Lin Ming [Thu, 18 Mar 2010 10:33:07 +0000 (18:33 +0800)]
perf, x86: Add a key to simplify template lookup in Pentium-4 PMU
Currently, we use opcode(Event and Event-Selector) + emask to
look up template in p4_templates.
But cache events (L1-dcache-load-misses, LLC-load-misses, etc)
use the same event(P4_REPLAY_EVENT) to do the counting, ie, they
have the same opcode and emask. So we can not use current lookup
mechanism to find the template for cache events.
This patch introduces a "key", which is the index into
p4_templates. The low 12 bits of CCCR are reserved, so we can
hide the "key" in the low 12 bits of hwc->config.
We extract the key from hwc->config and then quickly find the
template.
Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1268908387.13901.127.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ingo Molnar [Thu, 18 Mar 2010 15:51:16 +0000 (16:51 +0100)]
perf events: Fix false positive build warning with older GCC's
gcc 4.2.1 produces:
util/probe-event.c: In function 'add_perf_probe_events':
util/probe-event.c:883: warning: 'tev' may be used uninitialized in this function
make: *** [util/probe-event.o] Error 1
Newer GCCs get this right.
To work it around, initialize the variable to NULL so that older GCCs see
it as initialized too.
Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100316220612.32050.33806.stgit@localhost6.localdomain6> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cyrill Gorcunov [Wed, 17 Mar 2010 10:37:01 +0000 (13:37 +0300)]
x86, perf: Use apic_write unconditionally
Since apic_write() maps to a plain noop in the !CONFIG_X86_LOCAL_APIC
case we're safe to remove this conditional compilation and clean up
the code a bit.
Cyrill Gorcunov [Wed, 17 Mar 2010 10:37:00 +0000 (13:37 +0300)]
x86, apic: Allow to use certain functions without APIC built-in support
In case even if the kernel is configured so that
no APIC support is built-in we still may allow
to use certain apic functions as dummy calls.
In particular we start using it in perf-events code.
Note that this is not that same as NOOP apic driver (which
is used if APIC support is present but no physical APIC is
available), this is for the case when we don't have apic code
compiled in at all.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Yinghai Lu <yinghai@kernel.org> Cc: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <20100317104356.011052632@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Zhang, Yanmin [Thu, 18 Mar 2010 14:36:05 +0000 (11:36 -0300)]
perf events: Change perf parameter --pid to process-wide collection instead of thread-wide
Parameter --pid (or -p) of perf currently means a thread-wide
collection. For exmaple, if a process whose id is 8888 has 10
threads, 'perf top -p 8888' just collects the main thread
statistics. That's misleading. Users are used to attach a whole
process when debugging a process by gdb. To follow normal usage
style, the patch change --pid to process-wide collection and add
--tid (-t) to mean a thread-wide collection.
Usage example is:
# perf top -p 8888
# perf record -p 8888 -f sleep 10
# perf stat -p 8888 -f sleep 10
Above commands collect the statistics of all threads of process
8888.
Zhang, Yanmin [Thu, 18 Mar 2010 14:36:04 +0000 (11:36 -0300)]
perf record: Enable counters only when kernel is execing subcommand
'perf record' starts counters before subcommand is execed, so
the statistics is not precise because it includes data of some
preparation steps. I fix it with the patch.
In addition, change the condition to fork/exec subcommand. If
there is a subcommand parameter, perf always fork/exec it. The
usage example is:
# perf record -f -a sleep 10
So this command could collect statistics for 10 seconds
precisely. User still could stop it by CTRL+C. Without the new
capability, user could only input CTRL+C to stop it without
precise time clock.
Zhang, Yanmin [Thu, 18 Mar 2010 14:36:03 +0000 (11:36 -0300)]
perf stat: Enable counters when collecting process-wide or system-wide data
Command 'perf stat' doesn't enable counters when collecting an
existing (by -p) process or system-wide statistics. Fix the
issue.
Change the condition of fork/exec subcommand. If there is a
subcommand parameter, perf always forks/execs it. The usage
example is:
# perf stat -a sleep 10
So this command could collect statistics for 10 seconds
precisely. User still could stop it by CTRL+C. Without the new
capability, user could only use CTRL+C to stop it without
precise time clock.
Another issue is 'perf stat -a' consumes 100% time of a full
single logical cpu. It has a bad impact on running workload.
Fix it by adding a sleep(1) in the while(!done) loop in function
run_perf_stat.
Robert Richter [Wed, 17 Mar 2010 11:49:10 +0000 (12:49 +0100)]
perf/core, x86: Reduce number of CONFIG_X86_LOCAL_APIC macros
The function reserve_pmc_hardware() and release_pmc_hardware()
were hard to read. This patch improves readability of the code by
removing most of the CONFIG_X86_LOCAL_APIC macros.
Signed-off-by: Robert Richter <robert.richter@amd.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268826553-19518-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Masami Hiramatsu [Tue, 16 Mar 2010 22:06:26 +0000 (18:06 -0400)]
perf probe: Add data structure member access support
Support accessing members in the data structures. With this,
perf-probe accepts data-structure members(IOW, it now accepts
dot '.' and arrow '->' operators) as probe arguemnts.
Masami Hiramatsu [Tue, 16 Mar 2010 22:06:12 +0000 (18:06 -0400)]
perf probe: Introduce kprobe_trace_event and perf_probe_event
Introduce kprobe_trace_event and perf_probe_event and replace
old probe_point structure with it. probe_point structure is
not enough flexible nor extensible. New data structures
will help implementing further features.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100316220612.32050.33806.stgit@localhost6.localdomain6> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Masami Hiramatsu [Tue, 16 Mar 2010 22:05:51 +0000 (18:05 -0400)]
perf probe: Rename some die_get_* functions
Rename die_get_real_subprogram and die_get_inlinefunc to
die_find_real_subprogram and die_find_inlinefunc respectively,
because these functions search its children. After that,
'die_get_' means getting a property of that die, and
'die_find_' means searching DIE-tree to get an appropriate
child die.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100316220551.32050.36181.stgit@localhost6.localdomain6> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Masami Hiramatsu [Mon, 15 Mar 2010 17:00:54 +0000 (13:00 -0400)]
kprobes: Hide CONFIG_OPTPROBES and set if arch supports optimized kprobes
Hide CONFIG_OPTPROBES and set if the arch supports optimized
kprobes (IOW, HAVE_OPTPROBES=y), since this option doesn't
change the major behavior of kprobes, and workarounds for minor
changes are documented.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Dieter Ries <mail@dieterries.net> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> Cc: 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: <20100315170054.31593.3153.stgit@localhost6.localdomain6> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf annotate: Properly notify the user that vmlinux is missing
Before this patch we would not find a vmlinux, then try to pass
objdump "[kernel.kallsyms]" as the filename, it would get
confused and produce no output:
[root@doppio ~]# perf annotate n_tty_write
------------------------------------------------
Percent | Source code & Disassembly of [kernel.kallsyms]
------------------------------------------------
Now we check that and emit meaningful warning:
[root@doppio ~]# perf annotate n_tty_write
Can't annotate n_tty_write: No vmlinux file was found in the
path: [0] vmlinux
[1] /boot/vmlinux
[2] /boot/vmlinux-2.6.34-rc1-tip+
[3] /lib/modules/2.6.34-rc1-tip+/build/vmlinux
[4] /usr/lib/debug/lib/modules/2.6.34-rc1-tip+/vmlinux
[root@doppio ~]#
This bug was introduced when we added automatic search for
vmlinux, before that time the user had to specify a vmlinux
file.
v2: Print the warning just for the first symbol found when no
symbol name is specified, otherwise it will spam the screen
repeating the warning for each symbol.
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> Cc: <stable@kernel.org>
LKML-Reference: <1268669073-6856-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf top: Properly notify the user that vmlinux is missing
Before this patch this message would very briefly appear on the
screen and then the screen would get updates only on the top,
for number of interrupts received, etc, but no annotation would
be performed:
[root@doppio linux-2.6-tip]# perf top -s n_tty_write > /tmp/bla
objdump: '[kernel.kallsyms]': No such file
Now this is what the user gets:
[root@doppio linux-2.6-tip]# perf top -s n_tty_write
Can't annotate n_tty_write: No vmlinux file was found in the
path: [0] vmlinux
[1] /boot/vmlinux
[2] /boot/vmlinux-2.6.33-rc5
[3] /lib/modules/2.6.33-rc5/build/vmlinux
[4] /usr/lib/debug/lib/modules/2.6.33-rc5/vmlinux
[root@doppio linux-2.6-tip]#
This bug was introduced when we added automatic search for
vmlinux, before that time the user had to specify a vmlinux
file.
Reported-by: David S. Miller <davem@davemloft.net> 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> Cc: <stable@kernel.org>
LKML-Reference: <1268664418-28328-2-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Mon, 15 Mar 2010 14:46:57 +0000 (11:46 -0300)]
perf record: Enable the enable_on_exec flag if record forks the target
When forking its target, perf record can capture data from
before the target application is started. Perf stat uses the
enable_on_exec flag in the event attributes to keep from
displaying events from before the target program starts, this
patch adds the same functionality to perf record when it is will
fork the target process.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1268664418-28328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cyrill Gorcunov [Mon, 15 Mar 2010 04:58:22 +0000 (12:58 +0800)]
perf, x86: Enable not tagged retired instruction counting on P4s
This should turn on instruction counting on P4s, which was missing in
the first version of the new PMU driver.
It's inaccurate for now, we still need dependant event to tag mops
before we can count them precisely. The result is that the number of
instruction may be lifted up.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1268629102.3355.11.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cyrill Gorcunov [Sat, 13 Mar 2010 08:11:16 +0000 (11:11 +0300)]
x86, perf: Unmask LVTPC only if we have APIC supported
Ingo reported:
|
| There's a build failure on -tip with the P4 driver, on UP 32-bit, if
| PERF_EVENTS is enabled but UP_APIC is disabled:
|
| arch/x86/built-in.o: In function `p4_pmu_handle_irq':
| perf_event.c:(.text+0xa756): undefined reference to `apic'
| perf_event.c:(.text+0xa76e): undefined reference to `apic'
|
So we have to unmask LVTPC only if we're configured to have one.
Reported-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Lin Ming <ming.m.lin@intel.com> CC: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <20100313081116.GA5179@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Masami Hiramatsu [Fri, 12 Mar 2010 23:22:17 +0000 (18:22 -0500)]
perf probe: Fix probe_point buffer overrun
Fix probe_point array-size overrun problem. In some cases (e.g.
inline function), one user-specified probe-point can be
translated to many probe address, and it overruns pre-defined
array-size. This also removes redundant MAX_PROBES macro
definition.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: <stable@kernel.org>
LKML-Reference: <20100312232217.2017.45017.stgit@localhost6.localdomain6>
[ Note that only root can create new probes. Eventually we should remove
the MAX_PROBES limit, but that is a larger patch not eligible to
perf/urgent treatment. ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
Not good, the newt form for report works, but slang has to eat
the cost of the additional callgraph lines everytime it prints a
line, and the callgraph doesn't appear on the screen, so move
the callgraph printing to a separate function and don't use it
in newt.c.
Newt tree widgets are being investigated to properly support
callgraphs, but till that gets merged, lets remove this huge
overhead and show at least the symbol overheads for a callgraph
rich perf.data with good performance.
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: <1268408808-13595-2-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Newt has widespread availability and provides a rather simple
API as can be seen by the size of this patch.
The work needed to support it will benefit other frontends too.
In this initial patch it just checks if the output is a tty, if
not it falls back to the previous behaviour, also if
newt-devel/libnewt-dev is not installed the previous behaviour
is maintaned.
Pressing enter on a symbol will annotate it, ESC in the
annotation window will return to the report symbol list.
More work will be done to remove the special casing in
color_fprintf, stop using fmemopen/FILE in the printing of
hist_entries, etc.
Also the annotation doesn't need to be done via spawning "perf
annotate" and then browsing its output, we can do better by
calling directly the builtin-annotate.c functions, that would
then be moved to tools/perf/util/annotate.c and shared with perf
top, etc
But lets go by baby steps, this patch already improves perf
usability by allowing to quickly do annotations on symbols from
the report screen and provides a first experimentation with
libnewt/TUI integration of tools.
Tested on RHEL5 and Fedora12 X86_64 and on Debian PARISC64 to
browse a perf.data file collected on a Fedora12 x86_64 box.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Avi Kivity <avi@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: <1268349164-5822-5-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf record: Don't try to find buildids in a zero sized file
Fixing this symptom:
[acme@mica linux-2.6-tip]$ perf record -a -f
Fatal: Permission error - are you root?
Bus error
[acme@mica linux-2.6-tip]$
I.e. if for some reason no data is collected, in this case a non
root user trying to do systemwide profiling, no data will be
collected, and then we end up trying to mmap a zero sized file
and access the file header, b00m.
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> Cc: <stable@kernel.org>
LKML-Reference: <1268333592-30872-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cyrill Gorcunov [Thu, 11 Mar 2010 16:54:39 +0000 (19:54 +0300)]
perf, x86: Implement initial P4 PMU driver
The netburst PMU is way different from the "architectural
perfomance monitoring" specification that current CPUs use.
P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle
perfomance monitoring events.
A few implementational details:
1) We need a separate x86_pmu::hw_config helper in struct
x86_pmu since register bit-fields are quite different from P6,
Core and later cpu series.
2) For the same reason is a x86_pmu::schedule_events helper
introduced.
3) hw_perf_event::config consists of packed ESCR+CCCR values.
It's allowed since in reality both registers only use a half
of their size. Of course before making a real write into a
particular MSR we need to unpack the value and extend it to
a proper size.
4) The tuple of packed ESCR+CCCR in hw_perf_event::config
doesn't describe the memory address of ESCR MSR register
so that we need to keep a mapping between these tuples
used and available ESCR (various P4 events may use same
ESCRs but not simultaneously), for this sake every active
event has a per-cpu map of hw_perf_event::idx <--> ESCR
addresses.
5) Since hw_perf_event::idx is an offset to counter/control register
we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel
strips it down to 8 registers and event armed may never be turned
off (ie the bit in active_mask is set but the loop never reaches
this index to check), thanks to Peter Zijlstra
Restrictions:
- No cascaded counters support (do we ever need them?)
- No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS
doesn't work for now)
- There are events with same counters which can't work simultaneously
(need to use intersected ones due to broken counter 1)
- No PERF_COUNT_HW_CACHE_ events yet
Todo:
- Implement dependent events
- Need proper hashing for event opcodes (no linear search, good for
debugging stage but not in real loads)
- Some events counted during a clock cycle -- need to set threshold
for them and count every clock cycle just to get summary statistics
(ie to behave the same way as other PMUs do)
- Need to swicth to use event_constraints
- To support RAW events we need to encode a global list of P4 events
into p4_templates
- Cache events need to be added
Event support status matrix:
Event status
-----------------------------
cycles works
cache-references works
cache-misses works
branch-misses works
bus-cycles partially (does not work on 64bit cpu with HT enabled)
instruction doesnt work (needs dependent event [mop tagging])
branches doesnt work
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch is an optimization in perf_event_task_sched_in() to avoid
scheduling the events twice in a row.
Without it, the perf_disable()/perf_enable() pair is invoked twice,
thereby pinned events counts while scheduling flexible events and we go
throuh hw_perf_enable() twice.
By encapsulating, the whole sequence into perf_disable()/perf_enable() we
ensure, hw_perf_enable() is going to be invoked only once because of the
refcount protection.
This happens because we only iterate the n_running events in the first
pass, and reset their index to -1 if they don't match to force a
re-assignment.
Now, in our RR example, n_running == 0 because we fully unscheduled, so
event-50 will retain its idx==32, even though in scheduling it will have
gotten idx=0, and we don't trigger the re-assign path.
The easiest way to fix this is the below patch, which simply validates
the full assignment in the second pass.
Peter Zijlstra [Thu, 11 Mar 2010 12:06:56 +0000 (13:06 +0100)]
perf, ppc: Fix compile error due to new cpu notifiers
Fix:
arch/powerpc/kernel/perf_event.c:1334: error: 'power_pmu_notifier' undeclared (first use in this function)
arch/powerpc/kernel/perf_event.c:1334: error: (Each undeclared identifier is reported only once
arch/powerpc/kernel/perf_event.c:1334: error: for each function it appears in.)
arch/powerpc/kernel/perf_event.c:1334: error: implicit declaration of function 'power_pmu_notifier'
arch/powerpc/kernel/perf_event.c:1334: error: implicit declaration of function 'register_cpu_notifier'
Due to commit 3f6da390 (perf: Rework and fix the arch CPU-hotplug hooks).
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
kprobes: Calculate the index correctly when freeing the out-of-line execution slot
From : Ananth N Mavinakayanahalli <ananth@in.ibm.com>
When freeing the instruction slot, the arithmetic to calculate
the index of the slot in the page needs to account for the total
size of the instruction on the various architectures.
Calculate the index correctly when freeing the out-of-line
execution slot.
Paul Mackerras [Wed, 10 Mar 2010 09:36:09 +0000 (20:36 +1100)]
perf tools: Fix sparse CPU numbering related bugs
At present, the perf subcommands that do system-wide monitoring
(perf stat, perf record and perf top) don't work properly unless
the online cpus are numbered 0, 1, ..., N-1. These tools ask
for the number of online cpus with sysconf(_SC_NPROCESSORS_ONLN)
and then try to create events for cpus 0, 1, ..., N-1.
This creates problems for systems where the online cpus are
numbered sparsely. For example, a POWER6 system in
single-threaded mode (i.e. only running 1 hardware thread per
core) will have only even-numbered cpus online.
This fixes the problem by reading the /sys/devices/system/cpu/online
file to find out which cpus are online. The code that does that is in
tools/perf/util/cpumap.[ch], and consists of a read_cpu_map()
function that sets up a cpumap[] array and returns the number of
online cpus. If /sys/devices/system/cpu/online can't be read or
can't be parsed successfully, it falls back to using sysconf to
ask how many cpus are online and sets up an identity map in cpumap[].
The perf record, perf stat and perf top code then calls
read_cpu_map() in the system-wide monitoring case (instead of
sysconf) and uses cpumap[] to get the cpu numbers to pass to
perf_event_open.
Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Paul Mackerras [Wed, 10 Mar 2010 09:45:52 +0000 (20:45 +1100)]
perf_event: Fix oops triggered by cpu offline/online
Anton Blanchard found that he could reliably make the kernel hit a
BUG_ON in the slab allocator by taking a cpu offline and then online
while a system-wide perf record session was running.
The reason is that when the cpu comes up, we completely reinitialize
the ctx field of the struct perf_cpu_context for the cpu. If there is
a system-wide perf record session running, then there will be a struct
perf_event that has a reference to the context, so its refcount will
be 2. (The perf_event has been removed from the context's group_entry
and event_entry lists by perf_event_exit_cpu(), but that doesn't
remove the perf_event's reference to the context and doesn't decrement
the context's refcount.)
When the cpu comes up, perf_event_init_cpu() gets called, and it calls
__perf_event_init_context() on the cpu's context. That resets the
refcount to 1. Then when the perf record session finishes and the
perf_event is closed, the refcount gets decremented to 0 and the
context gets kfreed after an RCU grace period. Since the context
wasn't kmalloced -- it's part of a per-cpu variable -- bad things
happen.
In fact we don't need to completely reinitialize the context when the
cpu comes up. It's sufficient to initialize the context once at boot,
but we need to do it for all possible cpus.
This moves the context initialization to happen at boot time. With
this, we don't trash the refcount and the context never gets kfreed,
and we don't hit the BUG_ON.
Reported-by: Anton Blanchard <anton@samba.org> Signed-off-by: Paul Mackerras <paulus@samba.org> Tested-by: Anton Blanchard <anton@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
We are taking a wrong regs snapshot when a trace event triggers.
Either we use get_irq_regs(), which gives us the interrupted
registers if we are in an interrupt, or we use task_pt_regs()
which gives us the state before we entered the kernel, assuming
we are lucky enough to be no kernel thread, in which case
task_pt_regs() returns the initial set of regs when the kernel
thread was started.
What we want is different. We need a hot snapshot of the regs,
so that we can get the instruction pointer to record in the
sample, the frame pointer for the callchain, and some other
things.
Let's use the new perf_fetch_caller_regs() for that.
Comparison with perf record -e lock: -R -a -f -g
Before:
perf: Introduce new perf_fetch_caller_regs() for hot regs snapshot
Events that trigger overflows by interrupting a context can
use get_irq_regs() or task_pt_regs() to retrieve the state
when the event triggered. But this is not the case for some
other class of events like trace events as tracepoints are
executed in the same context than the code that triggered
the event.
It means we need a different api to capture the regs there,
namely we need a hot snapshot to get the most important
informations for perf: the instruction pointer to get the
event origin, the frame pointer for the callchain, the code
segment for user_mode() tests (we always use __KERNEL_CS as
trace events always occur from the kernel) and the eflags
for further purposes.
v2: rename perf_save_regs to perf_fetch_caller_regs as per
Masami's suggestion.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Archs <linux-arch@vger.kernel.org>
perf/x86-64: Use frame pointer to walk on irq and process stacks
We were using the frame pointer based stack walker on every
contexts in x86-32, but not in x86-64 where we only use the
seven-league boots on the exception stacks.
Use it also on irq and process stacks. This utterly accelerate
the captures.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
lockdep: Move lock events under lockdep recursion protection
There are rcu locked read side areas in the path where we submit
a trace event. And these rcu_read_(un)lock() trigger lock events,
which create recursive events.
The problem is not that much the trace recursion itself, as we have a
recursion protection already (though it's always wasteful to recurse).
But the trace events are outside the lockdep recursion protection, then
each lockdep event triggers a lock trace, which will trigger two
other lockdep events. Here the recursive lock trace event won't
be taken because of the trace recursion, so the recursion stops there
but lockdep will still analyse these new events:
perf report: Print the map table just after samples for which no map was found
If -vv is used just the map table will be printed, -vvv will
print the symbol table too, with it we can see that we have a
bug where some samples are not being resolved to a map when we
get them in the perf.data stream, but after we have it all
processed, we can find the right map, some reordering probably
is happening.
Upcoming patches will provide ways to ask for most PERF_SAMPLE_
conditional samples to be taken for !PERF_RECORD_SAMPLE events
too, then we'll be able to ask for PERF_SAMPLE_TIME and
PERF_SAMPLE_CPU to help diagnose this.
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: <1268161097-17761-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Fri, 5 Mar 2010 15:51:09 +0000 (12:51 -0300)]
perf report: Add multiple event support
Perf report does not handle multiple events being reported, even
though perf record stores them properly on disk. This patch
addresses that issue by adding the logic to perf report to use
the event stream id that is saved by record and the new data
structures to seperate the event streams and report them
individually.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267804269-22660-6-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Fri, 5 Mar 2010 15:51:08 +0000 (12:51 -0300)]
perf session: Change perf_session post processing functions to take histogram tree
Now that report can store historgrams for multiple events we
need to be able to do the post processing work for each
histogram. This patch changes the post processing functions so
that they can be called individually for each event's histogram.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com>
[ Guarantee bisectabilty by fixing up builtin-report.c ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267804269-22660-5-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Fri, 5 Mar 2010 15:51:07 +0000 (12:51 -0300)]
perf session: Add storage for seperating event types in report
This patch adds the structures necessary to count each event
type independently in perf report.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267804269-22660-4-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Fri, 5 Mar 2010 15:51:06 +0000 (12:51 -0300)]
perf session: Change add_hist_entry to take the tree root instead of session
In order to minimize the impact of storing multiple events in a
report this function will now take the root of the histogram
tree so that the logic for selecting the proper tree can be
inserted before the call.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267804269-22660-3-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Eric B Munson [Fri, 5 Mar 2010 15:51:05 +0000 (12:51 -0300)]
perf record: Add ID and to recorded event data when recording multiple events
Currently perf record does not write the ID or the to disk for
events. This doesn't allow report to tell if an event stream
contains one or more types of events. This patch adds this
entry to the list of data that record will write to disk if more
than one event was requested.
Signed-off-by: Eric B Munson <ebmunson@us.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267804269-22660-2-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
cc1: warnings being treated as errors
util/probe-finder.c: In function 'find_line_range':
util/probe-finder.c:172: warning: 'src' may be used
uninitialized in this function make: *** [util/probe-finder.o]
Error 1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: 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: <1267804269-22660-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf tools: Don't trow away old map slices not overlapped by new maps
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: David S. Miller <davem@davemloft.net> 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: <1267800842-22324-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Mon, 8 Mar 2010 12:51:12 +0000 (13:51 +0100)]
perf, x86: Don't reset the LBR as frequently
If we reset the LBR on each first counter, simple counter rotation which
first deschedules all counters and then reschedules the new ones will
lead to LBR reset, even though we're still in the same task context.
Reduce this by not flushing on the first counter but only flushing on
different task contexts.
Peter Zijlstra [Sat, 6 Mar 2010 17:57:38 +0000 (18:57 +0100)]
perf, x86: Deal with multiple state bits for pebs-fmt1
Its unclear if the PEBS state record will have only a single bit set, in
case it does not and accumulates bits, deal with that by only processing
each event once.
Peter Zijlstra [Sat, 6 Mar 2010 12:26:11 +0000 (13:26 +0100)]
perf, x86: Fix pebs drains
I overlooked the perf_disable()/perf_enable() calls in
intel_pmu_handle_irq(), (pointed out by Markus) so we should not
explicitly disable_all/enable_all pebs counters in the drain functions,
these are already disabled and enabling them early is confusing.
Peter Zijlstra [Mon, 8 Mar 2010 16:51:33 +0000 (17:51 +0100)]
perf, x86: Avoid double disable on throttle vs ioctl(PERF_IOC_DISABLE)
Calling ioctl(PERF_EVENT_IOC_DISABLE) on a thottled counter would result
in a double disable, cure this by using x86_pmu_{start,stop} for
throttle/unthrottle and teach x86_pmu_stop() to check ->active_mask.
Peter Zijlstra [Wed, 3 Mar 2010 16:07:40 +0000 (17:07 +0100)]
perf, x86: Clean up IA32_PERF_CAPABILITIES usage
Saner PERF_CAPABILITIES support, which also exposes pebs_trap. Use that
latter to make PEBS's use of LBR conditional since a fault-like pebs
should already report the correct IP.
( As of this writing there is no known hardware that implements
!pebs_trap )
Peter Zijlstra [Wed, 3 Mar 2010 12:12:23 +0000 (13:12 +0100)]
perf, x86: use LBR for PEBS IP+1 fixup
Use the LBR to fix up the PEBS IP+1 issue.
As said, PEBS reports the next instruction, here we use the LBR to find
the last branch and from that construct the actual IP. If the IP matches
the LBR-TO, we use LBR-FROM, otherwise we use the LBR-TO address as the
beginning of the last basic block and decode forward.
Once we find a match to the current IP, we use the previous location.
This patch introduces a new ABI element: PERF_RECORD_MISC_EXACT, which
conveys that the reported IP (PERF_SAMPLE_IP) is the exact instruction
that caused the event (barring CPU errata).
The fixup can fail due to various reasons:
1) LBR contains invalid data (quite possible)
2) part of the basic block got paged out
3) the reported IP isn't part of the basic block (see 1)
Peter Zijlstra [Wed, 3 Mar 2010 11:02:30 +0000 (12:02 +0100)]
perf, x86: Implement simple LBR support
Implement simple suport Intel Last-Branch-Record, it supports all
hardware that implements FREEZE_LBRS_ON_PMI, but does not (yet) implement
the LBR config register.
The Intel LBR is a FIFO of From,To addresses describing the last few
branches the hardware took.
This patch does not add perf interface to the LBR, but merely provides an
interface for internal use.
projects / karo-tx-linux.git / log