pandora-kernel.git
10 years agoperf stat: Add support for --initial-delay option
Andi Kleen [Sat, 3 Aug 2013 00:41:11 +0000 (17:41 -0700)]
perf stat: Add support for --initial-delay option

When measuring workloads the startup phase -- doing page faults, dynamic
linking, opening files -- is often very different from the rest of the
workload.  Especially with smaller kernels and using counter
multiplexing this can give significant measurement errors.

Multiplexing assumes that the workload is mostly the same over longer
periods. But at startup there is typically some spike of activity which
is relatively short.  If many groups are multiplexing the one group
seeing the spike, and which is then scaled up over the time to run all
groups, may see a significant error.

Also in general it's often not useful to measure the startup, because it
is so different from the rest.

One way around this is to use interval mode and discard the first
sample, but this can be awkward because interval mode doesn't support
intervals of less than 100ms, and also a useful interval is not
necessarily the same as a useful startup delay.

This patch adds a new --initial-delay / -D option to skip measuring for
the startup phase. The time can be specified in ms

Here's a simple example:

perf stat -e page-faults bash -c 'for i in $(seq 100000) ; do true ; done'
...
             3,721 page-faults
...

If we just wait 20 ms the number of page faults is 1/3 less:

perf stat -D 20 -e page-faults bash -c 'for i in $(seq 100000) ; do true ; done'
...
             2,823 page-faults
...

So we filtered out most of the startup noise from bash.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1375490473-1503-4-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evsel: Add support for enabling counters
Andi Kleen [Sat, 3 Aug 2013 00:41:10 +0000 (17:41 -0700)]
perf evsel: Add support for enabling counters

Add support for enabling already set up counters by using an
ioctl. I share some code with the filter setup.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1375490473-1503-3-git-send-email-andi@firstfloor.org
[ Fixed up 'err' variable indentation ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evlist: Remove obsolete dummy execve
Andi Kleen [Sat, 3 Aug 2013 00:41:09 +0000 (17:41 -0700)]
perf evlist: Remove obsolete dummy execve

Minor cleanup.

The dummy execve to pre-resolve the PLT is obsolete since
"enable_on_execve" was added. The counters are only
running after the execve anyways. So just remove it.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1375490473-1503-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf kvm: Split out tracepoints from record args
David Ahern [Fri, 2 Aug 2013 20:05:42 +0000 (14:05 -0600)]
perf kvm: Split out tracepoints from record args

Needed by kvm live command. Make record_args a local while we are
messing with the args.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1375473947-64285-5-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf session: Export a few functions for event processing
David Ahern [Fri, 2 Aug 2013 20:05:41 +0000 (14:05 -0600)]
perf session: Export a few functions for event processing

Allows kvm live mode to reuse the event processing and ordered samples
processing used by the perf-report path.

v2: removed flush_sample_queue as noticed by Jiri

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1375473947-64285-4-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf stats: Add max and min stats
David Ahern [Fri, 2 Aug 2013 20:05:40 +0000 (14:05 -0600)]
perf stats: Add max and min stats

Need an initialization function to set min to -1 to
differentiate from an actual min of 0.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1375473947-64285-3-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf top: move CONSOLE_CLEAR to header file
David Ahern [Fri, 2 Aug 2013 20:05:39 +0000 (14:05 -0600)]
perf top: move CONSOLE_CLEAR to header file

For use with kvm-live mode.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1375473947-64285-2-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf util: Add parse_nsec_time() function
Namhyung Kim [Tue, 4 Jun 2013 01:50:29 +0000 (10:50 +0900)]
perf util: Add parse_nsec_time() function

The parse_nsec_time() function is for parsing a string of time into
64-bit nsec value.  It's a preparation of time filtering in some of perf
commands.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: David Ahern <dsahern@gmail.com>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1370310629-9642-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf python: Remove duplicate TID bit from mask
Arnaldo Carvalho de Melo [Thu, 1 Aug 2013 20:00:45 +0000 (17:00 -0300)]
perf python: Remove duplicate TID bit from mask

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>
Cc: Thiago Peixoto <thiagolcpeixoto@gmail.com>
Link: http://lkml.kernel.org/n/tip-jurgz6myq125o1ql6lldh6f7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf trace: Beautify 'connect' result
Arnaldo Carvalho de Melo [Tue, 30 Jul 2013 19:38:23 +0000 (16:38 -0300)]
perf trace: Beautify 'connect' result

It is an errno, so print an error string.

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-zt68gijvvoe8gd7kmclo43si@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tools: Fix compile of util/tsc.c
David Ahern [Fri, 26 Jul 2013 14:27:23 +0000 (08:27 -0600)]
perf tools: Fix compile of util/tsc.c

On Fedora 18, with gcc 4.6.4 compile fails with:

arch/x86/util/tsc.c: In function ‘perf_time_to_tsc’:
arch/x86/util/tsc.c:13:6: error: declaration of ‘time’ shadows a global declaration [-Werror=shadow]
cc1: all warnings being treated as errors
make: *** [/tmp/junk/arch/x86/util/tsc.o] Error 1
make: *** Waiting for unfinished jobs....

Fix by renaming the local variable.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Link: http://lkml.kernel.org/r/1374848843-43127-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evsel: Actually show symbol offset in stack trace when requested
David Ahern [Sun, 28 Jul 2013 15:14:34 +0000 (09:14 -0600)]
perf evsel: Actually show symbol offset in stack trace when requested

Symbol offset is one of the fields that can be requested in perf-script.
Currently you do not get that data when requested. e.g.,

perf script -f comm,tid,pid,time,cpu,sym,symoff,ip
...
gcc  6201/6201  [006] 762250.617897:
    ffffffff81090d95 update_curr
    ffffffff810911b8 dequeue_entity
    ffffffff81091825 dequeue_task_fair
    ffffffff81087163 dequeue_task
    ffffffff81087c03 deactivate_task
...

With this patch you get the offset:
...
gcc  6201/6201  [006] 762250.617897:
    ffffffff81090d95 update_curr+0x1c5
    ffffffff810911b8 dequeue_entity+0x28
    ffffffff81091825 dequeue_task_fair+0x45
    ffffffff81087163 dequeue_task+0x93
    ffffffff81087c03 deactivate_task+0x23
...

Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1375024474-45726-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Add parse events tests for leader sampling
Jiri Olsa [Fri, 1 Feb 2013 19:37:11 +0000 (20:37 +0100)]
perf tests: Add parse events tests for leader sampling

Adding 2 more tests to the automated parse events suite for following
event config:

  '{cycles,cache-misses,branch-misses}:S'
  '{instructions,branch-misses}:Su'

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-tmcy0ir7i8id2t54qg5ifbio@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Add attr record group sampling test
Jiri Olsa [Fri, 1 Feb 2013 18:33:31 +0000 (19:33 +0100)]
perf tests: Add attr record group sampling test

Adding test to validate perf_event_attr data for command:

  'record -e '{cycles,cache-misses}:S'

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-9eppxvhkly6gse5ximudckrp@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tools: Add 'S' event/group modifier to read sample value
Jiri Olsa [Wed, 10 Oct 2012 15:39:03 +0000 (17:39 +0200)]
perf tools: Add 'S' event/group modifier to read sample value

Adding 'S' event/group modifier to specify that the event value/s are
read by PERF_SAMPLE_READ sample type processing, instead of the period
value offered by lower layers.

There's additional behaviour change for 'S' modifier being specified on
event group:

Currently all the events within a group makes samples. If user now
specifies 'S' within group modifier, only the leader will trigger
samples. The rest of events in the group will have sampling disabled.

And same as for single events, values of all events within the group
(including leader) are read by PERF_SAMPLE_READ sample type processing.

Following example will create event group with cycles and cache-misses
events, setting the cycles as group leader and the only event to
actually sample. Both cycles and cache-misses event period values are
read by PERF_SAMPLE_READ sample type processing with PERF_FORMAT_GROUP
read format.

Example:

  $ perf record -e '{cycles,cache-misses}:S' ls
  ...
  $ perf report --group --show-total-period --stdio
  ...
  # Samples: 36  of event 'anon group { cycles, cache-misses }'
  # Event count (approx.): 12585593
  #
  #       Overhead          Period  Command      Shared Object                      Symbol
  # ..............  ..............  .......  .................  ..........................
  #
    19.92%   1.20%  2505936     31       ls  [kernel.kallsyms]  [k] mark_held_locks
    13.74%   0.47%  1729327     12       ls  [kernel.kallsyms]  [k] sched_clock_local
    13.64%  23.72%  1716147    612       ls  ld-2.14.90.so      [.] check_match.10805
    13.12%  23.22%  1650778    599       ls  libc-2.14.90.so    [.] _nl_intern_locale_data
    11.24%  29.19%  1414554    753       ls  [kernel.kallsyms]  [k] sched_clock_cpu
     8.50%   0.35%  1070150      9       ls  [kernel.kallsyms]  [k] check_chain_key
  ...

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-iyoinu3axi11mymwnh2b7fxj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evsel: Add PERF_SAMPLE_READ sample related processing
Jiri Olsa [Wed, 10 Oct 2012 16:52:24 +0000 (18:52 +0200)]
perf evsel: Add PERF_SAMPLE_READ sample related processing

For sample with sample type PERF_SAMPLE_READ the period value is stored
in the 'struct sample_read'.

Moreover if the read format has PERF_FORMAT_GROUP, the 'struct
sample_read' contains period values for all events in the group (for
which the sample's event is a leader).

We deliver separated samples for all the values contained within the
'struct sample_read'.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-6mdm5xkrm6kypouh1c33cyys@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evlist: Add perf_evlist__id2sid method to get event ID related data
Jiri Olsa [Thu, 11 Oct 2012 12:10:35 +0000 (14:10 +0200)]
perf evlist: Add perf_evlist__id2sid method to get event ID related data

This will be helpful for PERF_FORMAT_GROUP samples where we need to
store ID related period value for each event.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-twmlgsbyim97p7cyohjwb1df@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evlist: Fix event ID retrieval for group format read case
Jiri Olsa [Fri, 12 Oct 2012 11:02:21 +0000 (13:02 +0200)]
perf evlist: Fix event ID retrieval for group format read case

We need to fail the event ID retrieval in case both following conditions
are true:

  - we are on kernel with no PERF_EVENT_IOC_ID support
  - PERF_FORMAT_GROUP read format is set

The PERF_FORMAT_GROUP read format bit is the killer for retrieving event
ID out of the read syscall, because we have no guarantee of the event
placement within leader kernel sibling list.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-e93pgyj20rqx48qzw10vj4r4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tools: Add support for parsing PERF_SAMPLE_READ sample type
Jiri Olsa [Wed, 10 Oct 2012 15:38:13 +0000 (17:38 +0200)]
perf tools: Add support for parsing PERF_SAMPLE_READ sample type

Adding support to parse out the PERF_SAMPLE_READ sample bits.  The code
contains both single and group format specification.

This code parse out and prepare PERF_SAMPLE_READ data into the
perf_sample struct. It will be used for group leader sampling feature
comming in shortly.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-0tgdoln5rwk3wocshb442cl3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evlist: Use PERF_EVENT_IOC_ID perf ioctl to read event id
Jiri Olsa [Wed, 4 Apr 2012 17:32:27 +0000 (19:32 +0200)]
perf evlist: Use PERF_EVENT_IOC_ID perf ioctl to read event id

Changing the way we retrieve the event ID. Instead of parsing out
the ID out of the read data, using the PERF_EVENT_IOC_ID ioctl.

Keeping the old way in place to support kernels without
PERF_EVENT_IOC_ID ioctl support.

This will be useful for retrieving the event ID for events
with PERF_FORMAT_GROUP read format set, where it's impossible
to get correct event id out of the read call data.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-psgb4n7kte8e6tfenbe7nj2h@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf: Do not get values from disabled counters in group format read
Jiri Olsa [Mon, 15 Oct 2012 18:13:45 +0000 (20:13 +0200)]
perf: Do not get values from disabled counters in group format read

It's possible some of the counters in the group could be
disabled when sampling member of the event group is reading
the rest via PERF_SAMPLE_READ sample type processing. Disabled
counters could then produce wrong numbers.

Fixing that by reading only enabled counters for PERF_SAMPLE_READ
sample type processing.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-wwkjb0bbcuslnz0klrmqi26r@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf: Add PERF_EVENT_IOC_ID ioctl to return event ID
Jiri Olsa [Wed, 24 Oct 2012 11:37:58 +0000 (13:37 +0200)]
perf: Add PERF_EVENT_IOC_ID ioctl to return event ID

The only way to get the event ID is by reading the event fd,
followed by parsing the ID value out of the returned data.

While this is ok for current read format used by perf tool,
it is not ok when we use PERF_FORMAT_GROUP format.

With this format the data are returned for the whole group
and there's no way to find out what ID belongs to our fd
(if we are not group leader event).

Adding a simple ioctl that returns event primary ID for given fd.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-v1bn5cto707jn0bon34afqr1@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agowatchdog: Make it work under full dynticks
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:06 +0000 (02:31 +0200)]
watchdog: Make it work under full dynticks

A perf event can be used without forcing the tick to
stay alive if it doesn't use a frequency but a sample
period and if it doesn't throttle (raise storm of events).

Since the lockup detector neither use a perf event frequency
nor should ever throttle due to its high period, it can now
run concurrently with the full dynticks feature.

So remove the hack that disabled the watchdog.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: Anish Singh <anish198519851985@gmail.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-9-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Implement finer grained full dynticks kick
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:05 +0000 (02:31 +0200)]
perf: Implement finer grained full dynticks kick

Currently the full dynticks subsystem keep the
tick alive as long as there are perf events running.

This prevents the tick from being stopped as long as features
such that the lockup detectors are running. As a temporary fix,
the lockup detector is disabled by default when full dynticks
is built but this is not a long term viable solution.

To fix this, only keep the tick alive when an event configured
with a frequency rather than a period is running on the CPU,
or when an event throttles on the CPU.

These are the only purposes of the perf tick, especially now that
the rotation of flexible events is handled from a seperate hrtimer.
The tick can be shutdown the rest of the time.

Original-patch-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-8-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Account freq events per cpu
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:04 +0000 (02:31 +0200)]
perf: Account freq events per cpu

This is going to be used by the full dynticks subsystem
as a finer-grained information to know when to keep and
when to stop the tick.

Original-patch-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-7-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Migrate per cpu event accounting
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:03 +0000 (02:31 +0200)]
perf: Migrate per cpu event accounting

When an event is migrated, move the event per-cpu
accounting accordingly so that branch stack and cgroup
events work correctly on the new CPU.

Original-patch-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-6-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Split the per-cpu accounting part of the event accounting code
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:02 +0000 (02:31 +0200)]
perf: Split the per-cpu accounting part of the event accounting code

This way we can use the per-cpu handling seperately.
This is going to be used by to fix the event migration
code accounting.

Original-patch-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-5-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Factor out event accounting code to account_event()/__free_event()
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:01 +0000 (02:31 +0200)]
perf: Factor out event accounting code to account_event()/__free_event()

Gather all the event accounting code to a single place,
once all the prerequisites are completed. This simplifies
the refcounting.

Original-patch-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-4-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Sanitize get_callchain_buffer()
Frederic Weisbecker [Tue, 23 Jul 2013 00:31:00 +0000 (02:31 +0200)]
perf: Sanitize get_callchain_buffer()

In case of allocation failure, get_callchain_buffer() keeps the
refcount incremented for the current event.

As a result, when get_callchain_buffers() returns an error,
we must cleanup what it did by cancelling its last refcount
with a call to put_callchain_buffers().

This is a hack in order to be able to call free_event()
after that failure.

The original purpose of that was to simplify the failure
path. But this error handling is actually counter intuitive,
ugly and not very easy to follow because one expect to
see the resources used to perform a service to be cleaned
by the callee if case of failure, not by the caller.

So lets clean this up by cancelling the refcount from
get_callchain_buffer() in case of failure. And correctly free
the event accordingly in perf_event_alloc().

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-3-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Fix branch stack refcount leak on callchain init failure
Frederic Weisbecker [Tue, 23 Jul 2013 00:30:59 +0000 (02:30 +0200)]
perf: Fix branch stack refcount leak on callchain init failure

On callchain buffers allocation failure, free_event() is
called and all the accounting performed in perf_event_alloc()
for that event is cancelled.

But if the event has branch stack sampling, it is unaccounted
as well from the branch stack sampling events refcounts.

This is a bug because this accounting is performed after the
callchain buffer allocation. As a result, the branch stack sampling
events refcount can become negative.

To fix this, move the branch stack event accounting before the
callchain buffer allocation.

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374539466-4799-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agosched: Micro-optimize the smart wake-affine logic
Peter Zijlstra [Thu, 4 Jul 2013 04:56:46 +0000 (12:56 +0800)]
sched: Micro-optimize the smart wake-affine logic

Smart wake-affine is using node-size as the factor currently, but the overhead
of the mask operation is high.

Thus, this patch introduce the 'sd_llc_size' percpu variable, which will record
the highest cache-share domain size, and make it to be the new factor, in order
to reduce the overhead and make it more reasonable.

Tested-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/51D5008E.6030102@linux.vnet.ibm.com
[ Tidied up the changelog. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agosched: Implement smarter wake-affine logic
Michael Wang [Thu, 4 Jul 2013 04:55:51 +0000 (12:55 +0800)]
sched: Implement smarter wake-affine logic

The wake-affine scheduler feature is currently always trying to pull
the wakee close to the waker. In theory this should be beneficial if
the waker's CPU caches hot data for the wakee, and it's also beneficial
in the extreme ping-pong high context switch rate case.

Testing shows it can benefit hackbench up to 15%.

However, the feature is somewhat blind, from which some workloads
such as pgbench suffer. It's also time-consuming algorithmically.

Testing shows it can damage pgbench up to 50% - far more than the
benefit it brings in the best case.

So wake-affine should be smarter and it should realize when to
stop its thankless effort at trying to find a suitable CPU to wake on.

This patch introduces 'wakee_flips', which will be increased each
time the task flips (switches) its wakee target.

So a high 'wakee_flips' value means the task has more than one
wakee, and the bigger the number, the higher the wakeup frequency.

Now when making the decision on whether to pull or not, pay attention to
the wakee with a high 'wakee_flips', pulling such a task may benefit
the wakee. Also imply that the waker will face cruel competition later,
it could be very cruel or very fast depends on the story behind
'wakee_flips', waker therefore suffers.

Furthermore, if waker also has a high 'wakee_flips', that implies that
multiple tasks rely on it, then waker's higher latency will damage all
of them, so pulling wakee seems to be a bad deal.

Thus, when 'waker->wakee_flips / wakee->wakee_flips' becomes
higher and higher, the cost of pulling seems to be worse and worse.

The patch therefore helps the wake-affine feature to stop its pulling
work when:

wakee->wakee_flips > factor &&
waker->wakee_flips > (factor * wakee->wakee_flips)

The 'factor' here is the number of CPUs in the current CPU's NUMA node,
so a bigger node will lead to more pulling since the trial becomes more
severe.

After applying the patch, pgbench shows up to 40% improvements and no regressions.

Tested with 12 cpu x86 server and tip 3.10.0-rc7.

The percentages in the final column highlight the areas with the biggest wins,
all other areas improved as well:

pgbench     base smart

| db_size | clients |  tps  | |  tps  |
+---------+---------+-------+   +-------+
| 22 MB   |       1 | 10598 |   | 10796 |
| 22 MB   |       2 | 21257 |   | 21336 |
| 22 MB   |       4 | 41386 |   | 41622 |
| 22 MB   |       8 | 51253 |   | 57932 |
| 22 MB   |      12 | 48570 |   | 54000 |
| 22 MB   |      16 | 46748 |   | 55982 | +19.75%
| 22 MB   |      24 | 44346 |   | 55847 | +25.93%
| 22 MB   |      32 | 43460 |   | 54614 | +25.66%
| 7484 MB |       1 |  8951 |   |  9193 |
| 7484 MB |       2 | 19233 |   | 19240 |
| 7484 MB |       4 | 37239 |   | 37302 |
| 7484 MB |       8 | 46087 |   | 50018 |
| 7484 MB |      12 | 42054 |   | 48763 |
| 7484 MB |      16 | 40765 |   | 51633 | +26.66%
| 7484 MB |      24 | 37651 |   | 52377 | +39.11%
| 7484 MB |      32 | 37056 |   | 51108 | +37.92%
| 15 GB   |       1 |  8845 |   |  9104 |
| 15 GB   |       2 | 19094 |   | 19162 |
| 15 GB   |       4 | 36979 |   | 36983 |
| 15 GB   |       8 | 46087 |   | 49977 |
| 15 GB   |      12 | 41901 |   | 48591 |
| 15 GB   |      16 | 40147 |   | 50651 | +26.16%
| 15 GB   |      24 | 37250 |   | 52365 | +40.58%
| 15 GB   |      32 | 36470 |   | 50015 | +37.14%

Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/51D50057.9000809@linux.vnet.ibm.com
[ Improved the changelog. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agosched: Move h_load calculation to task_h_load()
Vladimir Davydov [Mon, 15 Jul 2013 13:49:19 +0000 (17:49 +0400)]
sched: Move h_load calculation to task_h_load()

The bad thing about update_h_load(), which computes hierarchical load
factor for task groups, is that it is called for each task group in the
system before every load balancer run, and since rebalance can be
triggered very often, this function can eat really a lot of cpu time if
there are many cpu cgroups in the system.

Although the situation was improved significantly by commit a35b646
('sched, cgroup: Reduce rq->lock hold times for large cgroup
hierarchies'), the problem still can arise under some kinds of loads,
e.g. when cpus are switching from idle to busy and back very frequently.

For instance, when I start 1000 of processes that wake up every
millisecond on my 8 cpus host, 'top' and 'perf top' show:

Cpu(s): 17.8%us, 24.3%sy,  0.0%ni, 57.9%id,  0.0%wa,  0.0%hi,  0.0%si
Events: 243K cycles
  7.57%  [kernel]               [k] __schedule
  7.08%  [kernel]               [k] timerqueue_add
  6.13%  libc-2.12.so           [.] usleep

Then if I create 10000 *idle* cpu cgroups (no processes in them), cpu
usage increases significantly although the 'wakers' are still executing
in the root cpu cgroup:

Cpu(s): 19.1%us, 48.7%sy,  0.0%ni, 31.6%id,  0.0%wa,  0.0%hi,  0.7%si
Events: 230K cycles
 24.56%  [kernel]            [k] tg_load_down
  5.76%  [kernel]            [k] __schedule

This happens because this particular kind of load triggers 'new idle'
rebalance very frequently, which requires calling update_h_load(),
which, in turn, calls tg_load_down() for every *idle* cpu cgroup even
though it is absolutely useless, because idle cpu cgroups have no tasks
to pull.

This patch tries to improve the situation by making h_load calculation
proceed only when h_load is really necessary. To achieve this, it
substitutes update_h_load() with update_cfs_rq_h_load(), which computes
h_load only for a given cfs_rq and all its ascendants, and makes the
load balancer call this function whenever it considers if a task should
be pulled, i.e. it moves h_load calculations directly to task_h_load().
For h_load of the same cfs_rq not to be updated multiple times (in case
several tasks in the same cgroup are considered during the same balance
run), the patch keeps the time of the last h_load update for each cfs_rq
and breaks calculation when it finds h_load to be uptodate.

The benefit of it is that h_load is computed only for those cfs_rq's,
which really need it, in particular all idle task groups are skipped.
Although this, in fact, moves h_load calculation under rq lock, it
should not affect latency much, because the amount of work done under rq
lock while trying to pull tasks is limited by sched_nr_migrate.

After the patch applied with the setup described above (1000 wakers in
the root cgroup and 10000 idle cgroups), I get:

Cpu(s): 16.9%us, 24.8%sy,  0.0%ni, 58.4%id,  0.0%wa,  0.0%hi,  0.0%si
Events: 242K cycles
  7.57%  [kernel]                  [k] __schedule
  6.70%  [kernel]                  [k] timerqueue_add
  5.93%  libc-2.12.so              [.] usleep

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1373896159-1278-1-git-send-email-vdavydov@parallels.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf tools: Add test for converting perf time to/from TSC
Adrian Hunter [Fri, 28 Jun 2013 13:22:19 +0000 (16:22 +0300)]
perf tools: Add test for converting perf time to/from TSC

The test uses the newly added cap_usr_time_zero and time_zero of
perf_event_mmap_page.  TSC from rdtsc is compared with the time
from 2 perf events.  The test passes if the calculated times are
all in the correct order.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1372425741-1676-4-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf/x86: Add ability to calculate TSC from perf sample timestamps
Adrian Hunter [Fri, 28 Jun 2013 13:22:18 +0000 (16:22 +0300)]
perf/x86: Add ability to calculate TSC from perf sample timestamps

For modern CPUs, perf clock is directly related to TSC.  TSC
can be calculated from perf clock and vice versa using a simple
calculation.  Two of the three componenets of that calculation
are already exported in struct perf_event_mmap_page.  This patch
exports the third.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Link: http://lkml.kernel.org/r/1372425741-1676-3-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Fix broken union in 'struct perf_event_mmap_page'
Adrian Hunter [Fri, 28 Jun 2013 13:22:17 +0000 (16:22 +0300)]
perf: Fix broken union in 'struct perf_event_mmap_page'

The capabilities bits must not be "union'ed" together.
Put them in a separate struct.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1372425741-1676-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf: Update perf_event_type documentation
Peter Zijlstra [Tue, 16 Jul 2013 15:09:07 +0000 (17:09 +0200)]
perf: Update perf_event_type documentation

Due to a discussion with Adrian I had a good look at the perf_event_type record
layout and found the documentation to be somewhat unclear.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20130716150907.GL23818@dyad.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agokprobes/x86: Call out into INT3 handler directly instead of using notifier
Jiri Kosina [Tue, 23 Jul 2013 08:09:28 +0000 (10:09 +0200)]
kprobes/x86: Call out into INT3 handler directly instead of using notifier

In fd4363fff3d96 ("x86: Introduce int3 (breakpoint)-based
instruction patching"), the mechanism that was introduced for
notifying alternatives code from int3 exception handler that and
exception occured was die_notifier.

This is however problematic, as early code might be using jump
labels even before the notifier registration has been performed,
which will then lead to an oops due to unhandled exception. One
of such occurences has been encountered by Fengguang:

 int3: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
 Modules linked in:
 CPU: 1 PID: 0 Comm: swapper/1 Not tainted 3.11.0-rc1-01429-g04bf576 #8
 task: ffff88000da1b040 ti: ffff88000da1c000 task.ti: ffff88000da1c000
 RIP: 0010:[<ffffffff811098cc>]  [<ffffffff811098cc>] ttwu_do_wakeup+0x28/0x225
 RSP: 0000:ffff88000dd03f10  EFLAGS: 00000006
 RAX: 0000000000000000 RBX: ffff88000dd12940 RCX: ffffffff81769c40
 RDX: 0000000000000002 RSI: 0000000000000000 RDI: 0000000000000001
 RBP: ffff88000dd03f28 R08: ffffffff8176a8c0 R09: 0000000000000002
 R10: ffffffff810ff484 R11: ffff88000dd129e8 R12: ffff88000dbc90c0
 R13: ffff88000dbc90c0 R14: ffff88000da1dfd8 R15: ffff88000da1dfd8
 FS:  0000000000000000(0000) GS:ffff88000dd00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
 CR2: 00000000ffffffff CR3: 0000000001c88000 CR4: 00000000000006e0
 Stack:
  ffff88000dd12940 ffff88000dbc90c0 ffff88000da1dfd8 ffff88000dd03f48
  ffffffff81109e2b ffff88000dd12940 0000000000000000 ffff88000dd03f68
  ffffffff81109e9e 0000000000000000 0000000000012940 ffff88000dd03f98
 Call Trace:
  <IRQ>
  [<ffffffff81109e2b>] ttwu_do_activate.constprop.56+0x6d/0x79
  [<ffffffff81109e9e>] sched_ttwu_pending+0x67/0x84
  [<ffffffff8110c845>] scheduler_ipi+0x15a/0x2b0
  [<ffffffff8104dfb4>] smp_reschedule_interrupt+0x38/0x41
  [<ffffffff8173bf5d>] reschedule_interrupt+0x6d/0x80
  <EOI>
  [<ffffffff810ff484>] ? __atomic_notifier_call_chain+0x5/0xc1
  [<ffffffff8105cc30>] ? native_safe_halt+0xd/0x16
  [<ffffffff81015f10>] default_idle+0x147/0x282
  [<ffffffff81017026>] arch_cpu_idle+0x3d/0x5d
  [<ffffffff81127d6a>] cpu_idle_loop+0x46d/0x5db
  [<ffffffff81127f5c>] cpu_startup_entry+0x84/0x84
  [<ffffffff8104f4f8>] start_secondary+0x3c8/0x3d5
  [...]

Fix this by directly calling poke_int3_handler() from the int3
exception handler (analogically to what ftrace has been doing
already), instead of relying on notifier, registration of which
might not have yet been finalized by the time of the first trap.

Reported-and-tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/alpine.LNX.2.00.1307231007490.14024@pobox.suse.cz
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Tue, 23 Jul 2013 07:37:33 +0000 (09:37 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

  * Fix memcpy benchmark for large sizes, from Andi Kleen.

  * Support callchain sorting based on addresses, from Andi Kleen

  * Move weight back to common sort keys, From Andi Kleen.

  * Fix named threads support in 'perf script', from David Ahern.

  * Handle ENODEV on default cycles event, fix from David Ahern.

  * More install tests, from Jiri Olsa.

  * Fix build with perl 5.18, from Kirill A. Shutemov.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoperf tools: Move weight back to common sort keys
Andi Kleen [Thu, 18 Jul 2013 22:58:53 +0000 (15:58 -0700)]
perf tools: Move weight back to common sort keys

This is a partial revert of Namhyung's patch

 afab87b91f3f331d55664172dad8e476e6ffca9d
 perf sort: Separate out memory-specific sort keys

He wrote

 For global/local weights, I'm not entirely sure to place them into the
 memory dimension.  But it's the only user at this time.

Well TSX is another (in fact the original) user of the flags, and it
needs them to be common. So move local/global weight back to the common
sort keys.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Link: http://lkml.kernel.org/r/1374188333-17899-1-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Add broken install-* tests into tests/make
Jiri Olsa [Mon, 22 Jul 2013 12:43:34 +0000 (14:43 +0200)]
perf tests: Add broken install-* tests into tests/make

Adding install-* tests into tests/make. Those tests are
broken, so commenting them out right away.

* Nothing get installed for install-man, install_doc and
  install_html targets, they just rebuild the documentation.

* I've got following error for 'install-info':

  $ make -f tests/make make_install_info
  - make_install_info: cd . && make -f Makefile DESTDIR=/tmp/tmp.Xi4mb9J1a0 install-info

  $ tail -f make_install_info
  ...
  PERF_VERSION = 3.11.rc1.g9b3c2d
  make[2]: *** No rule to make target `user-manual.xml', needed by `user-manual.texi'.  Stop.
  make[1]: *** [install-info] Error 2

* I've got following error for 'install-pdf':

  $ make -f tests/make make_install_pdf
  - make_install_pdf: cd . && make -f Makefile DESTDIR=/tmp/tmp.fXseECBbt1 install-pdf

  $ tail -f make_install_pdf
  ...
  PERF_VERSION = 3.11.rc1.g9b3c2d
  make[2]: *** No rule to make target `user-manual.xml', needed by `user-manual.pdf'.  Stop.
  make[1]: *** [install-pdf] Error 2

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374497014-2817-6-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Add 'make install/install-bin' tests into tests/make
Jiri Olsa [Mon, 22 Jul 2013 12:43:33 +0000 (14:43 +0200)]
perf tests: Add 'make install/install-bin' tests into tests/make

Adding 'make install' and 'make install-bin' tests into tests/make. It's
run as part of the suite, but could be run separately like:

  $ make -f tests/make make_install
  - make_install: cd . && make -f Makefile DESTDIR=/tmp/tmp.LpkYbk5pfs install
    test: test -x /tmp/tmp.LpkYbk5pfs/bin/perf
  $ make -f tests/make make_install_bin
  - make_install_bin: cd . && make -f Makefile DESTDIR=/tmp/tmp.dMxePBMcFT
    install-bin
    test: test -x /tmp/tmp.dMxePBMcFT/bin/perf

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374497014-2817-5-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Add DESTDIR=TMP_DEST tests/make variable
Jiri Olsa [Mon, 22 Jul 2013 12:43:32 +0000 (14:43 +0200)]
perf tests: Add DESTDIR=TMP_DEST tests/make variable

Adding TMP_DEST tests/make variable to provide the DESTDIR directory for
installation tests.

Adding this to existing test targets, since DESTDIR variable 'should
not' affect other than install* targets. We can always separate this if
there's a need for DESTDIR-free build test.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374497014-2817-4-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Rename TMP to TMP_O tests/make variable
Jiri Olsa [Mon, 22 Jul 2013 12:43:31 +0000 (14:43 +0200)]
perf tests: Rename TMP to TMP_O tests/make variable

Renaming TMP to TMP_O tests/make variable to make a name space for other
temp variables.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374497014-2817-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tests: Run ctags/cscope make tests only with needed binaries
Jiri Olsa [Mon, 22 Jul 2013 12:43:30 +0000 (14:43 +0200)]
perf tests: Run ctags/cscope make tests only with needed binaries

Running tags and cscope make tests only if the 'ctags' and 'cscope'
binaries are installed, so we don't have false alarm test failures.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374497014-2817-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tools: Fix build with perl 5.18
Kirill A. Shutemov [Mon, 24 Jun 2013 08:43:14 +0000 (11:43 +0300)]
perf tools: Fix build with perl 5.18

perl.h from new Perl release doesn't like -Wundef and -Wswitch-default:

/usr/lib/perl5/core_perl/CORE/perl.h:548:5: error: "SILENT_NO_TAINT_SUPPORT" is not defined [-Werror=undef]
 #if SILENT_NO_TAINT_SUPPORT && !defined(NO_TAINT_SUPPORT)
     ^
/usr/lib/perl5/core_perl/CORE/perl.h:556:5: error: "NO_TAINT_SUPPORT" is not defined [-Werror=undef]
 #if NO_TAINT_SUPPORT
     ^
In file included from /usr/lib/perl5/core_perl/CORE/perl.h:3471:0,
                 from util/scripting-engines/trace-event-perl.c:30:
/usr/lib/perl5/core_perl/CORE/sv.h:1455:5: error: "NO_TAINT_SUPPORT" is not defined [-Werror=undef]
 #if NO_TAINT_SUPPORT
     ^
In file included from /usr/lib/perl5/core_perl/CORE/perl.h:3472:0,
                 from util/scripting-engines/trace-event-perl.c:30:
/usr/lib/perl5/core_perl/CORE/regexp.h:436:5: error: "NO_TAINT_SUPPORT" is not defined [-Werror=undef]
 #if NO_TAINT_SUPPORT
     ^
In file included from /usr/lib/perl5/core_perl/CORE/hv.h:592:0,
                 from /usr/lib/perl5/core_perl/CORE/perl.h:3480,
                 from util/scripting-engines/trace-event-perl.c:30:
/usr/lib/perl5/core_perl/CORE/hv_func.h: In function ‘S_perl_hash_siphash_2_4’:
/usr/lib/perl5/core_perl/CORE/hv_func.h:222:3: error: switch missing default case [-Werror=switch-default]
   switch( left )
   ^
/usr/lib/perl5/core_perl/CORE/hv_func.h: In function ‘S_perl_hash_superfast’:
/usr/lib/perl5/core_perl/CORE/hv_func.h:274:5: error: switch missing default case [-Werror=switch-default]
     switch (rem) { \
     ^
/usr/lib/perl5/core_perl/CORE/hv_func.h: In function ‘S_perl_hash_murmur3’:
/usr/lib/perl5/core_perl/CORE/hv_func.h:398:5: error: switch missing default case [-Werror=switch-default]
     switch(bytes_in_carry) { /* how many bytes in carry */
     ^

Let's disable the warnings for code which uses perl.h.

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1372063394-20126-1-git-send-email-kirill@shutemov.name
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf tools: Support callchain sorting based on addresses
Andi Kleen [Thu, 18 Jul 2013 22:33:57 +0000 (15:33 -0700)]
perf tools: Support callchain sorting based on addresses

With programs with very large functions it can be useful to distinguish
the callgraph nodes on more than just function names. So for example if
you have multiple calls to the same function, it ends up being separate
nodes in the chain.

This patch adds a new key field to the callgraph options, that allows
comparing nodes on functions (as today, default) and addresses.

Longer term it would be nice to also handle src lines, but that would
need more changes and address is a reasonable proxy for it today.

I right now reference the global params, as there was no simple way to
register a params pointer.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-0uskktybf0e7wrnoi5e9b9it@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf bench: Fix memcpy benchmark for large sizes
Andi Kleen [Thu, 18 Jul 2013 22:33:46 +0000 (15:33 -0700)]
perf bench: Fix memcpy benchmark for large sizes

The glibc calloc() function has an optimization to not explicitely
memset() very large calloc allocations that just came from mmap(),
because they are known to be zero.

This could result in the perf memcpy benchmark reading only from
the zero page, which gives unrealistic results.

Always call memset explicitly on the source area to avoid this problem.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Hitoshi Mitake <h.mitake@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Link: http://lkml.kernel.org/n/tip-pzz2qrdq9eymxda0y8yxdn33@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf evsel: Handle ENODEV on default cycles event
David Ahern [Thu, 18 Jul 2013 23:27:59 +0000 (17:27 -0600)]
perf evsel: Handle ENODEV on default cycles event

Some systems (e.g., VMs on qemu-0.13 with the default vcpu model) report
an unsupported CPU model:

Performance Events: unsupported p6 CPU model 2 no PMU driver, software events only.

Subsequent invocations of perf fail with:

The sys_perf_event_open() syscall returned with 19 (No such device) for event (cycles).
/bin/dmesg may provide additional information.
No CONFIG_PERF_EVENTS=y kernel support configured?

Add ENODEV to the list of errno's to fallback to cpu-clock.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1374190079-28507-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf script: Fix named threads support
David Ahern [Thu, 18 Jul 2013 22:06:15 +0000 (16:06 -0600)]
perf script: Fix named threads support

Commit 73994dc broke named thread support in perf-script. The thread
struct in al is the main thread for a multithreaded process. The thread
struct used for analysis (e.g., dumping events) should be the specific
thread for the sample.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Feng Tang <feng.tang@intel.com>
Link: http://lkml.kernel.org/r/1374185175-28272-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agokprobes/x86: Remove unused text_poke_smp() and text_poke_smp_batch() functions
Masami Hiramatsu [Thu, 18 Jul 2013 11:47:53 +0000 (20:47 +0900)]
kprobes/x86: Remove unused text_poke_smp() and text_poke_smp_batch() functions

Since introducing the text_poke_bp() for all text_poke_smp*()
callers, text_poke_smp*() are now unused. This patch basically
reverts:

  3d55cc8a058e ("x86: Add text_poke_smp for SMP cross modifying code")
  7deb18dcf047 ("x86: Introduce text_poke_smp_batch() for batch-code modifying")

and related commits.

This patch also fixes a Kconfig dependency issue on STOP_MACHINE
in the case of CONFIG_SMP && !CONFIG_MODULE_UNLOAD.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jiri Kosina <jkosina@suse.cz>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@akamai.com>
Cc: yrl.pp-manager.tt@hitachi.com
Cc: Borislav Petkov <bpetkov@suse.de>
Link: http://lkml.kernel.org/r/20130718114753.26675.18714.stgit@mhiramat-M0-7522
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agokprobes/x86: Use text_poke_bp() instead of text_poke_smp*()
Masami Hiramatsu [Thu, 18 Jul 2013 11:47:50 +0000 (20:47 +0900)]
kprobes/x86: Use text_poke_bp() instead of text_poke_smp*()

Use text_poke_bp() for optimizing kprobes instead of
text_poke_smp*(). Since the number of kprobes is usually not so
large (<100) and text_poke_bp() is much lighter than
text_poke_smp() [which uses stop_machine()], this just stops
using batch processing.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jiri Kosina <jkosina@suse.cz>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@akamai.com>
Cc: yrl.pp-manager.tt@hitachi.com
Cc: Borislav Petkov <bpetkov@suse.de>
Link: http://lkml.kernel.org/r/20130718114750.26675.9174.stgit@mhiramat-M0-7522
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agokprobes/x86: Remove an incorrect comment about int3 in NMI/MCE
Masami Hiramatsu [Thu, 18 Jul 2013 11:47:47 +0000 (20:47 +0900)]
kprobes/x86: Remove an incorrect comment about int3 in NMI/MCE

Remove a comment about an int3 issue in NMI/MCE, since
commit:

  3f3c8b8c4b2a ("x86: Add workaround to NMI iret woes")

already fixed that. Keeping this incorrect comment can mislead developers.

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jiri Kosina <jkosina@suse.cz>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@akamai.com>
Cc: yrl.pp-manager.tt@hitachi.com
Cc: Borislav Petkov <bpetkov@suse.de>
Link: http://lkml.kernel.org/r/20130718114747.26675.84110.stgit@mhiramat-M0-7522
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoMerge branch 'x86/jumplabel' into perf/core
Ingo Molnar [Fri, 19 Jul 2013 07:55:00 +0000 (09:55 +0200)]
Merge branch 'x86/jumplabel' into perf/core

Upcoming kprobes patches rely on the int3 code-patching machinery introduced by:

   fd4363fff3d9 x86: Introduce int3 (breakpoint)-based instruction patching

Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Fri, 19 Jul 2013 07:35:30 +0000 (09:35 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

 * Add missing 'finished_round' event forwarding in 'perf inject', from Adrian Hunter.

 * Assorted tidy ups, from Adrian Hunter.

 * Fall back to sysfs event names when parsing fails, from Andi Kleen.

 * List pmu events in perf list, from Andi Kleen.

 * Cleanup some memory allocation/freeing uses, from David Ahern.

 * Add option to collapse undesired parts of call graph, from Greg Price.

 * Prep work for multi perf data file storage, from Jiri Olsa.

 * Add support for more than two files comparision in 'perf diff', from Jiri Olsa

 * A few more 'perf test' improvements, from Jiri Olsa

 * libtraceevent cleanups, from Namhyung Kim.

 * Remove odd build stall in 'perf sched' by moving a large struct initialization
   from a local variable to a global one, from Namhyung Kim.

 * Add support for callchains in the gtk UI, from Namhyung Kim.

 * Do not apply symfs for an absolute vmlinux path, fix from Namhyung Kim.

 * Use default include path notation for libtraceevent, from Robert Richter.

 * Fix 'make tools/perf', from Robert Richter.

 * Make Power7 events available, from Runzhen Wang.

 * Add --objdump option to 'perf top', from Sukadev Bhattiprolu.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoMerge branch 'linus' into perf/core
Ingo Molnar [Fri, 19 Jul 2013 07:34:42 +0000 (09:34 +0200)]
Merge branch 'linus' into perf/core

Merge in a v3.11-rc1-ish branch to go from v3.10 based development
to a v3.11 based one.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 19 Jul 2013 03:08:47 +0000 (20:08 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:
 "A couple interesting SKB fragment handling fixes, plus the usual small
  bits here and there:

   1) Fix 64-bit divide build failure on 32-bit platforms in mlx5, from
      Tim Gardner.

   2) Get rid of a stupid reimplementation on "%*phC" in our sysfs MAC
      address printing helper.

   3) Fix NETIF_F_SG capability advertisement in hyperv driver, if the
      device can't do checksumming offloads then it shouldn't say it can
      do SG either.  From Haiyang Zhang.

   4) bgmac needs to depend on PHYLIB, from Hauke Mehrtens.

   5) Don't leak DMA mappings on mapping failures, from Neil Horman.

   6) We need to reset the transport header of SKBs in ipv4 before we
      attempt to perform early socket demux, just like ipv6 does.  From
      Eric Dumazet.

   7) Add missing locking on vxlan device removal, from Stephen
      Hemminger.

   8) xen-netfront has to make two passes over an SKB to prepare it for
      transfer.  One pass calculates the number of slots needed, the
      second massages the SKB and fills the slots.  Unfortunately, the
      first pass doesn't calculate the number of slots properly so we
      can end up trying to build a MAX_SKB_FRAGS + 1 SKB which doesn't
      work out so well.  Fix from Jan Beulich with help and discussion
      with several others.

   9) Fix a similar problem in tun and macvtap, which have to split up
      scatter-gather elements at PAGE_SIZE boundaries.  Don't do
      zerocopy if it would result in a > MAX_SKB_FRAGS skb.  Fixes from
      Jason Wang.

  10) On receive, once we've decoded the VLAN state completely, clear
      skb->vlan_tci.  Otherwise demuxed tunnels underneath can trigger
      the VLAN code again, corrupting the packet.  Fix from Eric
      Dumazet"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  vlan: fix a race in egress prio management
  vlan: mask vlan prio bits
  macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS
  tuntap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS
  pkt_sched: sch_qfq: remove a source of high packet delay/jitter
  xen-netfront: pull on receive skb may need to happen earlier
  vxlan: add necessary locking on device removal
  hyperv: Fix the NETIF_F_SG flag setting in netvsc
  net: Fix sysfs_format_mac() code duplication.
  be2net: Fix to avoid hardware workaround when not needed
  macvtap: do not assume 802.1Q when send vlan packets
  macvtap: fix the missing ret value of TUNSETQUEUE
  ipv4: set transport header earlier
  mlx5 core: Fix __udivdi3 when compiling for 32 bit arches
  bgmac: add dependency to phylib
  net/irda: fixed style issues in irlan_eth
  ethtool: fixed trailing statements in ethtool
  ndisc: bool initializations should use true and false
  atl1e: unmap partially mapped skb on dma error and free skb

10 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 19 Jul 2013 00:39:05 +0000 (17:39 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Peter Anvin:
 "Trying again to get the fixes queue, including the fixed IDT alignment
  patch.

  The UEFI patch is by far the biggest issue at hand: it is currently
  causing quite a few machines to boot.  Which is sad, because the only
  reason they would is because their BIOSes touch memory that has
  already been freed.  The other major issue is that we finally have
  tracked down the root cause of a significant number of machines
  failing to suspend/resume"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86: Make sure IDT is page aligned
  x86, suspend: Handle CPUs which fail to #GP on RDMSR
  x86/platform/ce4100: Add header file for reboot type
  Revert "UEFI: Don't pass boot services regions to SetVirtualAddressMap()"
  efivars: check for EFI_RUNTIME_SERVICES

10 years agoMerge tag 'md-3.11-fixes' of git://neil.brown.name/md
Linus Torvalds [Fri, 19 Jul 2013 00:37:46 +0000 (17:37 -0700)]
Merge tag 'md-3.11-fixes' of git://neil.brown.name/md

Pull md bug fixes from NeilBrown:
 "Sorry boss, back at work now boss.  Here's them nice shiny patches ya
  wanted.  All nicely tagged and justified for -stable and everyfing:

  Three bug fixes for md in 3.10

  3.10 wasn't a good release for md.  The bio changes left a couple of
  bugs, and an md "fix" created another one.

  These three patches appear to fix the issues and have been tagged for
  -stable"

* tag 'md-3.11-fixes' of git://neil.brown.name/md:
  md/raid1: fix bio handling problems in process_checks()
  md: Remove recent change which allows devices to skip recovery.
  md/raid10: fix two problems with RAID10 resync.

10 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Thu, 18 Jul 2013 21:01:08 +0000 (14:01 -0700)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "You'll be terribly disappointed in this, I'm not trying to sneak any
  features in or anything, its mostly radeon and intel fixes, a couple
  of ARM driver fixes"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (34 commits)
  drm/radeon/dpm: add debugfs support for RS780/RS880 (v3)
  drm/radeon/dpm/atom: fix broken gcc harder
  drm/radeon/dpm/atom: restructure logic to work around a compiler bug
  drm/radeon/dpm: fix atom vram table parsing
  drm/radeon: fix an endian bug in atom table parsing
  drm/radeon: add a module parameter to disable aspm
  drm/rcar-du: Use the GEM PRIME helpers
  drm/shmobile: Use the GEM PRIME helpers
  uvesafb: Really allow mtrr being 0, as documented and warn()ed
  radeon kms: do not flush uninitialized hotplug work
  drm/radeon/dpm/sumo: handle boost states properly when forcing a perf level
  drm/radeon: align VM PTBs (Page Table Blocks) to 32K
  drm/radeon: allow selection of alignment in the sub-allocator
  drm/radeon: never unpin UVD bo v3
  drm/radeon: fix UVD fence emit
  drm/radeon: add fault decode function for CIK
  drm/radeon: add fault decode function for SI (v2)
  drm/radeon: add fault decode function for cayman/TN (v2)
  drm/radeon: use radeon device for request firmware
  drm/radeon: add missing ttm_eu_backoff_reservation to radeon_bo_list_validate
  ...

10 years agovlan: fix a race in egress prio management
Eric Dumazet [Thu, 18 Jul 2013 16:35:10 +0000 (09:35 -0700)]
vlan: fix a race in egress prio management

egress_priority_map[] hash table updates are protected by rtnl,
and we never remove elements until device is dismantled.

We have to make sure that before inserting an new element in hash table,
all its fields are committed to memory or else another cpu could
find corrupt values and crash.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovlan: mask vlan prio bits
Eric Dumazet [Thu, 18 Jul 2013 14:19:26 +0000 (07:19 -0700)]
vlan: mask vlan prio bits

In commit 48cc32d38a52d0b68f91a171a8d00531edc6a46e
("vlan: don't deliver frames for unknown vlans to protocols")
Florian made sure we set pkt_type to PACKET_OTHERHOST
if the vlan id is set and we could find a vlan device for this
particular id.

But we also have a problem if prio bits are set.

Steinar reported an issue on a router receiving IPv6 frames with a
vlan tag of 4000 (id 0, prio 2), and tunneled into a sit device,
because skb->vlan_tci is set.

Forwarded frame is completely corrupted : We can see (8100:4000)
being inserted in the middle of IPv6 source address :

16:48:00.780413 IP6 2001:16d8:8100:4000:ee1c:0:9d9:bc87 >
9f94:4d95:2001:67c:29f4::: ICMP6, unknown icmp6 type (0), length 64
       0x0000:  0000 0029 8000 c7c3 7103 0001 a0ae e651
       0x0010:  0000 0000 ccce 0b00 0000 0000 1011 1213
       0x0020:  1415 1617 1819 1a1b 1c1d 1e1f 2021 2223
       0x0030:  2425 2627 2829 2a2b 2c2d 2e2f 3031 3233

It seems we are not really ready to properly cope with this right now.

We can probably do better in future kernels :
vlan_get_ingress_priority() should be a netdev property instead of
a per vlan_dev one.

For stable kernels, lets clear vlan_tci to fix the bugs.

Reported-by: Steinar H. Gunderson <sesse@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agomacvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS
Jason Wang [Thu, 18 Jul 2013 02:55:16 +0000 (10:55 +0800)]
macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
network.

Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than
MAX_SKB_FRAGS.

This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify
vhost.

We can do further optimization on top.

This bug were introduced from b92946e2919134ebe2a4083e4302236295ea2a73
(macvtap: zerocopy: validate vectors before building skb).

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotuntap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS
Jason Wang [Thu, 18 Jul 2013 02:55:15 +0000 (10:55 +0800)]
tuntap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
network.

Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than
MAX_SKB_FRAGS.

This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify
vhost.

We can do further optimization on top.

The bug were introduced from commit 0690899b4d4501b3505be069b9a687e68ccbe15b
(tun: experimental zero copy tx support)

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agopkt_sched: sch_qfq: remove a source of high packet delay/jitter
Paolo Valente [Tue, 16 Jul 2013 06:52:30 +0000 (08:52 +0200)]
pkt_sched: sch_qfq: remove a source of high packet delay/jitter

QFQ+ inherits from QFQ a design choice that may cause a high packet
delay/jitter and a severe short-term unfairness. As QFQ, QFQ+ uses a
special quantity, the system virtual time, to track the service
provided by the ideal system it approximates. When a packet is
dequeued, this quantity must be incremented by the size of the packet,
divided by the sum of the weights of the aggregates waiting to be
served. Tracking this sum correctly is a non-trivial task, because, to
preserve tight service guarantees, the decrement of this sum must be
delayed in a special way [1]: this sum can be decremented only after
that its value would decrease also in the ideal system approximated by
QFQ+. For efficiency, QFQ+ keeps track only of the 'instantaneous'
weight sum, increased and decreased immediately as the weight of an
aggregate changes, and as an aggregate is created or destroyed (which,
in its turn, happens as a consequence of some class being
created/destroyed/changed). However, to avoid the problems caused to
service guarantees by these immediate decreases, QFQ+ increments the
system virtual time using the maximum value allowed for the weight
sum, 2^10, in place of the dynamic, instantaneous value. The
instantaneous value of the weight sum is used only to check whether a
request of weight increase or a class creation can be satisfied.

Unfortunately, the problems caused by this choice are worse than the
temporary degradation of the service guarantees that may occur, when a
class is changed or destroyed, if the instantaneous value of the
weight sum was used to update the system virtual time. In fact, the
fraction of the link bandwidth guaranteed by QFQ+ to each aggregate is
equal to the ratio between the weight of the aggregate and the sum of
the weights of the competing aggregates. The packet delay guaranteed
to the aggregate is instead inversely proportional to the guaranteed
bandwidth. By using the maximum possible value, and not the actual
value of the weight sum, QFQ+ provides each aggregate with the worst
possible service guarantees, and not with service guarantees related
to the actual set of competing aggregates. To see the consequences of
this fact, consider the following simple example.

Suppose that only the following aggregates are backlogged, i.e., that
only the classes in the following aggregates have packets to transmit:
one aggregate with weight 10, say A, and ten aggregates with weight 1,
say B1, B2, ..., B10. In particular, suppose that these aggregates are
always backlogged. Given the weight distribution, the smoothest and
fairest service order would be:
A B1 A B2 A B3 A B4 A B5 A B6 A B7 A B8 A B9 A B10 A B1 A B2 ...

QFQ+ would provide exactly this optimal service if it used the actual
value for the weight sum instead of the maximum possible value, i.e.,
11 instead of 2^10. In contrast, since QFQ+ uses the latter value, it
serves aggregates as follows (easy to prove and to reproduce
experimentally):
A B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 A A A A A A A A A A B1 B2 ... B10 A A ...

By replacing 10 with N in the above example, and by increasing N, one
can increase at will the maximum packet delay and the jitter
experienced by the classes in aggregate A.

This patch addresses this issue by just using the above
'instantaneous' value of the weight sum, instead of the maximum
possible value, when updating the system virtual time.  After the
instantaneous weight sum is decreased, QFQ+ may deviate from the ideal
service for a time interval in the order of the time to serve one
maximum-size packet for each backlogged class. The worst-case extent
of the deviation exhibited by QFQ+ during this time interval [1] is
basically the same as of the deviation described above (but, without
this patch, QFQ+ suffers from such a deviation all the time). Finally,
this patch modifies the comment to the function qfq_slot_insert, to
make it coherent with the fact that the weight sum used by QFQ+ can
now be lower than the maximum possible value.

[1] P. Valente, "Extending WF2Q+ to support a dynamic traffic mix",
Proceedings of AAA-IDEA'05, June 2005.

Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge tag 'driver-core-3.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 18 Jul 2013 19:48:40 +0000 (12:48 -0700)]
Merge tag 'driver-core-3.11-rc2' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core patches from Greg KH:
 "Here are some driver core patches for 3.11-rc2.  They aren't really
  bugfixes, but a bunch of new helper macros for drivers to properly
  create attribute groups, which drivers and subsystems need to fix up a
  ton of race issues with incorrectly creating sysfs files (binary and
  normal) after userspace has been told that the device is present.

  Also here is the ability to create binary files as attribute groups,
  to solve that race condition, which was impossible to do before this,
  so that's my fault the drivers were broken.

  The majority of the .c changes is indenting and moving code around a
  bit.  It affects no existing code, but allows the large backlog of 70+
  patches that I already have created to start flowing into the
  different subtrees, instead of having to live in my driver-core tree,
  causing merge nightmares in linux-next for the next few months.

  These were finalized too late for the -rc1 merge window, which is why
  they were didn't make that pull request, testing and review from
  others didn't happen until a few weeks ago, and then there's the whole
  distraction of the past few days, which prevented these from getting
  to you sooner, sorry about that.

  Oh, and there's a bugfix for the documentation build warning in here
  as well.  All of these have been in linux-next this week, with no
  reported problems"

* tag 'driver-core-3.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  driver-core: fix new kernel-doc warning in base/platform.c
  sysfs: use file mode defines from stat.h
  sysfs: add more helper macro's for (bin_)attribute(_groups)
  driver core: add default groups to struct class
  driver core: Introduce device_create_groups
  sysfs: prevent warning when only using binary attributes
  sysfs: add support for binary attributes in groups
  driver core: device.h: add RW and RO attribute macros
  sysfs.h: add BIN_ATTR macro
  sysfs.h: add ATTRIBUTE_GROUPS() macro
  sysfs.h: add __ATTR_RW() macro

10 years agoMerge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
Linus Torvalds [Thu, 18 Jul 2013 18:32:36 +0000 (11:32 -0700)]
Merge tag 'hwmon-for-linus' of git://git./linux/kernel/git/groeck/linux-staging

Pull hwmon fix from Guenter Roeck:
 "Single patch to staticize a local variable"

* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (abx500) Staticize abx500_temp_attributes

10 years agoMerge branch 'cpuinit_phase2' of git://git.kernel.org/pub/scm/linux/kernel/git/paulg...
Linus Torvalds [Thu, 18 Jul 2013 17:50:26 +0000 (10:50 -0700)]
Merge branch 'cpuinit_phase2' of git://git./linux/kernel/git/paulg/linux

Pull phase two of __cpuinit removal from Paul Gortmaker:
 "With the __cpuinit infrastructure removed earlier, this group of
  commits only removes the function/data tagging that was done with the
  various (now no-op) __cpuinit related prefixes.

  Now that the dust has settled with yesterday's v3.11-rc1, there
  hopefully shouldn't be any new users leaking back in tree, but I think
  we can leave the harmless no-op stubs there for a release as a
  courtesy to those who still have out of tree stuff and weren't paying
  attention.

  Although the commits are against the recent tag to allow for minor
  context refreshes for things like yesterday's v3.11-rc1~ slab content,
  the patches have been largely unchanged for weeks, aside from such
  trivial updates.

  For detail junkies, the largely boring and mostly irrelevant history
  of the patches can be viewed at:

    http://git.kernel.org/cgit/linux/kernel/git/paulg/cpuinit-delete.git

  If nothing else, I guess it does at least demonstrate the level of
  involvement required to shepherd such a treewide change to completion.

  This is the same repository of patches that has been applied to the
  end of the daily linux-next branches for the past several weeks"

* 'cpuinit_phase2' of git://git.kernel.org/pub/scm/linux/kernel/git/paulg/linux: (28 commits)
  block: delete __cpuinit usage from all block files
  drivers: delete __cpuinit usage from all remaining drivers files
  kernel: delete __cpuinit usage from all core kernel files
  rcu: delete __cpuinit usage from all rcu files
  net: delete __cpuinit usage from all net files
  acpi: delete __cpuinit usage from all acpi files
  hwmon: delete __cpuinit usage from all hwmon files
  cpufreq: delete __cpuinit usage from all cpufreq files
  clocksource+irqchip: delete __cpuinit usage from all related files
  x86: delete __cpuinit usage from all x86 files
  score: delete __cpuinit usage from all score files
  xtensa: delete __cpuinit usage from all xtensa files
  openrisc: delete __cpuinit usage from all openrisc files
  m32r: delete __cpuinit usage from all m32r files
  hexagon: delete __cpuinit usage from all hexagon files
  frv: delete __cpuinit usage from all frv files
  cris: delete __cpuinit usage from all cris files
  metag: delete __cpuinit usage from all metag files
  tile: delete __cpuinit usage from all tile files
  sh: delete __cpuinit usage from all sh files
  ...

10 years agoMerge tag 'sound-3.11' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
Linus Torvalds [Thu, 18 Jul 2013 17:48:48 +0000 (10:48 -0700)]
Merge tag 'sound-3.11' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "Except for a slightly big OMAP changes, all rest are small, mostly
  boring changes; all either 3.11 regression fixes or stable materials.

   - ASoC OMAP fixes due to non-DT OMAP4 removals
   - Other ASoC driver changes (sglt5000, wm8978, wm8948, samsung)
   - Fix missing locking for snd_pcm_stop() calls in many drivers
   - Fix the blocking request_module() in OSS sequencer
   - Fix old OSS vwsnd driver builds
   - Add a new HD-audio HDMI codec ID"

* tag 'sound-3.11' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound: (23 commits)
  ALSA: seq-oss: Initialize MIDI clients asynchronously
  ALSA: hda - Add new GPU codec ID to snd-hda
  staging: line6: Fix unlocked snd_pcm_stop() call
  [media] saa7134: Fix unlocked snd_pcm_stop() call
  ASoC: s6000: Fix unlocked snd_pcm_stop() call
  ASoC: atmel: Fix unlocked snd_pcm_stop() call
  ALSA: pxa2xx: Fix unlocked snd_pcm_stop() call
  ALSA: usx2y: Fix unlocked snd_pcm_stop() call
  ALSA: ua101: Fix unlocked snd_pcm_stop() call
  ALSA: 6fire: Fix unlocked snd_pcm_stop() call
  ALSA: atiixp: Fix unlocked snd_pcm_stop() call
  ALSA: asihpi: Fix unlocked snd_pcm_stop() call
  sound: oss/vwsnd: Always define vwsnd_mutex
  sound: oss/vwsnd: Add missing inclusion of linux/delay.h
  ASoC: wm8978: enable symmetric rates
  ASoC: omap-mcbsp: Use different method for DMA request when booted with DT
  ASoC: omap-dmic: Do not use platform_get_resource_byname() for DMA
  ASoC: omap-mcpdm: Do not use platform_get_resource_byname() for DMA
  ASoC: omap-pcm: Request the DMA channel differently when DT is involved
  ASoC: Samsung: Set RFS and BFS in slave mode
  ...

10 years agoMerge branch 'drm/3.11/fixes' of git://linuxtv.org/pinchartl/fbdev into drm-fixes
Dave Airlie [Thu, 18 Jul 2013 10:04:50 +0000 (20:04 +1000)]
Merge branch 'drm/3.11/fixes' of git://linuxtv.org/pinchartl/fbdev into drm-fixes

Fixes builds
* 'drm/3.11/fixes' of git://linuxtv.org/pinchartl/fbdev:
  drm/rcar-du: Use the GEM PRIME helpers
  drm/shmobile: Use the GEM PRIME helpers

10 years agomd/raid1: fix bio handling problems in process_checks()
NeilBrown [Wed, 17 Jul 2013 05:19:29 +0000 (15:19 +1000)]
md/raid1: fix bio handling problems in process_checks()

Recent change to use bio_copy_data() in raid1 when repairing
an array is faulty.

The underlying may have changed the bio in various ways using
bio_advance and these need to be undone not just for the 'sbio' which
is being copied to, but also the 'pbio' (primary) which is being
copied from.

So perform the reset on all bios that were read from and do it early.

This also ensure that the sbio->bi_io_vec[j].bv_len passed to
memcmp is correct.

This fixes a crash during a 'check' of a RAID1 array.  The crash was
introduced in 3.10 so this is suitable for 3.10-stable.

Cc: stable@vger.kernel.org (3.10)
Reported-by: Joe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: NeilBrown <neilb@suse.de>
10 years agomd: Remove recent change which allows devices to skip recovery.
NeilBrown [Wed, 17 Jul 2013 04:55:31 +0000 (14:55 +1000)]
md: Remove recent change which allows devices to skip recovery.

commit 7ceb17e87bde79d285a8b988cfed9eaeebe60b86
    md: Allow devices to be re-added to a read-only array.

allowed a bit more than just that.  It also allows devices to be added
to a read-write array and to end up skipping recovery.

This patch removes the offending piece of code pending a rewrite for a
subsequent release.

More specifically:
 If the array has a bitmap, then the device will still need a bitmap
 based resync ('saved_raid_disk' is set under different conditions
 is a bitmap is present).
 If the array doesn't have a bitmap, then this is correct as long as
 nothing has been written to the array since the metadata was checked
 by ->validate_super.  However there is no locking to ensure that there
 was no write.

Bug was introduced in 3.10 and causes data corruption so
patch is suitable for 3.10-stable.

Cc: stable@vger.kernel.org (3.10)
Reported-by: Joe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: NeilBrown <neilb@suse.de>
10 years agomd/raid10: fix two problems with RAID10 resync.
NeilBrown [Tue, 16 Jul 2013 06:50:47 +0000 (16:50 +1000)]
md/raid10: fix two problems with RAID10 resync.

1/ When an different between blocks is found, data is copied from
   one bio to the other.  However bv_len is used as the length to
   copy and this could be zero.  So use r10_bio->sectors to calculate
   length instead.
   Using bv_len was probably always a bit dubious, but the introduction
   of bio_advance made it much more likely to be a problem.

2/ When preparing some blocks for sync, we don't set BIO_UPTODATE
   except on bios that we schedule for a read.  This ensures that
   missing/failed devices don't confuse the loop at the top of
   sync_request write.
   Commit 8be185f2c9d54d6 "raid10: Use bio_reset()"
   removed a loop which set BIO_UPTDATE on all appropriate bios.
   So we need to re-add that flag.

These bugs were introduced in 3.10, so this patch is suitable for
3.10-stable, and can remove a potential for data corruption.

Cc: stable@vger.kernel.org (3.10)
Reported-by: Brassow Jonathan <jbrassow@redhat.com>
Signed-off-by: NeilBrown <neilb@suse.de>
10 years agoMerge branch 'drm-fixes-3.11' of git://people.freedesktop.org/~agd5f/linux
Dave Airlie [Thu, 18 Jul 2013 00:19:46 +0000 (10:19 +1000)]
Merge branch 'drm-fixes-3.11' of git://people.freedesktop.org/~agd5f/linux

more DPM fixes for radeon.

* 'drm-fixes-3.11' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon/dpm: add debugfs support for RS780/RS880 (v3)
  drm/radeon/dpm/atom: fix broken gcc harder
  drm/radeon/dpm/atom: restructure logic to work around a compiler bug
  drm/radeon/dpm: fix atom vram table parsing
  drm/radeon: fix an endian bug in atom table parsing
  drm/radeon: add a module parameter to disable aspm

10 years agodrm/radeon/dpm: add debugfs support for RS780/RS880 (v3)
Alex Deucher [Tue, 2 Jul 2013 17:05:23 +0000 (13:05 -0400)]
drm/radeon/dpm: add debugfs support for RS780/RS880 (v3)

This allows you to look at the current DPM state via
debugfs.

Due to the way the hardware works on these asics, there's
no way to look up exactly what power state we are in, so
we make the best guess we can based on the current sclk.

v2: Anthoine's version
v3: fix ref div

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agoMerge branch 'for-3.11' of git://linux-nfs.org/~bfields/linux
Linus Torvalds [Wed, 17 Jul 2013 20:43:55 +0000 (13:43 -0700)]
Merge branch 'for-3.11' of git://linux-nfs.org/~bfields/linux

Pull nfsd bugfixes from Bruce Fields:
 "Just three minor bugfixes"

* 'for-3.11' of git://linux-nfs.org/~bfields/linux:
  svcrdma: underflow issue in decode_write_list()
  nfsd4: fix minorversion support interface
  lockd: protect nlm_blocked access in nlmsvc_retry_blocked

10 years agodrm/radeon/dpm/atom: fix broken gcc harder
Alex Deucher [Wed, 17 Jul 2013 20:34:12 +0000 (16:34 -0400)]
drm/radeon/dpm/atom: fix broken gcc harder

See bugs:
https://bugs.freedesktop.org/show_bug.cgi?id=66932
https://bugs.freedesktop.org/show_bug.cgi?id=66972
https://bugs.freedesktop.org/show_bug.cgi?id=66945

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agoperf header: Recognize version number for perf data file
Jiri Olsa [Wed, 17 Jul 2013 17:49:47 +0000 (19:49 +0200)]
perf header: Recognize version number for perf data file

Keep the recognized data file version within 'struct perf_header'.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374083403-14591-8-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoxen-netfront: pull on receive skb may need to happen earlier
Jan Beulich [Wed, 17 Jul 2013 07:09:37 +0000 (08:09 +0100)]
xen-netfront: pull on receive skb may need to happen earlier

Due to commit 3683243b ("xen-netfront: use __pskb_pull_tail to ensure
linear area is big enough on RX") xennet_fill_frags() may end up
filling MAX_SKB_FRAGS + 1 fragments in a receive skb, and only reduce
the fragment count subsequently via __pskb_pull_tail(). That's a
result of xennet_get_responses() allowing a maximum of one more slot to
be consumed (and intermediately transformed into a fragment) if the
head slot has a size less than or equal to RX_COPY_THRESHOLD.

Hence we need to adjust xennet_fill_frags() to pull earlier if we
reached the maximum fragment count - due to the described behavior of
xennet_get_responses() this guarantees that at least the first fragment
will get completely consumed, and hence the fragment count reduced.

In order to not needlessly call __pskb_pull_tail() twice, make the
original call conditional upon the pull target not having been reached
yet, and defer the newly added one as much as possible (an alternative
would have been to always call the function right before the call to
xennet_fill_frags(), but that would imply more frequent cases of
needing to call it twice).

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: stable@vger.kernel.org (3.6 onwards)
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovxlan: add necessary locking on device removal
stephen hemminger [Sat, 13 Jul 2013 17:18:18 +0000 (10:18 -0700)]
vxlan: add necessary locking on device removal

The socket management is now done in workqueue (outside of RTNL)
and protected by vn->sock_lock. There were two possible bugs, first
the vxlan device was removed from the VNI hash table per socket without
holding lock. And there was a race when device is created and the workqueue
could run after deletion.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoperf header: Introduce feat_offset into perf_header
Jiri Olsa [Wed, 17 Jul 2013 17:49:44 +0000 (19:49 +0200)]
perf header: Introduce feat_offset into perf_header

Introducing feat_offset into perf_header to make the location of the
features section clear.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374083403-14591-5-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf header: Remove attr_offset from perf_header
Jiri Olsa [Wed, 17 Jul 2013 17:49:43 +0000 (19:49 +0200)]
perf header: Remove attr_offset from perf_header

Removing attr_offset from perf_header as it's possible to use it as a
local variable.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374083403-14591-4-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf header: Remove data_offset seek as it's not needed
Jiri Olsa [Wed, 17 Jul 2013 17:49:42 +0000 (19:49 +0200)]
perf header: Remove data_offset seek as it's not needed

Removing data_offset seek as it's not needed, because data are not read
by syscall but mmaped instead.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374083403-14591-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf session: Use session->fd instead of passing fd as argument
Jiri Olsa [Wed, 17 Jul 2013 17:49:41 +0000 (19:49 +0200)]
perf session: Use session->fd instead of passing fd as argument

Using session->fd instead of passing fd as argument because it's always
session->fd that's passed as fd argument.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374083403-14591-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agodrm/radeon/dpm/atom: restructure logic to work around a compiler bug
Andre Heider [Wed, 17 Jul 2013 18:02:23 +0000 (14:02 -0400)]
drm/radeon/dpm/atom: restructure logic to work around a compiler bug

It seems gcc 4.8.1 generates bogus code for the old logic causing
part of the function to get skipped.

Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=66932
https://bugs.freedesktop.org/show_bug.cgi?id=66972
https://bugs.freedesktop.org/show_bug.cgi?id=66945

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agodrm/radeon/dpm: fix atom vram table parsing
Alex Deucher [Wed, 17 Jul 2013 14:52:43 +0000 (10:52 -0400)]
drm/radeon/dpm: fix atom vram table parsing

Parsing the table in incorrectly led to problems with
certain asics with mclk switching.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agodrm/radeon: fix an endian bug in atom table parsing
Alex Deucher [Wed, 17 Jul 2013 14:18:52 +0000 (10:18 -0400)]
drm/radeon: fix an endian bug in atom table parsing

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agodrm/radeon: add a module parameter to disable aspm
Alex Deucher [Tue, 16 Jul 2013 19:58:50 +0000 (15:58 -0400)]
drm/radeon: add a module parameter to disable aspm

Can cause hangs when enabled in certain motherboards.
Set radeon.aspm=0 to disable aspm.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
10 years agoperf tests: Check proper prev_state size for sched_switch tp
Jiri Olsa [Fri, 12 Jul 2013 14:29:06 +0000 (16:29 +0200)]
perf tests: Check proper prev_state size for sched_switch tp

The prev_state is defined as long which is 4 bytes long on 32-bit x86.
Changing the check against sizeof(long).

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1373639346-4547-1-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoperf symbols: Do not apply symfs for an absolute vmlinux path
Namhyung Kim [Wed, 17 Jul 2013 08:08:15 +0000 (17:08 +0900)]
perf symbols: Do not apply symfs for an absolute vmlinux path

If an user gives both of --symfs and --vmlinux option, the vmlinux will
be searched under the symfs directory.  This is somewhat confusing since
vmlinux often lives in kernel build directory or somewhere other than
user space binaries.

So it'd be better not adding symfs prefix for a vmlinux if it has an
absolute pathname.

Reported-by: Kwanghyun Yoo <ykh815.yoo@lge.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1374048495-3643-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agodrm/rcar-du: Use the GEM PRIME helpers
Laurent Pinchart [Wed, 10 Jul 2013 13:23:35 +0000 (15:23 +0200)]
drm/rcar-du: Use the GEM PRIME helpers

The GEM CMA PRIME import/export helpers have been removed in favor of
generic GEM PRIME helpers with GEM CMA low-level operations. Fix the
driver accordingly.

Reported-by: Mark Brown <broonie@linaro.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Tested-by: Mark Brown <broonie@linaro.org>
10 years agodrm/shmobile: Use the GEM PRIME helpers
Laurent Pinchart [Wed, 10 Jul 2013 13:28:15 +0000 (15:28 +0200)]
drm/shmobile: Use the GEM PRIME helpers

The GEM CMA PRIME import/export helpers have been removed in favor of
generic GEM PRIME helpers with GEM CMA low-level operations. Fix the
driver accordingly.

Reported-by: Mark Brown <broonie@linaro.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Tested-by: Mark Brown <broonie@linaro.org>
10 years agoperf tools: Fix 'make tools/perf'
Robert Richter [Tue, 16 Jul 2013 14:50:36 +0000 (16:50 +0200)]
perf tools: Fix 'make tools/perf'

Boris just raised another variant of building perf tools which is
broken:

 $ make tools/perf
 ...
     LINK /home/robert/cx/linux/tools/perf/perf
 gcc: error: ../linux/tools/lib/lk/liblk.a: No such file or directory

The variant wasn't considered by:

 107de37 perf tools: Fix build errors with O and DESTDIR make vars set

There are other variant of building perf too:

 $ make -C tools perf
 $ make -C tools/perf

Plus variants with O= and DESTDIR set.

This patch fixes the above and was tested with the following:

 $ make O=... DESTDIR=... tools/perf
 $ make O=... DESTDIR=... -C tools/ perf
 $ make O=... DESTDIR=... -C tools/perf
 $ make tools/perf
 $ make -C tools/ perf
 $ make -C tools/perf

Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Robert Richter <robert.richter@linaro.org>
Signed-off-by: Robert Richter <rric@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-tip-commits@vger.kernel.org
Link: http://lkml.kernel.org/r/20130716145036.GH8731@rric.localhost
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
10 years agoALSA: seq-oss: Initialize MIDI clients asynchronously
Takashi Iwai [Tue, 16 Jul 2013 10:17:49 +0000 (12:17 +0200)]
ALSA: seq-oss: Initialize MIDI clients asynchronously

We've got bug reports that the module loading stuck on Debian system
with 3.10 kernel.  The debugging session revealed that the initial
registration of OSS sequencer clients stuck at module loading time,
which involves again with request_module() at the init phase.  This is
triggered only by special --install stuff Debian is using, but it's
still not good to have such loops.

As a workaround, call the registration part asynchronously.  This is a
better approach irrespective of the hang fix, in anyway.

Reported-and-tested-by: Philipp Matthias Hahn <pmhahn@pmhahn.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
10 years agohyperv: Fix the NETIF_F_SG flag setting in netvsc
Haiyang Zhang [Wed, 17 Jul 2013 06:01:20 +0000 (23:01 -0700)]
hyperv: Fix the NETIF_F_SG flag setting in netvsc

SG mode is not currently supported by netvsc, so remove this flag for now.
Otherwise, it will be unconditionally enabled by commit ec5f0615642
    "Kill link between CSUM and SG features"
Previously, the SG feature is disabled because CSUM is not set here.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodriver-core: fix new kernel-doc warning in base/platform.c
Randy Dunlap [Mon, 15 Jul 2013 00:43:06 +0000 (17:43 -0700)]
driver-core: fix new kernel-doc warning in base/platform.c

Fix new kernel-doc warning in drivers/base/platform.c:

Warning(drivers/base/platform.c:528): No description found for parameter 'owner'

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agolinked-list: Remove __list_for_each
Dave Jones [Wed, 17 Jul 2013 02:44:08 +0000 (22:44 -0400)]
linked-list: Remove __list_for_each

__list_for_each used to be the non prefetch() aware list walking
primitive.  When we removed the prefetch macros from the list routines,
it became redundant.  Given it does exactly the same thing as
list_for_each now, we might as well remove it and call list_for_each
directly.

All users of __list_for_each have been converted to list_for_each calls
in the current merge window.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agox86: Make jump_label use int3-based patching
Jiri Kosina [Fri, 12 Jul 2013 09:22:09 +0000 (11:22 +0200)]
x86: Make jump_label use int3-based patching

Make jump labels use text_poke_bp() for text patching instead of
text_poke_smp(), avoiding the need for stop_machine().

Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Link: http://lkml.kernel.org/r/alpine.LNX.2.00.1307121120250.29788@pobox.suse.cz
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
10 years agox86: Introduce int3 (breakpoint)-based instruction patching
Jiri Kosina [Fri, 12 Jul 2013 09:21:48 +0000 (11:21 +0200)]
x86: Introduce int3 (breakpoint)-based instruction patching

Introduce a method for run-time instruction patching on a live SMP kernel
based on int3 breakpoint, completely avoiding the need for stop_machine().

The way this is achieved:

- add a int3 trap to the address that will be patched
- sync cores
- update all but the first byte of the patched range
- sync cores
- replace the first byte (int3) by the first byte of
  replacing opcode
- sync cores

According to

http://lkml.indiana.edu/hypermail/linux/kernel/1001.1/01530.html

synchronization after replacing "all but first" instructions should not
be necessary (on Intel hardware), as the syncing after the subsequent
patching of the first byte provides enough safety.
But there's not only Intel HW out there, and we'd rather be on a safe
side.

If any CPU instruction execution would collide with the patching,
it'd be trapped by the int3 breakpoint and redirected to the provided
"handler" (which would typically mean just skipping over the patched
region, acting as "nop" has been there, in case we are doing nop -> jump
and jump -> nop transitions).

Ftrace has been using this very technique since 08d636b ("ftrace/x86:
Have arch x86_64 use breakpoints instead of stop machine") for ages
already, and jump labels are another obvious potential user of this.

Based on activities of Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
a few years ago.

Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Link: http://lkml.kernel.org/r/alpine.LNX.2.00.1307121102440.29788@pobox.suse.cz
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
10 years agonet: Fix sysfs_format_mac() code duplication.
David S. Miller [Wed, 17 Jul 2013 00:09:15 +0000 (17:09 -0700)]
net: Fix sysfs_format_mac() code duplication.

It's just a duplicate implementation of "%*phC".  Thanks to Joe
Perches for showing that we had exactly this support in the
lib/vsprintf.c code already.

Signed-off-by: David S. Miller <davem@davemloft.net>