We cannot assume that because hwc->idx == assign[i], we can avoid
reprogramming the counter in hw_perf_enable().
The event may have been scheduled out and another event may have been
programmed into this counter. Thus, we need a more robust way of
verifying if the counter still contains config/data related to an event.
This patch adds a generation number to each counter on each cpu. Using
this mechanism we can verify reliabilty whether the content of a counter
corresponds to an event.
Signed-off-by: Stephane Eranian <eranian@google.com>
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: <4b66dc67.0b38560a.1635.ffffae18@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Pretty much all of the calls do perf_disable/perf_enable cycles, pull
that out to cut back on hardware programming.
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: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
By relying on logic in dso__load_kernel_sym(), we can
automatically load vmlinux.
The only thing which needs to be adjusted, is how --sym-annotate
option is handled - now we can't rely on vmlinux been loaded
until full successful pass of dso__load_vmlinux(), but that's
not the case if we'll do sym_filter_entry setup in
symbol_filter().
So move this step right after event__process_sample() where we
know the whole dso__load_kernel_sym() pass is done.
By the way, though conceptually similar `perf top` still can't
annotate userspace - see next patches with fixes.
Signed-off-by: Kirill Smelkov <kirr@landau.phys.spbu.ru>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1265223128-11786-9-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The problem was we were incorrectly calculating objdump
addresses for sym->start and sym->end, look:
For simple ET_DYN type DSO (*.so) with one function, objdump -dS
output is something like this:
000004ac <my_strlen>:
int my_strlen(const char *s)
4ac: 55 push %ebp
4ad: 89 e5 mov %esp,%ebp
4af: 83 ec 10 sub $0x10,%esp
{
i.e. we have relative-to-dso-mapping IPs (=RIP) there.
For ET_EXEC type and probably for prelinked libs as well (sorry
can't test - I don't use prelink) objdump outputs absolute IPs,
e.g.
08048604 <zz_strlen>:
extern "C"
int zz_strlen(const char *s)
8048604: 55 push %ebp
8048605: 89 e5 mov %esp,%ebp
8048607: 83 ec 10 sub $0x10,%esp
{
So, if sym->start is always relative to dso mapping(*), we'll
have to unmap it for ET_EXEC like cases, and leave as is for
ET_DYN cases.
(*) and it is - we've explicitely made it relative. Look for
adjust_symbols handling in dso__load_sym()
Previously we were always unmapping sym->start and for ET_DYN
dsos resulting addresses were wrong, and so objdump output was
empty.
The end result was that perf annotate output for symbols from
non-prelinked *.so had always 0.00% percents only, which is
wrong.
To fix it, let's introduce a helper for converting rip to
objdump address, and also let's document what map_ip() and
unmap_ip() do -- I had to study sources for several hours to
understand it.
Signed-off-by: Kirill Smelkov <kirr@landau.phys.spbu.ru>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1265223128-11786-8-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We want to stream events as fast as possible to perf.data, and
also in the future we want to have splice working, when no
interception will be possible.
Using build_id__mark_dso_hit_ops to create the list of DSOs that
back MMAPs we also optimize disk usage in the build-id cache by
only caching DSOs that had hits.
Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1265223128-11786-6-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We can check using strcmp, most DSOs don't start with '[' so the
test is cheap enough and we had to test it there anyway since
when reading perf.data files we weren't calling the routine that
created this global variable and thus weren't setting it as
"loaded", which was causing a bogus:
Failed to open [vdso], continuing without symbols
Message as the first line of '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: <1265223128-11786-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
While debugging a problem reported by Pekka Enberg by printing
the IP and all the maps for a thread when we don't find a map
for an IP I noticed that dso__load_sym needs to fixup these
extra maps it creates to hold symbols in different ELF sections
than the main kernel one.
Now we're back showing things like:
[root@doppio linux-2.6-tip]# perf report | grep vsyscall
0.02% mutt [kernel.kallsyms].vsyscall_fn [.] vread_hpet
0.01% named [kernel.kallsyms].vsyscall_fn [.] vread_hpet
0.01% NetworkManager [kernel.kallsyms].vsyscall_fn [.] vread_hpet
0.01% gconfd-2 [kernel.kallsyms].vsyscall_0 [.] vgettimeofday
0.01% hald-addon-rfki [kernel.kallsyms].vsyscall_fn [.] vread_hpet
0.00% dbus-daemon [kernel.kallsyms].vsyscall_fn [.] vread_hpet
[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: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1265223128-11786-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I noticed while writing the first test in 'perf regtest' that to
just test the symbol handling routines one needs to create a
perf session, that is a layer centered on a perf.data file,
events, etc, so I untied these layers.
This reduces the complexity for the users as the number of
parameters to most of the symbols and session APIs now was
reduced while not adding more state to all the map instances by
only having data that is needed to split the kernel (kallsyms
and ELF symtab sections) maps and do vmlinux relocation on the
main kernel map.
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: <1265223128-11786-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Open perf data file with O_LARGEFILE flag since its size is
easily larger that 2G.
For example:
# rm -rf perf.data
# ./perf kmem record sleep 300
[ perf record: Woken up 0 times to write data ]
[ perf record: Captured and wrote 3142.147 MB perf.data
(~137282513 samples) ]
# ll -h perf.data
-rw------- 1 root root 3.1G .....
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4B68F32A.9040203@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This reverts commit f5a2c3dce0.
This patch is required for making "perf lock rec" work.
The commit f5a2c3dce0 changes write_event() of builtin-record.c
. And changed write_event() sometimes doesn't stop with perf
lock rec.
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: <new-submission>
[ that commit also causes perf record to not be Ctrl-C-able,
and it's concetually wrong to parse the data at record time
(unconditionally - even when not needed), as we eventually
want to be able to do zero-copy recording, at least for
non-archive recordings. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Checked with:
./../scripts/checkpatch.pl --terse --file perf.c
perf.c: 51: ERROR: open brace '{' following function declarations go on the next line
perf.c: 73: ERROR: "foo*** bar" should be "foo ***bar"
perf.c:112: ERROR: space prohibited before that close parenthesis ')'
perf.c:127: ERROR: space prohibited before that close parenthesis ')'
perf.c:171: ERROR: "foo** bar" should be "foo **bar"
perf.c:213: ERROR: "(foo*)" should be "(foo *)"
perf.c:216: ERROR: "(foo*)" should be "(foo *)"
perf.c:217: ERROR: space required before that '*' (ctx:OxV)
perf.c:452: ERROR: do not initialise statics to 0 or NULL
perf.c:453: ERROR: do not initialise statics to 0 or NULL
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
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>
LKML-Reference: <1264633557-17597-7-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
One problem with frequency driven counters is that we cannot
predict the rate at which they trigger, therefore we have to
start them at period=1, this causes a ramp up effect. However,
if we fail to propagate the stable state on fork each new child
will have to ramp up again. This can lead to significant
artifacts in sample data.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: eranian@google.com
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: <1264752266.4283.2121.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
At enable time the counter might still have a ->idx pointing to
a previously occupied location that might now be taken by
another event. Resetting the counter at that location with data
from this event will destroy the other counter's count.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100127221122.261477183@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The new Intel documentation includes Westmere arch specific
event maps that are significantly different from the Nehalem
ones. Add support for this generation.
Found the CPUID model numbers on wikipedia.
Also ammend some Nehalem constraints, spotted those when looking
for the differences between Nehalem and Westmere.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <20100127221122.151865645@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Share the meat of the x86_pmu_disable() code with hw_perf_enable().
Also remove the barrier() from that code, since I could not convince
myself we actually need 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>
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>
