GitHub/LineageOS/android_kernel_samsung_universal7580.git
13 years agoperf tools: Make clean leaves some files
Kusanagi Kouichi [Sun, 7 Aug 2011 08:39:31 +0000 (17:39 +0900)]
perf tools: Make clean leaves some files

Use LIB_OBJS and BUILTIN_OBJS for .o files.

LIB_FILE is already prefixed with OUTPUT.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110807083932.9C0E514C03B@msa103.auone-net.jp
Signed-off-by: Kusanagi Kouichi <slash@ac.auone-net.jp>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf lock: Dropping unsupported ':r' modifier
Zhu Yanhai [Sat, 30 Jul 2011 14:13:52 +0000 (22:13 +0800)]
perf lock: Dropping unsupported ':r' modifier

Looks to me like the :r modifier is not supported anymore, so remove it
from the list of events. Without this fix 'perf lock record' doesn't
work.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Zhu Yanhai <gaoyang.zyh@taobao.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1312035232-9534-1-git-send-email-gaoyang.zyh@taobao.com
Signed-off-by: Zhu Yanhai <gaoyang.zyh@taobao.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf probe: Fix coredump introduced by probe module option
Jovi Zhang [Mon, 25 Jul 2011 14:08:08 +0000 (22:08 +0800)]
perf probe: Fix coredump introduced by probe module option

perf will coredump if the user doesn't give the "-m" option in probe
command, this patch fixes it.

[root@localhost perf]# ./perf probe --add='PROBE'
Segmentation fault (core dumped)

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1311602888-2389-1-git-send-email-bookjovi@gmail.com
Signed-off-by: Jovi Zhang <bookjovi@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf report: Use ui__warning in some more places
Arnaldo Carvalho de Melo [Wed, 3 Aug 2011 15:33:24 +0000 (12:33 -0300)]
perf report: Use ui__warning in some more places

So that we get a proper warning in the TUI in cases like:

 $ perf report --stdio -g fractal,0.5,caller --sort pid
 Selected -g but no callchain data. Did you call 'perf record' without -g?
 $

The --stdio case is ok because it uses fprintf, ui__warning is needed to
figure out if --stdio or --tui is being used.

Cc: Arun Sharma <asharma@fb.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sam Liao <phyomh@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ag9fz2wd17mbbfjsbznq1wms@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf python: Add PERF_RECORD_{LOST,READ,SAMPLE} routine tables
Arnaldo Carvalho de Melo [Mon, 25 Jul 2011 20:13:27 +0000 (17:13 -0300)]
perf python: Add PERF_RECORD_{LOST,READ,SAMPLE} routine tables

So those friggin "spurious" PERF_RECORD_MMAP events were actually a
brain fart copy'n'paste error in the python binding, doh. I.e. they
weren't MMAPs, just SAMPLEs.

Fix it by providing routines for these events instead of using the MMAP
ones.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-b0rc8y5jd03f9f11kftodvkm@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf evlist: Introduce 'disable' method
Arnaldo Carvalho de Melo [Mon, 25 Jul 2011 14:06:19 +0000 (11:06 -0300)]
perf evlist: Introduce 'disable' method

To remove the last case of access to the FD() macro outside the library.

Inspired by a patch by Borislav that moved the FD() macro to util.h, for
namespace concerns I rather preferred to constrain it to ev{sel,list}.c.

Cc: Borislav Petkov <bp@amd64.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-qn893qsstcg366tkucu649qj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf buildid-cache: Zero out buffer of filenames when adding/removing buildid
Han Pingtian [Mon, 18 Jul 2011 03:13:14 +0000 (11:13 +0800)]
perf buildid-cache: Zero out buffer of filenames when adding/removing buildid

The readlink() function doesn't append a null byte to buf. So we should
zero out buf with zalloc(). Or we'll see sometimes error like this:

[root@intel-s3e36-01]~# /usr/bin/perf buildid-cache -a /lib/modules/2.6.32-130.el6.x86_64/kernel/crypto/twofish_common.ko -v
Adding f64ba8efd5f53c7ad332fc17db1d21de309038e1 /lib/modules/2.6.32-130.el6.x86_64/kernel/crypto/twofish_common.ko: Ok
[root@intel-s3e36-01]~# /usr/bin/perf buildid-cache -r /lib/modules/2.6.32-130.el6.x86_64/kernel/crypto/twofish_common.ko -v
Removing f64ba8efd5f53c7ad332fc17db1d21de309038e1 /lib/modules/2.6.32-130.el6.x86_64/kernel/crypto/twofish_common.ko: FAIL
/lib/modules/2.6.32-130.el6.x86_64/kernel/crypto/twofish_common.ko wasn't in the cache

The change in build_id_cache__add_s() is a defense.

Tested-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/20110718031314.GA5802@hpt.nay.redhat.com
Signed-off-by: Han Pingtian <phan@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf: Remove the nmi parameter from the oprofile_perf backend
Will Deacon [Fri, 8 Jul 2011 17:34:42 +0000 (18:34 +0100)]
perf: Remove the nmi parameter from the oprofile_perf backend

In commit a8b0ca17b80e ("perf: Remove the nmi parameter from the
swevent and overflow interface") one site was overlooked.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110708173442.GB31972@e102144-lin.cambridge.arm.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agox86, perf: Make copy_from_user_nmi() a library function
Robert Richter [Tue, 7 Jun 2011 09:49:55 +0000 (11:49 +0200)]
x86, perf: Make copy_from_user_nmi() a library function

copy_from_user_nmi() is used in oprofile and perf. Moving it to other
library functions like copy_from_user(). As this is x86 code for 32
and 64 bits, create a new file usercopy.c for unified code.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110607172413.GJ20052@erda.amd.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf: Remove perf_event_attr::type check
Lin Ming [Thu, 30 Jun 2011 08:09:55 +0000 (08:09 +0000)]
perf: Remove perf_event_attr::type check

PMU type id can be allocated dynamically, so perf_event_attr::type check
when copying attribute from userspace to kernel is not valid.

Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Cc: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1309421396-17438-4-git-send-email-ming.m.lin@intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agox86, perf: P4 PMU - Fix typos in comments and style cleanup
Cyrill Gorcunov [Thu, 21 Jul 2011 16:06:25 +0000 (20:06 +0400)]
x86, perf: P4 PMU - Fix typos in comments and style cleanup

This patch:

 - fixes typos in comments and clarifies the text
 - renames obscure p4_event_alias::original and ::alter members to
   ::original and ::alternative as appropriate
 - drops parenthesis from the return of p4_get_alias_event()

No functional changes.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Link: http://lkml.kernel.org/r/20110721160625.GX7492@sun
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: Make test use the preset debugfs path
Jiri Olsa [Thu, 14 Jul 2011 09:25:34 +0000 (11:25 +0200)]
perf tools: Make test use the preset debugfs path

Use preset debugfs path instead of hardcoded one.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: acme@redhat.com
Cc: a.p.zijlstra@chello.nl
Cc: paulus@samba.org
Link: http://lkml.kernel.org/r/1310635534-4013-4-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: Add automated tests for events parsing
Jiri Olsa [Thu, 14 Jul 2011 09:25:33 +0000 (11:25 +0200)]
perf tools: Add automated tests for events parsing

Adding builtin test for parse_events function, which is
responsible for parsing/processing "-e" option for
stat/top/record commands.

This new test will run within the builtin test command suite
(perf test).

One or several tests were added for each type of event.
More tests could be added easily if needed.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: acme@redhat.com
Cc: a.p.zijlstra@chello.nl
Cc: paulus@samba.org
Link: http://lkml.kernel.org/r/1310635534-4013-3-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: De-opt the parse_events function
Jiri Olsa [Thu, 14 Jul 2011 09:25:32 +0000 (11:25 +0200)]
perf tools: De-opt the parse_events function

Moving out the option parameter from parse_events function,
and adding new parse_events_option function instead.

The option parameter is used only to carry "struct perf_evlist"
pointer for chaining new events. Putting it away, enable us
to call parse_events from other places without using the
option parameter.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: acme@redhat.com
Cc: a.p.zijlstra@chello.nl
Cc: paulus@samba.org
Link: http://lkml.kernel.org/r/1310635534-4013-2-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf script: Fix display of IP address for non-callchain path
David Ahern [Mon, 30 May 2011 15:16:27 +0000 (09:16 -0600)]
perf script: Fix display of IP address for non-callchain path

Non-callchain path is using al.addr which prints as:
  openssl 14564 17672.003587:       7862d _x86_64_AES_encrypt_compact

This should be sample->ip to print as:
  openssl 14564 17672.003587:  3f7867862d _x86_64_AES_encrypt_compact

Signed-off-by: David Ahern <dsahern@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: acme@ghostprotocols.net
Cc: peterz@infradead.org
Cc: paulus@samba.org
Link: http://lkml.kernel.org/r/1306768587-15376-1-git-send-email-dsahern@gmail.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: Fix endian conversion reading event attr from file header
David Ahern [Fri, 15 Jul 2011 18:34:09 +0000 (12:34 -0600)]
perf tools: Fix endian conversion reading event attr from file header

The perf_event_attr struct has two __u32's at the top and
they need to be swapped individually.

With this change I was able to analyze a perf.data collected in a
32-bit PPC VM on an x86 system. I tested both 32-bit and 64-bit
binaries for the Intel analysis side; both read the PPC perf.data
file correctly.

-v2:
 - changed the existing perf_event__attr_swap() to swap only elements
   of perf_event_attr and exported it for use in swapping the
   attributes in the file header
 - updated swap_ops used for processing events

Signed-off-by: David Ahern <dsahern@gmail.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: acme@ghostprotocols.net
Cc: peterz@infradead.org
Cc: paulus@samba.org
Cc: <stable@kernel.org>
Link: http://lkml.kernel.org/r/1310754849-12474-1-git-send-email-dsahern@gmail.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: Add missing 'node' alias to the hw_cache[] array
Jiri Olsa [Wed, 13 Jul 2011 20:58:18 +0000 (22:58 +0200)]
perf tools: Add missing 'node' alias to the hw_cache[] array

Add "node" as a simple alias for NODE cache events.

The addition of NODE cache events broke the parse_alias
function, so any mismatched event caused the segfault, like:

  # ./perf stat -e krava ls

The hw_cache/hw_cache_op/hw_cache_result arrays needs to follow
PERF_COUNT_HW_CACHE_*MAX enums. Adding those MAXs to be size
of those arrays, so possible ommision in future wil not lead to
segfault.

Adding read/write/prefetch as allowed operations for node cache
event.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: acme@redhat.com
Link: http://lkml.kernel.org/r/20110713205818.GB7827@jolsa.brq.redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoMerge branch 'tip/perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
Ingo Molnar [Thu, 21 Jul 2011 07:32:40 +0000 (09:32 +0200)]
Merge branch 'tip/perf/core' of git://git./linux/kernel/git/rostedt/linux-2.6-trace into perf/core

13 years agoMerge branch 'perf/urgent' into perf/core
Ingo Molnar [Thu, 21 Jul 2011 07:29:14 +0000 (09:29 +0200)]
Merge branch 'perf/urgent' into perf/core

Merge reason: pick up the latest fixes - they won't make v3.0.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf probe: Support adding probes on offline kernel modules
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:51 +0000 (16:27 +0900)]
perf probe: Support adding probes on offline kernel modules

Support adding probes on offline kernel modules. This enables
perf-probe to trace kernel-module init functions via perf-probe.
If user gives the path of module with -m option, perf-probe
expects the module is offline.
This feature works with --add, --funcs, and --vars.

E.g)
 # perf probe -m /lib/modules/`uname -r`/kernel/fs/btrfs/btrfs.ko \
   -a "extent_io_init:5 extent_state_cache"
 Add new events:
   probe:extent_io_init (on extent_io_init:5 with extent_state_cache)
   probe:extent_io_init_1 (on extent_io_init:5 with extent_state_cache)

 You can now use it on all perf tools, such as:

         perf record -e probe:extent_io_init_1 -aR sleep 1

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072751.6528.10230.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf probe: Add probed module in front of function
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:45 +0000 (16:27 +0900)]
perf probe: Add probed module in front of function

Add probed module name and ":" in front of function name
if -m module option is given. In the result, the symbol
name passed to kprobe-tracer becomes MODULE:FUNCTION,
so that kallsyms can solve it as a symbol in the module
correctly.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072745.6528.26416.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf probe: Introduce debuginfo to encapsulate dwarf information
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:39 +0000 (16:27 +0900)]
perf probe: Introduce debuginfo to encapsulate dwarf information

Introduce debuginfo to encapsulate dwarf information.
This new object allows us to reuse and expand debuginfo easily.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072739.6528.12438.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf-probe: Move dwarf library routines to dwarf-aux.{c, h}
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:27 +0000 (16:27 +0900)]
perf-probe: Move dwarf library routines to dwarf-aux.{c, h}

Move dwarf library related routines to dwarf-aux.{c,h}.
This includes several minor changes.
- Add simple documents for each API.
- Rename die_find_real_subprogram() to die_find_realfunc()
- Rename line_walk_handler_t to line_walk_callback_t.
- Minor cleanups.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072727.6528.57647.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf probe: Remove redundant dwarf functions
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:21 +0000 (16:27 +0900)]
perf probe: Remove redundant dwarf functions

Since there are dwarf_bitsize, dwarf_bitoffset and dwarf_bytesize
defined in libdw, we don't need die_get_bit_size, die_get_bit_offset
and die_get_byte_size anymore.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072721.6528.2747.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf probe: Move strtailcmp to string.c
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:15 +0000 (16:27 +0900)]
perf probe: Move strtailcmp to string.c

Since strtailcmp() is enough generic, it should be defined in string.c.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20110627072715.6528.10677.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf probe: Rename DIE_FIND_CB_FOUND to DIE_FIND_CB_END
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:09 +0000 (16:27 +0900)]
perf probe: Rename DIE_FIND_CB_FOUND to DIE_FIND_CB_END

Since die_find/walk* callbacks use DIE_FIND_CB_FOUND for
both of failed and found cases, it should be "END"
instead "FOUND" for avoiding confusion.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Link: http://lkml.kernel.org/r/20110627072709.6528.45706.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing/kprobe: Update symbol reference when loading module
Masami Hiramatsu [Mon, 27 Jun 2011 07:27:03 +0000 (16:27 +0900)]
tracing/kprobe: Update symbol reference when loading module

Since the address of a module-local variable can only be
solved after the target module is loaded, the symbol
fetch-argument should be updated when loading target
module.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/20110627072703.6528.75042.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing/kprobes: Support module init function probing
Masami Hiramatsu [Mon, 27 Jun 2011 07:26:56 +0000 (16:26 +0900)]
tracing/kprobes: Support module init function probing

To support probing module init functions, kprobe-tracer allows
user to define a probe on non-existed function when it is given
with a module name. This also enables user to set a probe on
a function on a specific module, even if a same name (but different)
function is locally defined in another module.

The module name must be in the front of function name and separated
by a ':'. e.g. btrfs:btrfs_init_sysfs

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/20110627072656.6528.89970.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agokprobes: Return -ENOENT if probe point doesn't exist
Masami Hiramatsu [Mon, 27 Jun 2011 07:26:50 +0000 (16:26 +0900)]
kprobes: Return -ENOENT if probe point doesn't exist

Return -ENOENT if probe point doesn't exist, but still returns
-EINVAL if both of kprobe->addr and kprobe->symbol_name are
specified or both are not specified.

Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Link: http://lkml.kernel.org/r/20110627072650.6528.67329.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing/kprobes: Merge trace probe enable/disable functions
Masami Hiramatsu [Mon, 27 Jun 2011 07:26:44 +0000 (16:26 +0900)]
tracing/kprobes: Merge trace probe enable/disable functions

Merge redundant enable/disable functions into enable_trace_probe()
and disable_trace_probe().

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: yrl.pp-manager.tt@hitachi.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Link: http://lkml.kernel.org/r/20110627072644.6528.26910.stgit@fedora15
[ converted kprobe selftest to use  enable_trace_probe ]

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Fix regression where ftrace breaks when modules are loaded
Steven Rostedt [Fri, 15 Jul 2011 03:02:27 +0000 (23:02 -0400)]
ftrace: Fix regression where ftrace breaks when modules are loaded

Enabling function tracer to trace all functions, then load a module and
then disable function tracing will cause ftrace to fail.

This can also happen by enabling function tracing on the command line:

  ftrace=function

and during boot up, modules are loaded, then you disable function tracing
with 'echo nop > current_tracer' you will trigger a bug in ftrace that
will shut itself down.

The reason is, the new ftrace code keeps ref counts of all ftrace_ops that
are registered for tracing. When one or more ftrace_ops are registered,
all the records that represent the functions that the ftrace_ops will
trace have a ref count incremented. If this ref count is not zero,
when the code modification runs, that function will be enabled for tracing.
If the ref count is zero, that function will be disabled from tracing.

To make sure the accounting was working, FTRACE_WARN_ON()s were added
to updating of the ref counts.

If the ref count hits its max (> 2^30 ftrace_ops added), or if
the ref count goes below zero, a FTRACE_WARN_ON() is triggered which
disables all modification of code.

Since it is common for ftrace_ops to trace all functions in the kernel,
instead of creating > 20,000 hash items for the ftrace_ops, the hash
count is just set to zero, and it represents that the ftrace_ops is
to trace all functions. This is where the issues arrise.

If you enable function tracing to trace all functions, and then add
a module, the modules function records do not get the ref count updated.
When the function tracer is disabled, all function records ref counts
are subtracted. Since the modules never had their ref counts incremented,
they go below zero and the FTRACE_WARN_ON() is triggered.

The solution to this is rather simple. When modules are loaded, and
their functions are added to the the ftrace pool, look to see if any
ftrace_ops are registered that trace all functions. And for those,
update the ref count for the module function records.

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing/kprobes: Rename probe_* to trace_probe_*
Masami Hiramatsu [Mon, 27 Jun 2011 07:26:36 +0000 (16:26 +0900)]
tracing/kprobes: Rename probe_* to trace_probe_*

Rename probe_* to trace_probe_* for avoiding namespace
confliction. This also fixes improper names of find_probe_event()
and cleanup_all_probes() to find_trace_probe() and
release_all_trace_probes() respectively.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/r/20110627072636.6528.60374.stgit@fedora15
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf, x86: P4 PMU - Introduce event alias feature
Cyrill Gorcunov [Fri, 8 Jul 2011 20:17:12 +0000 (00:17 +0400)]
perf, x86: P4 PMU - Introduce event alias feature

Instead of hw_nmi_watchdog_set_attr() weak function
and appropriate x86_pmu::hw_watchdog_set_attr() call
we introduce even alias mechanism which allow us
to drop this routines completely and isolate quirks
of Netburst architecture inside P4 PMU code only.

The main idea remains the same though -- to allow
nmi-watchdog and perf top run simultaneously.

Note the aliasing mechanism applies to generic
PERF_COUNT_HW_CPU_CYCLES event only because arbitrary
event (say passed as RAW initially) might have some
additional bits set inside ESCR register changing
the behaviour of event and we can't guarantee anymore
that alias event will give the same result.

P.S. Thanks a huge to Don and Steven for for testing
     and early review.

Acked-by: Don Zickus <dzickus@redhat.com>
Tested-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
CC: Ingo Molnar <mingo@elte.hu>
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
CC: Stephane Eranian <eranian@google.com>
CC: Lin Ming <ming.m.lin@intel.com>
CC: Arnaldo Carvalho de Melo <acme@redhat.com>
CC: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/r/20110708201712.GS23657@sun
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing: Have dynamic size event stack traces
Steven Rostedt [Thu, 14 Jul 2011 20:36:53 +0000 (16:36 -0400)]
tracing: Have dynamic size event stack traces

Currently the stack trace per event in ftace is only 8 frames.
This can be quite limiting and sometimes useless. Especially when
the "ignore frames" is wrong and we also use up stack frames for
the event processing itself.

Change this to be dynamic by adding a percpu buffer that we can
write a large stack frame into and then copy into the ring buffer.

For interrupts and NMIs that come in while another event is being
process, will only get to use the 8 frame stack. That should be enough
as the task that it interrupted will have the full stack frame anyway.

Requested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoperf: Robustify proc and debugfs file recording
Sonny Rao [Thu, 14 Jul 2011 03:34:43 +0000 (13:34 +1000)]
perf: Robustify proc and debugfs file recording

While attempting to create a timechart of boot up I found perf didn't
tolerate modules being loaded/unloaded.  This patch fixes this by
reading the file once and then writing the size read at the correct
point in the file.  It also simplifies the code somewhat.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: Sonny Rao <sonnyrao@chromium.org>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Link: http://lkml.kernel.org/r/10011.1310614483@neuling.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Fix dynamic selftest failure on some archs
Steven Rostedt [Wed, 13 Jul 2011 19:11:02 +0000 (15:11 -0400)]
ftrace: Fix dynamic selftest failure on some archs

Archs that do not implement CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST, will
fail the dynamic ftrace selftest.

The function tracer has a quick 'off' variable that will prevent
the call back functions from being called. This variable is called
function_trace_stop. In x86, this is implemented directly in the mcount
assembly, but for other archs, an intermediate function is used called
ftrace_test_stop_func().

In dynamic ftrace, the function pointer variable ftrace_trace_function is
used to update the caller code in the mcount caller. But for archs that
do not have CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST set, it only calls
ftrace_test_stop_func() instead, which in turn calls __ftrace_trace_function.

When more than one ftrace_ops is registered, the function it calls is
ftrace_ops_list_func(), which will iterate over all registered ftrace_ops
and call the callbacks that have their hash matching.

The issue happens when two ftrace_ops are registered for different functions
and one is then unregistered. The __ftrace_trace_function is then pointed
to the remaining ftrace_ops callback function directly. This mean it will
be called for all functions that were registered to trace by both ftrace_ops
that were registered.

This is not an issue for archs with CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST,
because the update of ftrace_trace_function doesn't happen until after all
functions have been updated, and then the mcount caller is updated. But
for those archs that do use the ftrace_test_stop_func(), the update is
immediate.

The dynamic selftest fails because it hits this situation, and the
ftrace_ops that it registers fails to only trace what it was suppose to
and instead traces all other functions.

The solution is to delay the setting of __ftrace_trace_function until
after all the functions have been updated according to the registered
ftrace_ops. Also, function_trace_stop is set during the update to prevent
function tracing from calling code that is caused by the function tracer
itself.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Update filter when tracing enabled in set_ftrace_filter()
Steven Rostedt [Wed, 13 Jul 2011 19:08:31 +0000 (15:08 -0400)]
ftrace: Update filter when tracing enabled in set_ftrace_filter()

Currently, if set_ftrace_filter() is called when the ftrace_ops is
active, the function filters will not be updated. They will only be updated
when tracing is disabled and re-enabled.

Update the functions immediately during set_ftrace_filter().

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Balance records when updating the hash
Steven Rostedt [Wed, 13 Jul 2011 19:03:44 +0000 (15:03 -0400)]
ftrace: Balance records when updating the hash

Whenever the hash of the ftrace_ops is updated, the record counts
must be balance. This requires disabling the records that are set
in the original hash, and then enabling the records that are set
in the updated hash.

Moving the update into ftrace_hash_move() removes the bug where the
hash was updated but the records were not, which results in ftrace
triggering a warning and disabling itself because the ftrace_ops filter
is updated while the ftrace_ops was registered, and then the failure
happens when the ftrace_ops is unregistered.

The current code will not trigger this bug, but new code will.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Fix warning when CONFIG_FUNCTION_TRACER is not defined
Steven Rostedt [Mon, 11 Jul 2011 14:12:59 +0000 (10:12 -0400)]
ftrace: Fix warning when CONFIG_FUNCTION_TRACER is not defined

The struct ftrace_hash was declared within CONFIG_FUNCTION_TRACER
but was referenced outside of it.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoftrace: Do not disable interrupts for modules in mcount update
Steven Rostedt [Sat, 25 Jun 2011 03:28:13 +0000 (23:28 -0400)]
ftrace: Do not disable interrupts for modules in mcount update

When I mounted an NFS directory, it caused several modules to be loaded. At the
time I was running the preemptirqsoff tracer, and it showed the following
output:

# tracer: preemptirqsoff
#
# preemptirqsoff latency trace v1.1.5 on 2.6.33.9-rt30-mrg-test
# --------------------------------------------------------------------
# latency: 1177 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
#    -----------------
#    | task: modprobe-19370 (uid:0 nice:0 policy:0 rt_prio:0)
#    -----------------
#  => started at: ftrace_module_notify
#  => ended at:   ftrace_module_notify
#
#
#                  _------=> CPU#
#                 / _-----=> irqs-off
#                | / _----=> need-resched
#                || / _---=> hardirq/softirq
#                ||| / _--=> preempt-depth
#                |||| /_--=> lock-depth
#                |||||/     delay
#  cmd     pid   |||||| time  |   caller
#     \   /      ||||||   \   |   /
modprobe-19370   3d....    0us!: ftrace_process_locs <-ftrace_module_notify
modprobe-19370   3d.... 1176us : ftrace_process_locs <-ftrace_module_notify
modprobe-19370   3d.... 1178us : trace_hardirqs_on <-ftrace_module_notify
modprobe-19370   3d.... 1178us : <stack trace>
 => ftrace_process_locs
 => ftrace_module_notify
 => notifier_call_chain
 => __blocking_notifier_call_chain
 => blocking_notifier_call_chain
 => sys_init_module
 => system_call_fastpath

That's over 1ms that interrupts are disabled on a Real-Time kernel!

Looking at the cause (being the ftrace author helped), I found that the
interrupts are disabled before the code modification of mcounts into nops. The
interrupts only need to be disabled on start up around this code, not when
modules are being loaded.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing: Still trace filtered irq functions when irq trace is disabled
Steven Rostedt [Tue, 14 Jun 2011 23:02:29 +0000 (19:02 -0400)]
tracing: Still trace filtered irq functions when irq trace is disabled

If a function is set to be traced by the set_graph_function, but the
option funcgraph-irqs is zero, and the traced function happens to be
called from a interrupt, it will not be traced.

The point of funcgraph-irqs is to not trace interrupts when we are
preempted by an irq, not to not trace functions we want to trace that
happen to be *in* a irq.

Luckily the current->trace_recursion element is perfect to add a flag
to help us be able to trace functions within an interrupt even when
we are not tracing interrupts that preempt the trace.

Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing, x86/irq: Do not trace arch_local_{*,irq_*}() functions
Steven Rostedt [Sat, 2 Jul 2011 03:04:36 +0000 (23:04 -0400)]
tracing, x86/irq: Do not trace arch_local_{*,irq_*}() functions

I triggered a triple fault with gcc 4.5.1 because it did not
honor the inline annotation to arch_local_save_flags() function
and that function was added to the pool of functions traced by
the function tracer.

When preempt_schedule() called arch_local_save_flags() (called
by irqs_disabled()), it was traced, but the first thing the
function tracer does is disable preemption. When it enables
preemption, the NEED_RESCHED flag will not have been cleared and
the preemption check will trigger the call to preempt_schedule()
again.

Although the dynamic function tracer crashed immediately, the
static version of the function tracer (CONFIG_DYNAMIC_FTRACE is
not set) actually was able to show where the problem was.

 swapper-1       3.N.. 103885us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103886us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103886us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103887us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103887us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103888us : arch_local_save_flags <-preempt_schedule
 swapper-1       3.N.. 103888us : arch_local_save_flags <-preempt_schedule

It went on for a while before it triple faulted with a corrupted
stack.

The arch_local_save_flags and arch_local_irq_* functions should
not be traced. Even though they are marked as inline, gcc may
still make them a function and enable tracing of them.

The simple solution is to just mark them as notrace. I had to
add the <linux/types.h> for this file to include the notrace
tag.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20110702033852.733414762@goodmis.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoftrace: Fix regression of :mod:module function enabling
Steven Rostedt [Thu, 7 Jul 2011 15:09:22 +0000 (11:09 -0400)]
ftrace: Fix regression of :mod:module function enabling

The new code that allows different utilities to pick and choose
what functions they trace broke the :mod: hook that allows users
to trace only functions of a particular module.

The reason is that the :mod: hook bypasses the hash that is setup
to allow individual users to trace their own functions and uses
the global hash directly. But if the global hash has not been
set up, it will cause a bug:

echo '*:mod:radeon' > /sys/kernel/debug/set_ftrace_filter

produces:

 [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
 [drm:radeon_crtc_page_flip] *ERROR* failed to reserve new rbo buffer before flip
 BUG: unable to handle kernel paging request at ffffffff8160ec90
 IP: [<ffffffff810d9136>] add_hash_entry+0x66/0xd0
 PGD 1a05067 PUD 1a09063 PMD 80000000016001e1
 Oops: 0003 [#1] SMP Jul  7 04:02:28 phyllis kernel: [55303.858604] CPU 1
 Modules linked in: cryptd aes_x86_64 aes_generic binfmt_misc rfcomm bnep ip6table_filter hid radeon r8169 ahci libahci mii ttm drm_kms_helper drm video i2c_algo_bit intel_agp intel_gtt

 Pid: 10344, comm: bash Tainted: G        WC  3.0.0-rc5 #1 Dell Inc. Inspiron N5010/0YXXJJ
 RIP: 0010:[<ffffffff810d9136>]  [<ffffffff810d9136>] add_hash_entry+0x66/0xd0
 RSP: 0018:ffff88003a96bda8  EFLAGS: 00010246
 RAX: ffff8801301735c0 RBX: ffffffff8160ec80 RCX: 0000000000306ee0
 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff880137c92940
 RBP: ffff88003a96bdb8 R08: ffff880137c95680 R09: 0000000000000000
 R10: 0000000000000001 R11: 0000000000000000 R12: ffffffff81c9df78
 R13: ffff8801153d1000 R14: 0000000000000000 R15: 0000000000000000
 FS: 00007f329c18a700(0000) GS:ffff880137c80000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: ffffffff8160ec90 CR3: 000000003002b000 CR4: 00000000000006e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
 Process bash (pid: 10344, threadinfo ffff88003a96a000, task ffff88012fcfc470)
 Stack:
  0000000000000fd0 00000000000000fc ffff88003a96be38 ffffffff810d92f5
  ffff88011c4c4e00 ffff880000000000 000000000b69f4d0 ffffffff8160ec80
  ffff8800300e6f06 0000000081130295 0000000000000282 ffff8800300e6f00
 Call Trace:
  [<ffffffff810d92f5>] match_records+0x155/0x1b0
  [<ffffffff810d940c>] ftrace_mod_callback+0xbc/0x100
  [<ffffffff810dafdf>] ftrace_regex_write+0x16f/0x210
  [<ffffffff810db09f>] ftrace_filter_write+0xf/0x20
  [<ffffffff81166e48>] vfs_write+0xc8/0x190
  [<ffffffff81167001>] sys_write+0x51/0x90
  [<ffffffff815c7e02>] system_call_fastpath+0x16/0x1b
 Code: 48 8b 33 31 d2 48 85 f6 75 33 49 89 d4 4c 03 63 08 49 8b 14 24 48 85 d2 48 89 10 74 04 48 89 42 08 49 89 04 24 4c 89 60 08 31 d2
 RIP [<ffffffff810d9136>] add_hash_entry+0x66/0xd0
  RSP <ffff88003a96bda8>
 CR2: ffffffff8160ec90
 ---[ end trace a5d031828efdd88e ]---

Reported-by: Brian Marete <marete@toshnix.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing: Have "enable" file use refcounts like the "filter" file
Steven Rostedt [Tue, 5 Jul 2011 18:32:51 +0000 (14:32 -0400)]
tracing: Have "enable" file use refcounts like the "filter" file

The "enable" file for the event system can be removed when a module
is unloaded and the event system only has events from that module.
As the event system nr_events count goes to zero, it may be freed
if its ref_count is also set to zero.

Like the "filter" file, the "enable" file may be opened by a task and
referenced later, after a module has been unloaded and the events for
that event system have been removed.

Although the "filter" file referenced the event system structure,
the "enable" file only references a pointer to the event system
name. Since the name is freed when the event system is removed,
it is possible that an access to the "enable" file may reference
a freed pointer.

Update the "enable" file to use the subsystem_open() routine that
the "filter" file uses, to keep a reference to the event system
structure while the "enable" file is opened.

Cc: <stable@kernel.org>
Reported-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agotracing: Fix bug when reading system filters on module removal
Steven Rostedt [Tue, 5 Jul 2011 15:36:06 +0000 (11:36 -0400)]
tracing: Fix bug when reading system filters on module removal

The event system is freed when its nr_events is set to zero. This happens
when a module created an event system and then later the module is
removed. Modules may share systems, so the system is allocated when
it is created and freed when the modules are unloaded and all the
events under the system are removed (nr_events set to zero).

The problem arises when a task opened the "filter" file for the
system. If the module is unloaded and it removed the last event for
that system, the system structure is freed. If the task that opened
the filter file accesses the "filter" file after the system has
been freed, the system will access an invalid pointer.

By adding a ref_count, and using it to keep track of what
is using the event system, we can free it after all users
are finished with the event system.

Cc: <stable@kernel.org>
Reported-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoMerge branch 'tip/perf/core-2' of git://git.kernel.org/pub/scm/linux/kernel/git/roste...
Ingo Molnar [Tue, 5 Jul 2011 09:55:43 +0000 (11:55 +0200)]
Merge branch 'tip/perf/core-2' of git://git./linux/kernel/git/rostedt/linux-2.6-trace into perf/core

13 years agoperf report/annotate/script: Add option to specify a CPU range
Anton Blanchard [Mon, 4 Jul 2011 11:57:50 +0000 (21:57 +1000)]
perf report/annotate/script: Add option to specify a CPU range

Add an option to perf report/annotate/script to specify which
CPUs to operate on. This enables us to take a single system wide
profile and analyse each CPU (or group of CPUs) in isolation.

This was useful when profiling a multiprocess workload where the
bottleneck was on one CPU but this was hidden in the overall
profile. Per process and per thread breakdowns didn't help
because multiple processes were running on each CPU and no
single process consumed an entire CPU.

The patch converts the list of CPUs returned by cpu_map__new
into a bitmap for fast lookup. I wanted to use -C to be
consistent with perf top/record/stat, but unfortunately perf
report already uses -C <comms>.

 v2: Incorporate suggestions from David Ahern:
- Added -c to perf script
- Check that SAMPLE_CPU is set when -c is used
- Update documentation

 v3: Create perf_session__cpu_bitmap()

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Link: http://lkml.kernel.org/r/20110704215750.11647eb9@kryten
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoMerge branch 'core' of git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile...
Ingo Molnar [Mon, 4 Jul 2011 10:27:28 +0000 (12:27 +0200)]
Merge branch 'core' of git://git./linux/kernel/git/rric/oprofile into perf/core

13 years agoMerge branch 'perf/stacktrace' of git://git.kernel.org/pub/scm/linux/kernel/git/frede...
Ingo Molnar [Sun, 3 Jul 2011 18:39:40 +0000 (20:39 +0200)]
Merge branch 'perf/stacktrace' of git://git./linux/kernel/git/frederic/random-tracing into perf/core

13 years agox86: Don't use frame pointer to save old stack on irq entry
Frederic Weisbecker [Sat, 2 Jul 2011 14:52:45 +0000 (16:52 +0200)]
x86: Don't use frame pointer to save old stack on irq entry

rbp is used in SAVE_ARGS_IRQ to save the old stack pointer
in order to restore it later in ret_from_intr.

It is convenient because we save its value in the irq regs
and it's easily restored using the leave instruction.

However this is a kind of abuse of the frame pointer which
role is to help unwinding the kernel by chaining frames
together, each node following the return address to the
previous frame.

But although we are breaking the frame by changing the stack
pointer, there is no preceding return address before the new
frame. Hence using the frame pointer to link the two stacks
breaks the stack unwinders that find a random value instead of
a return address here.

There is no workaround that can work in every case. We are using
the fixup_bp_irq_link() function to dereference that abused frame
pointer in the case of non nesting interrupt (which means stack
changed).
But that doesn't fix the case of interrupts that don't change the
stack (but we still have the unconditional frame link), which is
the case of hardirq interrupting softirq. We have no way to detect
this transition so the frame irq link is considered as a real frame
pointer and the return address is dereferenced but it is still a
spurious one.

There are two possible results of this: either the spurious return
address, a random stack value, luckily belongs to the kernel text
and then the unwinding can continue and we just have a weird entry
in the stack trace. Or it doesn't belong to the kernel text and
unwinding stops there.

This is the reason why stacktraces (including perf callchains) on
irqs that interrupted softirqs don't work very well.

To solve this, we don't save the old stack pointer on rbp anymore
but we save it to a scratch register that we push on the new
stack and that we pop back later on irq return.

This preserves the whole frame chain without spurious return addresses
in the middle and drops the need for the horrid fixup_bp_irq_link()
workaround.

And finally irqs that interrupt softirq are sanely unwinded.

Before:

    99.81%         perf  [kernel.kallsyms]  [k] perf_pending_event
                   |
                   --- perf_pending_event
                       irq_work_run
                       smp_irq_work_interrupt
                       irq_work_interrupt
                      |
                      |--41.60%-- __read
                      |          |
                      |          |--99.90%-- create_worker
                      |          |          bench_sched_messaging
                      |          |          cmd_bench
                      |          |          run_builtin
                      |          |          main
                      |          |          __libc_start_main
                      |           --0.10%-- [...]

After:

     1.64%  swapper  [kernel.kallsyms]  [k] perf_pending_event
            |
            --- perf_pending_event
                irq_work_run
                smp_irq_work_interrupt
                irq_work_interrupt
               |
               |--95.00%-- arch_irq_work_raise
               |          irq_work_queue
               |          __perf_event_overflow
               |          perf_swevent_overflow
               |          perf_swevent_event
               |          perf_tp_event
               |          perf_trace_softirq
               |          __do_softirq
               |          call_softirq
               |          do_softirq
               |          irq_exit
               |          |
               |          |--73.68%-- smp_apic_timer_interrupt
               |          |          apic_timer_interrupt
               |          |          |
               |          |          |--96.43%-- amd_e400_idle
               |          |          |          cpu_idle
               |          |          |          start_secondary

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jan Beulich <JBeulich@novell.com>
13 years agox86: Remove useless unwinder backlink from irq regs saving
Frederic Weisbecker [Sat, 2 Jul 2011 13:03:44 +0000 (15:03 +0200)]
x86: Remove useless unwinder backlink from irq regs saving

The unwinder backlink in interrupt entry is very useless.
It's actually not part of the stack frame chain and thus is
never used.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jan Beulich <JBeulich@novell.com>
13 years agox86,64: Separate arg1 from rbp handling in SAVE_REGS_IRQ
Frederic Weisbecker [Fri, 1 Jul 2011 00:25:17 +0000 (02:25 +0200)]
x86,64: Separate arg1 from rbp handling in SAVE_REGS_IRQ

Just for clarity in the code. Have a first block that handles
the frame pointer and a separate one that handles pt_regs
pointer and its use.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jan Beulich <JBeulich@novell.com>
13 years agox86,64: Simplify save_regs()
Frederic Weisbecker [Thu, 30 Jun 2011 23:51:22 +0000 (01:51 +0200)]
x86,64: Simplify save_regs()

The save_regs function that saves the regs on low level
irq entry is complicated because of the fact it changes
its stack in the middle and also because it manipulates
data allocated in the caller frame and accesses there
are directly calculated from callee rsp value with the
return address in the middle of the way.

This complicates the static stack offsets calculation and
require more dynamic ones. It also needs a save/restore
of the function's return address.

To simplify and optimize this, turn save_regs() into a
macro.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jan Beulich <JBeulich@novell.com>
13 years agox86: Fetch stack from regs when possible in dump_trace()
Frederic Weisbecker [Thu, 30 Jun 2011 17:04:56 +0000 (19:04 +0200)]
x86: Fetch stack from regs when possible in dump_trace()

When regs are passed to dump_stack(), we fetch the frame
pointer from the regs but the stack pointer is taken from
the current frame.

Thus the frame and stack pointers may not come from the same
context. For example this can result in the unwinder to
think the context is in irq, due to the current value of
the stack, but the frame pointer coming from the regs points
to a frame from another place. It then tries to fix up
the irq link but ends up dereferencing a random frame
pointer that doesn't belong to the irq stack:

[ 9131.706906] ------------[ cut here ]------------
[ 9131.707003] WARNING: at arch/x86/kernel/dumpstack_64.c:129 dump_trace+0x2aa/0x330()
[ 9131.707003] Hardware name: AMD690VM-FMH
[ 9131.707003] Perf: bad frame pointer = 0000000000000005 in callchain
[ 9131.707003] Modules linked in:
[ 9131.707003] Pid: 1050, comm: perf Not tainted 3.0.0-rc3+ #181
[ 9131.707003] Call Trace:
[ 9131.707003]  <IRQ>  [<ffffffff8104bd4a>] warn_slowpath_common+0x7a/0xb0
[ 9131.707003]  [<ffffffff8104be21>] warn_slowpath_fmt+0x41/0x50
[ 9131.707003]  [<ffffffff8178b873>] ? bad_to_user+0x6d/0x10be
[ 9131.707003]  [<ffffffff8100c2da>] dump_trace+0x2aa/0x330
[ 9131.707003]  [<ffffffff810107d3>] ? native_sched_clock+0x13/0x50
[ 9131.707003]  [<ffffffff8101b164>] perf_callchain_kernel+0x54/0x70
[ 9131.707003]  [<ffffffff810d391f>] perf_prepare_sample+0x19f/0x2a0
[ 9131.707003]  [<ffffffff810d546c>] __perf_event_overflow+0x16c/0x290
[ 9131.707003]  [<ffffffff810d5430>] ? __perf_event_overflow+0x130/0x290
[ 9131.707003]  [<ffffffff810107d3>] ? native_sched_clock+0x13/0x50
[ 9131.707003]  [<ffffffff8100fbb9>] ? sched_clock+0x9/0x10
[ 9131.707003]  [<ffffffff810752e5>] ? T.375+0x15/0x90
[ 9131.707003]  [<ffffffff81084da4>] ? trace_hardirqs_on_caller+0x64/0x180
[ 9131.707003]  [<ffffffff810817bd>] ? trace_hardirqs_off+0xd/0x10
[ 9131.707003]  [<ffffffff810d5764>] perf_event_overflow+0x14/0x20
[ 9131.707003]  [<ffffffff810d588c>] perf_swevent_hrtimer+0x11c/0x130
[ 9131.707003]  [<ffffffff817821a1>] ? error_exit+0x51/0xb0
[ 9131.707003]  [<ffffffff81072e93>] __run_hrtimer+0x83/0x1e0
[ 9131.707003]  [<ffffffff810d5770>] ? perf_event_overflow+0x20/0x20
[ 9131.707003]  [<ffffffff81073256>] hrtimer_interrupt+0x106/0x250
[ 9131.707003]  [<ffffffff812a3bfd>] ? trace_hardirqs_off_thunk+0x3a/0x3c
[ 9131.707003]  [<ffffffff81024833>] smp_apic_timer_interrupt+0x53/0x90
[ 9131.707003]  [<ffffffff81789053>] apic_timer_interrupt+0x13/0x20
[ 9131.707003]  <EOI>  [<ffffffff817821a1>] ? error_exit+0x51/0xb0
[ 9131.707003]  [<ffffffff8178219c>] ? error_exit+0x4c/0xb0
[ 9131.707003] ---[ end trace b2560d4876709347 ]---

Fix this by simply taking the stack pointer from regs->sp
when regs are provided.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agox86: Save stack pointer in perf live regs savings
Frederic Weisbecker [Sat, 2 Jul 2011 13:00:52 +0000 (15:00 +0200)]
x86: Save stack pointer in perf live regs savings

In order to prepare for fetching the stack pointer from the
regs when possible in dump_trace() instead of taking the
local one, save the current stack pointer in perf live regs saving.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
13 years agoperf, powerpc: Fix build borkage
Peter Zijlstra [Fri, 1 Jul 2011 13:24:53 +0000 (15:24 +0200)]
perf, powerpc: Fix build borkage

The patch a8b0ca17b80e ("perf: Remove the nmi parameter from the swevent
and overflow interface") missed a spot in the ppc hw_breakpoint code,
fix this up so things compile again.

Reported-by: Ingo Molnar <mingo@elte.hu>
Cc: Anton Blanchard <anton@samba.org>
Cc: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-09pfip95g88s70iwkxu6nnbt@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf stat: Add noise output for csv mode
Zhengyu He [Thu, 23 Jun 2011 20:45:42 +0000 (13:45 -0700)]
perf stat: Add noise output for csv mode

Previously, when you want perf-stat to output the statistics in
csv mode, no information of the noise will be printed out.

For example right now we output this --repeat information:

 ./perf stat -r3 -x, sleep 1
 1.164789,task-clock
 8,context-switches
 0,CPU-migrations
 219,page-faults
 3337800,cycles

With this patch, the output will be appended with an additional
entry for the noise value:

 ./perf stat -r3 -x, sleep 1
 1.164789,task-clock,3.75%
 8,context-switches,75.00%
 0,CPU-migrations,100.00%
 219,page-faults,0.00%
 3337800,cycles,3.36%

Signed-off-by: Zhengyu He <zhengyuh@google.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Stephane Eranian <eranian@google.com>
Cc: Venkatesh Pallipadi <venki@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/r/1308861942-4945-1-git-send-email-zhengyuh@google.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoMerge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
Ingo Molnar [Fri, 1 Jul 2011 09:51:58 +0000 (11:51 +0200)]
Merge branch 'perf/core' of git://git./linux/kernel/git/frederic/random-tracing into perf/core

13 years agoperf: export perf_event_refresh() to modules
Avi Kivity [Wed, 29 Jun 2011 15:42:37 +0000 (18:42 +0300)]
perf: export perf_event_refresh() to modules

KVM needs one-shot samples, since a PMC programmed to -X will fire after X
events and then again after 2^40 events (i.e. variable period).

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1309362157-6596-4-git-send-email-avi@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agox86, perf: Add constraints for architectural PMU
Avi Kivity [Wed, 29 Jun 2011 15:42:36 +0000 (18:42 +0300)]
x86, perf: Add constraints for architectural PMU

The v1 PMU does not have any fixed counters.  Using the v2 constraints,
which do have fixed counters, causes an additional choice to be present
in the weight calculation, but not when actually scheduling the event,
leading to an event being not scheduled at all.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1309362157-6596-3-git-send-email-avi@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf: Add context field to perf_event
Avi Kivity [Wed, 29 Jun 2011 15:42:35 +0000 (18:42 +0300)]
perf: Add context field to perf_event

The perf_event overflow handler does not receive any caller-derived
argument, so many callers need to resort to looking up the perf_event
in their local data structure.  This is ugly and doesn't scale if a
single callback services many perf_events.

Fix by adding a context parameter to perf_event_create_kernel_counter()
(and derived hardware breakpoints APIs) and storing it in the perf_event.
The field can be accessed from the callback as event->overflow_handler_context.
All callers are updated.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1309362157-6596-2-git-send-email-avi@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf, arch: Add generic NODE cache events
Peter Zijlstra [Fri, 22 Apr 2011 21:37:06 +0000 (23:37 +0200)]
perf, arch: Add generic NODE cache events

Add a NODE level to the generic cache events which is used to measure
local vs remote memory accesses. Like all other cache events, an
ACCESS is HIT+MISS, if there is no way to distinguish between reads
and writes do reads only etc..

The below needs filling out for !x86 (which I filled out with
unsupported events).

I'm fairly sure ARM can leave it like that since it doesn't strike me as
an architecture that even has NUMA support. SH might have something since
it does appear to have some NUMA bits.

Sparc64, PowerPC and MIPS certainly want a good look there since they
clearly are NUMA capable.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: David Miller <davem@davemloft.net>
Cc: Anton Blanchard <anton@samba.org>
Cc: David Daney <ddaney@caviumnetworks.com>
Cc: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1303508226.4865.8.camel@laptop
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf, intel: Try alternative OFFCORE encodings
Peter Zijlstra [Mon, 23 May 2011 09:08:15 +0000 (11:08 +0200)]
perf, intel: Try alternative OFFCORE encodings

Since the OFFCORE registers are fully symmetric, try the other one
when the specified one is already in use.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1306141897.18455.8.camel@twins
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf_events: Add Intel Sandy Bridge offcore_response low-level support
Stephane Eranian [Mon, 6 Jun 2011 14:57:12 +0000 (16:57 +0200)]
perf_events: Add Intel Sandy Bridge offcore_response low-level support

This patch adds Intel Sandy Bridge offcore_response support by
providing the low-level constraint table for those events.

On Sandy Bridge, there are two offcore_response events. Each uses
its own dedictated extra register. But those registers are NOT shared
between sibling CPUs when HT is on unlike Nehalem/Westmere. They are
always private to each CPU. But they still need to be controlled within
an event group. All events within an event group must use the same
value for the extra MSR. That's not controlled by the second patch in
this series.

Furthermore on Sandy Bridge, the offcore_response events have NO
counter constraints contrary to what the official documentation
indicates, so drop the events from the contraint table.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110606145712.GA7304@quad
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf_events: Fix validation of events using an extra reg
Stephane Eranian [Mon, 6 Jun 2011 14:57:08 +0000 (16:57 +0200)]
perf_events: Fix validation of events using an extra reg

The validate_group() function needs to validate events with
extra shared regs. Within an event group, only events with
the same value for the extra reg can co-exist. This was not
checked by validate_group() because it was missing the
shared_regs logic.

This patch changes the allocation of the fake cpuc used for
validation to also point to a fake shared_regs structure such
that group events be properly testing.

It modifies __intel_shared_reg_get_constraints() to use
spin_lock_irqsave() to avoid lockdep issues.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110606145708.GA7279@quad
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf_events: Update Intel extra regs shared constraints management
Stephane Eranian [Mon, 6 Jun 2011 14:57:03 +0000 (16:57 +0200)]
perf_events: Update Intel extra regs shared constraints management

This patch improves the code managing the extra shared registers
used for offcore_response events on Intel Nehalem/Westmere. The
idea is to use static allocation instead of dynamic allocation.
This simplifies greatly the get and put constraint routines for
those events.

The patch also renames per_core to shared_regs because the same
data structure gets used whether or not HT is on. When HT is
off, those events still need to coordination because they use
a extra MSR that has to be shared within an event group.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110606145703.GA7258@quad
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf: Remove the perf_output_begin(.sample) argument
Peter Zijlstra [Mon, 27 Jun 2011 14:47:16 +0000 (16:47 +0200)]
perf: Remove the perf_output_begin(.sample) argument

Since only samples call perf_output_sample() its much saner (and more
correct) to put the sample logic in there than in the
perf_output_begin()/perf_output_end() pair.

Saves a useless argument, reduces conditionals and shrinks
struct perf_output_handle, win!

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-2crpvsx3cqu67q3zqjbnlpsc@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf: Remove the nmi parameter from the swevent and overflow interface
Peter Zijlstra [Mon, 27 Jun 2011 12:41:57 +0000 (14:41 +0200)]
perf: Remove the nmi parameter from the swevent and overflow interface

The nmi parameter indicated if we could do wakeups from the current
context, if not, we would set some state and self-IPI and let the
resulting interrupt do the wakeup.

For the various event classes:

  - hardware: nmi=0; PMI is in fact an NMI or we run irq_work_run from
    the PMI-tail (ARM etc.)
  - tracepoint: nmi=0; since tracepoint could be from NMI context.
  - software: nmi=[0,1]; some, like the schedule thing cannot
    perform wakeups, and hence need 0.

As one can see, there is very little nmi=1 usage, and the down-side of
not using it is that on some platforms some software events can have a
jiffy delay in wakeup (when arch_irq_work_raise isn't implemented).

The up-side however is that we can remove the nmi parameter and save a
bunch of conditionals in fast paths.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Michael Cree <mcree@orcon.net.nz>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jason Wessel <jason.wessel@windriver.com>
Cc: Don Zickus <dzickus@redhat.com>
Link: http://lkml.kernel.org/n/tip-agjev8eu666tvknpb3iaj0fg@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf, x86: Add hw_watchdog_set_attr() in a sake of nmi-watchdog on P4
Cyrill Gorcunov [Thu, 23 Jun 2011 12:49:18 +0000 (16:49 +0400)]
perf, x86: Add hw_watchdog_set_attr() in a sake of nmi-watchdog on P4

Due to restriction and specifics of Netburst PMU we need a separated
event for NMI watchdog. In particular every Netburst event
consumes not just a counter and a config register, but also an
additional ESCR register.

Since ESCR registers are grouped upon counters (i.e. if ESCR is occupied
for some event there is no room for another event to enter until its
released) we need to pick up the "least" used ESCR (or the most available
one) for nmi-watchdog purposes -- so MSR_P4_CRU_ESCR2/3 was chosen.

With this patch nmi-watchdog and perf top should be able to run simultaneously.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
CC: Lin Ming <ming.m.lin@intel.com>
CC: Arnaldo Carvalho de Melo <acme@redhat.com>
CC: Frederic Weisbecker <fweisbec@gmail.com>
Tested-and-reviewed-by: Don Zickus <dzickus@redhat.com>
Tested-and-reviewed-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110623124918.GC13050@sun
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoevents: Ensure that timers are updated without requiring read() call
Eric B Munson [Fri, 24 Jun 2011 16:26:26 +0000 (12:26 -0400)]
events: Ensure that timers are updated without requiring read() call

The event tracing infrastructure exposes two timers which should be updated
each time the value of the counter is updated.  Currently, these counters are
only updated when userspace calls read() on the fd associated with an event.
This means that counters which are read via the mmap'd page exclusively never
have their timers updated.  This patch adds ensures that the timers are updated
each time the values in the mmap'd page are updated.

Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1308932786-5111-1-git-send-email-emunson@mgebm.net
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoevents: Move lockless timer calculation into helper function
Eric B Munson [Thu, 23 Jun 2011 20:34:38 +0000 (16:34 -0400)]
events: Move lockless timer calculation into helper function

Take the timer calculation from perf_output_read and move it to a helper
function for any place that needs timer values but cannot take the ctx->lock.

Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1308861279-15216-2-git-send-email-emunson@mgebm.net
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoevents: Add note to update_event_times comment about holding ctx->lock
Eric B Munson [Thu, 23 Jun 2011 20:34:37 +0000 (16:34 -0400)]
events: Add note to update_event_times comment about holding ctx->lock

Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1308861279-15216-1-git-send-email-emunson@mgebm.net
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf: Remove 64-bit alignment padding from perf_event_context
Richard Kennedy [Tue, 7 Jun 2011 15:33:38 +0000 (16:33 +0100)]
perf: Remove 64-bit alignment padding from perf_event_context

Reorder perf_event_context to remove 8 bytes of 64 bit alignment padding
shrinking its size to 192 bytes, allowing it to fit into a smaller slab
and use one fewer cache lines.

Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1307460819.1950.5.camel@castor.rsk
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf_events: Fix perf buffer watermark setting
Vince Weaver [Wed, 1 Jun 2011 19:15:36 +0000 (15:15 -0400)]
perf_events: Fix perf buffer watermark setting

Since 2.6.36 (specifically commit d57e34fdd60b ("perf: Simplify the
ring-buffer logic: make perf_buffer_alloc() do everything needed"),
the perf_buffer_init_code() has been mis-setting the buffer watermark
if perf_event_attr.wakeup_events has a non-zero value.

This is because perf_event_attr.wakeup_events is a union with
perf_event_attr.wakeup_watermark.

This commit re-enables the check for perf_event_attr.watermark being
set before continuing with setting a non-default watermark.

This bug is most noticable when you are trying to use PERF_IOC_REFRESH
with a value larger than one and perf_event_attr.wakeup_events is set to
one.  In this case the buffer watermark will be set to 1 and you will
get extraneous POLL_IN overflows rather than POLL_HUP as expected.

[ avoid using attr.wakeup_events when attr.watermark is set ]

Signed-off-by: Vince Weaver <vweaver1@eecs.utk.edu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org>
Link: http://lkml.kernel.org/r/alpine.DEB.2.00.1106011506390.5384@cl320.eecs.utk.edu
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoirq_work, alpha: Fix up arch hooks
Peter Zijlstra [Tue, 28 Jun 2011 10:15:51 +0000 (12:15 +0200)]
irq_work, alpha: Fix up arch hooks

Commit e360adbe29 ("irq_work: Add generic hardirq context
callbacks") fouled up the Alpha bit, not properly naming the
arch specific function that raises the 'self-IPI'.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Michael Cree <mcree@orcon.net.nz>
Cc: stable@kernel.org # 37+
Link: http://lkml.kernel.org/n/tip-gukh0txmql2l4thgrekzzbfy@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoirq_work, ppc: Fix up arch hooks
Peter Zijlstra [Mon, 27 Jun 2011 15:22:43 +0000 (17:22 +0200)]
irq_work, ppc: Fix up arch hooks

Commit e360adbe29 ("irq_work: Add generic hardirq context
callbacks") fouled up the ppc bit, not properly naming the
arch specific function that raises the 'self-IPI'.

Cc: Huang Ying <ying.huang@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: stable@kernel.org # 37+
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-eg0aqien8p1aqvzu9dft6dtv@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoMerge commit 'v3.0-rc5' into perf/core
Ingo Molnar [Fri, 1 Jul 2011 08:28:42 +0000 (10:28 +0200)]
Merge commit 'v3.0-rc5' into perf/core

Merge reason: Pick up the latest fixes.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
13 years agoperf tools: Only display parent field if explictly sorted
Frederic Weisbecker [Wed, 29 Jun 2011 21:52:52 +0000 (23:52 +0200)]
perf tools: Only display parent field if explictly sorted

We don't need to display the parent field if the parent
sorting machinery is only used for parent filtering
(as in "-p foo").

However if parent filtering is used in combination with
explicit parent sorting ( -s parent), we want to
display it.

Result with:

  perf report -p kernel_thread -s parent

Before:

 # Overhead  Parent symbol
 # ........  .............
 #
     0.07%
            |
            --- ioread8
                ata_sff_check_status
                ata_sff_tf_load
                ata_sff_qc_issue
                ata_bmdma_qc_issue
                ata_qc_issue
                ata_scsi_translate
                ata_scsi_queuecmd
                scsi_dispatch_cmd
                scsi_request_fn
                __blk_run_queue
                __make_request
                generic_make_request
                submit_bio
                submit_bh
                journal_submit_commit_record
                jbd2_journal_commit_transaction
                kjournald2
                kthread
                kernel_thread_helpe

After:

 # Overhead  Parent symbol
 # ........  .............
 #
     0.07%  kernel_thread_helper
            |
            --- ioread8
                ata_sff_check_status
                ata_sff_tf_load
                ata_sff_qc_issue
                ata_bmdma_qc_issue
                ata_qc_issue
                ata_scsi_translate
                ata_scsi_queuecmd
                scsi_dispatch_cmd
                scsi_request_fn
                __blk_run_queue
                __make_request
                generic_make_request
                submit_bio
                submit_bh
                journal_submit_commit_record
                jbd2_journal_commit_transaction
                kjournald2
                kthread
                kernel_thread_helper

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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Sam Liao <phyomh@gmail.com>
13 years agoperf tools: Allow sort dimensions to be registered more than once
Frederic Weisbecker [Wed, 29 Jun 2011 21:08:14 +0000 (23:08 +0200)]
perf tools: Allow sort dimensions to be registered more than once

So that the parent sort dimension can be registered twice: once
if we add it as an explicit sort dimension (-s parent) and twice
if we request a parent filter (-p foo).

We'll have only one parent sort dimension in the end but this
allows to override the default parent filter with we gave in "-p"
option. The goal of this is to prepare to allow the use of
"-s parent" and "-p foo" at the same time, ie: sort by filtered
parent.

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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Sam Liao <phyomh@gmail.com>
13 years agoperf tools: Don't display ignored entries on stdio ui
Frederic Weisbecker [Wed, 29 Jun 2011 20:23:03 +0000 (22:23 +0200)]
perf tools: Don't display ignored entries on stdio ui

As for newt ui, don't display entries that have been marked
as ignored.

The practical current effect of this is to make parent
filtering really working. Before, entries that were ignored
were given a null parent but were still displayed. This
resulted in some weird effects:

 # Overhead      Command      Shared Object        Symbol
 # ........  ...........  .................  ............
 #
^A
                   |
                   --- __lock_acquire
                      |
                      |--95.97%-- lock_acquire
                      |          |
                      |          |--30.75%-- _raw_spin_lock

Discard these from the stdio display.

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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Sam Liao <phyomh@gmail.com>
13 years agoperf tools: Remove sort print helpers declarations
Frederic Weisbecker [Wed, 29 Jun 2011 01:25:14 +0000 (03:25 +0200)]
perf tools: Remove sort print helpers declarations

These are probably some old leftovers.

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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Sam Liao <phyomh@gmail.com>
13 years agoperf tools: Make sort operations static
Frederic Weisbecker [Wed, 29 Jun 2011 01:14:52 +0000 (03:14 +0200)]
perf tools: Make sort operations static

These don't need to be globally visible.

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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Sam Liao <phyomh@gmail.com>
13 years agoperf tools: Add inverted call graph report support.
Sam Liao [Tue, 7 Jun 2011 15:49:46 +0000 (23:49 +0800)]
perf tools: Add inverted call graph report support.

Add "caller/callee" option to support inverted butterfly report,
in the inverted report (with caller option), the call graph start
from the callee's ancestor. Users can use such view to catch system's
performance bottleneck from a sysprof like view. Using this option
with specified sort order like pid gives us high level view of call
graph statistics.

Also add "-G" alias for inverted call graph.

Signed-off-by: Sam Liao <phyomh@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: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
13 years agojump_label: Fix jump_label update for modules
Xiao Guangrong [Tue, 21 Jun 2011 02:35:55 +0000 (10:35 +0800)]
jump_label: Fix jump_label update for modules

The jump labels entries for modules do not stop at __stop__jump_table,
but after mod->jump_entries + mod_num_jump_entries.

By checking the wrong end point, module trace events never get enabled.

Cc: Ingo Molnar <mingo@elte.hu>
Acked-by: Jason Baron <jbaron@redhat.com>
Tested-by: Avi Kivity <avi@redhat.com>
Tested-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Link: http://lkml.kernel.org/r/4E00038B.2060404@cn.fujitsu.com
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
13 years agoLinux 3.0-rc5
Linus Torvalds [Tue, 28 Jun 2011 02:12:22 +0000 (19:12 -0700)]
Linux 3.0-rc5

13 years agodrm/i915: more struct_mutex locking
Hugh Dickins [Mon, 27 Jun 2011 23:18:20 +0000 (16:18 -0700)]
drm/i915: more struct_mutex locking

When auditing the locking in i915_gem.c (for a prospective change which
I then abandoned), I noticed two places where struct_mutex is not held
across GEM object manipulations that would usually require it.

Since one is in initial setup and the other in driver unload, I'm
guessing the mutex is not required for either; but post a patch in case
it is.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Keith Packard <keithp@keithp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrm/i915: use shmem_truncate_range
Hugh Dickins [Mon, 27 Jun 2011 23:18:19 +0000 (16:18 -0700)]
drm/i915: use shmem_truncate_range

The interface to ->truncate_range is changing very slightly: once "tmpfs:
take control of its truncate_range" has been applied, this can be applied.
 For now there is only a slight inefficiency while this remains unapplied,
but it will soon become essential for managing shmem's use of swap.

Change i915_gem_object_truncate() to use shmem_truncate_range() directly:
which should also spare i915 later change if we switch from
inode_operations->truncate_range to file_operations->fallocate.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Keith Packard <keithp@keithp.com>
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrm/i915: use shmem_read_mapping_page
Hugh Dickins [Mon, 27 Jun 2011 23:18:18 +0000 (16:18 -0700)]
drm/i915: use shmem_read_mapping_page

Soon tmpfs will stop supporting ->readpage and read_cache_page_gfp(): once
"tmpfs: add shmem_read_mapping_page_gfp" has been applied, this patch can
be applied to ease the transition.

Make i915_gem_object_get_pages_gtt() use shmem_read_mapping_page_gfp() in
the one place it's needed; elsewhere use shmem_read_mapping_page(), with
the mapping's gfp_mask properly initialized.

Forget about __GFP_COLD: since tmpfs initializes its pages with memset,
asking for a cold page is counter-productive.

Include linux/shmem_fs.h also in drm_gem.c: with shmem_file_setup() now
declared there too, we shall remove the prototype from linux/mm.h later.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Keith Packard <keithp@keithp.com>
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrm/ttm: use shmem_read_mapping_page
Hugh Dickins [Mon, 27 Jun 2011 23:18:17 +0000 (16:18 -0700)]
drm/ttm: use shmem_read_mapping_page

Soon tmpfs will stop supporting ->readpage and read_mapping_page(): once
"tmpfs: add shmem_read_mapping_page_gfp" has been applied, this patch can
be applied to ease the transition.

ttm_tt_swapin() and ttm_tt_swapout() use shmem_read_mapping_page() in
place of read_mapping_page(), since their swap_space has been created with
shmem_file_setup().

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrivers/tty/serial/8250_pci.c: fix warning
Andrew Morton [Mon, 27 Jun 2011 23:18:16 +0000 (16:18 -0700)]
drivers/tty/serial/8250_pci.c: fix warning

Fis the warning

  drivers/tty/serial/8250_pci.c:1457: warning: initialization from incompatible pointer type

Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrivers/misc/ioc4.c: fix section mismatch / race condition
Ralf Baechle [Mon, 27 Jun 2011 23:18:15 +0000 (16:18 -0700)]
drivers/misc/ioc4.c: fix section mismatch / race condition

Fix this section mismatch:

  WARNING: drivers/misc/ioc4.o(.data+0x144): Section mismatch in reference from the variable ioc4_load_modules_work to the function .devinit.text:ioc4_load_modules()
  The variable ioc4_load_modules_work references
  the function __devinit ioc4_load_modules()
  If the reference is valid then annotate the
  variable with __init* or __refdata (see linux/init.h) or name the variable:
  *driver, *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console

This one is potentially fatal; by the time ioc4_load_modules is invoked
it may already have been freed.  For that reason ioc4_load_modules_work
can't be turned to __devinitdata but also because it's referenced in
ioc4_exit.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Acked-by: Brent Casavant <bcasavan@sgi.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrivers/leds/leds-lp5523.c: fix section mismatches
Ralf Baechle [Mon, 27 Jun 2011 23:18:14 +0000 (16:18 -0700)]
drivers/leds/leds-lp5523.c: fix section mismatches

Fix this section mismatch:

  WARNING: drivers/leds/leds-lp5523.o(.text+0x12f4): Section mismatch in reference from the function lp5523_probe() to the function .init.text:lp5523_init_led()
  The function lp5523_probe() references
  the function __init lp5523_init_led().
  This is often because lp5523_probe lacks a __init
  annotation or the annotation of lp5523_init_led is wrong.

Fixing this one triggers one more mismatch, fix that one as well.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrivers/leds/leds-lp5521.c: fix section mismatches
Ralf Baechle [Mon, 27 Jun 2011 23:18:13 +0000 (16:18 -0700)]
drivers/leds/leds-lp5521.c: fix section mismatches

Fix this section mismatch:

  WARNING: drivers/leds/leds-lp5521.o(.text+0xf2c): Section mismatch in reference from the function lp5521_probe() to the function .init.text:lp5521_init_led()
  The function lp5521_probe() references
  the function __init lp5521_init_led().
  This is often because lp5521_probe lacks a __init
  annotation or the annotation of lp5521_init_led is wrong.

Fixing this mismatch triggers one more mismatch, fix that one as well.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomemcg: fix direct softlimit reclaim to be called in limit path
KAMEZAWA Hiroyuki [Mon, 27 Jun 2011 23:18:12 +0000 (16:18 -0700)]
memcg: fix direct softlimit reclaim to be called in limit path

Commit d149e3b25d7c ("memcg: add the soft_limit reclaim in global direct
reclaim") adds a softlimit hook to shrink_zones().  By this, soft limit
is called as

   try_to_free_pages()
       do_try_to_free_pages()
           shrink_zones()
               mem_cgroup_soft_limit_reclaim()

Then, direct reclaim is memcg softlimit hint aware, now.

But, the memory cgroup's "limit" path can call softlimit shrinker.

   try_to_free_mem_cgroup_pages()
       do_try_to_free_pages()
           shrink_zones()
               mem_cgroup_soft_limit_reclaim()

This will cause a global reclaim when a memcg hits limit.

This is bug. soft_limit_reclaim() should be called when
scanning_global_lru(sc) == true.

And the commit adds a variable "total_scanned" for counting softlimit
scanned pages....it's not "total".  This patch removes the variable and
update sc->nr_scanned instead of it.  This will affect shrink_slab()'s
scan condition but, global LRU is scanned by softlimit and I think this
change makes sense.

TODO: avoid too much scanning of a zone when softlimit did enough work.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Ying Han <yinghan@google.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agotaskstats: don't allow duplicate entries in listener mode
Vasiliy Kulikov [Mon, 27 Jun 2011 23:18:11 +0000 (16:18 -0700)]
taskstats: don't allow duplicate entries in listener mode

Currently a single process may register exit handlers unlimited times.
It may lead to a bloated listeners chain and very slow process
terminations.

Eg after 10KK sent TASKSTATS_CMD_ATTR_REGISTER_CPUMASKs ~300 Mb of
kernel memory is stolen for the handlers chain and "time id" shows 2-7
seconds instead of normal 0.003.  It makes it possible to exhaust all
kernel memory and to eat much of CPU time by triggerring numerous exits
on a single CPU.

The patch limits the number of times a single process may register
itself on a single CPU to one.

One little issue is kept unfixed - as taskstats_exit() is called before
exit_files() in do_exit(), the orphaned listener entry (if it was not
explicitly deregistered) is kept until the next someone's exit() and
implicit deregistration in send_cpu_listeners().  So, if a process
registered itself as a listener exits and the next spawned process gets
the same pid, it would inherit taskstats attributes.

Signed-off-by: Vasiliy Kulikov <segooon@gmail.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: fix assertion mapping->nrpages == 0 in end_writeback()
Jan Kara [Mon, 27 Jun 2011 23:18:10 +0000 (16:18 -0700)]
mm: fix assertion mapping->nrpages == 0 in end_writeback()

Under heavy memory and filesystem load, users observe the assertion
mapping->nrpages == 0 in end_writeback() trigger.  This can be caused by
page reclaim reclaiming the last page from a mapping in the following
race:

CPU0 CPU1
  ...
  shrink_page_list()
    __remove_mapping()
      __delete_from_page_cache()
        radix_tree_delete()
evict_inode()
  truncate_inode_pages()
    truncate_inode_pages_range()
      pagevec_lookup() - finds nothing
  end_writeback()
    mapping->nrpages != 0 -> BUG
        page->mapping = NULL
        mapping->nrpages--

Fix the problem by doing a reliable check of mapping->nrpages under
mapping->tree_lock in end_writeback().

Analyzed by Jay <jinshan.xiong@whamcloud.com>, lost in LKML, and dug out
by Miklos Szeredi <mszeredi@suse.de>.

Cc: Jay <jinshan.xiong@whamcloud.com>
Cc: Miklos Szeredi <mszeredi@suse.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm/memory-failure.c: fix spinlock vs mutex order
Peter Zijlstra [Mon, 27 Jun 2011 23:18:09 +0000 (16:18 -0700)]
mm/memory-failure.c: fix spinlock vs mutex order

We cannot take a mutex while holding a spinlock, so flip the order and
fix the locking documentation.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodrivers/misc/lkdtm.c: fix race when crashpoint is hit multiple times before checking...
Josh Hunt [Mon, 27 Jun 2011 23:18:08 +0000 (16:18 -0700)]
drivers/misc/lkdtm.c: fix race when crashpoint is hit multiple times before checking count

We observed the crash point count going negative in cases where the
crash point is hit multiple times before the check of "count == 0" is
done.  Because of this we never call lkdtm_do_action().  This patch just
adds a spinlock to protect count.

Reported-by: Tapan Dhimant <tdhimant@akamai.com>
Signed-off-by: Josh Hunt <johunt@akamai.com>
Acked-by: Ankita Garg <ankita@in.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoinclude/linux/compat.h: declare compat_sys_sendmmsg()
Chris Metcalf [Mon, 27 Jun 2011 23:18:07 +0000 (16:18 -0700)]
include/linux/compat.h: declare compat_sys_sendmmsg()

This is required for tilegx to be able to use the compat unistd.h header
where compat_sys_sendmmsg() is now mentioned.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoromfs: fix romfs_get_unmapped_area() argument check
Bob Liu [Mon, 27 Jun 2011 23:18:06 +0000 (16:18 -0700)]
romfs: fix romfs_get_unmapped_area() argument check

romfs_get_unmapped_area() checks argument `len' without considering
PAGE_ALIGN which will cause do_mmap_pgoff() return -EINVAL error after
commit f67d9b1576c ("nommu: add page_align to mmap").

Fix the check by changing it in same way ramfs_nommu_get_unmapped_area()
was changed in ramfs/file-nommu.c.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Acked-by: Greg Ungerer <gerg@snapgear.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>