Support sample context in perf report
[Changes:
v4:
Address review comments.
Fix --cpu filtering.
Fix a sampling bug.
Add support for configuring custom script menu entries in perfconfig. Fix display of more samples than fit on screen.
Fix some buffer overruns in legacy code.
Add more tips
]
We currently have two ways to look at sample data in perf:
either use perf report to aggregate everything, or use perf script to look at all individual samples.
Both ways are useful. Of course aggregation is useful to quickly find the most expensive part of the code.
But sometimes a single sample is not good enough to determine the problem and we need to look at context, either through branch contexts, or other previous samples (e.g. for correlating different micro architecture events or computing metrics)
This can be done through perf script today, but it can be rather cumbersome to find the right samples to look at.
Another problem with perf report is that it aggregates the whole measurement period. But many real workloads have phases where they behave quite differently, and it is often not useful to combine them into a single histogram.
While this can be worked around with the --time option to report, it can be quite cumbersome.
This patch kit attempts to address some of these problems in perf report by making it time aware.
- It adds a new time sort key that allows perf report to separate samples from different regions. The time regions can be defined with a new --time-quantum option.
- Then it extends the perf script support in the tui record browser to allow browsing samples for different time regions from within a perf report session.
- Extends the report browser script display
to automatically select sensible defaults
based on what was recorded. For example it will
automatically show branch contexts with -b.
- Support browsing the context of individual samples. perf report can save a limited number of random samples per histogram entry with the new --samples option. Then the browser allows directly jumping to any
of the saved samples and browsing the context on the current thread or CPU.
There could be probably be done more to make perf report even better for such use cases (e.g. a real time line display), but this basic support is good enough for many useful usages.
Also available ingit://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc.git perf/streams-4
PATCH v4 1/15 perf tools: Add utility function to fetch executable
From: Andi Kleen <[email protected]>
Add a utility function to fetch executable code. Convert one user over to it. There are more places doing that, but they do significantly different actions, so they are not easy to fit into a single library function.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/util/Build | 1 + tools/perf/util/fetch.c | 28 ++++++++++++++++++++++++++++ tools/perf/util/fetch.h | 7 +++++++ tools/perf/util/intel-bts.c | 21 +++------------------ 4 files changed, 39 insertions(+), 18 deletions(-) create mode 100644 tools/perf/util/fetch.c create mode 100644 tools/perf/util/fetch.h
tools/perf/util/Build +1 -0
@@ -65,6 +65,7 @@ perf-y += trace-event-scripting.o perf-y += trace-event.o perf-y += svghelper.o perf-y += sort.o +perf-y += fetch.o perf-y += hist.o perf-y += util.o perf-y += xyarray.o
tools/perf/util/fetch.c new file +28 -0
@@ -0,0 +1,28 @@ +#include "perf.h" +#include "machine.h" +#include "thread.h" +#include "symbol.h" +#include "map.h" +#include "fetch.h" + +int fetch_exe (u64 ip,struct thread *thread,struct machine *machine, +char *buf,int len,bool *is64bit) +{ +struct addr_location al; +u8 cpumode; +long offset; + +if (machine__kernel_ip (machine, ip)) + cpumode =PERF_RECORD_MISC_KERNEL ; +else + cpumode =PERF_RECORD_MISC_USER ; +if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso) +return -1 ; +if (al.map->dso->data.status ==DSO_DATA_STATUS_ERROR ) +return -1 ; +map__load (al.map); + offset = al.map->map_ip (al.map, ip); +if (is64bit) + *is64bit = al.map->dso->is_64_bit; +return dso__data_read_offset (al.map->dso, machine, offset, (u8 *)buf, len); +}
tools/perf/util/fetch.h new file +7 -0
@@ -0,0 +1,7 @@ +#ifndef FETCH_H +#define FETCH_H 1 + +int fetch_exe (u64 ip,struct thread *thread,struct machine *machine, +char *buf,int len,bool *is64bit); + +#endif
tools/perf/util/intel-bts.c +3 -18
@@ -38,6 +38,7 @@#include "auxtrace.h" #include "intel-pt-decoder/intel-pt-insn-decoder.h" #include "intel-bts.h" +#include "fetch.h" #define MAX_TIMESTAMP (~0ULL)
@@ -328,35 +329,19 @@ static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip) {struct machine *machine = btsq->bts->machine;struct thread *thread; -struct addr_location al;unsigned char buf[INTEL_PT_INSN_BUF_SZ ];ssize_t len; -int x86_64; -uint8_t cpumode; +bool x86_64;int err =-1 ; -if (machine__kernel_ip (machine, ip)) - cpumode =PERF_RECORD_MISC_KERNEL ; -else - cpumode =PERF_RECORD_MISC_USER ; - thread =machine__find_thread (machine,-1 , btsq->tid);if (!thread)return -1 ; -if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso) - goto out_put; - - len =dso__data_read_addr (al.map->dso, al.map, machine, ip, buf, -INTEL_PT_INSN_BUF_SZ ); + len =fetch_exe (ip, thread, machine, (char *)buf,INTEL_PT_INSN_BUF_SZ , &x86_64);if (len <=0 ) goto out_put; -/* Load maps to ensure dso->is_64_bit has been updated */ -map__load (al.map); - - x86_64 = al.map->dso->is_64_bit; -if (intel_pt_get_insn (buf, len, x86_64, &btsq->intel_pt_insn)) goto out_put;
Signature · 1 line
2.20.1
Re: [PATCH v4 01/15] perf tools: Add utility function to fetch executable
Em Tue, Mar 05, 2019 at 06:47:44AM -0800, Andi Kleen escreveu:
Quoted 8 lines
From: Andi Kleen <[email protected]>
Add a utility function to fetch executable code. Convert one user over to it. There are more places doing that, but they do significantly different actions, so they are not easy to fit into a single library function.
Signed-off-by Andi Kleen <[email protected]>
Quoted 8 lines
--- tools/perf/util/Build | 1 + tools/perf/util/fetch.c | 28 ++++++++++++++++++++++++++++ tools/perf/util/fetch.h | 7 +++++++ tools/perf/util/intel-bts.c | 21 +++------------------ 4 files changed, 39 insertions(+), 18 deletions(-) create mode 100644 tools/perf/util/fetch.c create mode 100644 tools/perf/util/fetch.h
So I've made some changes and the end result is below, will fixup the following patches, holler if you find any problems:
Changes:
. No need to cast around, make 'buf' be a void pointer
. Rename it to thread__memcpy() to reflect the fact it is about copying
a chunk of memory from a thread, i.e. from its address space.
. No need to have it in a separate object file, move it to thread.[ch]
. Check the return of map__load(), the original code didn't do it, but
since we're moving this around, check that as well, could be moved to a separate patch tho.
- Arnaldo
tools/perf/util/intel-bts.c +2 -18
@@ -328,35 +328,19 @@ static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip) {struct machine *machine = btsq->bts->machine;struct thread *thread; -struct addr_location al;unsigned char buf[INTEL_PT_INSN_BUF_SZ ];ssize_t len; -int x86_64; -uint8_t cpumode; +bool x86_64;int err =-1 ; -if (machine__kernel_ip (machine, ip)) - cpumode =PERF_RECORD_MISC_KERNEL ; -else - cpumode =PERF_RECORD_MISC_USER ; - thread =machine__find_thread (machine,-1 , btsq->tid);if (!thread)return -1 ; -if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso) - goto out_put; - - len =dso__data_read_addr (al.map->dso, al.map, machine, ip, buf, -INTEL_PT_INSN_BUF_SZ ); + len =thread__memcpy (thread, machine, buf, ip,INTEL_PT_INSN_BUF_SZ , &x86_64);if (len <=0 ) goto out_put; -/* Load maps to ensure dso->is_64_bit has been updated */ -map__load (al.map); - - x86_64 = al.map->dso->is_64_bit; -if (intel_pt_get_insn (buf, len, x86_64, &btsq->intel_pt_insn)) goto out_put;
tools/perf/util/thread.c +23 -0
@@ -12,6 +12,7 @@#include "debug.h" #include "namespaces.h" #include "comm.h" +#include "map.h" #include "symbol.h" #include "unwind.h"
@@ -393,3 +394,25 @@ struct thread *thread__main_thread(struct machine *machine, struct thread *threareturn machine__find_thread (machine, thread->pid_, thread->pid_); } + +int thread__memcpy (struct thread *thread,struct machine *machine, +void *buf,u64 ip,int len,bool *is64bit) +{ +u8 cpumode =PERF_RECORD_MISC_USER ; +struct addr_location al; +long offset; + +if (machine__kernel_ip (machine, ip)) + cpumode =PERF_RECORD_MISC_KERNEL ; + +if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso || + al.map->dso->data.status ==DSO_DATA_STATUS_ERROR || +map__load (al.map) <0 ) +return -1 ; + + offset = al.map->map_ip (al.map, ip); +if (is64bit) + *is64bit = al.map->dso->is_64_bit; + +return dso__data_read_offset (al.map->dso, machine, offset, buf, len); +}
tools/perf/util/thread.h +3 -0
@@ -113,6 +113,9 @@ struct symbol *thread__find_symbol_fb(struct thread *thread, u8 cpumode,void thread__find_cpumode_addr_location (struct thread *thread,u64 addr,struct addr_location *al); +int thread__memcpy (struct thread *thread,struct machine *machine, +void *buf,u64 ip,int len,bool *is64bit); +static inline void *thread__priv (struct thread *thread) {return thread->priv;
Re: [PATCH v4 01/15] perf tools: Add utility function to fetch executable
Quoted 10 lines
. No need to cast around, make 'buf' be a void pointer
. Rename it to
thread__memcpy()to reflect the fact it is about copying
a chunk of memory from a thread, i.e. from its address space.. No need to have it in a separate object file, move it to thread.[ch]
. Check the return of
map__load(), the original code didn't do it, but
since we're moving this around, check that as well, could be moved to a separate patch tho.
Changes look good. Thanks Arnaldo.
-Andi
tip-bot for Andi Kleen
merge bot[tip:perf/urgent] perf thread: Generalize function to copy from thread addr space from intel-bts code
In reply to Andi Kleenperf thread: Generalize function to copy from thread addr space from intel-bts code
Commit-ID: 153259382633ecbbc0af4f3f0b6515757ebe2984 Gitweb: https://git.kernel.org/tip/153259382633ecbbc0af4f3f0b6515757ebe2984 Author: Andi Kleen <[email protected]> AuthorDate: Wed, 6 Mar 2019 17:55:35 -0300 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Wed, 6 Mar 2019 17:55:35 -0300
perf thread: Generalize function to copy from thread addr space from intel-bts code
Add a utility function to fetch executable code. Convert one user over to it. There are more places doing that, but they do significantly different actions, so they are not easy to fit into a single library function.
Committer changes:
. No need to cast around, make 'buf' be a void pointer.
. Rename it to thread__memcpy() to reflect the fact it is about copying
a chunk of memory from a thread, i.e. from its address space.
. No need to have it in a separate object file, move it to thread.[ch]
. Check the return of map__load(), the original code didn't do it, but
since we're moving this around, check that as well.
Signed-off-by Andi Kleen <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
--- tools/perf/util/intel-bts.c | 20 ++------------------ tools/perf/util/thread.c | 23 +++++++++++++++++++++++ tools/perf/util/thread.h | 3 +++ 3 files changed, 28 insertions(+), 18 deletions(-)
tools/perf/util/intel-bts.c +2 -18
@@ -328,35 +328,19 @@ static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip) {struct machine *machine = btsq->bts->machine;struct thread *thread; -struct addr_location al;unsigned char buf[INTEL_PT_INSN_BUF_SZ ];ssize_t len; -int x86_64; -uint8_t cpumode; +bool x86_64;int err =-1 ; -if (machine__kernel_ip (machine, ip)) - cpumode =PERF_RECORD_MISC_KERNEL ; -else - cpumode =PERF_RECORD_MISC_USER ; - thread =machine__find_thread (machine,-1 , btsq->tid);if (!thread)return -1 ; -if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso) - goto out_put; - - len =dso__data_read_addr (al.map->dso, al.map, machine, ip, buf, -INTEL_PT_INSN_BUF_SZ ); + len =thread__memcpy (thread, machine, buf, ip,INTEL_PT_INSN_BUF_SZ , &x86_64);if (len <=0 ) goto out_put; -/* Load maps to ensure dso->is_64_bit has been updated */ -map__load (al.map); - - x86_64 = al.map->dso->is_64_bit; -if (intel_pt_get_insn (buf, len, x86_64, &btsq->intel_pt_insn)) goto out_put;
tools/perf/util/thread.c +23 -0
@@ -12,6 +12,7 @@#include "debug.h" #include "namespaces.h" #include "comm.h" +#include "map.h" #include "symbol.h" #include "unwind.h"
@@ -393,3 +394,25 @@ struct thread *thread__main_thread(struct machine *machine, struct thread *threareturn machine__find_thread (machine, thread->pid_, thread->pid_); } + +int thread__memcpy (struct thread *thread,struct machine *machine, +void *buf,u64 ip,int len,bool *is64bit) +{ +u8 cpumode =PERF_RECORD_MISC_USER ; +struct addr_location al; +long offset; + +if (machine__kernel_ip (machine, ip)) + cpumode =PERF_RECORD_MISC_KERNEL ; + +if (!thread__find_map (thread, cpumode, ip, &al) || !al.map->dso || + al.map->dso->data.status ==DSO_DATA_STATUS_ERROR || +map__load (al.map) <0 ) +return -1 ; + + offset = al.map->map_ip (al.map, ip); +if (is64bit) + *is64bit = al.map->dso->is_64_bit; + +return dso__data_read_offset (al.map->dso, machine, offset, buf, len); +}
tools/perf/util/thread.h +3 -0
@@ -113,6 +113,9 @@ struct symbol *thread__find_symbol_fb(struct thread *thread, u8 cpumode,void thread__find_cpumode_addr_location (struct thread *thread,u64 addr,struct addr_location *al); +int thread__memcpy (struct thread *thread,struct machine *machine, +void *buf,u64 ip,int len,bool *is64bit); +static inline void *thread__priv (struct thread *thread) {return thread->priv;
PATCH v4 2/15 perf tools script: Support insn output for normal samples
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
perf script -F +insn was only working for PT traces because the PT instruction decoder was filling in the insn/insn_len sample attributes. Support it for non PT samples too on x86 using the existing x86 instruction decoder.
This adds some extra checking to ensure that we don't try to decode instructions when using perf.data from a different architecture.
% perf record -a sleep 1
% perf script -F ip,sym,insn --xed
ffffffff811704c9 remote_function movl %eax, 0x18(%rbx)
ffffffff8100bb50 intel_bts_enable_local retq ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi)
ffffffff810f1f79 generic_exec_single xor %eax, %eax
ffffffff811704c9 remote_function movl %eax, 0x18(%rbx)
ffffffff8100bb34 intel_bts_enable_local movl 0x2000(%rax), %edx
ffffffff81048610 native_apic_mem_write mov %edi, %edi
...
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Avoid printing instruction when empty
Only decode when perf.data file was collected on same architecture
--- tools/perf/arch/x86/util/Build | 1 + tools/perf/arch/x86/util/archinsn.c | 28 ++++++++++++++++++++++++++++ tools/perf/builtin-script.c | 21 ++++++++++++++++++++- tools/perf/util/archinsn.h | 12 ++++++++++++ 4 files changed, 61 insertions(+), 1 deletion(-) create mode 100644 tools/perf/arch/x86/util/archinsn.c create mode 100644 tools/perf/util/archinsn.h
tools/perf/arch/x86/util/Build +1 -0
@@ -6,6 +6,7 @@ perf-y += perf_regs.o perf-y += group.o perf-y += machine.o perf-y += event.o +perf-y += archinsn.o perf-$(CONFIG_DWARF) += dwarf-regs.o perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
tools/perf/arch/x86/util/archinsn.c new file +28 -0
@@ -0,0 +1,28 @@ +// SPDX-License-Identifier: GPL-2.0 +#include "perf.h" +#include "archinsn.h" +#include "fetch.h" +#include "util/intel-pt-decoder/insn.h" +#include "machine.h" +#include "thread.h" +#include "symbol.h" + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine) +{ +struct insn insn; +int len; +bool is64bit = false; + +if (!sample->ip) +return ; + len =fetch_exe (sample->ip, thread, machine, sample->insn, +sizeof (sample->insn), &is64bit); +if (len <=0 ) +return ; +insn_init (&insn, sample->insn, len, is64bit); +insn_get_length (&insn); +if (insn_complete (&insn) && insn.length <= len) + sample->insn_len = insn.length; +}
tools/perf/builtin-script.c +20 -1
@@ -29,10 +29,12 @@#include "util/time-utils.h" #include "util/path.h" #include "print_binary.h" +#include "archinsn.h" #include <linux/bitmap.h> #include <linux/kernel.h> #include <linux/stringify.h> #include <linux/time64.h> +#include <sys/utsname.h> #include "asm/bug.h" #include "util/mem-events.h" #include "util/dump-insn.h"
@@ -63,6 +65,7 @@ static const char *cpu_list; static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);static struct perf_stat_config stat_config;static int max_blocks; +static bool native_arch;unsigned int scripting_max_stack =PERF_MAX_STACK_DEPTH ;
@@ -1227,6 +1230,12 @@ static int perf_sample__fprintf_callindent(struct perf_sample *sample,return len + dlen; } +__weak void arch_fetch_insn (struct perf_sample *sample__maybe_unused , +struct thread *thread__maybe_unused , +struct machine *machine__maybe_unused ) +{ +} + static int perf_sample__fprintf_insn(struct perf_sample *sample, struct perf_event_attr *attr, struct thread *thread,
@@ -1234,9 +1243,12 @@ static int perf_sample__fprintf_insn(struct perf_sample *sample, {int printed =0 ; +if (sample->insn_len ==0 && native_arch) +arch_fetch_insn (sample, thread, machine); +if (PRINT_FIELD (INSNLEN )) printed +=fprintf (fp," ilen: %d" , sample->insn_len); -if (PRINT_FIELD (INSN )) { +if (PRINT_FIELD (INSN ) && sample->insn_len) {int i; printed +=fprintf (fp," insn:" );
@@ -3277,6 +3289,7 @@ int cmd_script(int argc, const char **argv) .set = false, .default_no_sample = true, }; +struct utsname uts;char *script_path =NULL ;const char **__argv;int i, j, err =0 ;
@@ -3615,6 +3628,12 @@ int cmd_script(int argc, const char **argv)if (symbol__init (&session->header.env) <0 ) goto out_delete; +uname (&uts); +if (!strcmp (uts.machine, session->header.env.arch) || + (!strcmp (uts.machine,"x86_64" ) && + !strcmp (session->header.env.arch,"i386" ))) + native_arch = true; + script.session = session;script__setup_sample_type (&script);
tools/perf/util/archinsn.h new file +12 -0
@@ -0,0 +1,12 @@ +#ifndef INSN_H +#define INSN_H 1 + +struct perf_sample ; +struct machine ; +struct thread ; + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine); + +#endif
Signature · 1 line
2.20.1
Re: [PATCH v4 02/15] perf tools script: Support insn output for normal samples
Em Tue, Mar 05, 2019 at 06:47:45AM -0800, Andi Kleen escreveu:
Quoted 24 lines
From: Andi Kleen <[email protected]>
perf script -F +insn was only working for PT traces because the PT instruction decoder was filling in the insn/insn_len sample attributes. Support it for non PT samples too on x86 using the existing x86 instruction decoder.
This adds some extra checking to ensure that we don't try to decode instructions when using perf.data from a different architecture.
% perf record -a sleep 1
% perf script -F ip,sym,insn --xed
ffffffff811704c9remote_functionmovl %eax, 0x18(%rbx)ffffffff8100bb50 intel_bts_enable_local retq ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi)
ffffffff810f1f79generic_exec_singlexor %eax, %eax
ffffffff811704c9remote_functionmovl %eax, 0x18(%rbx)
ffffffff8100bb34intel_bts_enable_localmovl 0x2000(%rax), %edx
ffffffff81048610native_apic_mem_writemov %edi, %edi
...
Quoted 6 lines
Signed-off-by Andi Kleen <[email protected]>---
v2:
Avoid printing instruction when empty
Only decode when perf.data file was collected on same architecture
Thanks, applied and added some testing notes comparing the output with the one from 'perf annotate --stdio2':
commit 2109d7fa73200bc6ff7abe909b5fe821224f98b3
Author: Andi Kleen <[email protected]> Date: Tue Mar 5 06:47:45 2019 -0800
perf script: Support insn output for normal samples
perf script -F +insn was only working for PT traces because the PT
instruction decoder was filling in the insn/insn_len sample attributes.
Support it for non PT samples too on x86 using the existing x86
instruction decoder.
This adds some extra checking to ensure that we don't try to decode
instructions when using perf.data from a different architecture.
% perf record -a sleep 1
% perf script -F ip,sym,insn --xed
ffffffff811704c9 remote_function movl %eax, 0x18(%rbx)
ffffffff8100bb50 intel_bts_enable_local retq
ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi)
ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi)
ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi)
ffffffff810f1f79 generic_exec_single xor %eax, %eax
ffffffff811704c9 remote_function movl %eax, 0x18(%rbx)
ffffffff8100bb34 intel_bts_enable_local movl 0x2000(%rax), %edx
ffffffff81048610 native_apic_mem_write mov %edi, %edi
...
Committer testing:
Before:
# perf script -F ip,sym,insn --xed | head -5
ffffffffa4068804 native_write_msr addb %al, (%rax)
ffffffffa4068804 native_write_msr addb %al, (%rax)
ffffffffa4068804 native_write_msr addb %al, (%rax)
ffffffffa4068806 native_write_msr addb %al, (%rax)
ffffffffa4068806 native_write_msr addb %al, (%rax)
# perf script -F ip,sym,insn --xed | grep -v "addb %al, (%rax)"
#
After:
# perf script -F ip,sym,insn --xed | head -5
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1)
ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1)
# perf script -F ip,sym,insn --xed | grep -v "addb %al, (%rax)" | head -5
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068804 native_write_msr wrmsr
ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1)
ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1)
#
More examples:
# perf script -F ip,sym,insn --xed | grep -v native_write_msr | head
ffffffffa416b90e tick_check_broadcast_expired btq %rax, 0x1a5f42a(%rip)
ffffffffa4956bd0 nmi_cpu_backtrace pushq %r13
ffffffffa415b95e __hrtimer_next_event_base movq 0x18(%rax), %rdx
ffffffffa4956bf3 nmi_cpu_backtrace popq %r12
ffffffffa4171d5c smp_call_function_single pause
ffffffffa4956bdd nmi_cpu_backtrace mov %ebp, %r12d
ffffffffa4797e4d menu_select cmp $0x190, %rax
ffffffffa4171d5c smp_call_function_single pause
ffffffffa405a7d8 nmi_cpu_backtrace_handler callq 0xffffffffa4956bd0
ffffffffa4797f7a menu_select shr $0x3, %rax
#
Which matches the annotate output modulo resolving callqs:
# perf annotate --stdio2 nmi_cpu_backtrace_handler
Samples: 4 of event 'cycles:ppp', 4000 Hz, Event count (approx.): 35908, [percent: local period]
nmi_cpu_backtrace_handler() /lib/modules/5.0.0+/build/vmlinux
Percent
Disassembly of section .text:
ffffffff8105a7d0 <nmi_cpu_backtrace_handler>:
nmi_cpu_backtrace_handler():
nmi_trigger_cpumask_backtrace(mask, exclude_self,
nmi_raise_cpu_backtrace);
}
static int nmi_cpu_backtrace_handler(unsigned int cmd, struct pt_regs *regs)
{
24.45 → callq __fentry__
if (nmi_cpu_backtrace(regs))
mov %rsi,%rdi
75.55 → callq nmi_cpu_backtrace
return NMI_HANDLED;
movzbl %al,%eax
return NMI_DONE;
}
← retq
#
# perf annotate --stdio2 __hrtimer_next_event_base
Samples: 4 of event 'cycles:ppp', 4000 Hz, Event count (approx.): 767977, [percent: local period]
__hrtimer_next_event_base() /lib/modules/5.0.0+/build/vmlinux
Percent
Disassembly of section .text:
ffffffff8115b910 <__hrtimer_next_event_base>:
__hrtimer_next_event_base():
static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
const struct hrtimer *exclude,
unsigned int active,
ktime_t expires_next)
{
→ callq __fentry__
<SNIP>
4a: add $0x1,%r14
77.31 mov 0x18(%rax),%rdx
shl $0x6,%r14
sub 0x38(%rbx,%r14,1),%rdx
if (expires < expires_next) {
cmp %r12,%rdx
↓ jge 68
<SNIP>
Signed-off-by: Andi Kleen <[email protected]>
Tested-by: Arnaldo Carvalho de Melo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
[ Converted fetch_exe() to use the name it ended up having when merged: thread__memcpy() ]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>tools/perf/arch/x86/util/Build +1 -0
@@ -6,6 +6,7 @@ perf-y += perf_regs.o perf-y += group.o perf-y += machine.o perf-y += event.o +perf-y += archinsn.o perf-$(CONFIG_DWARF) += dwarf-regs.o perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
tools/perf/arch/x86/util/archinsn.c new file +26 -0
@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: GPL-2.0 +#include "perf.h" +#include "archinsn.h" +#include "util/intel-pt-decoder/insn.h" +#include "machine.h" +#include "thread.h" +#include "symbol.h" + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine) +{ +struct insn insn; +int len; +bool is64bit = false; + +if (!sample->ip) +return ; + len =thread__memcpy (thread, machine, sample->insn, sample->ip,sizeof (sample->insn), &is64bit); +if (len <=0 ) +return ; +insn_init (&insn, sample->insn, len, is64bit); +insn_get_length (&insn); +if (insn_complete (&insn) && insn.length <= len) + sample->insn_len = insn.length; +}
tools/perf/builtin-script.c +20 -1
@@ -29,10 +29,12 @@#include "util/time-utils.h" #include "util/path.h" #include "print_binary.h" +#include "archinsn.h" #include <linux/bitmap.h> #include <linux/kernel.h> #include <linux/stringify.h> #include <linux/time64.h> +#include <sys/utsname.h> #include "asm/bug.h" #include "util/mem-events.h" #include "util/dump-insn.h"
@@ -63,6 +65,7 @@ static const char *cpu_list; static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);static struct perf_stat_config stat_config;static int max_blocks; +static bool native_arch;unsigned int scripting_max_stack =PERF_MAX_STACK_DEPTH ;
@@ -1227,6 +1230,12 @@ static int perf_sample__fprintf_callindent(struct perf_sample *sample,return len + dlen; } +__weak void arch_fetch_insn (struct perf_sample *sample__maybe_unused , +struct thread *thread__maybe_unused , +struct machine *machine__maybe_unused ) +{ +} + static int perf_sample__fprintf_insn(struct perf_sample *sample, struct perf_event_attr *attr, struct thread *thread,
@@ -1234,9 +1243,12 @@ static int perf_sample__fprintf_insn(struct perf_sample *sample, {int printed =0 ; +if (sample->insn_len ==0 && native_arch) +arch_fetch_insn (sample, thread, machine); +if (PRINT_FIELD (INSNLEN )) printed +=fprintf (fp," ilen: %d" , sample->insn_len); -if (PRINT_FIELD (INSN )) { +if (PRINT_FIELD (INSN ) && sample->insn_len) {int i; printed +=fprintf (fp," insn:" );
@@ -3277,6 +3289,7 @@ int cmd_script(int argc, const char **argv) .set = false, .default_no_sample = true, }; +struct utsname uts;char *script_path =NULL ;const char **__argv;int i, j, err =0 ;
@@ -3615,6 +3628,12 @@ int cmd_script(int argc, const char **argv)if (symbol__init (&session->header.env) <0 ) goto out_delete; +uname (&uts); +if (!strcmp (uts.machine, session->header.env.arch) || + (!strcmp (uts.machine,"x86_64" ) && + !strcmp (session->header.env.arch,"i386" ))) + native_arch = true; + script.session = session;script__setup_sample_type (&script);
tools/perf/util/archinsn.h new file +12 -0
@@ -0,0 +1,12 @@ +#ifndef INSN_H +#define INSN_H 1 + +struct perf_sample ; +struct machine ; +struct thread ; + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine); + +#endif
tip-bot for Andi Kleen
merge bot[tip:perf/urgent] perf script: Support insn output for normal samples
In reply to Andi Kleenperf script: Support insn output for normal samples
Commit-ID: 3ab481a1cfe1511b94e142b648e2c5ade9175ed3 Gitweb: https://git.kernel.org/tip/3ab481a1cfe1511b94e142b648e2c5ade9175ed3 Author: Andi Kleen <[email protected]> AuthorDate: Tue, 5 Mar 2019 06:47:45 -0800 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Mon, 11 Mar 2019 11:56:02 -0300
perf script: Support insn output for normal samples
perf script -F +insn was only working for PT traces because the PT instruction decoder was filling in the insn/insn_len sample attributes. Support it for non PT samples too on x86 using the existing x86 instruction decoder.
This adds some extra checking to ensure that we don't try to decode instructions when using perf.data from a different architecture.
% perf record -a sleep 1 % perf script -F ip,sym,insn --xed ffffffff811704c9 remote_function movl %eax, 0x18(%rbx) ffffffff8100bb50 intel_bts_enable_local retq ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff81048612 native_apic_mem_write movl %esi, -0xa04000(%rdi) ffffffff810f1f79 generic_exec_single xor %eax, %eax ffffffff811704c9 remote_function movl %eax, 0x18(%rbx) ffffffff8100bb34 intel_bts_enable_local movl 0x2000(%rax), %edx ffffffff81048610 native_apic_mem_write mov %edi, %edi ...
Committer testing:
Before:
# perf script -F ip,sym,insn --xed | head -5 ffffffffa4068804 native_write_msr addb %al, (%rax) ffffffffa4068804 native_write_msr addb %al, (%rax) ffffffffa4068804 native_write_msr addb %al, (%rax) ffffffffa4068806 native_write_msr addb %al, (%rax) ffffffffa4068806 native_write_msr addb %al, (%rax) # perf script -F ip,sym,insn --xed | grep -v "addb %al, (%rax)" #
After:
# perf script -F ip,sym,insn --xed | head -5 ffffffffa4068804 native_write_msr wrmsr ffffffffa4068804 native_write_msr wrmsr ffffffffa4068804 native_write_msr wrmsr ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1) ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1) # perf script -F ip,sym,insn --xed | grep -v "addb %al, (%rax)" | head -5 ffffffffa4068804 native_write_msr wrmsr ffffffffa4068804 native_write_msr wrmsr ffffffffa4068804 native_write_msr wrmsr ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1) ffffffffa4068806 native_write_msr nopl %eax, (%rax,%rax,1) #
More examples:
# perf script -F ip,sym,insn --xed | grep -v native_write_msr | head ffffffffa416b90e tick_check_broadcast_expired btq %rax, 0x1a5f42a(%rip) ffffffffa4956bd0 nmi_cpu_backtrace pushq %r13 ffffffffa415b95e __hrtimer_next_event_base movq 0x18(%rax), %rdx ffffffffa4956bf3 nmi_cpu_backtrace popq %r12 ffffffffa4171d5c smp_call_function_single pause ffffffffa4956bdd nmi_cpu_backtrace mov %ebp, %r12d ffffffffa4797e4d menu_select cmp $0x190, %rax ffffffffa4171d5c smp_call_function_single pause ffffffffa405a7d8 nmi_cpu_backtrace_handler callq 0xffffffffa4956bd0 ffffffffa4797f7a menu_select shr $0x3, %rax #
Which matches the annotate output modulo resolving callqs:
# perf annotate --stdio2 nmi_cpu_backtrace_handler
Samples: 4 of event 'cycles:ppp', 4000 Hz, Event count (approx.): 35908, [percent: local period]
nmi_cpu_backtrace_handler() /lib/modules/5.0.0+/build/vmlinux
Percent
Disassembly of section .text: ffffffff8105a7d0 <nmi_cpu_backtrace_handler>:
nmi_cpu_backtrace_handler():
nmi_trigger_cpumask_backtrace(mask, exclude_self,
nmi_raise_cpu_backtrace);
}
static int nmi_cpu_backtrace_handler(unsigned int cmd, struct pt_regs *regs)
{
24.45 → callq __fentry__
if (nmi_cpu_backtrace(regs))
mov %rsi,%rdi
75.55 → callq nmi_cpu_backtrace
return NMI_HANDLED;
movzbl %al,%eax
return NMI_DONE;
}
← retq
# # perf annotate --stdio2 __hrtimer_next_event_base
Samples: 4 of event 'cycles:ppp', 4000 Hz, Event count (approx.): 767977, [percent: local period]
__hrtimer_next_event_base() /lib/modules/5.0.0+/build/vmlinux
Percent
Disassembly of section .text: ffffffff8115b910 <__hrtimer_next_event_base>:
__hrtimer_next_event_base():
static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
const struct hrtimer *exclude, unsigned int active, ktime_t expires_next)
{
→ callq __fentry__
<SNIP>
4a: add $0x1,%r14
77.31 mov 0x18(%rax),%rdx
shl $0x6,%r14
sub 0x38(%rbx,%r14,1),%rdx
if (expires < expires_next) {
cmp %r12,%rdx
↓ jge 68<SNIP>
Signed-off-by Andi Kleen <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
[ Converted fetch_exe() to use the name it ended up having when merged: thread__memcpy() ]
[ archinsn.c needs the instruction decoder that is only build when CONFIG_AUXTRACE=y, fix that ]
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
--- tools/perf/arch/x86/util/Build | 1 + tools/perf/arch/x86/util/archinsn.c | 26 ++++++++++++++++++++++++++ tools/perf/builtin-script.c | 21 ++++++++++++++++++++- tools/perf/util/archinsn.h | 12 ++++++++++++ 4 files changed, 59 insertions(+), 1 deletion(-)
tools/perf/arch/x86/util/Build +1 -0
@@ -14,5 +14,6 @@ perf-$(CONFIG_LOCAL_LIBUNWIND) += unwind-libunwind.o perf-$(CONFIG_LIBDW_DWARF_UNWIND) += unwind-libdw.o perf-$(CONFIG_AUXTRACE) += auxtrace.o +perf-$(CONFIG_AUXTRACE) += archinsn.o perf-$(CONFIG_AUXTRACE) += intel-pt.o perf-$(CONFIG_AUXTRACE) += intel-bts.o
tools/perf/arch/x86/util/archinsn.c new file +26 -0
@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: GPL-2.0 +#include "perf.h" +#include "archinsn.h" +#include "util/intel-pt-decoder/insn.h" +#include "machine.h" +#include "thread.h" +#include "symbol.h" + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine) +{ +struct insn insn; +int len; +bool is64bit = false; + +if (!sample->ip) +return ; + len =thread__memcpy (thread, machine, sample->insn, sample->ip,sizeof (sample->insn), &is64bit); +if (len <=0 ) +return ; +insn_init (&insn, sample->insn, len, is64bit); +insn_get_length (&insn); +if (insn_complete (&insn) && insn.length <= len) + sample->insn_len = insn.length; +}
tools/perf/builtin-script.c +20 -1
@@ -29,10 +29,12 @@#include "util/time-utils.h" #include "util/path.h" #include "print_binary.h" +#include "archinsn.h" #include <linux/bitmap.h> #include <linux/kernel.h> #include <linux/stringify.h> #include <linux/time64.h> +#include <sys/utsname.h> #include "asm/bug.h" #include "util/mem-events.h" #include "util/dump-insn.h"
@@ -63,6 +65,7 @@ static const char *cpu_list; static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);static struct perf_stat_config stat_config;static int max_blocks; +static bool native_arch;unsigned int scripting_max_stack =PERF_MAX_STACK_DEPTH ;
@@ -1227,6 +1230,12 @@ static int perf_sample__fprintf_callindent(struct perf_sample *sample,return len + dlen; } +__weak void arch_fetch_insn (struct perf_sample *sample__maybe_unused , +struct thread *thread__maybe_unused , +struct machine *machine__maybe_unused ) +{ +} + static int perf_sample__fprintf_insn(struct perf_sample *sample, struct perf_event_attr *attr, struct thread *thread,
@@ -1234,9 +1243,12 @@ static int perf_sample__fprintf_insn(struct perf_sample *sample, {int printed =0 ; +if (sample->insn_len ==0 && native_arch) +arch_fetch_insn (sample, thread, machine); +if (PRINT_FIELD (INSNLEN )) printed +=fprintf (fp," ilen: %d" , sample->insn_len); -if (PRINT_FIELD (INSN )) { +if (PRINT_FIELD (INSN ) && sample->insn_len) {int i; printed +=fprintf (fp," insn:" );
@@ -3277,6 +3289,7 @@ int cmd_script(int argc, const char **argv) .set = false, .default_no_sample = true, }; +struct utsname uts;char *script_path =NULL ;const char **__argv;int i, j, err =0 ;
@@ -3615,6 +3628,12 @@ int cmd_script(int argc, const char **argv)if (symbol__init (&session->header.env) <0 ) goto out_delete; +uname (&uts); +if (!strcmp (uts.machine, session->header.env.arch) || + (!strcmp (uts.machine,"x86_64" ) && + !strcmp (session->header.env.arch,"i386" ))) + native_arch = true; + script.session = session;script__setup_sample_type (&script);
tools/perf/util/archinsn.h new file +12 -0
@@ -0,0 +1,12 @@ +#ifndef INSN_H +#define INSN_H 1 + +struct perf_sample ; +struct machine ; +struct thread ; + +void arch_fetch_insn (struct perf_sample *sample, +struct thread *thread, +struct machine *machine); + +#endif
PATCH v4 3/15 perf tools script: Filter COMM/FORK/.. events by CPU
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
The --cpu option only filtered samples. Filter other perf events, such as COMM, FORK, SWITCH by the CPU too.
Reported-by Jiri Olsa <[email protected]>
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/builtin-script.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)
tools/perf/builtin-script.c +24 -0
@@ -2038,6 +2038,9 @@ static int process_comm_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->comm.pid, event->comm.tid);if (thread ==NULL ) {pr_debug ("problem processing COMM event, skipping it.\n" );
@@ -2073,6 +2076,9 @@ static int process_namespaces_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->namespaces.pid, event->namespaces.tid);if (thread ==NULL ) {
@@ -2108,6 +2114,9 @@ static int process_fork_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_fork (tool, event, sample, machine) <0 )return -1 ;
@@ -2141,6 +2150,9 @@ static int process_exit_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->fork.pid, event->fork.tid);if (thread ==NULL ) {pr_debug ("problem processing EXIT event, skipping it.\n" );
@@ -2174,6 +2186,9 @@ static int process_mmap_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap (tool, event, sample, machine) <0 )return -1 ;
@@ -2206,6 +2221,9 @@ static int process_mmap2_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap2 (tool, event, sample, machine) <0 )return -1 ;
@@ -2238,6 +2256,9 @@ static int process_switch_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_switch (tool, event, sample, machine) <0 )return -1 ;
@@ -2266,6 +2287,9 @@ process_lost_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);struct thread *thread; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, sample->pid, sample->tid);if (thread ==NULL )
Signature · 1 line
2.20.1
Adrian Hunter
reviewerRe: [PATCH v4 03/15] perf tools script: Filter COMM/FORK/.. events by CPU
Quoted 4 lines · Andi Kleen
From: Andi Kleen <[email protected]>
The --cpu option only filtered samples. Filter other perf events, such as COMM, FORK, SWITCH by the CPU too.
Because tasks can migrate from cpu to cpu, we probably need to process most of the events anyway, even if they are not printed.
Quoted 92 lines
Reported-by Jiri Olsa <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/builtin-script.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)tools/perf/builtin-script.c +24
-0@@ -2038,6 +2038,9 @@ static int process_comm_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->comm.pid, event->comm.tid);if (thread ==NULL ) {pr_debug ("problem processing COMM event, skipping it.\n" );@@ -2073,6 +2076,9 @@ static int process_namespaces_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->namespaces.pid, event->namespaces.tid);if (thread ==NULL ) {@@ -2108,6 +2114,9 @@ static int process_fork_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_fork (tool, event, sample, machine) <0 )return -1 ;@@ -2141,6 +2150,9 @@ static int process_exit_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->fork.pid, event->fork.tid);if (thread ==NULL ) {pr_debug ("problem processing EXIT event, skipping it.\n" );@@ -2174,6 +2186,9 @@ static int process_mmap_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap (tool, event, sample, machine) <0 )return -1 ;@@ -2206,6 +2221,9 @@ static int process_mmap2_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap2 (tool, event, sample, machine) <0 )return -1 ;@@ -2238,6 +2256,9 @@ static int process_switch_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_switch (tool, event, sample, machine) <0 )return -1 ;@@ -2266,6 +2287,9 @@ process_lost_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);struct thread *thread; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, sample->pid, sample->tid);if (thread ==NULL )
Re: [PATCH v4 03/15] perf tools script: Filter COMM/FORK/.. events by CPU
Quoted 8 lines · Adrian Hunter
Quoted 4 lines · Andi Kleen
From: Andi Kleen <[email protected]>
The --cpu option only filtered samples. Filter other perf events, such as COMM, FORK, SWITCH by the CPU too.
Because tasks can migrate from cpu to cpu, we probably need to process most of the events anyway, even if they are not printed.
agreed, I wonder we could just make the perf_event__fprintf conditional
jirka
Quoted 94 lines
Quoted 93 lines
Reported-by Jiri Olsa <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/builtin-script.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)tools/perf/builtin-script.c +24
-0@@ -2038,6 +2038,9 @@ static int process_comm_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->comm.pid, event->comm.tid);if (thread ==NULL ) {pr_debug ("problem processing COMM event, skipping it.\n" );@@ -2073,6 +2076,9 @@ static int process_namespaces_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->namespaces.pid, event->namespaces.tid);if (thread ==NULL ) {@@ -2108,6 +2114,9 @@ static int process_fork_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_fork (tool, event, sample, machine) <0 )return -1 ;@@ -2141,6 +2150,9 @@ static int process_exit_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->fork.pid, event->fork.tid);if (thread ==NULL ) {pr_debug ("problem processing EXIT event, skipping it.\n" );@@ -2174,6 +2186,9 @@ static int process_mmap_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap (tool, event, sample, machine) <0 )return -1 ;@@ -2206,6 +2221,9 @@ static int process_mmap2_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap2 (tool, event, sample, machine) <0 )return -1 ;@@ -2238,6 +2256,9 @@ static int process_switch_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_switch (tool, event, sample, machine) <0 )return -1 ;@@ -2266,6 +2287,9 @@ process_lost_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);struct thread *thread; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, sample->pid, sample->tid);if (thread ==NULL )
Re: [PATCH v4 03/15] perf tools script: Filter COMM/FORK/.. events by CPU
Em Thu, Mar 07, 2019 at 12:02:31PM +0100, Jiri Olsa escreveu:
Quoted 11 lines · Adrian Hunter
Quoted 8 lines · Andi Kleen
Quoted 4 lines
From: Andi Kleen <[email protected]>
The --cpu option only filtered samples. Filter other perf events, such as COMM, FORK, SWITCH by the CPU too.
Because tasks can migrate from cpu to cpu, we probably need to process most of the events anyway, even if they are not printed.
agreed, I wonder we could just make the
perf_event__fprintfconditional
Humm, probably just do the filtering on PERF_RECORD_SAMPLE is enough? I.e. having the other PERF_RECORD_{COMM,MMAP,} etc is required in face of migration.
- Arnaldo
Quoted 96 lines
jirka
Quoted 94 lines
Quoted 93 lines
Reported-by Jiri Olsa <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/builtin-script.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)tools/perf/builtin-script.c +24
-0@@ -2038,6 +2038,9 @@ static int process_comm_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->comm.pid, event->comm.tid);if (thread ==NULL ) {pr_debug ("problem processing COMM event, skipping it.\n" );@@ -2073,6 +2076,9 @@ static int process_namespaces_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);int ret =-1 ; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->namespaces.pid, event->namespaces.tid);if (thread ==NULL ) {@@ -2108,6 +2114,9 @@ static int process_fork_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_fork (tool, event, sample, machine) <0 )return -1 ;@@ -2141,6 +2150,9 @@ static int process_exit_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, event->fork.pid, event->fork.tid);if (thread ==NULL ) {pr_debug ("problem processing EXIT event, skipping it.\n" );@@ -2174,6 +2186,9 @@ static int process_mmap_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap (tool, event, sample, machine) <0 )return -1 ;@@ -2206,6 +2221,9 @@ static int process_mmap2_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_mmap2 (tool, event, sample, machine) <0 )return -1 ;@@ -2238,6 +2256,9 @@ static int process_switch_event(struct perf_tool *tool,struct perf_session *session = script->session;struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id); +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; +if (perf_event__process_switch (tool, event, sample, machine) <0 )return -1 ;@@ -2266,6 +2287,9 @@ process_lost_event(struct perf_tool *tool,struct perf_evsel *evsel =perf_evlist__id2evsel (session->evlist, sample->id);struct thread *thread; +if (cpu_list && !test_bit (sample->cpu, cpu_bitmap)) +return 0 ; + thread =machine__findnew_thread (machine, sample->pid, sample->tid);if (thread ==NULL )
Signature · 1 line
- Arnaldo
Re: [PATCH v4 03/15] perf tools script: Filter COMM/FORK/.. events by CPU
Quoted 8 lines · Arnaldo Carvalho de Melo
Em Thu, Mar 07, 2019 at 12:02:31PM +0100, Jiri Olsa escreveu:
Quoted 7 lines · Adrian Hunter
Quoted 6 lines · Andi Kleen
Quoted 4 lines
From: Andi Kleen <[email protected]>
The --cpu option only filtered samples. Filter other perf events, such as COMM, FORK, SWITCH by the CPU too.
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
Because tasks can migrate from cpu to cpu, we probably need to process most of the events anyway, even if they are not printed.
agreed, I wonder we could just make the
perf_event__fprintfconditionalHumm, probably just do the filtering on
PERF_RECORD_SAMPLEis enough? I.e. having the other PERF_RECORD_{COMM,MMAP,} etc is required in face of migration.
The goal was to only show the output for the correct CPU in the perf sample context browser. Otherwise the output on larger systems is very confusing because most of it is for irrelevant CPUs.
-andi
PATCH v4 4/15 perf tools report: Support nano seconds
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Upcoming changes add timestamp output in perf report. Add a --ns argument similar to perf script to support nanoseconds resolution when needed.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Move flag into symbol_conf and change all users
--- tools/perf/Documentation/perf-report.txt | 3 +++ tools/perf/builtin-report.c | 1 + tools/perf/builtin-script.c | 11 +++++------ tools/perf/util/symbol.c | 1 + tools/perf/util/symbol_conf.h | 1 + 5 files changed, 11 insertions(+), 6 deletions(-)
tools/perf/Documentation/perf-report.txt +3 -0
@@ -477,6 +477,9 @@ include::itrace.txt[] Please note that not all mmaps are stored, options affecting which ones are include 'perf record --data', for instance. +--ns:: + Show time stamps in nanoseconds. + --stats:: Display overall events statistics without any further processing. (like the one at the end of the perf report -D command)
tools/perf/builtin-report.c +1 -0
@@ -1147,6 +1147,7 @@ int cmd_report(int argc, const char **argv)OPT_CALLBACK (0 ,"percent-type" , &report.annotation_opts,"local-period" ,"Set percent type local/global-period/hits" , annotate_parse_percent_type), +OPT_BOOLEAN (0 ,"ns" , &symbol_conf.nanosecs,"Show times in nanosecs" ),OPT_END () }; struct perf_data data = {
tools/perf/builtin-script.c +5 -6
@@ -60,7 +60,6 @@ static bool no_callchain;static bool latency_format;static bool system_wide;static bool print_flags; -static bool nanosecs;static const char *cpu_list; static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);static struct perf_stat_config stat_config;
@@ -691,7 +690,7 @@ static int perf_sample__fprintf_start(struct perf_sample *sample, secs = nsecs /NSEC_PER_SEC ; nsecs -= secs *NSEC_PER_SEC ; -if (nanosecs) +if (symbol_conf.nanosecs) printed +=fprintf (fp,"%5lu.%09llu: " , secs, nsecs);else {char sample_time[32 ];
@@ -3268,7 +3267,7 @@ static int parse_insn_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"+insn,-event,-period" ,0 );itrace_parse_synth_opts (opt,"i0ns" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3286,7 +3285,7 @@ static int parse_call_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"-ip,-addr,-event,-period,+callindent" ,0 );itrace_parse_synth_opts (opt,"cewp" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3296,7 +3295,7 @@ static int parse_callret_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"-ip,-addr,-event,-period,+callindent,+flags" ,0 );itrace_parse_synth_opts (opt,"crewp" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3432,7 +3431,7 @@ int cmd_script(int argc, const char **argv)OPT_BOOLEAN ('f' ,"force" , &symbol_conf.force,"don't complain, do it" ),OPT_INTEGER (0 ,"max-blocks" , &max_blocks,"Maximum number of code blocks to dump with brstackinsn" ), -OPT_BOOLEAN (0 ,"ns" , &nanosecs, +OPT_BOOLEAN (0 ,"ns" , &symbol_conf.nanosecs,"Use 9 decimal places when displaying time" ), OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", "Instruction Tracing options\n" ITRACE_HELP,
tools/perf/util/symbol.c +1 -0
@@ -39,6 +39,7 @@ int vmlinux_path__nr_entries;char **vmlinux_path;struct symbol_conf symbol_conf = { + .nanosecs = false, .use_modules = true, .try_vmlinux_path = true, .demangle = true,
tools/perf/util/symbol_conf.h +1 -0
@@ -8,6 +8,7 @@ struct strlist;struct intlist ;struct symbol_conf { +bool nanosecs;unsigned short priv_size;bool try_vmlinux_path, init_annotation,
Signature · 1 line
2.20.1
tip-bot for Andi Kleen
merge bot[tip:perf/urgent] perf report: Support output in nanoseconds
perf report: Support output in nanoseconds
Commit-ID: 52bab8868211b7c504146f6239e101421d4d125b Gitweb: https://git.kernel.org/tip/52bab8868211b7c504146f6239e101421d4d125b Author: Andi Kleen <[email protected]> AuthorDate: Tue, 5 Mar 2019 06:47:47 -0800 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Mon, 11 Mar 2019 11:56:02 -0300
perf report: Support output in nanoseconds
Upcoming changes add timestamp output in perf report. Add a --ns argument similar to perf script to support nanoseconds resolution when needed.
Signed-off-by Andi Kleen <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
--- tools/perf/Documentation/perf-report.txt | 3 +++ tools/perf/builtin-report.c | 1 + tools/perf/builtin-script.c | 11 +++++------ tools/perf/util/symbol.c | 1 + tools/perf/util/symbol_conf.h | 1 + 5 files changed, 11 insertions(+), 6 deletions(-)
tools/perf/Documentation/perf-report.txt +3 -0
@@ -477,6 +477,9 @@ include::itrace.txt[] Please note that not all mmaps are stored, options affecting which ones are include 'perf record --data', for instance. +--ns:: + Show time stamps in nanoseconds. + --stats:: Display overall events statistics without any further processing. (like the one at the end of the perf report -D command)
tools/perf/builtin-report.c +1 -0
@@ -1147,6 +1147,7 @@ int cmd_report(int argc, const char **argv)OPT_CALLBACK (0 ,"percent-type" , &report.annotation_opts,"local-period" ,"Set percent type local/global-period/hits" , annotate_parse_percent_type), +OPT_BOOLEAN (0 ,"ns" , &symbol_conf.nanosecs,"Show times in nanosecs" ),OPT_END () }; struct perf_data data = {
tools/perf/builtin-script.c +5 -6
@@ -60,7 +60,6 @@ static bool no_callchain;static bool latency_format;static bool system_wide;static bool print_flags; -static bool nanosecs;static const char *cpu_list; static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);static struct perf_stat_config stat_config;
@@ -691,7 +690,7 @@ static int perf_sample__fprintf_start(struct perf_sample *sample, secs = nsecs /NSEC_PER_SEC ; nsecs -= secs *NSEC_PER_SEC ; -if (nanosecs) +if (symbol_conf.nanosecs) printed +=fprintf (fp,"%5lu.%09llu: " , secs, nsecs);else {char sample_time[32 ];
@@ -3244,7 +3243,7 @@ static int parse_insn_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"+insn,-event,-period" ,0 );itrace_parse_synth_opts (opt,"i0ns" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3262,7 +3261,7 @@ static int parse_call_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"-ip,-addr,-event,-period,+callindent" ,0 );itrace_parse_synth_opts (opt,"cewp" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3272,7 +3271,7 @@ static int parse_callret_trace(const struct option *opt __maybe_unused, {parse_output_fields (NULL ,"-ip,-addr,-event,-period,+callindent,+flags" ,0 );itrace_parse_synth_opts (opt,"crewp" ,0 ); - nanosecs = true; + symbol_conf.nanosecs = true;return 0 ; }
@@ -3408,7 +3407,7 @@ int cmd_script(int argc, const char **argv)OPT_BOOLEAN ('f' ,"force" , &symbol_conf.force,"don't complain, do it" ),OPT_INTEGER (0 ,"max-blocks" , &max_blocks,"Maximum number of code blocks to dump with brstackinsn" ), -OPT_BOOLEAN (0 ,"ns" , &nanosecs, +OPT_BOOLEAN (0 ,"ns" , &symbol_conf.nanosecs,"Use 9 decimal places when displaying time" ), OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", "Instruction Tracing options\n" ITRACE_HELP,
tools/perf/util/symbol.c +1 -0
@@ -39,6 +39,7 @@ int vmlinux_path__nr_entries;char **vmlinux_path;struct symbol_conf symbol_conf = { + .nanosecs = false, .use_modules = true, .try_vmlinux_path = true, .demangle = true,
tools/perf/util/symbol_conf.h +1 -0
@@ -8,6 +8,7 @@ struct strlist;struct intlist ;struct symbol_conf { +bool nanosecs;unsigned short priv_size;bool try_vmlinux_path, init_annotation,
PATCH v4 5/15 perf tools report: Parse time quantum
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Many workloads change over time. perf report currently aggregates the whole time range reported in perf.data.
This patch adds an option for a time quantum to quantisize the perf.data over time.
This just adds the option, will be used in follow on patches for a time sort key.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Move time_quantum to symbol_conf. check for zero time quantum v3:
Document s unit
--- tools/perf/Documentation/perf-report.txt | 4 +++ tools/perf/builtin-report.c | 41 ++++++++++++++++++++++++ tools/perf/util/symbol.c | 1 + tools/perf/util/symbol_conf.h | 1 + 4 files changed, 47 insertions(+)
tools/perf/Documentation/perf-report.txt +4 -0
@@ -497,6 +497,10 @@ include::itrace.txt[] The period/hits keywords set the base the percentage is computed on - the samples period or the number of samples (hits). +--time-quantum:: + Configure time quantum for time sort key. Default 100ms. + Accepts s, us, ms, ns units. + include::callchain-overhead-calculation.txt[] SEE ALSO
tools/perf/builtin-report.c +41 -0
59 lines not on this page · raw message
tools/perf/util/symbol.c +1 -0
@@ -45,6 +45,7 @@ struct symbol_conf symbol_conf = { .demangle = true, .demangle_kernel = false, .cumulate_callchain = true, + .time_quantum = 100000000, /* 100ms */ .show_hist_headers = true, .symfs = "", .event_group = true,
tools/perf/util/symbol_conf.h +1 -0
@@ -56,6 +56,7 @@ struct symbol_conf { *sym_list_str, *col_width_list_str, *bt_stop_list_str; +unsigned long time_quantum;struct strlist *dso_list, *comm_list, *sym_list,
Signature · 1 line
2.20.1
Re: [PATCH v4 05/15] perf tools report: Parse time quantum
Em Tue, Mar 05, 2019 at 06:47:48AM -0800, Andi Kleen escreveu:
Quoted 64 lines
From: Andi Kleen <[email protected]>
Many workloads change over time. perf report currently aggregates the whole time range reported in perf.data.
This patch adds an option for a time quantum to quantisize the perf.data over time.
This just adds the option, will be used in follow on patches for a time sort key.
Signed-off-by Andi Kleen <[email protected]>---
v2:
Movetime_quantumtosymbol_conf. check for zero time quantum v3:
Document s unit--- tools/perf/Documentation/perf-report.txt | 4 +++ tools/perf/builtin-report.c | 41 ++++++++++++++++++++++++ tools/perf/util/symbol.c | 1 + tools/perf/util/symbol_conf.h | 1 + 4 files changed, 47 insertions(+)tools/perf/Documentation/perf-report.txt +4
-0@@ -497,6 +497,10 @@ include::itrace.txt[] The period/hits keywords set the base the percentage is computed on - the samples period or the number of samples (hits). +--time-quantum:: + Configure time quantum for time sort key. Default 100ms. + Accepts s, us, ms, ns units. + include::callchain-overhead-calculation.txt[] SEE ALSOtools/perf/builtin-report.c +9
-0@@ -47,6 +47,7 @@#include <errno.h> #include <inttypes.h> #include <regex.h> +#include "sane_ctype.h" #include <signal.h> #include <linux/bitmap.h> #include <linux/stringify.h> @@ -926,6 +927,43 @@ report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)return parse_callchain_report_opt (arg); } +static int +parse_time_quantum (const struct option *opt,const char *arg, +int unset__maybe_unused ) +{ +unsigned long *time_q = opt->value; +char *end; + + *time_q =strtoul (arg, &end,0 );
tools/perf/builtin-report.c report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) · line 947
Quoted 10 lines
tools/perf/builtin-report.c +10
-0+if (end == arg) + goto parse_err; +if (*time_q ==0 ) { +pr_err ("time quantum cannot be 0" ); +return -1 ; + } +while (isspace (*end)) + end++; +if (*end ==0 ) +return 0 ;
We have tools/include/linux/time64.h, just like the kernel, so please use:
#include <linux/time64.h>
tools/perf/builtin-report.c report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) · line 949
Quoted 2 lines
tools/perf/builtin-report.c +2
-0+if (!strcmp (end,"s" )) { + *time_q *=1000000000 ;
NSEC_PER_SEC;
tools/perf/builtin-report.c report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) · line 953
Quoted 4 lines
tools/perf/builtin-report.c +4
-0+return 0 ; + } +if (!strcmp (end,"ms" )) { + *time_q *=1000000 ;
NSEC_PER_MSEC;
tools/perf/builtin-report.c report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) · line 957
Quoted 4 lines
tools/perf/builtin-report.c +4
-0+return 0 ; + } +if (!strcmp (end,"us" )) { + *time_q *=1000 ;
NSEC_PER_USEC;
one more note below
Quoted 27 lines
tools/perf/builtin-report.c +12
-0+return 0 ; + } +if (!strcmp (end,"ns" )) +return 0 ; +parse_err: +pr_err ("Cannot parse time quantum `%s'\n" , arg); +return -1 ; +} +int report_parse_ignore_callees_opt (const struct option *opt__maybe_unused ,const char *arg,int unset__maybe_unused )@@ -1148,6 +1186,9 @@ int cmd_report(int argc, const char **argv) "Set percent type local/global-period/hits", annotate_parse_percent_type),OPT_BOOLEAN (0 ,"ns" , &symbol_conf.nanosecs,"Show times in nanosecs" ), +OPT_CALLBACK (0 ,"time-quantum" , &symbol_conf.time_quantum,"time (ms|us|ns|s)" , +"Set time quantum for time sort key (default 100ms)" , + parse_time_quantum),OPT_END () }; struct perf_data data = {tools/perf/util/symbol.c +0
-0
tools/perf/util/symbol.c struct symbol_conf symbol_conf = { · line 48
Quoted 4 lines
tools/perf/util/symbol.c +1
-0@@ -45,6 +45,7 @@ struct symbol_conf symbol_conf = { .demangle = true, .demangle_kernel = false, .cumulate_callchain = true, + .time_quantum = 100000000, /* 100ms */
100 * NSEC_PER_MSEC;
Quoted 17 lines
tools/perf/util/symbol.c +0
-0.show_hist_headers = true, .symfs = "", .event_group = true,tools/perf/util/symbol_conf.h +1
-0@@ -56,6 +56,7 @@ struct symbol_conf { *sym_list_str, *col_width_list_str, *bt_stop_list_str; +unsigned long time_quantum;struct strlist *dso_list, *comm_list, *sym_list,Signature · 1 line
2.20.1
Signature · 1 line
- Arnaldo
Re: [PATCH v4 05/15] perf tools report: Parse time quantum
Em Fri, Mar 08, 2019 at 10:45:32AM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 9 lines
Em Tue, Mar 05, 2019 at 06:47:48AM -0800, Andi Kleen escreveu:
Review
Quoted 3 lines
From: Andi Kleen <[email protected]>
(quoted diff) +2
-0+ if (*end == 0) + return 0;We have
tools/include/linux/time64.h, just like the kernel, so please use:#include <
linux/time64.h>
I'll do that this time.
- Arnado
Re: [PATCH v4 05/15] perf tools report: Parse time quantum
Quoted 9 lines · Arnaldo Carvalho de Melo
Em Fri, Mar 08, 2019 at 10:45:32AM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 8 lines
Em Tue, Mar 05, 2019 at 06:47:48AM -0800, Andi Kleen escreveu:
Review
Quoted 3 lines
From: Andi Kleen <[email protected]>
(quoted diff) +2
-0+ if (*end == 0) + return 0;We have
tools/include/linux/time64.h, just like the kernel, so please use:
Quoted 3 lines
Quoted 1 line
#include <
linux/time64.h>I'll do that this time.
Thanks. I already did it in my copy too.
-Andi
tip-bot for Andi Kleen
merge bot[tip:perf/urgent] perf report: Parse time quantum
In reply to Andi Kleenperf report: Parse time quantum
Commit-ID: 2a1292cbd4e5c81edbf815a410fa2072c341db1e Gitweb: https://git.kernel.org/tip/2a1292cbd4e5c81edbf815a410fa2072c341db1e Author: Andi Kleen <[email protected]> AuthorDate: Tue, 5 Mar 2019 06:47:48 -0800 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Mon, 11 Mar 2019 11:56:03 -0300
perf report: Parse time quantum
Many workloads change over time. 'perf report' currently aggregates the whole time range reported in perf.data.
This patch adds an option for a time quantum to quantisize the perf.data over time.
This just adds the option, will be used in follow on patches for a time sort key.
Signed-off-by Andi Kleen <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
[ Use NSEC_PER_[MU]SEC ]
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
--- tools/perf/Documentation/perf-report.txt | 4 +++ tools/perf/builtin-report.c | 42 ++++++++++++++++++++++++++++++++ tools/perf/util/symbol.c | 2 ++ tools/perf/util/symbol_conf.h | 1 + 4 files changed, 49 insertions(+)
tools/perf/Documentation/perf-report.txt +4 -0
@@ -497,6 +497,10 @@ include::itrace.txt[] The period/hits keywords set the base the percentage is computed on - the samples period or the number of samples (hits). +--time-quantum:: + Configure time quantum for time sort key. Default 100ms. + Accepts s, us, ms, ns units. + include::callchain-overhead-calculation.txt[] SEE ALSO
tools/perf/builtin-report.c +42 -0
63 lines not on this page · raw message
tools/perf/util/symbol.c +2 -0
@@ -6,6 +6,7 @@#include <string.h> #include <linux/kernel.h> #include <linux/mman.h> +#include <linux/time64.h> #include <sys/types.h> #include <sys/stat.h> #include <sys/param.h>
@@ -45,6 +46,7 @@ struct symbol_conf symbol_conf = { .demangle = true, .demangle_kernel = false, .cumulate_callchain = true, + .time_quantum = 100 * NSEC_PER_MSEC, /* 100ms */ .show_hist_headers = true, .symfs = "", .event_group = true,
tools/perf/util/symbol_conf.h +1 -0
@@ -56,6 +56,7 @@ struct symbol_conf { *sym_list_str, *col_width_list_str, *bt_stop_list_str; +unsigned long time_quantum;struct strlist *dso_list, *comm_list, *sym_list,
PATCH v4 6/15 perf tools report: Support time sort key
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Add a time sort key to perf report to display samples for different time quantums separately. This allows easier analysis of workloads that change over time, and also will allow looking at the context of samples.
% perf record ...
% perf report --sort time,overhead,symbol --time-quantum 1ms --stdio
...
0.67% 277061.87300 [.] _dl_start
0.50% 277061.87300 [.] f1
0.50% 277061.87300 [.] f2
0.33% 277061.87300 [.] main
0.29% 277061.87300 [.] _dl_lookup_symbol_x
0.29% 277061.87300 [.] dl_main
0.29% 277061.87300 [.] do_lookup_x
0.17% 277061.87300 [.] _dl_debug_initialize
0.17% 277061.87300 [.] _dl_init_paths
0.08% 277061.87300 [.] check_match
0.04% 277061.87300 [.] _dl_count_modids
1.33% 277061.87400 [.] f1
1.33% 277061.87400 [.] f2
1.33% 277061.87400 [.] main
1.17% 277061.87500 [.] main
1.08% 277061.87500 [.] f1
1.08% 277061.87500 [.] f2
1.00% 277061.87600 [.] main
0.83% 277061.87600 [.] f1
0.83% 277061.87600 [.] f2
1.00% 277061.87700 [.] mainSigned-off-by Andi Kleen <[email protected]>
---
v2:
Use symbol_conf.time_quantum
--- tools/perf/Documentation/perf-report.txt | 2 ++ tools/perf/util/hist.c | 10 +++++++ tools/perf/util/hist.h | 1 + tools/perf/util/sort.c | 38 ++++++++++++++++++++++++ tools/perf/util/sort.h | 2 ++ 5 files changed, 53 insertions(+)
tools/perf/Documentation/perf-report.txt +2 -0
@@ -105,6 +105,8 @@ OPTIONS guest machine - sample: Number of sample - period: Raw number of event count of sample + - time: Separate the samples by time stamp with the resolution specified by + --time-quantum (default 100ms). Specify with overhead and before it. By default, comm, dso and symbol keys are used. (i.e. --sort comm,dso,symbol)
tools/perf/util/hist.c +10 -0
@@ -192,6 +192,7 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *h)hists__new_col_len (hists,HISTC_MEM_LVL ,21 +3 );hists__new_col_len (hists,HISTC_LOCAL_WEIGHT ,12 );hists__new_col_len (hists,HISTC_GLOBAL_WEIGHT ,12 ); +hists__new_col_len (hists,HISTC_TIME ,12 );if (h->srcline) { len =MAX (strlen (h->srcline),strlen (sort_srcline.se_header));
@@ -246,6 +247,14 @@ static void he_stat__add_cpumode_period(struct he_stat *he_stat, } } +static long hist_time (unsigned long time) +{ +unsigned long time_quantum = symbol_conf.time_quantum; +if (time_quantum) +return (time / time_quantum) * time_quantum; +return (time /1000000 ) *1000000 ; +} +static void he_stat__add_period (struct he_stat *he_stat,u64 period,u64 weight) {
@@ -626,6 +635,7 @@ __hists__add_entry(struct hists *hists, .raw_data = sample->raw_data, .raw_size = sample->raw_size, .ops = ops, + .time =hist_time (sample->time), }, *he = hists__findnew_entry(hists, &entry, al, sample_self);if (!hists->has_callchains && he && he->callchain_size !=0 )
tools/perf/util/hist.h +1 -0
@@ -31,6 +31,7 @@ enum hist_filter {enum hist_column {HISTC_SYMBOL , +HISTC_TIME ,HISTC_DSO ,HISTC_THREAD ,HISTC_COMM ,
tools/perf/util/sort.c +38 -0
@@ -15,6 +15,7 @@#include <traceevent/event-parse.h> #include "mem-events.h" #include "annotate.h" +#include "time-utils.h" #include <linux/kernel.h> regex_t parent_regex;
@@ -654,6 +655,42 @@ struct sort_entry sort_socket = { .se_width_idx = HISTC_SOCKET, }; +/* --sort time */ + +static int64_t +sort__time_cmp (struct hist_entry *left,struct hist_entry *right) +{ +return right->time - left->time; +} + +static int hist_entry__time_snprintf (struct hist_entry *he,char *bf, +size_t size,unsigned int width) +{ +unsigned long secs; +unsigned long long nsecs; +char he_time[32 ]; + + nsecs = he->time; + secs = nsecs /1000000000 ; + nsecs -= secs *1000000000 ; + +if (symbol_conf.nanosecs) +snprintf (he_time,sizeof he_time,"%5lu.%09llu: " , + secs, nsecs); +else +timestamp__scnprintf_usec (he->time, he_time, +sizeof (he_time)); + +return repsep_snprintf (bf, size,"%-.*s" , width, he_time); +} + +struct sort_entry sort_time = { + .se_header ="Time" , + .se_cmp = sort__time_cmp, + .se_snprintf = hist_entry__time_snprintf, + .se_width_idx =HISTC_TIME , +}; +/* --sort trace */ static char *get_trace_output (struct hist_entry *he)
@@ -1634,6 +1671,7 @@ static struct sort_dimension common_sort_dimensions[] = {DIM (SORT_DSO_SIZE ,"dso_size" , sort_dso_size),DIM (SORT_CGROUP_ID ,"cgroup_id" , sort_cgroup_id),DIM (SORT_SYM_IPC_NULL ,"ipc_null" , sort_sym_ipc_null), + DIM(SORT_TIME, "time", sort_time), };#undef DIM
tools/perf/util/sort.h +2 -0
@@ -135,6 +135,7 @@ struct hist_entry {char *srcfile;struct symbol *parent;struct branch_info *branch_info; +long time;struct hists *hists;struct mem_info *mem_info;void *raw_data;
@@ -231,6 +232,7 @@ enum sort_type {SORT_DSO_SIZE ,SORT_CGROUP_ID ,SORT_SYM_IPC_NULL , +SORT_TIME ,/* branch stack specific sort keys */ __SORT_BRANCH_STACK,
Signature · 1 line
2.20.1
PATCH v4 7/15 perf tools report: Use less for scripts output
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
The UI viewer for scripts output has a lot of limitations: limited size, no search or save function, slow, and various other issues.
Just use 'less' to display directly on the terminal instead.
This won't work in gtk mode, but gtk doesn't support these context menus anyways. If that is ever done could use an terminal for the output.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Remove some unneeded headers
--- tools/perf/ui/browsers/scripts.c | 130 ++++--------------------------- 1 file changed, 17 insertions(+), 113 deletions(-)
tools/perf/ui/browsers/scripts.c +17 -113
@@ -1,35 +1,12 @@// SPDX-License-Identifier: GPL-2.0 -#include <elf.h> -#include <inttypes.h> -#include <sys/ttydefaults.h> -#include <string.h> #include "../../util/sort.h" #include "../../util/util.h" #include "../../util/hist.h" #include "../../util/debug.h" #include "../../util/symbol.h" #include "../browser.h" -#include "../helpline.h" #include "../libslang.h" -/* 2048 lines should be enough for a script output */ -#define MAX_LINES 2048 - -/* 160 bytes for one output line */ -#define AVERAGE_LINE_LEN 160 - -struct script_line { -struct list_head node; -char line[AVERAGE_LINE_LEN ]; -}; - -struct perf_script_browser { -struct ui_browser b; -struct list_head entries; -const char *script_name; -int nr_lines; -}; -#define SCRIPT_NAMELEN 128#define SCRIPT_MAX_NO 64/*
@@ -73,69 +50,29 @@ static int list_scripts(char *script_name)return ret; } -static void script_browser__write (struct ui_browser *browser, -void *entry,int row) +static void run_script (char *cmd) { - struct script_line *sline = list_entry(entry, struct script_line, node); -bool current_entry =ui_browser__is_current_entry (browser, row); - -ui_browser__set_color (browser, current_entry ?HE_COLORSET_SELECTED : -HE_COLORSET_NORMAL ); - -ui_browser__write_nstring (browser, sline->line, browser->width); +pr_debug ("Running %s\n" , cmd); +SLang_reset_tty (); +if (system (cmd) <0 ) +pr_warning ("Cannot run %s\n" , cmd); +/* +* SLang doesn't seem to reset the whole terminal, so be more +* forceful to get back to the original state. +*/ +printf ("\033[c\033[H\033[J" ); +fflush (stdout); +SLang_init_tty (0 ,0 ,0 ); +SLsmg_refresh (); } -static int script_browser__run (struct perf_script_browser *browser) -{ -int key; - -if (ui_browser__show (&browser->b, browser->script_name, -"Press ESC to exit" ) <0 ) -return -1 ; - -while (1 ) { - key =ui_browser__run (&browser->b,0 ); - -/* We can add some special key handling here if needed */ -break ; - } - -ui_browser__hide (&browser->b); -return key; -} - -int script_browse (const char *script_opt) {char cmd[SCRIPT_FULLPATH_LEN *2 ], script_name[SCRIPT_FULLPATH_LEN ]; -char *line =NULL ; -size_t len =0 ; -ssize_t retlen; -int ret =-1 , nr_entries =0 ; -FILE *fp; -void *buf; -struct script_line *sline; - -struct perf_script_browser script = { - .b = { - .refresh = ui_browser__list_head_refresh, - .seek = ui_browser__list_head_seek, - .write = script_browser__write, - }, - .script_name = script_name, - }; - -INIT_LIST_HEAD (&script.entries); - -/* Save each line of the output in one struct script_line object. */ - buf =zalloc ((sizeof (*sline)) *MAX_LINES ); -if (!buf) -return -1 ; - sline = buf;memset (script_name,0 ,SCRIPT_FULLPATH_LEN );if (list_scripts (script_name)) - goto exit; +return -1 ;sprintf (cmd,"perf script -s %s " , script_name);
@@ -147,42 +84,9 @@ int script_browse(const char *script_opt)strcat (cmd, input_name); } -strcat (cmd," 2>&1" ); - - fp =popen (cmd,"r" ); -if (!fp) - goto exit; - -while ((retlen =getline (&line, &len, fp)) !=-1 ) { -strncpy (sline->line, line,AVERAGE_LINE_LEN ); - -/* If one output line is very large, just cut it short */ -if (retlen >=AVERAGE_LINE_LEN ) { - sline->line[AVERAGE_LINE_LEN -1 ] ='\0' ; - sline->line[AVERAGE_LINE_LEN -2 ] ='\n' ; - } -list_add_tail (&sline->node, &script.entries); - -if (script.b.width < retlen) - script.b.width = retlen; - -if (nr_entries++ >=MAX_LINES -1 ) -break ; - sline++; - } - -if (script.b.width >AVERAGE_LINE_LEN ) - script.b.width =AVERAGE_LINE_LEN ; - -free (line); -pclose (fp); +strcat (cmd," 2>&1 | less" ); - script.nr_lines = nr_entries; - script.b.nr_entries = nr_entries; - script.b.entries = &script.entries; +run_script (cmd); - ret =script_browser__run (&script); -exit: -free (buf); -return ret; +return 0 ; }
Signature · 1 line
2.20.1
Re: [PATCH v4 07/15] perf tools report: Use less for scripts output
Em Tue, Mar 05, 2019 at 06:47:50AM -0800, Andi Kleen escreveu:
Quoted 6 lines
From: Andi Kleen <[email protected]>
The UI viewer for scripts output has a lot of limitations: limited size, no search or save function, slow, and various other issues.
Just use 'less' to display directly on the terminal instead.
I'm ok with this, CCing Feng tho since he contributed this browser, to let him know.
- Arnaldo
Quoted 186 lines
This won't work in gtk mode, but gtk doesn't support these context menus anyways. If that is ever done could use an terminal for the output.
Signed-off-by Andi Kleen <[email protected]>---
v2:
Remove some unneeded headers--- tools/perf/ui/browsers/scripts.c | 130 ++++--------------------------- 1 file changed, 17 insertions(+), 113 deletions(-)tools/perf/ui/browsers/scripts.c +17
-113@@ -1,35 +1,12 @@// SPDX-License-Identifier: GPL-2.0 -#include <elf.h> -#include <inttypes.h> -#include <sys/ttydefaults.h> -#include <string.h> #include "../../util/sort.h" #include "../../util/util.h" #include "../../util/hist.h" #include "../../util/debug.h" #include "../../util/symbol.h" #include "../browser.h" -#include "../helpline.h" #include "../libslang.h" -/* 2048 lines should be enough for a script output */ -#define MAX_LINES 2048 - -/* 160 bytes for one output line */ -#define AVERAGE_LINE_LEN 160 - -struct script_line { -struct list_head node; -char line[AVERAGE_LINE_LEN ]; -}; - -struct perf_script_browser { -struct ui_browser b; -struct list_head entries; -const char *script_name; -int nr_lines; -}; -#define SCRIPT_NAMELEN 128#define SCRIPT_MAX_NO 64/* @@ -73,69 +50,29 @@ static int list_scripts(char *script_name)return ret; } -static void script_browser__write (struct ui_browser *browser, -void *entry,int row) +static void run_script (char *cmd) { - struct script_line *sline = list_entry(entry, struct script_line, node); -bool current_entry =ui_browser__is_current_entry (browser, row); - -ui_browser__set_color (browser, current_entry ?HE_COLORSET_SELECTED : -HE_COLORSET_NORMAL ); - -ui_browser__write_nstring (browser, sline->line, browser->width); +pr_debug ("Running %s\n" , cmd); +SLang_reset_tty (); +if (system (cmd) <0 ) +pr_warning ("Cannot run %s\n" , cmd); +/* +* SLang doesn't seem to reset the whole terminal, so be more +* forceful to get back to the original state. +*/ +printf ("\033[c\033[H\033[J" ); +fflush (stdout); +SLang_init_tty (0 ,0 ,0 ); +SLsmg_refresh (); } -static int script_browser__run (struct perf_script_browser *browser) -{ -int key; - -if (ui_browser__show (&browser->b, browser->script_name, -"Press ESC to exit" ) <0 ) -return -1 ; - -while (1 ) { - key =ui_browser__run (&browser->b,0 ); - -/* We can add some special key handling here if needed */ -break ; - } - -ui_browser__hide (&browser->b); -return key; -} - -int script_browse (const char *script_opt) {char cmd[SCRIPT_FULLPATH_LEN *2 ], script_name[SCRIPT_FULLPATH_LEN ]; -char *line =NULL ; -size_t len =0 ; -ssize_t retlen; -int ret =-1 , nr_entries =0 ; -FILE *fp; -void *buf; -struct script_line *sline; - -struct perf_script_browser script = { - .b = { - .refresh = ui_browser__list_head_refresh, - .seek = ui_browser__list_head_seek, - .write = script_browser__write, - }, - .script_name = script_name, - }; - -INIT_LIST_HEAD (&script.entries); - -/* Save each line of the output in one struct script_line object. */ - buf =zalloc ((sizeof (*sline)) *MAX_LINES ); -if (!buf) -return -1 ; - sline = buf;memset (script_name,0 ,SCRIPT_FULLPATH_LEN );if (list_scripts (script_name)) - goto exit; +return -1 ;sprintf (cmd,"perf script -s %s " , script_name);@@ -147,42 +84,9 @@ int script_browse(const char *script_opt)strcat (cmd, input_name); } -strcat (cmd," 2>&1" ); - - fp =popen (cmd,"r" ); -if (!fp) - goto exit; - -while ((retlen =getline (&line, &len, fp)) !=-1 ) { -strncpy (sline->line, line,AVERAGE_LINE_LEN ); - -/* If one output line is very large, just cut it short */ -if (retlen >=AVERAGE_LINE_LEN ) { - sline->line[AVERAGE_LINE_LEN -1 ] ='\0' ; - sline->line[AVERAGE_LINE_LEN -2 ] ='\n' ; - } -list_add_tail (&sline->node, &script.entries); - -if (script.b.width < retlen) - script.b.width = retlen; - -if (nr_entries++ >=MAX_LINES -1 ) -break ; - sline++; - } - -if (script.b.width >AVERAGE_LINE_LEN ) - script.b.width =AVERAGE_LINE_LEN ; - -free (line); -pclose (fp); +strcat (cmd," 2>&1 | less" ); - script.nr_lines = nr_entries; - script.b.nr_entries = nr_entries; - script.b.entries = &script.entries; +run_script (cmd); - ret =script_browser__run (&script); -exit: -free (buf); -return ret; +return 0 ; }Signature · 1 line
2.20.1
Signature · 1 line
- Arnaldo
Feng Tang
Acked-byRe: [PATCH v4 07/15] perf tools report: Use less for scripts output
Quoted 10 lines · Arnaldo Carvalho de Melo
Em Tue, Mar 05, 2019 at 06:47:50AM -0800, Andi Kleen escreveu:
Quoted 6 lines
From: Andi Kleen <[email protected]>
The UI viewer for scripts output has a lot of limitations: limited size, no search or save function, slow, and various other issues.
Just use 'less' to display directly on the terminal instead.
I'm ok with this, CCing Feng tho since he contributed this browser, to let him know.
Thanks for the note!
Quoted 9 lines
- Arnaldo
Quoted 6 lines
This won't work in gtk mode, but gtk doesn't support these context menus anyways. If that is ever done could use an terminal for the output.
Signed-off-by Andi Kleen <[email protected]>
Looks good to me.
Acked-by Feng Tang <[email protected]>
Thanks,
Feng
PATCH v4 8/15 perf tools report: Support running scripts for current time range
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
When using the time sort key, add new context menus to run scripts for only the currently selected time range. Compute the correct range for the selection add pass it as the --time option to perf script.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Use symbol_conf.time_quantum
--- tools/perf/ui/browsers/hists.c | 82 +++++++++++++++++++++++++++++----- 1 file changed, 71 insertions(+), 11 deletions(-)
tools/perf/ui/browsers/hists.c +71 -11
@@ -30,6 +30,7 @@#include "srcline.h" #include "string2.h" #include "units.h" +#include "time-utils.h" #include "sane_ctype.h"
@@ -2338,6 +2339,7 @@ static int switch_data_file(void) }struct popup_action { +unsigned long time;struct thread *thread;struct map_symbol ms;int socket;
@@ -2527,36 +2529,64 @@ static int do_run_script(struct hist_browser *browser __maybe_unused, struct popup_action *act) { - char script_opt[64]; -memset (script_opt,0 ,sizeof (script_opt)); + char *script_opt; +int len; +int n =0 ; + + len =100 ; +if (act->thread) + len +=strlen (thread__comm_str (act->thread)); +else if (act->ms.sym) + len +=strlen (act->ms.sym->name); + script_opt =malloc (len); +if (!script_opt) +return -1 ; + script_opt[0 ] =0 ;if (act->thread) { -scnprintf (script_opt,sizeof (script_opt)," -c %s " , + n =scnprintf (script_opt, len," -c %s " ,thread__comm_str (act->thread)); }else if (act->ms.sym) { -scnprintf (script_opt,sizeof (script_opt)," -S %s " , + n =scnprintf (script_opt, len," -S %s " , act->ms.sym->name); } +if (act->time) { +char start[64 ], end[64 ]; +unsigned long starttime = act->time; +unsigned long endtime = act->time + symbol_conf.time_quantum; + +if (starttime == endtime) {/* Display 1ms as fallback */ + starttime -=1 *1000000 ; + endtime +=1 *1000000 ; + } +timestamp__scnprintf_usec (starttime, start,sizeof start); +timestamp__scnprintf_usec (endtime, end,sizeof end); + n +=snprintf (script_opt + n, len - n," --time %s,%s" , start, end); + } +script_browse (script_opt); +free (script_opt);return 0 ; }static int -add_script_opt (struct hist_browser *browser__maybe_unused , +add_script_opt_2 (struct hist_browser *browser__maybe_unused ,struct popup_action *act,char **optstr, -struct thread *thread,struct symbol *sym) +struct thread *thread,struct symbol *sym, +const char *tstr) { +if (thread) { -if (asprintf (optstr,"Run scripts for samples of thread [%s]" , -thread__comm_str (thread)) <0 ) +if (asprintf (optstr,"Run scripts for samples of thread [%s]%s" , +thread__comm_str (thread), tstr) <0 )return 0 ; }else if (sym) { -if (asprintf (optstr,"Run scripts for samples of symbol [%s]" , - sym->name) <0 ) +if (asprintf (optstr,"Run scripts for samples of symbol [%s]%s" , + sym->name, tstr) <0 )return 0 ; }else { -if (asprintf (optstr,"Run scripts for all samples" ) <0 ) +if (asprintf (optstr,"Run scripts for all samples%s" , tstr) <0 )return 0 ; }
@@ -2566,6 +2596,36 @@ add_script_opt(struct hist_browser *browser __maybe_unused,return 1 ; } +static int +add_script_opt (struct hist_browser *browser, +struct popup_action *act,char **optstr, +struct thread *thread,struct symbol *sym) +{ +int n, j; +struct hist_entry *he; + + n =add_script_opt_2 (browser, act, optstr, thread, sym,"" ); + + he =hist_browser__selected_entry (browser); +if (sort_order &&strstr (sort_order,"time" )) { +char tstr[128 ]; + + optstr++; + act++; + j =sprintf (tstr," in " ); + j +=timestamp__scnprintf_usec (he->time, tstr + j, +sizeof tstr - j); + j +=sprintf (tstr + j,"-" ); +timestamp__scnprintf_usec (he->time + symbol_conf.time_quantum, + tstr + j, +sizeof tstr - j); + n +=add_script_opt_2 (browser, act, optstr, thread, sym, + tstr); + act->time = he->time; + } +return n; +} +static int do_switch_data (struct hist_browser *browser__maybe_unused ,struct popup_action *act__maybe_unused )
Signature · 1 line
2.20.1
PATCH v4 9/15 perf tools report: Support builtin perf script in scripts menu
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
The scripts menu traditionally only showed custom perf scripts.
Allow to run standard perf script with useful default options too.
- Normal perf script
- perf script with assembler (needs xed installed)
- perf script with source code output (needs debuginfo)
- perf script with custom arguments
Then we automatically select the right options to
display the information in the perf.data file.
For example with -b display branch contexts.
It's not easily possible to check for xed's existence
in advance. perf script usually gives sensible error messages when it's not available.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Pass --inline if needed
v3:
Avoid compiler warnings. Allocate cmd buffer dynamically. Change formatting
--- tools/perf/ui/browsers/annotate.c | 2 +- tools/perf/ui/browsers/hists.c | 23 +++--- tools/perf/ui/browsers/scripts.c | 127 +++++++++++++++++++++++------- tools/perf/util/hist.h | 8 +- 4 files changed, 120 insertions(+), 40 deletions(-)
tools/perf/ui/browsers/annotate.c +1 -1
@@ -750,7 +750,7 @@ static int annotate_browser__run(struct annotate_browser *browser,continue ;case 'r' : { -script_browse (NULL ); +script_browse (NULL ,NULL );continue ; }case 'k' :
tools/perf/ui/browsers/hists.c +13 -10
@@ -2343,6 +2343,7 @@ struct popup_action {struct thread *thread;struct map_symbol ms;int socket; +struct perf_evsel *evsel;int (*fn)(struct hist_browser *browser,struct popup_action *act); };
@@ -2565,7 +2566,7 @@ do_run_script(struct hist_browser *browser __maybe_unused, n +=snprintf (script_opt + n, len - n," --time %s,%s" , start, end); } -script_browse (script_opt); +script_browse (script_opt, act->evsel);free (script_opt);return 0 ; }
@@ -2574,7 +2575,7 @@ static int add_script_opt_2(struct hist_browser *browser __maybe_unused, struct popup_action *act, char **optstr, struct thread *thread, struct symbol *sym, - const char *tstr) + struct perf_evsel *evsel, const char *tstr) { if (thread) {
@@ -2592,6 +2593,7 @@ add_script_opt_2(struct hist_browser *browser __maybe_unused, act->thread = thread; act->ms.sym = sym; + act->evsel = evsel; act->fn = do_run_script;return 1 ; }
@@ -2599,12 +2601,13 @@ add_script_opt_2(struct hist_browser *browser __maybe_unused,static int add_script_opt (struct hist_browser *browser,struct popup_action *act,char **optstr, -struct thread *thread,struct symbol *sym) +struct thread *thread,struct symbol *sym, +struct perf_evsel *evsel) {int n, j;struct hist_entry *he; - n =add_script_opt_2 (browser, act, optstr, thread, sym,"" ); + n =add_script_opt_2 (browser, act, optstr, thread, sym, evsel,"" ); he =hist_browser__selected_entry (browser);if (sort_order &&strstr (sort_order,"time" )) {
@@ -2617,10 +2620,9 @@ add_script_opt(struct hist_browser *browser, sizeof tstr - j); j +=sprintf (tstr + j,"-" );timestamp__scnprintf_usec (he->time + symbol_conf.time_quantum, - tstr + j, -sizeof tstr - j); + tstr + j,sizeof tstr - j); n +=add_script_opt_2 (browser, act, optstr, thread, sym, - tstr); + evsel, tstr); act->time = he->time; }return n;
@@ -3091,7 +3093,7 @@ static int perf_evsel__hists_browse(struct perf_evsel *evsel, int nr_events, nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options], - thread,NULL ); + thread,NULL , evsel); }/* * Note that browser->selection != NULL
@@ -3106,11 +3108,12 @@ static int perf_evsel__hists_browse(struct perf_evsel *evsel, int nr_events, nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options], -NULL , browser->selection->sym); +NULL , browser->selection->sym, + evsel); } } nr_options +=add_script_opt (browser, &actions[nr_options], - &options[nr_options],NULL ,NULL ); + &options[nr_options],NULL ,NULL , evsel); nr_options +=add_switch_opt (browser, &actions[nr_options], &options[nr_options]); skip_scripting:
tools/perf/ui/browsers/scripts.c +100 -27
163 lines not on this page · raw message
tools/perf/util/hist.h +6 -2
@@ -436,6 +436,8 @@ struct annotation_options;#ifdef HAVE_SLANG_SUPPORT #include "../ui/keysyms.h" +void attr_to_script (char *buf,struct perf_event_attr *attr); +int map_symbol__tui_annotate (struct map_symbol *ms,struct perf_evsel *evsel,struct hist_browser_timer *hbt,struct annotation_options *annotation_opts);
@@ -450,7 +452,8 @@ int perf_evlist__tui_browse_hists(struct perf_evlist *evlist, const char *help,struct perf_env *env, bool warn_lost_event, struct annotation_options *annotation_options); -int script_browse (const char *script_opt); + +int script_browse (const char *script_opt,struct perf_evsel *evsel);#else static inline int perf_evlist__tui_browse_hists(struct perf_evlist *evlist __maybe_unused,
@@ -479,7 +482,8 @@ static inline int hist_entry__tui_annotate(struct hist_entry *he __maybe_unused,return 0 ; } -static inline int script_browse (const char *script_opt__maybe_unused ) +static inline int script_browse (const char *script_opt__maybe_unused , +struct perf_evsel *evsel__maybe_unused ) {return 0 ; }
Signature · 1 line
2.20.1
PATCH v4 10/15 perf tools: Add utility function to print ns time stamps
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Add a utility function to print nanosecond timestamps.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/util/time-utils.c | 8 ++++++++ tools/perf/util/time-utils.h | 1 + 2 files changed, 9 insertions(+)
tools/perf/util/time-utils.c +8 -0
@@ -404,6 +404,14 @@ int timestamp__scnprintf_usec(u64 timestamp, char *buf, size_t sz)return scnprintf (buf, sz,"%" PRIu64".%06" PRIu64, sec, usec); } +int timestamp__scnprintf_nsec (u64 timestamp,char *buf,size_t sz) +{ +u64 sec = timestamp /NSEC_PER_SEC ; +u64 nsec = timestamp %NSEC_PER_SEC ; + +return scnprintf (buf, sz,"%" PRIu64".%09" PRIu64, sec, nsec); +} +int fetch_current_timestamp (char *buf,size_t sz) {struct timeval tv;
tools/perf/util/time-utils.h +1 -0
@@ -24,6 +24,7 @@ bool perf_time__ranges_skip_sample(struct perf_time_interval *ptime_buf, int num, u64 timestamp);int timestamp__scnprintf_usec (u64 timestamp,char *buf,size_t sz); +int timestamp__scnprintf_nsec (u64 timestamp,char *buf,size_t sz);int fetch_current_timestamp (char *buf,size_t sz);
Signature · 1 line
2.20.1
Arnaldo Carvalho de Melo
AppliedRe: [PATCH v4 10/15] perf tools: Add utility function to print ns time stamps
Em Tue, Mar 05, 2019 at 06:47:53AM -0800, Andi Kleen escreveu:
Quoted 3 lines
From: Andi Kleen <[email protected]>
Add a utility function to print nanosecond timestamps.
Applied.
- Arnaldo
tip-bot for Andi Kleen
merge bot[tip:perf/urgent] perf time-utils: Add utility function to print time stamps in nanoseconds
In reply to Andi Kleenperf time-utils: Add utility function to print time stamps in nanoseconds
Commit-ID: f8c856cb2c947f4fad0a2dff5e95cdcddb801303 Gitweb: https://git.kernel.org/tip/f8c856cb2c947f4fad0a2dff5e95cdcddb801303 Author: Andi Kleen <[email protected]> AuthorDate: Tue, 5 Mar 2019 06:47:53 -0800 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Mon, 11 Mar 2019 11:56:02 -0300
perf time-utils: Add utility function to print time stamps in nanoseconds
Add a utility function to print nanosecond timestamps.
Signed-off-by Andi Kleen <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
--- tools/perf/util/time-utils.c | 8 ++++++++ tools/perf/util/time-utils.h | 1 + 2 files changed, 9 insertions(+)
tools/perf/util/time-utils.c +8 -0
@@ -453,6 +453,14 @@ int timestamp__scnprintf_usec(u64 timestamp, char *buf, size_t sz)return scnprintf (buf, sz,"%" PRIu64".%06" PRIu64, sec, usec); } +int timestamp__scnprintf_nsec (u64 timestamp,char *buf,size_t sz) +{ +u64 sec = timestamp /NSEC_PER_SEC , + nsec = timestamp %NSEC_PER_SEC ; + +return scnprintf (buf, sz,"%" PRIu64".%09" PRIu64, sec, nsec); +} +int fetch_current_timestamp (char *buf,size_t sz) {struct timeval tv;
tools/perf/util/time-utils.h +1 -0
@@ -30,6 +30,7 @@ int perf_time__parse_for_ranges(const char *str, struct perf_session *session, int *range_size, int *range_num);int timestamp__scnprintf_usec (u64 timestamp,char *buf,size_t sz); +int timestamp__scnprintf_nsec (u64 timestamp,char *buf,size_t sz);int fetch_current_timestamp (char *buf,size_t sz);
PATCH v4 11/15 perf tools report: Implement browsing of individual samples
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Now report can show whole time periods with perf script, but the user still has to find individual samples of interest manually.
It would be expensive and complicated to search for the right samples in the whole perf file. Typically users only need to look at a small number of samples for useful analysis.
Also the full scripts tend to show samples of all CPUs and all threads mixed up, which can be very confusing on larger systems.
Add a new --samples option to save a small random number of samples per hist entry
Use a reservoir sample technique to select a representatve number of samples.
Then allow browsing the samples using perf script
as part of the hist entry context menu. This automatically adds the right filters, so only the thread or cpu of the sample is displayed. Then we use less' search functionality to directly jump the to the time stamp of the selected sample.
It uses different menus for assembler and source display. Assembler needs xed installed and source needs debuginfo.
Currently it only supports as many samples as fit on the screen due to some limitations in the slang ui code.
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Free names on error path
Pass --inline and --show-*-event to child perf as needed. v3:
Don't use num_res to keep track of total samples Don't pass -1 cpus to perf script.
Don't pass extra event option with tid filtering. Fix sampling bug.
--- tools/perf/Documentation/perf-report.txt | 4 ++ tools/perf/builtin-report.c | 2 + tools/perf/ui/browsers/Build | 1 + tools/perf/ui/browsers/hists.c | 45 +++++++++++++ tools/perf/ui/browsers/res_sample.c | 81 ++++++++++++++++++++++++ tools/perf/ui/browsers/scripts.c | 2 +- tools/perf/util/hist.c | 36 +++++++++++ tools/perf/util/hist.h | 19 ++++++ tools/perf/util/sort.h | 8 +++ tools/perf/util/symbol.c | 1 + tools/perf/util/symbol_conf.h | 1 + 11 files changed, 199 insertions(+), 1 deletion(-) create mode 100644 tools/perf/ui/browsers/res_sample.c
tools/perf/Documentation/perf-report.txt +4 -0
@@ -461,6 +461,10 @@ include::itrace.txt[] --socket-filter:: Only report the samples on the processor socket that match with this filter +--samples=N:: + Save N individual samples for each histogram entry to show context in perf + report tui browser. + --raw-trace:: When displaying traceevent output, do not use print fmt or plugins.
tools/perf/builtin-report.c +2 -0
@@ -1158,6 +1158,8 @@ int cmd_report(int argc, const char **argv)OPT_BOOLEAN (0 ,"demangle-kernel" , &symbol_conf.demangle_kernel,"Enable kernel symbol demangling" ),OPT_BOOLEAN (0 ,"mem-mode" , &report.mem_mode,"mem access profile" ), +OPT_INTEGER (0 ,"samples" , &symbol_conf.res_sample, +"Number of samples to save per histogram entry for individual browsing" ),OPT_CALLBACK (0 ,"percent-limit" , &report,"percent" ,"Don't show entries under that percent" , parse_percent_limit), OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
tools/perf/ui/browsers/Build +1 -0
@@ -3,6 +3,7 @@ perf-y += hists.o perf-y += map.o perf-y += scripts.o perf-y += header.o +perf-y += res_sample.o CFLAGS_annotate.o += -DENABLE_SLFUTURE_CONST CFLAGS_hists.o += -DENABLE_SLFUTURE_CONST
tools/perf/ui/browsers/hists.c +45 -0
69 lines not on this page · raw message
tools/perf/ui/browsers/res_sample.c new file +81 -0
81 lines not on this page · raw message
tools/perf/ui/browsers/scripts.c +1 -1
@@ -125,7 +125,7 @@ static int list_scripts(char *script_name, bool *custom,return ret; } -static void run_script (char *cmd) +void run_script (char *cmd) {pr_debug ("Running %s\n" , cmd);SLang_reset_tty ();
tools/perf/util/hist.c +36 -0
@@ -444,6 +444,14 @@ static int hist_entry__init(struct hist_entry *he,return -ENOMEM ; } } + +if (symbol_conf.res_sample) { + he->res_samples =calloc (sizeof (struct res_sample ), + symbol_conf.res_sample); +if (!he->res_samples) +return -ENOMEM ; + } +INIT_LIST_HEAD (&he->pairs.node);thread__get (he->thread); he->hroot_in =RB_ROOT_CACHED ;
@@ -593,6 +601,32 @@ static struct hist_entry *hists__findnew_entry(struct hists *hists,return he; } +static unsigned random_max (unsigned high) +{ +unsigned thresh = -high % high; +for (;;) { +unsigned r =random (); +if (r >= thresh) +return r % high; + } +} + +static void hists__res_sample (struct hist_entry *he,struct perf_sample *sample) +{ +struct res_sample *r; +int j; + +if (he->num_res < symbol_conf.res_sample) { + j = he->num_res++; + }else { + j =random_max (symbol_conf.res_sample); + } + r = &he->res_samples[j]; + r->time = sample->time; + r->cpu = sample->cpu; + r->tid = sample->tid; +} +static struct hist_entry * __hists__add_entry(struct hists *hists, struct addr_location *al,
@@ -640,6 +674,8 @@ __hists__add_entry(struct hists *hists,if (!hists->has_callchains && he && he->callchain_size !=0 ) hists->has_callchains = true; +if (he && symbol_conf.res_sample) +hists__res_sample (he, sample);return he; }
tools/perf/util/hist.h +19 -0
@@ -433,6 +433,13 @@ struct hist_browser_timer { };struct annotation_options ; +struct res_sample ; + +enum rstype { +A_NORMAL , +A_ASM , +A_SOURCE +};#ifdef HAVE_SLANG_SUPPORT #include "../ui/keysyms.h"
@@ -454,6 +461,10 @@ int perf_evlist__tui_browse_hists(struct perf_evlist *evlist, const char *help, struct annotation_options *annotation_options);int script_browse (const char *script_opt,struct perf_evsel *evsel); + +void run_script (char *cmd); +int res_sample_browse (struct res_sample *res_samples,int num_res, +struct perf_evsel *evsel,enum rstype rstype);#else static inline int perf_evlist__tui_browse_hists(struct perf_evlist *evlist __maybe_unused,
@@ -488,6 +499,14 @@ static inline int script_browse(const char *script_opt __maybe_unused,return 0 ; } +static inline int res_sample_browse (struct res_sample *res_samples__maybe_unused , +int num_res__maybe_unused , +struct perf_evsel *evsel__maybe_unused , +enum rstype rstype__maybe_unused ) +{ +return 0 ; +} +#define K_LEFT -1000#define K_RIGHT -2000#define K_SWITCH_INPUT_DATA -3000
tools/perf/util/sort.h +8 -0
@@ -47,6 +47,12 @@ extern struct sort_entry sort_srcline;extern enum sort_type sort__first_dimension;extern const char default_mem_sort_order[]; +struct res_sample { +u64 time; +int cpu; +int tid; +}; +struct he_stat {u64 period;u64 period_sys;
@@ -140,6 +146,8 @@ struct hist_entry {struct mem_info *mem_info;void *raw_data;u32 raw_size; +int num_res; +struct res_sample *res_samples;void *trace_output;struct perf_hpp_list *hpp_list;struct hist_entry *parent_he;
tools/perf/util/symbol.c +1 -0
@@ -50,6 +50,7 @@ struct symbol_conf symbol_conf = { .symfs = "", .event_group = true, .inline_name = true, + .res_sample = 0, };static enum dso_binary_type binary_type_symtab[] = {
tools/perf/util/symbol_conf.h +1 -0
@@ -68,6 +68,7 @@ struct symbol_conf {struct intlist *pid_list, *tid_list;const char *symfs; +int res_sample; };extern struct symbol_conf symbol_conf;
Signature · 1 line
2.20.1
PATCH v4 12/15 perf tools: Add some new tips describing the new options
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Signed-off-by Andi Kleen <[email protected]>
---
v2:
Even more tips.
--- tools/perf/Documentation/tips.txt | 6 ++++++ 1 file changed, 6 insertions(+)
tools/perf/Documentation/tips.txt +6 -0
@@ -15,6 +15,7 @@ To see callchains in a more compact form: perf report -g folded Show individual samples with: perf script Limit to show entries above 5% only: perf report --percent-limit 5 Profiling branch (mis)predictions with: perf record -b / perf report +To show assembler sample contexts use perf record -b / perf script -F +brstackinsn --xed Treat branches as callchains: perf report --branch-history To count events in every 1000 msec: perf stat -I 1000 Print event counts in CSV format with: perf stat -x,
@@ -34,3 +35,8 @@ Show current config key-value pairs: perf config --list Show user configuration overrides: perf config --user --list To add Node.js USDT(User-Level Statically Defined Tracing): perf buildid-cache --add `which node` To report cacheline events from previous recording: perf c2c report +To browse sample contexts use perf report --sample 10 and select in context menu +To separate samples by time use perf report --sort time,overhead,sym +To set sample time separation other than 100ms with --sort time use --time-quantum +Add -I to perf report to sample register values visible in perf report context. +To show IPC for sampling periods use perf record -e '{cycles,instructions}:S' and then browse context
Signature · 1 line
2.20.1
PATCH v4 13/15 perf tools report: Add custom scripts to script menu
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Add a way to define custom scripts through ~/.perfconfig, which are then added to the scripts menu. The scripts get the same arguments as perf script, in particular -i, --cpu, --tid.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/Documentation/perf-config.txt | 8 ++++++++ tools/perf/ui/browsers/scripts.c | 20 ++++++++++++++++++++ 2 files changed, 28 insertions(+)
tools/perf/Documentation/perf-config.txt +8 -0
@@ -584,6 +584,14 @@ llvm.*:: llvm.opts:: Options passed to llc. +script.*:: + + Any option defines a script that is added to the scripts menu + in the interactive perf browser and whose output is displayed. + The name of the option is the name, the value is a script command line. + The script gets the same options passed as a full perf script, + in particular -i perfdata file, --cpu, --tid + SEE ALSO -------- linkperf:perf[1]
tools/perf/ui/browsers/scripts.c +20 -0
@@ -6,6 +6,7 @@#include "../../util/symbol.h" #include "../browser.h" #include "../libslang.h" +#include "config.h" #define SCRIPT_NAMELEN 128#define SCRIPT_MAX_NO 64
@@ -53,6 +54,24 @@ static int add_script_option(const char *name, const char *opt,return 0 ; } +static int scripts_config (const char *var,const char *value,void *data) +{ +struct script_config *c = data; + +if (!strstarts (var,"script." )) +return -1 ; +if (c->index >=SCRIPT_MAX_NO ) +return -1 ; + c->names[c->index] =strdup (var +7 ); +if (!c->names[c->index]) +return -1 ; +if (asprintf (&c->paths[c->index],"%s %s" , value, + c->extra_format) <0 ) +return -1 ; + c->index++; +return 0 ; +} +/* * When success, will copy the full path of the selected script * into the buffer pointed by script_name, and return 0.
@@ -87,6 +106,7 @@ static int list_scripts(char *script_name, bool *custom, &scriptc);add_script_option ("Show individual samples with source" ,"-F +srcline,+srccode" , &scriptc); +perf_config (scripts_config, &scriptc); custom_perf = scriptc.index;add_script_option ("Show samples with custom perf script arguments" ,"" , &scriptc); i = scriptc.index;
Signature · 1 line
2.20.1
PATCH v4 14/15 perf tools script: Add array bound checking to list_scripts
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Don't overflow array when the scripts directory is too large, or the script file name is too long.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/builtin-script.c | 8 ++++++-- tools/perf/builtin.h | 3 ++- tools/perf/ui/browsers/scripts.c | 3 ++- 3 files changed, 10 insertions(+), 4 deletions(-)
tools/perf/builtin-script.c +6 -2
@@ -2983,7 +2983,8 @@ static int check_ev_match(char *dir_name, char *scriptname,* will list all statically runnable scripts, select one, execute it and * show the output in a perf browser. */ -int find_scripts (char **scripts_array,char **scripts_path_array) +int find_scripts (char **scripts_array,char **scripts_path_array,int num, +int pathlen) {struct dirent *script_dirent, *lang_dirent;char scripts_path[MAXPATHLEN ], lang_path[MAXPATHLEN ];
@@ -3028,7 +3029,10 @@ int find_scripts(char **scripts_array, char **scripts_path_array)/* Skip those real time scripts: xxxtop.p[yl] */ if (strstr (script_dirent->d_name,"top." ))continue ; -sprintf (scripts_path_array[i],"%s/%s" , lang_path, +if (i >= num) +break ; +snprintf (scripts_path_array[i], pathlen,"%s/%s" , + lang_path, script_dirent->d_name); temp =strchr (script_dirent->d_name,'.' ); snprintf(scripts_array[i],
tools/perf/builtin.h +2 -1
@@ -40,5 +40,6 @@ int cmd_mem(int argc, const char **argv);int cmd_data (int argc,const char **argv);int cmd_ftrace (int argc,const char **argv); -int find_scripts (char **scripts_array,char **scripts_path_array); +int find_scripts (char **scripts_array,char **scripts_path_array,int num, +int pathlen);#endif
tools/perf/ui/browsers/scripts.c +2 -1
@@ -117,7 +117,8 @@ static int list_scripts(char *script_name, bool *custom, paths[i] = names[i] +SCRIPT_NAMELEN ; } - num =find_scripts (names + max_std, paths + max_std); + num =find_scripts (names + max_std, paths + max_std,SCRIPT_MAX_NO - max_std, +SCRIPT_FULLPATH_LEN );if (num <0 ) num =0 ; choice =ui__popup_menu (num + max_std, (char *const *)names);
Signature · 1 line
2.20.1
PATCH v4 15/15 perf tools ui: Fix ui popup browser for many entries
In reply to Andi KleenFrom: Andi Kleen <[email protected]>
Fix the argv ui browser code to correctly display more entries than fit on the screen without crashing. The problem was some type confusion with pointer types in the ->seek function. Do the argv arithmetic correctly with char ** pointers. Also add some asserts to find overruns and limit the display function correctly.
Then finally remove a workaround for this in the res sample browser.
Signed-off-by Andi Kleen <[email protected]>
--- tools/perf/ui/browser.c | 10 +++++++--- tools/perf/ui/browsers/res_sample.c | 3 --- 2 files changed, 7 insertions(+), 6 deletions(-)
tools/perf/ui/browser.c +7 -3
@@ -611,14 +611,16 @@ void ui_browser__argv_seek(struct ui_browser *browser, off_t offset, int whence) browser->top = browser->entries;break ;case SEEK_CUR : - browser->top = browser->top + browser->top_idx + offset; + browser->top = (char **)browser->top + offset;break ;case SEEK_END : - browser->top = browser->top + browser->nr_entries -1 + offset; + browser->top = (char **)browser->entries + browser->nr_entries -1 + offset;break ;default :return ; } +assert ((char **)browser->top < (char **)browser->entries + browser->nr_entries); +assert ((char **)browser->top >= (char **)browser->entries); }unsigned int ui_browser__argv_refresh (struct ui_browser *browser)
@@ -630,7 +632,9 @@ unsigned int ui_browser__argv_refresh(struct ui_browser *browser) browser->top = browser->entries; pos = (char **)browser->top; -while (idx < browser->nr_entries) { +while (idx < browser->nr_entries && + row < (unsigned )SLtt_Screen_Rows -1 ) { +assert (pos < (char **)browser->entries + browser->nr_entries);if (!browser->filter || !browser->filter (browser, *pos)) {ui_browser__gotorc (browser, row,0 ); browser->write (browser, pos, row);
tools/perf/ui/browsers/res_sample.c +0 -3
@@ -23,9 +23,6 @@ int res_sample_browse(struct res_sample *res_samples, int num_res,struct res_sample *r;char extra_format[256 ]; -/* For now since ui__popup_menu doesn't like lists that don't fit */ - num_res =max (min (SLtt_Screen_Rows -4 , num_res),0 ); - names =calloc (num_res,sizeof (char *));if (!names)return -1 ;
Signature · 1 line
2.20.1
Quoted 9 lines · Andi Kleen
[Changes:
v4:
Address review comments.
Fix --cpu filtering.
Fix a sampling bug.
Add support for configuring custom script menu entries in perfconfig. Fix display of more samples than fit on screen.
Fix some buffer overruns in legacy code.
Add more tips
hi,
getting gcc error on your branch, similar like last time:
ui/browsers/hists.c: In function ‘perf_evsel__hists_browse’:ui/browsers/hists.c:2567:8: error: ‘%s’ directive output may be truncated writing up to 63 bytes into a region of size between 28 and 91 [-Werror=format-truncation=]
n += snprintf(script_opt + n, len - n, " --time %s,%s", start, end);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~In file included from /usr/include/stdio.h:862,
from ui/browsers/hists.c:5:/usr/include/bits/stdio2.h:64:10: note: ‘__builtin___snprintf_chk’ output between 10 and 136 bytes into a destination of size 100
return __builtin___snprintf_chk (__s, __n, __USE_FORTIFY_LEVEL - 1,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
__bos (__s), __fmt, __va_arg_pack ());
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~cc1: all warnings being treated as errors
jirka
Quoted 11 lines · Jiri Olsa
Quoted 9 lines · Andi Kleen
[Changes:
v4:
Address review comments.
Fix --cpu filtering.
Fix a sampling bug.
Add support for configuring custom script menu entries in perfconfig. Fix display of more samples than fit on screen.
Fix some buffer overruns in legacy code.
Add more tips
Quoted 2 lines
hi,
getting gcc error on your branch, similar like last time:
Okay I figured out now why I'm not seeing this. It's because my perf builds are with DEBUG=1, and Makefile.config has this insanity:
ifeq ($(DEBUG),0)
ifeq ($(feature-fortify-source), 1)
CFLAGS += -D_FORTIFY_SOURCE=2
endif
endif
Anyways the warnings are false positives because the strings can never be that big. In fact I think they're harmful because it discourages adding safety margins to stack buffers.
Anyways you can use this patch as a workaround.
tools/perf/ui/browsers/hists.c +1 -1
@@ -2554,7 +2554,7 @@ do_run_script(struct hist_browser *browser __maybe_unused, }if (act->time) { -char start[64 ], end[64 ]; +char start[32 ], end[32 ];unsigned long starttime = act->time;unsigned long endtime = act->time + symbol_conf.time_quantum;