openwrt/staging/blogic.git
14 years agoperf probe: Add kernel source path option
Chase Douglas [Mon, 14 Jun 2010 19:26:30 +0000 (15:26 -0400)]
perf probe: Add kernel source path option

The probe plugin requires access to the source code for some operations.  The
source code must be in the exact same location as specified by the DWARF tags,
but sometimes the location is an absolute path that cannot be replicated by a
normal user. This change adds the -s|--source option to allow the user to
specify the root of the kernel source tree.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
LKML-Reference: <1276543590-10486-1-git-send-email-chase.douglas@canonical.com>
Signed-off-by: Chase Douglas <chase.douglas@canonical.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf tools: .gitignore += config.make config.make.autogen
Kirill Smelkov [Mon, 14 Jun 2010 12:00:47 +0000 (16:00 +0400)]
perf tools: .gitignore += config.make config.make.autogen

These are local-configuration files and should be ignored.

LKML-Reference: <1276516847-25817-1-git-send-email-kirr@landau.phys.spbu.ru>
Signed-off-by: Kirill Smelkov <kirr@landau.phys.spbu.ru>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf record: Add option to avoid updating buildid cache
Stephane Eranian [Thu, 17 Jun 2010 09:39:01 +0000 (11:39 +0200)]
perf record: Add option to avoid updating buildid cache

There are situations where there is enough information in the perf.data
to process the samples. Updating the buildid cache may add unecessary
overhead in terms of disk space and time (copying large elf images).

A persistent option to do this already exists via the perfconfig file,
simply do:

[buildid]
dir = /dev/null

This patch provides a way to suppress builid cache updates on a per-run
basis.  It addds a new option, -N, to perf record. Buildids are still
generated in the perf.data file.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4c19ef89.93ecd80a.40dc.fffff8e9@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf symbols: Function descriptor symbol lookup
Eric B Munson [Mon, 14 Jun 2010 13:56:33 +0000 (14:56 +0100)]
perf symbols: Function descriptor symbol lookup

Currently symbol resolution does not work for 64-bit programs on architectures
that use function descriptors such as ppc64.

The problem is that a symbol doesn't point to a text address, it points to a
data area that contains (amongst other things) a pointer to the text address.

We look for a section called ".opd" which is the function descriptor area. To
create the full symbol table, when we see a symbol in the function descriptor
section we load the first pointer and use that as the text address.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1276523793-15422-1-git-send-email-ebmunson@us.ibm.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf record: Avoid synthesizing mmap() for all processes in per-thread mode
Stephane Eranian [Wed, 16 Jun 2010 18:59:01 +0000 (20:59 +0200)]
perf record: Avoid synthesizing mmap() for all processes in per-thread mode

A bug was introduced by commit c45c6ea2e5c57960dc67e00294c2b78e9540c007.

Perf record was scanning /proc/PID to create synthetic PERF_RECOR_MMAP
entries even though it was running in per-thread mode. There was a bogus
check to select what mmaps to synthesize. We only need all processes in
system-wide mode.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <4c192107.4f1ee30a.4316.fffff98e@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf tools: Reorganize the Makefile feature tests
Arnaldo Carvalho de Melo [Wed, 9 Jun 2010 19:57:39 +0000 (16:57 -0300)]
perf tools: Reorganize the Makefile feature tests

Moving the tests to a separate file, feature-tests.mak and using a try-cc
function similar to the try-run in Kbuild.

This also makes the output more quiet as we can stop using the INTERMEDIATE
target to remove the .perf.dev.null file needed for some gcc versions where
/dev/null can't be used as the output file name.

As the tests get shorter by uninlining the source code used to test for
features, we can more properly use identation.

The feature tests itself can be made more clear and reused, like when trying to
see what is needed to have bfd_demangle.

We also get a bit closer to reusing scripts/Kbuild.include, reducing the
distance from the kernel build system.

Tests performed:

[root@emilia perf]# make -j9 O=/tmp/perf
PERF_VERSION = 0.0.2.PERF
    GEN /tmp/perf/common-cmds.h
    * new build flags or prefix
    GEN perf-archive
    CC /tmp/perf/builtin-annotate.o
    CC /tmp/perf/bench/sched-messaging.o
    CC /tmp/perf/builtin-diff.o
<SNIP>
    CC /tmp/perf/scripts/python/Perf-Trace-Util/Context.o
    CC /tmp/perf/perf.o
    CC /tmp/perf/builtin-help.o
    AR /tmp/perf/libperf.a
    LINK /tmp/perf/perf
[root@emilia perf]#

If we uninstall, for instance newt-devel we get:

[root@emilia perf]# rpm -e newt-devel
[root@emilia perf]# make -j9 O=/tmp/perf
Makefile:564: newt not found, disables TUI support. Please install newt-devel or libnewt-dev
    * new build flags or prefix
    GEN perf-archive
    CC /tmp/perf/perf.o
    CC /tmp/perf/builtin-annotate.o
<SNIP>
    AR /tmp/perf/libperf.a
    LINK /tmp/perf/perf
[root@emilia perf]#

And then binutils-devel:

[root@emilia perf]# make -j9 O=/tmp/perf
Makefile:564: newt not found, disables TUI support. Please install newt-devel or libnewt-dev
Makefile:632: No bfd.h/libbfd found, install binutils-dev[el]/zlib-static to gain symbol demangling
    * new build flags or prefix
    GEN perf-archive
    CC /tmp/perf/perf.o
<SNIP>
    AR /tmp/perf/libperf.a
    LINK /tmp/perf/perf
[root@emilia perf]#

And then strictly required devel packages:

[root@emilia perf]# rpm -e elfutils-libelf-devel elfutils-devel
[root@emilia perf]# make -j9 O=/tmp/perf
Makefile:509: No libdw.h found or old libdw.h found or elfutils is older than 0.138, disables dwarf support. Please install new elfutils-devel/libdw-dev
Makefile:542: *** No libelf.h/libelf found, please install libelf-dev/elfutils-libelf-devel.  Stop.
[root@emilia perf]#

After installing everything back on:

[root@emilia perf]# yum install elfutils-devel binutils-devel newt-devel
<SNIP>
Installed:
  binutils-devel.x86_64 0:2.20.51.0.2-5.11.el6
  elfutils-devel.x86_64 0:0.147-1.el6
  elfutils-libelf-devel.x86_64 0:0.147-1.el6
  newt-devel.x86_64 0:0.52.11-1.el6

Complete!
[root@emilia perf]# make -j9
PERF_VERSION = 0.0.2.PERF
    GEN common-cmds.h
    * new build flags or prefix
    GEN perf-archive
    CC builtin-annotate.o
<SNIP>
    AR libperf.a
    LINK perf
[root@emilia perf]# make -j9
[root@emilia perf]#

Thanks to Sam for pointing me to try-run.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoMerge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
Ingo Molnar [Wed, 9 Jun 2010 16:55:20 +0000 (18:55 +0200)]
Merge branch 'perf/core' of git://git./linux/kernel/git/frederic/random-tracing into perf/core

14 years agox86: Unify save_stack_address() and save_stack_address_nosched()
Oleg Nesterov [Thu, 3 Jun 2010 19:32:43 +0000 (21:32 +0200)]
x86: Unify save_stack_address() and save_stack_address_nosched()

Cleanup. Factor the common code in save_stack_address() and
save_stack_address_nosched().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
LKML-Reference: <20100603193243.GA31534@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
14 years agox86: Make save_stack_address() !CONFIG_FRAME_POINTER friendly
Oleg Nesterov [Thu, 3 Jun 2010 19:32:39 +0000 (21:32 +0200)]
x86: Make save_stack_address() !CONFIG_FRAME_POINTER friendly

If CONFIG_FRAME_POINTER=n, print_context_stack() shouldn't neglect the
non-reliable addresses on stack, this is all we have if dump_trace(bp)
is called with the wrong or zero bp.

For example, /proc/pid/stack doesn't work if CONFIG_FRAME_POINTER=n.

This patch obviously has no effect if CONFIG_FRAME_POINTER=y, otherwise
it reverts 1650743c "x86: don't save unreliable stack trace entries".

Also, remove the unnecessary type-cast.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <20100603193239.GA31530@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
14 years agotracing: Remove kmemtrace ftrace plugin
Li Zefan [Wed, 26 May 2010 09:22:17 +0000 (17:22 +0800)]
tracing: Remove kmemtrace ftrace plugin

We have been resisting new ftrace plugins and removing existing
ones, and kmemtrace has been superseded by kmem trace events
and perf-kmem, so we remove it.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
[ remove kmemtrace from the makefile, handle slob too ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
14 years agoperf: Fix build breakage for architecutes without atomic64_t
Peter Zijlstra [Wed, 9 Jun 2010 09:57:23 +0000 (11:57 +0200)]
perf: Fix build breakage for architecutes without atomic64_t

The local64.h include dependency was not dependent on PERF_EVENT=y,
which meant that arch's without atomic64_t support ended up including
it and failed to build.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>

14 years agoperf: Convert perf_event to local_t
Peter Zijlstra [Fri, 21 May 2010 12:43:08 +0000 (14:43 +0200)]
perf: Convert perf_event to local_t

Since now all modification to event->count (and ->prev_count
and ->period_left) are local to a cpu, change then to local64_t so we
avoid the LOCK'ed ops.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Add perf_event::child_count
Peter Zijlstra [Fri, 21 May 2010 12:27:58 +0000 (14:27 +0200)]
perf: Add perf_event::child_count

Only child counters adding back their values into the parent counter
are responsible for cross-cpu updates to event->count.

So if we pull that out into a new child_count variable, we get an
event->count that is only modified locally.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Add perf_event_count()
Peter Zijlstra [Fri, 21 May 2010 12:43:12 +0000 (14:43 +0200)]
perf: Add perf_event_count()

Create a helper function for those sites that want to read the event count.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoarch: Implement local64_t
Peter Zijlstra [Fri, 21 May 2010 12:05:13 +0000 (14:05 +0200)]
arch: Implement local64_t

On 64bit, local_t is of size long, and thus we make local64_t an alias.
On 32bit, we fall back to atomic64_t. (architecture can provide optimized
32-bit version)

(This new facility is to be used by perf events optimizations.)

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: linux-arch@vger.kernel.org
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Simplify the ring-buffer logic: make perf_buffer_alloc() do everything needed
Peter Zijlstra [Fri, 28 May 2010 17:41:35 +0000 (19:41 +0200)]
perf: Simplify the ring-buffer logic: make perf_buffer_alloc() do everything needed

Currently there are perf_buffer_alloc() + perf_buffer_init() + some
separate bits, fold it all into a single perf_buffer_alloc() and only
leave the attachment to the event separate.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Rename perf_mmap_data to perf_buffer
Peter Zijlstra [Fri, 28 May 2010 17:33:23 +0000 (19:33 +0200)]
perf: Rename perf_mmap_data to perf_buffer

Rename to clarify code.

s/perf_mmap_data/perf_buffer/g and selective s/data/buffer/g

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf, x86: Make a second write to performance counter if needed
Cyrill Gorcunov [Wed, 2 Jun 2010 21:23:04 +0000 (01:23 +0400)]
perf, x86: Make a second write to performance counter if needed

On Netburst PMU we need a second write to a performance counter
due to cpu erratum.

A simple flag test instead of alternative instructions was choosen
because wrmsrl is already a macro and if virtualization is turned
on will need an additional wrapper call which is more expencise.

nb: we should propably switch to jump-labels as only this facility
reach the mainline.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20100602212304.GC5264@lenovo>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Cleanup {start,commit,cancel}_txn details
Peter Zijlstra [Tue, 25 May 2010 15:49:05 +0000 (17:49 +0200)]
perf: Cleanup {start,commit,cancel}_txn details

Clarify some of the transactional group scheduling API details
and change it so that a successfull ->commit_txn also closes
the transaction.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1274803086.5882.1752.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf: Add non-exec mmap() tracking
Eric B Munson [Tue, 18 May 2010 14:30:49 +0000 (15:30 +0100)]
perf: Add non-exec mmap() tracking

Add the capacility to track data mmap()s. This can be used together
with PERF_SAMPLE_ADDR for data profiling.

Signed-off-by: Anton Blanchard <anton@samba.org>
[Updated code for stable perf ABI]
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1274193049-25997-1-git-send-email-ebmunson@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf, trace: Remove superfluous rcu_read_lock()
Peter Zijlstra [Fri, 21 May 2010 13:13:59 +0000 (15:13 +0200)]
perf, trace: Remove superfluous rcu_read_lock()

__DO_TRACE() already calls the callbacks under rcu_read_lock_sched(),
which is sufficient for our needs, avoid doing it again.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf, trace: Inline perf_swevent_put_recursion_context()
Peter Zijlstra [Fri, 21 May 2010 13:11:34 +0000 (15:11 +0200)]
perf, trace: Inline perf_swevent_put_recursion_context()

Inline perf_swevent_put_recursion_context into perf_tp_event(), this
shrinks the per trace template code footprint and saves a function
call.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agotracing: Remove boot tracer
Américo Wang [Wed, 26 May 2010 10:57:53 +0000 (18:57 +0800)]
tracing: Remove boot tracer

The boot tracer is useless. It simply logs the initcalls
but in fact these initcalls are also logged through printk
while using the initcall_debug kernel parameter.

Nobody seem to be using it so far. Then just remove it.

Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Chase Douglas <chase.douglas@canonical.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <20100526105753.GA5677@cr0.nay.redhat.com>
[ remove the hooks in main.c, and the headers ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
14 years agoperf: Drop the skip argument from perf_arch_fetch_regs_caller
Frederic Weisbecker [Thu, 20 May 2010 05:47:21 +0000 (07:47 +0200)]
perf: Drop the skip argument from perf_arch_fetch_regs_caller

Drop this argument now that we always want to rewind only to the
state of the first caller.
It means frame pointers are not necessary anymore to reliably get
the source of an event. But this also means we need this helper
to be a macro now, as an inline function is not an option since
we need to know when to provide a default implentation.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agox86: Unify dumpstack.h and stacktrace.h
Frederic Weisbecker [Wed, 19 May 2010 19:35:17 +0000 (21:35 +0200)]
x86: Unify dumpstack.h and stacktrace.h

arch/x86/include/asm/stacktrace.h and arch/x86/kernel/dumpstack.h
declare headers of objects that deal with the same topic.
Actually most of the files that include stacktrace.h also include
dumpstack.h

Although dumpstack.h seems more reserved for internals of stack
traces, those are quite often needed to define specialized stack
trace operations. And perf event arch headers are going to need
access to such low level operations anyway. So don't continue to
bother with dumpstack.h as it's not anymore about isolated deep
internals.

v2: fix struct stack_frame definition conflict in sysprof

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Soeren Sandmann <sandmann@daimi.au.dk>
14 years agoperf, x86: Small fix to cpuid10_edx
Livio Soares [Thu, 3 Jun 2010 19:00:31 +0000 (15:00 -0400)]
perf, x86: Small fix to cpuid10_edx

Fixes to 'cpuid10_edx' to comply with Intel documentation.
According to the Intel Manual, Volume 2A, Table 3-12, the cpuid for
architecture performance monitoring returns, in EDX, two pieces of
information:

  1) Number of fixed-function counters (5 bits, not 4)
  2) Width of fixed-function counters (8 bits)

Signed-off-by: Livio Soares <livio@eecg.toronto.edu>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoMerge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux...
Ingo Molnar [Tue, 8 Jun 2010 17:42:08 +0000 (19:42 +0200)]
Merge branch 'perf/core' of git://git./linux/kernel/git/acme/linux-2.6 into perf/core

14 years agoMerge branch 'tip/perf/core-3' of git://git.kernel.org/pub/scm/linux/kernel/git/roste...
Ingo Molnar [Tue, 8 Jun 2010 17:34:40 +0000 (19:34 +0200)]
Merge branch 'tip/perf/core-3' of git://git./linux/kernel/git/rostedt/linux-2.6-trace into perf/core

14 years agoMerge branch 'for-tip' of git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile...
Ingo Molnar [Tue, 8 Jun 2010 17:15:37 +0000 (19:15 +0200)]
Merge branch 'for-tip' of git://git./linux/kernel/git/rric/oprofile into perf/urgent

14 years agoperf: Fix signed comparison in perf_adjust_period()
Peter Zijlstra [Fri, 4 Jun 2010 13:18:01 +0000 (15:18 +0200)]
perf: Fix signed comparison in perf_adjust_period()

Frederic reported that frequency driven swevents didn't work properly
and even caused a division-by-zero error.

It turns out there are two bugs, the division-by-zero comes from a
failure to deal with that in perf_calculate_period().

The other was more interesting and turned out to be a wrong comparison
in perf_adjust_period(). The comparison was between an s64 and u64 and
got implicitly converted to an unsigned comparison. The problem is
that period_left is typically < 0, so it ended up being always true.

Cure this by making the local period variables s64.

Reported-by: Frederic Weisbecker <fweisbec@gmail.com>
Tested-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf tools: Allow building perf source tarballs on non-configured tree
Arnaldo Carvalho de Melo [Mon, 7 Jun 2010 10:44:25 +0000 (07:44 -0300)]
perf tools: Allow building perf source tarballs on non-configured tree

So that we don't require that the kernel be configured first, and as we
don't use KERNELRELEASE at all in the -src-pkg targets, we need o add a
new wildcard for targets ending in src-pkg:

On a make mrproper'ed kernel we get this without this patch:

[linux-2.6-tip]$ LANG= make perf-tarbz2-src-pkg
/bin/sh: include/config/kernel.release: No such file or directory
make: *** [include/config/kernel.release] Error 1
[acme@emilia linux-2.6-tip]$

Acked-by: Michal Marek <mmarek@suse.cz>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <20100604173552.GA875@ghostprotocols.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agopowerpc/oprofile: fix potential buffer overrun in op_model_cell.c
Denis Kirjanov [Tue, 1 Jun 2010 19:43:34 +0000 (15:43 -0400)]
powerpc/oprofile: fix potential buffer overrun in op_model_cell.c

Fix potential initial_lfsr buffer overrun.
Writing past the end of the buffer could happen when index == ENTRIES

Signed-off-by: Denis Kirjanov <dkirjanov@kernel.org>
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
14 years agoperf report: Implement --sort cpu
Arun Sharma [Fri, 4 Jun 2010 14:27:10 +0000 (11:27 -0300)]
perf report: Implement --sort cpu

In a shared multi-core environment, users want to analyze why their
program was slow. In particular, if the code ran slower only on certain
CPUs due to interference from other programs or kernel threads, the user
should be able to notice that.

Sample usage:

perf record -f -a -- sleep 3
perf report --sort cpu,comm

Workload:

program is running on 16 CPUs
Experiencing interference from an antagonist only on 4 CPUs.

  Samples: 106218177676 cycles

  Overhead  CPU          Command
  ........  ...  ...............

     6.25%  2            program
     6.24%  6            program
     6.24%  11           program
     6.24%  5            program
     6.24%  9            program
     6.24%  10           program
     6.23%  15           program
     6.23%  7            program
     6.23%  3            program
     6.23%  14           program
     6.22%  1            program
     6.20%  13           program
     3.17%  12           program
     3.15%  8            program
     3.14%  0            program
     3.13%  4            program
     3.11%  4         antagonist
     3.11%  0         antagonist
     3.10%  8         antagonist
     3.07%  12        antagonist

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <20100505181612.GA5091@sharma-home.net>
Signed-off-by: Arun Sharma <aruns@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf tools: Make event__preprocess_sample parse the sample
Arnaldo Carvalho de Melo [Fri, 4 Jun 2010 11:02:07 +0000 (08:02 -0300)]
perf tools: Make event__preprocess_sample parse the sample

Simplifying the tools that were using both in sequence and allowing
upcoming simplifications, such as Arun's patch to sort by cpus.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf annotate: Ask objdump to demangle symbols
Stephane Eranian [Thu, 3 Jun 2010 13:50:01 +0000 (15:50 +0200)]
perf annotate: Ask objdump to demangle symbols

Perf report is demangling symbols but not annotate.

The former uses internal demangling via libbdf or libiberty. The latter
executes objdump which by default does not demangle symbols.

This patch adds the -C option to the objdump cmdline to enable symbol
demangling.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4c07b323.2126e30a.6245.0e1e@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf buildid: add perfconfig option to specify buildid cache dir
Stephane Eranian [Tue, 1 Jun 2010 19:25:01 +0000 (21:25 +0200)]
perf buildid: add perfconfig option to specify buildid cache dir

This patch adds the ability to specify an alternate directory to store the
buildid cache (buildids, copy of binaries). By default, it is hardcoded to
$HOME/.debug. This directory contains immutable data. The layout of the
directory is such that no conflicts in filenames are possible. A modification
in a file, yields a different buildid and thus a different location in the
subdir hierarchy.

You may want to put the buildid cache elsewhere because of disk space
limitation or simply to share the cache between users. It is also useful for
remote collect vs. local analysis of profiles.

This patch adds a new config option to the perfconfig file.  Under the tag
'buildid', there is a dir option. For instance, if you have:

$ cat /etc/perfconfig
[buildid]
dir = /var/cache/perf-buildid

All buildids and binaries are be saved in the directory specified. The perf
record, buildid-list, buildid-cache, report, annotate, and archive commands
will it to pull information out.

The option can be set in the system-wide perfconfig file or in the
$HOME/.perfconfig file.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4c055fb7.df0ce30a.5f0d.ffffae52@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf tools: Make target to generate self contained source tarball
Arnaldo Carvalho de Melo [Mon, 31 May 2010 14:13:21 +0000 (11:13 -0300)]
perf tools: Make target to generate self contained source tarball

Useful for when people want to try some version of the perf tools and don't
wants to download the kernel tarball.

Here is a session using this new target:

  [root@emilia linux-2.6-tip]# make help | grep -i perf
    perf-tar-src-pkg    - Build perf-2.6.35-rc1.tar source tarball
    perf-targz-src-pkg  - Build perf-2.6.35-rc1.tar.gz source tarball
    perf-tarbz2-src-pkg - Build perf-2.6.35-rc1.tar.bz2 source tarball
  [root@emilia linux-2.6-tip]# make perf-tarbz2-src-pkg
    TAR
  [root@emilia linux-2.6-tip]# ls -la perf-2.6.35-rc1.tar.bz2
  -rw-r--r-- 1 root root 295731 May 31 11:18 perf-2.6.35-rc1.tar.bz2
  [root@emilia linux-2.6-tip]# tar xf perf-2.6.35-rc1.tar.bz2
  [root@emilia linux-2.6-tip]# cd perf-2.6.35-rc1
  [root@emilia perf-2.6.35-rc1]# ls
  arch  HEAD  include  lib  tools
  [root@emilia perf-2.6.35-rc1]# cd tools/perf
  [root@emilia perf]# make -j9 2>&1 | tail
      CC arch/x86/util/dwarf-regs.o
      CC util/probe-finder.o
      CC util/newt.o
      CC util/scripting-engines/trace-event-perl.o
      CC scripts/perl/Perf-Trace-Util/Context.o
      CC perf.o
      CC builtin-help.o
      AR libperf.a
      LINK perf
  rm .perf.dev.null
  [root@emilia perf]# ./perf record -a sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.262 MB perf.data (~11457 samples) ]
  [root@emilia perf]# ./perf report | head -12
  # Events: 6K cycles
  #
  # Overhead          Command       Shared Object  Symbol
  # ........  ...............  ..................  ......
  #
       4.73%             perf  [kernel.kallsyms]   [k] format_decode
       4.49%             perf  libc-2.12.so        [.] _IO_file_underflow_internal
       4.38%             init  [kernel.kallsyms]   [k] mwait_idle
       3.29%             perf  [kernel.kallsyms]   [k] vsnprintf
       2.38%             init  [kernel.kallsyms]   [k] sched_clock_local
       2.35%             init  [kernel.kallsyms]   [k] apic_timer_interrupt
       1.86%     sirq-timer/5  [kernel.kallsyms]   [k] find_busiest_group
  [root@emilia perf]#

Acked-by: Michal Marek <mmarek@suse.cz>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <20100528185357.GA28009@ghostprotocols.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf tools: Add the ability to specify list of cpus to monitor
Stephane Eranian [Fri, 28 May 2010 10:00:01 +0000 (12:00 +0200)]
perf tools: Add the ability to specify list of cpus to monitor

This patch adds a -C option to stat, record, top to designate a list of CPUs to
monitor. CPUs can be specified as a comma-separated list or ranges, no space
allowed.

Examples:
$ perf record -a -C0-1,4-7 sleep 1
$ perf top -C0-4
$ perf stat -a -C1,2,3,4 sleep 1

With perf record in per-thread mode with inherit mode on, samples are collected
only when the thread runs on the designated CPUs.

The -C option does not turn on system-wide mode automatically.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4bff9496.d345d80a.41fe.7b00@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf report: Make -D print sampled CPU
Stephane Eranian [Fri, 28 May 2010 10:08:01 +0000 (12:08 +0200)]
perf report: Make -D print sampled CPU

It is useful to know on which CPU a sample was captured on.
The information is captured with perf record -R but it was
not printed out by perf report -D. This patch adds this.

When -R is not used, cpu is set to -1to indicate that
the CPU is unknown (it is not captured).

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4bff964c.e88cd80a.3106.7d31@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoMerge branch 'perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/acme...
Ingo Molnar [Fri, 4 Jun 2010 14:55:19 +0000 (16:55 +0200)]
Merge branch 'perf/urgent' of git://git./linux/kernel/git/acme/linux-2.6 into perf/urgent

14 years agoperf symbols: Set the DSO long name when using symbol_conf.vmlinux_name
Arnaldo Carvalho de Melo [Thu, 3 Jun 2010 21:35:55 +0000 (18:35 -0300)]
perf symbols: Set the DSO long name when using symbol_conf.vmlinux_name

We need to set the long name to the name specified via, for instance,
'perf annotate --vmlinux /path/to/vmlinux', if not it will remain as
'[kernel.kallsyms]' and that will make annotate fail when passing this
as the vmlinux name in the call to objdump.

The way this is setup grew unwieldly and dso__load_vmlinux is the
function that should allocate space for the long name, with callers not
assuming that filenames should be allocated somehow by then (strdup,
dso__build_id_filename, etc).

For now this is the minimalistic patch, a proper fix for .36 will be
made.

Reported-by: Stephane Eranian <eranian@google.com>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <20100604003900.GD10469@ghostprotocols.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agotracing: Remove ftrace_preempt_disable/enable
Steven Rostedt [Thu, 3 Jun 2010 13:36:50 +0000 (09:36 -0400)]
tracing: Remove ftrace_preempt_disable/enable

The ftrace_preempt_disable/enable functions were to address a
recursive race caused by the function tracer. The function tracer
traces all functions which makes it easily susceptible to recursion.
One area was preempt_enable(). This would call the scheduler and
the schedulre would call the function tracer and loop.
(So was it thought).

The ftrace_preempt_disable/enable was made to protect against recursion
inside the scheduler by storing the NEED_RESCHED flag. If it was
set before the ftrace_preempt_disable() it would not call schedule
on ftrace_preempt_enable(), thinking that if it was set before then
it would have already scheduled unless it was already in the scheduler.

This worked fine except in the case of SMP, where another task would set
the NEED_RESCHED flag for a task on another CPU, and then kick off an
IPI to trigger it. This could cause the NEED_RESCHED to be saved at
ftrace_preempt_disable() but the IPI to arrive in the the preempt
disabled section. The ftrace_preempt_enable() would not call the scheduler
because the flag was already set before entring the section.

This bug would cause a missed preemption check and cause lower latencies.

Investigating further, I found that the recusion caused by the function
tracer was not due to schedule(), but due to preempt_schedule(). Now
that preempt_schedule is completely annotated with notrace, the recusion
no longer is an issue.

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
14 years agotracing/sched: Make preempt_schedule() notrace
Steven Rostedt [Thu, 3 Jun 2010 01:52:29 +0000 (21:52 -0400)]
tracing/sched: Make preempt_schedule() notrace

The function tracer code uses ftrace_preempt_disable() to disable
preemption instead of normal preempt_disable(). But there's a slight
race condition that may cause it to lose a preemption check.

This was made to keep the function tracer from recursing on itself
by disabling preemption then having the enable call the function tracer
again, causing infinite recursion.

The bug was assumed to happen if the call was just in schedule, but
this is incorrect. The bug is caused by preempt_schedule() which
is called by preempt_enable(). The calling of preempt_enable() when
NEED_RESCHED was set would call preempt_schedule() which would call
the function tracer again.

By making the preempt_schedule() and add_preempt_count() notrace
then this will prevent the inifinite recursion. This is because
the add_preempt_count() would stop the preempt_enable() in the
function tracer from calling preempt_schedule() again.

The sub_preempt_count() is also made notrace just to keep it
symmetric.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
14 years agoperf: Fix crash in swevents
Peter Zijlstra [Thu, 3 Jun 2010 09:21:20 +0000 (11:21 +0200)]
perf: Fix crash in swevents

Frederic reported that because swevents handling doesn't disable IRQs
anymore, we can get a recursion of perf_adjust_period(), once from
overflow handling and once from the tick.

If both call ->disable, we get a double hlist_del_rcu() and trigger
a LIST_POISON2 dereference.

Since we don't actually need to stop/start a swevent to re-programm
the hardware (lack of hardware to program), simply nop out these
callbacks for the swevent pmu.

Reported-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1275557609.27810.35218.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoMerge branch 'perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/acme...
Ingo Molnar [Wed, 2 Jun 2010 07:13:12 +0000 (09:13 +0200)]
Merge branch 'perf/urgent' of git://git./linux/kernel/git/acme/linux-2.6 into perf/urgent

14 years agoperf buildid-list: Fix --with-hits event processing
Arnaldo Carvalho de Melo [Tue, 1 Jun 2010 15:37:05 +0000 (12:37 -0300)]
perf buildid-list: Fix --with-hits event processing

When we use plain 'perf buildid-list' we use only what is in the buildid
table in the perf.data header. And those have absolute pathnames because
at 'perf record' time we used __perf_session__process_events and that
doesn't sets up the path shortening code in map__new() that happens if
symbol_conf.full_paths is false, the default.

On the other hand, when we use 'perf buildid-list --with-hits' we
process all the events using perf_session__process_events, adding
entries to the global DSO list _after_ removing the current directory
from the DSO name, for presentation purposes.

Because of that we end up having two entries in the DSO list when
recording events for binaries using relative pathnames.

Fix it minimally by setting symbol_conf.full_paths to true when marking
the DSOs with hits in 'perf buildid-list --with-hits', as used by 'perf
archive'

Right fix longer term is to shorten the path only at presentation time.
Will be done for 2.6.36.

Reported-by: Stephane Eranian <eranian@google.com>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <20100601183837.GC4093@ghostprotocols.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf scripts python: Give field dict to unhandled callback
Pierre Tardy [Mon, 31 May 2010 21:12:09 +0000 (23:12 +0200)]
perf scripts python: Give field dict to unhandled callback

trace_unhandled() callback does not allow to access event fields, this patch
resolves the problem.

It can also been used as a more pythonic and flexible way for script writters
to demux event types

This will for example greatly simplify pytimechart event demux.

Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Tom Zanussi <tzanussi@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>,
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <1275340329-2397-1-git-send-email-tardyp@gmail.com>
Signed-off-by: Pierre Tardy <tardyp@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf hist: fix objdump output parsing
Konstantin Stepanyuk [Wed, 26 May 2010 22:10:11 +0000 (02:10 +0400)]
perf hist: fix objdump output parsing

hist_entry__annotate() runs objdump with -S option so the output may contain
lines of any format. If a line starts with a colon strtoull() returns 0 and
calculated offset will be negative. This causes perf annotate segfaults.

Make sure that strtoull() has parsed at least one digit.

Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Konstantin Stepanyuk <konstantin.stepanyuk@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
14 years agoperf-record: Check correct pid when forking
Borislav Petkov [Mon, 31 May 2010 21:18:18 +0000 (23:18 +0200)]
perf-record: Check correct pid when forking

When forking the child to be traced, we should check the correct
return value from fork() and not a local variable which is otherwise
unused.

Signed-off-by: Borislav Petkov <bp@alien8.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100531211818.GA30175@liondog.tnic>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
14 years agoperf: Do the comm inheritance per thread in event__process_task
Frederic Weisbecker [Mon, 31 May 2010 22:43:07 +0000 (00:43 +0200)]
perf: Do the comm inheritance per thread in event__process_task

event__process_task() doesn't propagate the comm copy on clone,
but only on process fork. So we loose all the tid:comm resolution
for tasks that aren't a main process thread.

Progragate the per thread granularity to event__process_task for
pid resolution.

This fixes various unresolved pids in perf sched, especially when
we trace multithread processes. The problem is quickly reproducible
with the messaging benchmark using the multithread mode "-t" :

perf sched record perf bench sched messaging -t

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
14 years agoperf: Use event__process_task from perf sched
Frederic Weisbecker [Mon, 31 May 2010 22:03:40 +0000 (00:03 +0200)]
perf: Use event__process_task from perf sched

perf sched uses event__process_comm(), which means it can resolve
comms from:

- tasks that have exec'ed (kernel comm events)
- tasks that were running when perf record started the actual
  recording (synthetized comm events)

But perf sched can't resolve the pids of tasks that were created
after the recording started.

To solve this, we need to inherit the comms on fork events using
event__process_task().

This fixes various unresolved pids in perf sched, easily visible
with:
perf sched record perf bench sched messaging

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
14 years agoperf: Process comm events by tid
Frederic Weisbecker [Mon, 31 May 2010 21:38:44 +0000 (23:38 +0200)]
perf: Process comm events by tid

When we synthetize the existing running tasks though procfs,
we walk through every threads of a process, queuing one comm
events per tid.

But then on report time, event__process_comm() only creates and
sets the comm on a per process granularity. This is the right
thing for comm events that came from the kernel, as they are
only created on exec. Sub-threads then inherit their comm
from fork events. But that doesn't work with our synthetized
comm events taken from procfs informations as the per thread
granularity is done on comm events directly there.

Hence we need event__process_comm() to work with the tid rather
than the pid. It won't change anything for comm events coming
from the kernel but this will fix the synthetized ones.

Before:

$ ./perf report -D | grep COMM | grep firefox

0x2c7b8 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c7d0 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c7e8 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c800 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c818 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c830 [0x18]: PERF_RECORD_COMM: firefox:5297

After:
$ ./perf report -D | grep COMM | grep firefox

0x2c7b8 [0x18]: PERF_RECORD_COMM: firefox:5297
0x2c7d0 [0x18]: PERF_RECORD_COMM: firefox:5299
0x2c7e8 [0x18]: PERF_RECORD_COMM: firefox:5300
0x2c800 [0x18]: PERF_RECORD_COMM: firefox:5308
0x2c818 [0x18]: PERF_RECORD_COMM: firefox:5309
0x2c830 [0x18]: PERF_RECORD_COMM: firefox:5312

This fixes various unresolved pid on perf sched.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
14 years agoblktrace: Fix new kernel-doc warnings
Randy Dunlap [Sat, 29 May 2010 18:45:07 +0000 (11:45 -0700)]
blktrace: Fix new kernel-doc warnings

Fix blktrace.c kernel-doc warnings:
 Warning(kernel/trace/blktrace.c:858): No description found for parameter 'ignore'
 Warning(kernel/trace/blktrace.c:890): No description found for parameter 'ignore'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20100529114507.c466fc1e.randy.dunlap@oracle.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events: Fix unincremented buffer base on partial copy
Frederic Weisbecker [Thu, 27 May 2010 19:34:58 +0000 (21:34 +0200)]
perf_events: Fix unincremented buffer base on partial copy

If a sample size crosses to the next page boundary, the copy
will be made in more than one step. However we forget to advance
the source offset for the next copy, leading to unexpected double
copies that completely mess up the traces.

This fixes various kinds of bad traces that have irrelevant
data inside, as an example:

geany-4979  [001]  5758.077775: sched_switch: prev_comm=! prev_pid=121
prev_prio=0 prev_state=S|D|Z|X|x ==> next_comm= next_pid=7497072
next_prio=0

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1274988898-5639-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events: Fix event scheduling issues introduced by transactional API
Stephane Eranian [Tue, 25 May 2010 14:23:10 +0000 (16:23 +0200)]
perf_events: Fix event scheduling issues introduced by transactional API

The transactional API patch between the generic and model-specific
code introduced several important bugs with event scheduling, at
least on X86. If you had pinned events, e.g., watchdog,  and were
over-committing the PMU, you would get bogus counts. The bug was
showing up on Intel CPU because events would move around more
often that on AMD. But the problem also existed on AMD, though
harder to expose.

The issues were:

 - group_sched_in() was missing a cancel_txn() in the error path

 - cpuc->n_added was not properly maintained, leading to missing
   actions in hw_perf_enable(), i.e., n_running being 0. You cannot
   update n_added until you know the transaction has succeeded. In
   case of failed transaction n_added was not adjusted back.

 - in case of failed transactions, event_sched_out() was called
   and eventually invoked x86_disable_event() to touch the HW reg.
   But with transactions, on X86, event_sched_in() does not touch
   HW registers, it simply collects events into a list. Thus, you
   could end up calling x86_disable_event() on a counter which
   did not correspond to the current event when idx != -1.

The patch modifies the generic and X86 code to avoid all those problems.

First, we keep track of the number of events added last. In case the
transaction fails, we substract them from n_added. This approach is
necessary (as opposed to delaying updates to n_added) because not all
event updates use the transaction API, e.g., single events.

Second, we encapsulate the event_sched_in() and event_sched_out() in
group_sched_in() inside the transaction. That makes the operations
symmetrical and you can also detect that you are inside a transaction
and skip the HW reg access by checking cpuc->group_flag.

With this patch, you can now overcommit the PMU even with pinned
system-wide events present and still get valid counts.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1274796225.5882.1389.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events, trace: Fix perf_trace_destroy(), mutex went missing
Peter Zijlstra [Fri, 21 May 2010 14:22:33 +0000 (16:22 +0200)]
perf_events, trace: Fix perf_trace_destroy(), mutex went missing

Steve spotted I forgot to do the destroy under event_mutex.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1274451913.1674.1707.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events, trace: Fix probe unregister race
Peter Zijlstra [Fri, 21 May 2010 10:31:09 +0000 (12:31 +0200)]
perf_events, trace: Fix probe unregister race

tracepoint_probe_unregister() does not synchronize against the probe
callbacks, so do that explicitly. This properly serializes the callbacks
and the free of the data used therein.

Also, use this_cpu_ptr() where possible.

Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1274438476.1674.1702.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events: Fix races in group composition
Peter Zijlstra [Thu, 27 May 2010 13:47:49 +0000 (15:47 +0200)]
perf_events: Fix races in group composition

Group siblings don't pin each-other or the parent, so when we destroy
events we must make sure to clean up all cross referencing pointers.

In particular, for destruction of a group leader we must be able to
find all its siblings and remove their reference to it.

This means that detaching an event from its context must not detach it
from the group, otherwise we can end up failing to clear all pointers.

Solve this by clearly separating the attachment to a context and
attachment to a group, and keep the group composed until we destroy
the events.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_events: Fix races and clean up perf_event and perf_mmap_data interaction
Peter Zijlstra [Thu, 27 May 2010 10:54:41 +0000 (12:54 +0200)]
perf_events: Fix races and clean up perf_event and perf_mmap_data interaction

In order to move toward separate buffer objects, rework the whole
perf_mmap_data construct to be a more self-sufficient entity, one
with its own lifetime rules.

This greatly sanitizes the whole output redirection code, which
was riddled with bugs and races.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoLinux 2.6.35-rc1
Linus Torvalds [Sun, 30 May 2010 20:21:02 +0000 (13:21 -0700)]
Linux 2.6.35-rc1

.. and thus endeth the merge window.

14 years agoMerge branch 'slub/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/penberg...
Linus Torvalds [Sun, 30 May 2010 19:46:17 +0000 (12:46 -0700)]
Merge branch 'slub/urgent' of git://git./linux/kernel/git/penberg/slab-2.6

* 'slub/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/penberg/slab-2.6:
  SLUB: Allow full duplication of kmalloc array for 390
  slub: move kmem_cache_node into it's own cacheline

14 years agoMerge branch 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 30 May 2010 19:35:15 +0000 (12:35 -0700)]
Merge branch 'core-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  mutex: Fix optimistic spinning vs. BKL

14 years agoMerge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 30 May 2010 19:35:01 +0000 (12:35 -0700)]
Merge branch 'perf-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  perf tui: Fix last use_browser problem related to .perfconfig
  perf symbols: Add the build id cache to the vmlinux path
  perf tui: Reset use_browser if stdout is not a tty
  ring-buffer: Move zeroing out excess in page to ring buffer code
  ring-buffer: Reset "real_end" when page is filled

14 years agoia64: revert __node_random addition
Linus Torvalds [Sun, 30 May 2010 17:08:03 +0000 (10:08 -0700)]
ia64: revert __node_random addition

This partially reverts commit 4ec37de89d8c758ee8115e0e64b3f994910789ee
("[IA64] Fix build breakage"), since the commit that made it necessary
got reverted earlier (see commit 35926ff5fba8, 'Revert "cpusets:
randomize node rotor used in cpuset_mem_spread_node()"')

Even if we ever re-introduce this, there is no reason to make
__node_random be some architecture-specific function.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi...
Linus Torvalds [Sun, 30 May 2010 16:16:14 +0000 (09:16 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mszeredi/fuse

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  mm: export generic_pipe_buf_*() to modules
  fuse: support splice() reading from fuse device
  fuse: allow splice to move pages
  mm: export remove_from_page_cache() to modules
  mm: export lru_cache_add_*() to modules
  fuse: support splice() writing to fuse device
  fuse: get page reference for readpages
  fuse: use get_user_pages_fast()
  fuse: remove unneeded variable

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
Linus Torvalds [Sun, 30 May 2010 16:13:43 +0000 (09:13 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/rostedt/linux-2.6-kconfig

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-2.6-kconfig:
  kconfig: Hide error output in find command in streamline_config.pl
  kconfig: Fix typo in comment in streamline_config.pl
  kconfig: Make a variable local in streamline_config.pl

14 years agoMerge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/mfd-2.6
Linus Torvalds [Sun, 30 May 2010 16:13:08 +0000 (09:13 -0700)]
Merge branch 'for-next' of git://git./linux/kernel/git/sameo/mfd-2.6

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/mfd-2.6: (47 commits)
  mfd: Rename twl5031 sih modules
  mfd: Storage class for timberdale should be before const qualifier
  mfd: Remove unneeded and dangerous clearing of clientdata
  mfd: New AB8500 driver
  gpio: Fix inverted rdc321x gpio data out registers
  mfd: Change rdc321x resources flags to IORESOURCE_IO
  mfd: Move pcf50633 irq related functions to its own file.
  mfd: Use threaded irq for pcf50633
  mfd: pcf50633-adc: Fix potential race in pcf50633_adc_sync_read
  mfd: Fix pcf50633 bitfield logic in interrupt handler
  gpio: rdc321x needs to select MFD_CORE
  mfd: Use menuconfig for quicker config editing
  ARM: AB3550 board configuration and irq for U300
  mfd: AB3550 core driver
  mfd: AB3100 register access change to abx500 API
  mfd: Renamed ab3100.h to abx500.h
  gpio: Add TC35892 GPIO driver
  mfd: Add Toshiba's TC35892 MFD core
  mfd: Delay to mask tsc irq in max8925
  mfd: Remove incorrect wm8350 kfree
  ...

14 years agoMerge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx
Linus Torvalds [Sun, 30 May 2010 16:12:43 +0000 (09:12 -0700)]
Merge branch 'next' of git://git./linux/kernel/git/djbw/async_tx

* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx:
  DMAENGINE: DMA40 U8500 platform configuration
  DMA: PL330: Add dma api driver

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Sun, 30 May 2010 16:12:16 +0000 (09:12 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  IB/qib: Remove DCA support until feature is finished
  IB/qib: Use a single txselect module parameter for serdes tuning
  IB/qib: Don't rely on (undefined) order of function parameter evaluation
  IB/ucm: Use memdup_user()
  IB/qib: Fix undefined symbol error when CONFIG_PCI_MSI=n

14 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux...
Linus Torvalds [Sun, 30 May 2010 16:11:11 +0000 (09:11 -0700)]
Merge branch 'for_linus' of git://git./linux/kernel/git/jack/linux-fs-2.6

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs-2.6:
  quota: Convert quota statistics to generic percpu_counter
  ext3 uses rb_node = NULL; to zero rb_root.
  quota: Fixup dquot_transfer
  reiserfs: Fix resuming of quotas on remount read-write
  pohmelfs: Remove dead quota code
  ufs: Remove dead quota code
  udf: Remove dead quota code
  quota: rename default quotactl methods to dquot_
  quota: explicitly set ->dq_op and ->s_qcop
  quota: drop remount argument to ->quota_on and ->quota_off
  quota: move unmount handling into the filesystem
  quota: kill the vfs_dq_off and vfs_dq_quota_on_remount wrappers
  quota: move remount handling into the filesystem
  ocfs2: Fix use after free on remount read-only

Fix up conflicts in fs/ext4/super.c and fs/ufs/file.c

14 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 30 May 2010 16:06:13 +0000 (09:06 -0700)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86, cpufeature: Unbreak compile with gcc 3.x
  x86, pat: Fix memory leak in free_memtype
  x86, k8: Fix section mismatch for powernowk8_exit()
  lib/atomic64_test: fix missing include of linux/kernel.h
  x86: remove last traces of quicklist usage
  x86, setup: Phoenix BIOS fixup is needed on Dell Inspiron Mini 1012
  x86: "nosmp" command line option should force the system into UP mode
  arch/x86/pci: use kasprintf
  x86, apic: ack all pending irqs when crashed/on kexec

14 years agoCPU: Avoid using unititialized error variable in disable_nonboot_cpus()
Rafael J. Wysocki [Thu, 27 May 2010 20:16:22 +0000 (22:16 +0200)]
CPU: Avoid using unititialized error variable in disable_nonboot_cpus()

If there's only one CPU online when disable_nonboot_cpus() is called,
the error variable will not be initialized and that may lead to
erroneous behavior.  Fix this issue by initializing error in
disable_nonboot_cpus() as appropriate.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agorapidio: fix new kernel-doc warnings
Randy Dunlap [Fri, 28 May 2010 22:08:08 +0000 (15:08 -0700)]
rapidio: fix new kernel-doc warnings

Fix a bunch of new rapidio kernel-doc warnings:

Warning(include/linux/rio.h:123): No description found for parameter 'comp_tag'
Warning(include/linux/rio.h:123): No description found for parameter 'phys_efptr'
Warning(include/linux/rio.h:123): No description found for parameter 'em_efptr'
Warning(include/linux/rio.h:123): No description found for parameter 'pwcback'
Warning(include/linux/rio.h:247): No description found for parameter 'set_domain'
Warning(include/linux/rio.h:247): No description found for parameter 'get_domain'
Warning(drivers/rapidio/rio-scan.c:1133): No description found for parameter 'rdev'
Warning(drivers/rapidio/rio-scan.c:1133): Excess function parameter 'port' description in 'rio_init_em'
Warning(drivers/rapidio/rio.c:349): No description found for parameter 'rdev'
Warning(drivers/rapidio/rio.c:349): Excess function parameter 'mport' description in 'rio_request_inb_pwrite'
Warning(drivers/rapidio/rio.c:393): No description found for parameter 'port'
Warning(drivers/rapidio/rio.c:393): No description found for parameter 'local'
Warning(drivers/rapidio/rio.c:393): No description found for parameter 'destid'
Warning(drivers/rapidio/rio.c:393): No description found for parameter 'hopcount'
Warning(drivers/rapidio/rio.c:393): Excess function parameter 'rdev' description in 'rio_mport_get_physefb'
Warning(drivers/rapidio/rio.c:845): Excess function parameter 'local' description in 'rio_std_route_clr_table'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Alexandre Bounine <alexandre.bounine@idt.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/kyle/parisc-2.6
Linus Torvalds [Sun, 30 May 2010 16:02:02 +0000 (09:02 -0700)]
Merge branch 'fixes' of git://git./linux/kernel/git/kyle/parisc-2.6

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/kyle/parisc-2.6:
  parisc: Call pagefault_disable/pagefault_enable in kmap_atomic/kunmap_atomic
  parisc: Remove unnecessary macros from entry.S
  parisc: LWS fixes for syscall.S
  parisc: Delete unnecessary nop's in entry.S
  parisc: Avoid interruption in critical region in entry.S
  parisc: invoke oom-killer from page fault
  parisc: clear floating point exception flag on SIGFPE signal
  parisc: Use of align_frame provides stack frame.

14 years agoRevert "cpusets: randomize node rotor used in cpuset_mem_spread_node()"
Linus Torvalds [Sun, 30 May 2010 16:00:03 +0000 (09:00 -0700)]
Revert "cpusets: randomize node rotor used in cpuset_mem_spread_node()"

This reverts commit 0ac0c0d0f837c499afd02a802f9cf52d3027fa3b, which
caused cross-architecture build problems for all the wrong reasons.
IA64 already added its own version of __node_random(), but the fact is,
there is nothing architectural about the function, and the original
commit was just badly done. Revert it, since no fix is forthcoming.

Requested-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Sun, 30 May 2010 15:56:39 +0000 (08:56 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  ceph: clean up on forwarded aborted mds request
  ceph: fix leak of osd authorizer
  ceph: close out mds, osd connections before stopping auth
  ceph: make lease code DN specific
  fs/ceph: Use ERR_CAST
  ceph: renew auth tickets before they expire
  ceph: do not resend mon requests on auth ticket renewal
  ceph: removed duplicated #includes
  ceph: avoid possible null dereference
  ceph: make mds requests killable, not interruptible
  sched: add wait_for_completion_killable_timeout

14 years agoSLUB: Allow full duplication of kmalloc array for 390
Christoph Lameter [Thu, 27 May 2010 14:17:17 +0000 (09:17 -0500)]
SLUB: Allow full duplication of kmalloc array for 390

Commit 756dee75872a2a764b478e18076360b8a4ec9045 ("SLUB: Get rid of dynamic DMA
kmalloc cache allocation") makes S390 run out of kmalloc caches.  Increase the
number of kmalloc caches to a safe size.

Cc: <stable@kernel.org> [ .33 and .34 ]
Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
14 years agoparisc: Call pagefault_disable/pagefault_enable in kmap_atomic/kunmap_atomic
John David Anglin [Sun, 11 Apr 2010 16:36:14 +0000 (16:36 +0000)]
parisc: Call pagefault_disable/pagefault_enable in kmap_atomic/kunmap_atomic

Based on the generic implementation of kmap_atomic and kunmap_atomic,
we should call pagefault_disable and pagefault_enable in our PA8000
implementation.

The define for kmap_atomic_prot was also missing, and I updated
kmap_atomic_pfn to use the generic implementation because of the
change to kmap_atomic.

I believe that this change is needed to fix the fork copy-on-write
bug.

Signed-off-by: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoparisc: Remove unnecessary macros from entry.S
John David Anglin [Sun, 11 Apr 2010 17:03:54 +0000 (17:03 +0000)]
parisc: Remove unnecessary macros from entry.S

The EXTR, DEP and DEPI macros are unnecessary.  There are PA 1.X
pneumonics available with the same functionality, and the DEP and DEPI
macros conflict with assembler pneumonics.

Tested on a variety of 32 and 64-bit systems.

Signed-off-by: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoparisc: LWS fixes for syscall.S
John David Anglin [Sun, 11 Apr 2010 17:26:34 +0000 (17:26 +0000)]
parisc: LWS fixes for syscall.S

1) Gate immediately and save a branch.
2) Fix off by one error in checking entry number.
3) Use sr7 instead of sr3 in error return path as sr3 might not
   contain correct value.
4) Enable locking on UP systems to prevent incorrect operation of
   the cas_action critical region on page faults.

Tested on several systems, including UP c3750 with 2.6.33.2 kernel.

Signed-off-by: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoparisc: Delete unnecessary nop's in entry.S
John David Anglin [Sun, 11 Apr 2010 17:08:51 +0000 (17:08 +0000)]
parisc: Delete unnecessary nop's in entry.S

Signed-off-by: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoparisc: Avoid interruption in critical region in entry.S
John David Anglin [Sun, 11 Apr 2010 17:12:56 +0000 (17:12 +0000)]
parisc: Avoid interruption in critical region in entry.S

Signed-off-by: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoparisc: invoke oom-killer from page fault
Nick Piggin [Thu, 22 Apr 2010 16:06:23 +0000 (16:06 +0000)]
parisc: invoke oom-killer from page fault

As explained in commit 1c0fe6e3bd, we want to call the architecture independent
oom killer when getting an unexplained OOM from handle_mm_fault, rather than
simply killing current.

Cc: linux-parisc@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Kyle McMartin <kyle@mcmartin.ca>
14 years agoparisc: clear floating point exception flag on SIGFPE signal
Helge Deller [Mon, 3 May 2010 20:44:21 +0000 (20:44 +0000)]
parisc: clear floating point exception flag on SIGFPE signal

Clear the floating point exception flag before returning to
user space. This is needed, else the libc trampoline handler
may hit the same SIGFPE again while building up a trampoline
to a signal handler.

Fixes debian bug #559406.

Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Kyle McMartin <kyle@mcmartin.ca>
14 years agoparisc: Use of align_frame provides stack frame.
Carlos O'Donell [Wed, 31 Mar 2010 20:42:33 +0000 (20:42 +0000)]
parisc: Use of align_frame provides stack frame.

Any assembly constant generated with the use of
align_frame includes size for a full stack frame.

Signed-off-by: Carlos O'Donell <carlos@systemhalted.org>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Sat, 29 May 2010 22:31:57 +0000 (15:31 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6: (26 commits)
  ALSA: snd-usb-caiaq: Bump version number to 1.3.21
  ALSA: Revert "ALSA: snd-usb-caiaq: Set default input mode of A4DJ"
  ALSA: snd-usb-caiaq: Simplify single case to an 'if'
  ALSA: snd-usb-caiaq: Restore 'Control vinyl' input mode on A4DJ
  ALSA: hda: Use LPIB for a Shuttle device
  ALSA: hda: Add support for another Lenovo ThinkPad Edge in conexant codec
  ALSA: hda: Use LPIB for Sony VPCS11V9E
  ALSA: usb-audio: fix feature unit parser for UAC2
  ALSA: asihpi - Minor code cleanup
  ALSA: asihpi - Add support for new ASI8800 family
  ALSA: asihpi - Fix bug preventing outstream_write preload from happening
  ALSA: asihpi - Fix imbalanced lock path in hw_message
  ALSA: asihpi - Remove support for old ASI8800 family
  ALSA: asihpi - Add hd radio blend functions
  ALSA: asihpi - Remove unused io map functions
  ALSA: usb-audio: add support for UAC2 pitch control
  ALSA: usb-audio: parse UAC2 endpoint descriptors correctly
  ALSA: usb-audio: fix return values
  ALSA: usb-audio: parse more format descriptors with structs
  sound: Add missing spin_unlock
  ...

14 years agoMerge branch 'fix/hda' into for-linus
Takashi Iwai [Sat, 29 May 2010 19:50:36 +0000 (21:50 +0200)]
Merge branch 'fix/hda' into for-linus

14 years agoMerge branch 'fix/misc' into for-linus
Takashi Iwai [Sat, 29 May 2010 19:50:33 +0000 (21:50 +0200)]
Merge branch 'fix/misc' into for-linus

14 years agoMerge branch 'fix/asoc' into for-linus
Takashi Iwai [Sat, 29 May 2010 19:50:27 +0000 (21:50 +0200)]
Merge branch 'fix/asoc' into for-linus

14 years agoALSA: snd-usb-caiaq: Bump version number to 1.3.21
Mark Hills [Sat, 29 May 2010 15:53:26 +0000 (16:53 +0100)]
ALSA: snd-usb-caiaq: Bump version number to 1.3.21

Acked-by: Daniel Mack <daniel@caiaq.de>
Signed-off-by: Mark Hills <mark@pogo.org.uk>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoALSA: Revert "ALSA: snd-usb-caiaq: Set default input mode of A4DJ"
Mark Hills [Sat, 29 May 2010 15:53:25 +0000 (16:53 +0100)]
ALSA: Revert "ALSA: snd-usb-caiaq: Set default input mode of A4DJ"

Do not explicity set the default input mode. Use the hardware default
of mode 0 ('Control vinyl'), which is now available.

This reverts commit e3ca4c9.

Acked-by: Daniel Mack <daniel@caiaq.de>
Signed-off-by: Mark Hills <mark@pogo.org.uk>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoALSA: snd-usb-caiaq: Simplify single case to an 'if'
Mark Hills [Sat, 29 May 2010 15:53:24 +0000 (16:53 +0100)]
ALSA: snd-usb-caiaq: Simplify single case to an 'if'

After removing code, only one case remains. So use an 'if' instead.

Acked-by: Daniel Mack <daniel@caiaq.de>
Signed-off-by: Mark Hills <mark@pogo.org.uk>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoALSA: snd-usb-caiaq: Restore 'Control vinyl' input mode on A4DJ
Mark Hills [Sat, 29 May 2010 15:53:23 +0000 (16:53 +0100)]
ALSA: snd-usb-caiaq: Restore 'Control vinyl' input mode on A4DJ

This feature was undocumented on early A4DJ units. It is indicated
by lighting both the 'line' and 'phono' lamps at the same time.
Newer units document this and the newer Windows drivers enable this
for all units, so restore the functionality.

This patch simplifies the code and changes the mode mapping to match
the A8DJ, favouring simpler code and consistency over keeping the
existing mapping.

Both 'Control vinyl' and 'Phono' input modes enable the hardware
preamp. The difference is the input impedance.

This reverts commit 9a9527e.

Acked-by: Daniel Mack <daniel@caiaq.de>
Signed-off-by: Mark Hills <mark@pogo.org.uk>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoALSA: hda: Use LPIB for a Shuttle device
Daniel T Chen [Sat, 29 May 2010 15:04:11 +0000 (11:04 -0400)]
ALSA: hda: Use LPIB for a Shuttle device

BugLink: https://launchpad.net/bugs/551949
Symptom: On the reporter's Shuttle device, using PulseAudio in Ubuntu
10.04 LTS results in "popping clicking" audio with the PA crashing
shortly thereafter.

Test case: Using Ubuntu 10.04 LTS (Linux 2.6.32.12), Linux 2.6.33, or
Linux 2.6.34, adjust the HDA device's volume with PulseAudio.

Resolution: add SSID for this machine to the position_fix quirk table,
explicitly specifying the LPIB method.

Reported-and-Tested-By: Christian Mehlis <mehlis@inf.fu-berlin.de>
Cc: <stable@kernel.org>
Signed-off-by: Daniel T Chen <crimsun@ubuntu.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoceph: clean up on forwarded aborted mds request
Sage Weil [Fri, 28 May 2010 23:43:16 +0000 (16:43 -0700)]
ceph: clean up on forwarded aborted mds request

If an mds request is aborted (timeout, SIGKILL), it is left registered to
keep our state in sync with the mds.  If we get a forward notification,
though, we know the request didn't succeed and we can unregister it
safely.  We were trying to resend it, but then bailing out (and not
unregistering) in __do_request.

Signed-off-by: Sage Weil <sage@newdream.net>
14 years agoceph: fix leak of osd authorizer
Sage Weil [Thu, 27 May 2010 21:15:49 +0000 (14:15 -0700)]
ceph: fix leak of osd authorizer

Release the ceph_authorizer when releasing osd state.

Signed-off-by: Sage Weil <sage@newdream.net>
14 years agoceph: close out mds, osd connections before stopping auth
Sage Weil [Sat, 29 May 2010 16:41:23 +0000 (09:41 -0700)]
ceph: close out mds, osd connections before stopping auth

The auth module (part of the mon_client) is needed to free any
ceph_authorizer(s) used by the mds and osd connections.  Flush the msgr
workqueue before stopping monc to ensure that the destroy_authorizer
auth op is available when those connections are closed out.

Signed-off-by: Sage Weil <sage@newdream.net>
14 years agoceph: make lease code DN specific
Sage Weil [Tue, 25 May 2010 23:45:25 +0000 (16:45 -0700)]
ceph: make lease code DN specific

The lease code includes a mask in the CEPH_LOCK_* namespace, but that
namespace is changing, and only one mask (formerly _DN == 1) is used, so
hard code for that value for now.

If we ever extend this code to handle leases over different data types we
can extend it accordingly.

Signed-off-by: Sage Weil <sage@newdream.net>
14 years agofs/ceph: Use ERR_CAST
Julia Lawall [Sat, 22 May 2010 10:01:14 +0000 (12:01 +0200)]
fs/ceph: Use ERR_CAST

Use ERR_CAST(x) rather than ERR_PTR(PTR_ERR(x)).  The former makes more
clear what is the purpose of the operation, which otherwise looks like a
no-op.

In the case of fs/ceph/inode.c, ERR_CAST is not needed, because the type of
the returned value is the same as the type of the enclosing function.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
type T;
T x;
identifier f;
@@

T f (...) { <+...
- ERR_PTR(PTR_ERR(x))
+ x
 ...+> }

@@
expression x;
@@

- ERR_PTR(PTR_ERR(x))
+ ERR_CAST(x)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Sage Weil <sage@newdream.net>
14 years agoceph: renew auth tickets before they expire
Sage Weil [Tue, 25 May 2010 22:39:06 +0000 (15:39 -0700)]
ceph: renew auth tickets before they expire

We were only requesting renewal after our tickets expire; do so before
that.  Most of the low-level logic for this was already there; just use
it.

Signed-off-by: Sage Weil <sage@newdream.net>