]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
perf trace: Introduce --max-events
authorArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 18 Oct 2018 19:38:27 +0000 (16:38 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 19 Oct 2018 14:58:33 +0000 (11:58 -0300)
Allow stopping tracing after a number of events take place, considering
strace-like syscalls formatting as one event per enter/exit pair or when
in a multi-process tracing session a syscall is interrupted and printed
ending with '...'.

Examples included in the documentation:

Trace the first 4 open, openat or open_by_handle_at syscalls (in the future more syscalls may match here):

  $ perf trace -e open* --max-events 4
  [root@jouet perf]# trace -e open* --max-events 4
  2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
  2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
  3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
  4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
  $

Trace the first minor page fault when running a workload:

  # perf trace -F min --max-stack=7 --max-events 1 sleep 1
     0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
                                       __clear_user ([kernel.kallsyms])
                                       load_elf_binary ([kernel.kallsyms])
                                       search_binary_handler ([kernel.kallsyms])
                                       __do_execve_file.isra.33 ([kernel.kallsyms])
                                       __x64_sys_execve ([kernel.kallsyms])
                                       do_syscall_64 ([kernel.kallsyms])
                                       entry_SYSCALL_64 ([kernel.kallsyms])
  #

Trace the next min page page fault to take place on the first CPU:

  # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
     0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
                                       js::gc::FreeSpan::initAsEmpty (inlined)
                                       js::gc::Arena::setAsNotAllocated (inlined)
                                       js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
                                       js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
                                       js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
                                       js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
                                       js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
                                       js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
                                       js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
                                       JSThinInlineString::new_<(js::AllowGC)1> (inlined)
                                       AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
                                       js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
                                       [0x18b26e6bc2bd] (/tmp/perf-17136.map)

Tracing the next four ext4 operations on a specific CPU:

  # perf trace -e ext4:*/call-graph=fp/ --max-events 4 --cpu 3
     0.000 mutt/3849 ext4:ext4_es_lookup_extent_enter:dev 253,2 ino 57277 lblk 0
                                       ext4_es_lookup_extent ([kernel.kallsyms])
                                       read (/usr/lib64/libc-2.26.so)
     0.097 mutt/3849 ext4:ext4_es_lookup_extent_exit:dev 253,2 ino 57277 found 0 [0/0) 0
                                       ext4_es_lookup_extent ([kernel.kallsyms])
                                       read (/usr/lib64/libc-2.26.so)
     0.141 mutt/3849 ext4:ext4_ext_map_blocks_enter:dev 253,2 ino 57277 lblk 0 len 1 flags
                                       ext4_ext_map_blocks ([kernel.kallsyms])
                                       read (/usr/lib64/libc-2.26.so)
     0.184 mutt/3849 ext4:ext4_ext_load_extent:dev 253,2 ino 57277 lblk 1516511 pblk 18446744071750013657
                                       __read_extent_tree_block ([kernel.kallsyms])
                                       __read_extent_tree_block ([kernel.kallsyms])
                                       ext4_find_extent ([kernel.kallsyms])
                                       ext4_ext_map_blocks ([kernel.kallsyms])
                                       ext4_map_blocks ([kernel.kallsyms])
                                       ext4_mpage_readpages ([kernel.kallsyms])
                                       read_pages ([kernel.kallsyms])
                                       __do_page_cache_readahead ([kernel.kallsyms])
                                       ondemand_readahead ([kernel.kallsyms])
                                       generic_file_read_iter ([kernel.kallsyms])
                                       __vfs_read ([kernel.kallsyms])
                                       vfs_read ([kernel.kallsyms])
                                       ksys_read ([kernel.kallsyms])
                                       do_syscall_64 ([kernel.kallsyms])
                                       entry_SYSCALL_64 ([kernel.kallsyms])
                                       read (/usr/lib64/libc-2.26.so)
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Rudá Moura <ruda.moura@gmail.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-sweh107bs7ol5bzls0m4tqdz@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/Documentation/perf-trace.txt
tools/perf/builtin-trace.c

index 115db9e06ecd8a8d3f53059b8a8aa28c2337584a..0d1a1cd4d32817fec1014c9c7f75883565ac6860 100644 (file)
@@ -171,6 +171,11 @@ the thread executes on the designated CPUs. Default is to monitor all CPUs.
 --kernel-syscall-graph::
         Show the kernel callchains on the syscall exit path.
 
+--max-events=N::
+       Stop after processing N events. Note that strace-like events are considered
+       only at exit time or when a syscall is interrupted, i.e. in those cases this
+       option is equivalent to the number of lines printed.
+
 --max-stack::
         Set the stack depth limit when parsing the callchain, anything
         beyond the specified depth will be ignored. Note that at this point
@@ -238,6 +243,48 @@ Trace syscalls, major and minor pagefaults:
   As you can see, there was major pagefault in python process, from
   CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
 
+Trace the first 4 open, openat or open_by_handle_at syscalls (in the future more syscalls may match here):
+
+  $ perf trace -e open* --max-events 4
+  [root@jouet perf]# trace -e open* --max-events 4
+  2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
+  2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
+  3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
+  4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
+  $
+
+Trace the first minor page fault when running a workload:
+
+  # perf trace -F min --max-stack=7 --max-events 1 sleep 1
+     0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
+                                       __clear_user ([kernel.kallsyms])
+                                       load_elf_binary ([kernel.kallsyms])
+                                       search_binary_handler ([kernel.kallsyms])
+                                       __do_execve_file.isra.33 ([kernel.kallsyms])
+                                       __x64_sys_execve ([kernel.kallsyms])
+                                       do_syscall_64 ([kernel.kallsyms])
+                                       entry_SYSCALL_64 ([kernel.kallsyms])
+  #
+
+Trace the next min page page fault to take place on the first CPU:
+
+  # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
+     0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
+                                       js::gc::FreeSpan::initAsEmpty (inlined)
+                                       js::gc::Arena::setAsNotAllocated (inlined)
+                                       js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
+                                       js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
+                                       js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
+                                       js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
+                                       js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
+                                       js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
+                                       js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
+                                       JSThinInlineString::new_<(js::AllowGC)1> (inlined)
+                                       AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
+                                       js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
+                                       [0x18b26e6bc2bd] (/tmp/perf-17136.map)
+  #
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-script[1]
index 90289f31dd87c774ef882c24b1cdf55da9edcca3..74638034861ca168e1fadac5cb8951318d2629e5 100644 (file)
@@ -89,6 +89,8 @@ struct trace {
        u64                     base_time;
        FILE                    *output;
        unsigned long           nr_events;
+       unsigned long           nr_events_printed;
+       unsigned long           max_events;
        struct strlist          *ev_qualifier;
        struct {
                size_t          nr;
@@ -1664,6 +1666,8 @@ static int trace__printf_interrupted_entry(struct trace *trace)
        printed += fprintf(trace->output, "%-70s) ...\n", ttrace->entry_str);
        ttrace->entry_pending = false;
 
+       ++trace->nr_events_printed;
+
        return printed;
 }
 
@@ -1940,6 +1944,13 @@ errno_print: {
 
        fputc('\n', trace->output);
 
+       /*
+        * We only consider an 'event' for the sake of --max-events a non-filtered
+        * sys_enter + sys_exit and other tracepoint events.
+        */
+       if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
+               interrupted = true;
+
        if (callchain_ret > 0)
                trace__fprintf_callchain(trace, sample);
        else if (callchain_ret < 0)
@@ -2072,6 +2083,7 @@ static void bpf_output__fprintf(struct trace *trace,
 {
        binary__fprintf(sample->raw_data, sample->raw_size, 8,
                        bpf_output__printer, NULL, trace->output);
+       ++trace->nr_events_printed;
 }
 
 static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
@@ -2127,6 +2139,7 @@ static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
                        event_format__fprintf(evsel->tp_format, sample->cpu,
                                              sample->raw_data, sample->raw_size,
                                              trace->output);
+                       ++trace->nr_events_printed;
                }
        }
 
@@ -2225,6 +2238,8 @@ static int trace__pgfault(struct trace *trace,
                trace__fprintf_callchain(trace, sample);
        else if (callchain_ret < 0)
                pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
+
+       ++trace->nr_events_printed;
 out:
        err = 0;
 out_put:
@@ -2402,6 +2417,9 @@ static void trace__handle_event(struct trace *trace, union perf_event *event, st
                tracepoint_handler handler = evsel->handler;
                handler(trace, evsel, event, sample);
        }
+
+       if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
+               interrupted = true;
 }
 
 static int trace__add_syscall_newtp(struct trace *trace)
@@ -3249,6 +3267,7 @@ int cmd_trace(int argc, const char **argv)
                .trace_syscalls = false,
                .kernel_syscallchains = false,
                .max_stack = UINT_MAX,
+               .max_events = ULONG_MAX,
        };
        const char *output_name = NULL;
        const struct option trace_options[] = {
@@ -3301,6 +3320,8 @@ int cmd_trace(int argc, const char **argv)
                     &record_parse_callchain_opt),
        OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
                    "Show the kernel callchains on the syscall exit path"),
+       OPT_ULONG(0, "max-events", &trace.max_events,
+               "Set the maximum number of events to print, exit after that is reached. "),
        OPT_UINTEGER(0, "min-stack", &trace.min_stack,
                     "Set the minimum stack depth when parsing the callchain, "
                     "anything below the specified depth will be ignored."),