- 22 Feb, 2016 13 commits
-
-
Wang Nan authored
This patch introduces a new syntax to perf event parser: # perf record -e './test_bpf_map_3.c/map:channel.value[0,1,2,3...5]=101/' usleep 2 By utilizing the basic facilities in bpf-loader.c which allow setting different slots in a BPF map separately, the newly introduced syntax allows perf to control specific elements in a BPF map. Test result: # cat ./test_bpf_map_3.c /************************ BEGIN **************************/ #include <uapi/linux/bpf.h> #define SEC(NAME) __attribute__((section(NAME), used)) struct bpf_map_def { unsigned int type; unsigned int key_size; unsigned int value_size; unsigned int max_entries; }; static void *(*map_lookup_elem)(struct bpf_map_def *, void *) = (void *)BPF_FUNC_map_lookup_elem; static int (*trace_printk)(const char *fmt, int fmt_size, ...) = (void *)BPF_FUNC_trace_printk; struct bpf_map_def SEC("maps") channel = { .type = BPF_MAP_TYPE_ARRAY, .key_size = sizeof(int), .value_size = sizeof(unsigned char), .max_entries = 100, }; SEC("func=hrtimer_nanosleep rqtp->tv_nsec") int func(void *ctx, int err, long nsec) { char fmt[] = "%ld\n"; long usec = nsec * 0x10624dd3 >> 38; // nsec / 1000 int key = (int)usec; unsigned char *pval = map_lookup_elem(&channel, &key); if (!pval) return 0; trace_printk(fmt, sizeof(fmt), (unsigned char)*pval); return 0; } char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /************************* END ***************************/ Normal case: # echo "" > /sys/kernel/debug/tracing/trace # ./perf record -e './test_bpf_map_3.c/map:channel.value[0,1,2,3...5]=101/' usleep 2 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] # cat /sys/kernel/debug/tracing/trace | grep usleep usleep-405 [004] d... 2745423.547822: : 101 # ./perf record -e './test_bpf_map_3.c/map:channel.value[0...9,20...29]=102,map:channel.value[10...19]=103/' usleep 3 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] # ./perf record -e './test_bpf_map_3.c/map:channel.value[0...9,20...29]=102,map:channel.value[10...19]=103/' usleep 15 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] # cat /sys/kernel/debug/tracing/trace | grep usleep usleep-405 [004] d... 2745423.547822: : 101 usleep-655 [006] d... 2745434.122814: : 102 usleep-904 [006] d... 2745439.916264: : 103 # ./perf record -e './test_bpf_map_3.c/map:channel.value[all]=104/' usleep 99 # cat /sys/kernel/debug/tracing/trace | grep usleep usleep-405 [004] d... 2745423.547822: : 101 usleep-655 [006] d... 2745434.122814: : 102 usleep-904 [006] d... 2745439.916264: : 103 usleep-1537 [003] d... 2745538.053737: : 104 Error case: # ./perf record -e './test_bpf_map_3.c/map:channel.value[10...1000]=104/' usleep 99 event syntax error: '..annel.value[10...1000]=104/' \___ Index too large Hint: Valid config terms: map:[<arraymap>].value<indices>=[value] map:[<eventmap>].event<indices>=[event] where <indices> is something like [0,3...5] or [all] (add -v to see detail) Run 'perf list' for a list of valid events Usage: perf record [<options>] [<command>] or: perf record [<options>] -- <command> [<options>] -e, --event <event> event selector. use 'perf list' to list available events Signed-off-by:Wang Nan <wangnan0@huawei.com> Acked-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Cody P Schafer <dev@codyps.com> Cc: He Kuang <hekuang@huawei.com> Cc: Jeremie Galarneau <jeremie.galarneau@efficios.com> Cc: Kirill Smelkov <kirr@nexedi.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1456132275-98875-9-git-send-email-wangnan0@huawei.comSigned-off-by:
-
Wang Nan authored
This patch introduces basic facilities to support config different slots in a BPF map one by one. array.nr_ranges and array.ranges are introduced into 'struct parse_events_term', where ranges is an array of indices range (start, length) which will be configured by this config term. nr_ranges is the size of the array. The array is passed to 'struct bpf_map_priv'. To indicate the new type of configuration, BPF_MAP_KEY_RANGES is added as a new key type. bpf_map_config_foreach_key() is extended to iterate over those indices instead of all possible keys. Code in this commit will be enabled by following commit which enables the indices syntax for array configuration. Signed-off-by:
-
Wang Nan authored
A new syntax is added to the parser so that the user can access predefined perf events in BPF objects. After this patch, BPF programs for perf are finally able to utilize bpf_perf_event_read() introduced in commit 35578d79 ("bpf: Implement function bpf_perf_event_read() that get the selected hardware PMU counter"). Test result: # cat test_bpf_map_2.c /************************ BEGIN **************************/ #include <uapi/linux/bpf.h> #define SEC(NAME) __attribute__((section(NAME), used)) struct bpf_map_def { unsigned int type; unsigned int key_size; unsigned int value_size; unsigned int max_entries; }; static int (*trace_printk)(const char *fmt, int fmt_size, ...) = (void *)BPF_FUNC_trace_printk; static int (*get_smp_processor_id)(void) = (void *)BPF_FUNC_get_smp_processor_id; static int (*perf_event_read)(struct bpf_map_def *, int) = (void *)BPF_FUNC_perf_event_read; struct bpf_map_def SEC("maps") pmu_map = { .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY, .key_size = sizeof(int), .value_size = sizeof(int), .max_entries = __NR_CPUS__, }; SEC("func_write=sys_write") int func_write(void *ctx) { unsigned long long val; char fmt[] = "sys_write: pmu=%llu\n"; val = perf_event_read(&pmu_map, get_smp_processor_id()); trace_printk(fmt, sizeof(fmt), val); return 0; } SEC("func_write_return=sys_write%return") int func_write_return(void *ctx) { unsigned long long val = 0; char fmt[] = "sys_write_return: pmu=%llu\n"; val = perf_event_read(&pmu_map, get_smp_processor_id()); trace_printk(fmt, sizeof(fmt), val); return 0; } char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /************************* END ***************************/ Normal case: # echo "" > /sys/kernel/debug/tracing/trace # perf record -i -e cycles -e './test_bpf_map_2.c/map:pmu_map.event=cycles/' ls / [SNIP] [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.013 MB perf.data (7 samples) ] # cat /sys/kernel/debug/tracing/trace | grep ls ls-17066 [000] d... 938449.863301: : sys_write: pmu=1157327 ls-17066 [000] dN.. 938449.863342: : sys_write_return: pmu=1225218 ls-17066 [000] d... 938449.863349: : sys_write: pmu=1241922 ls-17066 [000] dN.. 938449.863369: : sys_write_return: pmu=1267445 Normal case (system wide): # echo "" > /sys/kernel/debug/tracing/trace # perf record -i -e cycles -e './test_bpf_map_2.c/map:pmu_map.event=cycles/' -a ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.811 MB perf.data (120 samples) ] # cat /sys/kernel/debug/tracing/trace | grep -v '18446744073709551594' | grep -v perf | head -n 20 [SNIP] # TASK-PID CPU# |||| TIMESTAMP FUNCTION # | | | |||| | | gmain-30828 [002] d... 2740551.068992: : sys_write: pmu=84373 gmain-30828 [002] d... 2740551.068992: : sys_write_return: pmu=87696 gmain-30828 [002] d... 2740551.068996: : sys_write: pmu=100658 gmain-30828 [002] d... 2740551.068997: : sys_write_return: pmu=102572 Error case 1: # perf record -e './test_bpf_map_2.c' ls / [SNIP] [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.014 MB perf.data ] # cat /sys/kernel/debug/tracing/trace | grep ls ls-17115 [007] d... 2724279.665625: : sys_write: pmu=18446744073709551614 ls-17115 [007] dN.. 2724279.665651: : sys_write_return: pmu=18446744073709551614 ls-17115 [007] d... 2724279.665658: : sys_write: pmu=18446744073709551614 ls-17115 [007] dN.. 2724279.665677: : sys_write_return: pmu=18446744073709551614 (18446744073709551614 is 0xfffffffffffffffe (-2)) Error case 2: # perf record -e cycles -e './test_bpf_map_2.c/map:pmu_map.event=evt/' -a event syntax error: '..ps:pmu_map.event=evt/' \___ Event not found for map setting Hint: Valid config terms: map:[<arraymap>].value=[value] map:[<eventmap>].event=[event] [SNIP] Error case 3: # ls /proc/2348/task/ 2348 2505 2506 2507 2508 # perf record -i -e cycles -e './test_bpf_map_2.c/map:pmu_map.event=cycles/' -p 2348 ERROR: Apply config to BPF failed: Cannot set event to BPF map in multi-thread tracing Error case 4: # perf record -e cycles -e './test_bpf_map_2.c/map:pmu_map.event=cycles/' ls / ERROR: Apply config to BPF failed: Doesn't support inherit event (Hint: use -i to turn off inherit) Error case 5: # perf record -i -e raw_syscalls:sys_enter -e './test_bpf_map_2.c/map:pmu_map.event=raw_syscalls:sys_enter/' ls ERROR: Apply config to BPF failed: Can only put raw, hardware and BPF output event into a BPF map Error case 6: # perf record -i -e './test_bpf_map_2.c/map:pmu_map.event=123/' ls / event syntax error: '.._map.event=123/' \___ Incorrect value type for map [SNIP] Signed-off-by:
He Kuang <hekuang@huawei.com> Signed-off-by:
-
Wang Nan authored
bpf__apply_obj_config() is introduced as the core API to apply object config options to all BPF objects. This patch also does the real work for setting values for BPF_MAP_TYPE_PERF_ARRAY maps by inserting value stored in map's private field into the BPF map. This patch is required because we are not always able to set all BPF config during parsing. Further patch will set events created by perf to BPF_MAP_TYPE_PERF_EVENT_ARRAY maps, which is not exist until perf_evsel__open(). bpf_map_foreach_key() is introduced to iterate over each key needs to be configured. This function would be extended to support more map types and different key settings. In perf record, before start recording, call bpf__apply_config() to turn on all BPF config options. Test result: # cat ./test_bpf_map_1.c /************************ BEGIN **************************/ #include <uapi/linux/bpf.h> #define SEC(NAME) __attribute__((section(NAME), used)) struct bpf_map_def { unsigned int type; unsigned int key_size; unsigned int value_size; unsigned int max_entries; }; static void *(*map_lookup_elem)(struct bpf_map_def *, void *) = (void *)BPF_FUNC_map_lookup_elem; static int (*trace_printk)(const char *fmt, int fmt_size, ...) = (void *)BPF_FUNC_trace_printk; struct bpf_map_def SEC("maps") channel = { .type = BPF_MAP_TYPE_ARRAY, .key_size = sizeof(int), .value_size = sizeof(int), .max_entries = 1, }; SEC("func=sys_nanosleep") int func(void *ctx) { int key = 0; char fmt[] = "%d\n"; int *pval = map_lookup_elem(&channel, &key); if (!pval) return 0; trace_printk(fmt, sizeof(fmt), *pval); return 0; } char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /************************* END ***************************/ # echo "" > /sys/kernel/debug/tracing/trace # ./perf record -e './test_bpf_map_1.c/map:channel.value=11/' usleep 10 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] # cat /sys/kernel/debug/tracing/trace # tracer: nop # # entries-in-buffer/entries-written: 1/1 #P:8 [SNIP] # TASK-PID CPU# |||| TIMESTAMP FUNCTION # | | | |||| | | usleep-18593 [007] d... 2394714.395539: : 11 # ./perf record -e './test_bpf_map_1.c/map:channel.value=101/' usleep 10 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] # cat /sys/kernel/debug/tracing/trace # tracer: nop # # entries-in-buffer/entries-written: 1/1 #P:8 [SNIP] # TASK-PID CPU# |||| TIMESTAMP FUNCTION # | | | |||| | | usleep-18593 [007] d... 2394714.395539: : 11 usleep-19000 [006] d... 2394831.057840: : 101 Signed-off-by: -
Wang Nan authored
This patch adds the final step for BPF map configuration. A new syntax is appended into parser so user can config BPF objects through '/' '/' enclosed config terms. After this patch, following syntax is available: # perf record -e ./test_bpf_map_1.c/map:channel.value=10/ ... It would takes effect after appling following commits. Test result: # cat ./test_bpf_map_1.c /************************ BEGIN **************************/ #include <uapi/linux/bpf.h> #define SEC(NAME) __attribute__((section(NAME), used)) struct bpf_map_def { unsigned int type; unsigned int key_size; unsigned int value_size; unsigned int max_entries; }; static void *(*map_lookup_elem)(struct bpf_map_def *, void *) = (void *)BPF_FUNC_map_lookup_elem; static int (*trace_printk)(const char *fmt, int fmt_size, ...) = (void *)BPF_FUNC_trace_printk; struct bpf_map_def SEC("maps") channel = { .type = BPF_MAP_TYPE_ARRAY, .key_size = sizeof(int), .value_size = sizeof(int), .max_entries = 1, }; SEC("func=sys_nanosleep") int func(void *ctx) { int key = 0; char fmt[] = "%d\n"; int *pval = map_lookup_elem(&channel, &key); if (!pval) return 0; trace_printk(fmt, sizeof(fmt), *pval); return 0; } char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /************************* END ***************************/ - Normal case: # ./perf record -e './test_bpf_map_1.c/map:channel.value=10/' usleep 10 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data ] - Error case: # ./perf record -e './test_bpf_map_1.c/map:channel.value/' usleep 10 event syntax error: '..ps:channel:value/' \___ Config value not set (missing '=') Hint: Valid config term: map:[<arraymap>]:value=[value] (add -v to see detail) Run 'perf list' for a list of valid events Usage: perf record [<options>] [<command>] or: perf record [<options>] -- <command> [<options>] -e, --event <event> event selector. use 'perf list' to list available events # ./perf record -e './test_bpf_map_1.c/xmap:channel.value=10/' usleep 10 event syntax error: '..pf_map_1.c/xmap:channel.value=10/' \___ Invalid object config option [SNIP] # ./perf record -e './test_bpf_map_1.c/map:xchannel.value=10/' usleep 10 event syntax error: '..p_1.c/map:xchannel.value=10/' \___ Target map not exist [SNIP] # ./perf record -e './test_bpf_map_1.c/map:channel.xvalue=10/' usleep 10 event syntax error: '..ps:channel.xvalue=10/' \___ Invalid object map config option [SNIP] # ./perf record -e './test_bpf_map_1.c/map:channel.value=x10/' usleep 10 event syntax error: '..nnel.value=x10/' \___ Incorrect value type for map [SNIP] Change BPF_MAP_TYPE_ARRAY to '1' in test_bpf_map_1.c: # ./perf record -e './test_bpf_map_1.c/map:channel.value=10/' usleep 10 event syntax error: '..ps:channel.value=10/' \___ Can't use this config term to this type of map Hint: Valid config term: map:[<arraymap>].value=[value] (add -v to see detail) Signed-off-by: -
Wang Nan authored
bpf__config_obj() is introduced as a core API to config BPF object after loading. One configuration option of maps is introduced. After this patch BPF object can accept assignments like: map:my_map.value=1234 (map.my_map.value looks pretty. However, there's a small but hard to fix problem related to flex's greedy matching. Please see [1]. Choose ':' to avoid it in a simpler way.) This patch is more complex than the work it does because the consideration of extension. In designing BPF map configuration, the following things should be considered: 1. Array indices selection: perf should allow user setting different value for different slots in an array, with syntax like: map:my_map.value[0,3...6]=1234; 2. A map should be set by different config terms, each for a part of it. For example, set each slot to the pid of a thread; 3. Type of value: integer is not the only valid value type. A perf counter can also be put into a map after commit 35578d79 ("bpf: Implement function bpf_perf_event_read() that get the selected hardware PMU counter") 4. For a hash table, it should be possible to use a string or other value as a key; 5. It is possible that map configuration is unable to be setup during parsing. A perf counter is an example. Therefore, this patch does the following: 1. Instead of updating map element during parsing, this patch stores map config options in 'struct bpf_map_priv'. Following patches will apply those configs at an appropriate time; 2. Link map operations in a list so a map can have multiple config terms attached, so different parts can be configured separately; 3. Make 'struct bpf_map_priv' extensible so that the following patches can add new types of keys and operations; 4. Use bpf_obj_config__map_funcs array to support more map config options. Since the patch changing the event parser to parse BPF object config is relative large, I've put it in another commit. Code in this patch can be tested after applying the next patch. [1] http://lkml.kernel.org/g/564ED621.4050500@huawei.comSigned-off-by: -
Namhyung Kim authored
The dynamic entry is created for each field in a tracepoint event. Since they have no fixed hpp format index, it should skip when perf_hpp__reset_width() is called. This caused following assertion failure.. $ perf record -e sched:sched_switch -a sleep 1 $ perf report -s comm,next_pid --stdio perf: ui/hist.c:651: perf_hpp__reset_width: Assertion `!(fmt->idx >= PERF_HPP__MAX_INDEX)' failed. Signed-off-by:Namhyung Kim <namhyung@kernel.org> Acked-by:
-
Namhyung Kim authored
It missed to update column length of the 'trace' sort key in the hists__calc_col_len() so it might truncate the output. It calculated the column length in the ->cmp() callback originally but it doesn't guarantee it's called always. Signed-off-by:
-
Namhyung Kim authored
The srcline, srcfile and trace sort keys can have long entries. With commit 89fee709 ("perf hists: Do column alignment on the format iterator"), it now aligns output with hist_entry__snprintf_alignment(). So each (possibly long) sort entries don't need to do it themselves. Signed-off-by:
-
Namhyung Kim authored
Normally the hist entry's srcline and/or srcfile is set during sorting. However sometime it's possible to a hist entry's srcline is not set yet after the sorting. This is because the entry is so unique and other sort keys already make it distinct. Then the srcline/file sort didn't have a chance to be called during the sorting. In that case it has NULL srcline/srcfile field and shows nothing. Before: $ perf report -s comm,sym,srcline ... Overhead Command Symbol ----------------------------------------------------------------- 34.42% swapper [k] intel_idle intel_idle.c:0 2.44% perf [.] __poll_nocancel (null) 1.70% gnome-shell [k] fw_domains_get (null) 1.04% Xorg [k] sock_poll (null) After: 34.42% swapper [k] intel_idle intel_idle.c:0 2.44% perf [.] __poll_nocancel .:0 1.70% gnome-shell [k] fw_domains_get fw_domains_get+42 1.04% Xorg [k] sock_poll socket.c:0 Signed-off-by: -
Namhyung Kim authored
A dynamic entry is created for each tracepoint event. When it sets up the sort key, it checks with existing keys using ->equal() callback. But it missed to set the ->equal for dynamic entries. The following segfault was due to the missing ->equal() callback. (gdb) bt #0 0x0000000000140003 in ?? () #1 0x0000000000537769 in fmt_equal (b=0x2106980, a=0x21067a0) at ui/hist.c:548 #2 perf_hpp__setup_output_field (list=0x8c6d80 <perf_hpp_list>) at ui/hist.c:560 #3 0x00000000004e927e in setup_sorting (evlist=<optimized out>) at util/sort.c:2642 #4 0x000000000043cf50 in cmd_report (argc=<optimized out>, argv=<optimized out>, prefix=<optimized out>) at builtin-report.c:932 #5 0x00000000004865a1 in run_builtin (p=p@entry=0x8bbce0 <commands+192>, argc=argc@entry=7, argv=argv@entry=0x7ffd24d56ce0) at perf.c:390 #6 0x000000000042dc1f in handle_internal_command (argv=0x7ffd24d56ce0, argc=7) at perf.c:451 #7 run_argv (argv=0x7ffd24d56a70, argcp=0x7ffd24d56a7c) at perf.c:495 #8 main (argc=7, argv=0x7ffd24d56ce0) at perf.c:620 Signed-off-by: -
Arnaldo Carvalho de Melo authored
Older compilers don't like this, for instance, on RHEL6.7: CC /tmp/build/perf/util/parse-events.o util/parse-events.c:844: error: redefinition of typedef ‘config_term_func_t’ util/parse-events.c:353: note: previous declaration of ‘config_term_func_t’ was here So remove the second definition, that should've been just moved in 43d0b978 ("perf tools: Enable config and setting names for legacy cache events"), not copied. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Fixes: 43d0b978 ("perf tools: Enable config and setting names for legacy cache events") Signed-off-by: -
Arnaldo Carvalho de Melo authored
In RHEL 6.7: CC /tmp/build/perf/util/parse-events.o cc1: warnings being treated as errors util/parse-events.c: In function ‘parse_events_add_cache’: util/parse-events.c:366: error: declaration of ‘error’ shadows a global declaration util/util.h:136: error: shadowed declaration is here Rename it to 'err'. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Fixes: 43d0b978 ("perf tools: Enable config and setting names for legacy cache events") Signed-off-by:
-
- 20 Feb, 2016 1 commit
-
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo: User visible changes: - Add 'perf record' --all-user/--all-kernel options, so that one can tell that all the events in the command line should be restricted to the user or kernel levels (Jiri Olsa), i.e.: perf record -e cycles:u,instructions:u is equivalent to: perf record --all-user -e cycles,instructions - Fix percentage update on key press, due to the buffering code (that creates hist_entries that will later be consumed) touching per hists state that is used by the display thread (Namhyung Kim) - Bail out when event modifiers not supported by 'perf stat' are specified, i.e.: (Wang Nan) # perf stat -e cycles/no-inherit/ usleep 1 event syntax error: 'cycles/no-inherit/' \___ 'no-inherit' is not usable in 'perf stat' # perf stat -e cycles/foo/ usleep 1 event syntax error: 'cycles/foo/' \___ unknown term valid terms: config,config1,config2,name # - Enable setting names for legacy cache, raw and numeric events, e.g: (Wang Nan) # perf record -e cycles -e 4:0x6530160/name=evtx,call-graph=fp/ -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.659 MB perf.data (844 samples) ] # perf evlist cycles evtx # Miscelaneous/Infrastructure changes: - Handled scaled == -1 case for counters in 'perf stat', fixing recent, only in perf/core, regression (Andi Kleen) - Reference count the cpu and thread maps at set_maps(), fixing the 'object code reading' 'perf test' entry when it was requesting a perf_event_attr.sample_freq > /proc/sys/kernel/perf_event_max_sample_rate (Arnaldo Carvalho de Melo) - Improve perf_evlist__strerror_open() to provide hints for -EINVAL due to perf_event_attr.sample_freq > /proc/sys/kernel/perf_event_max_sample_rate (Arnaldo Carvalho de Melo) - Add checks to various callchain and histogram routines (Namhyung Kim) - Fix checking asprintf return value when parsing additional event config terms (Wang Nan) Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
- 19 Feb, 2016 20 commits
-
-
Namhyung Kim authored
If it returns an error, warn user and bail out instead of silently ignoring it. Signed-off-by:
-
Namhyung Kim authored
Currently hists__collapse_resort() and hists__collapse_insert_entry() don't return an error code. Now that callchain_merge() can check for errors, abort and pass the error to the user. A later patch can add more work which also can fail. Signed-off-by:
-
Namhyung Kim authored
Now it can check the error case, so check and pass it to the caller. Signed-off-by:
-
Namhyung Kim authored
Now create_child() and add_child() return errors so check and pass it to the caller. Signed-off-by:
-
Namhyung Kim authored
The append_chain() might return either result of match_chain() or other (error) code. But match_chain() can return any value in s64 type so it's hard to check the error case. Add new enum match_result and make match_chain() return non-negative values only so that we can check the error cases. Signed-off-by:
-
Namhyung Kim authored
Memory allocation in the fill_node() can fail so change its return type to int and check it in add_child() too. Signed-off-by:
-
Namhyung Kim authored
The create_child() in add_child() can return NULL in case of memory allocation failure. So check the return value and bail out. The proper error handling will be added later. Signed-off-by:
-
Namhyung Kim authored
Currently 'perf top --tui' decrements percentage of all entries on any key press. This is because it adds total period as new samples are added to hists. As perf-top does it currently but added samples are not passed to the display thread, the percentages are decresing continuously. So separate total period stat into a different variable so that it cannot affect the output total period. This new total period stats are used only for calcualating callchain percent limit. Signed-off-by:
-
Wang Nan authored
This patch allows setting config terms for legacy cache events. For example: # perf stat -e L1-icache-misses/name=valA/ -e branches/name=valB/ ls ... Performance counter stats for 'ls': 11299 valA 451605 valB 0.000779091 seconds time elapsed # perf record -e cache-misses/name=inh/ -e cache-misses/name=noinh,no-inherit/ bash # ls # exit [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.023 MB perf.data (131 samples) ] # perf report --stdio | grep -B 1 'Event count' # Samples: 105 of event 'inh' # Event count (approx.): 109118 -- # Samples: 26 of event 'noinh' # Event count (approx.): 48302 A test case is introduced to test this feature. Signed-off-by: -
Wang Nan authored
This patch allows setting config terms for raw and numeric events. For example: # perf stat -e cycles/name=cyc/ ls ... 1821108 cyc ... # perf stat -e r6530160/name=event/ ls ... 1103195 event ... # perf record -e cycles -e 4:0x6530160/name=evtx,call-graph=fp/ -a sleep 1 ... # perf report --stdio ... # Samples: 124 of event 'cycles' 46.61% 0.00% swapper [kernel.vmlinux] [k] cpu_startup_entry 41.26% 0.00% swapper [kernel.vmlinux] [k] start_secondary ... # Samples: 91 of event 'evtx' ... 93.76% 0.00% swapper [kernel.vmlinux] [k] cpu_startup_entry | ---cpu_startup_entry | |--66.63%--call_cpuidle | cpuidle_enter | | ... 3 test cases are introduced to test config terms for symbol, raw and numeric events. Committer note: Further testing shows that we can retrieve the event name using 'perf evlist -v' and looking at the 'config' perf_event_attr field, i.e.: # perf record -e cycles -e 4:0x6530160/name=evtx,call-graph=fp/ -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.724 MB perf.data (2076 samples) ] # perf evlist cycles evtx # perf evlist -v cycles: size: 112, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|CPU|PERIOD|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1 evtx: type: 4, size: 112, config: 0x6530160, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|CALLCHAIN|CPU|PERIOD|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, freq: 1, sample_id_all: 1, exclude_guest: 1 # Signed-off-by: -
Arnaldo Carvalho de Melo authored
To remove duplicated code that differs only in using the matching '/a,b,c/' part or NULL if no event configuration is done ('//' or no pair of slashes at all). Will be used by some new targets allowing the configuration of hardware events, etc. Lifted part of the 'opt_event_config' nonterminal from a patch by Wang Nan. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Cody P Schafer <dev@codyps.com> Cc: He Kuang <hekuang@huawei.com> Cc: Jeremie Galarneau <jeremie.galarneau@efficios.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kirill Smelkov <kirr@nexedi.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/n/tip-e3xzpx9cqsmwnaguaxyw6r42@git.kernel.orgSigned-off-by: -
Wang Nan authored
Following commits will make more events obey /name=newname/ options. This patch makes pmu_event_name() a generic helper. Makes new get_config_name() accept NULL input to make life easier. Signed-off-by:
-
Wang Nan authored
'perf stat' accepts some config terms but doesn't apply them. For example: # perf stat -e 'instructions/no-inherit/' -e 'instructions/inherit/' bash # ls # exit Performance counter stats for 'bash': 266258061 instructions/no-inherit/ 266258061 instructions/inherit/ 1.402183915 seconds time elapsed The result is confusing, because user may expect the first 'instructions' event exclude the 'ls' command. This patch forbid most of these config terms for 'perf stat'. Result: # ./perf stat -e 'instructions/no-inherit/' -e 'instructions/inherit/' bash event syntax error: 'instructions/no-inherit/' \___ 'no-inherit' is not usable in 'perf stat' ... We can add blocked config terms back when 'perf stat' really supports them. This patch also removes unavailable config term from error message: # ./perf stat -e 'instructions/badterm/' ls event syntax error: 'instructions/badterm/' \___ unknown term valid terms: config,config1,config2,name # ./perf stat -e 'cpu/badterm/' ls event syntax error: 'cpu/badterm/' \___ unknown term valid terms: pc,any,inv,edge,cmask,event,in_tx,ldlat,umask,in_tx_cp,offcore_rsp,config,config1,config2,name Signed-off-by: -
Wang Nan authored
config_term_names[] is introduced for future commits which will be able to retrieve the config name through the config term. Utilize this array in parse_events_formats_error_string() so the missing '{,no-}inherit' terms are added. Signed-off-by: -
Wang Nan authored
According to man pages, asprintf returns -1 when failure. This patch fixes two incorrect return value checker. Signed-off-by:
-
Wang Nan authored
The name of bpf_prog_priv__clear() doesn't follow perf's naming convention. bpf_prog_priv__delete() seems to be a better name. However, bpf_prog_priv__delete() should be a method of 'struct bpf_prog_priv', but its first parameter is 'struct bpf_program'. It is callback from libbpf to clear priv structures when destroying a bpf program. It is actually a method of bpf_program (libbpf object), but bpf_program__ functions should be provided by libbpf. This patch removes the prefix of that function. Signed-off-by:
-
Andi Kleen authored
Arnaldo pointed out that the earlier cb110f47 ("perf stat: Move noise/running printing into printout") change changed behavior for not counted counters. This patch fixes it again. Signed-off-by:
Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Stephane Eranian <eranian@google.com> Fixes: cb110f47 ("perf stat: Move noise/running printing into printout") Link: http://lkml.kernel.org/r/1455749045-18098-2-git-send-email-andi@firstfloor.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Using 4 kHz is not necessary and sometimes is more than what was auto-tuned: # dmesg | grep max_sample_rate | tail -2 [ 2499.144373] perf interrupt took too long (2501 > 2500), lowering kernel.perf_event_max_sample_rate to 50000 [ 3592.413606] perf interrupt took too long (5069 > 5000), lowering kernel.perf_event_max_sample_rate to 25000 Simulating a auto-tune of 2000 we make the test fail, as reported by Steven Noonan for one of his machines, so reduce it to 500 HZ, it is enough to get a good number of samples for this test: # perf test -v 21 2>&1 | grep '^Reading object code for memory address' | tee /tmp/out | tail -5 Reading object code for memory address: 0x479f40 Reading object code for memory address: 0x7f29b7eea80d Reading object code for memory address: 0x7f29b7eea80d Reading object code for memory address: 0x7f29b7eea800 Reading object code for memory address: 0xffffffff813b2f23 [root@jouet ~]# wc -l /tmp/out 40 /tmp/out [root@jouet ~]# For systems that auto-tune below that, the previous patches will tell the user what is happening so that he may either ignore the result of this test or bump /proc/sys/kernel/perf_event_max_sample_rate. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Steven Noonan <steven@uplinklabs.net> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-6kufyy1iprdfzrbtuqgxir70@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Before: # perf test -v "code reading" 2>&1 | tail -4 perf_evlist__open failed test child finished with -1 ---- end ---- Test object code reading: FAILED! # After: # perf test -v "code reading" 2>&1 | tail -7 perf_evlist__open() failed! Error: Invalid argument. Hint: Check /proc/sys/kernel/perf_event_max_sample_rate. Hint: The current value is 1000 and 4000 is being requested. test child finished with -1 ---- end ---- Test object code reading: FAILED! # Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Steven Noonan <steven@uplinklabs.net> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-ifbx7vmrc38loe6317owz2jx@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
When running the "code reading" test we get: # perf test -v "code reading" 2>&1 | tail -5 Parsing event 'cycles:u' perf_evlist__open failed test child finished with -1 ---- end ---- Test object code reading: FAILED! # And with -vv we get the errno value, -22, i.e. -EINVAL, but we can do better and handle the case at hand, with this patch it becomes: # perf test -v "code reading" 2>&1 | tail -7 perf_evlist__open() failed! Error: Invalid argument. Hint: Check /proc/sys/kernel/perf_event_max_sample_rate. Hint: The current value is 1000 and 4000 is being requested. test child finished with -1 ---- end ---- Test object code reading: FAILED! # Next patch will make this 'perf test' entry to use perf_evlist__strerror() Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Steven Noonan <steven@uplinklabs.net> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-i31ai6kfefn75eapejjokfhc@git.kernel.orgSigned-off-by:
-
- 18 Feb, 2016 2 commits
-
-
Jiri Olsa authored
Allow user to easily switch all events to user or kernel space with simple --all-user or --all-kernel options. This will be handy within perf mem/c2c wrappers to switch easily monitoring modes. Committer note: Testing it: # perf record --all-kernel --all-user -a sleep 2 Error: option `all-user' cannot be used with all-kernel Usage: perf record [<options>] [<command>] or: perf record [<options>] -- <command> [<options>] --all-user Configure all used events to run in user space. --all-kernel Configure all used events to run in kernel space. # perf record --all-user --all-kernel -a sleep 2 Error: option `all-kernel' cannot be used with all-user Usage: perf record [<options>] [<command>] or: perf record [<options>] -- <command> [<options>] --all-kernel Configure all used events to run in kernel space. --all-user Configure all used events to run in user space. # perf record --all-user -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.416 MB perf.data (162 samples) ] # perf report | grep '\[k\]' # perf record --all-kernel -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.423 MB perf.data (296 samples) ] # perf report | grep '\[\.\]' # Signed-off-by: -
Arnaldo Carvalho de Melo authored
We were dropping the reference we possibly held but not obtaining one for the new maps, which we will drop at perf_evlist__delete(), fix it. This was caught by Steven Noonan in some of the machines which would produce this output when caught by glibc debug mechanisms: $ sudo perf test 21 21: Test object code reading :*** Error in `perf': corrupted double-linked list: 0x00000000023ffcd0 *** ======= Backtrace: ========= /usr/lib/libc.so.6(+0x72055)[0x7f25be0f3055] /usr/lib/libc.so.6(+0x779b6)[0x7f25be0f89b6] /usr/lib/libc.so.6(+0x7a0ed)[0x7f25be0fb0ed] /usr/lib/libc.so.6(__libc_calloc+0xba)[0x7f25be0fceda] perf(parse_events_lex_init_extra+0x38)[0x4cfff8] perf(parse_events+0x55)[0x4a0615] perf(perf_evlist__config+0xcf)[0x4eeb2f] perf[0x479f82] perf(test__code_reading+0x1e)[0x47ad4e] perf(cmd_test+0x5dd)[0x46452d] perf[0x47f4e3] perf(main+0x603)[0x42c723] /usr/lib/libc.so.6(__libc_start_main+0xf0)[0x7f25be0a1610] perf(_start+0x29)[0x42c859] Further investigation using valgrind led to the reference count imbalance fixed in this patch. Reported-and-Tested-by:
Steven Noonan <steven@uplinklabs.net> Report-Link: http://lkml.kernel.org/r/CAKbGBLjC2Dx5vshxyGmQkcD+VwiAQLbHoXA9i7kvRB2-2opHZQ@mail.gmail.com 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> Fixes: f30a79b0 ("perf tools: Add reference counting for cpu_map object") Link: http://lkml.kernel.org/n/tip-j0u1bdhr47sa511sgg76kb8h@git.kernel.orgSigned-off-by:
-
- 17 Feb, 2016 4 commits
-
-
Thomas Gleixner authored
No functional change, just less confusing to read. Signed-off-by:
Thomas Gleixner <tglx@linutronix.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Vince Weaver <vincent.weaver@maine.edu> Link: http://lkml.kernel.org/r/20160209201007.921540566@linutronix.deSigned-off-by:
-
Ingo Molnar authored
Signed-off-by: -
Thomas Gleixner authored
If CPU_UP_PREPARE is called it is not guaranteed, that a previously allocated and assigned hash has been freed already, but perf_event_init_cpu() unconditionally allocates and assignes a new hash if the swhash is referenced. By overwriting the pointer the existing hash is not longer accessible. Verify that there is no hash assigned on this cpu before allocating and assigning a new one. Signed-off-by:
-
Thomas Gleixner authored
If CPU_DOWN_PREPARE fails the perf hotplug notifier is called for CPU_DOWN_FAILED and calls perf_event_init_cpu(), which checks whether the swhash is referenced. If yes it allocates a new hash and stores the pointer in the per cpu data structure. But at this point the cpu is still online, so there must be a valid hash already. By overwriting the pointer the existing hash is not longer accessible. Remove the CPU_DOWN_FAILED state, as there is nothing to (re)allocate. Signed-off-by:
-
