Perf hooks allow hooking user code at perf events. They can be used for
manipulation of BPF maps, taking snapshot and reporting results. In this
patch two perf hook points are introduced: record_start and record_end.
To avoid buggy user actions, a SIGSEGV signal handler is introduced into
'perf record'. It turns off perf hook if it causes a segfault and report
an error to help debugging.
A test case for perf hook is introduced.
Test result:
$ ./buildperf/perf test -v hook
50: Test perf hooks :
--- start ---
test child forked, pid 10311
SIGSEGV is observed as expected, try to recover.
Fatal error (SEGFAULT) in perf hook 'test'
test child finished with 0
---- end ----
Test perf hooks: Ok
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Joe Stringer <joe@ovn.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/20161126070354.141764-5-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
New features:
- Improve ARM support in the annotation code, affecting 'perf annotate', 'perf
report' and live annotation in 'perf top' (Kim Phillips)
- Initial support for PowerPC in the annotation code (Ravi Bangoria)
- Skip repetitive scheduler function on the top of the stack in
'perf sched timehist' (Namhyung Kim)
Fixes:
- Fix maps resolution in libbpf (Eric Leblond)
- Get the kernel signature via /proc/version_signature, available on
Ubuntu systems, to make sure BPF proggies works, as the one provided
via 'uname -r' doesn't (Wang Nan)
- Fix segfault in 'perf record' when running with suid and kptr_restrict
is 1 (Wang Nan)
Infrastructure changes:
- Support per-arch instruction tables, kept via a static or dynamic table
(Arnaldo Carvalho de Melo)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
It is not correct to assimilate the elf data of the maps section to an
array of map definition. In fact the sizes differ. The offset provided
in the symbol section has to be used instead.
This patch fixes a bug causing a elf with two maps not to load
correctly.
Wang Nan added:
This patch requires a name for each BPF map, so array of BPF maps is not
allowed. This restriction is reasonable, because kernel verifier forbid
indexing BPF map from such array unless the index is a fixed value, but
if the index is fixed why not merging it into name?
For example:
Program like this:
...
unsigned long cpu = get_smp_processor_id();
int *pval = map_lookup_elem(&map_array[cpu], &key);
...
Generates bytecode like this:
0: (b7) r1 = 0
1: (63) *(u32 *)(r10 -4) = r1
2: (b7) r1 = 680997
3: (63) *(u32 *)(r10 -8) = r1
4: (85) call 8
5: (67) r0 <<= 4
6: (18) r1 = 0x112dd000
8: (0f) r0 += r1
9: (bf) r2 = r10
10: (07) r2 += -4
11: (bf) r1 = r0
12: (85) call 1
Where instruction 8 is the computation, 8 and 11 render r1 to an invalid
value for function map_lookup_elem, causes verifier report error.
Signed-off-by: Eric Leblond <eric@regit.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Wang Nan <wangnan0@huawei.com>
[ Merge bpf_object__init_maps_name into bpf_object__init_maps.
Fix segfault for buggy BPF script Validate obj->maps ]
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/20161115040617.69788-5-wangnan0@huawei.com
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Before this patch perf panics if kptr_restrict is set to 1 and perf is
owned by root with suid set:
$ whoami
wangnan
$ ls -l ./perf
-rwsr-xr-x 1 root root 19781908 Sep 21 19:29 /home/wangnan/perf
$ cat /proc/sys/kernel/kptr_restrict
1
$ cat /proc/sys/kernel/perf_event_paranoid
-1
$ ./perf record -a
Segmentation fault (core dumped)
$
The reason is that perf assumes it is allowed to read kptr from
/proc/kallsyms when euid is root, but in fact the kernel doesn't allow
reading kptr when euid and uid do not match with each other:
$ cp /bin/cat .
$ sudo chown root:root ./cat
$ sudo chmod u+s ./cat
$ cat /proc/kallsyms | grep do_fork
0000000000000000 T _do_fork <--- kptr is hidden even euid is root
$ sudo cat /proc/kallsyms | grep do_fork
ffffffff81080230 T _do_fork
See lib/vsprintf.c for kernel side code.
This patch fixes this problem by checking both uid and euid.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/20161115040617.69788-3-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
On ubuntu the internal kernel version code is different from what can
be retrived from uname:
$ uname -r
4.4.0-47-generic
$ cat /lib/modules/`uname -r`/build/include/generated/uapi/linux/version.h
#define LINUX_VERSION_CODE 263192
#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
$ cat /lib/modules/`uname -r`/build/include/generated/utsrelease.h
#define UTS_RELEASE "4.4.0-47-generic"
#define UTS_UBUNTU_RELEASE_ABI 47
$ cat /proc/version_signature
Ubuntu 4.4.0-47.68-generic 4.4.24
The macro LINUX_VERSION_CODE is set to 4.4.24 (263192 == 0x40418), but
`uname -r` reports 4.4.0.
This mismatch causes LINUX_VERSION_CODE macro passed to BPF script become
an incorrect value, results in magic failure in BPF loading:
$ sudo ./buildperf/perf record -e ./tools/perf/tests/bpf-script-example.c ls
event syntax error: './tools/perf/tests/bpf-script-example.c'
\___ Failed to load program for unknown reason
According to Ubuntu document (https://wiki.ubuntu.com/Kernel/FAQ), the
correct kernel version can be retrived through /proc/version_signature, which
is ubuntu specific.
This patch checks the existance of /proc/version_signature, and returns
version number through parsing this file instead of uname. Version string
is untouched (value returns from uname) because `uname -r` is required
to be consistence with path of kbuild directory in /lib/module.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/20161115040617.69788-2-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
New tool:
- 'perf sched timehist' provides an analysis of scheduling events.
Example usage:
perf sched record -- sleep 1
perf sched timehist
By default it shows the individual schedule events, including the wait
time (time between sched-out and next sched-in events for the task), the
task scheduling delay (time between wakeup and actually running) and run
time for the task:
time cpu task name wait time sch delay run time
[tid/pid] (msec) (msec) (msec)
-------- ------ ---------------- --------- --------- --------
1.874569 [0011] gcc[31949] 0.014 0.000 1.148
1.874591 [0010] gcc[31951] 0.000 0.000 0.024
1.874603 [0010] migration/10[59] 3.350 0.004 0.011
1.874604 [0011] <idle> 1.148 0.000 0.035
1.874723 [0005] <idle> 0.016 0.000 1.383
1.874746 [0005] gcc[31949] 0.153 0.078 0.022
...
Times are in msec.usec. (David Ahern, Namhyung Kim)
Improvements:
- Make 'perf c2c report' support -f/--force, to allow skipping the
ownership check for root users, for instance, just like the other
tools (Jiri Olsa)
- Allow sorting cachelines by total number of HITMs, in addition to
local and remote numbers (Jiri Olsa)
Fixes:
- Make sure errors aren't suppressed by the TUI reset at the end of
a 'perf c2c report' session (Jiri Olsa)
Infrastructure changes:
- Initial work on having the annotate code better support multiple
architectures, including the ability to cross-annotate, i.e. to
annotate perf.data files collected on an ARM system on a x86_64
workstation (Arnaldo Carvalho de Melo, Ravi Bangoria, Kim Phillips)
- Use USECS_PER_SEC instead of hard coded number in libtraceevent (Steven Rostedt)
- Add retrieval of preempt count and latency flags in libtraceevent (Steven Rostedt)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull NFS client bugfixes from Anna Schumaker:
"Most of these fix regressions or races, but there is one patch for
stable that Arnd sent me
Stable bugfix:
- Hide array-bounds warning
Bugfixes:
- Keep a reference on lock states while checking
- Handle NFS4ERR_OLD_STATEID in nfs4_reclaim_open_state
- Don't call close if the open stateid has already been cleared
- Fix CLOSE rases with OPEN
- Fix a regression in DELEGRETURN"
* tag 'nfs-for-4.9-4' of git://git.linux-nfs.org/projects/anna/linux-nfs:
NFSv4.x: hide array-bounds warning
NFSv4.1: Keep a reference on lock states while checking
NFSv4.1: Handle NFS4ERR_OLD_STATEID in nfs4_reclaim_open_state
NFSv4: Don't call close if the open stateid has already been cleared
NFSv4: Fix CLOSE races with OPEN
NFSv4.1: Fix a regression in DELEGRETURN
Pull arch/tile bugfix from Chris Metcalf:
"This fixes a bug that causes reboots after 208 days of uptime :-)"
* 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
tile: avoid using clocksource_cyc2ns with absolute cycle count
For large values of "mult" and long uptimes, the intermediate
result of "cycles * mult" can overflow 64 bits. For example,
the tile platform calls clocksource_cyc2ns with a 1.2 GHz clock;
we have mult = 853, and after 208.5 days, we overflow 64 bits.
Since clocksource_cyc2ns() is intended to be used for relative
cycle counts, not absolute cycle counts, performance is more
importance than accepting a wider range of cycle values. So,
just use mult_frac() directly in tile's sched_clock().
Commit 4cecf6d401 ("sched, x86: Avoid unnecessary overflow
in sched_clock") by Salman Qazi results in essentially the same
generated code for x86 as this change does for tile. In fact,
a follow-on change by Salman introduced mult_frac() and switched
to using it, so the C code was largely identical at that point too.
Peter Zijlstra then added mul_u64_u32_shr() and switched x86
to use it. This is, in principle, better; by optimizing the
64x64->64 multiplies to be 32x32->64 multiplies we can potentially
save some time. However, the compiler piplines the 64x64->64
multiplies pretty well, and the conditional branch in the generic
mul_u64_u32_shr() causes some bubbles in execution, with the
result that it's pretty much a wash. If tilegx provided its own
implementation of mul_u64_u32_shr() without the conditional branch,
we could potentially save 3 cycles, but that seems like small gain
for a fair amount of additional build scaffolding; no other platform
currently provides a mul_u64_u32_shr() override, and tile doesn't
currently have an <asm/div64.h> header to put the override in.
Additionally, gcc currently has an optimization bug that prevents
it from recognizing the opportunity to use a 32x32->64 multiply,
and so the result would be no better than the existing mult_frac()
until such time as the compiler is fixed.
For now, just using mult_frac() seems like the right answer.
Cc: stable@kernel.org [v3.4+]
Signed-off-by: Chris Metcalf <cmetcalf@mellanox.com>
Pull perf fixes from Ingo Molnar:
"Six fixes for bugs that were found via fuzzing, and a trivial
hw-enablement patch for AMD Family-17h CPU PMUs"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/x86/intel/uncore: Allow only a single PMU/box within an events group
perf/x86/intel: Cure bogus unwind from PEBS entries
perf/x86: Restore TASK_SIZE check on frame pointer
perf/core: Fix address filter parser
perf/x86: Add perf support for AMD family-17h processors
perf/x86/uncore: Fix crash by removing bogus event_list[] handling for SNB client uncore IMC
perf/core: Do not set cpuctx->cgrp for unscheduled cgroups
If callchains were recorded they are appended to the line with a default stack depth of 5:
1.874569 [0011] gcc[31949] 0.014 0.000 1.148 wait_for_completion_killable <- do_fork <- sys_vfork <- stub_vfork <- __vfork
1.874591 [0010] gcc[31951] 0.000 0.000 0.024 __cond_resched <- _cond_resched <- wait_for_completion <- stop_one_cpu <- sched_exec
1.874603 [0010] migration/10[59] 3.350 0.004 0.011 smpboot_thread_fn <- kthread <- ret_from_fork
1.874604 [0011] <idle> 1.148 0.000 0.035 cpu_startup_entry <- start_secondary
1.874723 [0005] <idle> 0.016 0.000 1.383 cpu_startup_entry <- start_secondary
1.874746 [0005] gcc[31949] 0.153 0.078 0.022 do_wait sys_wait4 <- system_call_fastpath <- __GI___waitpid
--no-call-graph can be used to not show the callchains. --max-stack is used
to control the number of frames shown (default of 5). -x/--excl options can
be used to collapse redundant callchains to get more relevant data on screen.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20161116060634.28477-7-namhyung@kernel.org
[ Add documentation based on above commit message ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The -s/--summary option is to show process runtime statistics. And the
-S/--with-summary option is to show the stats with the normal output.
$ perf sched timehist -s
Runtime summary
comm parent sched-in run-time min-run avg-run max-run stddev
(count) (msec) (msec) (msec) (msec) %
---------------------------------------------------------------------------------------------------------
ksoftirqd/0[3] 2 2 0.011 0.004 0.005 0.006 14.87
rcu_preempt[7] 2 11 0.071 0.002 0.006 0.017 20.23
watchdog/0[11] 2 1 0.002 0.002 0.002 0.002 0.00
watchdog/1[12] 2 1 0.004 0.004 0.004 0.004 0.00
...
Terminated tasks:
sleep[7220] 7219 3 0.770 0.087 0.256 0.576 62.28
Idle stats:
CPU 0 idle for 2352.006 msec
CPU 1 idle for 2764.497 msec
CPU 2 idle for 2998.229 msec
CPU 3 idle for 2967.800 msec
Total number of unique tasks: 52
Total number of context switches: 2532
Total run time (msec): 218.036
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20161116060634.28477-5-namhyung@kernel.org
[ Add documentation from last commit, so that docs comes with the cset that introduces the feature ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
'perf sched timehist' provides an analysis of scheduling events.
Example usage:
perf sched record -- sleep 1
perf sched timehist
By default it shows the individual schedule events, including the wait
time (time between sched-out and next sched-in events for the task), the
task scheduling delay (time between wakeup and actually running) and run
time for the task:
time cpu task name wait time sch delay run time
[tid/pid] (msec) (msec) (msec)
-------------- ------ -------------------- --------- --------- ---------
79371.874569 [0011] gcc[31949] 0.014 0.000 1.148
79371.874591 [0010] gcc[31951] 0.000 0.000 0.024
79371.874603 [0010] migration/10[59] 3.350 0.004 0.011
79371.874604 [0011] <idle> 1.148 0.000 0.035
79371.874723 [0005] <idle> 0.016 0.000 1.383
79371.874746 [0005] gcc[31949] 0.153 0.078 0.022
...
Times are in msec.usec.
Committer note:
Add above explanation as the 'perf sched timehist' entry for 'man
perf-sched'.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20161116060634.28477-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It is useful for debug to see file descriptors for each event.
Before:
$ perf stat -vvv -e cycles,cache-misses ls
...
sys_perf_event_open: pid 12146 cpu -1 group_fd -1 flags 0x8
...
sys_perf_event_open: pid 12146 cpu -1 group_fd 3 flags 0x8
sys_perf_event_open failed, error -13
Now:
$ perf stat -vvv -e cycles,cache-misses ls
...
sys_perf_event_open: pid 12858 cpu -1 group_fd -1 flags 0x8 = 3
...
sys_perf_event_open: pid 12858 cpu -1 group_fd 3 flags 0x8
sys_perf_event_open failed, error -13
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Joe Mario <jmario@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1479764011-10732-2-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull crypto fixes from Herbert Xu:
"The last push broke algif_hash for all shash implementations, so this
is a follow-up to fix that.
This also fixes a problem in the crypto scatterwalk that triggers a
BUG_ON with certain debugging options due to the new vmalloced-stack
code"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: scatterwalk - Remove unnecessary aliasing check in map_and_copy
crypto: algif_hash - Fix result clobbering in recvmsg
Pull thermal management fix from Zhang Rui:
"We only have one urgent fix this time.
Commit 3105f234e0 ("thermal/powerclamp: correct cpu support check"),
which is shipped in 4.9-rc3, fixed a problem introduced by commit
b721ca0d19 ("thermal/powerclamp: remove cpu whitelist").
But unfortunately, it broke intel_powerclamp driver module auto-
loading at the same time. Thus we need this change to add back module
auto-loading for 4.9"
* 'for-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux:
thermal/powerclamp: add back module device table
Pull SCSI fixes from James Bottomley:
"Two small fixes.
One prevents timeouts on mpt3sas when trying to use the secure erase
protocol which causes the erase protocol to be aborted. The second is
a regression in a prior fix which causes all commands to abort during
PCI extended error recovery, which is incorrect because PCI EEH is
independent from what's happening on the FC transport"
* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
scsi: qla2xxx: do not abort all commands in the adapter during EEH recovery
scsi: mpt3sas: Fix secure erase premature termination
Pull clk fixes from Stephen Boyd:
"A handful of driver fixes.
The sunxi fixes are for an incorrect clk tree configuration and a bad
frequency calculation. The other two are fixes for passing the wrong
pointer in drivers recently converted to clk_hw style registration"
* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
clk: efm32gg: Pass correct type to hw provider registration
clk: berlin: Pass correct type to hw provider registration
clk: sunxi: Fix M factor computation for APB1
clk: sunxi-ng: sun6i-a31: Force AHB1 clock to use PLL6 as parent
A correct bugfix introduced a harmless warning that shows up with gcc-7:
fs/nfs/callback.c: In function 'nfs_callback_up':
fs/nfs/callback.c:214:14: error: array subscript is outside array bounds [-Werror=array-bounds]
What happens here is that the 'minorversion == 0' check tells the
compiler that we assume minorversion can be something other than 0,
but when CONFIG_NFS_V4_1 is disabled that would be invalid and
result in an out-of-bounds access.
The added check for IS_ENABLED(CONFIG_NFS_V4_1) tells gcc that this
really can't happen, which makes the code slightly smaller and also
avoids the warning.
The bugfix that introduced the warning is marked for stable backports,
we want this one backported to the same releases.
Fixes: 98b0f80c23 ("NFSv4.x: Fix a refcount leak in nfs_callback_up_net")
Cc: stable@vger.kernel.org # v3.7+
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Pull scheduler fixes from Ingo Molnar:
"Two fixes for autogroup scheduling, for races when turning the feature
on/off via /proc/sys/kernel/sched_autogroup_enabled"
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/autogroup: Do not use autogroup->tg in zombie threads
sched/autogroup: Fix autogroup_move_group() to never skip sched_move_task()
Pull x86 fixes from Ingo Molnar:
"Misc fixes:
- two fixes to make (very) old Intel CPUs boot reliably
- fix the intel-mid driver and rename it
- two KASAN false positive fixes
- an FPU fix
- two sysfb fixes
- two build fixes related to new toolchain versions"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/platform/intel-mid: Rename platform_wdt to platform_mrfld_wdt
x86/build: Build compressed x86 kernels as PIE when !CONFIG_RELOCATABLE as well
x86/platform/intel-mid: Register watchdog device after SCU
x86/fpu: Fix invalid FPU ptrace state after execve()
x86/boot: Fail the boot if !M486 and CPUID is missing
x86/traps: Ignore high word of regs->cs in early_fixup_exception()
x86/dumpstack: Prevent KASAN false positive warnings
x86/unwind: Prevent KASAN false positive warnings in guess unwinder
x86/boot: Avoid warning for zero-filling .bss
x86/sysfb: Fix lfb_size calculation
x86/sysfb: Add support for 64bit EFI lfb_base
