- 30 Jul, 2020 11 commits
-
-
Jiri Olsa authored
Collecting referenced metrics in struct metric_ref_node object, so we can process them later on. The change will parse nested metric names out of expression and 'resolve' them. All referenced metrics are dissolved into one context, meaning all nested metrics events and added to the parent context. Signed-off-by:
Jiri Olsa <jolsa@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: John Garry <john.garry@huawei.com> Cc: Michael Petlan <mpetlan@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lore.kernel.org/lkml/20200719181320.785305-9-jolsa@kernel.orgSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Jiri Olsa authored
Renaming __metricgroup__add_metric to __add_metric to fit in the current function names. Signed-off-by:
Kajol Jain <kjain@linux.ibm.com> Acked-by:
Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: John Garry <john.garry@huawei.com> Cc: Michael Petlan <mpetlan@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lore.kernel.org/lkml/20200719181320.785305-8-jolsa@kernel.orgSigned-off-by:
-
Jiri Olsa authored
Decouple metric adding logging into add_metric function, so it can be used from other places in following changes. Signed-off-by:
-
Jiri Olsa authored
Adding following macros to iterate events and metric: map_for_each_event(__pe, __idx, __map) - iterates over all pmu_events_map events map_for_each_metric(__pe, __idx, __map, __metric) - iterates over all metrics that match __metric argument and use it in metricgroup__add_metric function. Macros will be be used from other places in following changes. Signed-off-by: -
Jiri Olsa authored
Adding expr__del_id function to remove ID from hashmap. It will save us few lines in following changes. Signed-off-by:
-
Jiri Olsa authored
Changing expr__get_id to use and return struct expr_id_data pointer as value for the ID. This way we can access data other than value for given ID in following changes. Signed-off-by:
-
Jiri Olsa authored
Add the expr__add_id() function to data for ID with zero value, which is used when scanning the expression for IDs. Signed-off-by:
-
Jiri Olsa authored
Arnaldo found that we don't release value data in case the hashmap__set fails. Releasing it in case of an error. Reported-by:
-
Ian Rogers authored
Test that a command line option doesn't override the period set on a libpfm4 event. Without libpfm4 test passes as unsupported. Signed-off-by:
-
Jiri Olsa authored
Jin Yao reported issue with possible conflict between raw events and term values in pmu event syntax. Currently following syntax is resolved as raw event with 0xead value: uncore_imc_free_running/read/ instead of using 'read' term from uncore_imc_free_running pmu, because 'read' is correct raw event syntax with 0xead value. To solve this issue we do following: - check existing terms during rXXXX syntax processing and make them priority in case of conflict - allow pmu/r0x1234/ syntax to be able to specify conflicting raw event (implemented in previous patch) Also add automated tests for this and perf_pmu__parse_cleanup call to parse_events_terms, so the test gets properly cleaned up. Fixes: 3a6c51e4 ("perf parser: Add support to specify rXXX event with pmu") Reported-by:Jin Yao <yao.jin@linux.intel.com> Signed-off-by:
-
Jiri Olsa authored
Add support to specify raw event with 'r0<HEX>' syntax within pmu term syntax like: -e cpu/r0xdead/ It will be used to specify raw events in cases where they conflict with real pmu terms, like 'read', which is valid raw event syntax, but also a possible pmu term name as reported by Jin Yao. Reported-by:
-
- 29 Jul, 2020 2 commits
-
-
Wei Li authored
- auxtrace_record__init() is called only once, so there is no point in using a static variable to cache the results of find_all_arm_spe_pmus(), make it local and free the results after use. - Another reason is, even though SPE is micro-architecture dependent, but so far it only supports "statistical-profiling-extension-v1" and we have no chance to use multiple SPE's PMU events in Perf command. So remove the useless check code to make it clear. Signed-off-by:
Wei Li <liwei391@huawei.com> Reviewed-by:
Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Hanjun Guo <guohanjun@huawei.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kim Phillips <kim.phillips@arm.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: linux-arm-kernel@lists.infradead.org Link: http://lore.kernel.org/lkml/20200724071111.35593-3-liwei391@huawei.comSigned-off-by:
-
Wei Li authored
When recording with cache-misses and arm_spe_x event, I found that it will just fail without showing any error info if i put cache-misses after 'arm_spe_x' event. [root@localhost 0620]# perf record -e cache-misses \ -e arm_spe_0/ts_enable=1,pct_enable=1,pa_enable=1,load_filter=1,jitter=1,store_filter=1,min_latency=0/ sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.067 MB perf.data ] [root@localhost 0620]# [root@localhost 0620]# perf record -e arm_spe_0/ts_enable=1,pct_enable=1,pa_enable=1,load_filter=1,jitter=1,store_filter=1,min_latency=0/ \ -e cache-misses sleep 1 [root@localhost 0620]# The current code can only work if the only event to be traced is an 'arm_spe_x', or if it is the last event to be specified. Otherwise the last event type will be checked against all the arm_spe_pmus[i]->types, none will match and an out of bound 'i' index will be used in arm_spe_recording_init(). We don't support concurrent multiple arm_spe_x events currently, that is checked in arm_spe_recording_options(), and it will show the relevant info. So add the check and record of the first found 'arm_spe_pmu' to fix this issue here. Fixes: ffd3d18c ("perf tools: Add ARM Statistical Profiling Extensions (SPE) support") Signed-off-by:
Leo Yan <leo.yan@linaro.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Hanjun Guo <guohanjun@huawei.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kim Phillips <kim.phillips@arm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: linux-arm-kernel@lists.infradead.org Link: http://lore.kernel.org/lkml/20200724071111.35593-2-liwei391@huawei.comSigned-off-by:
-
- 28 Jul, 2020 1 commit
-
-
Davidlohr Bueso authored
The usefulness of having a standard way of testing syscall performance has come up from time to time[0]. Furthermore, some of our testing machinery (such as 'mmtests') already makes use of a simplified version of the microbenchmark. This patch mainly takes the same idea to measure syscall throughput compatible with 'perf-bench' via getppid(2), yet without any of the additional template stuff from Ingo's version (based on numa.c). The code is identical to what mmtests uses. [0] https://lore.kernel.org/lkml/20160201074156.GA27156@gmail.com/ Committer notes: Add mising stdlib.h and unistd.h to get the prototypes for exit() and getppid(). Committer testing: $ perf bench Usage: perf bench [<common options>] <collection> <benchmark> [<options>] # List of all available benchmark collections: sched: Scheduler and IPC benchmarks syscall: System call benchmarks mem: Memory access benchmarks numa: NUMA scheduling and MM benchmarks futex: Futex stressing benchmarks epoll: Epoll stressing benchmarks internals: Perf-internals benchmarks all: All benchmarks $ $ perf bench syscall # List of available benchmarks for collection 'syscall': basic: Benchmark for basic getppid(2) calls all: Run all syscall benchmarks $ perf bench syscall basic # Running 'syscall/basic' benchmark: # Executed 10000000 getppid() calls Total time: 3.679 [sec] 0.367957 usecs/op 2717708 ops/sec $ perf bench syscall all # Running syscall/basic benchmark... # Executed 10000000 getppid() calls Total time: 3.644 [sec] 0.364456 usecs/op 2743815 ops/sec $ Signed-off-by:
Davidlohr Bueso <dbueso@suse.de> Acked-by:
Josh Poimboeuf <jpoimboe@redhat.com> Acked-by:
Mel Gorman <mgorman@techsingularity.net> Tested-by:
-
- 22 Jul, 2020 8 commits
-
-
Alexey Budankov authored
Implement handling of 'enable' and 'disable' control commands coming from control file descriptor. If poll event splits initiated timeout interval then the reminder is calculated and still waited in the following evlist__poll() call. Committer testing: The testing instructions came in the cover letter, here I'll extract the parts that are needed to test this specific patch, so that we don't introduce bisection regressions by testing only the patch series as a whole: <FILL IN THE TEST INSTRUCTIONS> Signed-off-by:
Alexey Budankov <alexey.budankov@linux.intel.com> Acked-by:
Jiri Olsa <jolsa@redhat.com> Acked-by:
Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lore.kernel.org/lkml/3cb8a826-145f-81f4-fcb2-fa20045c6957@linux.intel.comSigned-off-by:
-
Alexey Budankov authored
Extend -D,--delay option with -1 value to start monitoring with events disabled to be enabled later by enable command provided via control file descriptor. Signed-off-by:
-
Alexey Budankov authored
Consolidate event dispatching loops for fork, attach and system wide monitoring use cases into common dispatch_events() function. Signed-off-by:
-
Alexey Budankov authored
Factor out body of event handling loop for fork case reusing handle_interval() function. Signed-off-by:
-
Alexey Budankov authored
Check for target existence in loop control statement jointly external asynchronous 'done' signal. Signed-off-by:
-
Alexey Budankov authored
Introduce handle_interval() function that factors out body of event handling loop for attach and system wide monitoring use cases. Signed-off-by:
-
Alexey Budankov authored
Implement functions of initialization, finalization and processing of control command messages coming from control file descriptors. Allocate control file descriptor as descriptor at struct pollfd object of evsel_list for atomic poll() operation. Signed-off-by:
-
Alexey Budankov authored
Define and initialize control file descriptors. Signed-off-by:
-
- 21 Jul, 2020 3 commits
-
-
Alexey Budankov authored
Avoid counting of struct pollfd *entries objects with fdarray_flag__nonfilterable flag by fdarray__filter(). Nonfilterable objects are still processed if requested revents have been signaled for them. Signed-off-by:
-
Alexey Budankov authored
Store flags per struct pollfd *entries object in a bitmap of int size. Implement fdarray_flag__nonfilterable flag to skip object from counting by fdarray__filter(). Fixed fdarray test issue reported by kernel test robot. Reported-by:
kernel test robot <rong.a.chen@intel.com> Signed-off-by:
-
Alexey Budankov authored
Avoid moving of fds by fdarray__filter() so fds indices returned by fdarray__add() can be used for access and processing of objects at struct pollfd *entries. Signed-off-by:
-
- 17 Jul, 2020 6 commits
-
-
Jiri Olsa authored
Add 'struct expr_id_data' to keep an expr value instead of just a simple double pointer, so we can store more data for ID in the following changes. Signed-off-by:
-
Jiri Olsa authored
Rename expr__add_id() to expr__add_val() so we can use expr__add_id() to actually add just the id without any value in following changes. There's no functional change. Signed-off-by:
-
Masami Hiramatsu authored
Warn if the probe target function is a GNU indirect function (GNU_IFUNC) because it may not be what the user wants to probe. The GNU indirect function ( https://sourceware.org/glibc/wiki/GNU_IFUNC ) is the dynamic symbol solved at runtime. An IFUNC function is a selector which is invoked from the ELF loader, but the symbol address of the function which will be modified by the IFUNC is the same as the IFUNC in the symbol table. This can confuse users trying to probe such functions. For example, memcpy is an IFUNC. probe_libc:memcpy (on __new_memcpy_ifunc@x86_64/multiarch/memcpy.c in /usr/lib64/libc-2.30.so) the probe is put on an IFUNC. perf 1742 [000] 26201.715632: probe_libc:memcpy: (7fdaa53824c0) 7fdaa53824c0 __new_memcpy_ifunc+0x0 (inlined) 7fdaa5d4a980 elf_machine_rela+0x6c0 (inlined) 7fdaa5d4a980 elf_dynamic_do_Rela+0x6c0 (inlined) 7fdaa5d4a980 _dl_relocate_object+0x6c0 (/usr/lib64/ld-2.30.so) 7fdaa5d42155 dl_main+0x1cc5 (/usr/lib64/ld-2.30.so) 7fdaa5d5831a _dl_sysdep_start+0x54a (/usr/lib64/ld-2.30.so) 7fdaa5d3ffeb _dl_start_final+0x25b (inlined) 7fdaa5d3ffeb _dl_start+0x25b (/usr/lib64/ld-2.30.so) 7fdaa5d3f117 .annobin_rtld.c+0x7 (inlined) And the event is invoked from the ELF loader instead of the target program's main code. Moreover, at this moment, we can not probe on the function which will be selected by the IFUNC, because it is determined at runtime. But uprobe will be prepared before running the target binary. Thus, I decided to warn user when 'perf probe' detects that the probe point is on an GNU IFUNC symbol. Someone who wants to probe an IFUNC symbol to debug the IFUNC function can ignore this warning. Committer notes: I.e., this warning will be emitted if the probe point is an IFUNC: "Warning: The probe function (%s) is a GNU indirect function.\n" "Consider identifying the final function used at run time and set the probe directly on that.\n" Complete set of steps: # readelf -sW /lib64/libc-2.29.so | grep IFUNC | tail 22196: 0000000000109a80 183 IFUNC GLOBAL DEFAULT 14 __memcpy_chk 22214: 00000000000b7d90 191 IFUNC GLOBAL DEFAULT 14 __gettimeofday 22336: 000000000008b690 60 IFUNC GLOBAL DEFAULT 14 memchr 22350: 000000000008b9b0 89 IFUNC GLOBAL DEFAULT 14 __stpcpy 22420: 000000000008bb10 76 IFUNC GLOBAL DEFAULT 14 __strcasecmp_l 22582: 000000000008a970 60 IFUNC GLOBAL DEFAULT 14 strlen 22585: 00000000000a54d0 92 IFUNC WEAK DEFAULT 14 wmemset 22600: 000000000010b030 92 IFUNC GLOBAL DEFAULT 14 __wmemset_chk 22618: 000000000008b8a0 183 IFUNC GLOBAL DEFAULT 14 __mempcpy 22675: 000000000008ba70 76 IFUNC WEAK DEFAULT 14 strcasecmp # # perf probe -x /lib64/libc-2.29.so strlen Warning: The probe function (strlen) is a GNU indirect function. Consider identifying the final function used at run time and set the probe directly on that. Added new event: probe_libc:strlen (on strlen in /usr/lib64/libc-2.29.so) You can now use it in all perf tools, such as: perf record -e probe_libc:strlen -aR sleep 1 # Reported-by:
Andi Kleen <andi@firstfloor.org> Signed-off-by:
Masami Hiramatsu <mhiramat@kernel.org> Tested-by:
-
Masami Hiramatsu authored
Fix the memory leakage in debuginfo__find_trace_events() when the probe point is not found in the debuginfo. If there is no probe point found in the debuginfo, debuginfo__find_probes() will NOT return -ENOENT, but 0. Thus the caller of debuginfo__find_probes() must check the tf.ntevs and release the allocated memory for the array of struct probe_trace_event. The current code releases the memory only if the debuginfo__find_probes() hits an error but not checks tf.ntevs. In the result, the memory allocated on *tevs are not released if tf.ntevs == 0. This fixes the memory leakage by checking tf.ntevs == 0 in addition to ret < 0. Fixes: ff741783 ("perf probe: Introduce debuginfo to encapsulate dwarf information") Signed-off-by:
Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: stable@vger.kernel.org Link: http://lore.kernel.org/lkml/159438668346.62703.10887420400718492503.stgit@devnote2Signed-off-by:
-
Masami Hiramatsu authored
Fix a wrong "variable not found" warning when the probe point is not found in the debuginfo. Since the debuginfo__find_probes() can return 0 even if it does not find given probe point in the debuginfo, fill_empty_trace_arg() can be called with tf.ntevs == 0 and it can emit a wrong warning. To fix this, reject ntevs == 0 in fill_empty_trace_arg(). E.g. without this patch; # perf probe -x /lib64/libc-2.30.so -a "memcpy arg1=%di" Failed to find the location of the '%di' variable at this address. Perhaps it has been optimized out. Use -V with the --range option to show '%di' location range. Added new events: probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) You can now use it in all perf tools, such as: perf record -e probe_libc:memcpy -aR sleep 1 With this; # perf probe -x /lib64/libc-2.30.so -a "memcpy arg1=%di" Added new events: probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) You can now use it in all perf tools, such as: perf record -e probe_libc:memcpy -aR sleep 1 Fixes: cb402730 ("perf probe: Trace a magic number if variable is not found") Reported-by:Andi Kleen <ak@linux.intel.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: stable@vger.kernel.org Link: http://lore.kernel.org/lkml/159438667364.62703.2200642186798763202.stgit@devnote2Signed-off-by:
-
Masami Hiramatsu authored
There is a case that several same-name symbols points to the same address. In that case, 'perf probe' returns an error. E.g. # perf probe -x /lib64/libc-2.30.so -v -a "memcpy arg1=%di" probe-definition(0): memcpy arg1=%di symbol:memcpy file:(null) line:0 offset:0 return:0 lazy:(null) parsing arg: arg1=%di into name:arg1 %di 1 arguments symbol:setjmp file:(null) line:0 offset:0 return:0 lazy:(null) symbol:longjmp file:(null) line:0 offset:0 return:0 lazy:(null) symbol:longjmp_target file:(null) line:0 offset:0 return:0 lazy:(null) symbol:lll_lock_wait_private file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_arena_max file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_arena_test file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_tunable_tcache_max_bytes file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_tunable_tcache_count file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_tunable_tcache_unsorted_limit file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_trim_threshold file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_top_pad file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_mmap_threshold file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_mmap_max file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_perturb file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_mxfast file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_heap_new file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_arena_reuse_free_list file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_arena_reuse file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_arena_reuse_wait file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_arena_new file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_arena_retry file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_sbrk_less file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_heap_free file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_heap_less file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_tcache_double_free file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_heap_more file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_sbrk_more file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_malloc_retry file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_memalign_retry file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt_free_dyn_thresholds file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_realloc_retry file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_calloc_retry file:(null) line:0 offset:0 return:0 lazy:(null) symbol:memory_mallopt file:(null) line:0 offset:0 return:0 lazy:(null) Open Debuginfo file: /usr/lib/debug/usr/lib64/libc-2.30.so.debug Try to find probe point from debuginfo. Opening /sys/kernel/debug/tracing//README write=0 Failed to find the location of the '%di' variable at this address. Perhaps it has been optimized out. Use -V with the --range option to show '%di' location range. An error occurred in debuginfo analysis (-2). Trying to use symbols. Opening /sys/kernel/debug/tracing//uprobe_events write=1 Writing event: p:probe_libc/memcpy /usr/lib64/libc-2.30.so:0x914c0 arg1=%di Writing event: p:probe_libc/memcpy /usr/lib64/libc-2.30.so:0x914c0 arg1=%di Failed to write event: File exists Error: Failed to add events. Reason: File exists (Code: -17) You can see that perf tried to write completely the same probe definition twice, which caused an error. To fix this issue, check the symbol list and drop duplicated symbols (which has the same symbol name and address) from it. With this patch: # perf probe -x /lib64/libc-2.30.so -a "memcpy arg1=%di" Failed to find the location of the '%di' variable at this address. Perhaps it has been optimized out. Use -V with the --range option to show '%di' location range. Added new events: probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) probe_libc:memcpy (on memcpy in /usr/lib64/libc-2.30.so with arg1=%di) You can now use it in all perf tools, such as: perf record -e probe_libc:memcpy -aR sleep 1 Committer notes: Fix this build error on 32-bit arches by using PRIx64 for symbol->start, that is an u64: In file included from util/probe-event.c:27: util/probe-event.c: In function 'find_probe_trace_events_from_map': util/probe-event.c:2978:14: error: format '%lx' expects argument of type 'long unsigned int', but argument 5 has type 'u64' {aka 'long long unsigned int'} [-Werror=format=] pr_debug("Found duplicated symbol %s @ %lx\n", ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ util/debug.h:17:21: note: in definition of macro 'pr_fmt' #define pr_fmt(fmt) fmt ^~~ util/probe-event.c:2978:5: note: in expansion of macro 'pr_debug' pr_debug("Found duplicated symbol %s @ %lx\n", ^~~~~~~~ Reported-by:
-
- 10 Jul, 2020 7 commits
-
-
Ian Rogers authored
'perf kmem' has an input file option but current an output file option fails: $ sudo perf kmem record -o /tmp/p.data sleep 1 Error: unknown switch `o' Usage: perf kmem [<options>] {record|stat} -f, --force don't complain, do it -i, --input <file> input file name -l, --line <num> show n lines -s, --sort <key[,key2...]> sort by keys: ptr, callsite, bytes, hit, pingpong, frag, page, order, mig> -v, --verbose be more verbose (show symbol address, etc) --alloc show per-allocation statistics --caller show per-callsite statistics --live Show live page stat --page Analyze page allocator --raw-ip show raw ip instead of symbol --slab Analyze slab allocator --time <str> Time span of interest (start,stop) 'perf sched' is similar in implementation and avoids the problem by passing additional arguments to 'perf record'. This change makes 'perf kmem' parse command line options consistently with 'perf sched', although neither actually list that -o is a supported option. Signed-off-by: -
Ian Rogers authored
Setting the parse_events_error directly doesn't increment num_errors causing the error message not to be displayed. Use the parse_events__handle_error function that sets num_errors and handle multiple errors. Committer notes: Ian provided a before/after upon request: Before: $ /tmp/perf/perf record -e /tmp/perf/util/parse-events.o 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 event After: $ /tmp/perf/perf record -e /tmp/perf/util/parse-events.o event syntax error: '/tmp/perf/util/parse-events.o' \___ Failed to load /tmp/perf/util/parse-events.o: BPF object format invalid (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: -
Adrian Hunter authored
It is generally more useful to show the symbol with an address. In this case, the print function requires the 'machine' which means changing callers to provide it as a parameter. It is optional because most events do not need it and the callers that matter can provide it. Committer notes: Made 'union perf_event' continue to be the first parameter to the perf_event__fprintf() and perf_event__fprintf_text_poke() events. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Borislav Petkov <bp@alien8.de> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: x86@kernel.org Link: http://lore.kernel.org/lkml/20200512121922.8997-16-adrian.hunter@intel.comSigned-off-by:
-
Adrian Hunter authored
Consistent with other new events, add an option to perf script to display text poke events and ksymbol events. Both text poke events and ksymbol events are displayed because some text pokes (e.g. ftrace trampolines) have corresponding ksymbol events. Signed-off-by:
-
Adrian Hunter authored
Select text poke events when available and the kernel is being traced. Process text poke events to invalidate entries in Intel PT's instruction cache. Example: The example requires kernel config: CONFIG_PROC_SYSCTL=y CONFIG_SCHED_DEBUG=y CONFIG_SCHEDSTATS=y Before: # perf record -o perf.data.before --kcore -a -e intel_pt//k -m,64M & # cat /proc/sys/kernel/sched_schedstats 0 # echo 1 > /proc/sys/kernel/sched_schedstats # cat /proc/sys/kernel/sched_schedstats 1 # echo 0 > /proc/sys/kernel/sched_schedstats # cat /proc/sys/kernel/sched_schedstats 0 # kill %1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 3.341 MB perf.data.before ] [1]+ Terminated perf record -o perf.data.before --kcore -a -e intel_pt//k -m,64M # perf script -i perf.data.before --itrace=e >/dev/null Warning: 474 instruction trace errors After: # perf record -o perf.data.after --kcore -a -e intel_pt//k -m,64M & # cat /proc/sys/kernel/sched_schedstats 0 # echo 1 > /proc/sys/kernel/sched_schedstats # cat /proc/sys/kernel/sched_schedstats 1 # echo 0 > /proc/sys/kernel/sched_schedstats # cat /proc/sys/kernel/sched_schedstats 0 # kill %1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 2.646 MB perf.data.after ] [1]+ Terminated perf record -o perf.data.after --kcore -a -e intel_pt//k -m,64M # perf script -i perf.data.after --itrace=e >/dev/null Example: The example requires kernel config: # CONFIG_FUNCTION_TRACER is not set Before: # perf record --kcore -m,64M -o t1 -a -e intel_pt//k & # perf probe __schedule Added new event: probe:__schedule (on __schedule) You can now use it in all perf tools, such as: perf record -e probe:__schedule -aR sleep 1 # perf record -e probe:__schedule -aR sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.026 MB perf.data (68 samples) ] # perf probe -d probe:__schedule Removed event: probe:__schedule # kill %1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 41.268 MB t1 ] [1]+ Terminated perf record --kcore -m,64M -o t1 -a -e intel_pt//k # perf script -i t1 --itrace=e >/dev/null Warning: 207 instruction trace errors After: # perf record --kcore -m,64M -o t1 -a -e intel_pt//k & # perf probe __schedule Added new event: probe:__schedule (on __schedule) You can now use it in all perf tools, such as: perf record -e probe:__schedule -aR sleep 1 # perf record -e probe:__schedule -aR sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.028 MB perf.data (107 samples) ] # perf probe -d probe:__schedule Removed event: probe:__schedule # kill %1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 39.978 MB t1 ] [1]+ Terminated perf record --kcore -m,64M -o t1 -a -e intel_pt//k # perf script -i t1 --itrace=e >/dev/null # perf script -i t1 --no-itrace -D | grep 'POKE\|KSYMBOL' 6 565303693547 0x291f18 [0x50]: PERF_RECORD_KSYMBOL addr ffffffffc027a000 len 4096 type 2 flags 0x0 name kprobe_insn_page 6 565303697010 0x291f68 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffc027a000 old len 0 new len 6 6 565303838278 0x291fa8 [0x50]: PERF_RECORD_KSYMBOL addr ffffffffc027c000 len 4096 type 2 flags 0x0 name kprobe_optinsn_page 6 565303848286 0x291ff8 [0xa0]: PERF_RECORD_TEXT_POKE addr 0xffffffffc027c000 old len 0 new len 106 6 565369336743 0x292af8 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffff88ab8890 old len 5 new len 5 7 566434327704 0x217c208 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffff88ab8890 old len 5 new len 5 6 566456313475 0x293198 [0xa0]: PERF_RECORD_TEXT_POKE addr 0xffffffffc027c000 old len 106 new len 0 6 566456314935 0x293238 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffc027a000 old len 6 new len 0 Example: The example requires kernel config: CONFIG_FUNCTION_TRACER=y Before: # perf record --kcore -m,64M -o t1 -a -e intel_pt//k & # perf probe __kmalloc Added new event: probe:__kmalloc (on __kmalloc) You can now use it in all perf tools, such as: perf record -e probe:__kmalloc -aR sleep 1 # perf record -e probe:__kmalloc -aR sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.022 MB perf.data (6 samples) ] # perf probe -d probe:__kmalloc Removed event: probe:__kmalloc # kill %1 [ perf record: Woken up 2 times to write data ] [ perf record: Captured and wrote 43.850 MB t1 ] [1]+ Terminated perf record --kcore -m,64M -o t1 -a -e intel_pt//k # perf script -i t1 --itrace=e >/dev/null Warning: 8 instruction trace errors After: # perf record --kcore -m,64M -o t1 -a -e intel_pt//k & # perf probe __kmalloc Added new event: probe:__kmalloc (on __kmalloc) You can now use it in all perf tools, such as: perf record -e probe:__kmalloc -aR sleep 1 # perf record -e probe:__kmalloc -aR sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.037 MB perf.data (206 samples) ] # perf probe -d probe:__kmalloc Removed event: probe:__kmalloc # kill %1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 41.442 MB t1 ] [1]+ Terminated perf record --kcore -m,64M -o t1 -a -e intel_pt//k # perf script -i t1 --itrace=e >/dev/null # perf script -i t1 --no-itrace -D | grep 'POKE\|KSYMBOL' 5 312216133258 0x8bafe0 [0x50]: PERF_RECORD_KSYMBOL addr ffffffffc0360000 len 415 type 2 flags 0x0 name ftrace_trampoline 5 312216133494 0x8bb030 [0x1d8]: PERF_RECORD_TEXT_POKE addr 0xffffffffc0360000 old len 0 new len 415 5 312216229563 0x8bb208 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac6016f5 old len 5 new len 5 5 312216239063 0x8bb248 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac601803 old len 5 new len 5 5 312216727230 0x8bb288 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffabbea190 old len 5 new len 5 5 312216739322 0x8bb2c8 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac6016f5 old len 5 new len 5 5 312216748321 0x8bb308 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac601803 old len 5 new len 5 7 313287163462 0x2817430 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac6016f5 old len 5 new len 5 7 313287174890 0x2817470 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac601803 old len 5 new len 5 7 313287818979 0x28174b0 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffabbea190 old len 5 new len 5 7 313287829357 0x28174f0 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac6016f5 old len 5 new len 5 7 313287841246 0x2817530 [0x40]: PERF_RECORD_TEXT_POKE addr 0xffffffffac601803 old len 5 new len 5 Signed-off-by: -
Adrian Hunter authored
PERF_RECORD_KSYMBOL_TYPE_OOL marks an executable page. Create a map backed only by memory, which will be populated as necessary by text poke events. Committer notes: From the patch: OOL stands for "Out of line" code such as kprobe-replaced instructions or optimized kprobes or ftrace trampolines. Signed-off-by:
-
Adrian Hunter authored
Add processing for PERF_RECORD_TEXT_POKE events. When a text poke event is processed, then the kernel dso data cache is updated with the poked bytes. Signed-off-by:
-
- 09 Jul, 2020 1 commit
-
-
Numfor Mbiziwo-Tiapo authored
Our local MSAN (Memory Sanitizer) build of perf throws a warning that comes from the "dso__disassemble_filename" function in "tools/perf/util/annotate.c" when running perf record. The warning stems from the call to readlink, in which "build_id_path" was being read into "linkname". Since readlink does not null terminate, an uninitialized memory access would later occur when "linkname" is passed into the strstr function. This is simply fixed by null-terminating "linkname" after the call to readlink. To reproduce this warning, build perf by running: $ make -C tools/perf CLANG=1 CC=clang EXTRA_CFLAGS="-fsanitize=memory -fsanitize-memory-track-origins" (Additionally, llvm might have to be installed and clang might have to be specified as the compiler - export CC=/usr/bin/clang) Then running: tools/perf/perf record -o - ls / | tools/perf/perf --no-pager annotate -i - --stdio Please see the cover letter for why false positive warnings may be generated. Signed-off-by:
Numfor Mbiziwo-Tiapo <nums@google.com> Acked-by:
-
- 08 Jul, 2020 1 commit
-
-
Steve MacLean authored
**perf-<pid>.map and jit-<pid>.dump designs: When a JIT generates code to be executed, it must allocate memory and mark it executable using an mmap call. *** perf-<pid>.map design The perf-<pid>.map assumes that any sample recorded in an anonymous memory page is JIT code. It then tries to resolve the symbol name by looking at the process' perf-<pid>.map. *** jit-<pid>.dump design The jit-<pid>.dump mechanism takes a different approach. It requires a JIT to write a `<path>/jit-<pid>.dump` file. This file must also be mmapped so that perf inject -jit can find the file. The JIT must also add JIT_CODE_LOAD records for any functions it generates. The records are timestamped using a clock which can be correlated to the perf record clock. After perf record, the `perf inject -jit` pass parses the recording looking for a `<path>/jit-<pid>.dump` file. When it finds the file, it parses it and for each JIT_CODE_LOAD record: * creates an elf file `<path>/jitted-<pid>-<code_index>.so * injects a new mmap record mapping the new elf file into the process. *** Coexistence design The kernel and perf support both of these mechanisms. We need to make sure perf works on an app supporting either or both of these mechanisms. Both designs rely on mmap records to determine how to resolve an ip address. The mmap records of both techniques by definition overlap. When the JIT compiles a method, it must: * allocate memory (mmap) * add execution privilege (mprotect or mmap. either will generate an mmap event form the kernel to perf) * compile code into memory * add a function record to perf-<pid>.map and/or jit-<pid>.dump Because the jit-<pid>.dump mechanism supports greater capabilities, perf prefers the symbols from jit-<pid>.dump. It implements this based on timestamp ordering of events. There is an implicit ASSUMPTION that the JIT_CODE_LOAD record timestamp will be after the // anon mmap event that was generated during memory allocation or adding the execution privilege setting. *** Problems with the ASSUMPTION The ASSUMPTION made in the Coexistence design section above is violated in the following scenario. *** Scenario While a JIT is jitting code it will eventually need to commit more pages and change these pages to executable permissions. Typically the JIT will want these collocated to minimize branch displacements. The kernel will coalesce these anonymous mapping with identical permissions before sending an MMAP event for the new pages. The address range of the new mmap will not be just the most recently mmap pages. It will include the entire coalesced mmap region. See mm/mmap.c unsigned long mmap_region(struct file *file, unsigned long addr, unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, struct list_head *uf) { ... /* * Can we just expand an old mapping? */ ... perf_event_mmap(vma); ... } *** Symptoms The coalesced // anon mmap event will be timestamped after the JIT_CODE_LOAD records. This means it will be used as the most recent mapping for that entire address range. For remaining events it will look at the inferior perf-<pid>.map for symbols. If both mechanisms are supported, the symbol will appear twice with different module names. This causes weird behavior in reporting. If only jit-<pid>.dump is supported, the symbol will no longer be resolved. ** Implemented solution This patch solves the issue by removing // anon mmap events for any process which has a valid jit-<pid>.dump file. It tracks on a per process basis to handle the case where some running apps support jit-<pid>.dump, but some only support perf-<pid>.map. It adds new assumptions: * // anon mmap events are only required for perf-<pid>.map support. * An app that uses jit-<pid>.dump, no longer needs perf-<pid>.map support. It assumes that any perf-<pid>.map info is inferior. *** Details Use thread->priv to store whether a jitdump file has been processed During "perf inject --jit", discard "//anon*" mmap events for any pid which has sucessfully processed a jitdump file. ** Testing: // jitdump case perf record <app with jitdump> perf inject --jit --input perf.data --output perfjit.data // verify mmap "//anon" events present initially perf script --input perf.data --show-mmap-events | grep '//anon' // verify mmap "//anon" events removed perf script --input perfjit.data --show-mmap-events | grep '//anon' // no jitdump case perf record <app without jitdump> perf inject --jit --input perf.data --output perfjit.data // verify mmap "//anon" events present initially perf script --input perf.data --show-mmap-events | grep '//anon' // verify mmap "//anon" events not removed perf script --input perfjit.data --show-mmap-events | grep '//anon' ** Repro: This issue was discovered while testing the initial CoreCLR jitdump implementation. https://github.com/dotnet/coreclr/pull/26897. ** Alternate solutions considered These were also briefly considered: * Change kernel to not coalesce mmap regions. * Change kernel reporting of coalesced mmap regions to perf. Only include newly mapped memory. * Only strip parts of // anon mmap events overlapping existing jitted-<pid>-<code_index>.so mmap events. Signed-off-by:Steve MacLean <Steve.MacLean@Microsoft.com> Acked-by:
-
