[ Upstream commit 0e501a65d3 ]
The BTF func proto for a tracepoint has one more argument than the
actual tracepoint function since it has a context argument at the
begining. So it should compare to 5 when the tracepoint has 4
arguments.
typedef void (*btf_trace_sched_switch)(void *, bool, struct task_struct *, struct task_struct *, unsigned int);
Also, recent change in the perf tool would use a hand-written minimal
vmlinux.h to generate BTF in the skeleton. So it won't have the info
of the tracepoint. Anyway it should use the kernel's vmlinux BTF to
check the type in the kernel.
Fixes: b36888f71c ("perf record: Handle argument change in sched_switch")
Reviewed-by: Ian Rogers <irogers@google.com>
Acked-by: Song Liu <song@kernel.org>
Cc: Hao Luo <haoluo@google.com>
CC: bpf@vger.kernel.org
Link: https://lore.kernel.org/r/20230922234444.3115821-1-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f9cdeb58a9 ]
Dummy events are created with an attribute where the period and freq
are zero. evsel__config will then see the uninitialized values and
initialize them in evsel__default_freq_period. As fequency mode is
used by default the dummy event would be set to use frequency
mode. However, this has no effect on the dummy event but does cause
unnecessary timers/interrupts. Avoid this overhead by setting the
period to 1 for dummy events.
evlist__add_aux_dummy calls evlist__add_dummy then sets freq=0 and
period=1. This isn't necessary after this change and so the setting is
removed.
From Stephane:
The dummy event is not counting anything. It is used to collect mmap
records and avoid a race condition during the synthesize mmap phase of
perf record. As such, it should not cause any overhead during active
profiling. Yet, it did. Because of a bug the dummy event was
programmed as a sampling event in frequency mode. Events in that mode
incur more kernel overheads because on timer tick, the kernel has to
look at the number of samples for each event and potentially adjust
the sampling period to achieve the desired frequency. The dummy event
was therefore adding a frequency event to task and ctx contexts we may
otherwise not have any, e.g.,
perf record -a -e cpu/event=0x3c,period=10000000/.
On each timer tick the perf_adjust_freq_unthr_context() is invoked and
if ctx->nr_freq is non-zero, then the kernel will loop over ALL the
events of the context looking for frequency mode ones. In doing, so it
locks the context, and enable/disable the PMU of each hw event. If all
the events of the context are in period mode, the kernel will have to
traverse the list for nothing incurring overhead. The overhead is
multiplied by a very large factor when this happens in a guest kernel.
There is no need for the dummy event to be in frequency mode, it does
not count anything and therefore should not cause extra overhead for
no reason.
Fixes: 5bae025023 ("perf evlist: Introduce perf_evlist__new_dummy constructor")
Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Yang Jihong <yangjihong1@huawei.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20230916035640.1074422-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 9bf63282ea upstream.
The PERF_RECORD_ATTR is used for a pipe mode to describe an event with
attribute and IDs. The ID table comes after the attr and it calculate
size of the table using the total record size and the attr size.
n_ids = (total_record_size - end_of_the_attr_field) / sizeof(u64)
This is fine for most use cases, but sometimes it saves the pipe output
in a file and then process it later. And it becomes a problem if there
is a change in attr size between the record and report.
$ perf record -o- > perf-pipe.data # old version
$ perf report -i- < perf-pipe.data # new version
For example, if the attr size is 128 and it has 4 IDs, then it would
save them in 168 byte like below:
8 byte: perf event header { .type = PERF_RECORD_ATTR, .size = 168 },
128 byte: perf event attr { .size = 128, ... },
32 byte: event IDs [] = { 1234, 1235, 1236, 1237 },
But when report later, it thinks the attr size is 136 then it only read
the last 3 entries as ID.
8 byte: perf event header { .type = PERF_RECORD_ATTR, .size = 168 },
136 byte: perf event attr { .size = 136, ... },
24 byte: event IDs [] = { 1235, 1236, 1237 }, // 1234 is missing
So it should use the recorded version of the attr. The attr has the
size field already then it should honor the size when reading data.
Fixes: 2c46dbb517 ("perf: Convert perf header attrs into attr events")
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Zanussi <zanussi@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20230825152552.112913-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 979e9c9fc9 ]
In 616b14b47a ("perf build: Conditionally define NDEBUG") we
started using NDEBUG=1 when DEBUG=1 isn't present, so code that is
enclosed with assert() is not called.
In dd317df072 ("perf build: Make binutil libraries opt in") we
stopped linking against binutils-devel, for licensing reasons.
Recently people asked me why annotation of BPF programs wasn't working,
i.e. this:
$ perf annotate bpf_prog_5280546344e3f45c_kfree_skb
was returning:
case SYMBOL_ANNOTATE_ERRNO__NO_LIBOPCODES_FOR_BPF:
scnprintf(buf, buflen, "Please link with binutils's libopcode to enable BPF annotation");
This was on a fedora rpm, so its new enough that I had to try to test by
rebuilding using BUILD_NONDISTRO=1, only to get it segfaulting on me.
This combination made this libopcode function not to be called:
assert(bfd_check_format(bfdf, bfd_object));
Changing it to:
if (!bfd_check_format(bfdf, bfd_object))
abort();
Made it work, looking at this "check" function made me realize it
changes the 'bfdf' internal state, i.e. we better call it.
So stop using assert() on it, just call it and abort if it fails.
Probably it is better to propagate the error, etc, but it seems it is
unlikely to fail from the usage done so far and we really need to stop
using libopcodes, so do the quick fix above and move on.
With it we have BPF annotation back working when built with
BUILD_NONDISTRO=1:
⬢[acme@toolbox perf-tools-next]$ perf annotate --stdio2 bpf_prog_5280546344e3f45c_kfree_skb | head
No kallsyms or vmlinux with build-id 939bc71a1a51cdc434e60af93c7e734f7d5c0e7e was found
Samples: 12 of event 'cpu-clock:ppp', 4000 Hz, Event count (approx.): 3000000, [percent: local period]
bpf_prog_5280546344e3f45c_kfree_skb() bpf_prog_5280546344e3f45c_kfree_skb
Percent int kfree_skb(struct trace_event_raw_kfree_skb *args) {
nop
33.33 xchg %ax,%ax
push %rbp
mov %rsp,%rbp
sub $0x180,%rsp
push %rbx
push %r13
⬢[acme@toolbox perf-tools-next]$
Fixes: 6987561c9e ("perf annotate: Enable annotation of BPF programs")
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mohamed Mahmoud <mmahmoud@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Dave Tucker <datucker@redhat.com>
Cc: Derek Barbosa <debarbos@redhat.com>
Cc: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/lkml/ZMrMzoQBe0yqMek1@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 1c24956542 upstream.
This problem was encountered on an arm64 system with a lot of memory.
Without kernel debug symbols installed, and with both kcore and kallsyms
available, perf managed to get confused and returned "unknown" for all
of the kernel symbols that it tried to look up.
On this system, stext fell within the vmalloc segment. The kcore symbol
matching code tries to find the first segment that contains stext and
uses that to replace the segment generated from just the kallsyms
information. In this case, however, there were two: a very large
vmalloc segment, and the text segment. This caused perf to get confused
because multiple overlapping segments were inserted into the RB tree
that holds the discovered segments. However, that alone wasn't
sufficient to cause the problem. Even when we could find the segment,
the offsets were adjusted in such a way that the newly generated symbols
didn't line up with the instruction addresses in the trace. The most
obvious solution would be to consult which segment type is text from
kcore, but this information is not exposed to users.
Instead, select the smallest matching segment that contains stext
instead of the first matching segment. This allows us to match the text
segment instead of vmalloc, if one is contained within the other.
Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Reaver <me@davidreaver.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20230125183418.GD1963@templeofstupid.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit ecc68ee216 ]
It seems that perf stat -b <prog id> doesn't produce any results:
$ perf stat -e cycles -b 4 -I 10000 -vvv
Control descriptor is not initialized
cycles: 0 0 0
time counts unit events
10.007641640 <not supported> cycles
Looks like this happens because fentry/fexit progs are getting loaded, but the
corresponding perf event is not enabled and not added into the events bpf map.
I think there is some mixing up between two type of bpf support, one for bperf
and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which
perf events are enabled, but for the latter (bpf_profiler) a perf event is
required. Using evsel__is_bperf to check only bperf produces expected results:
$ perf stat -e cycles -b 4 -I 10000 -vvv
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
size 136
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 3
------------------------------------------------------------
[...perf_event_attr for other CPUs...]
------------------------------------------------------------
cycles: 309426 169009 169009
time counts unit events
10.010091271 309426 cycles
The final numbers correspond (at least in the level of magnitude) to the
same metric obtained via bpftool.
Fixes: 112cb56164 ("perf stat: Introduce config stat.bpf-counter-events")
Reviewed-by: Song Liu <song@kernel.org>
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Tested-by: Song Liu <song@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230412182316.11628-1-9erthalion6@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 57f14b5ae1 ]
An audit showed just this one problem with zfree(), fix it.
Fixes: 9fbc61f832 ("perf pmu: Add support for PMU capabilities")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 25f69c69bc ]
When creating counters with initial delay configured, the enable_on_exec
field is not set. So we need to enable the counters later. The problem
is, when a workload is specified the target__none() is true. So we also
need to check stat_config.initial_delay.
In this change, we add a new field 'initial_delay' for struct target
which could be shared by other subcommands. And define
target__enable_on_exec() which returns whether enable_on_exec should be
set on normal cases.
Before this fix the event is not counted:
$ ./perf stat -e instructions -D 100 sleep 2
Events disabled
Events enabled
Performance counter stats for 'sleep 2':
<not counted> instructions
1.901661124 seconds time elapsed
0.001602000 seconds user
0.000000000 seconds sys
After fix it works:
$ ./perf stat -e instructions -D 100 sleep 2
Events disabled
Events enabled
Performance counter stats for 'sleep 2':
404,214 instructions
1.901743475 seconds time elapsed
0.001617000 seconds user
0.000000000 seconds sys
Fixes: c587e77e10 ("perf stat: Do not delay the workload with --delay")
Signed-off-by: Changbin Du <changbin.du@huawei.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Hui Wang <hw.huiwang@huawei.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20230302031146.2801588-2-changbin.du@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit aeb802f872 ]
When it processes AUXTRACE_INFO, it calls to auxtrace_queue_data() to
collect AUXTRACE data first. That won't work with pipe since it needs
lseek() to read the scattered aux data.
$ perf record -o- -e intel_pt// true | perf report -i- --itrace=i100
# To display the perf.data header info, please use --header/--header-only options.
#
0x4118 [0xa0]: failed to process type: 70
Error:
failed to process sample
For the pipe mode, it can handle the aux data as it gets. But there's
no guarantee it can get the aux data in time. So the following warning
will be shown at the beginning:
WARNING: Intel PT with pipe mode is not recommended.
The output cannot relied upon. In particular,
time stamps and the order of events may be incorrect.
Fixes: dbd134322e ("perf intel-pt: Add support for decoding AUX area samples")
Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: James Clark <james.clark@arm.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: https://lore.kernel.org/r/20230131023350.1903992-3-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit cf129830ee upstream.
When a match has been made to the nth duplicate symbol, return
success not error.
Example:
Before:
$ cat file.c
cat: file.c: No such file or directory
$ cat file1.c
#include <stdio.h>
static void func(void)
{
printf("First func\n");
}
void other(void);
int main()
{
func();
other();
return 0;
}
$ cat file2.c
#include <stdio.h>
static void func(void)
{
printf("Second func\n");
}
void other(void)
{
func();
}
$ gcc -Wall -Wextra -o test file1.c file2.c
$ perf record -e intel_pt//u --filter 'filter func @ ./test' -- ./test
Multiple symbols with name 'func'
#1 0x1149 l func
which is near main
#2 0x1179 l func
which is near other
Disambiguate symbol name by inserting #n after the name e.g. func #2
Or select a global symbol by inserting #0 or #g or #G
Failed to parse address filter: 'filter func @ ./test'
Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]
Where multiple filters are separated by space or comma.
$ perf record -e intel_pt//u --filter 'filter func #2 @ ./test' -- ./test
Failed to parse address filter: 'filter func #2 @ ./test'
Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]
Where multiple filters are separated by space or comma.
After:
$ perf record -e intel_pt//u --filter 'filter func #2 @ ./test' -- ./test
First func
Second func
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.016 MB perf.data ]
$ perf script --itrace=b -Ftime,flags,ip,sym,addr --ns
1231062.526977619: tr strt 0 [unknown] => 558495708179 func
1231062.526977619: tr end call 558495708188 func => 558495708050 _init
1231062.526979286: tr strt 0 [unknown] => 55849570818d func
1231062.526979286: tr end return 55849570818f func => 55849570819d other
Fixes: 1b36c03e35 ("perf record: Add support for using symbols in address filters")
Reported-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Dmitry Dolgov <9erthalion6@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20230110185659.15979-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 54b353a20c ]
The --for-each-cgroup can have the same cgroup multiple times, but this
confuses BPF counters (since they have the same cgroup id), making only
the last cgroup events to be counted.
Let's check the cgroup name before adding a new entry to the cgroups
list.
Before:
$ sudo ./perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Performance counter stats for 'system wide':
<not counted> msec cpu-clock /
<not counted> context-switches /
<not counted> cpu-migrations /
<not counted> page-faults /
<not counted> cycles /
<not counted> instructions /
<not counted> branches /
<not counted> branch-misses /
8,016.04 msec cpu-clock / # 7.998 CPUs utilized
6,152 context-switches / # 767.461 /sec
250 cpu-migrations / # 31.187 /sec
442 page-faults / # 55.139 /sec
613,111,487 cycles / # 0.076 GHz
280,599,604 instructions / # 0.46 insn per cycle
57,692,724 branches / # 7.197 M/sec
3,385,168 branch-misses / # 5.87% of all branches
1.002220125 seconds time elapsed
After it becomes similar to the non-BPF mode:
$ sudo ./perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Performance counter stats for 'system wide':
8,013.38 msec cpu-clock / # 7.998 CPUs utilized
6,859 context-switches / # 855.944 /sec
334 cpu-migrations / # 41.680 /sec
345 page-faults / # 43.053 /sec
782,326,119 cycles / # 0.098 GHz
471,645,724 instructions / # 0.60 insn per cycle
94,963,430 branches / # 11.851 M/sec
3,685,511 branch-misses / # 3.88% of all branches
1.001864539 seconds time elapsed
Committer notes:
As a reminder, to test with BPF counters one has to use BUILD_BPF_SKEL=1
in the make command line and have clang/llvm installed when building
perf, otherwise the --bpf-counters option will not be available:
# perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Error: unknown option `bpf-counters'
Usage: perf stat [<options>] [<command>]
-a, --all-cpus system-wide collection from all CPUs
<SNIP>
#
Fixes: bb1c15b60b ("perf stat: Support regex pattern in --for-each-cgroup")
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: bpf@vger.kernel.org
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/r/20230104064402.1551516-5-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2d656b0f81 ]
When --for-each-cgroup option is used, it fails when any of events is
not supported and exits immediately. This is not how 'perf stat'
handles unsupported events.
Let's ignore the failure and proceed with others so that the output is
similar to when BPF counters are not used:
Before:
$ sudo ./perf stat -a --bpf-counters -e L1-icache-loads,L1-dcache-loads --for-each-cgroup system.slice,user.slice sleep 1
Failed to open first cgroup events
$
After it shows output similat to when --bpf-counters isn't specified:
$ sudo ./perf stat -a --bpf-counters -e L1-icache-loads,L1-dcache-loads --for-each-cgroup system.slice,user.slice sleep 1
Performance counter stats for 'system wide':
<not supported> L1-icache-loads system.slice
29,892,418 L1-dcache-loads system.slice
<not supported> L1-icache-loads user.slice
52,497,220 L1-dcache-loads user.slice
$
Fixes: 944138f048 ("perf stat: Enable BPF counter with --for-each-cgroup")
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/r/20230104064402.1551516-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 6f520ce179 ]
perf doesn't provide proper symbol information for specially crafted
.debug files.
Sometimes .debug file may not have similar program header as runtime
ELF file. For example if we generate .debug file using objcopy
--only-keep-debug resulting file will not contain .text, .data and
other runtime sections. That means corresponding program headers will
have zero FileSiz and modified Offset.
Example: program header of text section of libxxx.so:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x00000000003d3000 0x00000000003d3000 0x00000000003d3000
0x000000000055ae80 0x000000000055ae80 R E 0x1000
Same program header after executing:
objcopy --only-keep-debug libxxx.so libxxx.so.debug
LOAD 0x0000000000001000 0x00000000003d3000 0x00000000003d3000
0x0000000000000000 0x000000000055ae80 R E 0x1000
Offset and FileSiz have been changed.
Following formula will not provide correct value, if program header
taken from .debug file (syms_ss):
sym.st_value -= phdr.p_vaddr - phdr.p_offset;
Correct program header information is located inside runtime ELF
file (runtime_ss).
Fixes: 2d86612aac ("perf symbol: Correct address for bss symbols")
Signed-off-by: Ajay Kaher <akaher@vmware.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Makhalov <amakhalov@vmware.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srivatsa S. Bhat <srivatsab@vmware.com>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Cc: Vasavi Sirnapalli <vsirnapalli@vmware.com>
Link: http://lore.kernel.org/lkml/1669198696-50547-1-git-send-email-akaher@vmware.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit f4a2aade68 ("perf tests powerpc: Fix branch stack sampling
test to include sanity check for branch filter") added a skip if certain
branch options aren't available.
But the change added both -b (--branch-any) and --branch-filter options
at the same time, which will always result in a failure on any platform
because the arguments can't be used together.
Fix this by removing -b (--branch-any) and leaving --branch-filter which
already specifies 'any'. Also add warning messages to the test and perf
tool.
Output on x86 before this fix:
$ sudo ./perf test branch
108: Check branch stack sampling : Skip
After:
$ sudo ./perf test branch
108: Check branch stack sampling : Ok
Fixes: f4a2aade68 ("perf tests powerpc: Fix branch stack sampling test to include sanity check for branch filter")
Signed-off-by: James Clark <james.clark@arm.com>
Tested-by: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anshuman.Khandual@arm.com
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kajol Jain <kjain@linux.ibm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20221028121913.745307-1-james.clark@arm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
'perf stat' with CSV output option prints an extra empty string as first
field in metrics output line. Sample output below:
# ./perf stat -x, --per-socket -a -C 1 ls
S0,1,1.78,msec,cpu-clock,1785146,100.00,0.973,CPUs utilized
S0,1,26,,context-switches,1781750,100.00,0.015,M/sec
S0,1,1,,cpu-migrations,1780526,100.00,0.561,K/sec
S0,1,1,,page-faults,1779060,100.00,0.561,K/sec
S0,1,875807,,cycles,1769826,100.00,0.491,GHz
S0,1,85281,,stalled-cycles-frontend,1767512,100.00,9.74,frontend cycles idle
S0,1,576839,,stalled-cycles-backend,1766260,100.00,65.86,backend cycles idle
S0,1,288430,,instructions,1762246,100.00,0.33,insn per cycle
====> ,S0,1,,,,,,,2.00,stalled cycles per insn
The above command line uses field separator as "," via "-x," option and
per-socket option displays socket value as first field. But here the
last line for "stalled cycles per insn" has "," in the beginning.
Sample output using interval mode:
# ./perf stat -I 1000 -x, --per-socket -a -C 1 ls
0.001813453,S0,1,1.87,msec,cpu-clock,1872052,100.00,0.002,CPUs utilized
0.001813453,S0,1,2,,context-switches,1868028,100.00,1.070,K/sec
------
0.001813453,S0,1,85379,,instructions,1856754,100.00,0.32,insn per cycle
====> 0.001813453,,S0,1,,,,,,,1.34,stalled cycles per insn
Above result also has an extra CSV separator after
the timestamp. Patch addresses extra field separator
in the beginning of the metric output line.
The counter stats are displayed by function
"perf_stat__print_shadow_stats" in code
"util/stat-shadow.c". While printing the stats info
for "stalled cycles per insn", function "new_line_csv"
is used as new_line callback.
The new_line_csv function has check for "os->prefix"
and if prefix is not null, it will be printed along
with cvs separator.
Snippet from "new_line_csv":
if (os->prefix)
fprintf(os->fh, "%s%s", os->prefix, config->csv_sep);
Here os->prefix gets printed followed by ","
which is the cvs separator. The os->prefix is
used in interval mode option ( -I ), to print
time stamp on every new line. But prefix is
already set to contain CSV separator when used
in interval mode for CSV option.
Reference: Function "static void print_interval"
Snippet:
sprintf(prefix, "%6lu.%09lu%s", ts->tv_sec, ts->tv_nsec, config->csv_sep);
Also if prefix is not assigned (if not used with
-I option), it gets set to empty string.
Reference: function printout() in util/stat-display.c
Snippet:
.prefix = prefix ? prefix : "",
Since prefix already set to contain cvs_sep in interval
option, patch removes printing config->csv_sep in
new_line_csv function to avoid printing extra field.
After the patch:
# ./perf stat -x, --per-socket -a -C 1 ls
S0,1,2.04,msec,cpu-clock,2045202,100.00,1.013,CPUs utilized
S0,1,2,,context-switches,2041444,100.00,979.289,/sec
S0,1,0,,cpu-migrations,2040820,100.00,0.000,/sec
S0,1,2,,page-faults,2040288,100.00,979.289,/sec
S0,1,254589,,cycles,2036066,100.00,0.125,GHz
S0,1,82481,,stalled-cycles-frontend,2032420,100.00,32.40,frontend cycles idle
S0,1,113170,,stalled-cycles-backend,2031722,100.00,44.45,backend cycles idle
S0,1,88766,,instructions,2030942,100.00,0.35,insn per cycle
S0,1,,,,,,,1.27,stalled cycles per insn
Fixes: 92a61f6412 ("perf stat: Implement CSV metrics output")
Reported-by: Disha Goel <disgoel@linux.vnet.ibm.com>
Reviewed-By: Kajol Jain <kjain@linux.ibm.com>
Signed-off-by: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Tested-by: Disha Goel <disgoel@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: James Clark <james.clark@arm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nageswara R Sastry <rnsastry@linux.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20221018085605.63834-1-atrajeev@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We also need to add SYM_TYPED_FUNC_START() to util/include/linux/linkage.h
and update tools/perf/check_headers.sh to ignore the include cfi_types.h
line when checking if the kernel original files drifted from the copies
we carry.
This is to get the changes from:
ccace936ee ("x86: Add types to indirectly called assembly functions")
Addressing these tools/perf build warnings:
Warning: Kernel ABI header at 'tools/arch/x86/lib/memcpy_64.S' differs from latest version at 'arch/x86/lib/memcpy_64.S'
diff -u tools/arch/x86/lib/memcpy_64.S arch/x86/lib/memcpy_64.S
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/lkml/Y1f3VRIec9EBgX6F@kernel.org/
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
During the transition to libbpf 1.0 some functions that perf used were
deprecated and finally removed from libbpf, so bpf_program__set_insns()
was introduced for perf to continue to use its bpf loader.
But when build with LIBBPF_DYNAMIC=1 we now need to check if that
function is available so that perf can build with older libbpf versions,
even if the end result is emitting a warning to the user that the use
of the perf BPF loader requires a newer libbpf, since bpf_program__set_insns()
touches libbpf objects internal state.
This affects only 'perf trace' when using bpf C code or pre-compiled
bytecode as an event.
Noticed on RHEL9, that has libbpf 0.7.0, where bpf_program__set_insns()
isn't available.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The bpf_load_program() prototype appeared in tools/lib/bpf/bpf.h as
deprecated, but nowadays its completely removed, so add it back for
building with the system libbpf when using 'make LIBBPF_DYNAMIC=1'.
This is a stop gap hack till we do like tools/bpf does with bpftool,
i.e. bootstrap the libbpf build and install it in the perf build
directory when not using 'make LIBBPF_DYNAMIC=1'.
That has to be done to all libraries in tools/lib/, so tha we can
remove -Itools/lib/ from the tools/perf CFLAGS.
Noticed when building with LIBBPF_DYNAMIC=1 and libbpf 0.7.0 on RHEL9.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
