perf, tools: Add support for skipping itrace instructions
PATCH 1/3 perf, tools: Add support for skipping itrace instructions
From: Andi Kleen <[email protected]>
When using perf script to look at PT traces it is often useful to ignore the initialization code at the beginning.
On larger traces which may have many millions of instructions in initialization code doing that in a pipeline can be very slow, with perf script spending a lot of CPU time calling printf and writing data.
This patch adds an extension to the --itrace argument that skips 'n' events (instructions, branches or transactions) at the beginning. This is much more efficient.
Cc Adrian Hunter <[email protected]>
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/Documentation/intel-pt.txt | 7 +++++++ tools/perf/util/auxtrace.c | 7 +++++++ tools/perf/util/auxtrace.h | 2 ++ tools/perf/util/intel-pt.c | 14 ++++++++++++-- 4 files changed, 28 insertions(+), 2 deletions(-)
tools/perf/Documentation/intel-pt.txt +7 -0
@@ -672,6 +672,7 @@ The letters are: d create a debug log g synthesize a call chain (use with i or x) l synthesize last branch entries (use with i or x) + s skip initial number of events "Instructions" events look like they were recorded by "perf record -e instructions".
@@ -730,6 +731,12 @@ from one sample to the next. To disable trace decoding entirely, use the option --no-itrace. +It is also possible to skip events generated (instructions, branches, transactions) +at the beginning. This is useful to ignore initialization code. + + --itrace=i0nss1000000 + +skips the first million instructions. dump option -----------
tools/perf/util/auxtrace.c +7 -0
@@ -939,6 +939,7 @@ void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts) synth_opts->period =PERF_ITRACE_DEFAULT_PERIOD ; synth_opts->callchain_sz =PERF_ITRACE_DEFAULT_CALLCHAIN_SZ ; synth_opts->last_branch_sz =PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ ; + synth_opts->initial_skip =0 ; }/*
@@ -1063,6 +1064,12 @@ int itrace_parse_synth_opts(const struct option *opt, const char *str, synth_opts->last_branch_sz = val; }break ; +case 's' : + synth_opts->initial_skip =strtoul (p, &endptr,10 ); +if (p == endptr) + goto out_err; + p = endptr; +break ;case ' ' :case ',' :break ;
tools/perf/util/auxtrace.h +2 -0
@@ -68,6 +68,7 @@ enum itrace_period_type {* @last_branch_sz: branch context size * @period: 'instructions' events period * @period_type: 'instructions' events period type +* @initial_skip: skip N events at the beginning. */ struct itrace_synth_opts {bool set;
@@ -86,6 +87,7 @@ struct itrace_synth_opts {unsigned int last_branch_sz;unsigned long long period;enum itrace_period_type period_type; +unsigned long initial_skip; };/**
tools/perf/util/intel-pt.c +12 -2
@@ -100,6 +100,8 @@ struct intel_pt {u64 cyc_bit;u64 noretcomp_bit;unsigned max_non_turbo_ratio; + +unsigned long num_events; };enum switch_state {
@@ -972,6 +974,10 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)if (pt->branches_filter && !(pt->branches_filter & ptq->flags))return 0 ; +if (pt->synth_opts.initial_skip && + pt->num_events++ >= pt->synth_opts.initial_skip) +return 0 ; + event->sample.header.type =PERF_RECORD_SAMPLE ; event->sample.header.misc =PERF_RECORD_MISC_USER ; event->sample.header.size =sizeof (struct perf_event_header );
@@ -1196,14 +1202,18 @@ static int intel_pt_sample(struct intel_pt_queue *ptq) ptq->have_sample = false;if (pt->sample_instructions && - (state->type &INTEL_PT_INSTRUCTION )) { + (state->type &INTEL_PT_INSTRUCTION ) && + (!pt->synth_opts.initial_skip || + pt->num_events++ >= pt->synth_opts.initial_skip)) { err =intel_pt_synth_instruction_sample (ptq);if (err)return err; }if (pt->sample_transactions && - (state->type &INTEL_PT_TRANSACTION )) { + (state->type &INTEL_PT_TRANSACTION ) && + (!pt->synth_opts.initial_skip || + pt->num_events++ >= pt->synth_opts.initial_skip)) { err =intel_pt_synth_transaction_sample (ptq);if (err)return err;
Signature · 1 line
2.5.0
PATCH 2/3 perf, tools: Add support for probing for udev86
From: Andi Kleen <[email protected]>
Add support for autoprobing for the udev86 disassembler library.
Signed-off-by Andi Kleen <[email protected]>
--- tools/build/Makefile.feature | 6 ++++-- tools/build/feature/Makefile | 8 ++++++-- tools/build/feature/test-all.c | 5 +++++ tools/perf/config/Makefile | 5 +++++ 4 files changed, 20 insertions(+), 4 deletions(-)
tools/build/Makefile.feature +4 -2
@@ -54,7 +54,8 @@ FEATURE_TESTS ?= \ zlib \ lzma \ get_cpuid \ - bpf + bpf \ + udis86 FEATURE_DISPLAY ?= \ dwarf \
@@ -73,7 +74,8 @@ FEATURE_DISPLAY ?= \ zlib \ lzma \ get_cpuid \ - bpf + bpf \ + udis86 # Set FEATURE_CHECK_(C|LD)FLAGS-all for all FEATURE_TESTS features. # If in the future we need per-feature checks/flags for features not
tools/build/feature/Makefile +6 -2
@@ -35,7 +35,8 @@ FILES= \ test-zlib.bin \ test-lzma.bin \ test-bpf.bin \ - test-get_cpuid.bin + test-get_cpuid.bin \ + test-udis86.bin FILES := $(addprefix $(OUTPUT),$(FILES))
@@ -50,7 +51,7 @@ __BUILD = $(CC) $(CFLAGS) -Wall -Werror -o $@ $(patsubst %.bin,%.c,$(@F)) $(LDFL ############################### $(OUTPUT)test-all.bin: - $(BUILD) -fstack-protector-all -O2 -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma -lelf -laudit -I/usr/include/slang -lslang $(shell $(PKG_CONFIG) --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl -lz -llzma + $(BUILD) -fstack-protector-all -O2 -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma -lelf -laudit -I/usr/include/slang -lslang $(shell $(PKG_CONFIG) --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl -lz -llzma -ludis86 $(OUTPUT)test-hello.bin: $(BUILD)
@@ -96,6 +97,9 @@ $(OUTPUT)test-numa_num_possible_cpus.bin: $(OUTPUT)test-libunwind.bin: $(BUILD) -lelf +$(OUTPUT)test-udis86.bin: + $(BUILD) -ludis86 + $(OUTPUT)test-libunwind-debug-frame.bin: $(BUILD) -lelf
tools/build/feature/test-all.c +5 -0
@@ -129,6 +129,10 @@# include "test-bpf.c" #undef main +#define main main_test_udis86 +# include "test-udis86.c" +#endif +int main (int argc,char *argv[]) {main_test_libpython ();
@@ -158,6 +162,7 @@ int main(int argc, char *argv[])main_test_lzma ();main_test_get_cpuid ();main_test_bpf (); +main_test_udis86 ();return 0 ; }
tools/perf/config/Makefile +5 -0
@@ -576,6 +576,11 @@ ifneq ($(filter -lbfd,$(EXTLIBS)),) CFLAGS += -DHAVE_LIBBFD_SUPPORT endif +ifeq ($(feature-udis86), 1) + CFLAGS += -DHAVE_UDIS86 + EXTLIBS += -ludis86 +endif + ifndef NO_ZLIB ifeq ($(feature-zlib), 1) CFLAGS += -DHAVE_ZLIB_SUPPORT
Signature · 1 line
2.5.0
PATCH 3/3 perf, tools, script: Add support for printing assembler
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
When dumping PT traces with perf script it is very useful to see the assembler for each sample, so that it is easily possible to follow the control flow.
As using objdump is difficult and inefficient from perf script this patch uses the udis86 library to implement assembler output. The library can be downloaded from http://udis86.sourceforge.net/
The library is probed as an external dependency in the usual way. Then perf script calls into it when needed, and handles callbacks to resolve symbols.
% perf record -e intel_pt//u true
% perf script -F sym,symoff,ip,asm --itrace=i0ns | head
7fc7188b4190 _start+0x0 mov %rsp, %rdi
7fc7188b4193 _start+0x3 call _dl_start
7fc7188b7710 _dl_start+0x0 push %rbp
7fc7188b7711 _dl_start+0x1 mov %rsp, %rbp
7fc7188b7714 _dl_start+0x4 push %r15
7fc7188b7716 _dl_start+0x6 push %r14
7fc7188b7718 _dl_start+0x8 push %r13
7fc7188b771a _dl_start+0xa push %r12
7fc7188b771c _dl_start+0xc mov %rdi, %r12
7fc7188b771f _dl_start+0xf push %rbxCurrent issues:
- Some jump references do not get resolved to symbols.
- udis86 release does not support STAC/CLAC, which are used in the kernel, but there is a pending patch for it.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/Documentation/perf-script.txt | 4 +- tools/perf/builtin-script.c | 108 +++++++++++++++++++++++++++++-- 2 files changed, 106 insertions(+), 6 deletions(-)
tools/perf/Documentation/perf-script.txt +3 -1
@@ -116,7 +116,7 @@ OPTIONS --fields:: Comma separated list of fields to print. Options are: comm, tid, pid, time, cpu, event, trace, ip, sym, dso, addr, symoff, - srcline, period, iregs, brstack, brstacksym, flags. + srcline, period, iregs, brstack, brstacksym, flags, asm. Field list can be prepended with the type, trace, sw or hw, to indicate to which event type the field list applies. e.g., -f sw:comm,tid,time,ip,sym and -f trace:time,cpu,trace
@@ -185,6 +185,8 @@ OPTIONS The brstacksym is identical to brstack, except that the FROM and TO addresses are printed in a symbolic form if possible. + When asm is specified the assembler instruction of each sample is printed in disassembled form. + -k:: --vmlinux=<file>:: vmlinux pathname
tools/perf/builtin-script.c +103 -5
173 lines not on this page · raw message
Signature · 1 line
2.5.0
Adrian Hunter
reviewerQuoted 38 lines · Andi Kleen
From: Andi Kleen <[email protected]>
When using perf script to look at PT traces it is often useful to ignore the initialization code at the beginning.
On larger traces which may have many millions of instructions in initialization code doing that in a pipeline can be very slow, with perf script spending a lot of CPU time calling printf and writing data.
This patch adds an extension to the --itrace argument that skips 'n' events (instructions, branches or transactions) at the beginning. This is much more efficient.
Cc Adrian Hunter <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/Documentation/intel-pt.txt | 7 +++++++ tools/perf/util/auxtrace.c | 7 +++++++ tools/perf/util/auxtrace.h | 2 ++ tools/perf/util/intel-pt.c | 14 ++++++++++++-- 4 files changed, 28 insertions(+), 2 deletions(-)tools/perf/Documentation/intel-pt.txt +1
-0@@ -672,6 +672,7 @@ The letters are: d create a debug log g synthesize a call chain (use with i or x) l synthesize last branch entries (use with i or x) + s skip initial number of events "Instructions" events look like they were recorded by "perf record -e instructions".@@ -730,6 +731,12 @@ from one sample to the next. To disable trace decoding entirely, use the option --no-itrace.
tools/perf/Documentation/intel-pt.txt from one sample to the next. · line 742
Quoted 10 lines
tools/perf/Documentation/intel-pt.txt +6
-0+It is also possible to skip events generated (instructions, branches, transactions) +at the beginning. This is useful to ignore initialization code. + + --itrace=i0nss1000000 + +skips the first million instructions. dump option -----------
Also need to update tools/perf/Documentation/itrace.txt
Quoted 59 lines
tools/perf/util/auxtrace.c +7
-0@@ -939,6 +939,7 @@ void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts) synth_opts->period =PERF_ITRACE_DEFAULT_PERIOD ; synth_opts->callchain_sz =PERF_ITRACE_DEFAULT_CALLCHAIN_SZ ; synth_opts->last_branch_sz =PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ ; + synth_opts->initial_skip =0 ; }/* @@ -1063,6 +1064,12 @@ int itrace_parse_synth_opts(const struct option *opt, const char *str, synth_opts->last_branch_sz = val; }break ; +case 's' : + synth_opts->initial_skip =strtoul (p, &endptr,10 ); +if (p == endptr) + goto out_err; + p = endptr; +break ;case ' ' :case ',' :break ;tools/perf/util/auxtrace.h +2
-0@@ -68,6 +68,7 @@ enum itrace_period_type {* @last_branch_sz: branch context size * @period: 'instructions' events period * @period_type: 'instructions' events period type +* @initial_skip: skip N events at the beginning. */ struct itrace_synth_opts {bool set;@@ -86,6 +87,7 @@ struct itrace_synth_opts {unsigned int last_branch_sz;unsigned long long period;enum itrace_period_type period_type; +unsigned long initial_skip; };/** tools/perf/util/intel-pt.c +2
-0@@ -100,6 +100,8 @@ struct intel_pt {u64 cyc_bit;u64 noretcomp_bit;unsigned max_non_turbo_ratio; + +unsigned long num_events; };enum switch_state {
tools/perf/util/intel-pt.c static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq) · line 978
Quoted 5 lines
tools/perf/util/intel-pt.c +2
-0@@ -972,6 +974,10 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)if (pt->branches_filter && !(pt->branches_filter & ptq->flags))return 0 ; +if (pt->synth_opts.initial_skip && + pt->num_events++ >= pt->synth_opts.initial_skip)
The '>' looks the wrong way around
tools/perf/util/intel-pt.c static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq) · line 980
Quoted 2 lines
tools/perf/util/intel-pt.c +2
-0+return 0 ; +
For consistency, wouldn't it be better to do it this way for "instructions" and "transactions" too. i.e. in intel_pt_synth_instruction_sample() and intel_pt_synth_transaction_sample()
Quoted 15 lines
tools/perf/util/intel-pt.c +3
-1event->sample.header.type =PERF_RECORD_SAMPLE ; event->sample.header.misc =PERF_RECORD_MISC_USER ; event->sample.header.size =sizeof (struct perf_event_header );@@ -1196,14 +1202,18 @@ static int intel_pt_sample(struct intel_pt_queue *ptq) ptq->have_sample = false;if (pt->sample_instructions && - (state->type &INTEL_PT_INSTRUCTION )) { + (state->type &INTEL_PT_INSTRUCTION ) && + (!pt->synth_opts.initial_skip || + pt->num_events++ >= pt->synth_opts.initial_skip)) { err =intel_pt_synth_instruction_sample (ptq);if (err)return err;
tools/perf/util/intel-pt.c static int intel_pt_sample(struct intel_pt_queue *ptq) · line 1219
Quoted 10 lines
tools/perf/util/intel-pt.c +3
-1}if (pt->sample_transactions && - (state->type &INTEL_PT_TRANSACTION )) { + (state->type &INTEL_PT_TRANSACTION ) && + (!pt->synth_opts.initial_skip || + pt->num_events++ >= pt->synth_opts.initial_skip)) { err =intel_pt_synth_transaction_sample (ptq);if (err)return err;
The BTS code is almost the same, so I would suggest doing BTS as well as PT. i.e.
tools/perf/util/intel-bts.c +5 -0
@@ -66,6 +66,7 @@ struct intel_bts {u64 branches_id;size_t branches_event_size;bool synth_needs_swap; +unsigned long num_events; };struct intel_bts_queue {
@@ -275,6 +276,10 @@ static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,union perf_event event;struct perf_sample sample = { .ip =0 , }; +if (bts->synth_opts.initial_skip && + bts->num_events++ <= bts->synth_opts.initial_skip) +return 0 ; + event.sample.header.type =PERF_RECORD_SAMPLE ; event.sample.header.misc =PERF_RECORD_MISC_USER ; event.sample.header.size =sizeof (struct perf_event_header );