- 04 May, 2015 21 commits
-
-
Namhyung Kim authored
Currently 'perf kmem stat --page' shows total (page) allocation stat by default, but sometimes one might want to see live (total alloc-only) requests/pages only. The new --live option does this by subtracting freed allocation from the stat. E.g.: # perf kmem stat --page SUMMARY (page allocator) ======================== Total allocation requests : 988,858 [ 4,045,368 KB ] Total free requests : 886,484 [ 3,624,996 KB ] Total alloc+freed requests : 885,969 [ 3,622,628 KB ] Total alloc-only requests : 102,889 [ 422,740 KB ] Total free-only requests : 515 [ 2,368 KB ] Total allocation failures : 0 [ 0 KB ] Order Unmovable Reclaimable Movable Reserved CMA/Isolated ----- ------------ ------------ ------------ ------------ ------------ 0 172,173 3,083 806,686 . . 1 284 . . . . 2 6,124 58 . . . 3 114 335 . . . 4 . . . . . 5 . . . . . 6 . . . . . 7 . . . . . 8 . . . . . 9 . . 1 . . 10 . . . . . # perf kmem stat --page --live SUMMARY (page allocator) ======================== Total allocation requests : 988,858 [ 4,045,368 KB ] Total free requests : 886,484 [ 3,624,996 KB ] Total alloc+freed requests : 885,969 [ 3,622,628 KB ] Total alloc-only requests : 102,889 [ 422,740 KB ] Total free-only requests : 515 [ 2,368 KB ] Total allocation failures : 0 [ 0 KB ] Order Unmovable Reclaimable Movable Reserved CMA/Isolated ----- ------------ ------------ ------------ ------------ ------------ 0 2,214 3,025 97,156 . . 1 59 . . . . 2 19 58 . . . 3 23 335 . . . 4 . . . . . 5 . . . . . 6 . . . . . 7 . . . . . 8 . . . . . 9 . . . . . 10 . . . . . # Signed-off-by:Namhyung Kim <namhyung@kernel.org> Acked-by:
Pekka Enberg <penberg@kernel.org> Tested-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Joonsoo Kim <js1304@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linux-mm@kvack.org Link: http://lkml.kernel.org/r/1429592107-1807-4-git-send-email-namhyung@kernel.org [ Added examples to the changeset log ] Signed-off-by:
-
Namhyung Kim authored
Add new sort keys for page: page, order, migtype, gfp - existing 'bytes', 'hit' and 'callsite' sort keys also work for page. Note that -s/--sort option should be preceded by either of --slab or --page option to determine where the sort keys applies. Now it properly groups and sorts allocation stats - so same page/caller with different order/migtype/gfp will be printed on a different line. # perf kmem stat --page --caller -l 10 -s order,hit ----------------------------------------------------------------------------- Total alloc (KB) | Hits | Order | Mig.type | GFP flags | Callsite ----------------------------------------------------------------------------- 64 | 4 | 2 | RECLAIM | 00285250 | new_slab 50,144 | 12,536 | 0 | MOVABLE | 0102005a | __page_cache_alloc 52 | 13 | 0 | UNMOVABL | 002084d0 | pte_alloc_one 40 | 10 | 0 | MOVABLE | 000280da | handle_mm_fault 28 | 7 | 0 | UNMOVABL | 000000d0 | __pollwait 20 | 5 | 0 | MOVABLE | 000200da | do_wp_page 20 | 5 | 0 | MOVABLE | 000200da | do_cow_fault 16 | 4 | 0 | UNMOVABL | 00000200 | __tlb_remove_page 16 | 4 | 0 | UNMOVABL | 000084d0 | __pmd_alloc 8 | 2 | 0 | UNMOVABL | 000084d0 | __pud_alloc ... | ... | ... | ... | ... | ... ----------------------------------------------------------------------------- Signed-off-by: -
Namhyung Kim authored
It is 'perf kmem' support caller statistics for page. Unlike slab case, the tracepoints in page allocator don't provide callsite info. So it records with callchain and extracts callsite info. Note that the callchain contains several memory allocation functions which has no meaning for users. So skip those functions to get proper callsites. I used following regex pattern to skip the allocator functions: ^_?_?(alloc|get_free|get_zeroed)_pages? This gave me a following list of functions: # perf kmem record --page sleep 3 # perf kmem stat --page -v ... alloc func: __get_free_pages alloc func: get_zeroed_page alloc func: alloc_pages_exact alloc func: __alloc_pages_direct_compact alloc func: __alloc_pages_nodemask alloc func: alloc_page_interleave alloc func: alloc_pages_current alloc func: alloc_pages_vma alloc func: alloc_page_buffers alloc func: alloc_pages_exact_nid ... The output looks mostly same as --alloc (I also added callsite column to that) but groups entries by callsite. Currently, the order, migrate type and GFP flag info is for the last allocation and not guaranteed to be same for all allocations from the callsite. --------------------------------------------------------------------------------------------- Total_alloc (KB) | Hits | Order | Mig.type | GFP flags | Callsite --------------------------------------------------------------------------------------------- 1,064 | 266 | 0 | UNMOVABL | 000000d0 | __pollwait 52 | 13 | 0 | UNMOVABL | 002084d0 | pte_alloc_one 44 | 11 | 0 | MOVABLE | 000280da | handle_mm_fault 20 | 5 | 0 | MOVABLE | 000200da | do_cow_fault 20 | 5 | 0 | MOVABLE | 000200da | do_wp_page 16 | 4 | 0 | UNMOVABL | 000084d0 | __pmd_alloc 16 | 4 | 0 | UNMOVABL | 00000200 | __tlb_remove_page 12 | 3 | 0 | UNMOVABL | 000084d0 | __pud_alloc 8 | 2 | 0 | UNMOVABL | 00000010 | bio_copy_user_iov 4 | 1 | 0 | UNMOVABL | 000200d2 | pipe_write 4 | 1 | 0 | MOVABLE | 000280da | do_wp_page 4 | 1 | 0 | UNMOVABL | 002084d0 | pgd_alloc --------------------------------------------------------------------------------------------- Signed-off-by: -
Masami Hiramatsu authored
Currently, perf-probe --list option ignores given event filter. ---- # ./perf probe -l vfs\* probe:vfs_read (on vfs_read@ksrc/linux-3/fs/read_write.c) probe_libc:malloc (on __libc_malloc@malloc/malloc.c in /usr/lib64/libc-2.17.so) ---- This changes --list option to accept the event filter argument as below. ---- # ./perf probe -l vfs\* probe:vfs_read (on vfs_read@ksrc/linux-3/fs/read_write.c) # ./perf probe -l \*libc:\* probe_libc:malloc (on __libc_malloc@malloc/malloc.c in /usr/lib64/libc-2.17.so) ---- Signed-off-by:Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Tested-by:
-
Masami Hiramatsu authored
Accept multiple filter options. Each filters are combined by logical-or. E.g. --filter abc* --filter *def is same as --filter abc*|*def Signed-off-by:
-
Masami Hiramatsu authored
Add strfilter__string to recover rules string from strfilter. This will be good for debugging. Signed-off-by:
-
Masami Hiramatsu authored
Add strfilter__or/and to append additional rules to existing strfilter. Signed-off-by:
-
Naveen N. Rao authored
Currently, perf probe considers patterns including a '.' to be a file. However, this causes problems on powerpc ABIv1 where all functions have a leading '.': $ perf probe -F | grep schedule_timeout_interruptible .schedule_timeout_interruptible $ perf probe .schedule_timeout_interruptible Semantic error :File always requires line number or lazy pattern. Error: Command Parse Error. Fix this: - by checking the probe pattern in more detail, and - skipping leading dot if one exists when creating/deleting events. Signed-off-by:Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Reviewed-by:
Srikar Dronamraju <srikar@linux.vnet.ibm.com> Acked-by:
-
Namhyung Kim authored
The YYLTYPE_IS_TRIVIAL is defined in the Build file, but unlike pmu-bison.c, gcc complained about it for parse-events-bison.c: CC util/parse-events-bison.o In file included from util/parse-events.y:16: util/parse-events-bison.h:101:1: error: "YYLTYPE_IS_TRIVIAL" redefined <command-line>: error: this is the location of the previous definition make[3]: *** [util/parse-events-bison.o] Error 1 Comments from Jiri Olsa: "Reason is the parse error handling that was added just recently: it adds YYLTYPE type (which is not present in pmu-bison.h), so YYLTYPE_IS_TRIVIAL gets redefined, which is ok in F20 that handle the error via '-w' option, but it's not ok for RHEL6 where the '-w' does not work for this kind of error." Signed-off-by:Jiri Olsa <jolsa@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1430322871-18107-1-git-send-email-namhyung@kernel.orgSigned-off-by:
-
Naveen N. Rao authored
Add a debug message to indicate that the build id didn't match. Signed-off-by:
-
Adrian Hunter authored
Add support for decoding an AUX area assuming it contains instruction tracing data. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1429903807-20559-4-git-send-email-adrian.hunter@intel.com [ Do not use -Z as an alternative to --itrace ] [ Fixed initialization of itrace_synth_opts struct fields on older gcc versions ] Signed-off-by:
-
Adrian Hunter authored
If a file contains AUX area tracing data then always allow fields 'addr' and 'cpu' to be selected as options for perf script. This is necessary because AUX area decoding may synthesize events with that information. Signed-off-by:
-
Adrian Hunter authored
Reference counting of the mmap buffer does not work correctly when there is an AUX area mmap also. In snapshot mode it is not easy to know if the AUX area mmap buffer contains usefull information. Equally the evlist does not know if the recording is in sanpshot mode anyway. Consequently, for now just assume the AUX area mmap always has data, which will just cause the mmap buffer to remain mmapped for the duration of the recording. Signed-off-by:
-
Naveen N. Rao authored
On powerpc ABIv2, if no debug-info is found and we use kallsyms, we need to fixup the function entry to point to the local entry point. Use offset of 8 since current toolchains always generate 2 instructions (8 bytes). Signed-off-by:
-
Naveen N. Rao authored
Use symbol table lookups by default if DWARF is not necessary, since powerpc ABIv2 encodes local entry points in the symbol table and the function entry address in DWARF may not be appropriate for kprobes, as described here: https://sourceware.org/bugzilla/show_bug.cgi?id=17638 "The DWARF address ranges deliberately include the *whole* function, both global and local entry points." ... "If you want to set probes on a local entry point, you should look up the symbol in the main symbol table (not DWARF), and check the st_other bits; they will indicate whether the function has a local entry point, and what its offset from the global entry point is. Note that GDB does the same when setting a breakpoint on a function entry." Signed-off-by:
-
Ananth N Mavinakayanahalli authored
ppc64 ELF ABIv2 has a Global Entry Point (GEP) and a Local Entry Point (LEP). For purposes of probing, we need the LEP - the offset to which is encoded in st_other. Signed-off-by:
Ananth N Mavinakayanahalli <ananth@in.ibm.com> Reviewed-by:
-
Naveen N. Rao authored
Allow perf probe to work on ppc ABIv1 without the need to specify the leading dot '.' for functions. 'perf probe do_fork' works with this patch. We do this by changing how symbol name comparison works on ppc ABIv1 - we simply ignore and skip over the initial dot, if one exists, during symbol name comparison. Signed-off-by:
-
Naveen N. Rao authored
Use the proper prefix when ignoring SyS symbols on ppc ABIv1. While at it, generalize symbol selection so architectures can implement their own logic. Signed-off-by:
-
Naveen N. Rao authored
If using the symbol table, symbol addresses are not being fixed up properly, resulting in probes being placed at wrong addresses: # perf probe do_fork Added new event: probe:do_fork (on do_fork) You can now use it in all perf tools, such as: perf record -e probe:do_fork -aR sleep 1 # cat /sys/kernel/debug/tracing/kprobe_events p:probe/do_fork _text+635952 # printf "%x" 635952 9b430 # grep do_fork /boot/System.map c0000000000ab430 T .do_fork Fix by checking for ELF type ET_DYN used by ppc64 kernels. Signed-off-by: -
Petr Holasek authored
In verbose mode perf bench numa shows also GB/s speed, system and user cpu time for each particular thread. Using of getrusage() can provide much more per process or per thread stats in future. Signed-off-by:
Petr Holasek <pholasek@redhat.com> Reviewed-by:
Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Link: http://lkml.kernel.org/r/1429198699-25039-3-git-send-email-pholasek@redhat.com [ Rename 'usage' variable to not shadow util.h's usage() ] Signed-off-by:
-
Will Deacon authored
Parsing /proc/cpuinfo is a fiddly, arch-dependent business and a recent change to get it working for Sparc broke arm and arm64 platforms. Use sysconf to determine the number of online CPUs only parsing /proc/cpuinfo when sysconf is not available. Signed-off-by:
Will Deacon <will.deacon@arm.com> Acked-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
-
- 29 Apr, 2015 19 commits
-
-
Namhyung Kim authored
As the --children option changes the output of perf report (and perf top) it sometimes confuses users. Add more words and examples to help understanding of the option's behavior - and how to disable it ;-). Signed-off-by:
-
Jiri Olsa authored
Separating metrics values for exclude_idle bit. Signed-off-by:
-
Jiri Olsa authored
Separating metrics values for guest and host, so we get proper values. Signed-off-by:
-
Jiri Olsa authored
Separating metrics values for exclude_hv bit. Signed-off-by:
-
Jiri Olsa authored
Changing metrics context calculation to allow more than 2 types of context. Following patches will add support for the rest of the exclude_* bits so we need separate array element for all context combinations. Signed-off-by:
-
Andi Kleen authored
Currently in perf IPC and other metrics cannot be directly shown separately for both user and kernel in a single run. The problem was that the metrics matching code did not check event qualifiers. With this patch the following case works correctly. % perf stat -e cycles:k,cycles:u,instructions:k,instructions:u true Performance counter stats for 'true': 531,718 cycles:k 203,895 cycles:u 338,151 instructions:k # 0.64 insns per cycle 105,961 instructions:u # 0.52 insns per cycle 0.002989739 seconds time elapsed Previously it would misreport the ratios because they were matching the wrong value. The patch is fairly big, but quite mechanic as it just adds context indexes everywhere. Reported-by:William Cohen <wcohen@redhat.com> Signed-off-by:
Andi Kleen <ak@linux.intel.com> Acked-by:
-
Arnaldo Carvalho de Melo authored
This comes from the desire of having -e/--expr to have the same meaning as for 'strace', while other perf tools use it for --event, which 'trace' honours, i.e. all perf tools have --event in common, but trace uses -e for strace's --expr. Clarify it in the --help output. Reported-by:
David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-5j94bcsdmcbeu2xthnzsj60d@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Cut't'paste error, fix it. Reported-by:
-
Masami Hiramatsu authored
Fix perf-probe --del option to delete all matched probes in both of kprobes and uprobes at once. When we have 2 or more events on different binaries as below, ---- # ./perf probe -l probe:vfs_read (on vfs_read@ksrc/linux-3/fs/read_write.c) probe_libc:malloc (on __libc_malloc@malloc/malloc.c in /usr/lib64/libc-2.17 ---- Trying to remove all event with '*' just removes kprobe events at first. ---- # ./perf probe -d \* Removed event: probe:vfs_read ---- And in 2nd try, it removes all uprobe events. ---- # ./perf probe -d \* Removed event: probe_libc:malloc ---- This fixes to remove all event at once as below. ---- # ./perf probe -d \* Removed event: probe:vfs_read Removed event: probe_libc:malloc ---- Reported-by:
Arnaldo Carvalho de Melo <acme@kernel.org> Signed-off-by:
-
Masami Hiramatsu authored
The --funcs option should be given exclusively. This adds PARSE_OPT_EXCUSIVE flag on --funcs (-F) option. Without this, 'perf probe --funcs -l' just shows the list of probes. With this, it shows error message correctly. This also fixes the help message and the documentation. Signed-off-by:
-
Jiri Olsa authored
Allowing symbolic events processing to report back error. $ perf record -e 'cycles/period=krava/' ls event syntax error: '../period=krava/' \___ expected numeric value $ perf record -e 'cycles/name=1/' ls event syntax error: '..es/name=1/' \___ expected string value Signed-off-by: -
Jiri Olsa authored
Allowing tracepoint events processing to report back error. $ perf record -e 'sched:krava' ls event syntax error: 'sched:krava' \___ unknown tracepoint ... Signed-off-by: -
Jiri Olsa authored
Allowing static terms like 'name,period,config,config1..' processing to report back error. $ perf record -e 'cpu/event=1,name=1/' ls event syntax error: '..=1,name=1/' \___ expected string value $ perf record -e 'cpu/event=1,period=krava/' ls event syntax error: '..,period=krava/' \___ expected numeric value $ perf record -e 'cpu/config=krava1/' ls event syntax error: '../config=krava1/' \___ expected numeric value Signed-off-by: -
Jiri Olsa authored
Allowing event's term processing to report back error, like: $ perf record -e 'cpu/even=0x1/' ls event syntax error: 'cpu/even=0x1/' \___ unknown term valid terms: pc,any,inv,edge,cmask,event,in_tx,ldlat,umask,in_tx_cp,offcore_rsp,config,config1,config2,name,period,branch_type Signed-off-by: -
Jiri Olsa authored
Saving the terms location within term struct, so it could be used later for report. Signed-off-by:
-
Jiri Olsa authored
Changing parse_events_add_pmu interface to allow propagating of the parse_events_error info. Signed-off-by:
-
Jiri Olsa authored
Not sure why we allowed the fail state, but it's wrong. Wrong type for 'name' term can cause segfault, and there's probably more fun hidden. Signed-off-by:
-
Jiri Olsa authored
Allowing flex parser to report back event parsing error, like: $ perf record -e cycles,cache-mises ls event syntax error: '..es,cache-mises' \___ parser error ... Signed-off-by: -
Jiri Olsa authored
Adding support to return error information from parse_events function. Following struct will be populated by parse_events function on return: struct parse_events_error { int idx; char *str; char *help; }; where 'idx' is the position in the string where the parsing failed, 'str' contains dynamically allocated error string describing the error and 'help' is optional help string. The change contains reporting function, which currently does not display anything. The code changes to supply error data for specific event types are coming in next patches. However this is what the expected output is: $ sudo perf record -e 'sched:krava' ls event syntax error: 'sched:krava' \___ unknown tracepoint ... $ perf record -e 'cpu/even=0x1/' ls event syntax error: 'cpu/even=0x1/' \___ unknown term valid terms: pc,any,inv,edge,cmask,event,in_tx,ldlat,umask,in_tx_cp,offcore_rsp,config,config1,config2,name,period,branch_type ... $ perf record -e cycles,cache-mises ls event syntax error: '..es,cache-mises' \___ parser error ... The output functions cut the beginning of the event string so the error starts up to 10th character and cut the end of the string of it crosses the terminal width. Signed-off-by:
-
