The code to update the print formats for events requires a vprintf
format in the trace_seq. This patch adds that interface.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
TRACE_EVENT is a more generic way to define tracepoints. Doing so adds
these new capabilities to this tracepoint:
- zero-copy and per-cpu splice() tracing
- binary tracing without printf overhead
- structured logging records exposed under /debug/tracing/events
- trace events embedded in function tracer output and other plugins
- user-defined, per tracepoint filter expressions
...
Cons:
- no dev_t info for the output of plug, unplug_timer and unplug_io events.
no dev_t info for getrq and sleeprq events if bio == NULL.
no dev_t info for rq_abort,...,rq_requeue events if rq->rq_disk == NULL.
This is mainly because we can't get the deivce from a request queue.
But this may change in the future.
- A packet command is converted to a string in TP_assign, not TP_print.
While blktrace do the convertion just before output.
Since pc requests should be rather rare, this is not a big issue.
- In blktrace, an event can have 2 different print formats, but a TRACE_EVENT
has a unique format, which means we have some unused data in a trace entry.
The overhead is minimized by using __dynamic_array() instead of __array().
I've benchmarked the ioctl blktrace vs the splice based TRACE_EVENT tracing:
dd dd + ioctl blktrace dd + TRACE_EVENT (splice)
1 7.36s, 42.7 MB/s 7.50s, 42.0 MB/s 7.41s, 42.5 MB/s
2 7.43s, 42.3 MB/s 7.48s, 42.1 MB/s 7.43s, 42.4 MB/s
3 7.38s, 42.6 MB/s 7.45s, 42.2 MB/s 7.41s, 42.5 MB/s
So the overhead of tracing is very small, and no regression when using
those trace events vs blktrace.
And the binary output of TRACE_EVENT is much smaller than blktrace:
# ls -l -h
-rw-r--r-- 1 root root 8.8M 06-09 13:24 sda.blktrace.0
-rw-r--r-- 1 root root 195K 06-09 13:24 sda.blktrace.1
-rw-r--r-- 1 root root 2.7M 06-09 13:25 trace_splice.out
Following are some comparisons between TRACE_EVENT and blktrace:
plug:
kjournald-480 [000] 303.084981: block_plug: [kjournald]
kjournald-480 [000] 303.084981: 8,0 P N [kjournald]
unplug_io:
kblockd/0-118 [000] 300.052973: block_unplug_io: [kblockd/0] 1
kblockd/0-118 [000] 300.052974: 8,0 U N [kblockd/0] 1
remap:
kjournald-480 [000] 303.085042: block_remap: 8,0 W 102736992 + 8 <- (8,8) 33384
kjournald-480 [000] 303.085043: 8,0 A W 102736992 + 8 <- (8,8) 33384
bio_backmerge:
kjournald-480 [000] 303.085086: block_bio_backmerge: 8,0 W 102737032 + 8 [kjournald]
kjournald-480 [000] 303.085086: 8,0 M W 102737032 + 8 [kjournald]
getrq:
kjournald-480 [000] 303.084974: block_getrq: 8,0 W 102736984 + 8 [kjournald]
kjournald-480 [000] 303.084975: 8,0 G W 102736984 + 8 [kjournald]
bash-2066 [001] 1072.953770: 8,0 G N [bash]
bash-2066 [001] 1072.953773: block_getrq: 0,0 N 0 + 0 [bash]
rq_complete:
konsole-2065 [001] 300.053184: block_rq_complete: 8,0 W () 103669040 + 16 [0]
konsole-2065 [001] 300.053191: 8,0 C W 103669040 + 16 [0]
ksoftirqd/1-7 [001] 1072.953811: 8,0 C N (5a 00 08 00 00 00 00 00 24 00) [0]
ksoftirqd/1-7 [001] 1072.953813: block_rq_complete: 0,0 N (5a 00 08 00 00 00 00 00 24 00) 0 + 0 [0]
rq_insert:
kjournald-480 [000] 303.084985: block_rq_insert: 8,0 W 0 () 102736984 + 8 [kjournald]
kjournald-480 [000] 303.084986: 8,0 I W 102736984 + 8 [kjournald]
Changelog from v2 -> v3:
- use the newly introduced __dynamic_array().
Changelog from v1 -> v2:
- use __string() instead of __array() to minimize the memory required
to store hex dump of rq->cmd().
- support large pc requests.
- add missing blk_fill_rwbs_rq() in block_rq_requeue TRACE_EVENT.
- some cleanups.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A2DF669.5070905@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The update of ret got mistakenly added to the if statement of
rb_try_to_discard. The variable ret should be 1 on commit and zero
otherwise.
[ Impact: fix compiler warning and real bug ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
These are defined as static cpumask_var_t so if MAXSMP is not used,
they are cleared already. Avoid surprises when MAXSMP is enabled.
Signed-off-by: Yinghai Lu <yinghai.lu@kernel.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
On Sun, 7 Jun 2009, Ingo Molnar wrote:
> Testing tracer sched_switch: <6>Starting ring buffer hammer
> PASSED
> Testing tracer sysprof: PASSED
> Testing tracer function: PASSED
> Testing tracer irqsoff:
> =============================================
> PASSED
> Testing tracer preemptoff: PASSED
> Testing tracer preemptirqsoff: [ INFO: possible recursive locking detected ]
> PASSED
> Testing tracer branch: 2.6.30-rc8-tip-01972-ge5b9078-dirty #5760
> ---------------------------------------------
> rb_consumer/431 is trying to acquire lock:
> (&cpu_buffer->reader_lock){......}, at: [<c109eef7>] ring_buffer_reset_cpu+0x37/0x70
>
> but task is already holding lock:
> (&cpu_buffer->reader_lock){......}, at: [<c10a019e>] ring_buffer_consume+0x7e/0xc0
>
> other info that might help us debug this:
> 1 lock held by rb_consumer/431:
> #0: (&cpu_buffer->reader_lock){......}, at: [<c10a019e>] ring_buffer_consume+0x7e/0xc0
The ring buffer is a generic structure, and can be used outside of
ftrace. If ftrace traces within the use of the ring buffer, it can produce
false positives with lockdep.
This patch passes in a static lock key into the allocation of the ring
buffer, so that different ring buffers will have their own lock class.
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1244477919.13761.9042.camel@twins>
[ store key in ring buffer descriptor ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Our async work synchronization was broken by "async: make sure
independent async domains can't accidentally entangle" (commit
d5a877e8dd), because it would report
the wrong lowest active async ID when there was both running and
pending async work.
This caused things like no being able to read the root filesystem,
resulting in missing console devices and inability to run 'init',
causing a boot-time panic.
This fixes it by properly returning the lowest pending async ID: if
there is any running async work, that will have a lower ID than any
pending work, and we should _not_ look at the pending work list.
There were alternative patches from Jaswinder and James, but this one
also cleans up the code by removing the pointless 'ret' variable and
the unnecesary testing for an empty list around 'for_each_entry()' (if
the list is empty, the for_each_entry() thing just won't execute).
Fixes-bug: http://bugzilla.kernel.org/show_bug.cgi?id=13474
Reported-and-tested-by: Chris Clayton <chris2553@googlemail.com>
Cc: Jaswinder Singh Rajput <jaswinder@kernel.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We want to use hrtimers in UBIFS (for write-buffer write-back timer).
We need the 'hrtimer_set_expires_range_ns()', which is an in-line
function which uses 'ktime_add_safe()'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Extend generic event enumeration with the PERF_TYPE_HW_CACHE
method.
This is a 3-dimensional space:
{ L1-D, L1-I, L2, ITLB, DTLB, BPU } x
{ load, store, prefetch } x
{ accesses, misses }
User-space passes in the 3 coordinates and the kernel provides
a counter. (if the hardware supports that type and if the
combination makes sense.)
Combinations that make no sense produce a -EINVAL.
Combinations that are not supported by the hardware produce -ENOTSUP.
Extend the tools to deal with this, and rewrite the event symbol
parsing code with various popular aliases for the units and
access methods above. So 'l1-cache-miss' and 'l1d-read-ops' are
both valid aliases.
( x86 is supported for now, with the Nehalem event table filled in,
and with Core2 and Atom having placeholder tables. )
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Counter type is a frequently used value and we do a lot of
bit juggling by encoding and decoding it from attr->config.
Clean this up by creating a separate attr->type field.
Also clean up the various similarly complex user-space bits
all around counter attribute management.
The net improvement is significant, and it will be easier
to add a new major type (which is what triggered this cleanup).
(This changes the ABI, all tools are adapted.)
(PowerPC build-tested.)
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to allow easy tracking of the period, also provide means of
adding it to the sample data.
Signed-off-by: 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>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The purpose of PERF_SAMPLE_CONFIG was to identify the counters,
since then we've added counter ids, use those instead.
Signed-off-by: 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>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to track the vdso also generate mmap events for
install_special_mapping().
Signed-off-by: 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>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Commit 8e3747c1 ("perf_counter: Change data head from u32 to u64")
changed the type of 'head' in struct perf_mmap_data from atomic_t
to atomic_long_t, but missed converting one use of atomic_read on
it to atomic_long_read. The effect of using atomic_read rather than
atomic_long_read on powerpc (and other big-endian architectures) is
that we get the high half of the 64-bit quantity, resulting in the
cmpxchg retry loop in perf_output_begin spinning forever as soon as
data->head becomes non-zero. On little-endian architectures such as
x86 we would get the low half, resulting in a lockup once data->head
becomes greater than 4G.
This fixes it by using atomic_long_read rather than atomic_read.
[ Impact: fix perfcounter lockup on PowerPC / big-endian systems ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18984.33964.21541.743096@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Commit 95a3540da9 ("ptrace_detach: the wrong
wakeup breaks the ERESTARTxxx logic") removed the "extra"
wake_up_process() from ptrace_detach(), but as Jan pointed out this breaks
the compatibility.
I believe the changelog is right and this wake_up() is wrong in many
ways, but GDB assumes that ptrace(PTRACE_DETACH, child, 0, 0) always
wakes up the tracee.
Despite the fact this breaks SIGNAL_STOP_STOPPED/group_stop_count logic,
and despite the fact this wake_up_process() can break another
assumption: PTRACE_DETACH with SIGSTOP should leave the tracee in
TASK_STOPPED case. Because the untraced child can dequeue SIGSTOP and
call do_signal_stop() before ptrace_detach() calls wake_up_process().
Revert this change for now. We need some fixes even if we we want to keep
the current behaviour, but these fixes are not for 2.6.30.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The "trace || CLONE_PTRACE" check in tracehook_report_clone() is not right,
- If the untraced task does clone(CLONE_PTRACE) the new child is not traced,
we must not queue SIGSTOP.
- If we forked the traced task, but the tracer exits and untraces both the
forking task and the new child (after copy_process() drops tasklist_lock),
we should not queue SIGSTOP too.
Change the code to check task_ptrace() != 0 instead. This is still racy, but
the race is harmless.
We can race with another tracer attaching to this child, or the tracer can
exit and detach in parallel. But giwen that we didn't do wake_up_new_task()
yet, the child must have the pending SIGSTOP anyway.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In name of keeping it simple, only track mmap events. Userspace
will have to remove old overlapping maps when it encounters them.
Signed-off-by: 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>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Create a fork event so that we can easily clone the comm and
dso maps without having to generate all those events.
Signed-off-by: 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>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch removes the dependency of mmap_min_addr on CONFIG_SECURITY.
It also sets a default mmap_min_addr of 4096.
mmapping of addresses below 4096 will only be possible for processes
with CAP_SYS_RAWIO.
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Eric Paris <eparis@redhat.com>
Looks-ok-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: James Morris <jmorris@namei.org>
Throttling logic is broken and we can lock up with too small
hw sampling intervals.
Make the throttling code more robust: disable counters even
if we already disabled them.
( Also clean up whitespace damage i noticed while reading
various pieces of code related to throttling. )
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The current method of printing out a stack trace is to add a new line
and print out the trace:
yum-updatesd-3120 [002] 573.691303:
=> do_softirq
=> irq_exit
=> smp_apic_timer_interrupt
=> apic_timer_interrupt
This looks a bit awkward, and if we have both stack and user stack traces
running, it would be nice to have a title to tell them apart, although
it is easy to tell by the output.
This patch adds an annotation to the start of the stack traces:
init-1 [003] 929.304979: <stack trace>
=> user_path_at
=> vfs_fstatat
=> vfs_stat
=> sys_newstat
=> system_call_fastpath
cat-3459 [002] 1016.824040: <user stack trace>
=> <0000003aae6c0250>
=> <00007ffff4b06ae4>
=> <69636172742f6775>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Here is an updated patch to include the extra call to
trace_seq_init() as requested. This is vs. the latest
-tip tree and fixes the use of multiple __print_flags
and __print_symbolic in a single tracer. Also tested
to ensure its working now:
mount.gfs2-2534 [000] 235.850587: gfs2_glock_queue: 8.7 glock 1:2 dequeue PR
mount.gfs2-2534 [000] 235.850591: gfs2_demote_rq: 8.7 glock 1:0 demote EX to NL flags:DI
mount.gfs2-2534 [000] 235.850591: gfs2_glock_queue: 8.7 glock 1:0 dequeue EX
glock_workqueue-2529 [000] 235.850666: gfs2_glock_state_change: 8.7 glock 1:0 state EX => NL tgt:NL dmt:NL flags:lDpI
glock_workqueue-2529 [000] 235.850672: gfs2_glock_put: 8.7 glock 1:0 state NL => IV flags:I
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
LKML-Reference: <1244037123.29604.603.camel@localhost.localdomain>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
According to "events/ftrace/user_stack/format", fix the output of
user stack.
before fix:
sh-1073 [000] 31.137561: <b7f274fe> <- <0804e33c> <- <080835c1>
after fix:
sh-1072 [000] 37.039329:
=> <b7f8a4fe>
=> <0804e33c>
=> <080835c1>
Signed-off-by: walimis <walimisdev@gmail.com>
LKML-Reference: <1244016090-7814-3-git-send-email-walimisdev@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
According to "events/ftrace/kernel_stack/format", output format of
kernel stack should use "=>" instead of "<=".
The second problem is that we shouldn't skip the first entry in the stack,
although it seems to be duplicated when used in the "function" tracer,
but events also use it. If we skip the first one, we will drop the topmost
entry of the stack.
The last problem is that if the last entry is ULONG_MAX(0xffffffff), we should
drop it, otherwise it will print a NULL name line.
before fix:
sh-1072 [000] 26.957239: sched_process_fork: parent sh:1072 child sh:1073
sh-1072 [000] 26.957262:
<= syscall_call
<=
sh-1072 [000] 26.957744: sched_switch: task sh:1072 [120] (R) ==> sh:1073 [120]
sh-1072 [000] 26.957752:
<= preempt_schedule
<= wake_up_new_task
<= do_fork
<= sys_clone
<= syscall_call
<=
After fix:
sh-1075 [000] 39.791848: sched_process_fork: parent sh:1075 child sh:1076
sh-1075 [000] 39.791871:
=> sys_clone
=> syscall_call
sh-1075 [000] 39.792713: sched_switch: task sh:1075 [120] (R) ==> sh:1076 [120]
sh-1075 [000] 39.792722:
=> schedule
=> preempt_schedule
=> wake_up_new_task
=> do_fork
=> sys_clone
=> syscall_call
Signed-off-by: walimis <walimisdev@gmail.com>
LKML-Reference: <1244016090-7814-2-git-send-email-walimisdev@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The last entry in the stack_dump_trace is ULONG_MAX, which is not
a valid entry, but max_stack_trace.nr_entries has accounted for it.
So when printing the header, we should decrease it by one.
Before fix, print as following, for example:
Depth Size Location (53 entries) <--- should be 52
----- ---- --------
0) 3264 108 update_wall_time+0x4d5/0x9a0
...
51) 80 80 syscall_call+0x7/0xb
^^^
it's correct.
Signed-off-by: walimis <walimisdev@gmail.com>
LKML-Reference: <1244016090-7814-1-git-send-email-walimisdev@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Every buffer page in the ring buffer includes its own time stamp.
When an event is recorded to the ring buffer with a delta time greater
than what can be held in the event header, a time stamp event is created.
If the the create timestamp falls over to the next buffer page, it is
redundant because the buffer page holds a full time stamp. This patch
will try to discard the time stamp when it falls to the start of the
next page.
This change also fixes a issues with disarding events. If most events are
discarded, timestamps will start to creep into the ring buffer. If we
do not discard the timestamps then they can fill up the ring buffer over
time and waste space.
This change will keep time stamps from filling up over another page. If
something is recorded in the buffer page, and the rest is filtered, then
the time stamps can only fill up to the end of the page.
[ Impact: prevent time stamps from filling ring buffer ]
Reported-by: Tim Bird <tim.bird@am.sony.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
There are times that a race may happen that we add a timestamp in a
nested write. This timestamp would just contain a zero delta and serves
no purpose.
Now that we have a way to discard events, this patch will try to discard
the timestamp instead of just wasting the space in the ring buffer.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
There's a bug in ring_buffer_discard_commit. The wrong
pointer is being compared in order to check if the event
can be freed from the buffer rather than discarded
(i.e. marked as PAD).
I noticed this when I was working on duration filtering.
The bug is not deadly - it just results in lots of wasted
space in the buffer. All filtered events are left in
the buffer and marked as discarded, rather than being
removed from the buffer to make space for other events.
Unfortunately, when I fixed this bug, I got errors doing a
filtered function trace. Multiple TIME_EXTEND
events pile up in the buffer, and trigger the
following loop overage warning in rb_iter_peek():
again:
...
if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
return NULL;
I'm not sure what the best way is to fix this. I don't
know if I should extend the loop threshhold, or if I should
make the test more complex (ignore TIME_EXTEND
events), or just get rid of this loop check completely.
Note that if I implement a workaround for this, then I
see another problem from rb_advance_iter(). I haven't
tracked that one down yet.
In general, it seems like the case of removing filtered
events has not been working properly, and so some assumptions
about buffer invariant conditions need to be revisited.
Here's the patch for the simple fix:
Compare correct pointer for checking if an event can be
freed rather than left as discarded in the buffer.
Signed-off-by: Tim Bird <tim.bird@am.sony.com>
LKML-Reference: <4A25BE9E.5090909@am.sony.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
We need the PID namespace and counter ID available when the
counter overflows and we need to generate a sample event.
[ Impact: fix kernel crash with high-frequency sampling ]
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
[ fixed a further crash and cleaned up the initialization a bit ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I noticed missing COMM events and found that we missed
reporting them for pure forks.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On creating a new task while running the function graph tracer, if
we fail to allocate the ret_stack, and then fail the fork, the
code will free the parent ret_stack. This is because the child
duplicated the parent and currently points to the parent's ret_stack.
This patch always initializes the task's ret_stack to NULL.
[ Impact: prevent crash of parent on low memory during fork ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When the function graph tracer is enabled, all new tasks must allocate
a ret_stack to place the return address of functions. This is because
the function graph tracer will replace the real return address with a
call to the tracing of the exit function.
This initialization happens in fork, but it happens too late. If fork
fails, then it will call free_task and that calls the freeing of this
ret_stack. But before initialization happens, the new (failed) task
points to its parents ret_stack. If a fork failure happens during
the function trace, it would be catastrophic for the parent.
Also, there's no need to call ftrace_graph_exit_task from fork, since
it is called by free_task which fork calls on failure.
[ Impact: prevent crash during failed fork running function graph tracer ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The structure isn't hw only and when I read event, I think about those
things that fall out the other end. Rename the thing.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
Cc: Stephane Eranian <eranian@googlemail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since some people worried that 4G might not be a large enough
as an mmap data window, extend it to 64 bit for capable
platforms.
Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
A few renames:
s/irq_period/sample_period/
s/irq_freq/sample_freq/
s/PERF_RECORD_/PERF_SAMPLE_/
s/record_type/sample_type/
And change both the new sample_type and read_format to u64.
Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Stephan raised the issue that we currently cannot distinguish between
similar counters within a group (PERF_RECORD_GROUP uses the config
value as identifier).
Therefore, generate a new ID for each counter using a global u64
sequence counter.
Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The code that handles the tasks ret_stack allocation for every task
assumes that only an interrupt can cause issues (even though interrupts
are disabled).
In reality, the code is allocating the ret_stack for tasks that may be
running on other CPUs and there are not efficient memory barriers to
handle this case.
[ Impact: prevent crash due to using of uninitialized ret_stack variables ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The function graph tracer checks if the task_struct has ret_stack defined
to know if it is OK or not to use it. The initialization is done for
all tasks by one process, but the idle tasks use the same initialization
used by new tasks.
If an interrupt happens on an idle task that just had the ret_stack
created, but before the rest of the initialization took place, then
we can corrupt the return address of the functions.
This patch moves the setting of the task_struct's ret_stack to after
the other variables have been initialized.
[ Impact: prevent kernel panic on idle task when starting function graph ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When the function graph tracer is enabled, it calls the initialization
needed for the init tasks that would be called on all created tasks.
The problem is that this is called every time the function graph tracer
is enabled, and the ret_stack is allocated for the idle tasks each time.
Thus, the old ret_stack is lost and a memory leak is created.
This is also dangerous because if an interrupt happened on another CPU
with the init task and the ret_stack is replaced, we then lose all the
return pointers for the interrupt, and a crash would take place.
[ Impact: fix memory leak and possible crash due to race ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Stop using task_struct::pid and start using PID namespaces.
PIDs will be reported in the PID namespace of the monitoring
task at the moment of counter creation.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This removes the prev_state field of struct perf_counter since
it is now unused. It was only used by the cpu migration
counter, which doesn't use it any more.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <18979.35052.915728.626374@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This fixes the cpu migration software counter to count
correctly even when contexts get swapped from one task to
another. Previously the cpu migration counts reported by perf
stat were bogus, ranging from negative to several thousand for
a single "lat_ctx 2 8 32" run. With this patch the cpu
migration count reported for "lat_ctx 2 8 32" is almost always
between 35 and 44.
This fixes the problem by adding a call into the perf_counter
code from set_task_cpu when tasks are migrated. This enables
us to use the generic swcounter code (with some modifications)
for the cpu migration counter.
This modifies the swcounter code to allow a NULL regs pointer
to be passed in to perf_swcounter_ctx_event() etc. The cpu
migration counter does this because there isn't necessarily a
pt_regs struct for the task available. In this case, the
counter will not have interrupt capability - but the migration
counter didn't have interrupt capability before, so this is no
loss.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <18979.35006.819769.416327@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This arranges for perf_counter's notifier for cpu hotplug
operations to be called earlier than the migration notifier in
sched.c by increasing its priority to 20, compared to the 10
for the migration notifier. The reason for doing this is that
a subsequent commit to convert the cpu migration counter to use
the generic swcounter infrastructure will add a call into the
perf_counter subsystem when tasks get migrated. Therefore the
perf_counter subsystem needs a chance to initialize its per-cpu
data for the new cpu before it can get called from the
migration code.
This also adds a comment to the migration notifier noting that
its priority needs to be lower than that of the perf_counter
notifier.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18981.1900.792795.836858@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
A race was found that if one were to enable and disable the function
profiler repeatedly, then the system can panic. This was because a profiled
function may be preempted just before disabling interrupts. While
the profiler is disabled and then reenabled, the preempted function
could start again, and access the hash as it is being initialized.
This just adds a check in the irq disabled part to check if the profiler
is enabled, and if it is not then it will just exit.
When the system is disabled, the profile_enabled variable is cleared
before calling the unregistering of the function profiler. This
unregistering calls stop machine which also acts as a synchronize schedule.
[ Impact: fix panic in enabling/disabling function profiler ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The trace_pipe did not recognize the latency format flag and would produce
different output than the trace file. The problem was partly due that
the trace flags in the iterator was not set as well as the trace_pipe
zeros out part of the iterator (including the flags) to be able to use
the same routines as the trace file. trace_flags of the iterator should
not cause any problems when not zeroed out by for trace_pipe.
Reported-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
A patch to allow the use of __print_symbolic and __print_flags
from a module. This allows the current GFS2 tracing patch to
build.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
LKML-Reference: <1243868015.29604.542.camel@localhost.localdomain>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
__string() is limited:
- it's a char array, but we may want to define array with other types
- a source string should be available, but we may just know the string size
We introduce __dynamic_array() to break those limitations, and __string()
becomes a wrapper of it. As a side effect, now __get_str() can be used
in TP_fast_assign but not only TP_print.
Take XFS for example, we have the string length in the dirent, but the
string itself is not NULL-terminated, so __dynamic_array() can be used:
TRACE_EVENT(xfs_dir2,
TP_PROTO(struct xfs_da_args *args),
TP_ARGS(args),
TP_STRUCT__entry(
__field(int, namelen)
__dynamic_array(char, name, args->namelen + 1)
...
),
TP_fast_assign(
char *name = __get_str(name);
if (args->namelen)
memcpy(name, args->name, args->namelen);
name[args->namelen] = '\0';
__entry->namelen = args->namelen;
),
TP_printk("name %.*s namelen %d",
__entry->namelen ? __get_str(name) : NULL
__entry->namelen)
);
[ Impact: allow defining dynamic size arrays ]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A2384D2.3080403@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Both event tracer and sched switch plugin are selected by default
by all generic tracers. But if no generic tracer is enabled, their options
appear. But ether one of them will select the other, thus it only
makes sense to have the default tracers be selected by one option.
[ Impact: clean up kconfig menu ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
There are two options that are selected by all tracers, but we want
to have those options available when no tracer is selected. These are
The event tracer and sched switch tracer.
The are enabled by all tracers, but if a tracer is not selected we want
the options to appear. All tracers including them select TRACING.
Thus what we would like to do is:
config EVENT_TRACER
bool "prompt"
depends on TRACING
select TRACING
But that gives us a bug in the kbuild system since we just created a
circular dependency. We only want the prompt to show when TRACING is off.
This patch adds GENERIC_TRACER that all tracers will select instead of
TRACING. The two options (sched switch and event tracer) will select
TRACING directly and depend on !GENERIC_TRACER. This solves the cicular
dependency.
[ Impact: hide options that are selected by default ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When using ftrace=function on the command line to trace functions
on boot up, one can not filter out functions that are commonly called.
This patch adds two new ftrace command line commands.
ftrace_notrace=function-list
ftrace_filter=function-list
Where function-list is a comma separated list of functions to filter.
The ftrace_notrace will make the functions listed not be included
in the function tracing, and ftrace_filter will only trace the functions
listed.
These two act the same as the debugfs/tracing/set_ftrace_notrace and
debugfs/tracing/set_ftrace_filter respectively.
The simple glob expressions that are allowed by the filter files can also
be used by the command line interface.
ftrace_notrace=rcu*,*lock,*spin*
Will not trace any function that starts with rcu, ends with lock, or has
the word spin in it.
Note, if the self tests are enabled, they may interfere with the filtering
set by the command lines.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
register_stat_tracer() uses list_for_each_entry_safe
to check whether a tracer is already present in the list.
But we don't delete anything from the list here, so
we don't need the safe version
[ Impact: cleanup list use is stat tracing ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
- remove duplicate code in stat_seq_init()
- update comments to reflect the change from stat list to stat rbtree
[ Impact: clean up ]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
When closing a trace_stat file, we destroy the rbtree constructed during
file open, but there is memory leak that the root node is not freed.
[ Impact: fix memory leak when closing a trace_stat file ]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Currently the output of trace_stat/workqueues is totally reversed:
# cat /debug/tracing/trace_stat/workqueues
...
1 17 17 210 37 `-blk_unplug_work+0x0/0x57
1 3779 3779 181 11 |-cfq_kick_queue+0x0/0x2f
1 3796 3796 kblockd/1:120
...
The correct output should be:
1 3796 3796 kblockd/1:120
1 3779 3779 181 11 |-cfq_kick_queue+0x0/0x2f
1 17 17 210 37 `-blk_unplug_work+0x0/0x57
It's caused by "tracing/stat: replace linked list by an rbtree for
sorting"
(53059c9b67a62a3dc8c80204d3da42b9267ea5a0).
dummpy_cmp() should return -1, so rb_node will always be inserted as
right-most node in the rbtree, thus we sort the output in ascending
order.
[ Impact: fix the output of trace_stat/workqueues ]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
When the stat tracing framework prepares the entries from a tracer
to output them to the user, it starts by computing a linear sort
through a linked list to give the entries ordered by relevance
to the user.
This is quite ugly and causes a small latency when we begin to
read the file.
This patch changes that by turning the linked list into a red-black
tree. Athough the whole iteration using the start and next tracer
callbacks while opening the file remain the same, it is now much
more fast and scalable.
The rbtree guarantees O(log(n)) insertions whereas a linked
list with linear sorting brought us a O(n) despair. Now the
(visible) latency has disapeared.
[ Impact: kill the latency while starting to read a stat tracer file ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
The "trace" prefix in struct trace_stat_session type is annoying while
reading the trace_stat.c file. It makes the lines longer, and
is not that much useful to explain the sense of this type.
Just keep "struct stat_session" for this type.
[ Impact: make the code a bit more readable ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
The blankline between each cpu's workqueue stat is not necessary, because
the cpu number is enough to part them by eye.
Old style also caused a blankline below headline, and made code complex
by using lock, disableirq and get cpu var.
Old style:
# CPU INSERTED EXECUTED NAME
# | | | |
0 8644 8644 events/0
0 0 0 cpuset
...
0 1 1 kdmflush
1 35365 35365 events/1
...
New style:
# CPU INSERTED EXECUTED NAME
# | | | |
0 8644 8644 events/0
0 0 0 cpuset
...
0 1 1 kdmflush
1 35365 35365 events/1
...
[ Impact: provide more readable code ]
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
cpu_workqueue_stats->first_entry is useless because we can retrieve the
header of a cpu workqueue using:
if (&cpu_workqueue_stats->list == workqueue_cpu_stat(cpu)->list.next)
[ Impact: cleanup ]
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
No need to use list_for_each_entry_safe() in iteration without deleting
any node, we can use list_for_each_entry() instead.
[ Impact: cleanup ]
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
v3: zhaolei@cn.fujitsu.com: Change TRACE_EVENT definition to new format
introduced by Steven Rostedt: consolidate trace and trace_event headers
v2: kosaki@jp.fujitsu.com: print the function names instead of addr, and zap
the work addr
v1: zhaolei@cn.fujitsu.com: Make workqueue tracepoints use TRACE_EVENT macro
TRACE_EVENT is a more generic way to define tracepoints.
Doing so adds these new capabilities to the tracepoints:
- zero-copy and per-cpu splice() tracing
- binary tracing without printf overhead
- structured logging records exposed under /debug/tracing/events
- trace events embedded in function tracer output and other plugins
- user-defined, per tracepoint filter expressions
Then, this patch converts DEFINE_TRACE to TRACE_EVENT in workqueue related
tracepoints.
[ Impact: expand workqueue tracer to events tracing ]
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Merge reason: arch/x86/kernel/irqinit_{32,64}.c unified in irq/numa
and modified in x86/mce3; this merge resolves the conflict.
Conflicts:
arch/x86/kernel/irqinit.c
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Conflicts:
arch/mips/sibyte/bcm1480/irq.c
arch/mips/sibyte/sb1250/irq.c
Merge reason: we gathered a few conflicts plus update to latest upstream fixes.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This changes perf_swcounter_match() so that per-task software
counters can count events that occur while their associated
task is not running. This will allow us to use the generic
software counter code for counting task migrations, which can
occur while the task is not scheduled in.
To do this, we have to distinguish between the situations where
the counter is inactive because its task has been scheduled
out, and those where the counter is inactive because it is part
of a group that was not able to go on the PMU. In the former
case we want the counter to count, but not in the latter case.
If the context is active, we have the latter case. If the
context is inactive then we need to know whether the counter
was counting when the context was last active, which we can
determine by comparing its ->tstamp_stopped timestamp with the
context's timestamp.
This also folds three checks in perf_swcounter_match, checking
perf_event_raw(), perf_event_type() and perf_event_id()
individually, into a single 64-bit comparison on
counter->hw_event.config, as an optimization.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <18979.34810.259718.955621@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This abstracts out the code for locking the context associated
with a task. Because the context might get transferred from
one task to another concurrently, we have to check after
locking the context that it is still the right context for the
task and retry if not. This was open-coded in
find_get_context() and perf_counter_init_task().
This adds a further function for pinning the context for a
task, i.e. marking it so it can't be transferred to another
task. This adds a 'pin_count' field to struct
perf_counter_context to indicate that a context is pinned,
instead of the previous method of setting the parent_gen count
to all 1s. Pinning the context with a pin_count is easier to
undo and doesn't require saving the parent_gen value. This
also adds a perf_unpin_context() to undo the effect of
perf_pin_task_context() and changes perf_counter_init_task to
use it.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <18979.34748.755674.596386@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Merge reason: merge almost-rc8 into perfcounters/core, which was -rc6
based - to pick up the latest upstream fixes.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When fork() fails we cannot use perf_counter_exit_task() since that
assumes to operate on current. Write a new helper that cleans up
unused/clean contexts.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove the local_irq_save() etc.. in routines that are smp function
calls, or have IRQs disabled by other means.
Then change the COMM, MMAP, and swcounter context iteration to
current->perf_counter_ctxp and RCU, since it really doesn't matter
which context they iterate, they're all folded.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Commit a63eaf34ae ("perf_counter: Dynamically allocate tasks'
perf_counter_context struct") broke COMM and MMAP notification for
cpu wide counters by dropping out early if there was no task context,
thereby also not iterating the cpu context.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This fixes a nasty crash and highlights a bug that we were
freeing failed-fork() counters incorrectly.
(the fix for that will come separately)
[ Impact: fix crashes/lockups with inherited counters ]
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter noticed that we are sometimes reading cpuctx->task_ctx with
interrupts enabled.
Noticed-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra pointed out that under some circumstances, we can take
the mutex in a context or a counter and then swap that context or
counter to another task, potentially leading to lock order inversions
or the mutexes not protecting what they are supposed to protect.
This fixes the problem by making sure that we never take a mutex in a
context or counter which could get swapped to another task. Most of
the cases where we take a mutex is on a top-level counter or context,
i.e. a counter which has an fd associated with it or a context that
contains such a counter. This adds WARN_ON_ONCE statements to verify
that.
The two cases where we need to take the mutex on a context that is a
clone of another are in perf_counter_exit_task and
perf_counter_init_task. The perf_counter_exit_task case is solved by
uncloning the context before starting to remove the counters from it.
The perf_counter_init_task is a little trickier; we temporarily
disable context swapping for the parent (forking) task by setting its
ctx->parent_gen to the all-1s value after locking the context, if it
is a cloned context, and restore the ctx->parent_gen value at the end
if the context didn't get uncloned in the meantime.
This also moves the increment of the context generation count to be
within the same critical section, protected by the context mutex, that
adds the new counter to the context. That way, taking the mutex is
sufficient to ensure that both the counter list and the generation
count are stable.
[ Impact: fix hangs, races with inherited and PID counters ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18975.31580.520676.619896@drongo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Needed in followon patch.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Commit 564c2b21 ("perf_counter: Optimize context switch between
identical inherited contexts") introduced a race where it is possible
that a counter being attached to a task could get attached to the
wrong task, if the task is one that has inherited its context from
another task via fork. This happens because the optimized context
switch could switch the context to another task after find_get_context
has read task->perf_counter_ctxp. In fact, it's possible that the
context could then get freed, if the other task then exits.
This fixes the problem by protecting both the context switch and the
critical code in find_get_context with spinlocks. The context switch
locks the cxt->lock of both the outgoing and incoming contexts before
swapping them. That means that once code such as find_get_context
has obtained the spinlock for the context associated with a task,
the context can't get swapped to another task. However, the context
may have been swapped in the interval between reading
task->perf_counter_ctxp and getting the lock, so it is necessary to
check and retry.
To make sure that none of the contexts being looked at in
find_get_context can get freed, this changes the context freeing code
to use RCU. Thus an rcu_read_lock() is sufficient to ensure that no
contexts can get freed. This part of the patch is lifted from a patch
posted by Peter Zijlstra.
This also adds a check to make sure that we can't add a counter to a
task that is exiting.
There is also a race between perf_counter_exit_task and
find_get_context; this solves the race by moving the get_ctx that
was in perf_counter_alloc into the locked region in find_get_context,
so that once find_get_context has got the context for a task, it
won't get freed even if the task calls perf_counter_exit_task. It
doesn't matter if new top-level (non-inherited) counters get attached
to the context after perf_counter_exit_task has detached the context
from the task. They will just stay there and never get scheduled in
until the counters' fds get closed, and then perf_release will remove
them from the context and eventually free the context.
With this, we are now doing the unclone in find_get_context rather
than when a counter was added to or removed from a context (actually,
we were missing the unclone_ctx() call when adding a counter to a
context). We don't need to unclone when removing a counter from a
context because we have no way to remove a counter from a cloned
context.
This also takes out the smp_wmb() in find_get_context, which Peter
Zijlstra pointed out was unnecessary because the cmpxchg implies a
full barrier anyway.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <18974.33033.667187.273886@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch adds __print_symbolic which is similar to __print_flags but
works for an enumeration type instead. That is, there is only a one to one
mapping between the values and the symbols. When a match is made, then
it is printed, otherwise the hex value is outputed.
[ Impact: add interface for showing symbol names in events ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Developers have been asking for the ability in the ftrace event tracer
to display names of bits in a flags variable.
Instead of printing out c2, it would be easier to read FOO|BAR|GOO,
assuming that FOO is bit 1, BAR is bit 6 and GOO is bit 7.
Some examples where this would be useful are the state flags in a context
switch, kmalloc flags, and even permision flags in accessing files.
[
v2 changes include:
Frederic Weisbecker's idea of using a mask instead of bits,
thus we can output GFP_KERNEL instead of GPF_WAIT|GFP_IO|GFP_FS.
Li Zefan's idea of allowing the caller of __print_flags to add their
own delimiter (or no delimiter) where we can get for file permissions
rwx instead of r|w|x.
]
[
v3 changes:
Christoph Hellwig's idea of using an array instead of va_args.
]
[ Impact: better displaying of flags in trace output ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
This shouldnt matter normally (and i have not seen any
misbehavior), because active counters always have a
proper ->oncpu value - but nevertheless initialize the
field properly to -1.
[ Impact: cleanup ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Always use ftrace_event_enable_disable() to enable/disable an event
so that we can factorize out the event toggling code.
[ Impact: factorize and cleanup event tracing code ]
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4A14FDFE.2080402@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
I found that there is nothing to protect event_hash in
ftrace_find_event(). Rcu protects the event hashlist
but not the event itself while we use it after its extraction
through ftrace_find_event().
This lack of a proper locking in this spot opens a race
window between any event dereferencing and module removal.
Eg:
--Task A--
print_trace_line(trace) {
event = find_ftrace_event(trace)
--Task B--
trace_module_remove_events(mod) {
list_trace_events_module(ev, mod) {
unregister_ftrace_event(ev->event) {
hlist_del(ev->event->node)
list_del(....)
}
}
}
|--> module removed, the event has been dropped
--Task A--
event->print(trace); // Dereferencing freed memory
If the event retrieved belongs to a module and this module
is concurrently removed, we may end up dereferencing a data
from a freed module.
RCU could solve this, but it would add latency to the kernel and
forbid tracers output callbacks to call any sleepable code.
So this fix converts 'trace_event_mutex' to a read/write semaphore,
and adds trace_event_read_lock() to protect ftrace_find_event().
[ Impact: fix possible freed memory dereference in ftrace ]
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <4A114806.7090302@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Introduce a generic per counter interrupt throttle.
This uses the perf_counter_overflow() quick disable to throttle a specific
counter when its going too fast when a pmu->unthrottle() method is provided
which can undo the quick disable.
Power needs to implement both the quick disable and the unthrottle method.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525153931.703093461@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ingo noticed that cpu counters had 0 context switches, even though
there was plenty scheduling on the cpu.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525124600.419025548@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fail fork() when we fail inheritance for some reason (-ENOMEM most likely).
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525124600.324656474@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Paul noted that the new ptcrl() didn't work on child counters.
Reported-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525124600.203151469@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Initialize a task's perfcounters (inherit from parent, etc.) after
the child task's scheduler fields have been initialized already.
[ Impact: cleanup ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The problem occurs when async_synchronize_full_domain() is called when
the async_pending list is not empty. This will cause lowest_running()
to return the cookie of the first entry on the async_pending list, which
might be nothing at all to do with the domain being asked for and thus
cause the domain synchronization to wait for an unrelated domain. This
can cause a deadlock if domain synchronization is used from one domain
to wait for another.
Fix by running over the async_pending list to see if any pending items
actually belong to our domain (and return their cookies if they do).
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We shouldn't hold dpm_list_mtx while executing
[disable|enable]_nonboot_cpus(), because theoretically this may lead
to a deadlock as shown by the following example (provided by Johannes
Berg):
CPU 3 CPU 2 CPU 1
suspend/hibernate
something:
rtnl_lock() device_pm_lock()
-> mutex_lock(&dpm_list_mtx)
mutex_lock(&dpm_list_mtx)
linkwatch_work
-> rtnl_lock()
disable_nonboot_cpus()
-> flush CPU 3 workqueue
Fortunately, device drivers are supposed to stop any activities that
might lead to the registration of new device objects way before
disable_nonboot_cpus() is called, so it shouldn't be necessary to
hold dpm_list_mtx over the entire late part of device suspend and
early part of device resume.
Thus, during the late suspend and the early resume of devices acquire
dpm_list_mtx only when dpm_list is going to be traversed and release
it right after that.
This patch is reported to fix the regressions tracked as
http://bugzilla.kernel.org/show_bug.cgi?id=13245.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Miles Lane <miles.lane@gmail.com>
Tested-by: Ming Lei <tom.leiming@gmail.com>
In a default 'perf top' run the tool will create a counter for
each online CPU. With enough CPUs this will eventually exhaust
the default limit.
So scale it up with the number of online CPUs.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
now that pctrl() no longer disables other people's counters,
remove the PMU cache code that deals with that.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163013.032998331@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Instead of en/dis-abling all counters acting on a particular
task, en/dis- able all counters we created.
[ v2: fix crash on first counter enable ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163012.916937244@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Use perf_counter_remove_from_context() to remove counters from
the context.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163012.796275849@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ensure we're consistent with the context locks.
context->mutex
context->lock
list_{add,del}_counter();
so that either lock is sufficient to stabilize the context.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163012.618790733@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
s/counter->mutex/counter->child_mutex/ and make sure its only
used to protect child_list.
The usage in __perf_counter_exit_task() doesn't appear to be
problematic since ctx->mutex also covers anything related to fd
tear-down.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163012.533186528@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We call perf_adjust_freq() from perf_counter_task_tick() which
is is called under the rq->lock causing lock recursion.
However, it's no longer required to be called under the
rq->lock, so remove it from under it.
Also, fix up some related comments.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090523163012.476197912@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Presently non-legacy IRQs have their irq_desc allocated with
kzalloc_node(). This assumes that all callers of irq_to_desc_node_alloc()
will be sufficiently late in the boot process that kmalloc is available.
While porting sparseirq support to sh this blew up immediately, as at the
time that we register the CPU's interrupt vector map only bootmem is
available. Check slab_is_available() to work out which path to use.
[ Impact: fix SH early boot crash with sparseirq enabled ]
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
LKML-Reference: <20090522014008.GA2806@linux-sh.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When monitoring a process and its descendants with a set of inherited
counters, we can often get the situation in a context switch where
both the old (outgoing) and new (incoming) process have the same set
of counters, and their values are ultimately going to be added together.
In that situation it doesn't matter which set of counters are used to
count the activity for the new process, so there is really no need to
go through the process of reading the hardware counters and updating
the old task's counters and then setting up the PMU for the new task.
This optimizes the context switch in this situation. Instead of
scheduling out the perf_counter_context for the old task and
scheduling in the new context, we simply transfer the old context
to the new task and keep using it without interruption. The new
context gets transferred to the old task. This means that both
tasks still have a valid perf_counter_context, so no special case
is introduced when the old task gets scheduled in again, either on
this CPU or another CPU.
The equivalence of contexts is detected by keeping a pointer in
each cloned context pointing to the context it was cloned from.
To cope with the situation where a context is changed by adding
or removing counters after it has been cloned, we also keep a
generation number on each context which is incremented every time
a context is changed. When a context is cloned we take a copy
of the parent's generation number, and two cloned contexts are
equivalent only if they have the same parent and the same
generation number. In order that the parent context pointer
remains valid (and is not reused), we increment the parent
context's reference count for each context cloned from it.
Since we don't have individual fds for the counters in a cloned
context, the only thing that can make two clones of a given parent
different after they have been cloned is enabling or disabling all
counters with prctl. To account for this, we keep a count of the
number of enabled counters in each context. Two contexts must have
the same number of enabled counters to be considered equivalent.
Here are some measurements of the context switch time as measured with
the lat_ctx benchmark from lmbench, comparing the times obtained with
and without this patch series:
-----Unmodified----- With this patch series
Counters: none 2 HW 4H+4S none 2 HW 4H+4S
2 processes:
Average 3.44 6.45 11.24 3.12 3.39 3.60
St dev 0.04 0.04 0.13 0.05 0.17 0.19
8 processes:
Average 6.45 8.79 14.00 5.57 6.23 7.57
St dev 1.27 1.04 0.88 1.42 1.46 1.42
32 processes:
Average 5.56 8.43 13.78 5.28 5.55 7.15
St dev 0.41 0.47 0.53 0.54 0.57 0.81
The numbers are the mean and standard deviation of 20 runs of
lat_ctx. The "none" columns are lat_ctx run directly without any
counters. The "2 HW" columns are with lat_ctx run under perfstat,
counting cycles and instructions. The "4H+4S" columns are lat_ctx run
under perfstat with 4 hardware counters and 4 software counters
(cycles, instructions, cache references, cache misses, task
clock, context switch, cpu migrations, and page faults).
[ Impact: performance optimization of counter context-switches ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <18966.10666.517218.332164@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This replaces the struct perf_counter_context in the task_struct with
a pointer to a dynamically allocated perf_counter_context struct. The
main reason for doing is this is to allow us to transfer a
perf_counter_context from one task to another when we do lazy PMU
switching in a later patch.
This has a few side-benefits: the task_struct becomes a little smaller,
we save some memory because only tasks that have perf_counters attached
get a perf_counter_context allocated for them, and we can remove the
inclusion of <linux/perf_counter.h> in sched.h, meaning that we don't
end up recompiling nearly everything whenever perf_counter.h changes.
The perf_counter_context structures are reference-counted and freed
when the last reference is dropped. A context can have references
from its task and the counters on its task. Counters can outlive the
task so it is possible that a context will be freed well after its
task has exited.
Contexts are allocated on fork if the parent had a context, or
otherwise the first time that a per-task counter is created on a task.
In the latter case, we set the context pointer in the task struct
locklessly using an atomic compare-and-exchange operation in case we
raced with some other task in creating a context for the subject task.
This also removes the task pointer from the perf_counter struct. The
task pointer was not used anywhere and would make it harder to move a
context from one task to another. Anything that needed to know which
task a counter was attached to was already using counter->ctx->task.
The __perf_counter_init_context function moves up in perf_counter.c
so that it can be called from find_get_context, and now initializes
the refcount, but is otherwise unchanged.
We were potentially calling list_del_counter twice: once from
__perf_counter_exit_task when the task exits and once from
__perf_counter_remove_from_context when the counter's fd gets closed.
This adds a check in list_del_counter so it doesn't do anything if
the counter has already been removed from the lists.
Since perf_counter_task_sched_in doesn't do anything if the task doesn't
have a context, and leaves cpuctx->task_ctx = NULL, this adds code to
__perf_install_in_context to set cpuctx->task_ctx if necessary, i.e. in
the case where the current task adds the first counter to itself and
thus creates a context for itself.
This also adds similar code to __perf_counter_enable to handle a
similar situation which can arise when the counters have been disabled
using prctl; that also leaves cpuctx->task_ctx = NULL.
[ Impact: refactor counter context management to prepare for new feature ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <18966.10075.781053.231153@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
register_module_notifier() returns zero in the success case.
So fix the inverted fail case check in trace events modules
handler.
[ Impact: fix spurious warning on ftrace initialization]
Reported-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Avoid a function call for !group counters by directly calling the counter
function.
[ Impact: micro-optimize the code ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090520102553.511933670@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently we call hrtimer_cancel() unconditionally on disable of time based
software counters. Avoid when possible.
[ Impact: micro-optimize the code ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090520102553.388185031@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
For the dynamic irq_period code, log whenever we change the period so that
analyzing code can normalize the event flow.
[ Impact: add new feature to allow more precise profiling ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090520102553.298769743@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Instead of disabling RR scheduling of the counters, use a different list
that does not get rotated to iterate the counters on inheritance.
[ Impact: cleanup, optimization ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090520102553.237504544@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
If the waiter has been requeued to the outer PI futex and is
interrupted by a signal and the thread handles the signal then
ERESTART_RESTARTBLOCK is changed to EINTR and the restart block is
discarded. That way we return an unexcpected EINTR to user space
instead of ending up in futex_lock_pi_restart.
But we do not need to restart the syscall because we know that the
condition has changed since we have been requeued. If we would simply
restart the syscall then we would drop out via the comparison of the
user space value with EWOULDBLOCK.
The user space side needs to handle EWOULDBLOCK anyway as the
enqueueing on the inner futex can race with a requeue/wake. So we can
simply return EWOULDBLOCK to user space which also signals that we did
not take the outer futex and let user space handle it in the same way
it has to handle the requeue/wake race.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The futex_wait_requeue_pi op should restart unconditionally like
futex_lock_pi. The user of that function e.g. pthread_cond_wait can
not be interrupted so we do not care about the SA_RESTART flag of the
signal. Clean up the FIXMEs.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Merge reason: this branch was on an pre -rc1 base, merge it up to -rc6+
to get the latest upstream fixes.
Conflicts:
kernel/futex.c
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Context rotation should not occur when we are in the middle of
walking the counter list when inheriting counters ...
[ Impact: fix occasionally incorrect perf stat results ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fix counter lifetime bugs which explain the crashes reported by
Marcelo Tosatti and Arnaldo Carvalho de Melo.
The new rule is: flushing + freeing is only done for a task's
own counters, never for other tasks.
[ Impact: fix crashes/lockups with inherited counters ]
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reported-by: Marcelo Tosatti <mtosatti@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The futex code installs a read only mapping via get_user_pages_fast()
even if the futex op function has to modify user space data. The
eventual fault was fixed up by futex_handle_fault() which walked the
VMA with mmap_sem held.
After the cleanup patches which removed the mmap_sem dependency of the
futex code commit 4dc5b7a36a49eff97050894cf1b3a9a02523717 (futex:
clean up fault logic) removed the private VMA walk logic from the
futex code. This change results in a stale RO mapping which is not
fixed up.
Instead of reintroducing the previous fault logic we set up the
mapping in get_user_pages_fast() read/write for all operations which
modify user space data. Also handle private futexes in the same way
and make the current unconditional access_ok(VERIFY_WRITE) depend on
the futex op.
Reported-by: Andreas Schwab <schwab@linux-m68k.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
CC: stable@kernel.org
debugfs directory entries for devices are not removed on some
of the failure pathes in do_blk_trace_setup().
One way to reproduce is to start blktrace on multiple devices
with insufficient Vmalloc space: Devices will fail with
a message like this:
BLKTRACESETUP(2) /dev/sdu failed: 5/Input/output error
If so, the respective entries in debugfs
(e.g. /sys/kernel/debug/block/sdu) will remain and subsequent
attempts to start blktrace on the respective devices will not
succeed due to existing directories.
[ Impact: fix /debug/tracing file cleanup corner case ]
Signed-off-by: Stefan Raspl <stefan.raspl@linux.vnet.ibm.com>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: schwidefsky@de.ibm.com
Cc: heiko.carstens@de.ibm.com
LKML-Reference: <4A1266CC.5040801@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Properly document the variable-size structure tricks we are doing
wrt. struct sched_group and sched_domain, and use the field[0] GCC
extension instead of defining a vla array.
Dont use unions for this, as pointed out by Linus.
[ Impact: cleanup, un-confuse Sparse and LLVM ]
Reported-by: Jeff Garzik <jeff@garzik.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
LKML-Reference: <alpine.LFD.2.01.0905180850110.3301@localhost.localdomain>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'sched-fixes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: Fix fallback sched_clock()'s offset when using jiffies
* 'core-fixes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
lockdep: increase MAX_LOCKDEP_ENTRIES and MAX_LOCKDEP_CHAINS
* 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
tracing: Append prompt in /debug/tracing/README file
x86/function-graph: fix constraint for recording old return value
return zero should be correct, so fix it.
[ Impact: eliminate incorrect syslog message ]
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: rostedt@goodmis.org
LKML-Reference: <1242545498-7285-1-git-send-email-tom.leiming@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Flushing counters in __exit_signal() with irqs disabled is not
a good idea as perf_counter_exit_task() acquires mutexes. So
flush it before acquiring the tasklist lock.
(Note, we still need a fix for when the PID has been unhashed.)
[ Impact: fix crash with inherited counters ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Clean up code that open-coded the list_{add,del}_counter() code in
__perf_counter_exit_task() which consequently diverged. This could
lead to software counter crashes.
Also, fold the ctx->nr_counter inc/dec into those functions and clean
up some of the related code.
[ Impact: fix potential sw counter crash, cleanup ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ian Campbell noticed that since "Eliminate thousands of warnings with
gcc 3.2 build" (commit 57adc4d2db) all
WARN_ON()'s currently appear to come from warn_slowpath_null(), eg:
WARNING: at kernel/softirq.c:143 warn_slowpath_null+0x1c/0x20()
because now that warn_slowpath_null() is in the call path, the
__builtin_return_address(0) returns that, rather than the place that
caused the warning.
Fix this by splitting up the warn_slowpath_null/fmt cases differently,
using a common helper function, and getting the return address in the
right place. This also happens to avoid the unnecessary stack usage for
the non-stdargs case, and just generally cleans things up.
Make the function name printout use %pS while at it.
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Check the return value of sysdev_suspend(). I think this was a typo.
Without this change, the following "if" check is always false.
I also changed the error message so it's distinguishable from the
similar message a few lines above.
Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb:
kgdb: gdb documentation fix
kgdb,i386: use address that SP register points to in the exception frame
sysrq, intel_fb: fix sysrq g collision
At present the values we put in overflow events for the misc
flags indicating processor mode and the instruction pointer are
obtained using the standard user_mode() and
instruction_pointer() functions. Those functions tell you where
the performance monitor interrupt was taken, which might not be
exactly where the counter overflow occurred, for example
because interrupts were disabled at the point where the
overflow occurred, or because the processor had many
instructions in flight and chose to complete some more
instructions beyond the one that caused the counter overflow.
Some architectures (e.g. powerpc) can supply more precise
information about where the counter overflow occurred and the
processor mode at that point. This introduces new functions,
perf_misc_flags() and perf_instruction_pointer(), which arch
code can override to provide more precise information if
available. They have default implementations which are
identical to the existing code.
This also adds a new misc flag value,
PERF_EVENT_MISC_HYPERVISOR, for the case where a counter
overflow occurred in the hypervisor. We encode the processor
mode in the 2 bits previously used to indicate user or kernel
mode; the values for user and kernel mode are unchanged and
hypervisor mode is indicated by both bits being set.
[ Impact: generalize perfcounter core facilities ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <18956.1272.818511.561835@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
avenrun is an rough estimate so we don't have to worry about
consistency of the three avenrun values. Remove the xtime lock
dependency and provide a function to scale the values. Cleanup the
users.
[ Impact: cleanup ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Dimitri Sivanich noticed that xtime_lock is held write locked across
calc_load() which iterates over all online CPUs. That can cause long
latencies for xtime_lock readers on large SMP systems.
The load average calculation is an rough estimate anyway so there is
no real need to protect the readers vs. the update. It's not a problem
when the avenrun array is updated while a reader copies the values.
Instead of iterating over all online CPUs let the scheduler_tick code
update the number of active tasks shortly before the avenrun update
happens. The avenrun update itself is handled by the CPU which calls
do_timer().
[ Impact: reduce xtime_lock write locked section ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Instead of specifying the irq_period for a counter, provide a target interrupt
frequency and dynamically adapt the irq_period to match this frequency.
[ Impact: new perf-counter attribute/feature ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090515132018.646195868@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Instead of a per-process mlock gift for perf-counters, use a
per-user gift so that there is less of a DoS potential.
[ Impact: allow less worst-case unprivileged memory consumption ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090515132018.496182835@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Now that ACPI idle doesn't use it anymore, remove the exports.
[ Impact: remove dead code/data ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090515132018.429826617@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The waitqueue which is used in struct futex_q is a leftover from the
futexfd implementation. There is no need to use a waitqueue at all, as
the waiting task is the only user of it. The waitqueue just adds
additional locking and a loop in the wake up path which both can be
avoided.
We have already a task reference in struct futex_q which is used for
PI futexes. Use it for normal futexes as well and just wake up the
task directly.
The logic of signalling the futex wakeup via setting q->lock_ptr to
NULL is kept with the difference that we set it NULL before doing the
wakeup. This opens an exit race window vs. a non futex wake up of the
to be woken up task, which we prevent with get_task_struct /
put_task_struct on the waiter.
[ Impact: simplification ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Commit 79e539453b introduced a
regression where you cannot use sysrq 'g' to enter kgdb. The solution
is to move the intel fb sysrq over to V for video instead of G for
graphics. The SMP VOYAGER code to register for the sysrq-v is not
anywhere to be found in the mainline kernel, so the comments in the
code were cleaned up as well.
This patch also cleans up the sysrq definitions for kgdb to make it
generic for the kernel debugger, such that the sysrq 'g' can be used
in the future to enter a gdbstub or another kernel debugger.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
This reverts commit fafd688e4c.
Work is progressing to switch away from pdflush as the process backing
for flushing out dirty data. So it seems pointless to add more knobs
to control pdflush threads. The original author of the patch did not
have any specific use cases for adding the knobs, so we can easily
revert this before 2.6.30 to avoid having to maintain this API
forever.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
The current disable/enable mechanism is:
token = hw_perf_save_disable();
...
/* do bits */
...
hw_perf_restore(token);
This works well, provided that the use nests properly. Except we don't.
x86 NMI/INT throttling has non-nested use of this, breaking things. Therefore
provide a reference counter disable/enable interface, where the first disable
disables the hardware, and the last enable enables the hardware again.
[ Impact: refactor, simplify the PMU disable/enable logic ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The simple reservation test in perf_output_copy() failed to take
unsigned int overflow into account, fix this.
[ Impact: fix false positive warning with more than 4GB of profiling data ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
