Implementing a new --smi-cost mode in perf stat to measure SMI cost.
During the measurement, the /sys/device/cpu/freeze_on_smi will be set.
The measurement can be done with one counter (unhalted core cycles), and
two free running MSR counters (IA32_APERF and SMI_COUNT).
In practice, the percentages of SMI core cycles should be more useful
than absolute value. So the output will be the percentage of SMI core
cycles and SMI#. metric_only will be set by default.
SMI cycles% = (aperf - unhalted core cycles) / aperf
Here is an example output.
Performance counter stats for 'sudo echo ':
SMI cycles% SMI#
0.1% 1
0.010858678 seconds time elapsed
Users who wants to get the actual value can apply additional
--no-metric-only.
Signed-off-by: Kan Liang <Kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1495825538-5230-3-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
While trying to reduce util.[ch] I noticed that fetch_kernel_version()
and fetch_ubuntu_kernel_version() do lots of operations only to check if
they are needed, i.e. it checks if the pointer where to return the
kernel version is NULL only after obtaining the kernel version from
/proc/version_signature or by parsing the results from uname().
Do it earlier not to confuse people reading this code in the future :-)
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-i94qwyekk4tzbu0b9ce1r1mz@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- Allow adding and removing fields to the default 'perf script' columns,
using + or - as field prefixes to do so (Andi Kleen)
- Display titles in left frame in the annotate browser (Jin Yao)
- Allow resolving the DSO name with 'perf script -F brstack{sym,off},dso'
(Mark Santaniello)
- Support function filtering in 'perf ftrace' (Namhyung Kim)
- Allow specifying function call depth in 'perf ftrace' (Namhyumg Kim)
Infrastructure changes:
- Adopt __noreturn, __printf, __scanf, noinline, __packed and __aligned
__alignment__(()) markers, to make the tools/ source code base to be
more compact and look more like kernel code (Arnaldo Carvalho de Melo)
- Remove unnecessary check in annotate_browser_write() (Jin Yao)
- Return arch from symbol__disassemble() so that callers, such as
the annotate TUI browser to use arch specific formattings, such
as the upcoming instruction micro-op fusion on Intel Core (Jin Yao)
- Remove superfluous check before use in the coresight code base (Kim
Phillips)
- Remove unused SAMPLE_SIZE defines and BTS priv array (Kim Phillips)
- Error handling fix/tidy ups in 'perf config' (Taeung Song)
- Avoid error in the BPF proggie built with clang in 'perf test llvm'
when PROFILE_ALL_BRANCHES is set (Wang Nan)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The idea here is to make AutoFDO easier in cloud environment with ASLR.
It's easiest to show how this is useful by example. I built a small test
akin to "while(1) { do_nothing(); }" where the do_nothing function is
loaded from a dso:
$ cat burncpu.cpp
#include <dlfcn.h>
int main() {
void* handle = dlopen("./dso.so", RTLD_LAZY);
if (!handle) return -1;
typedef void (*fp)();
fp do_nothing = (fp) dlsym(handle, "do_nothing");
while(1) {
do_nothing();
}
}
$ cat dso.cpp
extern "C" void do_nothing() {}
$ cat build.sh
#!/bin/bash
g++ -shared dso.cpp -o dso.so
g++ burncpu.cpp -o burncpu -ldl
I sampled the execution of this program with perf record -b.
Using the existing "brstack,dso", we get absolute addresses that are
affected by ASLR, and could be different on different hosts. The address
does not uniquely identify a branch/target in the binary:
$ perf script -F brstack,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x7f967139b6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
Using the existing "brstacksym,dso" is a little better, because the
symbol plus offset and dso name *does* uniquely identify a branch/target
in the binary. Ultimately, however, AutoFDO wants a simple offset into
the binary, so we'd have to undo all the work perf did to symbolize in
the first place:
$ perf script -F brstacksym,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
do_nothing+0x5(/tmp/burncpu/dso.so)/main+0x44(/tmp/burncpu/exe)/P/-/-/0
With the new "brstackoff,dso" we get what we need: a simple offset into a
specific dso/binary that uniquely identifies a branch/target:
$ perf script -F brstackoff,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
Signed-off-by: Mark Santaniello <marksan@fb.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170619163825.2012979-2-marksan@fb.com
[ Updated documentation about 'brstackoff' using text from above ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf script can report the dso for "addr" and "ip" fields.
This adds the same support for the "brstack" and "brstacksym" fields.
This can be helpful for AutoFDO: we can ignore LBR entries unless the
source and target address are both in the target module we are about to
build.
I built a small test akin to "while(1) { do_nothing(); }" where the
do_nothing function is loaded from a dso:
$ cat burncpu.cpp
#include <dlfcn.h>
int main() {
void* handle = dlopen("./dso.so", RTLD_LAZY);
if (!handle) return -1;
typedef void (*fp)();
fp do_nothing = (fp) dlsym(handle, "do_nothing");
while(1) {
do_nothing();
}
}
$ cat dso.cpp
extern "C" void do_nothing() {}
$ cat build.sh
#!/bin/bash
g++ -shared dso.cpp -o dso.so
g++ burncpu.cpp -o burncpu -ldl
I sampled the execution with perf record -b. Using the new perf script
functionality I can easily find cases where there was a transition from one
dso to another:
$ perf record -a -b -- sleep 5
[ perf record: Woken up 55 times to write data ]
[ perf record: Captured and wrote 18.815 MB perf.data (43593 samples) ]
$ perf script -F brstack,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x7f967139b6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
$ perf script -F brstacksym,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
do_nothing+0x5(/tmp/burncpu/dso.so)/main+0x44(/tmp/burncpu/exe)/P/-/-/0
Signed-off-by: Mark Santaniello <marksan@fb.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170619163825.2012979-1-marksan@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'if' keyword is a define that expands to complex code when
CONFIG_PROFILE_ALL_BRANCHES is selected, which causes a 'perf test LLVM'
failure like:
$ ./perf test LLVM
35: LLVM search and compile :
35.1: Basic BPF llvm compile : Ok
35.2: kbuild searching : Ok
35.3: Compile source for BPF prologue generation: FAILED!
35.4: Compile source for BPF relocation : Skip
The only affected test case is bpf-script-test-prologue.c
because it uses kernel headers and has 'if' inside.
This patch undefines 'if' to make it passes perf test.
More detailed analysis from a message in this thread, also by Wang:
The problem is caused by following relocation information:
$ readelf -a ./llvmsubtest3
...
[ 5] _ftrace_branch PROGBITS 0000000000000000 00000260
00000000000000a0 0000000000000000 WA 0 0 4
...
Relocation section '.relfunc=null_lseek file->f_mode offset orig' at
offset 0x490 contains 4 entries:
Offset Info Type Sym. Value Sym. Name
000000000038 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
0000000000b0 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
000000000128 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
0000000001c0 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
Relocation section '.rel_ftrace_branch' at offset 0x4d0 contains 8 entries:
Offset Info Type Sym. Value Sym. Name
000000000000 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000008 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000028 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000030 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000050 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000058 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000078 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000080 000100000001 unrecognized: 1 0000000000000015 .L.str
...
So I think the failure is because you enabled CONFIG_PROFILE_ALL_BRANCHES.
I can reproduce your buggy result by selecting
CONFIG_PROFILE_ALL_BRANCHES in my kbuild:
$ ./perf test LLVM
35: LLVM search and compile :
35.1: Basic BPF llvm compile : Ok
35.2: kbuild searching : Ok
35.3: Compile source for BPF prologue generation: FAILED!
35.4: Compile source for BPF relocation : Skip
Simply undef CONFIG_PROFILE_ALL_BRANCHES in clang opts not working
because it is introduced by "#include <uapi/linux/fs.h>", which override
cmdline options. So I think the best way is to undefine 'if' inside BPF
script.
Reported-and-Tested-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Zefan Li <lizefan@huawei.com>
Link: http://lkml.kernel.org/r/20170620183203.2517-1-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
With 'perf script' it is common that we just want to add or remove a field.
Currently this requires figuring out the long list of default fields and
specifying them first, and then adding/removing the new field.
This patch adds a new + - syntax to merely add or remove fields,
that allows more succint and clearer command lines
For example to remove the comm field from PMU samples:
Previously
$ perf script -F tid,cpu,time,event,sym,ip,dso,period | head -1
swapper 0 [000] 504345.383126: 1 cycles: ffffffff90060c66 native_write_msr ([kernel.kallsyms])
with the new syntax
perf script -F -comm | head -1
0 [000] 504345.383126: 1 cycles: ffffffff90060c66 native_write_msr ([kernel.kallsyms])
The new syntax cannot be mixed with normal overriding.
v2: Fix example in description. Use tid vs pid. No functional changes.
v3: Don't skip initialization when user specified explicit type.
v4: Rebase. Remove empty line.
Committer testing:
# perf record -a usleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 1.748 MB perf.data (14 samples) ]
Without a explicit field list specified via -F, defaults to:
# perf script | head -2
perf 6338 [000] 18467.058607: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
swapper 0 [001] 18467.058617: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
#
Which is equivalent to:
# perf script -F comm,tid,cpu,time,period,event,ip,sym,dso | head -2
perf 6338 [000] 18467.058607: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
swapper 0 [001] 18467.058617: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
#
So if we want to remove the comm, as in your original example, we would have to
figure out the default field list and remove ' comm' from it:
# perf script -F tid,cpu,time,period,event,ip,sym,dso | head -2
6338 [000] 18467.058607: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
0 [001] 18467.058617: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
#
With your patch this becomes simpler, one can remove fields by prefixing them
with '-':
# perf script -F -comm | head -2
6338 [000] 18467.058607: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
0 [001] 18467.058617: 1 cycles: ffffffff89060c36 native_write_msr (/lib/modules/4.11.0-rc8+/build/vmlinux)
#
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Milian Wolff <milian.wolff@kdab.com>
Link: http://lkml.kernel.org/r/20170602154810.15875-1-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The annotate browser is divided into 2 frames. Left frame contains 3
columns (some platforms only have one column).
For example:
│26 int compute_flag()
│27 {
22.80 1.20 │ sub $0x8,%rsp
│25 int i;
│
│27 i = rand() % 2;
22.78 1.20 1 │ → callq rand@plt
While it's hard for user to understand what the data is.
This patch adds the titles "Percent", "IPC" and "Cycle" on columns.
Percent IPC Cycle │
│25 __attribute__((noinline))
│26 int compute_flag()
│27 {
22.80 1.20 │ sub $0x8,%rsp
│25 int i;
│
│27 i = rand() % 2;
22.78 1.20 1 │ → callq rand@plt
The titles are displayed at row 0 of annotate browser if row 0 doesn't
have values of percent, ipc and cycle.
Signed-off-by: Yao Jin <yao.jin@linux.intel.com>
Acked-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yao Jin <yao.jin@linux.intel.com>
Link: http://lkml.kernel.org/r/1493909895-9668-3-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf fixes from Thomas Gleixner:
"Three fixes for the perf user space side:
- Fix the probing of precise_ip level, which got broken recently for
x86.
- Unbreak the ARCH=x86_64 build
- Report module before trying to unwind into the module code, which
avoids broken stack frames displayed"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf unwind: Report module before querying isactivation in dwfl unwind
perf tools: Fix build with ARCH=x86_64
perf evsel: Fix probing of precise_ip level for default cycles event
Pull objtool fix from Thomas Gleixner:
"A single fix which adds fortify_panic to the list of no return
functions"
* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
objtool: Add fortify_panic as __noreturn function
Pull networking fixes from David Miller:
1) The netlink attribute passed in to dev_set_alias() is not
necessarily NULL terminated, don't use strlcpy() on it. From
Alexander Potapenko.
2) Fix implementation of atomics in arm64 bpf JIT, from Daniel
Borkmann.
3) Correct the release of netdevs and driver private data in certain
circumstances.
4) Sanitize netlink message length properly in decnet, from Mateusz
Jurczyk.
5) Don't leak kernel data in rtnl_fill_vfinfo() netlink blobs. From
Yuval Mintz.
6) Hash secret is never initialized in ipv6 ILA translation code, from
Arnd Bergmann. I guess those clang warnings about unused inline
functions are useful for something!
7) Fix endian selection in bpf_endian.h, from Daniel Borkmann.
8) Sanitize sockaddr length before dereferncing any fields in AF_UNIX
and CAIF. From Mateusz Jurczyk.
9) Fix timestamping for GMAC3 chips in stmmac driver, from Mario
Molitor.
10) Do not leak netdev on dev_alloc_name() errors in mac80211, from
Johannes Berg.
11) Fix locking in sctp_for_each_endpoint(), from Xin Long.
12) Fix wrong memset size on 32-bit in snmp6, from Christian Perle.
13) Fix use after free in ip_mc_clear_src(), from WANG Cong.
14) Fix regressions caused by ICMP rate limiting changes in 4.11, from
Jesper Dangaard Brouer.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (91 commits)
i40e: Fix a sleep-in-atomic bug
net: don't global ICMP rate limit packets originating from loopback
net/act_pedit: fix an error code
net: update undefined ->ndo_change_mtu() comment
net_sched: move tcf_lock down after gen_replace_estimator()
caif: Add sockaddr length check before accessing sa_family in connect handler
qed: fix dump of context data
qmi_wwan: new Telewell and Sierra device IDs
net: phy: Fix MDIO_THUNDER dependencies
netconsole: Remove duplicate "netconsole: " logging prefix
igmp: acquire pmc lock for ip_mc_clear_src()
r8152: give the device version
net: rps: fix uninitialized symbol warning
mac80211: don't send SMPS action frame in AP mode when not needed
mac80211/wpa: use constant time memory comparison for MACs
mac80211: set bss_info data before configuring the channel
mac80211: remove 5/10 MHz rate code from station MLME
mac80211: Fix incorrect condition when checking rx timestamp
mac80211: don't look at the PM bit of BAR frames
i40e: fix handling of HW ATR eviction
...
Since commit 18e7a45af9 ("perf/x86: Reject non sampling events with
precise_ip") returns -EINVAL for sys_perf_event_open() with an attribute
with (attr.precise_ip > 0 && attr.sample_period == 0), just like is done
in the routine used to probe the max precise level when no events were
passed to 'perf record' or 'perf top', i.e.:
perf_evsel__new_cycles()
perf_event_attr__set_max_precise_ip()
The x86 code, in x86_pmu_hw_config(), which is called all the way from
sys_perf_event_open() did, starting with the aforementioned commit:
/* There's no sense in having PEBS for non sampling events: */
if (!is_sampling_event(event))
return -EINVAL;
Which makes it fail for cycles:ppp, cycles:pp and cycles:p, always using
just the non precise cycles variant.
To make sure that this is the case, I tested it, before this patch,
with:
# perf probe -L x86_pmu_hw_config
<x86_pmu_hw_config@/home/acme/git/linux/arch/x86/events/core.c:0>
0 int x86_pmu_hw_config(struct perf_event *event)
1 {
2 if (event->attr.precise_ip) {
<SNIP>
17 if (event->attr.precise_ip > precise)
18 return -EOPNOTSUPP;
/* There's no sense in having PEBS for non sampling events: */
21 if (!is_sampling_event(event))
22 return -EINVAL;
}
<SNIP>
# perf probe x86_pmu_hw_config:22
Added new events:
probe:x86_pmu_hw_config (on x86_pmu_hw_config:22)
probe:x86_pmu_hw_config_1 (on x86_pmu_hw_config:22)
You can now use it in all perf tools, such as:
perf record -e probe:x86_pmu_hw_config_1 -aR sleep 1
# perf trace -e perf_event_open,probe:x86_pmu_hwconfig*/max-stack=16/ perf record usleep 1
0.000 ( 0.015 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1 ) ...
0.015 ( ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
x86_pmu_hw_config ([kernel.kallsyms])
hsw_hw_config ([kernel.kallsyms])
x86_pmu_event_init ([kernel.kallsyms])
perf_try_init_event ([kernel.kallsyms])
perf_event_alloc ([kernel.kallsyms])
SYSC_perf_event_open ([kernel.kallsyms])
sys_perf_event_open ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
return_from_SYSCALL_64 ([kernel.kallsyms])
syscall (/usr/lib64/libc-2.24.so)
perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
perf_evsel__new_cycles (/home/acme/bin/perf)
perf_evlist__add_default (/home/acme/bin/perf)
cmd_record (/home/acme/bin/perf)
run_builtin (/home/acme/bin/perf)
handle_internal_command (/home/acme/bin/perf)
0.000 ( 0.021 ms): perf/4150 ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
0.023 ( 0.002 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1 ) ...
0.025 ( ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
x86_pmu_hw_config ([kernel.kallsyms])
hsw_hw_config ([kernel.kallsyms])
x86_pmu_event_init ([kernel.kallsyms])
perf_try_init_event ([kernel.kallsyms])
perf_event_alloc ([kernel.kallsyms])
SYSC_perf_event_open ([kernel.kallsyms])
sys_perf_event_open ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
return_from_SYSCALL_64 ([kernel.kallsyms])
syscall (/usr/lib64/libc-2.24.so)
perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
perf_evsel__new_cycles (/home/acme/bin/perf)
perf_evlist__add_default (/home/acme/bin/perf)
cmd_record (/home/acme/bin/perf)
run_builtin (/home/acme/bin/perf)
handle_internal_command (/home/acme/bin/perf)
0.023 ( 0.004 ms): perf/4150 ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
0.028 ( 0.002 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1 ) ...
0.030 ( ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
x86_pmu_hw_config ([kernel.kallsyms])
hsw_hw_config ([kernel.kallsyms])
x86_pmu_event_init ([kernel.kallsyms])
perf_try_init_event ([kernel.kallsyms])
perf_event_alloc ([kernel.kallsyms])
SYSC_perf_event_open ([kernel.kallsyms])
sys_perf_event_open ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
return_from_SYSCALL_64 ([kernel.kallsyms])
syscall (/usr/lib64/libc-2.24.so)
perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
perf_evsel__new_cycles (/home/acme/bin/perf)
perf_evlist__add_default (/home/acme/bin/perf)
cmd_record (/home/acme/bin/perf)
run_builtin (/home/acme/bin/perf)
handle_internal_command (/home/acme/bin/perf)
0.028 ( 0.004 ms): perf/4150 ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
41.018 ( 0.012 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8b5dd0, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
41.065 ( 0.011 ms): perf/4150 perf_event_open(attr_uptr: 0x3c7db78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
41.080 ( 0.006 ms): perf/4150 perf_event_open(attr_uptr: 0x3c7db78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
41.103 ( 0.010 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), group_fd: -1, flags: FD_CLOEXEC) = 4
41.115 ( 0.006 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 1, group_fd: -1, flags: FD_CLOEXEC) = 5
41.122 ( 0.004 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 2, group_fd: -1, flags: FD_CLOEXEC) = 6
41.128 ( 0.008 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 8
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.017 MB perf.data (2 samples) ]
#
I.e. that return -EINVAL in x86_pmu_hw_config() is hit three times.
So fix it by just setting attr.sample_period
Now, after this patch:
# perf trace --max-stack=2 -e perf_event_open,probe:x86_pmu_hw_config* perf record usleep 1
[ perf record: Woken up 1 times to write data ]
0.000 ( 0.017 ms): perf/8469 perf_event_open(attr_uptr: 0x7ffe36c27d10, pid: -1, cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 4
syscall (/usr/lib64/libc-2.24.so)
perf_event_open_cloexec_flag (/home/acme/bin/perf)
0.050 ( 0.031 ms): perf/8469 perf_event_open(attr_uptr: 0x24ebb78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
syscall (/usr/lib64/libc-2.24.so)
perf_evlist__config (/home/acme/bin/perf)
0.092 ( 0.040 ms): perf/8469 perf_event_open(attr_uptr: 0x24ebb78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
syscall (/usr/lib64/libc-2.24.so)
perf_evlist__config (/home/acme/bin/perf)
0.143 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, cpu: -1, group_fd: -1 ) = 4
syscall (/usr/lib64/libc-2.24.so)
perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
0.161 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), group_fd: -1, flags: FD_CLOEXEC) = 4
syscall (/usr/lib64/libc-2.24.so)
perf_evsel__open (/home/acme/bin/perf)
0.171 ( 0.005 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 1, group_fd: -1, flags: FD_CLOEXEC) = 5
syscall (/usr/lib64/libc-2.24.so)
perf_evsel__open (/home/acme/bin/perf)
0.180 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 2, group_fd: -1, flags: FD_CLOEXEC) = 6
syscall (/usr/lib64/libc-2.24.so)
perf_evsel__open (/home/acme/bin/perf)
0.190 ( 0.005 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 8
syscall (/usr/lib64/libc-2.24.so)
perf_evsel__open (/home/acme/bin/perf)
[ perf record: Captured and wrote 0.017 MB perf.data (7 samples) ]
#
The probe one called from perf_event_attr__set_max_precise_ip() works
the first time, with attr.precise_ip = 3, wit hthe next ones being the
per cpu ones for the cycles:ppp event.
And here is the text from a report and alternative proposed patch by
Thomas-Mich Richter:
---
On s390 the counter and sampling facility do not support a precise IP
skid level and sometimes returns EOPNOTSUPP when structure member
precise_ip in struct perf_event_attr is not set to zero.
On s390 commnd 'perf record -- true' fails with error EOPNOTSUPP. This
happens only when no events are specified on command line.
The functions called are
...
--> perf_evlist__add_default
--> perf_evsel__new_cycles
--> perf_event_attr__set_max_precise_ip
The last function determines the value of structure member precise_ip by
invoking the perf_event_open() system call and checking the return code.
The first successful open is the value for precise_ip.
However the value is determined without setting member sample_period and
indicates no sampling.
On s390 the counter facility and sampling facility are different. The
above procedure determines a precise_ip value of 3 using the counter
facility. Later it uses the sampling facility with a value of 3 and
fails with EOPNOTSUPP.
---
v2: Older compilers (e.g. gcc 4.4.7) don't support referencing members
of unnamed union members in the container struct initialization, so
move from:
struct perf_event_attr attr = {
...
.sample_period = 1,
};
to right after it as:
struct perf_event_attr attr = {
...
};
attr.sample_period = 1;
v3: We need to reset .sample_period to 0 to let the users of
perf_evsel__new_cycles() to properly setup attr.sample_period or
attr.sample_freq. Reported by Ingo Molnar.
Reported-and-Acked-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com>
Acked-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 18e7a45af9 ("perf/x86: Reject non sampling events with precise_ip")
Link: http://lkml.kernel.org/n/tip-yv6nnkl7tzqocrm0hl3x7vf1@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
I noticed that test_l4lb was failing in selftests:
# ./test_progs
test_pkt_access:PASS:ipv4 77 nsec
test_pkt_access:PASS:ipv6 44 nsec
test_xdp:PASS:ipv4 2933 nsec
test_xdp:PASS:ipv6 1500 nsec
test_l4lb:PASS:ipv4 377 nsec
test_l4lb:PASS:ipv6 544 nsec
test_l4lb:FAIL:stats 6297600000 200000
test_tcp_estats:PASS: 0 nsec
Summary: 7 PASSED, 1 FAILED
Tracking down the issue actually revealed that endianness selection
in bpf_endian.h is broken when compiled with clang with bpf target.
test_pkt_access.c, test_l4lb.c is compiled with __BYTE_ORDER as
__BIG_ENDIAN, test_xdp.c as __LITTLE_ENDIAN! test_l4lb noticeably
fails, because the test accounts bytes via bpf_ntohs(ip6h->payload_len)
and bpf_ntohs(iph->tot_len), and compares them against a defined
value and given a wrong endianness, the test outcome is different,
of course.
Turns out that there are actually two bugs: i) when we do __BYTE_ORDER
comparison with __LITTLE_ENDIAN/__BIG_ENDIAN, then depending on the
include order we see different outcomes. Reason is that __BYTE_ORDER
is undefined due to missing endian.h include. Before we include the
asm/byteorder.h (e.g. through linux/in.h), then __BYTE_ORDER equals
__LITTLE_ENDIAN since both are undefined, after the include which
correctly pulls in linux/byteorder/little_endian.h, __LITTLE_ENDIAN
is defined, but given __BYTE_ORDER is still undefined, we match on
__BYTE_ORDER equals to __BIG_ENDIAN since __BIG_ENDIAN is also
undefined at that point, sigh. ii) But even that would be wrong,
since when compiling the test cases with clang, one can select between
bpfeb and bpfel targets for cross compilation. Hence, we can also not
rely on what the system's endian.h provides, but we need to look at
the compiler's defined endianness. The compiler defines __BYTE_ORDER__,
and we can match __ORDER_LITTLE_ENDIAN__ and __ORDER_BIG_ENDIAN__,
which also reflects targets bpf (native), bpfel, bpfeb correctly,
thus really only rely on that. After patch:
# ./test_progs
test_pkt_access:PASS:ipv4 74 nsec
test_pkt_access:PASS:ipv6 42 nsec
test_xdp:PASS:ipv4 2340 nsec
test_xdp:PASS:ipv6 1461 nsec
test_l4lb:PASS:ipv4 400 nsec
test_l4lb:PASS:ipv6 530 nsec
test_tcp_estats:PASS: 0 nsec
Summary: 7 PASSED, 0 FAILED
Fixes: 43bcf707cc ("bpf: fix _htons occurences in test_progs")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
