- 25 Feb, 2016 1 commit
-
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo-20160224' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements from Arnaldo Carvalho de Melo: User visible changes: - Hierarchy histogram mode for 'perf top' and 'perf report', showing multiple levels, one per --sort entry: (Namhyung Kim) On a mostly idle system: # perf top --hierarchy -s comm,dso Then expand some levels and use 'P' to take a snapshot: # cat perf.hist.0 - 92.32% perf 58.20% perf 22.29% libc-2.22.so 5.97% [kernel] 4.18% libelf-0.165.so 1.69% [unknown] - 4.71% qemu-system-x86 3.10% [kernel] 1.60% qemu-system-x86_64 (deleted) + 2.97% swapper # - Check availability of memory events in 'perf mem': (Jiri Olsa) On a Intel Broadwell machine: # perf mem record -e list ldlat-loads : available ldlat-stores: available # - Decode data_src values (e.g. perf.data files generated by 'perf mem record') in 'perf script': (Jiri Olsa) # perf script perf 693 [1] 4.088652: 1 cpu/mem-loads,ldlat=30/P: ffff88007d0b0f40 68100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK No <SNIP> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - Print bpf-output events in 'perf script': (Wang Nan). # perf record -e bpf-output/no-inherit,name=evt/ -e ./test_bpf_output_3.c/map:channel.event=evt/ usleep 1000 # perf script usleep 4882 21384.532523: evt: ffffffff810e97d1 sys_nanosleep ([kernel.kallsyms]) BPF output: 0000: 52 61 69 73 65 20 61 20 Raise a 0008: 42 50 46 20 65 76 65 6e BPF even 0010: 74 21 00 00 t!.. BPF string: "Raise a BPF event!" # Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
- 24 Feb, 2016 32 commits
-
-
Namhyung Kim authored
Support hierarchy output for perf-top using --hierarchy option. Signed-off-by:
Namhyung Kim <namhyung@kernel.org> Acked-by:
Pekka Enberg <penberg@kernel.org> Cc: Andi Kleen <andi@firstfloor.org> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/r/1456326830-30456-19-git-send-email-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
In the hierarchy mode, hist entries should decay their children too. Also update hists__delete_entry() to be able to free child entries. Signed-off-by:
-
Namhyung Kim authored
The --hierarchy option is to show output in hierarchy mode. It extends folding/unfolding in the TUI and GTK browsers to support sort items as well as callchains. Users can toggle the items to see the performance result at wanted level. $ perf report --hierarchy --tui Overhead Command / Shared Object / Symbol -------------------------------------------------- + 32.96% gnome-shell - 15.11% swapper - 14.97% [kernel.vmlinux] 6.82% [k] intel_idle 0.66% [k] menu_select 0.43% [k] __hrtimer_start_range_ns ... Signed-off-by: -
Namhyung Kim authored
The hierarchy output mode is to group entries for each level so that user can see higher level picture more easily. Signed-off-by:
-
Namhyung Kim authored
Like in stdio, fit column header to hierarchy output. Merge column headers with "/" as a separator. Overhead Command / Shared Object / Symbol ... + 0.09% dwm + 0.06% emacs - 0.05% perf - 0.05% [kernel.vmlinux] + 0.03% [k] memcpy_orig + 0.01% [k] unmap_single_vma + 0.01% [k] smp_call_function_single + 0.00% [k] native_irq_return_iret + 0.00% [k] arch_trigger_all_cpu_backtrace_handler + 0.00% [k] native_write_msr_safe Signed-off-by: -
Namhyung Kim authored
Implement hierarchy mode in TUI. The output is look like stdio but it also supports to fold/unfold children dynamically. Signed-off-by:
-
Namhyung Kim authored
The 'C' and 'E' keys are to collapse/expand all hist entries. Update nr_hierarchy_entries properly in this case. Signed-off-by:
-
Namhyung Kim authored
Add nr_hierarchy_entries field to keep current number of (unfolded) hist entries. And the hist_entry->nr_rows carries number of direct children. But in the hierarchy mode, entry can have grand children and callchains. So update the number properly using hierarchy_count_rows() when toggling the folded state (by pressing ENTER key). Signed-off-by:
-
Namhyung Kim authored
The hierarchy output mode is to group entries so the existing columns won't fit to the new output. Treat all sort keys as a single column and separate headers by "/". # Overhead Command / Shared Object # ........... ................................ # 15.11% swapper 14.97% [kernel.vmlinux] 0.09% [libahci] 0.05% [iwlwifi] ... Signed-off-by: -
Namhyung Kim authored
The hierarchy output mode is to group entries for each level so that user can see higher level picture more easily. It also helps to find out which component is most costly. The output will look like below: 15.11% swapper 14.97% [kernel.vmlinux] 0.09% [libahci] 0.05% [iwlwifi] 10.29% irq/33-iwlwifi 6.45% [kernel.vmlinux] 1.41% [mac80211] 1.15% [iwldvm] 1.14% [iwlwifi] 0.14% [cfg80211] 4.81% firefox 3.92% libxul.so 0.34% [kernel.vmlinux] Signed-off-by: -
Namhyung Kim authored
It'll be used for hierarchy output mode to indent entries properly. Signed-off-by:
-
Namhyung Kim authored
In hierarchy mode, a filter can affect periods of entries in upper hierarchy. So it needs to resort the hists after filter. For example, let's look at following example: Overhead Command / Shared Object / Symbol ------------ -------------------------------- 30.00% perf 20.00% perf 10.00% main 5.00% pr_debug 5.00% memcpy 10.00% [kernel.vmlinux] 8.00% memset 2.00% cpu_idle If we apply simbol filter for 'mem' it should look like this 13.00% perf 8.00% [kernel.vmlinux] 8.00% memset 5.00% perf 5.00% memcpy Signed-off-by: -
Namhyung Kim authored
The hists__filter_hierarchy() function implements filtering in hierarchy mode. Now we have hist_entry__filter() so use it for entries in the hierarchy. It returns 3 kind of values. A negative value means that it's not filtered by this type. It marks current entry as filtered tentatively so if a lower level entry removes the filter it also removes the all parent so that we can find the entry in the output. Zero means it's filtered out by this type. A positive value means it's not filtered so it removes the filter and shows in the output. In these cases, it moves to next entry since lower level entry won't match by this type of filter anymore. Thus all children will be filtered or not together. Signed-off-by:
-
Namhyung Kim authored
The hist_entry__filter() function is to filter hist entries using sort key related info. This is needed to support hierarchy mode since each hist entry will be associated with a hpp fmt which has a sort key. So each entry should compare to only matching type of filters. To do that, add the ->se_filter callback field to struct sort_entry. This callback takes 'type' argument which determines whether it's matching sort key or not. It returns -1 for non-matching type, 0 for filtered entry and 1 for not filtered entries. Signed-off-by:
-
Namhyung Kim authored
The rb_hierarchy_{next,prev,last} functions are to traverse all hist entries in a hierarchy. They will be used by various function which supports hierarchy output. As the rb_hierarchy_next() is used to traverse the whole hierarchy, it sometime needs to visit entries regardless of current folding state. So add enum hierarchy_move_dir and pass it to __rb_hierarchy_next() for those cases. Signed-off-by: -
Namhyung Kim authored
For hierarchical output, each entry must be sorted in their rbtree (hroot) properly. Add hists__hierarchy_output_resort() to do the job. Note that those hierarchy entries share the period counts, it'd be important to update the hists->stats only once (for leaves). Signed-off-by:
-
Namhyung Kim authored
In the hierarchical view, entries will be grouped and sorted on the first key, and then on the second key, and so on. Add the he->hroot_{in,out} fields to keep the lower level entries. Actually this can share space, in a union, with callchain's 'sorted_root' since the hroots are only used by non-leaf entries and callchain is only used by leaf entries. It also adds the 'parent_he' and 'depth' fields which can be used by browsers. This patch only implements collapsing part which creates internal entries for each sort key. These need to be sorted by output_sort stage and to be displayed properly in the later patch(es). Signed-off-by: -
Namhyung Kim authored
The 'trace', 'srcline' and 'srcfile' sort keys updates hist entry's field later. With the hierarchy mode, those fields are passed to a matching entry so it needs to identify the sort keys. Signed-off-by:
-
Wang Nan authored
This patch allows 'perf script' output messages from BPF program. For example, use test_bpf_output_3.c at the end of this commit message, # ./perf record -e bpf-output/no-inherit,name=evt/ \ -e ./test_bpf_output_3.c/map:channel.event=evt/ \ usleep 100000 # ./perf script usleep 4882 21384.532523: evt: ffffffff810e97d1 sys_nanosleep ([kernel.kallsyms]) BPF output: 0000: 52 61 69 73 65 20 61 20 Raise a 0008: 42 50 46 20 65 76 65 6e BPF even 0010: 74 21 00 00 t!.. BPF string: "Raise a BPF event!" usleep 4882 21384.632606: evt: ffffffff8105c609 kretprobe_trampoline_holder ([kernel.kallsyms BPF output: 0000: 52 61 69 73 65 20 61 20 Raise a 0008: 42 50 46 20 65 76 65 6e BPF even 0010: 74 21 00 00 t!.. BPF string: "Raise a BPF event!" Two samples from BPF output are printed by both binary and string format. If BPF program output something unprintable, string format is suppressed. /************************ BEGIN **************************/ #include <uapi/linux/bpf.h> struct bpf_map_def { unsigned int type; unsigned int key_size; unsigned int value_size; unsigned int max_entries; }; #define SEC(NAME) __attribute__((section(NAME), used)) static u64 (*ktime_get_ns)(void) = (void *)BPF_FUNC_ktime_get_ns; 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_output)(void *, struct bpf_map_def *, int, void *, unsigned long) = (void *)BPF_FUNC_perf_event_output; struct bpf_map_def SEC("maps") channel = { .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY, .key_size = sizeof(int), .value_size = sizeof(u32), .max_entries = __NR_CPUS__, }; static inline int __attribute__((always_inline)) func(void *ctx, int type) { char output_str[] = "Raise a BPF event!"; perf_event_output(ctx, &channel, get_smp_processor_id(), &output_str, sizeof(output_str)); return 0; } SEC("func_begin=sys_nanosleep") int func_begin(void *ctx) {return func(ctx, 1);} SEC("func_end=sys_nanosleep%return") int func_end(void *ctx) { return func(ctx, 2);} char _license[] SEC("license") = "GPL"; int _version SEC("version") = LINUX_VERSION_CODE; /************************* END ***************************/ Signed-off-by:Wang Nan <wangnan0@huawei.com> Tested-by:
-
Wang Nan authored
Move code printing binray data from trace_event() to utils.c and allows passing different printer. Further commits will use this logic to print bpf output event. Signed-off-by:
-
Jiri Olsa authored
Adding support to display data_src values, for events with data_src data in sample. Example: $ perf script ... rcuos/3 32 [002] ... 68501042 Local RAM hit|SNP None or Hit|TLB L1 or L2 hit|LCK No ... rcuos/3 32 [002] ... 68100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK No ... swapper 0 [002] ... 68100242 LFB hit|SNP None|TLB L1 or L2 hit|LCK No ... swapper 0 [000] ... 68100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK No ... swapper 0 [000] ... 50100142 L1 hit|SNP None|TLB L2 miss|LCK No ... rcuos/3 32 [002] ... 68100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK No ... plugin-containe 16538 [000] ... 6a100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK Yes ... gkrellm 1736 [000] ... 68100242 LFB hit|SNP None|TLB L1 or L2 hit|LCK No ... gkrellm 1736 [000] ... 6a100142 L1 hit|SNP None|TLB L1 or L2 hit|LCK Yes ... ^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ data_src value data_src translation Signed-off-by:Jiri Olsa <jolsa@kernel.org> Tested-by:
-
Jiri Olsa authored
Moving strncat call into scnprintf to easily track number of displayed bytes. It will be used in following patch. Signed-off-by:
-
Jiri Olsa authored
Moving strncat/strcpy calls into scnprintf to easily track number of displayed bytes. It will be used in following patch. Signed-off-by:
-
Jiri Olsa authored
Moving strncat/strcpy calls into scnprintf to easily track number of displayed bytes. It will be used in following patch. Signed-off-by:
-
Jiri Olsa authored
Moving strncat/strcpy calls into scnprintf to easily track number of displayed bytes. It will be used in following patch. Signed-off-by:
-
Jiri Olsa authored
Move meminfo's lck display function into mem-events.c object, so it could be reused later from script code. Signed-off-by:
-
Jiri Olsa authored
Move meminfo's snp display function into mem-events.c object, so it could be reused later from script code. Signed-off-by:
-
Jiri Olsa authored
Move meminfo's lvl display function into mem-events.c object, so it could be reused later from script code. Signed-off-by:
-
Jiri Olsa authored
Move meminfo's tlb display function into mem-events.c object, so it could be reused later from script code. Signed-off-by:
-
Jiri Olsa authored
Wrap perf_mem_events[].name into perf_mem_events__name() so we could alter the events name if needed. This will be handy when changing latency settings for loads event in following patch. Signed-off-by:
-
Jiri Olsa authored
Check if current kernel support available memory events and display the status within -e list option: $ perf mem record -e list ldlat-loads : available ldlat-stores : available Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo-2' 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: New features: - Add API to set values of map entries in a BPF object, be it individual map slots or ranges (Wang Nan) - Introduce support for the 'bpf-output' event (Wang Nan) - Add glue to read perf events in a BPF program (Wang Nan) User visible changes: - Don't stop PMU parsing on alias parse error, allowing the addition of new sysfs PMU files without breaking old tools (Andi Kleen) - Implement '%' operation in libtraceevent (Daniel Bristot de Oliveira) - Allow specifying events via -e in 'perf mem record', also listing what events can be specified via 'perf mem record -e list' (Jiri Olsa) - Improve support to 'data_src', 'weight' and 'addr' fields in 'perf script' (Jiri Olsa) Infrastructure changes: - Export cacheline routines (Jiri Olsa) - Remove strbuf_{remove,splice}(), dead code (Arnaldo Carvalho de Melo) Fixes: - Sort key fixes: Alignment for srcline, file, trace; fix segfault for dynamic, trace events related sort keys (Namyung Kim) Build fixes: - Remove duplicate typedef config_term_func_t definition, fixing the build on older systems (Arnaldo Carvalho de Melo) Signed-off-by:
-
- 23 Feb, 2016 7 commits
-
-
Arnaldo Carvalho de Melo authored
No users, nuke them. 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> Link: http://lkml.kernel.org/n/tip-kfv2wo8xann8t97wdalttcx7@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
It is the only user of this function, just use the strlen() to skip the prefix. 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> Link: http://lkml.kernel.org/n/tip-blao710l5cd5hmwrhy51ftgq@git.kernel.orgSigned-off-by:
-
Andi Kleen authored
When an error happens during alias parsing currently the complete parsing of all attributes of the PMU is stopped. This is breaks old perf on a newer kernel that may have not-yet-know alias attributes (such as .scale or .per-pkg). Continue when some attribute is unparseable. This is IMHO a stable candidate and should be backported to older versions to avoid problems with newer kernels. v2: Print warnings when something goes wrong. v3: Change warning to debug output Signed-off-by:
Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: stable@vger.kernel.org # v3.6+ Link: http://lkml.kernel.org/r/1455749095-18358-1-git-send-email-andi@firstfloor.orgSigned-off-by:
-
Jiri Olsa authored
Adding addr/data_src/weight columns for raw events. Example: $ perf script ... true 11883 322960.489590: ... ffff8801aa0b8400 68501042 246 ffffffff813b2cd true 11883 322960.489600: ... ffff8800b90b38d8 68501042 251 ffffffff811d0b7 true 11883 322960.489612: ... ffff880196893130 6a100142 94 ffffffff8177fb8 true 11883 322960.489637: ... ffff880164277b40 68100842 101 ffffffff813b2cd true 11883 322960.489683: ... ffff880035d3d818 68501042 201 ffffffff811d0b7 true 11883 322960.489733: ... 7fb9616efcf0 68100242 199 7fb961aaba9 true 11883 322960.489818: ... ffffea000481c39c 6a100142 122 ffffffff811b634 ^^^^^^^^^^^^^^^^ ^^^^^^^^ ^^^ addr data_src weight Signed-off-by: -
Jiri Olsa authored
Adding data_src and weight column definitions, so it's displayed for related sample types. Signed-off-by:
-
Jiri Olsa authored
There's no need to define extra macros for that. Signed-off-by:
-
Jiri Olsa authored
Adding -e option for perf mem record command, to be able to specify memory event directly. Get list of available events: $ perf mem record -e list ldlat-loads ldlat-stores Monitor ldlat-loads: $ perf mem record -e ldlat-loads true Committer notes: Further testing: # perf mem record -e ldlat-loads true [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.020 MB perf.data (10 samples) ] # perf evlist cpu/mem-loads,ldlat=30/P # Signed-off-by:
-
