- 26 Feb, 2016 14 commits
-
-
Wang Nan authored
With this patch the contend of BPF output event is printed by 'perf trace'. For example: # ./perf trace -a --ev bpf-output/no-inherit,name=evt/ \ --ev ./test_bpf_trace.c/map:channel.event=evt/ \ usleep 100000 ... 1.787 ( 0.004 ms): usleep/3832 nanosleep(rqtp: 0x7ffc78b18980 ) ... 1.787 ( ): evt:Raise a BPF event!..) 1.788 ( ): perf_bpf_probe:func_begin:(ffffffff810e97d0)) ... 101.866 (87.038 ms): gmain/1654 poll(ufds: 0x7f57a80008c0, nfds: 2, timeout_msecs: 1000 ) ... 101.866 ( ): evt:Raise a BPF event!..) 101.867 ( ): perf_bpf_probe:func_end:(ffffffff810e97d0 <- ffffffff81796173)) 101.869 (100.087 ms): usleep/3832 ... [continued]: nanosleep()) = 0 ... (There is an extra ')' at the end of several lines. However, it is another problem, unrelated to this commit.) Where test_bpf_trace.c is: /************************ 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!"; char err_str[] = "BAD %d\n"; int err; err = perf_event_output(ctx, &channel, get_smp_processor_id(), &output_str, sizeof(output_str)); if (err) trace_printk(err_str, sizeof(err_str), err); return 1; } 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> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1456479154-136027-8-git-send-email-wangnan0@huawei.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Wang Nan authored
Without this patch BPF map configuration is not applied. Command like this: # ./perf trace --ev bpf-output/no-inherit,name=evt/ \ --ev ./test_bpf_trace.c/map:channel.event=evt/ \ usleep 100000 Load BPF files without error, but since map:channel.event=evt is not applied, bpf-output event not work. This patch allows 'perf trace' load and run BPF scripts. Signed-off-by: -
Wang Nan authored
perf_evlist__set_filter() tries to set filter to every evsel linked in the evlist. However, since filters can only be applied to tracepoints, checking type of evsel before calling perf_evsel__set_filter() would be better. Signed-off-by:
-
Wang Nan authored
Before this patch each subcommand calls perf_config() by themself, reading the default configuration together with subcommand specific options. If a subcommand doesn't have it own options, it needs to call 'perf_config(perf_default_config, NULL)' to ensure .perfconfig is loaded. This patch brings perf_config(perf_default_config, NULL) to the very start of main(), so subcommands don't need to do it. After this patch, 'llvm.clang-path' works for 'perf trace'. Signed-off-by:
-
Namhyung Kim authored
The column width of dynamic entries is updated when comparing hist entries. However some unique entries can miss the chance to update. So move the update to output resort stage to make sure every entry will get called before display. To do that, abuse ->sort callback to update the width when the third argument is NULL. When resorting entries in normal path, it never be NULL so it should be fine IMHO. Before: # Overhead ptr / bytes_req / gfp_flags # .............. .......................................... # 37.50% 0xffff8803f7669400 37.50% 448 37.50% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 10.42% 0xffff8803f766be00 8.33% 96 8.33% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 2.08% 512 2.08% GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP <-- here After: # Overhead ptr / bytes_req / gfp_flags # .............. ..................................................... # 37.50% 0xffff8803f7669400 37.50% 448 37.50% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 10.42% 0xffff8803f766be00 8.33% 96 8.33% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 2.08% 512 2.08% GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC Signed-off-by:Namhyung Kim <namhyung@kernel.org> Tested-by:
Jiri Olsa <jolsa@kernel.org> Cc: Andi Kleen <andi@firstfloor.org> Cc: David Ahern <dsahern@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/r/1456512767-1164-5-git-send-email-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
When dynamic sort key is used it might not show pretty printed output. This is because the trace output was not set only for the first dynamic sort key. During hierarchy_insert_entry() it missed to pass the trace_output to dynamic entries. Also even if it did, only first entry will have it. Subsequent entries might set it during collapsing stage but it's not guaranteed. Before: $ perf report --hierarchy --stdio -s ptr,bytes_req,gfp_flags -g none # # Overhead ptr / bytes_req / gfp_flags # .............. .......................................... # 37.50% 0xffff8803f7669400 37.50% 448 37.50% 66080 10.42% 0xffff8803f766be00 8.33% 96 8.33% 66080 2.08% 512 2.08% 67280 After: # # Overhead ptr / bytes_req / gfp_flags # .............. .......................................... # 37.50% 0xffff8803f7669400 37.50% 448 37.50% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 10.42% 0xffff8803f766be00 8.33% 96 8.33% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 2.08% 512 2.08% GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP Signed-off-by: -
Namhyung Kim authored
The dynamic entries are right-aligned unlike other entries since it usually has numeric value. But for the hierarchy mode, left alignment is more appropriate IMHO. Also trim spaces on the left so that we can easily identify the hierarchy. Before: $ perf report --hierarchy -i perf.data.kmem -s gfp_flags,ptr,bytes_req --stdio -g none ... # # Overhead gfp_flags / ptr / bytes_req # .............. ................................................................................................. # 91.67% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 37.50% 0xffff8803f7669400 37.50% 448 8.33% 0xffff8803f766be00 8.33% 96 4.17% 0xffff8800d156dc00 4.17% 704 After: # Overhead gfp_flags / ptr / bytes_req # .............. .................................... # 91.67% GFP_ATOMIC|GFP_NOWARN|GFP_NOMEMALLOC 37.50% 0xffff8803f7669400 37.50% 448 8.33% 0xffff8803f766be00 8.33% 96 4.17% 0xffff8800d156dc00 4.17% 704 Signed-off-by: -
Namhyung Kim authored
When dynamic entries are used in the hierarchy mode with multiple events, the output might not be aligned properly. In the hierarchy mode, the each sort column is indented using total number of sort keys. So it keeps track of number of sort keys when adding them. However a dynamic sort key can be added more than once when multiple events have same field names. This results in unnecessarily long indentation in the output. For example perf kmem records following events: $ perf evlist --trace-fields -i perf.data.kmem kmem:kmalloc: trace_fields: call_site,ptr,bytes_req,bytes_alloc,gfp_flags kmem:kmalloc_node: trace_fields: call_site,ptr,bytes_req,bytes_alloc,gfp_flags,node kmem:kfree: trace_fields: call_site,ptr kmem:kmem_cache_alloc: trace_fields: call_site,ptr,bytes_req,bytes_alloc,gfp_flags kmem:kmem_cache_alloc_node: trace_fields: call_site,ptr,bytes_req,bytes_alloc,gfp_flags,node kmem:kmem_cache_free: trace_fields: call_site,ptr kmem:mm_page_alloc: trace_fields: page,order,gfp_flags,migratetype kmem:mm_page_free: trace_fields: page,order As you can see, many field names shared between kmem events. So adding 'ptr' dynamic sort key alone will set nr_sort_keys to 6. And this adds many unnecessary spaces between columns. Before: $ perf report -i perf.data.kmem --hierarchy -s ptr -g none --stdio ... # Overhead ptr # ....................... ................................... # 99.89% 0xffff8803ffb79720 0.06% 0xffff8803d228a000 0.03% 0xffff8803f7678f00 0.00% 0xffff880401dc5280 0.00% 0xffff880406172380 0.00% 0xffff8803ffac3a00 0.00% 0xffff8803ffac1600 After: # Overhead ptr # ........ .................... # 99.89% 0xffff8803ffb79720 0.06% 0xffff8803d228a000 0.03% 0xffff8803f7678f00 0.00% 0xffff880401dc5280 0.00% 0xffff880406172380 0.00% 0xffff8803ffac3a00 0.00% 0xffff8803ffac1600 Signed-off-by: -
Namhyung Kim authored
When hist_entry__cmp() and hist_entry__collapse() are called, they should check if the dynamic entry is comparing matching hists only. Otherwise it might access different hists resulting in incorrect output. Signed-off-by:
-
Namhyung Kim authored
Like the stdio, it should show messages about omitted hierarchy entries. Please refer the previous commit for more details. Signed-off-by:
-
Namhyung Kim authored
Like the stdio, it should show messages about omitted hierarchy entries. Please refer the previous commit for more details. As it needs to check an entry is omitted or not multiple times, add the has_no_entry field in the hist entry. Suggested-and-Tested-by:
-
Namhyung Kim authored
The previous patch introduced __rb_hierarchy_next() function with various move direction like HMD_FORCE_CHILD but missed to change using it some place. Signed-off-by:
-
Namhyung Kim authored
When the hierarchy mode is used, some entries might be omiited due to a percent limit or filter. In this case the output hierarchy is different than other entries. Add an informative message to users about this. For example, when 4% of percent limit is applied: Before: # Overhead Command / Shared Object / Symbol # .............. .......................................... # 49.09% swapper 48.67% [kernel.vmlinux] 34.42% [k] intel_idle 11.51% firefox 8.87% libpthread-2.22.so 6.60% [.] __GI___libc_recvmsg 10.49% gnome-shell 4.74% libc-2.22.so 10.08% Xorg 6.11% libc-2.22.so 5.27% [.] __memcpy_sse2_unaligned 6.15% perf Note that, gnome-shell/libc has no symbols and perf has no dso/symbols. With that patch the output will look like below: After: # Overhead Command / Shared Object / Symbol # .............. .......................................... # 49.09% swapper 48.67% [kernel.vmlinux] 34.42% [k] intel_idle 11.51% firefox 8.87% libpthread-2.22.so 6.60% [.] __GI___libc_recvmsg 10.49% gnome-shell 4.74% libc-2.22.so no entry >= 4.00% 10.08% Xorg 6.11% libc-2.22.so 5.27% [.] __memcpy_sse2_unaligned 6.15% perf no entry >= 4.00% Suggested-and-Tested-by:Arnaldo Carvalho de Melo <acme@kernel.org> Signed-off-by:
-
Namhyung Kim authored
The hists__overhead_width() is to calculate width occupied by the overhead (and others) columns before the sort columns. The hist_entry__has_hiearchy_children() is to check whether an entry has lower entries (children) in the hierarchy to be shown in the output. This means the children should not be filtered out and above the percent limit. These two functions will be used to show information when all children of an entry is omitted by the percent limit (or filter). Signed-off-by:
-
- 25 Feb, 2016 6 commits
-
-
Taeung Song authored
script_spec_register() called two functions: script_spec__find() and script_spec__findnew(). But this way script_spec__find() gets called two times, directly and via script_spec__findnew(). So remove script_spec__findnew() and make script_spec_register() only call once script_spec__find(). Signed-off-by:
Taeung Song <treeze.taeung@gmail.com> Acked-by:
-
Taeung Song authored
After collecting samples for events 'syscalls:', perf-script with python script doesn't occasionally work generating a segmentation fault. The reason is that the print fmt is empty and a value of event->print_fmt.args is NULL, so dereferencing the null pointer results in a segmentation fault i.e.: # perf record -e syscalls:* # perf script -g python # perf script -s perf-script.py in trace_begin syscalls__sys_enter_brk 3 79841.832099154 3777 test.sh syscall_nr=12, brk=0 ... (omitted) ... Segmentation fault (core dumped) For example, a format of sys_enter_getuid() hasn't print fmt as below. # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_getuid/format name: sys_enter_getuid ID: 188 format: field:unsigned short common_type; offset:0; size:2; signed:0; field:unsigned char common_flags; offset:2; size:1; signed:0; field:unsigned char common_preempt_count; offset:3; size:1; signed:0; field:int common_pid; offset:4; size:4; signed:1; field:int syscall_nr; offset:8; size:4; signed:1; print fmt: "" So add exception handling to avoid this problem. Signed-off-by: -
Arnaldo Carvalho de Melo authored
In 1d55e8ef ("perf tools: Introduce opt_event_config nonterminal") I removed the unconditional "'/' '/'" for pmu events such as "intel_pt//" but forgot to use opt_event_config where it expected some event_config, oops. Fix it. Noticed when trying to use: # perf record -e intel_pt// -a sleep 1 event syntax error: 'intel_pt//' \___ parser error 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 # 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: 1d55e8ef ("perf tools: Introduce opt_event_config nonterminal") Signed-off-by:
-
Stephane Eranian authored
This patch improves the error message given by jvmti Makefile when the alternatives command cannot be found. It now suggests the user locates the root of their Java installation and pass it with JDIR= Signed-off-by:
Stephane Eranian <eranian@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1456378056-18812-1-git-send-email-eranian@google.comSigned-off-by:
-
Arnaldo Carvalho de Melo authored
No need to use strbuf there, its just a simple alloc+formatting, which asprintf does just fine. 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-6q6cxfhk8c8ypg3tfpo0i2iy@git.kernel.orgSigned-off-by:
-
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:
Ingo Molnar <mingo@kernel.org>
-
- 24 Feb, 2016 20 commits
-
-
Namhyung Kim authored
Support hierarchy output for perf-top using --hierarchy option. Signed-off-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 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:
-
