong calculation of RWBS
Li Zefan [Tue, 24 Mar 2009 08:05:06 +0000 (16:05 +0800)]
blktrace: fix wrong calculation of RWBS

Impact: fix the output of IO type category characters

Trace categories are the upper 16 bits, not the lower 16 bits.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <49C89432.8010805@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: mark ddir_act[] const
Li Zefan [Tue, 24 Mar 2009 08:04:37 +0000 (16:04 +0800)]
blktrace: mark ddir_act[] const

Impact: cleanup

ddir_act and what2act always stay immutable.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <49C89415.5080503@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: disallow integer values for string filters and vice versa
Tom Zanussi [Tue, 24 Mar 2009 07:14:42 +0000 (02:14 -0500)]
tracing/filters: disallow integer values for string filters and vice versa

Impact: fix filter use boundary condition / crash

Make sure filters for string fields don't use integer values and vice
versa.  Getting it wrong can crash the system or produce bogus
results.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878882.8339.61.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: use trace_seq_printf() to print filters
Tom Zanussi [Tue, 24 Mar 2009 07:14:31 +0000 (02:14 -0500)]
tracing/filters: use trace_seq_printf() to print filters

Impact: cleanup

Instead of just using the trace_seq buffer to print the filters, use
trace_seq_printf() as it was intended to be used.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878871.8339.59.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: free pred when clearing filters
Tom Zanussi [Tue, 24 Mar 2009 07:14:11 +0000 (02:14 -0500)]
tracing/filters: free pred when clearing filters

Impact: fix (small) per trace filter modification memory leak

Free the current pred when clearing the filters via the filter files.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878851.8339.58.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: use list_for_each_entry
Tom Zanussi [Tue, 24 Mar 2009 07:14:01 +0000 (02:14 -0500)]
tracing/filters: use list_for_each_entry

Impact: cleanup

No need to use the safe version here, so use list_for_each_entry instead
of list_for_each_entry_safe in find_event_field().

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878841.8339.57.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/function-graph-tracer: fix functions call traces imbalance
Frederic Weisbecker [Mon, 23 Mar 2009 21:17:01 +0000 (22:17 +0100)]
tracing/function-graph-tracer: fix functions call traces imbalance

Impact: fix traces output

Sometimes one can observe an imbalance in the traces between function
calls and function return traces:

func1() {
    }
}

The curly brace inside func1() is the return of another function nested
inside func1. The return trace have been inserted in the buffer but not
the entry.
We are storing a return address on the function traces stack while we
haven't inserted its entry on the buffer, hence the imbalance on the
traces.

This is because the tracers doesn't check all failures that can happen
on buffer insertion.

This patch reports the tracing recursion failures and the ring buffer
failures. In such cases, we now restore the original return address for
the function, giving up its return trace.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237843021-11695-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: Fix TRACING_SUPPORT dependency for PPC32
Anton Vorontsov [Mon, 23 Mar 2009 22:07:24 +0000 (01:07 +0300)]
tracing: Fix TRACING_SUPPORT dependency for PPC32

commit 40ada30f9621fbd831ac2437b9a2a399aa ("tracing: clean up menu"),
despite the "clean up" in its purpose, introduced a behavioural
change for Kconfig symbols: we no longer able to select tracing
support on PPC32 (because IRQFLAGS_SUPPORT isn't yet implemented).

The IRQFLAGS_SUPPORT is not mandatory for most tracers, tracing core
has a special case for platforms w/o irqflags (which, by the way, has
become useless as of the commit above).

Though according to Ingo Molnar, there was periodic build failures on
weird, unmaintained architectures that had no irqflags-tracing support
and hence didn't know the raw_irqs_save/restore primitives. Thus we'd
better not enable irqflags-less tracing for all architectures.

This patch restores the old behaviour for PPC32, and thus brings the
tracing back. Other architectures can either add themselves to the
exception list or (better) implement TRACE_IRQFLAGS_SUPPORT.

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-b: Steven Rostedt <rostedt@goodmis.org>
Cc: linuxppc-dev@ozlabs.org
LKML-Reference: <20090323220724.GA9851@oksana.dev.rtsoft.ru>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/ftrace: check if debugfs is registered before creating files
Frederic Weisbecker [Sun, 22 Mar 2009 22:10:45 +0000 (23:10 +0100)]
tracing/ftrace: check if debugfs is registered before creating files

Impact: fix a crash with ftrace={nop,boot} parameter

If the nop or initcall tracers are launched as boot tracers,
they will attempt to create their option directory and files.
But these tracers are registered very early and then assigned
as "boot tracers" very early if asked to.

Since they do this before debugfs has been registered (core initcall),
a crash is triggered.

Another early tracers could also come later. So we fix it by
checking if debugfs is initialized before creating the root
tracing directory.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237759847-21025-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agodebugfs: function to know if debugfs is initialized
Frederic Weisbecker [Sun, 22 Mar 2009 22:10:44 +0000 (23:10 +0100)]
debugfs: function to know if debugfs is initialized

Impact: add new debugfs API

With ftrace, some tracers are registered in early initcalls
and attempt to create files on the debugfs filesystem.
Depending on when they are activated, they can try to create their
file at any time. Some checks can be done on the tracing area
but providing a helper to know if debugfs is registered make it
really more easy.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237759847-21025-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: clean up filter_add_subsystem_pred()
Tom Zanussi [Mon, 23 Mar 2009 08:26:48 +0000 (03:26 -0500)]
tracing/filters: clean up filter_add_subsystem_pred()

Impact: cleanup, memory leak fix

This patch cleans up filter_add_subsystem_pred():

- searches for the field before creating a copy of the pred

- fixes memory leak in the case a predicate isn't applied

- if -ENOMEM, makes sure there's no longer a reference to the
  pred so the caller can free the half-finished filter

- changes the confusing i == MAX_FILTER_PRED - 1 comparison
  previously remarked upon

This affects only per-subsystem event filtering.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237796808.7527.40.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: fix bug in copy_pred()
Tom Zanussi [Mon, 23 Mar 2009 08:26:42 +0000 (03:26 -0500)]
tracing/filters: fix bug in copy_pred()

Impact: fix potential crash on subsystem filter expression freeing

When making a copy of the predicate, pred->field_name needs to be
duplicated in the copy as well, otherwise bad things can happen due to
later multiple frees of the same string.

This affects only per-subsystem event filtering.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237796802.7527.39.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/filters: use list_for_each_entry_safe
Tom Zanussi [Mon, 23 Mar 2009 08:26:28 +0000 (03:26 -0500)]
tracing/filters: use list_for_each_entry_safe

Impact: cleanup

Use list_for_each_entry_safe instead of list_for_each_entry in
find_event_field().

Reported-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237796788.7527.35.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/events: don't discard an event after commit
Frederic Weisbecker [Sun, 22 Mar 2009 23:18:39 +0000 (00:18 +0100)]
tracing/events: don't discard an event after commit

When we want to filter an event, the filter test is done after
the event is commited to the ring-buffer to be discarded later if
needed.

But a reader could be reading this event while we are trying to discard
it. Other kind of racy events can even happen because the event is
commited and can be read and/or consumed.

What we want is to discard the event before committing it.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <1237763919-21505-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/ftrace: make nop-tracer use polling wait for events on pipe
Frederic Weisbecker [Sun, 22 Mar 2009 22:10:47 +0000 (23:10 +0100)]
tracing/ftrace: make nop-tracer use polling wait for events on pipe

Impact: display events when they arrive

Now that the events don't use wake_up() anymore, we need the nop
tracer to poll waiting for events on the pipe. Especially because
nop is useful to look at orphan traces types (traces types that
don't rely on specific tracers) because it doesn't produce traces
itself.

And unlike other tracers that trigger specific traces periodically,
nop triggers no traces by itself that can wake him.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237759847-21025-5-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/events: don't use wake up for events
Frederic Weisbecker [Sun, 22 Mar 2009 22:10:46 +0000 (23:10 +0100)]
tracing/events: don't use wake up for events

Impact: fix hard-lockup with sched switch events

Some ftrace events, such as sched wakeup, can be traced
while the runqueue lock is hold. Since they are using
trace_current_buffer_unlock_commit(), they call wake_up()
which can try to grab the runqueue lock too, resulting in
a deadlock.

Now for all event, we call a new helper:
trace_nowake_buffer_unlock_commit() which do pretty the same than
trace_current_buffer_unlock_commit() except than it doesn't call
trace_wake_up().

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237759847-21025-4-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/events: make the filter files writable
Frederic Weisbecker [Sun, 22 Mar 2009 22:10:43 +0000 (23:10 +0100)]
tracing/events: make the filter files writable

We need the filter files to be writable, the current
filter file permissions are only set readable.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <1237759847-21025-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add run-time field descriptions for event filtering, kfree fix
Ingo Molnar [Sun, 22 Mar 2009 17:41:59 +0000 (18:41 +0100)]
tracing: add run-time field descriptions for event filtering, kfree fix

Impact: fix potential kfree of random data in (rare) failure path

Zero-initialize the field structure.

Reported-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <1237710639.7703.46.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add per-subsystem filtering
Tom Zanussi [Sun, 22 Mar 2009 08:31:17 +0000 (03:31 -0500)]
tracing: add per-subsystem filtering

This patch adds per-subsystem filtering to the event tracing subsystem.

It adds a 'filter' debugfs file to each subsystem directory.  This file
can be written to to set filters; reading from it will display the
current set of filters set for that subsystem.

Basically what it does is propagate the filter down to each event
contained in the subsystem.  If a particular event doesn't have a field
with the name specified in the filter, it simply doesn't get set for
that event.  You can verify whether or not the filter was set for a
particular event by looking at the filter file for that event.

As with per-event filters, compound expressions are supported, echoing
'0' to the subsystem's filter file clears all filters in the subsystem,
etc.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710677.7703.49.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add per-event filtering
Tom Zanussi [Sun, 22 Mar 2009 08:31:04 +0000 (03:31 -0500)]
tracing: add per-event filtering

This patch adds per-event filtering to the event tracing subsystem.

It adds a 'filter' debugfs file to each event directory.  This file can
be written to to set filters; reading from it will display the current
set of filters set for that event.

Basically, any field listed in the 'format' file for an event can be
filtered on (including strings, but not yet other array types) using
either matching ('==') or non-matching ('!=') 'predicates'.  A
'predicate' can be either a single expression:

 # echo pid != 0 > filter

 # cat filter
 pid != 0

or a compound expression of up to 8 sub-expressions combined using '&&'
or '||':

 # echo comm == Xorg > filter
 # echo "&& sig != 29" > filter

 # cat filter
 comm == Xorg
 && sig != 29

Only events having field values matching an expression will be available
in the trace output; non-matching events are discarded.

Note that a compound expression is built up by echoing each
sub-expression separately - it's not the most efficient way to do
things, but it keeps the parser simple and assumes that compound
expressions will be relatively uncommon.  In any case, a subsequent
patch introducing a way to set filters for entire subsystems should
mitigate any need to do this for lots of events.

Setting a filter without an '&&' or '||' clears the previous filter
completely and sets the filter to the new expression:

 # cat filter
 comm == Xorg
 && sig != 29

 # echo comm != Xorg

 # cat filter
 comm != Xorg

To clear a filter, echo 0 to the filter file:

 # echo 0 > filter
 # cat filter
 none

The limit of 8 predicates for a compound expression is arbitrary - for
efficiency, it's implemented as an array of pointers to predicates, and
8 seemed more than enough for any filter...

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710665.7703.48.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add ring_buffer_event_discard() to ring buffer
Tom Zanussi [Sun, 22 Mar 2009 08:30:49 +0000 (03:30 -0500)]
tracing: add ring_buffer_event_discard() to ring buffer

This patch overloads RINGBUF_TYPE_PADDING to provide a way to discard
events from the ring buffer, for the event-filtering mechanism
introduced in a subsequent patch.

I did the initial version but thanks to Steven Rostedt for adding
the parts that actually made it work. ;-)

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: fix four sparse warnings
Dmitri Vorobiev [Sun, 22 Mar 2009 17:11:11 +0000 (19:11 +0200)]
tracing: fix four sparse warnings

Impact: cleanup.

This patch fixes the following sparse warnings:

 kernel/trace/trace.c:385:9: warning: symbol 'trace_seq_to_buffer' was
 not declared. Should it be static?

 kernel/trace/trace_clock.c:29:13: warning: symbol 'trace_clock_local'
 was not declared. Should it be static?

 kernel/trace/trace_clock.c:54:13: warning: symbol 'trace_clock' was not
 declared. Should it be static?

 kernel/trace/trace_clock.c:74:13: warning: symbol 'trace_clock_global'
 was not declared. Should it be static?

Signed-off-by: Dmitri Vorobiev <dmitri.vorobiev@movial.com>
LKML-Reference: <1237741871-5827-4-git-send-email-dmitri.vorobiev@movial.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing, Text Edit Lock: Fix one sparse warning in kernel/extable.c
Dmitri Vorobiev [Sun, 22 Mar 2009 17:11:10 +0000 (19:11 +0200)]
tracing, Text Edit Lock: Fix one sparse warning in kernel/extable.c

Impact: cleanup.

The global mutex text_mutex if declared in linux/memory.h, so
this file needs to be included into kernel/extable.c, where the
same mutex is defined. This fixes the following sparse warning:

 kernel/extable.c:32:1: warning: symbol 'text_mutex' was not declared.
 Should it be static?

Signed-off-by: Dmitri Vorobiev <dmitri.vorobiev@movial.com>
LKML-Reference: <1237741871-5827-3-git-send-email-dmitri.vorobiev@movial.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: add run-time field descriptions for event filtering
Tom Zanussi [Sun, 22 Mar 2009 08:30:39 +0000 (03:30 -0500)]
tracing: add run-time field descriptions for event filtering

This patch makes the field descriptions defined for event tracing
available at run-time, for the event-filtering mechanism introduced
in a subsequent patch.

The common event fields are prepended with 'common_' in the format
display, allowing them to be distinguished from the other fields
that might internally have same name and can therefore be
unambiguously used in filters.

Signed-off-by: Tom Zanussi <tzanussi@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710639.7703.46.camel@charm-linux>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge branches 'tracing/ftrace', 'tracing/hw-breakpoints', 'tracing/ring-buffer'...
Ingo Molnar [Sun, 22 Mar 2009 17:10:02 +0000 (18:10 +0100)]
Merge branches 'tracing/ftrace', 'tracing/hw-breakpoints', 'tracing/ring-buffer', 'tracing/textedit' and 'linus' into tracing/core

15 years agotracing: keep the tracing buffer after self-test failure
Frederic Weisbecker [Sun, 22 Mar 2009 14:13:07 +0000 (15:13 +0100)]
tracing: keep the tracing buffer after self-test failure

Instead of using ftrace_dump_on_oops, it's far more convenient
to have the trace leading up to a self-test failure available
in /debug/tracing/trace.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237694675-23509-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/function-graph-tracer: prevent hangs during self-tests
Frederic Weisbecker [Sun, 22 Mar 2009 04:04:35 +0000 (05:04 +0100)]
tracing/function-graph-tracer: prevent hangs during self-tests

Impact: detect tracing related hangs

Sometimes, with some configs, the function graph tracer can make
the timer interrupt too much slow, hanging the kernel in an endless
loop of timer interrupts servicing.

As suggested by Ingo, this patch brings a watchdog which stops the
selftest after a defined number of functions traced, definitely
disabling this tracer.

For those who want to debug the cause of the function graph trace
hang, you can pass the ftrace_dump_on_oops kernel parameter to dump
the traces after this hang detection.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237694675-23509-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: avoid accessing NULL bdev->bd_disk
Li Zefan [Fri, 20 Mar 2009 02:34:00 +0000 (10:34 +0800)]
blktrace: avoid accessing NULL bdev->bd_disk

bdev->bd_disk can be NULL, if the block device is not opened.

Try this against an unmounted partition, and you'll see NULL dereference:

  # echo 1 > /sys/block/sda/sda5/enable

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C30098.6080107@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: remove sysfs_blk_trace_enable_show/store()
Li Zefan [Fri, 20 Mar 2009 03:33:55 +0000 (11:33 +0800)]
blktrace: remove sysfs_blk_trace_enable_show/store()

sysfs_blk_trace_enable_show()/store() share most of code with
sysfs_blk_trace_attr_show()/store().

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C30EA3.1060004@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: report EBUSY correctly
Li Zefan [Fri, 20 Mar 2009 01:49:08 +0000 (09:49 +0800)]
blktrace: report EBUSY correctly

blk_trace_remove_queue() returns EINVAL if q->blk_trace == NULL,
but blk_trace_setup_queue() doesn't return EBUSY if
q->blk_trace != NULL.

 # echo 0 > sdaX/trace/enable
 # echo 0 > sdaX/trace/enable
 bash: echo: write error: Invalid argument
 # echo 1 > sdaX/trace/enable
 # echo 1 > sdaX/trace/enable
 (should return EBUSY)

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C2F614.2010101@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: don't increase blk_probes_ref if failed to setup blk trace
Li Zefan [Fri, 20 Mar 2009 01:48:47 +0000 (09:48 +0800)]
blktrace: don't increase blk_probes_ref if failed to setup blk trace

do_blk_trace_setup() may return EBUSY, but the current code
doesn't decrease blk_probes_ref in this case.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C2F5FF.80002@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: remove blk_probe_mutex
Li Zefan [Fri, 20 Mar 2009 01:48:26 +0000 (09:48 +0800)]
blktrace: remove blk_probe_mutex

blk_register_tracepoints() always returns 0, so make it return void,
thus we don't need to use blk_probe_mutex to protect blk_probes_ref.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C2F5EA.8060606@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: make blk_tracer_enabled a bool flag
Li Zefan [Fri, 20 Mar 2009 01:48:03 +0000 (09:48 +0800)]
blktrace: make blk_tracer_enabled a bool flag

It doesn't have to be a counter, and it can be a bool flag instead.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <49C2F5D3.8090104@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoblktrace: fix possible memory leak
Li Zefan [Fri, 20 Mar 2009 01:47:30 +0000 (09:47 +0800)]
blktrace: fix possible memory leak

When we failed to create "block" debugfs dir, we should do some
cleanups.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <49C2F5B2.8000800@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing/ring-buffer: don't annotate rb_cpu_notify with __cpuinit
Frederic Weisbecker [Sat, 21 Mar 2009 03:33:36 +0000 (04:33 +0100)]
tracing/ring-buffer: don't annotate rb_cpu_notify with __cpuinit

Impact: remove a section warning

CONFIG_DEBUG_SECTION_MISMATCH raises the following warning on -tip:

  WARNING: kernel/trace/built-in.o(.text+0x5bc5): Section mismatch in
  reference from the function ring_buffer_alloc() to the function
  .cpuinit.text:rb_cpu_notify()
  The function ring_buffer_alloc() references
  the function __cpuinit rb_cpu_notify().

This is actually harmless. The code in the ring buffer don't build
rb_cpu_notify and other cpu hotplug stuffs when !CONFIG_HOTPLUG_CPU
so we have no risk to reference freed memory here (it would even
be harmless if we unconditionally build it because register_cpu_notifier
would do nothing when !CONFIG_HOTPLUG_CPU.

But since ring_buffer_alloc() can be called everytime, we don't want it
to be annotated with __cpuinit so we drop the __cpuinit from
rb_cpu_notify.

This is not a waste of memory because it is only defined and used on
CONFIG_HOTPLUG_CPU.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237606416-22268-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing, Text Edit Lock - kprobes architecture independent support, nommu fix
Ingo Molnar [Fri, 20 Mar 2009 10:05:04 +0000 (11:05 +0100)]
tracing, Text Edit Lock - kprobes architecture independent support, nommu fix

Impact: build fix on SH !CONFIG_MMU

Stephen Rothwell reported this linux-next build failure on the SH
architecture:

  kernel/built-in.o: In function `disable_all_kprobes':
  kernel/kprobes.c:1382: undefined reference to `text_mutex'
  [...]

And observed:

| Introduced by commit 4460fdad85becd569f11501ad5b91814814335ff ("tracing,
| Text Edit Lock - kprobes architecture independent support") from the
| tracing tree.  text_mutex is defined in mm/memory.c which is only built
| if CONFIG_MMU is defined, which is not true for sh allmodconfig.

Move this lock to kernel/extable.c (which is already home to various
kernel text related routines), which file is always built-in.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
LKML-Reference: <20090320110602.86351a91.sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoftrace: event profile hooks
Peter Zijlstra [Thu, 19 Mar 2009 19:26:15 +0000 (20:26 +0100)]
ftrace: event profile hooks

Impact: new tracing infrastructure feature

Provide infrastructure to generate software perf counter events
from tracepoints.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20090319194233.557364871@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoftrace: ensure every event gets an id
Peter Zijlstra [Thu, 19 Mar 2009 19:26:14 +0000 (20:26 +0100)]
ftrace: ensure every event gets an id

Impact: widen user-space visibe event IDs to all events

Previously only TRACE_EVENT events got ids, because only they
generated raw output which needs to be demuxed from the trace.

In order to provide a unique ID for each event, register everybody,
regardless.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20090319194233.464914218@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoftrace: provide an id file for each event
Peter Zijlstra [Thu, 19 Mar 2009 19:26:13 +0000 (20:26 +0100)]
ftrace: provide an id file for each event

Since not every event has a format file to read the id from,
expose it explicitly in a separate file.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20090319194233.372534033@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge branch 'tip/tracing/ftrace' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Fri, 20 Mar 2009 09:15:13 +0000 (10:15 +0100)]
Merge branch 'tip/tracing/ftrace' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-2.6-trace into tracing/ftrace

15 years agoMerge branches 'tracing/ftrace', 'tracing/kprobes', 'tracing/tasks' and 'linus' into...
Ingo Molnar [Fri, 20 Mar 2009 09:14:53 +0000 (10:14 +0100)]
Merge branches 'tracing/ftrace', 'tracing/kprobes', 'tracing/tasks' and 'linus' into tracing/core

15 years agoaio: lookup_ioctx can return the wrong value when looking up a bogus context
Jeff Moyer [Thu, 19 Mar 2009 00:04:21 +0000 (17:04 -0700)]
aio: lookup_ioctx can return the wrong value when looking up a bogus context

The libaio test harness turned up a problem whereby lookup_ioctx on a
bogus io context was returning the 1 valid io context from the list
(harness/cases/3.p).

Because of that, an extra put_iocontext was done, and when the process
exited, it hit a BUG_ON in the put_iocontext macro called from exit_aio
(since we expect a users count of 1 and instead get 0).

The problem was introduced by "aio: make the lookup_ioctx() lockless"
(commit abf137dd7712132ee56d5b3143c2ff61a72a5faa).

Thanks to Zach for pointing out that hlist_for_each_entry_rcu will not
return with a NULL tpos at the end of the loop, even if the entry was
not found.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Zach Brown <zach.brown@oracle.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoeventfd: remove fput() call from possible IRQ context
Davide Libenzi [Thu, 19 Mar 2009 00:04:19 +0000 (17:04 -0700)]
eventfd: remove fput() call from possible IRQ context

Remove a source of fput() call from inside IRQ context.  Myself, like Eric,
wasn't able to reproduce an fput() call from IRQ context, but Jeff said he was
able to, with the attached test program.  Independently from this, the bug is
conceptually there, so we might be better off fixing it.  This patch adds an
optimization similar to the one we already do on ->ki_filp, on ->ki_eventfd.
Playing with ->f_count directly is not pretty in general, but the alternative
here would be to add a brand new delayed fput() infrastructure, that I'm not
sure is worth it.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoMove cc-option to below arch-specific setup
Linus Torvalds [Thu, 19 Mar 2009 22:53:19 +0000 (15:53 -0700)]
Move cc-option to below arch-specific setup

Sam Ravnborg says:
 "We have several architectures that plays strange games with $(CC) and
  $(CROSS_COMPILE).

  So we need to postpone any use of $(call cc-option..) until we have
  included the arch specific Makefile so we try with the correct $(CC)
  version."

Requested-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoMerge branch 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6
Linus Torvalds [Thu, 19 Mar 2009 21:56:35 +0000 (14:56 -0700)]
Merge branch 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6

* 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6:
  [S390] make page table upgrade work again
  [S390] make page table walking more robust
  [S390] Dont check for pfn_valid() in uaccess_pt.c
  [S390] ftrace/mcount: fix kernel stack backchain
  [S390] topology: define SD_MC_INIT to fix performance regression
  [S390] __div64_31 broken for CONFIG_MARCH_G5

15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
Linus Torvalds [Thu, 19 Mar 2009 21:50:15 +0000 (14:50 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  HID: fix waitqueue usage in hiddev
  HID: fix incorrect free in hiddev

15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
Linus Torvalds [Thu, 19 Mar 2009 21:49:55 +0000 (14:49 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable

* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable:
  Btrfs: Clear space_info full when adding new devices
  Btrfs: Fix locking around adding new space_info

15 years agofunction-graph: show binary events as comments
Steven Rostedt [Thu, 19 Mar 2009 19:14:46 +0000 (15:14 -0400)]
function-graph: show binary events as comments

With the added TRACE_EVENT macro, the events no longer appear in
the function graph tracer. This was because the function graph
did not know how to display the entries. The graph tracer was
only aware of its own entries and the printk entries.

By using the event call back feature, the graph tracer can now display
the events.

 # echo irq > /debug/tracing/set_event

Which can show:

 0)               |          handle_IRQ_event() {
 0)               |            /* irq_handler_entry: irq=48 handler=eth0 */
 0)               |            e1000_intr() {
 0)   0.926 us    |              __napi_schedule();
 0)   3.888 us    |            }
 0)               |            /* irq_handler_exit: irq=48 return=handled */
 0)   0.655 us    |            runqueue_is_locked();
 0)               |            __wake_up() {
 0)   0.831 us    |              _spin_lock_irqsave();

The irq entry and exit events show up as comments.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agotracing: remove recording function depth from trace_printk
Steven Rostedt [Thu, 19 Mar 2009 18:03:53 +0000 (14:03 -0400)]
tracing: remove recording function depth from trace_printk

The function depth in trace_printk was to facilitate the function
graph output. Now that the function graph calculates the depth within
the trace output, we no longer need to record the depth when the
trace_printk is called.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agofunction-graph: calculate function depth within function graph tracer
Steven Rostedt [Thu, 19 Mar 2009 17:24:42 +0000 (13:24 -0400)]
function-graph: calculate function depth within function graph tracer

Currently, the function graph tracer depends on the trace_printk
to record the depth. All the information is already there in the trace
to calculate function depth, with the exception of having the printk
be the first item. But as soon as a entry or exit is reached, then
we know the depth.

This patch changes the iter->private data from recording a per cpu
last_pid, to a structure that holds both the last_pid and the current
depth. This data is used to determine the function depth for the
printks.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agotracing: make print_(b)printk_msg_only global
Steven Rostedt [Thu, 19 Mar 2009 16:20:38 +0000 (12:20 -0400)]
tracing: make print_(b)printk_msg_only global

This patch makes print_printk_msg_only and print_bprintk_msg_only
global for other functions to use. It also renames them by adding
a "trace_" to the beginning to avoid namespace collisions.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agoFix race in create_empty_buffers() vs __set_page_dirty_buffers()
Linus Torvalds [Thu, 19 Mar 2009 18:32:05 +0000 (11:32 -0700)]
Fix race in create_empty_buffers() vs __set_page_dirty_buffers()

Nick Piggin noticed this (very unlikely) race between setting a page
dirty and creating the buffers for it - we need to hold the mapping
private_lock until we've set the page dirty bit in order to make sure
that create_empty_buffers() might not build up a set of buffers without
the dirty bits set when the page is dirty.

I doubt anybody has ever hit this race (and it didn't solve the issue
Nick was looking at), but as Nick says: "Still, it does appear to solve
a real race, which we should close."

Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoAdd '-fwrapv' to gcc CFLAGS
Linus Torvalds [Thu, 19 Mar 2009 18:10:17 +0000 (11:10 -0700)]
Add '-fwrapv' to gcc CFLAGS

This makes sure that gcc doesn't try to optimize away wrapping
arithmetic, which the kernel occasionally uses for overflow testing, ie
things like

if (ptr + offset < ptr)

which technically is undefined for non-unsigned types. See

http://bugzilla.kernel.org/show_bug.cgi?id=12597

for details.

Not all versions of gcc support it, so we need to make it conditional
(it looks like it was introduced in gcc-3.4).

Reminded-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agotracing/ring-buffer: fix non cpu hotplug case
Frederic Weisbecker [Thu, 19 Mar 2009 13:47:33 +0000 (14:47 +0100)]
tracing/ring-buffer: fix non cpu hotplug case

Impact: fix warning with irqsoff tracer

The ring buffer allocates its buffers on pre-smp time (early_initcall).
It means that, at first, only the boot cpu buffer is allocated and
the ring-buffer cpumask only has the boot cpu set (cpu_online_mask).

Later, the secondary cpu will show up and the ring-buffer will be notified
about this event: the appropriate buffer will be allocated and the cpumask
will be updated.

Unfortunately, if !CONFIG_CPU_HOTPLUG, the ring-buffer will not be
notified about the secondary cpus, meaning that the cpumask will have
only the cpu boot set, and only one cpu buffer allocated.

We fix that by using cpu_possible_mask if !CONFIG_CPU_HOTPLUG.

This patch fixes the following warning with irqsoff tracer running:

[  169.317794] WARNING: at kernel/trace/trace.c:466 update_max_tr_single+0xcc/0xf3()
[  169.318002] Hardware name: AMILO Li 2727
[  169.318002] Modules linked in:
[  169.318002] Pid: 5624, comm: bash Not tainted 2.6.29-rc8-tip-02636-g6aafa6c #11
[  169.318002] Call Trace:
[  169.318002]  [<ffffffff81036182>] warn_slowpath+0xea/0x13d
[  169.318002]  [<ffffffff8100b9d6>] ? ftrace_call+0x5/0x2b
[  169.318002]  [<ffffffff8100b9d6>] ? ftrace_call+0x5/0x2b
[  169.318002]  [<ffffffff8100b9d1>] ? ftrace_call+0x0/0x2b
[  169.318002]  [<ffffffff8101ef10>] ? ftrace_modify_code+0xa9/0x108
[  169.318002]  [<ffffffff8106e27f>] ? trace_hardirqs_off+0x25/0x27
[  169.318002]  [<ffffffff8149afe7>] ? _spin_unlock_irqrestore+0x1f/0x2d
[  169.318002]  [<ffffffff81064f52>] ? ring_buffer_reset_cpu+0xf6/0xfb
[  169.318002]  [<ffffffff8106637c>] ? ring_buffer_reset+0x36/0x48
[  169.318002]  [<ffffffff8106aeda>] update_max_tr_single+0xcc/0xf3
[  169.318002]  [<ffffffff8100bc17>] ? sysret_check+0x22/0x5d
[  169.318002]  [<ffffffff8106e3ea>] stop_critical_timing+0x142/0x204
[  169.318002]  [<ffffffff8106e4cf>] trace_hardirqs_on_caller+0x23/0x25
[  169.318002]  [<ffffffff8149ac28>] trace_hardirqs_on_thunk+0x3a/0x3c
[  169.318002]  [<ffffffff8100bc17>] ? sysret_check+0x22/0x5d
[  169.318002] ---[ end trace db76cbf775a750cf ]---

Because this tracer may try to swap two cpu ring buffers for an
unregistered cpu on the ring buffer.

This patch might also fix a fair loss of traces due to unallocated buffers
for secondary cpus.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-b: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237470453-5427-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agofunction-graph: consolidate prologues for output
Steven Rostedt [Thu, 19 Mar 2009 15:29:23 +0000 (11:29 -0400)]
function-graph: consolidate prologues for output

Impact: clean up

The prologue of the function graph entry, return and comments all
start out pretty much the same. Each of these duplicate code and
do so slightly differently.

This patch consolidates the printing of the pid, absolute time,
cpu and proc (and for entry, the interrupt).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agoftrace: protect running nmi (V3)
Lai Jiangshan [Wed, 18 Mar 2009 08:42:57 +0000 (16:42 +0800)]
ftrace: protect running nmi (V3)

When I review the sensitive code ftrace_nmi_enter(), I found
the atomic variable nmi_running does protect NMI VS do_ftrace_mod_code(),
but it can not protects NMI(entered nmi) VS NMI(ftrace_nmi_enter()).

cpu#1                   | cpu#2                 | cpu#3
ftrace_nmi_enter()      | do_ftrace_mod_code()  |
  not modify            |                       |
------------------------|-----------------------|--
executing               | set mod_code_write = 1|
executing             --|-----------------------|--------------------
executing               |                       | ftrace_nmi_enter()
executing               |                       |    do modify
------------------------|-----------------------|-----------------
ftrace_nmi_exit()       |                       |

cpu#3 may be being modified the code which is still being executed on cpu#1,
it will have undefined results and possibly take a GPF, this patch
prevents it occurred.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <49C0B411.30003@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
15 years agotracepoints: dont update zero-sized tracepoint sections
Ingo Molnar [Wed, 18 Mar 2009 18:54:04 +0000 (19:54 +0100)]
tracepoints: dont update zero-sized tracepoint sections

Zero-sized tracepoint sections can occur if tracing is enabled but
no tracepoint is defined. Do not emit a warning in that case.

Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
LKML-Reference: <1237394936.3132.1.camel@localhost.localdomain>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agotracing: fix oops in tracepoint_update_probe_range()
Jaswinder Singh Rajput [Wed, 18 Mar 2009 16:48:56 +0000 (22:18 +0530)]
tracing: fix oops in tracepoint_update_probe_range()

Change this crash:

 BUG: unable to handle kernel NULL pointer dereference at (null)
 IP: [<ffffffff8107d4de>] tracepoint_update_probe_range+0x1f/0x9b
 PGD 13d5fb067 PUD 13d688067 PMD 0
 Oops: 0000 [#1] SMP

To a more debuggable WARN_ONCE().

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237394936.3132.1.camel@localhost.localdomain>
[ moved the check outside the lock and added a WARN_ON(). ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging-2.6
Linus Torvalds [Wed, 18 Mar 2009 16:34:17 +0000 (09:34 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging-2.6:
  Staging: benet: remove driver now that it is merged in drivers/net/