Some x86 microarchitectures fuse a subset of cmp/test/ALU instructions
with branch instructions, and thus perf annotate highlight such valid
pairs as fused.
When annotated with source, perf uses struct disasm_line to contain
either source or instruction line from objdump output. Usually, a C
statement generates multiple instructions which include such
cmp/test/ALU + branch instruction pairs. But in case of assembly
function, each individual assembly source line generate one
instruction.
The 'perf annotate' instruction fusion logic assumes the previous
disasm_line as the previous instruction line, which is wrong because,
for assembly function, previous disasm_line contains source line. And
thus perf fails to highlight valid fused instruction pairs for assembly
functions.
Fix it by searching backward until we find an instruction line and
consider that disasm_line as fused with current branch instruction.
Before:
│ cmpq %rcx, RIP+8(%rsp)
0.00 │ cmp %rcx,0x88(%rsp)
│ je .Lerror_bad_iret <--- Source line
0.14 │ ┌──je b4 <--- Instruction line
│ │movl %ecx, %eax
After:
│ cmpq %rcx, RIP+8(%rsp)
0.00 │ ┌──cmp %rcx,0x88(%rsp)
│ │je .Lerror_bad_iret
0.14 │ ├──je b4
│ │movl %ecx, %eax
Reviewed-by: Jin Yao <yao.jin@linux.intel.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@amd.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https //lore.kernel.org/r/20210911043854.8373-1-ravi.bangoria@amd.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If the disasm is empty, 's' should fail. Instead it seemingly works,
hiding the empty lines and causing an assertion error on the next time
annotate is called (from within perf report).
The problem is caused by a buffer overflow, caused by a wrong exit
condition in annotate_browser__find_next_asm_line, which checks
browser->b.top instead of browser->b.entries.
This patch fixes the issue, making annotate_browser__toggle_source
fail if the disasm is empty (nothing happens to the user).
Fixes: 6de249d66d ("perf annotate: Allow 's' on source code lines")
Signed-off-by: Riccardo Mancini <rickyman7@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Martin Liška <mliska@suse.cz>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20210705161524.72953-1-rickyman7@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To make the output more readable, I think it's better to remove 0's in
the output. Also the dummy event has no event stats so it just wasts
the space. Let's use the --skip-empty option to suppress it.
$ perf report --stat --skip-empty
Aggregated stats:
TOTAL events: 16530
MMAP events: 226
COMM events: 1596
EXIT events: 2
THROTTLE events: 121
UNTHROTTLE events: 117
FORK events: 1595
SAMPLE events: 719
MMAP2 events: 12147
CGROUP events: 2
FINISHED_ROUND events: 2
THREAD_MAP events: 1
CPU_MAP events: 1
TIME_CONV events: 1
cycles stats:
SAMPLE events: 719
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20210427013717.1651674-5-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Right now, when Line numbers are displayed, one can't easily find a
source file that the line corresponds to.
When a source line is selected and 'l' is pressed, full source file
location is displayed in perf UI footer line. The hotkey works only for
source code lines.
Signed-off-by: Martin Liška <mliska@suse.cz>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lore.kernel.org/lkml/25a6384f-d862-5dda-4fec-8f0555599c75@suse.cz
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Making perf with gcc-9.1.1 generates the following warning:
CC ui/browsers/hists.o
ui/browsers/hists.c: In function 'perf_evsel__hists_browse':
ui/browsers/hists.c:3078:61: error: '%d' directive output may be \
truncated writing between 1 and 11 bytes into a region of size \
between 2 and 12 [-Werror=format-truncation=]
3078 | "Max event group index to sort is %d (index from 0 to %d)",
| ^~
ui/browsers/hists.c:3078:7: note: directive argument in the range [-2147483648, 8]
3078 | "Max event group index to sort is %d (index from 0 to %d)",
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/include/stdio.h:937,
from ui/browsers/hists.c:5:
IOW, the string in line 3078 might be too long for buf[] of 64 bytes.
Fix this by increasing the size of buf[] to 128.
Fixes: dbddf17474 ("perf report/top TUI: Support hotkeys to let user select any event for sorting")
Signed-off-by: Song Liu <songliubraving@fb.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: stable@vger.kernel.org # v5.7+
Link: http://lore.kernel.org/lkml/20201030235431.534417-1-songliubraving@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
For a while we need to have a dummy event for doing things like
receiving PERF_RECORD_COMM, PERF_RECORD_EXEC, etc for threads being
created and dying while we synthesize the pre-existing ones at tool
start.
This 'dummy' event is needed for keeping track of thread lifetime events
early in the session but are uninteresting otherwise, i.e. no need to
have it in a initial events menu for the non-grouped case, i.e. for:
# perf top -e cycles,instructions
or even for plain:
# perf top
When 'cycles' and that 'dummy' event are in place.
The code to remove that 'dummy' event ended up creating an endless loop
for the grouped case, i.e.:
# perf top -e '{cycles,instructions}'
Fix it.
Fixes: bee9ca1c8a ("perf report TUI: Remove needless 'dummy' event from menu")
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>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To pick up fixes and move perf/core forward, minor conflict as
perf_evlist__add_dummy() lost its 'perf_' prefix as it operates on a
'struct evlist', not on a 'struct perf_evlist', i.e. its tools/perf/
specific, it is not in libperf.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fixing the common case of:
perf record
perf report
And getting just the cycles events.
We now have a 'dummy' event to get perf metadata events that take place
while we synthesize metadata records for pre-existing processes by
traversing procfs, so we always have this extra 'dummy' evsel, but we
don't have to offer it as there will be no samples on it, remove this
distraction.
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>
Link: https://lore.kernel.org/lkml/20200706115452.GA2772@redhat.com/
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
