Just like we do with the other DSOs. This also simplifies the
kernel_maps setup process, now all that the tools need to do is
to call kernel_maps__init and the maps for the modules and
kernel will be created, then, later, when
kernel_maps__find_symbol() is used, it will also call
maps__find_symbol that already checks if the symtab was loaded,
loading it if needed.
Now if one does 'perf top --hide_kernel_symbols' we won't pay
the price of loading the (many) symbols in /proc/kallsyms or
vmlinux.
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: <1258757489-5978-4-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In the kernel we have more than one notes section, so the linker
script combines all and puts them into a ".notes" combined
section. So we need to look at both sections and also traverse
them looking at multiple GElf_Nhdr entries till we find the one
we want, with the build_id.
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: <1258757489-5978-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
It should just load kernel symbols, not load the list of
modules. There are more stuff to move to other routines, but
lets do it in several steps.
End goal is to be able to defer symbol table loading till we
find a hit for that map address range. So that the kernel &
modules are handled just like all the other DSOs in the system.
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: <1258757489-5978-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
[root@doppio linux-2.6-tip]# perf record -a -f sleep 3s ; perf
buildid-list | grep vmlinux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.171 MB perf.data (~7489
samples) ] 18e7cc53db62a7d35e9d6f6c9ddc23017d38ee9a vmlinux
[root@doppio linux-2.6-tip]#
Several refactorings were needed so that we can have symmetry
between dsos__load_modules() and dsos__load_kernel(), i.e. those
functions will respectively create and add to the dsos list the
loaded modules and kernel, with its buildids, but not load its
symbols. That is something the subcomands that need will have to
call dso__load_kernel_sym(), just like we do with modules with
dsos__load_module_sym()/dso__load_module_sym().
Next csets will actually use this info to stop producing bogus
results using mismatched vmlinux and .ko files.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Roland McGrath <roland@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: <1258582853-8579-4-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
No need for this struct and its allocations, we can just use the
->build_id member we already have in struct dso, then ask for it
to be read, and later traverse the dsos list, writing the
buildid table to the perf.data file.
As a bonus, one more die() function got killed.
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: <1258582853-8579-2-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When we read the build_id from the DSO name to then index into
/usr/lib/debug/.buildid/DSO_BUILD_ID[0:2]/DSO_BUILD_ID[2:], we
were jumping directly to the comparision with the buildid we
already have in dso->build_id (that came from the perf.data
build_id section, collected at perf record time)
unconditionally, even if we didn't had recorded it, and
furthermore, comparing a formatted buildid with a rawbuildid, yikes.
Fix it by deleting the dso__read_build_id() function, that was
really misdesigned anyway, and do the necessary checks and
correct comparison of raw buildids.
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: <1258582853-8579-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We need to pass the symbol to the filter so that, for instance,
'perf top' can do filtering and also set the private area it
manages, setting the ->map pointer, etc.
I found this while running 'perf top' on a machine where hits
happened on PLT symbols, where ->map wasn't being set up and
segfaults thus happened.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1258386491-20278-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We are saving the build id once we stop the profiling. And only
after doing that we know if we need to set that feature in the
header through the feature bitmap.
But if we want a proper feature support in the headers, using a
rule of offset/size pairs in sections, we need to know in
advance how many features we need to set in the headers, so that
we can reserve rooms for their section headers.
The current state doesn't allow that, as it forces us to first
save the build-ids to the file right after the datas instead of
planning any structured layout.
That's why this splits up the build-ids processing in two parts:
one that fetches the build-ids from the Dso objects, and one
that saves them into the file.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
LKML-Reference: <1257911467-28276-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Conflicts:
tools/perf/Makefile
Merge reason: Resolve the conflict, merge to upstream and merge in
perf fixes so we can add a dependent patch.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We need this because we get section relative addresses when
reading the symtabs, but when a tool like 'perf annotate' needs
to match these address to what 'objdump -dS' produces we need
the address + section back again.
So in annotate now we look at the 'struct hist_entry' instances
(that weren't really being used) so that we iterate only over
the symbols that had some hit and get the map where that
particular hit happened so that we can get the right address to
match with annotate.
Verified that at least:
perf annotate mmap_read_counter # Uses the ~/bin/perf binary
perf annotate --vmlinux /home/acme/git/build/perf/vmlinux intel_pmu_enable_all
on a 'perf record perf top' session seems to work.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1255979877-12533-1-git-send-email-acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This removes the ovelapping of vmlinux addresses with modules,
using the ELF section name when using --vmlinux and creating a
unique DSO name when using /proc/kallsyms ([kernel].N).
This is done by creating multiple 'struct map' instances for
address ranges backed by DSOs that have just the symbols for that
range and a name that is derived from the ELF section name.o
Now it is possible to ask for just the symbols in some particular
kernel section:
$ perf report -m --vmlinux ../build/tip-recvmmsg/vmlinux \
--dsos [kernel].vsyscall_fn | head -15
52.73% Xorg [.] vread_hpet
18.61% firefox [.] vread_hpet
14.50% npviewer.bin [.] vread_hpet
6.83% compiz [.] vread_hpet
5.73% glxgears [.] vread_hpet
0.63% java [.] vread_hpet
0.30% gnome-terminal [.] vread_hpet
0.23% perf [.] vread_hpet
0.18% xchat [.] vread_hpet
$
Now we don't have to first lookup the list of modules and then, if
it fails, vmlinux symbols, its just a simple lookup for the map
then the symbols, just like for threads.
Reports generated using /proc/kallsyms and --vmlinux should provide
the same results, modulo the DSO name for sections other than
".text".
But they don't right now because things like:
ffffffff81011c20-ffffffff81012068 system_call
ffffffff81011c30-ffffffff81011c9b system_call_after_swapgs
ffffffff81011c9c-ffffffff81011cb6 system_call_fastpath
ffffffff81011cb7-ffffffff81011cbb ret_from_sys_call
I.e. overlapping symbols, again some ASM special case that we have
to fixup.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1254934136-8503-1-git-send-email-acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Arnaldo Carvalho de Melo
Frederic Weisbecker
Ingo Molnar
Marti Raudsepp