perf, x86: Clean up event constraints code a bit

- Remove stray debug code
 - Improve ugly macros a bit
 - Remove some whitespace damage
 - (Also fix up some accumulated damage in perf_event.h)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Stephane Eranian <eranian@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>

perf_event: x86: Optimize x86_pmu_disable()

x86_pmu_disable() removes the event from the cpuc->event_list[], however
since an event can only be on that list once, stop looking after we found
it.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Optimize the fast path a little more

Remove num from the fast path and save a few ops.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155536.056430539@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Optimize constraint weight computation

Add a weight member to the constraint structure and avoid recomputing the
weight at runtime.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.963944926@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Optimize the constraint searching bits

Instead of copying bitmasks around, pass pointers to the constraint
structure.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.887853503@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

bitops: Provide compile time HWEIGHT{8,16,32,64}

Provide compile time versions of hweight.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <20100122155535.797688466@chello.nl>
[ Remove some whitespace damage while we are at it ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Reduce some overly long lines with some MACROs

Introduce INTEL_EVENT_CONSTRAINT and FIXED_EVENT_CONSTRAINT to reduce
some line length and typing work.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.688730371@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Clean up some of the u64/long bitmask casting

We need this to be u64 for direct assigment, but the bitmask functions
all work on unsigned long, leading to cast heaven, solve this by using a
union.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.595961269@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Fixup constraints typing issue

Constraints gets defined an u64 but in long quantities and then cast to
long.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.504916780@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_event: x86: Allocate the fake_cpuc

GCC was complaining the stack usage was too large, so allocate the
structure.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100122155535.411197266@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_events: Add fast-path to the rescheduling code

Implement correct fastpath scheduling, i.e., reuse previous assignment.

Signed-off-by: Stephane Eranian <eranian@google.com>
[ split from larger patch]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4b588464.1818d00a.4456.383b@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_events, x86: Improve x86 event scheduling

This patch improves event scheduling by maximizing the use of PMU
registers regardless of the order in which events are created in a group.

The algorithm takes into account the list of counter constraints for each
event. It assigns events to counters from the most constrained, i.e.,
works on only one counter, to the least constrained, i.e., works on any
counter.

Intel Fixed counter events and the BTS special event are also handled via
this algorithm which is designed to be fairly generic.

The patch also updates the validation of an event to use the scheduling
algorithm. This will cause early failure in perf_event_open().

The 2nd version of this patch follows the model used by PPC, by running
the scheduling algorithm and the actual assignment separately. Actual
assignment takes place in hw_perf_enable() whereas scheduling is
implemented in hw_perf_group_sched_in() and x86_pmu_enable().

Signed-off-by: Stephane Eranian <eranian@google.com>
[ fixup whitespace and style nits as well as adding is_x86_event() ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4b5430c6.0f975e0a.1bf9.ffff85fe@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86/hw-breakpoints: Optimize return code from notifier chain in hw_breakpoint_handler

Processing of debug exceptions in do_debug() can stop if it
originated from a hw-breakpoint exception by returning NOTIFY_STOP
in most cases.

But for certain cases such as:

a) user-space breakpoints with pending SIGTRAP signal delivery (as
in the case of ptrace induced breakpoints).

b) exceptions due to other causes than breakpoints

We will continue to process the exception by returning NOTIFY_DONE.

Signed-off-by: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Jan Kiszka <jan.kiszka@siemens.com>
LKML-Reference: <20100128111415.GC13935@in.ibm.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

x86/debug: Clear reserved bits of DR6 in do_debug()

Clear the reserved bits from the stored copy of debug status
register (DR6).
This will help easy bitwise operations such as quick testing
of a debug event origin.

Signed-off-by: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Jan Kiszka <jan.kiszka@siemens.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Ingo Molnar <mingo@elte.hu>
LKML-Reference: <20100128111401.GB13935@in.ibm.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing/kprobe: Cleanup unused return value of tracing functions

The return values of the kprobe's tracing functions are meaningless,
lets remove these.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4B60E9A3.2040505@cn.fujitsu.com>
[fweisbec@gmail: whitespace fixes, drop useless void returns in end
of functions]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

perf: Factorize trace events raw sample buffer operations

Introduce ftrace_perf_buf_prepare() and ftrace_perf_buf_submit() to
gather the common code that operates on raw events sampling buffer.
This cleans up redundant code between regular trace events, syscall
events and kprobe events.

Changelog v1->v2:
- Rename function name as per Masami and Frederic's suggestion
- Add __kprobes for ftrace_perf_buf_prepare() and make
  ftrace_perf_buf_submit() inline as per Masami's suggestion
- Export ftrace_perf_buf_prepare since modules will use it

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4B60E92D.9000808@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

perf: Fix inconsistency between IP and callchain sampling

When running perf across all cpus with backtracing (-a -g), sometimes we
get samples without associated backtraces:

    23.44%         init  [kernel]                     [k] restore
    11.46%         init                       eeba0c  [k] 0x00000000eeba0c
     6.77%      swapper  [kernel]                     [k] .perf_ctx_adjust_freq
     5.73%         init  [kernel]                     [k] .__trace_hcall_entry
     4.69%         perf  libc-2.9.so                  [.] 0x0000000006bb8c
                       |
                       |--11.11%-- 0xfffa941bbbc

It turns out the backtrace code has a check for the idle task and the IP
sampling does not. This creates problems when profiling an interrupt
heavy workload (in my case 10Gbit ethernet) since we get no backtraces
for interrupts received while idle (ie most of the workload).

Right now x86 and sh check that current is not NULL, which should never
happen so remove that too.

Idle task's exclusion must be performed from the core code, on top
of perf_event_attr:exclude_idle.

Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mundt <lethal@linux-sh.org>
LKML-Reference: <20100118054707.GT12666@kryten>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

perf trace: Add -i option for choosing input file

perf trace lacks -i option for choosing input file.
This patch adds it to perf trace.

Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1264167929-6741-1-git-send-email-mitake@dcl.info.waseda.ac.jp>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Use the right variable to check for kallsyms in the cache

Probably this wasn't noticed when testing this on my parisc
machine because I must have copied manually to its cache the
vmlinux file used in the x86_64 machine, now that I tried
looking on a x86-32 machine with a fresh cache, kernel symbols
weren't being resolved even with the right kallsyms copy on its
cache, duh.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1264178102-4203-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Fix inverted logic for showing kallsyms as the source of symbols

Only if we parsed /proc/kallsyms (or a copy found in the buildid
cache) we should set the dso long name to "[kernel.kallsyms]".

Reported-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1264178102-4203-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf top: Handle PERF_RECORD_{FORK,EXIT} events

As noticed by Mike, symbols in new tasks were not being
processed as we weren't processing these events.

Reported-by: Mike Galbraith <efault@gmx.de>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1264086284-1431-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf top: Fix sample counting

Broken since "5b2bb75 perf top: Support userspace symbols too".

Reported-by: Mike Galbraith <efault@gmx.de>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1264086284-1431-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Reimplement frequency driven sampling

There was a bug in the old period code that caused intel_pmu_enable_all()
or native_write_msr_safe() to show up quite high in the profiles.

In staring at that code it made my head hurt, so I rewrote it in a
hopefully simpler fashion. Its now fully symetric between tick and
overflow driven adjustments and uses less data to boot.

The only complication is that it basically wants to do a u128 division.
The code approximates that in a rather simple truncate until it fits
fashion, taking care to balance the terms while truncating.

This version does not generate that sampling artefact.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf buildid-cache: Add new command to manage build-id cache

For now it just has operations to examine a given file, find its
build-id and add or remove it to/from the cache.

Useful, for instance, when adding binaries sent together with a
perf.data file, so that we can add them to the cache and have
the tools find it when resolving symbols.

It'll also manage the size of the cache like 'ccache' does.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1264008525-29025-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Use only --vmlinux if specified

Found while analysing a perf.data file collected on an ARM
machine where an explicitely specified vmlinux was being
disregarded.

Reported-by: Jamie Iles <jamie.iles@picochip.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263904574-30732-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Set dso->kernel when handling the fake vmlinux MMAP event

Because it may be possible that there was no buildid section,
where we would set this to 1.

Found while analysing a perf.data file collected on an ARM
machine where an explicitely specified vmlinux was being
disregarded.

Reported-by: Jamie Iles <jamie.iles@picochip.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263904574-30732-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf archive: Fix installation steps in the Makefile

Fix these warning:

acme@parisc:~/git/linux-2.6-tip$ make -C tools/perf/ install
make: Entering directory
`/home/acme/git/linux-2.6-tip/tools/perf' Makefile:833: warning:
overriding commands for target `perf-archive' Makefile:822:
warning: ignoring old commands for target `perf-archive'

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263846102-24841-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf archive: Add documentation

This also makes it appear on the 'perf --help' output, i.e.
util/generate-cmdlist.sh now takes it into account.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263837559-24168-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kprobetrace, doc: Add the explanation to remove probe points

Latest kprobetrace can remove probe points selectively, thus
the documentation should be updated too.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20100119023512.31880.35535.stgit@localhost6.localdomain6>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kprobetrace, doc: Shell needs single quote to use $ character

Shell interprets $val as shell variable, thus we need quote if
we use the echo command.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20100119023505.31880.17367.stgit@localhost6.localdomain6>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Merge branch 'perf/scheduling' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing into perf/core

perf: Better order flexible and pinned scheduling

When a task gets scheduled in. We don't touch the cpu bound events
so the priority order becomes:

cpu pinned, cpu flexible, task pinned, task flexible.

So schedule out cpu flexibles when a new task context gets in
and correctly order the groups to schedule in:

task pinned, cpu flexible, task flexible.

Cpu pinned groups don't need to be touched at this time.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf: Don't schedule out/in pinned events on task tick

We don't need to schedule in/out pinned events on task tick,
now that pinned and flexible groups can be scheduled separately.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf: Allow pinned and flexible groups to be scheduled separately

Tune the scheduling helpers so that we can choose to schedule either
pinned and/or flexible groups from a context.

And while at it, refactor a bit the naming of these helpers to make
these more consistent and flexible.

There is no (intended) change in scheduling behaviour in this
patch.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf: Make __perf_event_sched_out static

__perf_event_sched_out doesn't need to be globally available, make
it static.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

tracing/kprobe: Update kprobe tracing self test for new syntax

Update kprobe tracing self test for new syntax (it supports
deleting individual probes, and drops $argN support)
and behavior change (new probes are disabled in default).

This selftest includes the following checks:

 - Adding function-entry probe and return probe with arguments.
 - Enabling these probes.
 - Deleting it individually.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20100114051211.7814.29436.stgit@localhost6.localdomain6>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf probe: Fix build error of builtin-probe.c

I got this build error when building tip tree:

| cc1: warnings being treated as errors
| builtin-probe.c:123: error: 'opt_show_lines' defined but not used

This error is caused by:

| #ifndef NO_LIBDWARF
| OPT_CALLBACK('L', "line", NULL,
|      "FUNC[:RLN[+NUM|:RLN2]]|SRC:ALN[+NUM|:ALN2]",
|      "Show source code lines.", opt_show_lines),
| #endif

My environment defines NO_LIBDWARF, so gcc treated
opt_show_lines() as garbage. So I moved opt_show_lines() into
#ifndef NO_LIBDWARF ... #endif block.

Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1263645076-9993-1-git-send-email-mitake@dcl.info.waseda.ac.jp>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Fix implicit declaration of getline in util.c

getline() is considered as undeclared in util/util.c because
it includes string.h, that in turn includes stdio.h, without
having defined _GNU_SOURCE.

But util.c also includes util.h that handles the _GNU_SOURCE and
all the needed inclusions already. Let's include only util.h
and sys/mman.h which is the only one header not handled by
util.h

This fixes the following build error:

 util/util.c: In function 'slow_copyfile':
 util/util.c:49: erreur: implicit declaration of function
 'getline' util/util.c:49: erreur: nested extern declaration of 'getline'

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263648075-3858-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Export software-only event group characteristic as a flag

Before scheduling an event group, we first check if a group can go
on. We first check if the group is made of software only events
first, in which case it is enough to know if the group can be
scheduled in.

For that purpose, we iterate through the whole group, which is
wasteful as we could do this check when we add/delete an event to
a group.

So we create a group_flags field in perf event that can host
characteristics from a group of events, starting with a first
PERF_GROUP_SOFTWARE flag that reduces the check on the fast path.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf: Round robin flexible groups of events using list_rotate_left()

This is more proper that doing it through a list_for_each_entry()
that breaks after the first entry.

v2: Don't rotate pinned groups as its not needed to time share
them.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

list: Introduce list_rotate_left()

Bring a new list_rotate_left() helper that rotates a list to
the left. This is useful for codes that need to round roubin
elements which queue priority increases from tail to head.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf/core: Split context's event group list into pinned and non-pinned lists

Split-up struct perf_event_context::group_list into pinned_groups
and flexible_groups (non-pinned).

This first appears to be useless as it duplicates various loops around
the group list handlings.

But it scales better in the fast-path in perf_sched_in(). We don't
anymore iterate twice through the entire list to separate pinned and
non-pinned scheduling. Instead we interate through two distinct lists.

The another desired effect is that it makes easier to define distinct
scheduling rules on both.

Changes in v2:
- Respectively rename pinned_grp_list and
  volatile_grp_list into pinned_groups and flexible_groups as per
  Ingo suggestion.
- Various cleanups

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>

perf symbols: Accept an alias when looking for "_text"

As it is in PARISC64:

parisc:~# uname -a
Linux parisc 2.6.33-rc4-tip+ #1 SMP Thu Jan 14 13:33:34 BRST
2010 parisc64 GNU/Linux parisc:~# grep -w _text /proc/kallsyms
0000000040100000 A _text
parisc:~# grep 0000000040100000 /proc/kallsyms
0000000040100000 T stext
0000000040100000 T _stext
0000000040100000 A _text
parisc:~#

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263586107-1756-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf record: Intercept all events

The event interception we need to do in 'perf record' to create
a list of all DSOs in PERF_RECORD_MMAP events wasn't seeing all
events, make sure that happens by checking size agains
event_t->header.size.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263586107-1756-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf archive: Add helper script to package files needed to do analysis

It uses 'perf buildid-list --with-hits' to create a tarball with
what is needed to have in the destination machine ~/.debug
hierarchy to properly decode the perf.data file specified.

Here is an example where a perf.data file collected on a x86-64
machine running Fedora 12 is used and then the data is packaged,
transferred and decoded on a PARISC64 machine running Debian
Testing, 32-bit userspace:

[root@doppio linux-2.6-tip]# uname -a
Linux doppio.ghostprotocols.net 2.6.33-rc4-tip+ #3 SMP Wed Jan 13 11:58:15 BRST 2010 x86_64 x86_64 x86_64 GNU/Linux
[root@doppio linux-2.6-tip]# perf archive
[root@doppio linux-2.6-tip]# ls -la perf.data*
-rw------- 1 root root  737696 2010-01-14 23:36 perf.data
-rw-r--r-- 1 root root 8840025 2010-01-15 12:27 perf.data.tar.bz2
[root@doppio linux-2.6-tip]# scp perf.data.* parisc64:.
Password:
perf.data.tar.bz2                                      100% 8633KB   1.4MB/s   00:06
[root@doppio linux-2.6-tip]# ssh parisc64
Password:
Linux parisc 2.6.19-g2bbf29ac-dirty #1 Sun Dec 3 17:24:04 BRST 2006 parisc64

The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/*/copyright.

Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
Last login: Thu Jan 14 11:23:24 2010 from d
parisc:~# uname -a
Linux parisc 2.6.19-g2bbf29ac-dirty #1 Sun Dec 3 17:24:04 BRST 2006 parisc64 GNU/Linux
parisc:~# mkdir .debug
parisc:~# tar xvf perf.data.tar.bz2 -C ~/.debug
tar: Record size = 8 blocks
.build-id/74/f9930ee94475b6b3238caf3725a50d59cb994b
[kernel.kallsyms]/74f9930ee94475b6b3238caf3725a50d59cb994b
.build-id/9f/fdcac0a7935922d1f04b6cc9029dfef0f066ef
lib/modules/2.6.33-rc4-tip+/kernel/arch/x86/crypto/aes-x86_64.ko/9ffdcac0a7935922d1f04b6cc9029dfef0f066ef
.build-id/3a/af89c32ebfc438ff546c93597d41788e3e65f3
lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/wireless/iwlwifi/iwl3945.ko/3aaf89c32ebfc438ff546c93597d41788e3e65f3
.build-id/19/f46033f73e1ec612937189bb118c5daba5a0c8
lib/modules/2.6.33-rc4-tip+/kernel/net/mac80211/mac80211.ko/19f46033f73e1ec612937189bb118c5daba5a0c8
.build-id/17/72f014a7a7272859655acb0c64a20ab20b75ee
lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/e1000e/e1000e.ko/1772f014a7a7272859655acb0c64a20ab20b75ee
.build-id/eb/4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1
lib64/libc-2.10.2.so/eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1
.build-id/5c/68f7afeb33309c78037e374b0deee84dd441f6
lib64/libpthread-2.10.2.so/5c68f7afeb33309c78037e374b0deee84dd441f6
.build-id/e9/c9ad5c138ef882e4507d2605645b597da43873
bin/dbus-daemon/e9c9ad5c138ef882e4507d2605645b597da43873
.build-id/bc/da7d09eb6c9ee380dae0ed3d591d4311decc31
lib64/libdbus-1.so.3.4.0/bcda7d09eb6c9ee380dae0ed3d591d4311decc31
.build-id/7c/c449a77f48b85d6088114000e970ced613bed8
usr/lib64/libcrypto.so.0.9.8k/7cc449a77f48b85d6088114000e970ced613bed8
.build-id/fd/d1ccd1ff7917ab020653147ab3bacf0a85b5b9
lib64/libglib-2.0.so.0.2000.5/fdd1ccd1ff7917ab020653147ab3bacf0a85b5b9
.build-id/e4/417ebb8762e5f2eee93c8011a71115ff5edad8
lib64/libgobject-2.0.so.0.2000.5/e4417ebb8762e5f2eee93c8011a71115ff5edad8
.build-id/93/1e49461f6df99104f0febcc52f6fed5e2efce6
usr/sbin/sshd/931e49461f6df99104f0febcc52f6fed5e2efce6
.build-id/da/b5f724c088f89fbd8304da553ed6cb30bbec96
usr/lib64/libgdk-x11-2.0.so.0.1600.6/dab5f724c088f89fbd8304da553ed6cb30bbec96
.build-id/f2/037a091ef36b591187a858d75e203690ea9409
usr/sbin/openvpn/f2037a091ef36b591187a858d75e203690ea9409
.build-id/a8/e4f743b40fb1fd8b85e2f9b88d93b661472b8f
bin/find/a8e4f743b40fb1fd8b85e2f9b88d93b661472b8f
.build-id/81/120aada06e68b1e85882925a0fc6d7345ef59a
home/acme/bin/perf/81120aada06e68b1e85882925a0fc6d7345ef59a
parisc:~# perf report 2> /dev/null | head -25
     9.07%             find  find                               [.] 0x0000000000fb0e
     3.29%             perf  libc-2.10.2.so                     [.] __GI_strcmp
     3.19%             find  [kernel.kallsyms]                  [k] _raw_spin_unlock_irqrestore
     2.70%             find  libc-2.10.2.so                     [.] __GI_memmove
     2.62%             perf  [kernel.kallsyms]                  [k] vsnprintf
     2.03%             find  libc-2.10.2.so                     [.] _int_malloc
     2.02%             perf  [kernel.kallsyms]                  [k] format_decode
     1.70%             find  [kernel.kallsyms]                  [k] n_tty_write
     1.70%             find  [kernel.kallsyms]                  [k] half_md4_transform
     1.67%             find  libc-2.10.2.so                     [.] _IO_vfprintf_internal
     1.66%             perf  [kernel.kallsyms]                  [k] audit_free_aux
     1.62%          swapper  [kernel.kallsyms]                  [k] mwait_idle_with_hints
     1.58%             find  [kernel.kallsyms]                  [k] __kmalloc
     1.35%             find  [kernel.kallsyms]                  [k] sched_clock_local
     1.35%             find  [kernel.kallsyms]                  [k] ext4_check_dir_entry
     1.35%             find  [kernel.kallsyms]                  [k] ext4_htree_store_dirent
     1.35%             find  [kernel.kallsyms]                  [k] sys_write
     1.35%             find  [e1000e]                           [k] e1000_clean
     1.35%             find  [kernel.kallsyms]                  [k] _atomic_dec_and_lock
     1.34%             find  [kernel.kallsyms]                  [k] __d_lookup
parisc:~#

Probably the next step is to have 'perf report' notice that there is a
perf.data.tar.bz2 file in the same directory and look if it was already
added to ~/.debug/.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263568672-30323-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: The synthesized kernel modules MMAP must use the pathnames

Since we use ->long_name in dsos__find now.

Now 'perf buildid_list' is not duplicating those and managing to
show the proper build-ids for the DSOs with hits:

[root@doppio linux-2.6-tip]# perf buildid-list -H
74f9930ee94475b6b3238caf3725a50d59cb994b [kernel.kallsyms]
9ffdcac0a7935922d1f04b6cc9029dfef0f066ef /lib/modules/2.6.33-rc4-tip+/kernel/arch/x86/crypto/aes-x86_64.ko
3aaf89c32ebfc438ff546c93597d41788e3e65f3 /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/wireless/iwlwifi/iwl3945.ko
19f46033f73e1ec612937189bb118c5daba5a0c8 /lib/modules/2.6.33-rc4-tip+/kernel/net/mac80211/mac80211.ko
1772f014a7a7272859655acb0c64a20ab20b75ee /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/e1000e/e1000e.ko
eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 /lib64/libc-2.10.2.so
5c68f7afeb33309c78037e374b0deee84dd441f6 /lib64/libpthread-2.10.2.so
e9c9ad5c138ef882e4507d2605645b597da43873 /bin/dbus-daemon
bcda7d09eb6c9ee380dae0ed3d591d4311decc31 /lib64/libdbus-1.so.3.4.0
7cc449a77f48b85d6088114000e970ced613bed8 /usr/lib64/libcrypto.so.0.9.8k
fdd1ccd1ff7917ab020653147ab3bacf0a85b5b9 /lib64/libglib-2.0.so.0.2000.5
e4417ebb8762e5f2eee93c8011a71115ff5edad8 /lib64/libgobject-2.0.so.0.2000.5
931e49461f6df99104f0febcc52f6fed5e2efce6 /usr/sbin/sshd
dab5f724c088f89fbd8304da553ed6cb30bbec96 /usr/lib64/libgdk-x11-2.0.so.0.1600.6
f2037a091ef36b591187a858d75e203690ea9409 /usr/sbin/openvpn
a8e4f743b40fb1fd8b85e2f9b88d93b661472b8f /bin/find
81120aada06e68b1e85882925a0fc6d7345ef59a /home/acme/bin/perf
[root@doppio linux-2.6-tip]#

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263568672-30323-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf buildid-list: Introduce --with-hits option

Using this option 'perf buildid-list' will process all samples,
marking the DSOs that had some hits to list just them.

This in turn will be used by a new porcelain, 'perf archive',
that will be just a shell script to create a tarball from the
'perf buildid-list --with-hits' output and the files cached by
'perf record' in ~/.debug.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263519930-22803-4-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Create thread__find_addr_map from thread__find_addr_location

Because some tools will only want to know with maps had hits,
not needing the full symbol resolution done by
thread__find_addr_location.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263519930-22803-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf record: Encode the domain while synthesizing MMAP events

In the past 'perf record' had to process only userspace MMAP
events, the ones generated in the kernel, but after we reused
the MMAP events to encode the module mapings we ended up adding
them first to the list of userspace DSOs (dsos__user) and to the
kernel one (dsos__kernel).

Fix this by encoding the header.misc field and then using it,
like other parts to decide the right DSOs list to insert/find.

The gotcha here is that since the kernel puts zero in .misc,
which isn't PERF_RECORD_MISC_KERNEL (1 << 1), to differentiate,
we put 1 in .misc.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263519930-22803-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Use dso->long_name in dsos__find()

If not we end up duplicating the module DSOs because first we
insert them using the short name found in /proc/modules, then,
when processing synthesized MMAP events we add them again.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263519930-22803-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Cache /proc/kallsyms files by build-id

So that when we don't have a vmlinux handy we can store the
kallsyms for later use by 'perf report'.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263501006-14185-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Don't try to load kallsyms if doesn't match the record build-id

Now a perf.data file collected on a x86_64 fedora 12 machine
gets properly parsed on a Debian testing PARISC64 machine with
32-bit userland:

  acme@parisc:~/git/linux-2.6-tip$ perf report 2> /dev/null | head -15
  # Samples: 293085637
  #
  # Overhead          Command                      Shared Object  Symbol
  # ........  ...............  .................................  ......
  #
      35.11%             find  [kernel.kallsyms]                  [k] 0xffffffff81002b5a
      18.25%             perf  [kernel.kallsyms]                  [k] 0xffffffff8102235f
       9.07%             find  find                               [.] 0x0000000000fb0e
       5.80%          swapper  [kernel.kallsyms]                  [k] 0xffffffff8102235f
       3.29%             perf  libc-2.10.2.so                     [.] __GI_strcmp
       2.70%             find  libc-2.10.2.so                     [.] __GI_memmove
       2.33%             init  [kernel.kallsyms]                  [k] 0xffffffff810091b9
       2.03%             find  libc-2.10.2.so                     [.] _int_malloc
       1.67%             find  libc-2.10.2.so                     [.] _IO_vfprintf_internal
       1.65%             sshd  libcrypto.so.0.9.8k                [.] 0x00000000105440
  acme@parisc:~/git/linux-2.6-tip$

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263501006-14185-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Convert getpagesize() uses to sysconf(_SC_GETPAGESIZE)

Using the more portable and equivalent sysconf call.

Reported-by: Aristeu Rozanski <aris@redhat.com>
Reported-by: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Aristeu Rozanski <aris@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ulrich Drepper <drepper@redhat.com>
LKML-Reference: <1263501006-14185-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Cross platform perf.data analysis support

There are still some problems related to loading vmlinux files,
but those are unrelated to the feature implemented in this
patch, so will get fixed in the next patches, but here are some
results:

1. collect perf.data file on a Fedora 12 machine, x86_64, 64-bit
userland

2. transfer it to a Debian Testing machine, PARISC64, 32-bit
userland

  acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | head -5
  74f9930ee94475b6b3238caf3725a50d59cb994b [kernel.kallsyms]
  55fdd56670453ea66c011158c4b9d30179c1d049 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/ipt_MASQUERADE.ko
  41adff63c730890480980d5d8ba513f1c216a858 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/iptable_nat.ko
  90a33def1077bb8e97b8a78546dc96c2de62df46 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/nf_nat.ko
  984c7bea90ce1376d5c8e7ef43a781801286e62d /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/tun.ko

  acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | tail -5
  22492f3753c6a67de5c7ccbd6b863390c92c0723 /usr/lib64/libXt.so.6.0.0
  353802bb7e1b895ba43507cc678f951e778e4c6f /usr/lib64/libMagickCore.so.2.0.0
  d10c2897558595efe7be8b0584cf7e6398bc776c /usr/lib64/libfprint.so.0.0.0
  a83ecfb519a788774a84d5ddde633c9ba56c03ab /home/acme/bin/perf
  d3ca765a8ecf257d263801d7ad8c49c189082317 /usr/lib64/libdwarf.so.0.0
  acme@parisc:~/git/linux-2.6-tip$

  acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm
  The file [kernel.kallsyms] cannot be used, trying to use /proc/kallsyms...

  ^^^^ The problem related to vmlinux handling, it shouldn't be trying this
  ^^^^ rather alien /proc/kallsyms at all...

  /lib64/libpthread-2.10.2.so with build id 5c68f7afeb33309c78037e374b0deee84dd441f6 not found, continuing without symbols
  /lib64/libc-2.10.2.so with build id eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 not found, continuing without symbols
  /home/acme/bin/perf with build id a83ecfb519a788774a84d5ddde633c9ba56c03ab not found, continuing without symbols
  /usr/sbin/openvpn with build id f2037a091ef36b591187a858d75e203690ea9409 not found, continuing without symbols
  Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/e1000e/e1000e.ko, continuing without symbols
  Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/wireless/iwlwifi/iwlcore.ko, continuing without symbols

  <SNIP more complaints about not finding the right build-ids,
        those will have to wait for 'perf archive' or plain
        copying what was collected by 'perf record' on the x86_64,
        source machine, see further below for an example of this >

  # Samples: 293085637
  #
  # Overhead          Command
  # ........  ...............
  #
      61.70%             find
      23.50%             perf
       5.86%          swapper
       3.12%             sshd
       2.39%             init
       0.87%             bash
       0.86%            sleep
       0.59%      dbus-daemon
       0.25%             hald
       0.24%   NetworkManager
       0.19%  hald-addon-rfki
       0.15%          openvpn
       0.07%             phy0
       0.07%         events/0
       0.05%          iwl3945
       0.05%         events/1
       0.03%      kondemand/0
  acme@parisc:~/git/linux-2.6-tip$

Which matches what we get when running the same command for the
same perf.data file on the F12, x86_64, source machine:

  [root@doppio linux-2.6-tip]# perf report --sort comm
  # Samples: 293085637
  #
  # Overhead          Command
  # ........  ...............
  #
      61.70%             find
      23.50%             perf
       5.86%          swapper
       3.12%             sshd
       2.39%             init
       0.87%             bash
       0.86%            sleep
       0.59%      dbus-daemon
       0.25%             hald
       0.24%   NetworkManager
       0.19%  hald-addon-rfki
       0.15%          openvpn
       0.07%             phy0
       0.07%         events/0
       0.05%          iwl3945
       0.05%         events/1
       0.03%      kondemand/0
  [root@doppio linux-2.6-tip]#

The other modes work as well, modulo the problem with vmlinux:

  acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso 2> /dev/null | head -15
  # Samples: 293085637
  #
  # Overhead          Command                      Shared Object
  # ........  ...............  .................................
  #
      35.11%             find                   ffffffff81002b5a
      18.25%             perf                   ffffffff8102235f
      16.17%             find  libc-2.10.2.so
       9.07%             find  find
       5.80%          swapper                   ffffffff8102235f
       3.95%             perf  libc-2.10.2.so
       2.33%             init                   ffffffff810091b9
       1.65%             sshd  libcrypto.so.0.9.8k
       1.35%             find  [e1000e]
       0.68%            sleep  libc-2.10.2.so
  acme@parisc:~/git/linux-2.6-tip$

And the lack of the right buildids:

  acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso,symbol 2> /dev/null | head -15
  # Samples: 293085637
  #
  # Overhead          Command                      Shared Object  Symbol
  # ........  ...............  .................................  ......
  #
      35.11%             find                   ffffffff81002b5a  [k] 0xffffffff81002b5a
      18.25%             perf                   ffffffff8102235f  [k] 0xffffffff8102235f
      16.17%             find  libc-2.10.2.so                     [.] 0x00000000045782
       9.07%             find  find                               [.] 0x0000000000fb0e
       5.80%          swapper                   ffffffff8102235f  [k] 0xffffffff8102235f
       3.95%             perf  libc-2.10.2.so                     [.] 0x0000000007f398
       2.33%             init                   ffffffff810091b9  [k] 0xffffffff810091b9
       1.65%             sshd  libcrypto.so.0.9.8k                [.] 0x00000000105440
       1.35%             find  [e1000e]                           [k] 0x00000000010948
       0.68%            sleep  libc-2.10.2.so                     [.] 0x0000000011ad5b
  acme@parisc:~/git/linux-2.6-tip$

But if we:

  acme@parisc:~/git/linux-2.6-tip$ ls ~/.debug
  ls: cannot access /home/acme/.debug: No such file or directory
  acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/lib64/libc-2.10.2.so/
  acme@parisc:~/git/linux-2.6-tip$ scp doppio:.debug/lib64/libc-2.10.2.so/* ~/.debug/lib64/libc-2.10.2.so/
  acme@doppio's password:
  eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1              100% 1783KB 714.7KB/s   00:02
  acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/.build-id/eb
  acme@parisc:~/git/linux-2.6-tip$ ln -s ../../lib64/libc-2.10.2.so/eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 ~/.debug/.build-id/eb/4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1
  acme@parisc:~/git/linux-2.6-tip$ perf report --dsos libc-2.10.2.so 2> /dev/null
  # dso: libc-2.10.2.so
  # Samples: 64281170
  #
  # Overhead          Command  Symbol
  # ........  ...............  ......
  #
      14.98%             perf  [.] __GI_strcmp
      12.30%             find  [.] __GI_memmove
       9.25%             find  [.] _int_malloc
       7.60%             find  [.] _IO_vfprintf_internal
       6.10%             find  [.] _IO_new_file_xsputn
       6.02%             find  [.] __GI_close
       3.08%             find  [.] _IO_file_overflow_internal
       3.08%             find  [.] malloc_consolidate
       3.08%             find  [.] _int_free
       3.08%             find  [.] __strchrnul
       3.08%             find  [.] __getdents64
       3.08%             find  [.] __write_nocancel
       3.08%            sleep  [.] __GI__dl_addr
       3.08%             sshd  [.] __libc_select
       3.08%             find  [.] _IO_new_file_write
       3.07%             find  [.] _IO_new_do_write
       3.06%             find  [.] __GI___errno_location
       3.05%             find  [.] __GI___libc_malloc
       3.04%             perf  [.] __GI_memcpy
       1.71%             find  [.] __fprintf_chk
       1.29%             bash  [.] __gconv_transform_utf8_internal
       0.79%      dbus-daemon  [.] __GI_strlen
  #
  # (For a higher level overview, try: perf report --sort comm,dso)
  #
  acme@parisc:~/git/linux-2.6-tip$

Which matches what we get on the source, F12, x86_64 machine:

  [root@doppio linux-2.6-tip]# perf report --dsos libc-2.10.2.so
  # dso: libc-2.10.2.so
  # Samples: 64281170
  #
  # Overhead          Command  Symbol
  # ........  ...............  ......
  #
      14.98%             perf  [.] __GI_strcmp
      12.30%             find  [.] __GI_memmove
       9.25%             find  [.] _int_malloc
       7.60%             find  [.] _IO_vfprintf_internal
       6.10%             find  [.] _IO_new_file_xsputn
       6.02%             find  [.] __GI_close
       3.08%             find  [.] _IO_file_overflow_internal
       3.08%             find  [.] malloc_consolidate
       3.08%             find  [.] _int_free
       3.08%             find  [.] __strchrnul
       3.08%             find  [.] __getdents64
       3.08%             find  [.] __write_nocancel
       3.08%            sleep  [.] __GI__dl_addr
       3.08%             sshd  [.] __libc_select
       3.08%             find  [.] _IO_new_file_write
       3.07%             find  [.] _IO_new_do_write
       3.06%             find  [.] __GI___errno_location
       3.05%             find  [.] __GI___libc_malloc
       3.04%             perf  [.] __GI_memcpy
       1.71%             find  [.] __fprintf_chk
       1.29%             bash  [.] __gconv_transform_utf8_internal
       0.79%      dbus-daemon  [.] __GI_strlen
  #
  # (For a higher level overview, try: perf report --sort comm,dso)
  #
  [root@doppio linux-2.6-tip]#

So I think this is really, really nice in that it demonstrates
the portability of perf.data files and the use of build-ids
accross such aliens worlds :-)

There are some things to fix tho, like the bitmap on the header,
but things are looking good.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263478990-8200-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Don't cast RIP to pointers

Since they can come from another architecture with bigger
pointers, i.e. processing a 64-bit perf.data on a 32-bit arch.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263478990-8200-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf top: Fix code typo in prompt_symbol()

sym_filter is what was (if ever) passed with -s option. What was
typed by user, and what we were looking for, is in buf.

Signed-off-by: Kirill Smelkov <kirr@landau.phys.spbu.ru>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <1263396139-4798-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Fix few typos + cosmetics

Signed-off-by: Kirill Smelkov <kirr@landau.phys.spbu.ru>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <1263396139-4798-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Encode kernel module mappings in perf.data

We were always looking at the running machine /proc/modules,
even when processing a perf.data file, which only makes sense
when we're doing 'perf record' and 'perf report' on the same
machine, and in close sucession, or if we don't use modules at
all, right Peter? ;-)

Now, at 'perf record' time we read /proc/modules, find the long
path for modules, and put them as PERF_MMAP events, just like we
did to encode the reloc reference symbol for vmlinux. Talking
about that now it is encoded in .pgoff, so that we can use
.{start,len} to store the address boundaries for the kernel so
that when we reconstruct the kmaps tree we can do lookups right
away, without having to fixup the end of the kernel maps like we
did in the past (and now only in perf record).

One more step in the 'perf archive' direction when we'll finally
be able to collect data in one machine and analyse in another.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263396139-4798-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Make cmd_to_page() function more compact

Remove branch for is_perf_command.

Signed-off-by: Wenji Huang <wenji.huang@oracle.com>
Cc: fweisbec@gmail.com
Cc: jkacur@redhat.com
Cc: acme@redhat.com
LKML-Reference: <1263373298-13282-1-git-send-email-wenji.huang@oracle.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

sched/perf: Make sure irqs are disabled for perf_event_task_sched_in()

perf_event_task_sched_in() expects interrupts to be disabled,
but on architectures with __ARCH_WANT_INTERRUPTS_ON_CTXSW
defined, this isn't true. If this is defined, disable irqs
around the call in finish_task_switch().

Signed-off-by: Jamie Iles <jamie.iles@picochip.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
LKML-Reference: <1262964453-27370-1-git-send-email-jamie.iles@picochip.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Create typedef for common event synthesizing callback

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262901583-8074-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Record the domain of DSOs in HEADER_BUILD_ID header table

So that we can restore them to the right DSO list (either
dsos__kernel or dsos__user).

We do that just like the kernel does for the other events,
encoding PERF_RECORD_MISC_{KERNEL,USER} in perf_event_header.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262901583-8074-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf buildid-list: No need to process the header sections again

As it is already processed by:

perf_session__new
perf_session__open
perf_session__read

This was harmless, because we use dsos__findnew, that would
already find it, but is unnecessary work and removing it makes
builtin-buildid-list.c even shorter.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262901583-8074-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Fix memory leak: counterwidth

Signed-off-by: Alexander Beregalov <a.beregalov@gmail.com>
Cc: a.p.zijlstra@chello.nl
Cc: paulus@samba.org
LKML-Reference: <1262882447-23776-2-git-send-email-a.beregalov@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf probe: Support --line option to show probable source-code lines

Add --line option to support showing probable source-code lines.

  perf probe --line SRC:LN[-LN|+NUM]
   or
  perf probe --line FUNC[:LN[-LN|+NUM]]

This option shows source-code with line number if the line can
be probed. Lines without line number (and blue color) means that
the line can not be probed, because debuginfo doesn't have the
information of those lines.

The argument specifies the range of lines, "source.c:100-120"
shows lines between 100th to l20th in source.c file. And
"func:10+20" shows 20 lines from 10th line of func function.

e.g.
 # ./perf probe --line kernel/sched.c:1080
 <kernel/sched.c:1080>
          *
          * called with rq->lock held and irqs disabled
          */
         static void hrtick_start(struct rq *rq, u64 delay)
         {
                struct hrtimer *timer = &rq->hrtick_timer;
   1086         ktime_t time = ktime_add_ns(timer->base->get_time(), delay);

                hrtimer_set_expires(timer, time);

   1090         if (rq == this_rq()) {
   1091                 hrtimer_restart(timer);
   1092         } else if (!rq->hrtick_csd_pending) {
   1093                 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd,
   1094                 rq->hrtick_csd_pending = 1;

If you specifying function name, this shows function-relative
line number.

 # ./perf probe --line schedule
 <schedule:0>
         asmlinkage void __sched schedule(void)
      1  {
                struct task_struct *prev, *next;
                unsigned long *switch_count;
                struct rq *rq;
                int cpu;

         need_resched:
                preempt_disable();
      9         cpu = smp_processor_id();
     10         rq = cpu_rq(cpu);
     11         rcu_sched_qs(cpu);
     12         prev = rq->curr;
     13         switch_count = &prev->nivcsw;

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20100106144534.27218.77939.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Enhance glob string matching

Enhance strglobmatch() for supporting character classes([CHARS],
complementation and ranges are also supported) and escaped
special characters (\*, \? etc).

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20100105224724.19431.56271.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Support tracepoint glob matching

Support glob wildcard when selecting tracepoint events by -e
option. Without this patch, perf-tools supports 'GROUP:*:record'
syntax for selecting all tracepoints under GROUP group.

With this patch, user can choose tracepoints more flexibly by using
partial wildcards, e.g. 'block:*bio*:record'.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20100105224717.19431.68972.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf probe: Show probe list in pager

Show probe list in pager, because the list can be longer than
a page.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20100105224710.19431.61542.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf probe: Remove newline from die()

Remove newline from die(), because it is automatically added.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20100105224703.19431.42475.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86/ptrace: Remove unused regs_get_argument_nth API

Because of dropping function argument syntax from kprobe-tracer,
we don't need this API anymore.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Neuling <mikey@neuling.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: linuxppc-dev@ozlabs.org
LKML-Reference: <20100105224656.19431.92588.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/kprobe: Drop function argument access syntax

Drop function argument access syntax, because the function
arguments depend on not only architecture but also
compile-options and function API. And now, we have perf-probe
for finding register/memory assigned to each argument.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Neuling <mikey@neuling.org>
Cc: linuxppc-dev@ozlabs.org
LKML-Reference: <20100105224648.19431.52309.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/kprobe: Update example output in documentation

Update example output in documentation according to current
implementation.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: systemtap <systemtap@sources.redhat.com>
Cc: DLE <dle-develop@lists.sourceforge.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20100105224641.19431.34967.stgit@dhcp-100-2-132.bos.redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Handle relocatable kernels

DSOs don't have this problem because the kernel emits a
PERF_MMAP for each new executable mapping it performs on
monitored threads.

To fix the kernel case we simulate the same behaviour, by having
'perf record' to synthesize a PERF_MMAP for the kernel, encoded
like this:

[root@doppio ~]# perf record -a -f sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.344 MB perf.data (~15038 samples) ]
[root@doppio ~]# perf report -D | head -10

0xd0 [0x40]: event: 1
.
. ... raw event: size 64 bytes
.  0000:  01 00 00 00 00 00 40 00 00 00 00 00 00 00 00 00 ......@........
.  0010:  00 00 00 81 ff ff ff ff 00 00 00 00 00 00 00 00 ...............
.  0020:  00 00 00 00 00 00 00 00 5b 6b 65 72 6e 65 6c 2e ........  [kernel
.  0030:  6b 61 6c 6c 73 79 6d 73 2e 5f 74 65 78 74 5d 00  kallsyms._text]
.  0xd0
[0x40]: PERF_RECORD_MMAP 0/0: [0xffffffff81000000((nil)) @ (nil)]: [kernel.kallsyms._text]

I.e. we identify such event as having:

 .pid      = 0
 .filename = [kernel.kallsyms.REFNAME]
 .start    = REFNAME addr in /proc/kallsyms at 'perf record' time

and use now a hardcoded value of '.text' for REFNAME.

Then, later, in 'perf report', if there are any kernel hits and
thus we need to resolve kernel symbols, we search for REFNAME
and if its address changed, relocation happened and we thus must
change the kernel mapping routines to one that uses .pgoff as
the relocation to apply.

This way we use the same mechanism used for the other DSOs and
don't have to do a two pass in all the kernel symbols.

Reported-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
LKML-Reference: <1262717431-1246-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf report: Fix --no-call-chain option handling

To avoid the funny:

 [root@doppio ~]# perf record -a -f sleep 2s
 [ perf record: Woken up 1 times to write data ]
 [ perf record: Captured and wrote 0.334 MB perf.data (~14572 samples) ]
 [root@doppio ~]# perf report --no-call-graph
 selected -g but no callchain data. Did you call perf record without -g?

And fix the bug reported by peterz when we do indeed record with
callchains and then ask for a report without:

[root@doppio ~]# perf record -a -g -f sleep 2s
[root@doppio ~]# perf report --no-call-graph
Segmentation fault
[root@doppio ~]#

Reported-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262699685-27820-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf session: Keep pointers to the vmlinux maps

So that tools such as 'perf probe' don't have to lookup
'[kernel.kallsyms]' but instead access them directly after
perf_session__create_kernel_maps or
map_groups__create_kernel_maps.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262629169-22797-4-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Create write_padded routine out of __dsos__write_buildid_table

Will be used by other options where padding is needed.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262629169-22797-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Export symbol_type__is_a

Will be needed by the new HEADER_DSO_INFO feature that will be a
HEADER_BUILD_ID superset, replacing it.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262629169-22797-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf symbols: Generalise the kallsyms parsing routine

Will be used to find an specific symbol by name on 'perf record'
to support relocation reference symbols to support relocatable
kernels.

Still have to conver the perf trace tools to use it instead of
their current reimplementation.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262629169-22797-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Fix --pid option for stat

current pid option doesn't work for perf stat. Change it to what
perf record --pid acts as.

Signed-off-by: Liming Wang <liming.wang@windriver.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1262246750-2191-1-git-send-email-liming.wang@windriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Drop useless check for ignored frame

The check that ignores the debug and nmi stack frames is useless
now that we have a frame pointer that makes us start at the
right place. We don't anymore have to deal with these.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262235183-5320-2-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Merge branch 'perf/urgent' into perf/core

Merge reason: queue up dependent patch, update to -rc4

Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Check if /dev/null can be used as the -o gcc argument

At least on Debian PARISC64, using:

acme@parisc:~/git/linux-2.6-tip$ gcc -v
Using built-in specs.
Target: hppa-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian
4.3.4-6' --with-bugurl=file:///usr/share/doc/gcc-4.3/README.Bugs
--enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr
--enable-shared --enable-multiarch --enable-linker-build-id --with-system-zlib --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --enable-nls --with-gxx-include-dir=/usr/include/c++/4.3 --program-suffix=-4.3 --enable-clocale=gnu --enable-libstdcxx-debug --enable-objc-gc --enable-mpfr --disable-libssp --enable-checking=release --build=hppa-linux-gnu --host=hppa-linux-gnu --target=hppa-linux-gnu Thread model: posix gcc version 4.3.4 (Debian 4.3.4-6)

there are issues about using 'gcc -o /dev/null':

/usr/bin/ld: final link failed: File truncated
collect2: ld returned 1 exit status

So we test that and use /dev/null in environments where it
works, while using an .INTERMEDIATE file on those where it can't
be used, so that the .perf.dev.null file can be used instead and
then deleted when make exits.

Researched-with: Kyle McMartin <kyle@mcmartin.ca>
Researched-with: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263293910-8484-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Move QUIET_STDERR def to before first use

QUIET_STDERR is used when detecting if -fstack-protector-all can
be used.

Noticed while building the perf tools on a Debian PARISC64
machine.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1263293910-8484-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: Stop stack frame walking off kernel addresses boundaries

While processing kernel perf callchains, an bad entry can be
considered as a valid stack pointer but not as a kernel address.

In this case, we hang in an endless loop. This can happen in an
x86-32 kernel after processing the last entry in a kernel
stacktrace.

Just stop the stack frame walking after we encounter an invalid
kernel address.

This fixes a hard lockup in x86-32.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1262227945-27014-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Linux 2.6.33-rc4

Merge git://git.infradead.org/battery-2.6

* git://git.infradead.org/battery-2.6:
  pmu_battery: Fix battery full reporting

[SCSI] megaraid_sas: remove sysfs poll_mode_io world writeable permissions

/sys/bus/pci/drivers/megaraid_sas/poll_mode_io defaults to being
world-writable, which seems bad (letting any user affect kernel driver
behavior).

This turns off group and user write permissions, so that on typical
production systems only root can write to it.

Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'for-linus' of git://gitorious.org/linux-omap-dss2/linux

* 'for-linus' of git://gitorious.org/linux-omap-dss2/linux:
  OMAP: DSS2: OMAPFB: fix crash when panel driver was not loaded
  OMAP: DSS2: Reject scaling settings when they cannot be supported
  OMAP: DSS2: Make check-delay-loops consistent
  OMAP: DSS2: OMAPFB: fix omapfb_free_fbmem()
  video/omap: add __init/__exit macros to drivers/video/omap/lcd_htcherald.c
  OMAP: DSS2: Fix compile warning
  MAINTAINERS: Combine DSS2 and OMAPFB2 into one entry
  MAINTAINERS: change omapfb maintainer
  OMAP: OMAPFB: add dummy release function for omapdss
  OMAP: OMAPFB: fix clk_get for RFBI
  OMAP: DSS2: RFBI: convert to new kfifo API
  OMAP: DSS2: Fix crash when panel doesn't define enable_te()
  OMAP: DSS2: Collect interrupt statistics
  OMAP: DSS2: DSI: print debug DCS cmd in hex
  OMAP: DSS2: DSI: fix VC channels in send_short and send_null

lib: Introduce generic list_sort function

There are two copies of list_sort() in the tree already, one in the DRM
code, another in ubifs.  Now XFS needs this as well.  Create a generic
list_sort() function from the ubifs version and convert existing users
to it so we don't end up with yet another copy in the tree.

Signed-off-by: Dave Chinner <david@fromorbit.com>
Acked-by: Dave Airlie <airlied@redhat.com>
Acked-by: Artem Bityutskiy <dedekind@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove my email address from checkpatch.

Maybe this will stop people emailing me about it.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  libata: retry link resume if necessary
  ata_piix: enable 32bit PIO on SATA piix
  sata_promise: don't classify overruns as HSM errors

Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  ARM: Ensure ARMv6/7 mm files are built using appropriate assembler options
  ARM: Fix wrong dmb
  ARM: 5874/1: serial21285: fix disable_irq-from-interrupt-handler deadlock
  ARM: 5873/1: ARM: Fix the reset logic for ARM RealView boards
  ARM: 5872/1: ARM: include needed linux/cpu.h in asm/cpu.h
  ARM: 5871/1: arch/arm: Fix build failure for lpd7a404_defconfig caused by missing includes
  ARM: 5870/1: arch/arm: Fix build failure for defconfigs without CONFIG_ISA_DMA_API set
  ARM: 5868/1: ARM: fix "BUG: using smp_processor_id() in preemptible code"
  ARM: 5867/1: Update U300 defconfig
  ARM: 5866/1: arm ptrace: use unsigned types for kernel pt_regs
  [ARM] pxa: fix strange characters in zaurus gpio .desc
  ARM: add missing recvmmsg syscall number
  [ARM] pxa: fix compiler warnings of unused variable 'id' in cpu_is_pxa9*()
  [ARM] pxa: update pwm_backlight->notify() to include missed 'struct device *'
  [ARM] pxa: enable L2 if present in XSC3
  [ARM] pxa: do not enable L2 after MMU is enabled

Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus: (22 commits)
  MIPS: Ignore vmlinux.*
  MIPS: Move vmlinux.ecoff to arch/mips/boot
  MIPS: cpumask_of_node() should handle -1 as a node
  MIPS: Octeon: Use non-overflowing arithmetic in sched_clock
  MIPS: Malta, PowerTV: Remove unnecessary "Linux started"
  MIPS: BCM63xx: Remove duplicate CONFIG_CMDLINE.
  MIPS: AR7: Remove unused prom_getchar()
  MIPS: PowerTV: Remove extra r4k_clockevent_init() call
  MIPS: Cobalt use strlcat() for the command line arguments
  MIPS: Octeon: Add sched_clock() to csrc-octeon.c
  MIPS: TXx9: Cleanup builtin-cmdline processing
  MIPS: PowerTV: simplify prom_init_cmdline() and merge into prom_init()
  MIPS: PowerTV: Remove unused platform_die()
  MIPS: PowerTV: Remove mips_machine_halt()
  MIPS: PowerTV: Remove unused ptv_memsize
  MIPS: PowerTV: Remove unused prom_getcmdline()
  MIPS: AR7: Remove kgdb_enabled
  MIPS: Alchemy: Correct code taking the size of a pointer
  MIPS: BCM63xx: Fix whitespace damaged board_bcm963xx.c
  MIPS: VR41xx: Use strlcat() for the command line arguments
  ...

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  ALSA: hda - Fix ALC861-VD capture source mixer
  ALSA: ac97: add AC97 STMicroelectronics' codecs
  ALSA: ac97: Add Dell Dimension 2400 to Headphone/Line Jack Sense blacklist
  ASoC: Fix WM8350 DSP mode B configuration
  sbawe: fix memory detection part 2
  sound: oss: off by one bug
  ALSA: usb-audio - Avoid Oops after disconnect
  ALSA: test off by one in setsamplerate()
  ALSA: atiixp: Specify codec for Foxconn RC4107MA-RS2

Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (56 commits)
  sky2: Fix oops in sky2_xmit_frame() after TX timeout
  Documentation/3c509: document ethtool support
  af_packet: Don't use skb after dev_queue_xmit()
  vxge: use pci_dma_mapping_error to test return value
  netfilter: ebtables: enforce CAP_NET_ADMIN
  e1000e: fix and commonize code for setting the receive address registers
  e1000e: e1000e_enable_tx_pkt_filtering() returns wrong value
  e1000e: perform 10/100 adaptive IFS only on parts that support it
  e1000e: don't accumulate PHY statistics on PHY read failure
  e1000e: call pci_save_state() after pci_restore_state()
  netxen: update version to 4.0.72
  netxen: fix set mac addr
  netxen: fix smatch warning
  netxen: fix tx ring memory leak
  tcp: update the netstamp_needed counter when cloning sockets
  TI DaVinci EMAC: Handle emac module clock correctly.
  dmfe/tulip: Let dmfe handle DM910x except for SPARC on-board chips
  ixgbe: Fix compiler warning about variable being used uninitialized
  netfilter: nf_ct_ftp: fix out of bounds read in update_nl_seq()
  mv643xx_eth: don't include cache padding in rx desc buffer size
  ...

Fix trivial conflict in drivers/scsi/cxgb3i/cxgb3i_offload.c

m68knommu: fix definitions of __pa() and __va()

Fix compilation breakage of all m68knommu targets:

  CC      arch/m68knommu/kernel/asm-offsets.s
In file included from include/linux/sched.h:77,
                 from arch/m68knommu/kernel/asm-offsets.c:12:
include/linux/percpu.h: In function 'per_cpu_ptr_to_phys':
include/linux/percpu.h:161: error: implicit declaration of function 'virt_to_phy

This is broken in linux-2.6.33-rc3.

Change the definitions of __pa() and __va() to not use virt_to_phys()
and phys_to_virt(). Trivial 1:1 conversion required for the non-MMU case.

A side effect if this is that the m68knommu can now use asm/virtconvert.h
for the definition of virt_to_phys() and phys_to_virt().

Also cleaned up the definition of page_to_phys() when moving into
virtconvert.h.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

libata: retry link resume if necessary

Interestingly, when SIDPR is used in ata_piix, writes to DET in
SControl sometimes get ignored leading to detection failure.  Update
sata_link_resume() such that it reads back SControl after clearing DET
and retry if it's not clear.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: fengxiangjun <fengxiangjun@neusoft.com>
Reported-by: Jim Faulkner <jfaulkne@ccs.neu.edu>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

ata_piix: enable 32bit PIO on SATA piix

Commit 871af1210f13966ab911ed2166e4ab2ce775b99d enabled 32bit PIO for
PATA piix but didn't for SATA.  There's no reason not to use 32bit PIO
on SATA piix.  Enable it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

sata_promise: don't classify overruns as HSM errors

When sata_promise encounters an overrun or underrun error it
translates that to a libata AC_ERR_HSM, causing a hard reset.
Since over/under-runs were thought to be rare and transient,
this action seemed reasonable.

Unfortunately it turns out that the controller throws overrun
errors when e.g. hal polls a CD or DVD writer containing blank
media, causing long sequences of hard resets and retries before
EH finally gives up.

This patch updates sata_promise to classify over/under-runs as
AC_ERR_OTHER instead. This allows libata EH and upper layers to
retry or fail the operation as they see fit without the disruption
caused by repeated hard resets.

This fixes a problem using a DVD-RAM drive with sata_promise,
reported by Thomas Schorpp. I also tested it on a DVD-RW drive.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
Tested-by: thomas schorpp <thomas.schorpp@googlemail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

ARM: Ensure ARMv6/7 mm files are built using appropriate assembler options

A kernel with both ARMv6 and ARMv7 selected results in build errors.
Fix this by specifying the proper architectures for these assembly
files.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>