install_breakpoint() is called under mm->mmap_sem, this protects
set_swbp() but not prepare_uprobe(). Two or more different tasks
can call install_breakpoint()->prepare_uprobe() at the same time,
this leads to numerous problems if UPROBE_COPY_INSN is not set.
Just for example, the second copy_insn() can corrupt the already
analyzed/fixuped uprobe->arch.insn and race with handle_swbp().
This patch simply adds uprobe->copy_mutex to serialize this code.
We could probably reuse ->consumer_rwsem, but this would mean that
consumer->handler() can not use mm->mmap_sem, not good.
Note: this is another temporary ugly hack until we move this logic
into uprobe_register().
uprobes: Fix handle_swbp() vs unregister() + register() race
Strictly speaking this race was added by me in 56bb4cf6. However
I think that this bug is just another indication that we should
move copy_insn/uprobe_analyze_insn code from install_breakpoint()
to uprobe_register(), there are a lot of other reasons for that.
Until then, add a hack to close the race.
A task can hit uprobe U1, but before it calls find_uprobe() this
uprobe can be unregistered *AND* another uprobe U2 can be added to
uprobes_tree at the same inode/offset. In this case handle_swbp()
will use the not-fully-initialized U2, in particular its arch.insn
for xol.
Add the additional !UPROBE_COPY_INSN check into handle_swbp(),
if this flag is not set we simply restart as if the new uprobe was
not inserted yet. This is not very nice, we need barriers, but we
will remove this hack when we change uprobe_register().
Note: with or without this patch install_breakpoint() can race with
itself, yet another reson to kill UPROBE_COPY_INSN altogether. And
even the usage of uprobe->flags is not safe. See the next patches.
uprobes: Do not delete uprobe if uprobe_unregister() fails
delete_uprobe() must not be called if register_for_each_vma(false)
fails to remove all breakpoints, __uprobe_unregister() is correct.
The problem is that register_for_each_vma(false) always returns 0
and thus this logic does not work.
1. Change verify_opcode() to return 0 rather than -EINVAL when
unregister detects the !is_swbp insn, we can treat this case
as success and currently unregister paths ignore the error
code anyway.
2. Change remove_breakpoint() to propagate the error code from
write_opcode().
3. Change register_for_each_vma(is_register => false) to remove
as much breakpoints as possible but return non-zero if
remove_breakpoint() fails at least once.
uprobes/x86: Only rep+nop can be emulated correctly
__skip_sstep() correctly detects the "nontrivial" nop insns,
but since it doesn't update regs->ip we can not really skip
"0x0f 0x1f | 0x0f 0x19 | 0x87 0xc0", the probed application
is killed by SIGILL'ed handle_swbp().
Remove these additional checks. If we want to implement this
correctly we need to know the full insn length to update ->ip.
set_swbp()->is_swbp_at_addr() is not needed for correctness, it is
harmless to do the unnecessary __replace_page(old_page, new_page)
when these 2 pages are identical.
And it can not be counted as optimization. mmap/register races are
very unlikely, while in the likely case is_swbp_at_addr() adds the
extra get_user_pages() even if the caller is uprobe_mmap(current->mm)
and returns false.
Note also that the semantics/usage of is_swbp_at_addr() in uprobe.c
is confusing. set_swbp() uses it to detect the case when this insn
was already modified by uprobes, that is why it should always compare
the opcode with UPROBE_SWBP_INSN even if the hardware (like powerpc)
has other trap insns. It doesn't matter if this breakpoint was in fact
installed by gdb or application itself, we are going to "steal" this
breakpoint anyway and execute the original insn from vm_file even if
it no longer matches the memory.
OTOH, handle_swbp()->find_active_uprobe() uses is_swbp_at_addr() to
figure out whether we need to send SIGTRAP or not if we can not find
uprobe, so in this case it should return true for all trap variants,
not only for UPROBE_SWBP_INSN.
This patch removes set_swbp()->is_swbp_at_addr(), the next patches
will remove it from set_orig_insn() which is similar to set_swbp()
in this respect. So the only caller will be handle_swbp() and we
can make its semantics clear.
uprobes: Restrict valid_vma(false) to skip VM_SHARED vmas
valid_vma(false) ignores ->vm_flags, this is not actually right.
We should never try to write into MAP_SHARED mapping, this can
confuse an apllication which actually writes to ->vm_file.
With this patch valid_vma(false) ignores VM_WRITE only but checks
other (immutable) bits checked by valid_vma(true). This can also
speedup uprobe_munmap() and uprobe_unregister().
Note: even after this patch _unregister can confuse the probed
application if it does mprotect(PROT_WRITE) after _register and
installs "int3", but this is hardly possible to avoid and this
doesn't differ from gdb case.
uprobes: Change valid_vma() to demand VM_MAYEXEC rather than VM_EXEC
uprobe_register() or uprobe_mmap() requires VM_READ | VM_EXEC, this
is not right. An apllication can do mprotect(PROT_EXEC) later and
execute this code.
Change valid_vma(is_register => true) to check VM_MAYEXEC instead.
No need to check VM_MAYREAD, it is always set.
write_opcode()->get_user_pages() needs FOLL_FORCE to ensure we can
read the page even if the probed task did mprotect(PROT_NONE) after
uprobe_register(). Without FOLL_WRITE, FOLL_FORCE doesn't have any
side effect but allows to read the !VM_READ memory.
Otherwiese the subsequent uprobe_unregister()->set_orig_insn() fails
and we leak "int3". If that task does mprotect(PROT_READ | EXEC) and
execute the probed insn later it will be killed.
Note: in fact this is also needed for _register, see the next patch.
Kill UTASK_BP_HIT state, it buys nothing but complicates the code.
It is only used in uprobe_notify_resume() to decide who should be
called, we can check utask->active_uprobe != NULL instead. And this
allows us to simplify handle_swbp(), no need to clear utask->state.
Likewise we could kill UTASK_SSTEP, but UTASK_BP_HIT is worse and
imho should die. The problem is, it creates the special case when
task->utask is NULL, we can't distinguish RUNNING and BP_HIT. With
this patch utask == NULL always means RUNNING.
uprobes: Fix UPROBE_SKIP_SSTEP checks in handle_swbp()
If handle_swbp()->add_utask() fails but UPROBE_SKIP_SSTEP is set,
cleanup_ret: path do not restart the insn, this is wrong. Remove
this check and add the additional label for can_skip_sstep() = T
case.
Note also that UPROBE_SKIP_SSTEP can be false positive, we simply
can not trust it unless arch_uprobe_skip_sstep() was already called.
Also, move another UPROBE_SKIP_SSTEP check before can_skip_sstep()
into this helper, this looks more clean and understandable.
Note: probably we should rename "skip" to "emulate" and I think
that "clear UPROBE_SKIP_SSTEP" should be moved to arch_can_skip.
uprobes: Do not setup ->active_uprobe/state prematurely
handle_swbp() sets utask->active_uprobe before handler_chain(),
and UTASK_SSTEP before pre_ssout(). This complicates the code
for no reason, arch_ hooks or consumer->handler() should not
(and can't) use this info.
Change handle_swbp() to initialize them after pre_ssout(), and
remove the no longer needed cleanup-utask code.
What we get from audit_syscall_to_name isn't what we find in the
syscalls: tracepoint events, so add the alias that allows the tool to
find prctl, fstat, fstatat and stat.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-3m9su7jhwnxvepnr3ne1du5k@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Thu, 27 Sep 2012 11:23:38 +0000 (20:23 +0900)]
perf tools: Check libaudit availability for perf-trace builtin
The newly added trace command requires an external audit library.
However it can cause a build error because it's not checked whether the
libaudit is installed on system:
CC builtin-trace.o
builtin-trace.c:7:22: fatal error: libaudit.h: No such file or directory
compilation terminated.
make: *** [builtin-trace.o] Error 1
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1348745018-21744-1-git-send-email-namhyung@kernel.org
[ committer note: Added ", disables 'trace tool' to the feature warning msg ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
-p, --pid <pid> trace events on existing process id
--tid <tid> trace events on existing thread id
--all-cpus system-wide collection from all CPUs
--cpu <cpu> list of cpus to monitor
--no-inherit child tasks do not inherit counters
--mmap-pages <n> number of mmap data pages
--uid <user> user to profile
[acme@sandy linux]$
Those should have the same semantics as when using with 'perf record'.
It gets stuck sometimes, but hey, it works sometimes too!
In time it should support perf.data based workloads, i.e. it should have
a:
-o filename
Command line option that will produce a perf.data file that can then be
used with 'perf trace' or any of the other perf tools (script, report,
etc).
It will also eventually have the set of functionalities described in the
previous 'trace' prototype by Thomas Gleixner:
"Announcing a new utility: 'trace'"
http://lwn.net/Articles/415728/
Also planned is to have some of the features suggested in the comments
of that LWN article.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf evlist: Renane set_filters method to apply_filters
Because that is what it really does, i.e. it applies the filters that
were parsed from the command line and stashed into the evsels they refer
to.
We'll need the set_filter method name to actually apply a filter to all
the evsels in an evlist, for instance, to ask that a syswide tracer
doesn't trace itself.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf test: Add test to check we correctly parse and match syscall open parms
It will set up a syscall open tracepoint event, generate an open with
invalid flags, then check those flags were the ones reported in the
tracepoint fired.
For the filename we need vfs:getname, but that will go thru some more
iterations as the vfs getname codebase is going thru changes lately.
When that is in I'll just check that the perf_evsel__newtp constructor
is not bailing out and then add it to the evlist, catch the event and
check the filename against the one used in the 'open' call used to
trigger the event.
Cc: David Ahern <dsahern@gmail.com> Cc: Eric Paris <eparis@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jeff Layton <jlayton@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-p5w9aq0jcbb91ghzqomowm16@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We were relying on the info in pevent, but since we have it in
perf_evsel, set up by the perf_session routine if read from a perf.data
file or by whoever creates the evsels, use it.
New 'perf test' entries will use it to parse locally generated events,
in a non perf.data centered workflow.
As well as use byteswap.h to get per arch optimized swap routines, like
other parts of perf (header, perf_evsel__parse_sample, symbol, etc)
already do.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-8tjuxk09mlsfmh7macgkxsip@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Instead of passing it around for parsing as an explicit parameter, will
help with reading tracepoint fields when not using a perf session or
pevent structure, i.e. for non perf.data centered workflows.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-qa67ikv2sm49cwa7dyjhhp6g@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core
Pull perf/core fixes from Arnaldo Carvalho de Melo:
* The new perf_evsel__tp_sched_test 'perf test' broke the build by setting the
'ret' variable but not using it, caught by newer gcc
-Werror=unused-but-set-variable, fix from Namhyung Kim.
* pevent_parse_event should return a proper PEVENT_ERRNO__ and call
pevent_free_format on its failure path, fixes from Namhyung Kim.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Ingo Molnar <mingo@kernel.org>
Namhyung Kim [Tue, 25 Sep 2012 02:20:28 +0000 (11:20 +0900)]
perf test: Fix build failure
The commit 6a6cd11d4e57 ("perf test: Add test for the sched tracepoint
format fields") added following build error:
CC builtin-test.o
builtin-test.c: In function ‘perf_evsel__test_field’:
builtin-test.c:1216:6: error: variable ‘ret’ set but not used [-Werror=unused-but-set-variable]
builtin-test.c: In function ‘perf_evsel__tp_sched_test’:
builtin-test.c:1242:6: error: variable ‘ret’ set but not used [-Werror=unused-but-set-variable]
cc1: all warnings being treated as errors
make: *** [builtin-test.o] Error 1
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1348539628-3821-1-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
* Convert the trace builtins to use the growing evsel/evlist
tracepoint infrastructure, removing several open coded constructs
like switch like series of strcmp to dispatch events, etc.
Basically what had already been showcased in 'perf sched'.
* Add evsel constructor for tracepoints, that uses libtraceevent
just to parse the /format events file, use it in a new 'perf test'
to make sure the libtraceevent format parsing regressions can
be more readily caught.
* Some strange errors were happening in some builds, but not on the
next, reported by several people, problem was some parser related
files, generated during the build, didn't had proper make deps,
fix from Eric Sandeen.
* Fix some compiling errors on 32-bit, from Feng Tang.
* Don't use sscanf extension %as, not available on bionic, reimplementation
by Irina Tirdea.
* Fix bfd.h/libbfd detection with recent binutils, from Markus Trippelsdorf.
* Introduce struct and cache information about the environment where a
perf.data file was captured, from Namhyung Kim.
* Fix several error paths in libtraceevent, from Namhyung Kim.
Print event causing perf_event_open() to fail in 'perf record',
from Stephane Eranian.
* New 'kvm' analysis tool, from Xiao Guangrong.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In our application, we have trace markers spread through user-space.
We have markers in GL, X, etc. These are super handy for Chrome's
about:tracing feature (Chrome + system + kernel trace view), but
can be very distracting when you're trying to debug a kernel issue.
I normally, use "grep -v tracing_mark_write" but it would be nice
if I could just temporarily disable markers all together.
perf evsel: Provide a new constructor for tracepoints
The existing constructor receives a perf_event_attr filled with the
event type and the config.
To reduce the boilerplate for tracepoints, provide a new constructor,
perf_evsel__newtp() that receives the tracepoint name and will open
the debugfs file, call into libtraceevent new pevent_parse_format file
to fill its ->tp_format member, so that users can then just call
perf_evsel__field() to access its fields.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/n/tip-6du8dl1hz0y5l4cybodye7hn@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools lib traceevent: Carve out events format parsing routine
The pevent_parse_event() routine will parse a events/sys/tp/format file
and add an event_format instance to the pevent struct.
This patch introduces a pevent_parse_format() routine with just the bits
needed to parse the event/sys/tp/format file and just return the
event_format instance, useful for when all we want is to parse the
format file, without requiring the pevent struct.
Acked-by: Steven Rostedt <rostedt@goodmis.org> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/n/tip-lge0afl47arh86om0m6a5bqr@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Wed, 19 Sep 2012 06:58:42 +0000 (15:58 +0900)]
tools lib traceevent: Make sure that arg->op.right is set properly
When process_op failed, @arg will be freed on a caller with type of
PRINT_OP. Thus free_arg() will try to free ->op.right field which can
have stale value if something bad happens in the middle.
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1348037924-17568-3-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf uses sscanf extension %as to read and allocate a string in the same
step. This is a non-standard extension only present in new versions of
glibc.
Replacing the use of sscanf and %as with strtok_r calls in order to
parse a given string into its components. This is needed in Android
since bionic does not support
%as extension for sscanf.
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Tested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Irina Tirdea <irina.tirdea@intel.com> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1348173470-4936-1-git-send-email-irina.tirdea@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Mon, 24 Sep 2012 08:15:03 +0000 (17:15 +0900)]
perf header: Remove perf_header__read_feature
Because its only user builtin-kvm::get_cpu_isa() has gone, It can be
removed safely. In general, we have the feature information in
perf_session_env already, no need to read it again.
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Dong Hao <haodong@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Link: http://lkml.kernel.org/r/1348474503-15070-7-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Mon, 24 Sep 2012 08:15:00 +0000 (17:15 +0900)]
perf header: Use pre-processed session env when printing
From now on each feature information is processed and saved in perf
header so that it can be used for printing. The event desc and branch
stack features are not touched since they're not saved.
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1348474503-15070-4-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Mon, 24 Sep 2012 08:14:59 +0000 (17:14 +0900)]
perf header: Add ->process callbacks to most of features
From now on each feature information is processed and saved in perf
header so that it can be used wherever needed. The BRANCH_STACK feature
is an exception since it needs nothing to be done.
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1348474503-15070-3-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Mon, 24 Sep 2012 08:14:58 +0000 (17:14 +0900)]
perf header: Add struct perf_session_env
The struct perf_session_env will preserve environment information at the
time of perf record. It can be accessed anytime after parsing a
perf.data file if needed.
Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1348474503-15070-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Use zalloc for the malloc+memset open coded sequence.
Fix leak on the #ifdef'ed C state handling and when detecting invalid
data in p_state_change().
Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf lock: Use perf_evsel__intval and perf_session__set_tracepoints_handlers
Following the model of 'perf sched':
. raw_field_value searches first on the common fields, that are unused
in this tool
. Leave using perf_evsel__intval to the actual handlers, some may not
need to incur some of the cost because they may not need all the
fields values.
. Using perf_session__set_tracepoints_handlers will save all those
strcmp to find the right handler at sample processing time, do it just
once and get the handler from evsel->handler.func.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf kmem: Use perf_evsel__intval and perf_session__set_tracepoints_handlers
Following the model of 'perf sched':
. raw_field_value searches first on the common fields, that are unused
in this tool
. Using perf_session__set_tracepoints_handlers will save all those
strcmp to find the right handler at sample processing time, do it just
once and get the handler from evsel->handler.func.
Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Using plain raw_field_value(evsel->tp_format) will look at the common
fields as well, and since this tool doesn't need those, speed it up a
bit by looking at just the event specific fields.
Also in general use just evsel and sample, just like was done in 'perf
sched'.
v2: Fixed up test against evsel->name, that contains the subsys name
too, by David Ahern.
Cc: David Ahern <dsahern@gmail.com> Cc: Dong Hao <haodong@linux.vnet.ibm.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add 'perf kvm stat' support to analyze kvm vmexit/mmio/ioport smartly
Usage:
- kvm stat
run a command and gather performance counter statistics, it is the alias of
perf stat
- trace kvm events:
perf kvm stat record, or, if other tracepoints are interesting as well, we
can append the events like this:
perf kvm stat record -e timer:* -a
If many guests are running, we can track the specified guest by using -p or
--pid, -a is used to track events generated by all guests.
- show the result:
perf kvm stat report
The output example is following:
13005
13059
total 2 guests are running on the host
Then, track the guest whose pid is 13059:
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.253 MB perf.data.guest (~11065 samples) ]
Total Samples:195, Total events handled time:10959950.90us.
Signed-off-by: Dong Hao <haodong@linux.vnet.ibm.com> Signed-off-by: Runzhen Wang <runzhen@linux.vnet.ibm.com>
[ Dong Hao <haodong@linux.vnet.ibm.com>
Runzhen Wang <runzhen@linux.vnet.ibm.com>:
- rebase it on current acme's tree
- fix the compiling-error on i386 ] Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Acked-by: David Ahern <dsahern@gmail.com> Cc: Avi Kivity <avi@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: kvm@vger.kernel.org Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com> Link: http://lkml.kernel.org/r/1347870675-31495-4-git-send-email-haodong@linux.vnet.ibm.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adding the following dependencies fixes it for me.
Signed-off-by: Eric Sandeen <sandeen@redhat.com> Reviewed-by: Namhyung Kim <namhyung@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/505BD190.40707@redhat.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf record: Print event causing perf_event_open() to fail
Got tired of not getting the event that caused the perf_event_open()
syscall to fail. So I fixed the error message. This is very useful when
monitoring lots of events in a single run.
Signed-off-by: Stephane Eranian <eranian@google.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/20120920161945.GA7064@quad Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Feng Tang [Mon, 27 Aug 2012 07:38:26 +0000 (15:38 +0800)]
perf tools: Fix a compiling error in util/map.c
This patch fix a compile warning taken as error:
CC util/map.o
cc1: warnings being treated as errors
util/map.c: In function ‘map__fprintf_dsoname’:
util/map.c:240: error: ‘dsoname’ may be used uninitialized in this function
make: *** [util/map.o] Error 1
Signed-off-by: Feng Tang <feng.tang@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1346053107-11946-3-git-send-email-feng.tang@intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf tools: Fix a compiling error in trace-event-perl.c for 32 bits machine
On my x86_32 mahcine, there is a compile error:
CC util/scripting-engines/trace-event-perl.o
cc1: warnings being treated as errors
util/scripting-engines/trace-event-perl.c: In function
perl_process_tracepoint:
util/scripting-engines/trace-event-perl.c:285: error: format
expects type 'int', but argument 2 has type '__u64'
make: *** [util/scripting-engines/trace-event-perl.o] Error 1
Fix it by using the "%PRIu64" for __u64.
v2: use PRIu64 as suggested by Arnaldo.
Signed-off-by: Feng Tang <feng.tang@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20120828101730.6b2fd97e@feng-i7 Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Thu, 13 Sep 2012 04:14:30 +0000 (13:14 +0900)]
perf report: Add missing perf_hpp__init for pipe-mode
The perf_hpp__init() function was only called from setup_browser() so
that the pipe-mode missed the initialization thus didn't respond to
related options. Fix it.
Reported-by: Robert Richter <robert.richter@amd.com> Tested-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-tip-commits@vger.kernel.org Link: http://lkml.kernel.org/r/87txv28spl.fsf_-_@sejong.aot.lge.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf symbols: Filter samples with unresolved symbol when "--symbols" option is used
Report/top commands support to only handle specific symbols with
"--symbols" option, but current code will keep those samples whose
symbol can't be resolved, which should actually be filtered.
If we run following commands:
$perf record -a tree
$perf report --symbols intel_idle -n
the output will be:
Without the patch:
==================
46.27% 156 sshd [unknown]
26.05% 48 swapper [kernel.kallsyms]
17.26% 38 tree libc-2.12.1.so
7.69% 17 tree tree
2.73% 6 tree ld-2.12.1.so
With the patch:
===============
100.00% 48 swapper [kernel.kallsyms]
Signed-off-by: Feng Tang <feng.tang@intel.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1347007349-3102-2-git-send-email-feng.tang@intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Namhyung Kim [Fri, 14 Sep 2012 08:35:28 +0000 (17:35 +0900)]
perf report: Enable integrated annotation only if possible
The integrated annotation feature is supported only in TUI mode. Also
it should be enabled with 'symbol' sort key otherwise resulting hist
entry doesn't need to have same symbol as of a sample so that it can
fail on hist_entry__inc_addr_samples with -ERANGE.
You can easily see it when start perf report TUI without symbol* sort
key. This patch fixes the problem.
arch_uprobe_disable_step() should also take UTASK_SSTEP_TRAPPED into
account. In this case the probed insn was not executed, we need to
clear X86_EFLAGS_TF if it was set by us and that is all.
Again, this code will look more clean when we move it into
arch_uprobe_post_xol() and arch_uprobe_abort_xol().
uprobes/x86: Xol should send SIGTRAP if X86_EFLAGS_TF was set
arch_uprobe_disable_step() correctly preserves X86_EFLAGS_TF and
returns to user-mode. But this means the application gets SIGTRAP
only after the next insn.
This means that UPROBE_CLEAR_TF logic is not really right. _enable
should only record the state of X86_EFLAGS_TF, and _disable should
check it separately from UPROBE_FIX_SETF.
Remove arch_uprobe_task->restore_flags, add ->saved_tf instead, and
change enable/disable accordingly. This assumes that the probed insn
was not trapped, see the next patch.
arch_uprobe_skip_sstep() logic has the same problem, change it to
check X86_EFLAGS_TF and send SIGTRAP as well. We will cleanup this
all after we fold enable/disable_step into pre/post_hol hooks.
Note: send_sig(SIGTRAP) is not actually right, we need send_sigtrap().
But this needs more changes, handle_swbp() does the same and this is
equally wrong.
uprobes/x86: Do not (ab)use TIF_SINGLESTEP/user_*_single_step() for single-stepping
user_enable/disable_single_step() was designed for ptrace, it assumes
a single user and does unnecessary and wrong things for uprobes. For
example:
- arch_uprobe_enable_step() can't trust TIF_SINGLESTEP, an
application itself can set X86_EFLAGS_TF which must be
preserved after arch_uprobe_disable_step().
- we do not want to set TIF_SINGLESTEP/TIF_FORCED_TF in
arch_uprobe_enable_step(), this only makes sense for ptrace.
- otoh we leak TIF_SINGLESTEP if arch_uprobe_disable_step()
doesn't do user_disable_single_step(), the application will
be killed after the next syscall.
- arch_uprobe_enable_step() does access_process_vm() we do
not need/want.
Change arch_uprobe_enable/disable_step() to set/clear X86_EFLAGS_TF
directly, this is much simpler and more correct. However, we need to
clear TIF_BLOCKSTEP/DEBUGCTLMSR_BTF before executing the probed insn,
add set_task_blockstep(false).
Note: with or without this patch, there is another (hopefully minor)
problem. A probed "pushf" insn can see the wrong X86_EFLAGS_TF set by
uprobes. Perhaps we should change _disable to update the stack, or
teach arch_uprobe_skip_sstep() to emulate this insn.
Afaics the usage of update_debugctlmsr() and TIF_BLOCKSTEP in
step.c was always very wrong.
1. update_debugctlmsr() was simply unneeded. The child sleeps
TASK_TRACED, __switch_to_xtra(next_p => child) should notice
TIF_BLOCKSTEP and set/clear DEBUGCTLMSR_BTF after resume if
needed.
2. It is wrong. The state of DEBUGCTLMSR_BTF bit in CPU register
should always match the state of current's TIF_BLOCKSTEP bit.
3. Even get_debugctlmsr() + update_debugctlmsr() itself does not
look right. Irq can change other bits in MSR_IA32_DEBUGCTLMSR
register or the caller can be preempted in between.
4. It is not safe to play with TIF_BLOCKSTEP if task != current.
DEBUGCTLMSR_BTF and TIF_BLOCKSTEP should always match each
other if the task is running. The tracee is stopped but it
can be SIGKILL'ed right before set/clear_tsk_thread_flag().
However, now that uprobes uses user_enable_single_step(current)
we can't simply remove update_debugctlmsr(). So this patch adds
the additional "task == current" check and disables irqs to avoid
the race with interrupts/preemption.
Unfortunately this patch doesn't solve the last problem, we need
another fix. Probably we should teach ptrace_stop() to set/clear
single/block stepping after resume.
And afaics there is yet another problem: perf can play with
MSR_IA32_DEBUGCTLMSR from nmi, this obviously means that even
__switch_to_xtra() has problems.
uprobes/x86: Implement x86 specific arch_uprobe_*_step
The arch specific implementation behaves like user_enable_single_step()
except that it does not disable single stepping if it was already
enabled by ptrace. This allows the debugger to single step over an
uprobe. The state of block stepping is not restored. It makes only sense
together with TF and if that was enabled then the debugger is notified.
Note: this is still not correct. For example, TIF_SINGLESTEP check
is not right, the application itself can set X86_EFLAGS_TF. And otoh
we leak TIF_SINGLESTEP (set by enable) if the probed insn is "popf".
See the next patches, we need the changes in arch/x86/kernel/step.c
first.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Add the new "__weak arch" helpers which simply call user_*_single_step()
as a preparation. This is only needed to not break the powerpc port, we
will fold this logic into arch_uprobe_pre/post_xol() hooks later.
We should also change handle_singlestep(), _disable_step(&uprobe->arch)
should be called before put_uprobe().
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Oleg Nesterov [Sun, 19 Aug 2012 15:41:34 +0000 (17:41 +0200)]
uprobes: Teach find_active_uprobe() to clear MMF_HAS_UPROBES
The wrong MMF_HAS_UPROBES doesn't really hurt, just it triggers
the "slow" and unnecessary handle_swbp() path if the task hits
the non-uprobe breakpoint.
So this patch changes find_active_uprobe() to check every valid
vma and clear MMF_HAS_UPROBES if no uprobes were found. This is
adds the slow O(n) path, but it is only called in unlikely case
when the task hits the normal breakpoint first time after
uprobe_unregister().
Note the "not strictly accurate" comment in mmf_recalc_uprobes().
We can fix this, we only need to teach vma_has_uprobes() to return
a bit more more info, but I am not sure this worth the trouble.
Oleg Nesterov [Sun, 19 Aug 2012 14:15:09 +0000 (16:15 +0200)]
uprobes: Introduce MMF_RECALC_UPROBES
Add the new MMF_RECALC_UPROBES flag, it means that MMF_HAS_UPROBES
can be false positive after remove_breakpoint() or uprobe_munmap().
It is also set by uprobe_dup_mmap(), this is not optimal but simple.
We could add the new hook, uprobe_dup_vma(), to set MMF_HAS_UPROBES
only if the new mm actually has uprobes, but I don't think this
makes sense.
The next patch will use this flag to clear MMF_HAS_UPROBES.
projects / karo-tx-linux.git / log