linux/tools/perf/util/synthetic-events.c

2288 lines
55 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
#include "util/cgroup.h"
#include "util/data.h"
#include "util/debug.h"
#include "util/dso.h"
#include "util/event.h"
#include "util/evlist.h"
#include "util/machine.h"
#include "util/map.h"
#include "util/map_symbol.h"
#include "util/branch.h"
#include "util/memswap.h"
#include "util/namespaces.h"
#include "util/session.h"
#include "util/stat.h"
#include "util/symbol.h"
#include "util/synthetic-events.h"
#include "util/target.h"
#include "util/time-utils.h"
#include <linux/bitops.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/zalloc.h>
#include <linux/perf_event.h>
#include <asm/bug.h>
#include <perf/evsel.h>
#include <perf/cpumap.h>
#include <internal/lib.h> // page_size
#include <internal/threadmap.h>
#include <perf/threadmap.h>
#include <symbol/kallsyms.h>
#include <dirent.h>
#include <errno.h>
#include <inttypes.h>
#include <stdio.h>
#include <string.h>
#include <uapi/linux/mman.h> /* To get things like MAP_HUGETLB even on older libc headers */
#include <api/fs/fs.h>
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
#include <api/io.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#define DEFAULT_PROC_MAP_PARSE_TIMEOUT 500
unsigned int proc_map_timeout = DEFAULT_PROC_MAP_PARSE_TIMEOUT;
int perf_tool__process_synth_event(struct perf_tool *tool,
union perf_event *event,
struct machine *machine,
perf_event__handler_t process)
{
struct perf_sample synth_sample = {
.pid = -1,
.tid = -1,
.time = -1,
.stream_id = -1,
.cpu = -1,
.period = 1,
.cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK,
};
return process(tool, event, &synth_sample, machine);
};
/*
* Assumes that the first 4095 bytes of /proc/pid/stat contains
* the comm, tgid and ppid.
*/
static int perf_event__get_comm_ids(pid_t pid, pid_t tid, char *comm, size_t len,
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
pid_t *tgid, pid_t *ppid, bool *kernel)
{
char bf[4096];
int fd;
size_t size = 0;
ssize_t n;
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
char *name, *tgids, *ppids, *vmpeak, *threads;
*tgid = -1;
*ppid = -1;
if (pid)
snprintf(bf, sizeof(bf), "/proc/%d/task/%d/status", pid, tid);
else
snprintf(bf, sizeof(bf), "/proc/%d/status", tid);
fd = open(bf, O_RDONLY);
if (fd < 0) {
pr_debug("couldn't open %s\n", bf);
return -1;
}
n = read(fd, bf, sizeof(bf) - 1);
close(fd);
if (n <= 0) {
pr_warning("Couldn't get COMM, tigd and ppid for pid %d\n",
tid);
return -1;
}
bf[n] = '\0';
name = strstr(bf, "Name:");
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
tgids = strstr(name ?: bf, "Tgid:");
ppids = strstr(tgids ?: bf, "PPid:");
vmpeak = strstr(ppids ?: bf, "VmPeak:");
if (vmpeak)
threads = NULL;
else
threads = strstr(ppids ?: bf, "Threads:");
if (name) {
char *nl;
name = skip_spaces(name + 5); /* strlen("Name:") */
nl = strchr(name, '\n');
if (nl)
*nl = '\0';
size = strlen(name);
if (size >= len)
size = len - 1;
memcpy(comm, name, size);
comm[size] = '\0';
} else {
pr_debug("Name: string not found for pid %d\n", tid);
}
if (tgids) {
tgids += 5; /* strlen("Tgid:") */
*tgid = atoi(tgids);
} else {
pr_debug("Tgid: string not found for pid %d\n", tid);
}
if (ppids) {
ppids += 5; /* strlen("PPid:") */
*ppid = atoi(ppids);
} else {
pr_debug("PPid: string not found for pid %d\n", tid);
}
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
if (!vmpeak && threads)
*kernel = true;
else
*kernel = false;
return 0;
}
static int perf_event__prepare_comm(union perf_event *event, pid_t pid, pid_t tid,
struct machine *machine,
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
pid_t *tgid, pid_t *ppid, bool *kernel)
{
size_t size;
*ppid = -1;
memset(&event->comm, 0, sizeof(event->comm));
if (machine__is_host(machine)) {
if (perf_event__get_comm_ids(pid, tid, event->comm.comm,
sizeof(event->comm.comm),
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
tgid, ppid, kernel) != 0) {
return -1;
}
} else {
*tgid = machine->pid;
}
if (*tgid < 0)
return -1;
event->comm.pid = *tgid;
event->comm.header.type = PERF_RECORD_COMM;
size = strlen(event->comm.comm) + 1;
size = PERF_ALIGN(size, sizeof(u64));
memset(event->comm.comm + size, 0, machine->id_hdr_size);
event->comm.header.size = (sizeof(event->comm) -
(sizeof(event->comm.comm) - size) +
machine->id_hdr_size);
event->comm.tid = tid;
return 0;
}
pid_t perf_event__synthesize_comm(struct perf_tool *tool,
union perf_event *event, pid_t pid,
perf_event__handler_t process,
struct machine *machine)
{
pid_t tgid, ppid;
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
bool kernel_thread;
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
if (perf_event__prepare_comm(event, 0, pid, machine, &tgid, &ppid,
&kernel_thread) != 0)
return -1;
if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
return -1;
return tgid;
}
static void perf_event__get_ns_link_info(pid_t pid, const char *ns,
struct perf_ns_link_info *ns_link_info)
{
struct stat64 st;
char proc_ns[128];
sprintf(proc_ns, "/proc/%u/ns/%s", pid, ns);
if (stat64(proc_ns, &st) == 0) {
ns_link_info->dev = st.st_dev;
ns_link_info->ino = st.st_ino;
}
}
int perf_event__synthesize_namespaces(struct perf_tool *tool,
union perf_event *event,
pid_t pid, pid_t tgid,
perf_event__handler_t process,
struct machine *machine)
{
u32 idx;
struct perf_ns_link_info *ns_link_info;
if (!tool || !tool->namespace_events)
return 0;
memset(&event->namespaces, 0, (sizeof(event->namespaces) +
(NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
machine->id_hdr_size));
event->namespaces.pid = tgid;
event->namespaces.tid = pid;
event->namespaces.nr_namespaces = NR_NAMESPACES;
ns_link_info = event->namespaces.link_info;
for (idx = 0; idx < event->namespaces.nr_namespaces; idx++)
perf_event__get_ns_link_info(pid, perf_ns__name(idx),
&ns_link_info[idx]);
event->namespaces.header.type = PERF_RECORD_NAMESPACES;
event->namespaces.header.size = (sizeof(event->namespaces) +
(NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
machine->id_hdr_size);
if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
return -1;
return 0;
}
static int perf_event__synthesize_fork(struct perf_tool *tool,
union perf_event *event,
pid_t pid, pid_t tgid, pid_t ppid,
perf_event__handler_t process,
struct machine *machine)
{
memset(&event->fork, 0, sizeof(event->fork) + machine->id_hdr_size);
/*
* for main thread set parent to ppid from status file. For other
* threads set parent pid to main thread. ie., assume main thread
* spawns all threads in a process
*/
if (tgid == pid) {
event->fork.ppid = ppid;
event->fork.ptid = ppid;
} else {
event->fork.ppid = tgid;
event->fork.ptid = tgid;
}
event->fork.pid = tgid;
event->fork.tid = pid;
event->fork.header.type = PERF_RECORD_FORK;
event->fork.header.misc = PERF_RECORD_MISC_FORK_EXEC;
event->fork.header.size = (sizeof(event->fork) + machine->id_hdr_size);
if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
return -1;
return 0;
}
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
static bool read_proc_maps_line(struct io *io, __u64 *start, __u64 *end,
u32 *prot, u32 *flags, __u64 *offset,
u32 *maj, u32 *min,
__u64 *inode,
ssize_t pathname_size, char *pathname)
{
__u64 temp;
int ch;
char *start_pathname = pathname;
if (io__get_hex(io, start) != '-')
return false;
if (io__get_hex(io, end) != ' ')
return false;
/* map protection and flags bits */
*prot = 0;
ch = io__get_char(io);
if (ch == 'r')
*prot |= PROT_READ;
else if (ch != '-')
return false;
ch = io__get_char(io);
if (ch == 'w')
*prot |= PROT_WRITE;
else if (ch != '-')
return false;
ch = io__get_char(io);
if (ch == 'x')
*prot |= PROT_EXEC;
else if (ch != '-')
return false;
ch = io__get_char(io);
if (ch == 's')
*flags = MAP_SHARED;
else if (ch == 'p')
*flags = MAP_PRIVATE;
else
return false;
if (io__get_char(io) != ' ')
return false;
if (io__get_hex(io, offset) != ' ')
return false;
if (io__get_hex(io, &temp) != ':')
return false;
*maj = temp;
if (io__get_hex(io, &temp) != ' ')
return false;
*min = temp;
ch = io__get_dec(io, inode);
if (ch != ' ') {
*pathname = '\0';
return ch == '\n';
}
do {
ch = io__get_char(io);
} while (ch == ' ');
while (true) {
if (ch < 0)
return false;
if (ch == '\0' || ch == '\n' ||
(pathname + 1 - start_pathname) >= pathname_size) {
*pathname = '\0';
return true;
}
*pathname++ = ch;
ch = io__get_char(io);
}
}
static void perf_record_mmap2__read_build_id(struct perf_record_mmap2 *event,
bool is_kernel)
{
struct build_id bid;
int rc;
if (is_kernel)
rc = sysfs__read_build_id("/sys/kernel/notes", &bid);
else
rc = filename__read_build_id(event->filename, &bid) > 0 ? 0 : -1;
if (rc == 0) {
memcpy(event->build_id, bid.data, sizeof(bid.data));
event->build_id_size = (u8) bid.size;
event->header.misc |= PERF_RECORD_MISC_MMAP_BUILD_ID;
event->__reserved_1 = 0;
event->__reserved_2 = 0;
} else {
if (event->filename[0] == '/') {
pr_debug2("Failed to read build ID for %s\n",
event->filename);
}
}
}
int perf_event__synthesize_mmap_events(struct perf_tool *tool,
union perf_event *event,
pid_t pid, pid_t tgid,
perf_event__handler_t process,
struct machine *machine,
bool mmap_data)
{
unsigned long long t;
char bf[BUFSIZ];
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
struct io io;
bool truncation = false;
unsigned long long timeout = proc_map_timeout * 1000000ULL;
int rc = 0;
const char *hugetlbfs_mnt = hugetlbfs__mountpoint();
int hugetlbfs_mnt_len = hugetlbfs_mnt ? strlen(hugetlbfs_mnt) : 0;
if (machine__is_default_guest(machine))
return 0;
snprintf(bf, sizeof(bf), "%s/proc/%d/task/%d/maps",
machine->root_dir, pid, pid);
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
io.fd = open(bf, O_RDONLY, 0);
if (io.fd < 0) {
/*
* We raced with a task exiting - just return:
*/
pr_debug("couldn't open %s\n", bf);
return -1;
}
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
io__init(&io, io.fd, bf, sizeof(bf));
event->header.type = PERF_RECORD_MMAP2;
t = rdclock();
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
while (!io.eof) {
static const char anonstr[] = "//anon";
size_t size, aligned_size;
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
/* ensure null termination since stack will be reused. */
event->mmap2.filename[0] = '\0';
/* 00400000-0040c000 r-xp 00000000 fd:01 41038 /bin/cat */
if (!read_proc_maps_line(&io,
&event->mmap2.start,
&event->mmap2.len,
&event->mmap2.prot,
&event->mmap2.flags,
&event->mmap2.pgoff,
&event->mmap2.maj,
&event->mmap2.min,
&event->mmap2.ino,
sizeof(event->mmap2.filename),
event->mmap2.filename))
continue;
if ((rdclock() - t) > timeout) {
pr_warning("Reading %s/proc/%d/task/%d/maps time out. "
"You may want to increase "
"the time limit by --proc-map-timeout\n",
machine->root_dir, pid, pid);
truncation = true;
goto out;
}
perf tools: Give synthetic mmap events an inode generation When mmap2 events are synthesized the ino_generation field isn't being set leading to uninitialized memory being compared. Caught with clang's -fsanitize=memory: ==124733==WARNING: MemorySanitizer: use-of-uninitialized-value #0 0x55a96a6a65cc in __dso_id__cmp tools/perf/util/dsos.c:23:6 #1 0x55a96a6a81d5 in dso_id__cmp tools/perf/util/dsos.c:38:9 #2 0x55a96a6a717f in __dso__cmp_long_name tools/perf/util/dsos.c:74:15 #3 0x55a96a6a6c4c in __dsos__findnew_link_by_longname_id tools/perf/util/dsos.c:106:12 #4 0x55a96a6a851e in __dsos__findnew_by_longname_id tools/perf/util/dsos.c:178:9 #5 0x55a96a6a7798 in __dsos__find_id tools/perf/util/dsos.c:191:9 #6 0x55a96a6a7b57 in __dsos__findnew_id tools/perf/util/dsos.c:251:20 #7 0x55a96a6a7a57 in dsos__findnew_id tools/perf/util/dsos.c:259:17 #8 0x55a96a7776ae in machine__findnew_dso_id tools/perf/util/machine.c:2709:9 #9 0x55a96a77dfcf in map__new tools/perf/util/map.c:193:10 #10 0x55a96a77240a in machine__process_mmap2_event tools/perf/util/machine.c:1670:8 #11 0x55a96a7741a3 in machine__process_event tools/perf/util/machine.c:1882:9 #12 0x55a96a6aee39 in perf_event__process tools/perf/util/event.c:454:9 #13 0x55a96a87d633 in perf_tool__process_synth_event tools/perf/util/synthetic-events.c:63:9 #14 0x55a96a87f131 in perf_event__synthesize_mmap_events tools/perf/util/synthetic-events.c:403:7 #15 0x55a96a8815d6 in __event__synthesize_thread tools/perf/util/synthetic-events.c:548:9 #16 0x55a96a882bff in __perf_event__synthesize_threads tools/perf/util/synthetic-events.c:681:3 #17 0x55a96a881ec2 in perf_event__synthesize_threads tools/perf/util/synthetic-events.c:750:9 #18 0x55a96a562b26 in synth_all tools/perf/tests/mmap-thread-lookup.c:136:9 #19 0x55a96a5623b1 in mmap_events tools/perf/tests/mmap-thread-lookup.c:174:8 #20 0x55a96a561fa0 in test__mmap_thread_lookup tools/perf/tests/mmap-thread-lookup.c:230:2 #21 0x55a96a52c182 in run_test tools/perf/tests/builtin-test.c:378:9 #22 0x55a96a52afc1 in test_and_print tools/perf/tests/builtin-test.c:408:9 #23 0x55a96a52966e in __cmd_test tools/perf/tests/builtin-test.c:603:4 #24 0x55a96a52855d in cmd_test tools/perf/tests/builtin-test.c:747:9 #25 0x55a96a2844d4 in run_builtin tools/perf/perf.c:312:11 #26 0x55a96a282bd0 in handle_internal_command tools/perf/perf.c:364:8 #27 0x55a96a284097 in run_argv tools/perf/perf.c:408:2 #28 0x55a96a282223 in main tools/perf/perf.c:538:3 Uninitialized value was stored to memory at #1 0x55a96a6a18f7 in dso__new_id tools/perf/util/dso.c:1230:14 #2 0x55a96a6a78ee in __dsos__addnew_id tools/perf/util/dsos.c:233:20 #3 0x55a96a6a7bcc in __dsos__findnew_id tools/perf/util/dsos.c:252:21 #4 0x55a96a6a7a57 in dsos__findnew_id tools/perf/util/dsos.c:259:17 #5 0x55a96a7776ae in machine__findnew_dso_id tools/perf/util/machine.c:2709:9 #6 0x55a96a77dfcf in map__new tools/perf/util/map.c:193:10 #7 0x55a96a77240a in machine__process_mmap2_event tools/perf/util/machine.c:1670:8 #8 0x55a96a7741a3 in machine__process_event tools/perf/util/machine.c:1882:9 #9 0x55a96a6aee39 in perf_event__process tools/perf/util/event.c:454:9 #10 0x55a96a87d633 in perf_tool__process_synth_event tools/perf/util/synthetic-events.c:63:9 #11 0x55a96a87f131 in perf_event__synthesize_mmap_events tools/perf/util/synthetic-events.c:403:7 #12 0x55a96a8815d6 in __event__synthesize_thread tools/perf/util/synthetic-events.c:548:9 #13 0x55a96a882bff in __perf_event__synthesize_threads tools/perf/util/synthetic-events.c:681:3 #14 0x55a96a881ec2 in perf_event__synthesize_threads tools/perf/util/synthetic-events.c:750:9 #15 0x55a96a562b26 in synth_all tools/perf/tests/mmap-thread-lookup.c:136:9 #16 0x55a96a5623b1 in mmap_events tools/perf/tests/mmap-thread-lookup.c:174:8 #17 0x55a96a561fa0 in test__mmap_thread_lookup tools/perf/tests/mmap-thread-lookup.c:230:2 #18 0x55a96a52c182 in run_test tools/perf/tests/builtin-test.c:378:9 #19 0x55a96a52afc1 in test_and_print tools/perf/tests/builtin-test.c:408:9 Uninitialized value was stored to memory at #0 0x55a96a7725af in machine__process_mmap2_event tools/perf/util/machine.c:1646:25 #1 0x55a96a7741a3 in machine__process_event tools/perf/util/machine.c:1882:9 #2 0x55a96a6aee39 in perf_event__process tools/perf/util/event.c:454:9 #3 0x55a96a87d633 in perf_tool__process_synth_event tools/perf/util/synthetic-events.c:63:9 #4 0x55a96a87f131 in perf_event__synthesize_mmap_events tools/perf/util/synthetic-events.c:403:7 #5 0x55a96a8815d6 in __event__synthesize_thread tools/perf/util/synthetic-events.c:548:9 #6 0x55a96a882bff in __perf_event__synthesize_threads tools/perf/util/synthetic-events.c:681:3 #7 0x55a96a881ec2 in perf_event__synthesize_threads tools/perf/util/synthetic-events.c:750:9 #8 0x55a96a562b26 in synth_all tools/perf/tests/mmap-thread-lookup.c:136:9 #9 0x55a96a5623b1 in mmap_events tools/perf/tests/mmap-thread-lookup.c:174:8 #10 0x55a96a561fa0 in test__mmap_thread_lookup tools/perf/tests/mmap-thread-lookup.c:230:2 #11 0x55a96a52c182 in run_test tools/perf/tests/builtin-test.c:378:9 #12 0x55a96a52afc1 in test_and_print tools/perf/tests/builtin-test.c:408:9 #13 0x55a96a52966e in __cmd_test tools/perf/tests/builtin-test.c:603:4 #14 0x55a96a52855d in cmd_test tools/perf/tests/builtin-test.c:747:9 #15 0x55a96a2844d4 in run_builtin tools/perf/perf.c:312:11 #16 0x55a96a282bd0 in handle_internal_command tools/perf/perf.c:364:8 #17 0x55a96a284097 in run_argv tools/perf/perf.c:408:2 #18 0x55a96a282223 in main tools/perf/perf.c:538:3 Uninitialized value was created by a heap allocation #0 0x55a96a22f60d in malloc llvm/llvm-project/compiler-rt/lib/msan/msan_interceptors.cpp:925:3 #1 0x55a96a882948 in __perf_event__synthesize_threads tools/perf/util/synthetic-events.c:655:15 #2 0x55a96a881ec2 in perf_event__synthesize_threads tools/perf/util/synthetic-events.c:750:9 #3 0x55a96a562b26 in synth_all tools/perf/tests/mmap-thread-lookup.c:136:9 #4 0x55a96a5623b1 in mmap_events tools/perf/tests/mmap-thread-lookup.c:174:8 #5 0x55a96a561fa0 in test__mmap_thread_lookup tools/perf/tests/mmap-thread-lookup.c:230:2 #6 0x55a96a52c182 in run_test tools/perf/tests/builtin-test.c:378:9 #7 0x55a96a52afc1 in test_and_print tools/perf/tests/builtin-test.c:408:9 #8 0x55a96a52966e in __cmd_test tools/perf/tests/builtin-test.c:603:4 #9 0x55a96a52855d in cmd_test tools/perf/tests/builtin-test.c:747:9 #10 0x55a96a2844d4 in run_builtin tools/perf/perf.c:312:11 #11 0x55a96a282bd0 in handle_internal_command tools/perf/perf.c:364:8 #12 0x55a96a284097 in run_argv tools/perf/perf.c:408:2 #13 0x55a96a282223 in main tools/perf/perf.c:538:3 SUMMARY: MemorySanitizer: use-of-uninitialized-value tools/perf/util/dsos.c:23:6 in __dso_id__cmp Signed-off-by: Ian Rogers <irogers@google.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: clang-built-linux@googlegroups.com Link: http://lore.kernel.org/lkml/20200313053129.131264-1-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-03-13 13:31:29 +08:00
event->mmap2.ino_generation = 0;
/*
* Just like the kernel, see __perf_event_mmap in kernel/perf_event.c
*/
if (machine__is_host(machine))
event->header.misc = PERF_RECORD_MISC_USER;
else
event->header.misc = PERF_RECORD_MISC_GUEST_USER;
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
if ((event->mmap2.prot & PROT_EXEC) == 0) {
if (!mmap_data || (event->mmap2.prot & PROT_READ) == 0)
continue;
event->header.misc |= PERF_RECORD_MISC_MMAP_DATA;
}
out:
if (truncation)
event->header.misc |= PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT;
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
if (!strcmp(event->mmap2.filename, ""))
strcpy(event->mmap2.filename, anonstr);
if (hugetlbfs_mnt_len &&
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
!strncmp(event->mmap2.filename, hugetlbfs_mnt,
hugetlbfs_mnt_len)) {
strcpy(event->mmap2.filename, anonstr);
event->mmap2.flags |= MAP_HUGETLB;
}
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
size = strlen(event->mmap2.filename) + 1;
aligned_size = PERF_ALIGN(size, sizeof(u64));
event->mmap2.len -= event->mmap.start;
event->mmap2.header.size = (sizeof(event->mmap2) -
(sizeof(event->mmap2.filename) - aligned_size));
memset(event->mmap2.filename + size, 0, machine->id_hdr_size +
(aligned_size - size));
event->mmap2.header.size += machine->id_hdr_size;
event->mmap2.pid = tgid;
event->mmap2.tid = pid;
if (symbol_conf.buildid_mmap2)
perf_record_mmap2__read_build_id(&event->mmap2, false);
if (perf_tool__process_synth_event(tool, event, machine, process) != 0) {
rc = -1;
break;
}
if (truncation)
break;
}
perf synthetic events: Remove use of sscanf from /proc reading The synthesize benchmark, run on a single process and thread, shows perf_event__synthesize_mmap_events as the hottest function with fgets and sscanf taking the majority of execution time. fscanf performs similarly well. Replace the scanf call with manual reading of each field of the /proc/pid/maps line, and remove some unnecessary buffering. This change also addresses potential, but unlikely, buffer overruns for the string values read by scanf. Performance before is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 102.810 usec (+- 0.027 usec) Average num. events: 17.000 (+- 0.000) Average time per event 6.048 usec Average data synthesis took: 106.325 usec (+- 0.018 usec) Average num. events: 89.000 (+- 0.000) Average time per event 1.195 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 68103.100 usec (+- 441.234 usec) Average num. events: 30703.000 (+- 0.730) Average time per event 2.218 usec And after is: $ sudo perf bench internals synthesize -m 16 -M 16 -s -t \# Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 50.388 usec (+- 0.031 usec) Average num. events: 17.000 (+- 0.000) Average time per event 2.964 usec Average data synthesis took: 52.693 usec (+- 0.020 usec) Average num. events: 89.000 (+- 0.000) Average time per event 0.592 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 16 Average synthesis took: 45022.400 usec (+- 552.740 usec) Average num. events: 30624.200 (+- 10.037) Average time per event 1.470 usec On a Intel Xeon 6154 compiling with Debian gcc 9.2.1. Committer testing: On a AMD Ryzen 5 3600X 6-Core Processor: Before: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 267.491 usec (+- 0.176 usec) Average num. events: 56.000 (+- 0.000) Average time per event 4.777 usec Average data synthesis took: 277.257 usec (+- 0.169 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.966 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 81599.500 usec (+- 346.315 usec) Average num. events: 36096.100 (+- 2.523) Average time per event 2.261 usec # After: # perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 110.125 usec (+- 0.080 usec) Average num. events: 56.000 (+- 0.000) Average time per event 1.967 usec Average data synthesis took: 118.518 usec (+- 0.057 usec) Average num. events: 287.000 (+- 0.000) Average time per event 0.413 usec Computing performance of multi threaded perf event synthesis by synthesizing events on CPU 0: Number of synthesis threads: 12 Average synthesis took: 43490.700 usec (+- 284.527 usec) Average num. events: 37028.500 (+- 0.563) Average time per event 1.175 usec # Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andrey Zhizhikin <andrey.z@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Petr Mladek <pmladek@suse.com> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-04-15 13:40:50 +08:00
close(io.fd);
return rc;
}
#ifdef HAVE_FILE_HANDLE
static int perf_event__synthesize_cgroup(struct perf_tool *tool,
union perf_event *event,
char *path, size_t mount_len,
perf_event__handler_t process,
struct machine *machine)
{
size_t event_size = sizeof(event->cgroup) - sizeof(event->cgroup.path);
size_t path_len = strlen(path) - mount_len + 1;
struct {
struct file_handle fh;
uint64_t cgroup_id;
} handle;
int mount_id;
while (path_len % sizeof(u64))
path[mount_len + path_len++] = '\0';
memset(&event->cgroup, 0, event_size);
event->cgroup.header.type = PERF_RECORD_CGROUP;
event->cgroup.header.size = event_size + path_len + machine->id_hdr_size;
handle.fh.handle_bytes = sizeof(handle.cgroup_id);
if (name_to_handle_at(AT_FDCWD, path, &handle.fh, &mount_id, 0) < 0) {
pr_debug("stat failed: %s\n", path);
return -1;
}
event->cgroup.id = handle.cgroup_id;
strncpy(event->cgroup.path, path + mount_len, path_len);
memset(event->cgroup.path + path_len, 0, machine->id_hdr_size);
if (perf_tool__process_synth_event(tool, event, machine, process) < 0) {
pr_debug("process synth event failed\n");
return -1;
}
return 0;
}
static int perf_event__walk_cgroup_tree(struct perf_tool *tool,
union perf_event *event,
char *path, size_t mount_len,
perf_event__handler_t process,
struct machine *machine)
{
size_t pos = strlen(path);
DIR *d;
struct dirent *dent;
int ret = 0;
if (perf_event__synthesize_cgroup(tool, event, path, mount_len,
process, machine) < 0)
return -1;
d = opendir(path);
if (d == NULL) {
pr_debug("failed to open directory: %s\n", path);
return -1;
}
while ((dent = readdir(d)) != NULL) {
if (dent->d_type != DT_DIR)
continue;
if (!strcmp(dent->d_name, ".") ||
!strcmp(dent->d_name, ".."))
continue;
/* any sane path should be less than PATH_MAX */
if (strlen(path) + strlen(dent->d_name) + 1 >= PATH_MAX)
continue;
if (path[pos - 1] != '/')
strcat(path, "/");
strcat(path, dent->d_name);
ret = perf_event__walk_cgroup_tree(tool, event, path,
mount_len, process, machine);
if (ret < 0)
break;
path[pos] = '\0';
}
closedir(d);
return ret;
}
int perf_event__synthesize_cgroups(struct perf_tool *tool,
perf_event__handler_t process,
struct machine *machine)
{
union perf_event event;
char cgrp_root[PATH_MAX];
size_t mount_len; /* length of mount point in the path */
perf record: Synthesize cgroup events only if needed It didn't check the tool->cgroup_events bit which is set when the --all-cgroups option is given. Without it, samples will not have cgroup info so no reason to synthesize. We can check the PERF_RECORD_CGROUP records after running perf record *WITHOUT* the --all-cgroups option: Before: $ perf report -D | grep CGROUP 0 0 0x8430 [0x38]: PERF_RECORD_CGROUP cgroup: 1 / CGROUP events: 1 CGROUP events: 0 CGROUP events: 0 After: $ perf report -D | grep CGROUP CGROUP events: 0 CGROUP events: 0 CGROUP events: 0 Committer testing: Before: # perf record -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 2.208 MB perf.data (10003 samples) ] # perf report -D | grep "CGROUP events" CGROUP events: 146 CGROUP events: 0 CGROUP events: 0 # After: # perf record -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 2.208 MB perf.data (10448 samples) ] # perf report -D | grep "CGROUP events" CGROUP events: 0 CGROUP events: 0 CGROUP events: 0 # With all-cgroups: # perf record --all-cgroups -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 2.374 MB perf.data (11526 samples) ] # perf report -D | grep "CGROUP events" CGROUP events: 146 CGROUP events: 0 CGROUP events: 0 # Fixes: 8fb4b67939e16 ("perf record: Add --all-cgroups option") Signed-off-by: Namhyung Kim <namhyung@kernel.org> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lore.kernel.org/lkml/20201127054356.405481-1-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-11-27 13:43:56 +08:00
if (!tool || !tool->cgroup_events)
return 0;
if (cgroupfs_find_mountpoint(cgrp_root, PATH_MAX, "perf_event") < 0) {
pr_debug("cannot find cgroup mount point\n");
return -1;
}
mount_len = strlen(cgrp_root);
/* make sure the path starts with a slash (after mount point) */
strcat(cgrp_root, "/");
if (perf_event__walk_cgroup_tree(tool, &event, cgrp_root, mount_len,
process, machine) < 0)
return -1;
return 0;
}
#else
int perf_event__synthesize_cgroups(struct perf_tool *tool __maybe_unused,
perf_event__handler_t process __maybe_unused,
struct machine *machine __maybe_unused)
{
return -1;
}
#endif
int perf_event__synthesize_modules(struct perf_tool *tool, perf_event__handler_t process,
struct machine *machine)
{
int rc = 0;
struct map *pos;
struct maps *maps = machine__kernel_maps(machine);
union perf_event *event;
size_t size = symbol_conf.buildid_mmap2 ?
sizeof(event->mmap2) : sizeof(event->mmap);
event = zalloc(size + machine->id_hdr_size);
if (event == NULL) {
pr_debug("Not enough memory synthesizing mmap event "
"for kernel modules\n");
return -1;
}
/*
* kernel uses 0 for user space maps, see kernel/perf_event.c
* __perf_event_mmap
*/
if (machine__is_host(machine))
event->header.misc = PERF_RECORD_MISC_KERNEL;
else
event->header.misc = PERF_RECORD_MISC_GUEST_KERNEL;
maps__for_each_entry(maps, pos) {
if (!__map__is_kmodule(pos))
continue;
if (symbol_conf.buildid_mmap2) {
size = PERF_ALIGN(pos->dso->long_name_len + 1, sizeof(u64));
event->mmap2.header.type = PERF_RECORD_MMAP2;
event->mmap2.header.size = (sizeof(event->mmap2) -
(sizeof(event->mmap2.filename) - size));
memset(event->mmap2.filename + size, 0, machine->id_hdr_size);
event->mmap2.header.size += machine->id_hdr_size;
event->mmap2.start = pos->start;
event->mmap2.len = pos->end - pos->start;
event->mmap2.pid = machine->pid;
memcpy(event->mmap2.filename, pos->dso->long_name,
pos->dso->long_name_len + 1);
perf_record_mmap2__read_build_id(&event->mmap2, false);
} else {
size = PERF_ALIGN(pos->dso->long_name_len + 1, sizeof(u64));
event->mmap.header.type = PERF_RECORD_MMAP;
event->mmap.header.size = (sizeof(event->mmap) -
(sizeof(event->mmap.filename) - size));
memset(event->mmap.filename + size, 0, machine->id_hdr_size);
event->mmap.header.size += machine->id_hdr_size;
event->mmap.start = pos->start;
event->mmap.len = pos->end - pos->start;
event->mmap.pid = machine->pid;
memcpy(event->mmap.filename, pos->dso->long_name,
pos->dso->long_name_len + 1);
}
if (perf_tool__process_synth_event(tool, event, machine, process) != 0) {
rc = -1;
break;
}
}
free(event);
return rc;
}
static int filter_task(const struct dirent *dirent)
{
return isdigit(dirent->d_name[0]);
}
static int __event__synthesize_thread(union perf_event *comm_event,
union perf_event *mmap_event,
union perf_event *fork_event,
union perf_event *namespaces_event,
pid_t pid, int full, perf_event__handler_t process,
struct perf_tool *tool, struct machine *machine,
bool needs_mmap, bool mmap_data)
{
char filename[PATH_MAX];
struct dirent **dirent;
pid_t tgid, ppid;
int rc = 0;
int i, n;
/* special case: only send one comm event using passed in pid */
if (!full) {
tgid = perf_event__synthesize_comm(tool, comm_event, pid,
process, machine);
if (tgid == -1)
return -1;
if (perf_event__synthesize_namespaces(tool, namespaces_event, pid,
tgid, process, machine) < 0)
return -1;
/*
* send mmap only for thread group leader
* see thread__init_maps()
*/
if (pid == tgid && needs_mmap &&
perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
process, machine, mmap_data))
return -1;
return 0;
}
if (machine__is_default_guest(machine))
return 0;
snprintf(filename, sizeof(filename), "%s/proc/%d/task",
machine->root_dir, pid);
n = scandir(filename, &dirent, filter_task, NULL);
if (n < 0)
return n;
for (i = 0; i < n; i++) {
char *end;
pid_t _pid;
2021-03-09 19:04:47 +08:00
bool kernel_thread = false;
_pid = strtol(dirent[i]->d_name, &end, 10);
if (*end)
continue;
/* some threads may exit just after scan, ignore it */
if (perf_event__prepare_comm(comm_event, pid, _pid, machine,
perf tools: Skip PERF_RECORD_MMAP event synthesis for kernel threads To synthesize information to resolve sample IPs, it needs to scan task and mmap info from the /proc filesystem. For each process, it opens (and reads) status and maps file respectively. But as kernel threads don't have memory maps so we can skip the maps file. To find kernel threads, check "VmPeak:" line in /proc/<PID>/status file. It's about the peak virtual memory usage so only user-level tasks have that. Note that it's possible to miss the line due to partial reads. So we should double-check if it's a really kernel thread when there's no VmPeak line. Thus check "Threads:" line (which follows the VmPeak line whether or not it exists) to be sure it's read enough data - just in case of deeply nested pid namespaces or large number of supplementary groups are involved. This is for user process: $ head -40 /proc/1/status Name: systemd Umask: 0000 State: S (sleeping) Tgid: 1 Ngid: 0 Pid: 1 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 256 Groups: NStgid: 1 NSpid: 1 NSpgid: 1 NSsid: 1 VmPeak: 234192 kB <-- here VmSize: 169964 kB VmLck: 0 kB VmPin: 0 kB VmHWM: 29528 kB VmRSS: 6104 kB RssAnon: 2756 kB RssFile: 3348 kB RssShmem: 0 kB VmData: 19776 kB VmStk: 1036 kB VmExe: 784 kB VmLib: 9532 kB VmPTE: 116 kB VmSwap: 2400 kB HugetlbPages: 0 kB CoreDumping: 0 THP_enabled: 1 Threads: 1 <-- and here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 7be3c0fe28014a03 SigIgn: 0000000000001000 And this is for kernel thread: $ head -20 /proc/2/status Name: kthreadd Umask: 0000 State: S (sleeping) Tgid: 2 Ngid: 0 Pid: 2 PPid: 0 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: NStgid: 2 NSpid: 2 NSpgid: 0 NSsid: 0 Threads: 1 <-- here SigQ: 1/62808 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Jiri Olsa <jolsa@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20210202090118.2008551-3-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-02 17:01:17 +08:00
&tgid, &ppid, &kernel_thread) != 0)
continue;
rc = -1;
if (perf_event__synthesize_fork(tool, fork_event, _pid, tgid,
ppid, process, machine) < 0)
break;
if (perf_event__synthesize_namespaces(tool, namespaces_event, _pid,
tgid, process, machine) < 0)
break;
/*
* Send the prepared comm event
*/
if (perf_tool__process_synth_event(tool, comm_event, machine, process) != 0)
break;
rc = 0;
if (_pid == pid && !kernel_thread && needs_mmap) {
/* process the parent's maps too */
rc = perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
process, machine, mmap_data);
if (rc)
break;
}
}
for (i = 0; i < n; i++)
zfree(&dirent[i]);
free(dirent);
return rc;
}
int perf_event__synthesize_thread_map(struct perf_tool *tool,
struct perf_thread_map *threads,
perf_event__handler_t process,
struct machine *machine,
bool needs_mmap, bool mmap_data)
{
union perf_event *comm_event, *mmap_event, *fork_event;
union perf_event *namespaces_event;
int err = -1, thread, j;
comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
if (comm_event == NULL)
goto out;
mmap_event = malloc(sizeof(mmap_event->mmap2) + machine->id_hdr_size);
if (mmap_event == NULL)
goto out_free_comm;
fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
if (fork_event == NULL)
goto out_free_mmap;
namespaces_event = malloc(sizeof(namespaces_event->namespaces) +
(NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
machine->id_hdr_size);
if (namespaces_event == NULL)
goto out_free_fork;
err = 0;
for (thread = 0; thread < threads->nr; ++thread) {
if (__event__synthesize_thread(comm_event, mmap_event,
fork_event, namespaces_event,
perf_thread_map__pid(threads, thread), 0,
process, tool, machine,
needs_mmap, mmap_data)) {
err = -1;
break;
}
/*
* comm.pid is set to thread group id by
* perf_event__synthesize_comm
*/
if ((int) comm_event->comm.pid != perf_thread_map__pid(threads, thread)) {
bool need_leader = true;
/* is thread group leader in thread_map? */
for (j = 0; j < threads->nr; ++j) {
if ((int) comm_event->comm.pid == perf_thread_map__pid(threads, j)) {
need_leader = false;
break;
}
}
/* if not, generate events for it */
if (need_leader &&
__event__synthesize_thread(comm_event, mmap_event,
fork_event, namespaces_event,
comm_event->comm.pid, 0,
process, tool, machine,
needs_mmap, mmap_data)) {
err = -1;
break;
}
}
}
free(namespaces_event);
out_free_fork:
free(fork_event);
out_free_mmap:
free(mmap_event);
out_free_comm:
free(comm_event);
out:
return err;
}
static int __perf_event__synthesize_threads(struct perf_tool *tool,
perf_event__handler_t process,
struct machine *machine,
bool needs_mmap,
bool mmap_data,
struct dirent **dirent,
int start,
int num)
{
union perf_event *comm_event, *mmap_event, *fork_event;
union perf_event *namespaces_event;
int err = -1;
char *end;
pid_t pid;
int i;
comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
if (comm_event == NULL)
goto out;
mmap_event = malloc(sizeof(mmap_event->mmap2) + machine->id_hdr_size);
if (mmap_event == NULL)
goto out_free_comm;
fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
if (fork_event == NULL)
goto out_free_mmap;
namespaces_event = malloc(sizeof(namespaces_event->namespaces) +
(NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
machine->id_hdr_size);
if (namespaces_event == NULL)
goto out_free_fork;
for (i = start; i < start + num; i++) {
if (!isdigit(dirent[i]->d_name[0]))
continue;
pid = (pid_t)strtol(dirent[i]->d_name, &end, 10);
/* only interested in proper numerical dirents */
if (*end)
continue;
/*
* We may race with exiting thread, so don't stop just because
* one thread couldn't be synthesized.
*/
__event__synthesize_thread(comm_event, mmap_event, fork_event,
namespaces_event, pid, 1, process,
tool, machine, needs_mmap, mmap_data);
}
err = 0;
free(namespaces_event);
out_free_fork:
free(fork_event);
out_free_mmap:
free(mmap_event);
out_free_comm:
free(comm_event);
out:
return err;
}
struct synthesize_threads_arg {
struct perf_tool *tool;
perf_event__handler_t process;
struct machine *machine;
bool needs_mmap;
bool mmap_data;
struct dirent **dirent;
int num;
int start;
};
static void *synthesize_threads_worker(void *arg)
{
struct synthesize_threads_arg *args = arg;
__perf_event__synthesize_threads(args->tool, args->process,
args->machine,
args->needs_mmap, args->mmap_data,
args->dirent,
args->start, args->num);
return NULL;
}
int perf_event__synthesize_threads(struct perf_tool *tool,
perf_event__handler_t process,
struct machine *machine,
bool needs_mmap, bool mmap_data,
unsigned int nr_threads_synthesize)
{
struct synthesize_threads_arg *args = NULL;
pthread_t *synthesize_threads = NULL;
char proc_path[PATH_MAX];
struct dirent **dirent;
int num_per_thread;
int m, n, i, j;
int thread_nr;
int base = 0;
int err = -1;
if (machine__is_default_guest(machine))
return 0;
snprintf(proc_path, sizeof(proc_path), "%s/proc", machine->root_dir);
n = scandir(proc_path, &dirent, filter_task, NULL);
if (n < 0)
return err;
if (nr_threads_synthesize == UINT_MAX)
thread_nr = sysconf(_SC_NPROCESSORS_ONLN);
else
thread_nr = nr_threads_synthesize;
if (thread_nr <= 1) {
err = __perf_event__synthesize_threads(tool, process,
machine,
needs_mmap, mmap_data,
dirent, base, n);
goto free_dirent;
}
if (thread_nr > n)
thread_nr = n;
synthesize_threads = calloc(sizeof(pthread_t), thread_nr);
if (synthesize_threads == NULL)
goto free_dirent;
args = calloc(sizeof(*args), thread_nr);
if (args == NULL)
goto free_threads;
num_per_thread = n / thread_nr;
m = n % thread_nr;
for (i = 0; i < thread_nr; i++) {
args[i].tool = tool;
args[i].process = process;
args[i].machine = machine;
args[i].needs_mmap = needs_mmap;
args[i].mmap_data = mmap_data;
args[i].dirent = dirent;
}
for (i = 0; i < m; i++) {
args[i].num = num_per_thread + 1;
args[i].start = i * args[i].num;
}
if (i != 0)
base = args[i-1].start + args[i-1].num;
for (j = i; j < thread_nr; j++) {
args[j].num = num_per_thread;
args[j].start = base + (j - i) * args[i].num;
}
for (i = 0; i < thread_nr; i++) {
if (pthread_create(&synthesize_threads[i], NULL,
synthesize_threads_worker, &args[i]))
goto out_join;
}
err = 0;
out_join:
for (i = 0; i < thread_nr; i++)
pthread_join(synthesize_threads[i], NULL);
free(args);
free_threads:
free(synthesize_threads);
free_dirent:
for (i = 0; i < n; i++)
zfree(&dirent[i]);
free(dirent);
return err;
}
int __weak perf_event__synthesize_extra_kmaps(struct perf_tool *tool __maybe_unused,
perf_event__handler_t process __maybe_unused,
struct machine *machine __maybe_unused)
{
return 0;
}
static int __perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
perf_event__handler_t process,
struct machine *machine)
{
union perf_event *event;
size_t size = symbol_conf.buildid_mmap2 ?
sizeof(event->mmap2) : sizeof(event->mmap);
struct map *map = machine__kernel_map(machine);
struct kmap *kmap;
int err;
if (map == NULL)
return -1;
kmap = map__kmap(map);
if (!kmap->ref_reloc_sym)
return -1;
/*
* We should get this from /sys/kernel/sections/.text, but till that is
* available use this, and after it is use this as a fallback for older
* kernels.
*/
event = zalloc(size + machine->id_hdr_size);
if (event == NULL) {
pr_debug("Not enough memory synthesizing mmap event "
"for kernel modules\n");
return -1;
}
if (machine__is_host(machine)) {
/*
* kernel uses PERF_RECORD_MISC_USER for user space maps,
* see kernel/perf_event.c __perf_event_mmap
*/
event->header.misc = PERF_RECORD_MISC_KERNEL;
} else {
event->header.misc = PERF_RECORD_MISC_GUEST_KERNEL;
}
if (symbol_conf.buildid_mmap2) {
size = snprintf(event->mmap2.filename, sizeof(event->mmap2.filename),
"%s%s", machine->mmap_name, kmap->ref_reloc_sym->name) + 1;
size = PERF_ALIGN(size, sizeof(u64));
event->mmap2.header.type = PERF_RECORD_MMAP2;
event->mmap2.header.size = (sizeof(event->mmap2) -
(sizeof(event->mmap2.filename) - size) + machine->id_hdr_size);
event->mmap2.pgoff = kmap->ref_reloc_sym->addr;
event->mmap2.start = map->start;
event->mmap2.len = map->end - event->mmap.start;
event->mmap2.pid = machine->pid;
perf_record_mmap2__read_build_id(&event->mmap2, true);
} else {
size = snprintf(event->mmap.filename, sizeof(event->mmap.filename),
"%s%s", machine->mmap_name, kmap->ref_reloc_sym->name) + 1;
size = PERF_ALIGN(size, sizeof(u64));
event->mmap.header.type = PERF_RECORD_MMAP;
event->mmap.header.size = (sizeof(event->mmap) -
(sizeof(event->mmap.filename) - size) + machine->id_hdr_size);
event->mmap.pgoff = kmap->ref_reloc_sym->addr;
event->mmap.start = map->start;
event->mmap.len = map->end - event->mmap.start;
event->mmap.pid = machine->pid;
}
err = perf_tool__process_synth_event(tool, event, machine, process);
free(event);
return err;
}
int perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
perf_event__handler_t process,
struct machine *machine)
{
int err;
err = __perf_event__synthesize_kernel_mmap(tool, process, machine);
if (err < 0)
return err;
return perf_event__synthesize_extra_kmaps(tool, process, machine);
}
int perf_event__synthesize_thread_map2(struct perf_tool *tool,
struct perf_thread_map *threads,
perf_event__handler_t process,
struct machine *machine)
{
union perf_event *event;
int i, err, size;
size = sizeof(event->thread_map);
size += threads->nr * sizeof(event->thread_map.entries[0]);
event = zalloc(size);
if (!event)
return -ENOMEM;
event->header.type = PERF_RECORD_THREAD_MAP;
event->header.size = size;
event->thread_map.nr = threads->nr;
for (i = 0; i < threads->nr; i++) {
struct perf_record_thread_map_entry *entry = &event->thread_map.entries[i];
char *comm = perf_thread_map__comm(threads, i);
if (!comm)
comm = (char *) "";
entry->pid = perf_thread_map__pid(threads, i);
strncpy((char *) &entry->comm, comm, sizeof(entry->comm));
}
err = process(tool, event, NULL, machine);
free(event);
return err;
}
static void synthesize_cpus(struct cpu_map_entries *cpus,
struct perf_cpu_map *map)
{
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
int i, map_nr = perf_cpu_map__nr(map);
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
cpus->nr = map_nr;
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
for (i = 0; i < map_nr; i++)
cpus->cpu[i] = perf_cpu_map__cpu(map, i).cpu;
}
static void synthesize_mask(struct perf_record_record_cpu_map *mask,
struct perf_cpu_map *map, int max)
{
int i;
mask->nr = BITS_TO_LONGS(max);
mask->long_size = sizeof(long);
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
for (i = 0; i < perf_cpu_map__nr(map); i++)
set_bit(perf_cpu_map__cpu(map, i).cpu, mask->mask);
}
static size_t cpus_size(struct perf_cpu_map *map)
{
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
return sizeof(struct cpu_map_entries) + perf_cpu_map__nr(map) * sizeof(u16);
}
static size_t mask_size(struct perf_cpu_map *map, int *max)
{
int i;
*max = 0;
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
for (i = 0; i < perf_cpu_map__nr(map); i++) {
/* bit position of the cpu is + 1 */
perf cpumap: Migrate to libperf cpumap api Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: André Almeida <andrealmeid@collabora.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dmitriy Vyukov <dvyukov@google.com> Cc: Eric Dumazet <edumazet@google.com> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Miaoqian Lin <linmq006@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Stephen Brennan <stephen.s.brennan@oracle.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Yury Norov <yury.norov@gmail.com> Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-22 12:58:10 +08:00
int bit = perf_cpu_map__cpu(map, i).cpu + 1;
if (bit > *max)
*max = bit;
}
return sizeof(struct perf_record_record_cpu_map) + BITS_TO_LONGS(*max) * sizeof(long);
}
void *cpu_map_data__alloc(struct perf_cpu_map *map, size_t *size, u16 *type, int *max)
{
size_t size_cpus, size_mask;
bool is_dummy = perf_cpu_map__empty(map);
/*
* Both array and mask data have variable size based
* on the number of cpus and their actual values.
* The size of the 'struct perf_record_cpu_map_data' is:
*
* array = size of 'struct cpu_map_entries' +
* number of cpus * sizeof(u64)
*
* mask = size of 'struct perf_record_record_cpu_map' +
* maximum cpu bit converted to size of longs
*
* and finally + the size of 'struct perf_record_cpu_map_data'.
*/
size_cpus = cpus_size(map);
size_mask = mask_size(map, max);
if (is_dummy || (size_cpus < size_mask)) {
*size += size_cpus;
*type = PERF_CPU_MAP__CPUS;
} else {
*size += size_mask;
*type = PERF_CPU_MAP__MASK;
}
*size += sizeof(struct perf_record_cpu_map_data);
*size = PERF_ALIGN(*size, sizeof(u64));
return zalloc(*size);
}
void cpu_map_data__synthesize(struct perf_record_cpu_map_data *data, struct perf_cpu_map *map,
u16 type, int max)
{
data->type = type;
switch (type) {
case PERF_CPU_MAP__CPUS:
synthesize_cpus((struct cpu_map_entries *) data->data, map);
break;
case PERF_CPU_MAP__MASK:
synthesize_mask((struct perf_record_record_cpu_map *)data->data, map, max);
default:
break;
}
}
static struct perf_record_cpu_map *cpu_map_event__new(struct perf_cpu_map *map)
{
size_t size = sizeof(struct perf_record_cpu_map);
struct perf_record_cpu_map *event;
int max;
u16 type;
event = cpu_map_data__alloc(map, &size, &type, &max);
if (!event)
return NULL;
event->header.type = PERF_RECORD_CPU_MAP;
event->header.size = size;
event->data.type = type;
cpu_map_data__synthesize(&event->data, map, type, max);
return event;
}
int perf_event__synthesize_cpu_map(struct perf_tool *tool,
struct perf_cpu_map *map,
perf_event__handler_t process,
struct machine *machine)
{
struct perf_record_cpu_map *event;
int err;
event = cpu_map_event__new(map);
if (!event)
return -ENOMEM;
err = process(tool, (union perf_event *) event, NULL, machine);
free(event);
return err;
}
int perf_event__synthesize_stat_config(struct perf_tool *tool,
struct perf_stat_config *config,
perf_event__handler_t process,
struct machine *machine)
{
struct perf_record_stat_config *event;
int size, i = 0, err;
size = sizeof(*event);
size += (PERF_STAT_CONFIG_TERM__MAX * sizeof(event->data[0]));
event = zalloc(size);
if (!event)
return -ENOMEM;
event->header.type = PERF_RECORD_STAT_CONFIG;
event->header.size = size;
event->nr = PERF_STAT_CONFIG_TERM__MAX;
#define ADD(__term, __val) \
event->data[i].tag = PERF_STAT_CONFIG_TERM__##__term; \
event->data[i].val = __val; \
i++;
ADD(AGGR_MODE, config->aggr_mode)
ADD(INTERVAL, config->interval)
ADD(SCALE, config->scale)
WARN_ONCE(i != PERF_STAT_CONFIG_TERM__MAX,
"stat config terms unbalanced\n");
#undef ADD
err = process(tool, (union perf_event *) event, NULL, machine);
free(event);
return err;
}
int perf_event__synthesize_stat(struct perf_tool *tool,
perf cpumap: Give CPUs their own type A common problem is confusing CPU map indices with the CPU, by wrapping the CPU with a struct then this is avoided. This approach is similar to atomic_t. Committer notes: To make it build with BUILD_BPF_SKEL=1 these files needed the conversions to 'struct perf_cpu' usage: tools/perf/util/bpf_counter.c tools/perf/util/bpf_counter_cgroup.c tools/perf/util/bpf_ftrace.c Also perf_env__get_cpu() was removed back in "perf cpumap: Switch cpu_map__build_map to cpu function". Additionally these needed to be fixed for the ARM builds to complete: tools/perf/arch/arm/util/cs-etm.c tools/perf/arch/arm64/util/pmu.c Suggested-by: John Garry <john.garry@huawei.com> Signed-off-by: Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Vineet Singh <vineet.singh@intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-49-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-05 14:13:51 +08:00
struct perf_cpu cpu, u32 thread, u64 id,
struct perf_counts_values *count,
perf_event__handler_t process,
struct machine *machine)
{
struct perf_record_stat event;
event.header.type = PERF_RECORD_STAT;
event.header.size = sizeof(event);
event.header.misc = 0;
event.id = id;
perf cpumap: Give CPUs their own type A common problem is confusing CPU map indices with the CPU, by wrapping the CPU with a struct then this is avoided. This approach is similar to atomic_t. Committer notes: To make it build with BUILD_BPF_SKEL=1 these files needed the conversions to 'struct perf_cpu' usage: tools/perf/util/bpf_counter.c tools/perf/util/bpf_counter_cgroup.c tools/perf/util/bpf_ftrace.c Also perf_env__get_cpu() was removed back in "perf cpumap: Switch cpu_map__build_map to cpu function". Additionally these needed to be fixed for the ARM builds to complete: tools/perf/arch/arm/util/cs-etm.c tools/perf/arch/arm64/util/pmu.c Suggested-by: John Garry <john.garry@huawei.com> Signed-off-by: Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Vineet Singh <vineet.singh@intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-49-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-05 14:13:51 +08:00
event.cpu = cpu.cpu;
event.thread = thread;
event.val = count->val;
event.ena = count->ena;
event.run = count->run;
return process(tool, (union perf_event *) &event, NULL, machine);
}
int perf_event__synthesize_stat_round(struct perf_tool *tool,
u64 evtime, u64 type,
perf_event__handler_t process,
struct machine *machine)
{
struct perf_record_stat_round event;
event.header.type = PERF_RECORD_STAT_ROUND;
event.header.size = sizeof(event);
event.header.misc = 0;
event.time = evtime;
event.type = type;
return process(tool, (union perf_event *) &event, NULL, machine);
}
size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type, u64 read_format)
{
size_t sz, result = sizeof(struct perf_record_sample);
if (type & PERF_SAMPLE_IDENTIFIER)
result += sizeof(u64);
if (type & PERF_SAMPLE_IP)
result += sizeof(u64);
if (type & PERF_SAMPLE_TID)
result += sizeof(u64);
if (type & PERF_SAMPLE_TIME)
result += sizeof(u64);
if (type & PERF_SAMPLE_ADDR)
result += sizeof(u64);
if (type & PERF_SAMPLE_ID)
result += sizeof(u64);
if (type & PERF_SAMPLE_STREAM_ID)
result += sizeof(u64);
if (type & PERF_SAMPLE_CPU)
result += sizeof(u64);
if (type & PERF_SAMPLE_PERIOD)
result += sizeof(u64);
if (type & PERF_SAMPLE_READ) {
result += sizeof(u64);
if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
result += sizeof(u64);
if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
result += sizeof(u64);
/* PERF_FORMAT_ID is forced for PERF_SAMPLE_READ */
if (read_format & PERF_FORMAT_GROUP) {
sz = sample->read.group.nr *
sizeof(struct sample_read_value);
result += sz;
} else {
result += sizeof(u64);
}
}
if (type & PERF_SAMPLE_CALLCHAIN) {
sz = (sample->callchain->nr + 1) * sizeof(u64);
result += sz;
}
if (type & PERF_SAMPLE_RAW) {
result += sizeof(u32);
result += sample->raw_size;
}
if (type & PERF_SAMPLE_BRANCH_STACK) {
sz = sample->branch_stack->nr * sizeof(struct branch_entry);
perf tools: Add hw_idx in struct branch_stack The low level index of raw branch records for the most recent branch can be recorded in a sample with PERF_SAMPLE_BRANCH_HW_INDEX branch_sample_type. Extend struct branch_stack to support it. However, if the PERF_SAMPLE_BRANCH_HW_INDEX is not applied, only nr and entries[] will be output by kernel. The pointer of entries[] could be wrong, since the output format is different with new struct branch_stack. Add a variable no_hw_idx in struct perf_sample to indicate whether the hw_idx is output. Add get_branch_entry() to return corresponding pointer of entries[0]. To make dummy branch sample consistent as new branch sample, add hw_idx in struct dummy_branch_stack for cs-etm and intel-pt. Apply the new struct branch_stack for synthetic events as well. Extend test case sample-parsing to support new struct branch_stack. Committer notes: Renamed get_branch_entries() to perf_sample__branch_entries() to have proper namespacing and pave the way for this to be moved to libperf, eventually. Add 'static' to that inline as it is in a header. Add 'hw_idx' to 'struct dummy_branch_stack' in cs-etm.c to fix the build on arm64. Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexey Budankov <alexey.budankov@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Pavel Gerasimov <pavel.gerasimov@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com> Cc: Stephane Eranian <eranian@google.com> Cc: Vitaly Slobodskoy <vitaly.slobodskoy@intel.com> Link: http://lore.kernel.org/lkml/20200228163011.19358-2-kan.liang@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-02-29 00:30:00 +08:00
/* nr, hw_idx */
sz += 2 * sizeof(u64);
result += sz;
}
if (type & PERF_SAMPLE_REGS_USER) {
if (sample->user_regs.abi) {
result += sizeof(u64);
sz = hweight64(sample->user_regs.mask) * sizeof(u64);
result += sz;
} else {
result += sizeof(u64);
}
}
if (type & PERF_SAMPLE_STACK_USER) {
sz = sample->user_stack.size;
result += sizeof(u64);
if (sz) {
result += sz;
result += sizeof(u64);
}
}
perf tools: Support PERF_SAMPLE_WEIGHT_STRUCT The new sample type, PERF_SAMPLE_WEIGHT_STRUCT, is an alternative of the PERF_SAMPLE_WEIGHT sample type. Users can apply either the PERF_SAMPLE_WEIGHT sample type or the PERF_SAMPLE_WEIGHT_STRUCT sample type to retrieve the sample weight, but they cannot apply both sample types simultaneously. The new sample type shares the same space as the PERF_SAMPLE_WEIGHT sample type. The lower 32 bits are exactly the same for both sample type. The higher 32 bits may be different for different architecture. Add arch specific arch_evsel__set_sample_weight() to set the new sample type for X86. Only store the lower 32 bits for the sample->weight if the new sample type is applied. In practice, no memory access could last than 4G cycles. No data will be lost. If the kernel doesn't support the new sample type. Fall back to the PERF_SAMPLE_WEIGHT sample type. There is no impact for other architectures. Committer notes: Fixup related to PERF_SAMPLE_CODE_PAGE_SIZE, present in acme/perf/core but not upstream yet. Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lore.kernel.org/lkml/1612296553-21962-6-git-send-email-kan.liang@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-03 04:09:09 +08:00
if (type & PERF_SAMPLE_WEIGHT_TYPE)
result += sizeof(u64);
if (type & PERF_SAMPLE_DATA_SRC)
result += sizeof(u64);
if (type & PERF_SAMPLE_TRANSACTION)
result += sizeof(u64);
if (type & PERF_SAMPLE_REGS_INTR) {
if (sample->intr_regs.abi) {
result += sizeof(u64);
sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
result += sz;
} else {
result += sizeof(u64);
}
}
if (type & PERF_SAMPLE_PHYS_ADDR)
result += sizeof(u64);
if (type & PERF_SAMPLE_CGROUP)
result += sizeof(u64);
if (type & PERF_SAMPLE_DATA_PAGE_SIZE)
result += sizeof(u64);
if (type & PERF_SAMPLE_CODE_PAGE_SIZE)
result += sizeof(u64);
if (type & PERF_SAMPLE_AUX) {
result += sizeof(u64);
result += sample->aux_sample.size;
}
return result;
}
void __weak arch_perf_synthesize_sample_weight(const struct perf_sample *data,
__u64 *array, u64 type __maybe_unused)
{
*array = data->weight;
}
int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_format,
const struct perf_sample *sample)
{
__u64 *array;
size_t sz;
/*
* used for cross-endian analysis. See git commit 65014ab3
* for why this goofiness is needed.
*/
union u64_swap u;
array = event->sample.array;
if (type & PERF_SAMPLE_IDENTIFIER) {
*array = sample->id;
array++;
}
if (type & PERF_SAMPLE_IP) {
*array = sample->ip;
array++;
}
if (type & PERF_SAMPLE_TID) {
u.val32[0] = sample->pid;
u.val32[1] = sample->tid;
*array = u.val64;
array++;
}
if (type & PERF_SAMPLE_TIME) {
*array = sample->time;
array++;
}
if (type & PERF_SAMPLE_ADDR) {
*array = sample->addr;
array++;
}
if (type & PERF_SAMPLE_ID) {
*array = sample->id;
array++;
}
if (type & PERF_SAMPLE_STREAM_ID) {
*array = sample->stream_id;
array++;
}
if (type & PERF_SAMPLE_CPU) {
u.val32[0] = sample->cpu;
u.val32[1] = 0;
*array = u.val64;
array++;
}
if (type & PERF_SAMPLE_PERIOD) {
*array = sample->period;
array++;
}
if (type & PERF_SAMPLE_READ) {
if (read_format & PERF_FORMAT_GROUP)
*array = sample->read.group.nr;
else
*array = sample->read.one.value;
array++;
if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
*array = sample->read.time_enabled;
array++;
}
if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
*array = sample->read.time_running;
array++;
}
/* PERF_FORMAT_ID is forced for PERF_SAMPLE_READ */
if (read_format & PERF_FORMAT_GROUP) {
sz = sample->read.group.nr *
sizeof(struct sample_read_value);
memcpy(array, sample->read.group.values, sz);
array = (void *)array + sz;
} else {
*array = sample->read.one.id;
array++;
}
}
if (type & PERF_SAMPLE_CALLCHAIN) {
sz = (sample->callchain->nr + 1) * sizeof(u64);
memcpy(array, sample->callchain, sz);
array = (void *)array + sz;
}
if (type & PERF_SAMPLE_RAW) {
u.val32[0] = sample->raw_size;
*array = u.val64;
array = (void *)array + sizeof(u32);
memcpy(array, sample->raw_data, sample->raw_size);
array = (void *)array + sample->raw_size;
}
if (type & PERF_SAMPLE_BRANCH_STACK) {
sz = sample->branch_stack->nr * sizeof(struct branch_entry);
perf tools: Add hw_idx in struct branch_stack The low level index of raw branch records for the most recent branch can be recorded in a sample with PERF_SAMPLE_BRANCH_HW_INDEX branch_sample_type. Extend struct branch_stack to support it. However, if the PERF_SAMPLE_BRANCH_HW_INDEX is not applied, only nr and entries[] will be output by kernel. The pointer of entries[] could be wrong, since the output format is different with new struct branch_stack. Add a variable no_hw_idx in struct perf_sample to indicate whether the hw_idx is output. Add get_branch_entry() to return corresponding pointer of entries[0]. To make dummy branch sample consistent as new branch sample, add hw_idx in struct dummy_branch_stack for cs-etm and intel-pt. Apply the new struct branch_stack for synthetic events as well. Extend test case sample-parsing to support new struct branch_stack. Committer notes: Renamed get_branch_entries() to perf_sample__branch_entries() to have proper namespacing and pave the way for this to be moved to libperf, eventually. Add 'static' to that inline as it is in a header. Add 'hw_idx' to 'struct dummy_branch_stack' in cs-etm.c to fix the build on arm64. Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexey Budankov <alexey.budankov@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Pavel Gerasimov <pavel.gerasimov@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com> Cc: Stephane Eranian <eranian@google.com> Cc: Vitaly Slobodskoy <vitaly.slobodskoy@intel.com> Link: http://lore.kernel.org/lkml/20200228163011.19358-2-kan.liang@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-02-29 00:30:00 +08:00
/* nr, hw_idx */
sz += 2 * sizeof(u64);
memcpy(array, sample->branch_stack, sz);
array = (void *)array + sz;
}
if (type & PERF_SAMPLE_REGS_USER) {
if (sample->user_regs.abi) {
*array++ = sample->user_regs.abi;
sz = hweight64(sample->user_regs.mask) * sizeof(u64);
memcpy(array, sample->user_regs.regs, sz);
array = (void *)array + sz;
} else {
*array++ = 0;
}
}
if (type & PERF_SAMPLE_STACK_USER) {
sz = sample->user_stack.size;
*array++ = sz;
if (sz) {
memcpy(array, sample->user_stack.data, sz);
array = (void *)array + sz;
*array++ = sz;
}
}
perf tools: Support PERF_SAMPLE_WEIGHT_STRUCT The new sample type, PERF_SAMPLE_WEIGHT_STRUCT, is an alternative of the PERF_SAMPLE_WEIGHT sample type. Users can apply either the PERF_SAMPLE_WEIGHT sample type or the PERF_SAMPLE_WEIGHT_STRUCT sample type to retrieve the sample weight, but they cannot apply both sample types simultaneously. The new sample type shares the same space as the PERF_SAMPLE_WEIGHT sample type. The lower 32 bits are exactly the same for both sample type. The higher 32 bits may be different for different architecture. Add arch specific arch_evsel__set_sample_weight() to set the new sample type for X86. Only store the lower 32 bits for the sample->weight if the new sample type is applied. In practice, no memory access could last than 4G cycles. No data will be lost. If the kernel doesn't support the new sample type. Fall back to the PERF_SAMPLE_WEIGHT sample type. There is no impact for other architectures. Committer notes: Fixup related to PERF_SAMPLE_CODE_PAGE_SIZE, present in acme/perf/core but not upstream yet. Signed-off-by: Kan Liang <kan.liang@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lore.kernel.org/lkml/1612296553-21962-6-git-send-email-kan.liang@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-03 04:09:09 +08:00
if (type & PERF_SAMPLE_WEIGHT_TYPE) {
arch_perf_synthesize_sample_weight(sample, array, type);
array++;
}
if (type & PERF_SAMPLE_DATA_SRC) {
*array = sample->data_src;
array++;
}
if (type & PERF_SAMPLE_TRANSACTION) {
*array = sample->transaction;
array++;
}
if (type & PERF_SAMPLE_REGS_INTR) {
if (sample->intr_regs.abi) {
*array++ = sample->intr_regs.abi;
sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
memcpy(array, sample->intr_regs.regs, sz);
array = (void *)array + sz;
} else {
*array++ = 0;
}
}
if (type & PERF_SAMPLE_PHYS_ADDR) {
*array = sample->phys_addr;
array++;
}
if (type & PERF_SAMPLE_CGROUP) {
*array = sample->cgroup;
array++;
}
if (type & PERF_SAMPLE_DATA_PAGE_SIZE) {
*array = sample->data_page_size;
array++;
}
if (type & PERF_SAMPLE_CODE_PAGE_SIZE) {
*array = sample->code_page_size;
array++;
}
if (type & PERF_SAMPLE_AUX) {
sz = sample->aux_sample.size;
*array++ = sz;
memcpy(array, sample->aux_sample.data, sz);
array = (void *)array + sz;
}
return 0;
}
int perf_event__synthesize_id_index(struct perf_tool *tool, perf_event__handler_t process,
struct evlist *evlist, struct machine *machine)
{
union perf_event *ev;
struct evsel *evsel;
size_t nr = 0, i = 0, sz, max_nr, n;
int err;
pr_debug2("Synthesizing id index\n");
max_nr = (UINT16_MAX - sizeof(struct perf_record_id_index)) /
sizeof(struct id_index_entry);
evlist__for_each_entry(evlist, evsel)
nr += evsel->core.ids;
n = nr > max_nr ? max_nr : nr;
sz = sizeof(struct perf_record_id_index) + n * sizeof(struct id_index_entry);
ev = zalloc(sz);
if (!ev)
return -ENOMEM;
ev->id_index.header.type = PERF_RECORD_ID_INDEX;
ev->id_index.header.size = sz;
ev->id_index.nr = n;
evlist__for_each_entry(evlist, evsel) {
u32 j;
for (j = 0; j < evsel->core.ids; j++) {
struct id_index_entry *e;
struct perf_sample_id *sid;
if (i >= n) {
err = process(tool, ev, NULL, machine);
if (err)
goto out_err;
nr -= n;
i = 0;
}
e = &ev->id_index.entries[i++];
e->id = evsel->core.id[j];
sid = evlist__id2sid(evlist, e->id);
if (!sid) {
free(ev);
return -ENOENT;
}
e->idx = sid->idx;
perf cpumap: Give CPUs their own type A common problem is confusing CPU map indices with the CPU, by wrapping the CPU with a struct then this is avoided. This approach is similar to atomic_t. Committer notes: To make it build with BUILD_BPF_SKEL=1 these files needed the conversions to 'struct perf_cpu' usage: tools/perf/util/bpf_counter.c tools/perf/util/bpf_counter_cgroup.c tools/perf/util/bpf_ftrace.c Also perf_env__get_cpu() was removed back in "perf cpumap: Switch cpu_map__build_map to cpu function". Additionally these needed to be fixed for the ARM builds to complete: tools/perf/arch/arm/util/cs-etm.c tools/perf/arch/arm64/util/pmu.c Suggested-by: John Garry <john.garry@huawei.com> Signed-off-by: Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Vineet Singh <vineet.singh@intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-49-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-01-05 14:13:51 +08:00
e->cpu = sid->cpu.cpu;
e->tid = sid->tid;
}
}
sz = sizeof(struct perf_record_id_index) + nr * sizeof(struct id_index_entry);
ev->id_index.header.size = sz;
ev->id_index.nr = nr;
err = process(tool, ev, NULL, machine);
out_err:
free(ev);
return err;
}
int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
struct target *target, struct perf_thread_map *threads,
perf_event__handler_t process, bool needs_mmap,
bool data_mmap, unsigned int nr_threads_synthesize)
{
perf synthetic-events: Return error if procfs isn't mounted for PID namespaces For perf recording, it retrieves process info by iterating nodes in proc fs. If we run perf in a non-root PID namespace with command: # unshare --fork --pid perf record -e cycles -a -- test_program ... in this case, unshare command creates a child PID namespace and launches perf tool in it, but the issue is the proc fs is not mounted for the non-root PID namespace, this leads to the perf tool gathering process info from its parent PID namespace. We can use below command to observe the process nodes under proc fs: # unshare --pid --fork ls /proc 1 137 1968 2128 3 342 48 62 78 crypto kcore net uptime 10 138 2 2142 30 35 49 63 8 devices keys pagetypeinfo version 11 139 20 2143 304 36 50 64 82 device-tree key-users partitions vmallocinfo 12 14 2011 22 305 37 51 65 83 diskstats kmsg self vmstat 128 140 2038 23 307 39 52 656 84 driver kpagecgroup slabinfo zoneinfo 129 15 2074 24 309 4 53 67 9 execdomains kpagecount softirqs 13 16 2094 241 31 40 54 68 asound fb kpageflags stat 130 164 2096 242 310 41 55 69 buddyinfo filesystems loadavg swaps 131 17 2098 25 317 42 56 70 bus fs locks sys 132 175 21 26 32 43 57 71 cgroups interrupts meminfo sysrq-trigger 133 179 2102 263 329 44 58 75 cmdline iomem misc sysvipc 134 1875 2103 27 330 45 59 76 config.gz ioports modules thread-self 135 19 2117 29 333 46 6 77 consoles irq mounts timer_list 136 1941 2121 298 34 47 60 773 cpuinfo kallsyms mtd tty So it shows many existed tasks, since unshared command has not mounted the proc fs for the new created PID namespace, it still accesses the proc fs of the root PID namespace. This leads to two prominent issues: - Firstly, PID values are mismatched between thread info and samples. The gathered thread info are coming from the proc fs of the root PID namespace, but samples record its PID from the child PID namespace. - The second issue is profiled program 'test_program' returns its forked PID number from the child PID namespace, perf tool wrongly uses this PID number to retrieve the process info via the proc fs of the root PID namespace. To avoid issues, we need to mount proc fs for the child PID namespace with the option '--mount-proc' when use unshare command: # unshare --fork --pid --mount-proc perf record -e cycles -a -- test_program Conversely, when the proc fs of the root PID namespace is used by child namespace, perf tool can detect the multiple PID levels and nsinfo__is_in_root_namespace() returns false, this patch reports error for this case: # unshare --fork --pid perf record -e cycles -a -- test_program Couldn't synthesize bpf events. Perf runs in non-root PID namespace but it tries to gather process info from its parent PID namespace. Please mount the proc file system properly, e.g. add the option '--mount-proc' for unshare command. Reviewed-by: James Clark <james.clark@arm.com> Signed-off-by: Leo Yan <leo.yan@linaro.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Andrii Nakryiko <andrii@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Fastabend <john.fastabend@gmail.com> Cc: KP Singh <kpsingh@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Song Liu <songliubraving@fb.com> Cc: Yonghong Song <yhs@fb.com> Link: https://lore.kernel.org/r/20211224124014.2492751-1-leo.yan@linaro.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-12-24 20:40:13 +08:00
/*
* When perf runs in non-root PID namespace, and the namespace's proc FS
* is not mounted, nsinfo__is_in_root_namespace() returns false.
* In this case, the proc FS is coming for the parent namespace, thus
* perf tool will wrongly gather process info from its parent PID
* namespace.
*
* To avoid the confusion that the perf tool runs in a child PID
* namespace but it synthesizes thread info from its parent PID
* namespace, returns failure with warning.
*/
if (!nsinfo__is_in_root_namespace()) {
pr_err("Perf runs in non-root PID namespace but it tries to ");
pr_err("gather process info from its parent PID namespace.\n");
pr_err("Please mount the proc file system properly, e.g. ");
pr_err("add the option '--mount-proc' for unshare command.\n");
return -EPERM;
}
if (target__has_task(target))
return perf_event__synthesize_thread_map(tool, threads, process, machine,
needs_mmap, data_mmap);
else if (target__has_cpu(target))
return perf_event__synthesize_threads(tool, process, machine,
needs_mmap, data_mmap,
nr_threads_synthesize);
/* command specified */
return 0;
}
int machine__synthesize_threads(struct machine *machine, struct target *target,
struct perf_thread_map *threads, bool needs_mmap,
bool data_mmap, unsigned int nr_threads_synthesize)
{
return __machine__synthesize_threads(machine, NULL, target, threads,
perf_event__process, needs_mmap,
data_mmap, nr_threads_synthesize);
}
static struct perf_record_event_update *event_update_event__new(size_t size, u64 type, u64 id)
{
struct perf_record_event_update *ev;
size += sizeof(*ev);
size = PERF_ALIGN(size, sizeof(u64));
ev = zalloc(size);
if (ev) {
ev->header.type = PERF_RECORD_EVENT_UPDATE;
ev->header.size = (u16)size;
ev->type = type;
ev->id = id;
}
return ev;
}
int perf_event__synthesize_event_update_unit(struct perf_tool *tool, struct evsel *evsel,
perf_event__handler_t process)
{
size_t size = strlen(evsel->unit);
struct perf_record_event_update *ev;
int err;
ev = event_update_event__new(size + 1, PERF_EVENT_UPDATE__UNIT, evsel->core.id[0]);
if (ev == NULL)
return -ENOMEM;
strlcpy(ev->data, evsel->unit, size + 1);
err = process(tool, (union perf_event *)ev, NULL, NULL);
free(ev);
return err;
}
int perf_event__synthesize_event_update_scale(struct perf_tool *tool, struct evsel *evsel,
perf_event__handler_t process)
{
struct perf_record_event_update *ev;
struct perf_record_event_update_scale *ev_data;
int err;
ev = event_update_event__new(sizeof(*ev_data), PERF_EVENT_UPDATE__SCALE, evsel->core.id[0]);
if (ev == NULL)
return -ENOMEM;
ev_data = (struct perf_record_event_update_scale *)ev->data;
ev_data->scale = evsel->scale;
err = process(tool, (union perf_event *)ev, NULL, NULL);
free(ev);
return err;
}
int perf_event__synthesize_event_update_name(struct perf_tool *tool, struct evsel *evsel,
perf_event__handler_t process)
{
struct perf_record_event_update *ev;
size_t len = strlen(evsel->name);
int err;
ev = event_update_event__new(len + 1, PERF_EVENT_UPDATE__NAME, evsel->core.id[0]);
if (ev == NULL)
return -ENOMEM;
strlcpy(ev->data, evsel->name, len + 1);
err = process(tool, (union perf_event *)ev, NULL, NULL);
free(ev);
return err;
}
int perf_event__synthesize_event_update_cpus(struct perf_tool *tool, struct evsel *evsel,
perf_event__handler_t process)
{
size_t size = sizeof(struct perf_record_event_update);
struct perf_record_event_update *ev;
int max, err;
u16 type;
if (!evsel->core.own_cpus)
return 0;
ev = cpu_map_data__alloc(evsel->core.own_cpus, &size, &type, &max);
if (!ev)
return -ENOMEM;
ev->header.type = PERF_RECORD_EVENT_UPDATE;
ev->header.size = (u16)size;
ev->type = PERF_EVENT_UPDATE__CPUS;
ev->id = evsel->core.id[0];
cpu_map_data__synthesize((struct perf_record_cpu_map_data *)ev->data,
evsel->core.own_cpus, type, max);
err = process(tool, (union perf_event *)ev, NULL, NULL);
free(ev);
return err;
}
int perf_event__synthesize_attrs(struct perf_tool *tool, struct evlist *evlist,
perf_event__handler_t process)
{
struct evsel *evsel;
int err = 0;
evlist__for_each_entry(evlist, evsel) {
err = perf_event__synthesize_attr(tool, &evsel->core.attr, evsel->core.ids,
evsel->core.id, process);
if (err) {
pr_debug("failed to create perf header attribute\n");
return err;
}
}
return err;
}
static bool has_unit(struct evsel *evsel)
{
return evsel->unit && *evsel->unit;
}
static bool has_scale(struct evsel *evsel)
{
return evsel->scale != 1;
}
int perf_event__synthesize_extra_attr(struct perf_tool *tool, struct evlist *evsel_list,
perf_event__handler_t process, bool is_pipe)
{
struct evsel *evsel;
int err;
/*
* Synthesize other events stuff not carried within
* attr event - unit, scale, name
*/
evlist__for_each_entry(evsel_list, evsel) {
if (!evsel->supported)
continue;
/*
* Synthesize unit and scale only if it's defined.
*/
if (has_unit(evsel)) {
err = perf_event__synthesize_event_update_unit(tool, evsel, process);
if (err < 0) {
pr_err("Couldn't synthesize evsel unit.\n");
return err;
}
}
if (has_scale(evsel)) {
err = perf_event__synthesize_event_update_scale(tool, evsel, process);
if (err < 0) {
pr_err("Couldn't synthesize evsel evsel.\n");
return err;
}
}
if (evsel->core.own_cpus) {
err = perf_event__synthesize_event_update_cpus(tool, evsel, process);
if (err < 0) {
pr_err("Couldn't synthesize evsel cpus.\n");
return err;
}
}
/*
* Name is needed only for pipe output,
* perf.data carries event names.
*/
if (is_pipe) {
err = perf_event__synthesize_event_update_name(tool, evsel, process);
if (err < 0) {
pr_err("Couldn't synthesize evsel name.\n");
return err;
}
}
}
return 0;
}
int perf_event__synthesize_attr(struct perf_tool *tool, struct perf_event_attr *attr,
u32 ids, u64 *id, perf_event__handler_t process)
{
union perf_event *ev;
size_t size;
int err;
size = sizeof(struct perf_event_attr);
size = PERF_ALIGN(size, sizeof(u64));
size += sizeof(struct perf_event_header);
size += ids * sizeof(u64);
ev = zalloc(size);
if (ev == NULL)
return -ENOMEM;
ev->attr.attr = *attr;
memcpy(ev->attr.id, id, ids * sizeof(u64));
ev->attr.header.type = PERF_RECORD_HEADER_ATTR;
ev->attr.header.size = (u16)size;
if (ev->attr.header.size == size)
err = process(tool, ev, NULL, NULL);
else
err = -E2BIG;
free(ev);
return err;
}
int perf_event__synthesize_tracing_data(struct perf_tool *tool, int fd, struct evlist *evlist,
perf_event__handler_t process)
{
union perf_event ev;
struct tracing_data *tdata;
ssize_t size = 0, aligned_size = 0, padding;
struct feat_fd ff;
/*
* We are going to store the size of the data followed
* by the data contents. Since the fd descriptor is a pipe,
* we cannot seek back to store the size of the data once
* we know it. Instead we:
*
* - write the tracing data to the temp file
* - get/write the data size to pipe
* - write the tracing data from the temp file
* to the pipe
*/
tdata = tracing_data_get(&evlist->core.entries, fd, true);
if (!tdata)
return -1;
memset(&ev, 0, sizeof(ev));
ev.tracing_data.header.type = PERF_RECORD_HEADER_TRACING_DATA;
size = tdata->size;
aligned_size = PERF_ALIGN(size, sizeof(u64));
padding = aligned_size - size;
ev.tracing_data.header.size = sizeof(ev.tracing_data);
ev.tracing_data.size = aligned_size;
process(tool, &ev, NULL, NULL);
/*
* The put function will copy all the tracing data
* stored in temp file to the pipe.
*/
tracing_data_put(tdata);
ff = (struct feat_fd){ .fd = fd };
if (write_padded(&ff, NULL, 0, padding))
return -1;
return aligned_size;
}
int perf_event__synthesize_build_id(struct perf_tool *tool, struct dso *pos, u16 misc,
perf_event__handler_t process, struct machine *machine)
{
union perf_event ev;
size_t len;
if (!pos->hit)
return 0;
memset(&ev, 0, sizeof(ev));
len = pos->long_name_len + 1;
len = PERF_ALIGN(len, NAME_ALIGN);
memcpy(&ev.build_id.build_id, pos->bid.data, sizeof(pos->bid.data));
ev.build_id.header.type = PERF_RECORD_HEADER_BUILD_ID;
ev.build_id.header.misc = misc;
ev.build_id.pid = machine->pid;
ev.build_id.header.size = sizeof(ev.build_id) + len;
memcpy(&ev.build_id.filename, pos->long_name, pos->long_name_len);
return process(tool, &ev, NULL, machine);
}
int perf_event__synthesize_stat_events(struct perf_stat_config *config, struct perf_tool *tool,
struct evlist *evlist, perf_event__handler_t process, bool attrs)
{
int err;
if (attrs) {
err = perf_event__synthesize_attrs(tool, evlist, process);
if (err < 0) {
pr_err("Couldn't synthesize attrs.\n");
return err;
}
}
err = perf_event__synthesize_extra_attr(tool, evlist, process, attrs);
err = perf_event__synthesize_thread_map2(tool, evlist->core.threads, process, NULL);
if (err < 0) {
pr_err("Couldn't synthesize thread map.\n");
return err;
}
perf evlist: Rename cpus to user_requested_cpus evlist contains cpus and all_cpus. all_cpus is the union of the cpu maps of all evsels. For non-task targets, cpus is set to be cpus requested from the command line, defaulting to all online cpus if no cpus are specified. For an uncore event, all_cpus may be just CPU 0 or every online CPU. This causes all_cpus to have fewer values than the cpus variable which is confusing given the 'all' in the name. To try to make the behavior clearer, rename cpus to user_requested_cpus and add comments on the two struct variables. Signed-off-by: Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Antonov <alexander.antonov@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrii Nakryiko <andrii@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: German Gomez <german.gomez@arm.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Fastabend <john.fastabend@gmail.com> Cc: John Garry <john.garry@huawei.com> Cc: KP Singh <kpsingh@kernel.org> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Song Liu <songliubraving@fb.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Will Deacon <will@kernel.org> Cc: Yonghong Song <yhs@fb.com> Cc: bpf@vger.kernel.org Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: netdev@vger.kernel.org Link: http://lore.kernel.org/lkml/20220328232648.2127340-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-03-29 07:26:44 +08:00
err = perf_event__synthesize_cpu_map(tool, evlist->core.user_requested_cpus, process, NULL);
if (err < 0) {
pr_err("Couldn't synthesize thread map.\n");
return err;
}
err = perf_event__synthesize_stat_config(tool, config, process, NULL);
if (err < 0) {
pr_err("Couldn't synthesize config.\n");
return err;
}
return 0;
}
extern const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
int perf_event__synthesize_features(struct perf_tool *tool, struct perf_session *session,
struct evlist *evlist, perf_event__handler_t process)
{
struct perf_header *header = &session->header;
struct perf_record_header_feature *fe;
struct feat_fd ff;
size_t sz, sz_hdr;
int feat, ret;
sz_hdr = sizeof(fe->header);
sz = sizeof(union perf_event);
/* get a nice alignment */
sz = PERF_ALIGN(sz, page_size);
memset(&ff, 0, sizeof(ff));
ff.buf = malloc(sz);
if (!ff.buf)
return -ENOMEM;
ff.size = sz - sz_hdr;
ff.ph = &session->header;
for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
if (!feat_ops[feat].synthesize) {
pr_debug("No record header feature for header :%d\n", feat);
continue;
}
ff.offset = sizeof(*fe);
ret = feat_ops[feat].write(&ff, evlist);
if (ret || ff.offset <= (ssize_t)sizeof(*fe)) {
pr_debug("Error writing feature\n");
continue;
}
/* ff.buf may have changed due to realloc in do_write() */
fe = ff.buf;
memset(fe, 0, sizeof(*fe));
fe->feat_id = feat;
fe->header.type = PERF_RECORD_HEADER_FEATURE;
fe->header.size = ff.offset;
ret = process(tool, ff.buf, NULL, NULL);
if (ret) {
free(ff.buf);
return ret;
}
}
/* Send HEADER_LAST_FEATURE mark. */
fe = ff.buf;
fe->feat_id = HEADER_LAST_FEATURE;
fe->header.type = PERF_RECORD_HEADER_FEATURE;
fe->header.size = sizeof(*fe);
ret = process(tool, ff.buf, NULL, NULL);
free(ff.buf);
return ret;
}
int perf_event__synthesize_for_pipe(struct perf_tool *tool,
struct perf_session *session,
struct perf_data *data,
perf_event__handler_t process)
{
int err;
int ret = 0;
struct evlist *evlist = session->evlist;
/*
* We need to synthesize events first, because some
* features works on top of them (on report side).
*/
err = perf_event__synthesize_attrs(tool, evlist, process);
if (err < 0) {
pr_err("Couldn't synthesize attrs.\n");
return err;
}
ret += err;
err = perf_event__synthesize_features(tool, session, evlist, process);
if (err < 0) {
pr_err("Couldn't synthesize features.\n");
return err;
}
ret += err;
if (have_tracepoints(&evlist->core.entries)) {
int fd = perf_data__fd(data);
/*
* FIXME err <= 0 here actually means that
* there were no tracepoints so its not really
* an error, just that we don't need to
* synthesize anything. We really have to
* return this more properly and also
* propagate errors that now are calling die()
*/
err = perf_event__synthesize_tracing_data(tool, fd, evlist,
process);
if (err <= 0) {
pr_err("Couldn't record tracing data.\n");
return err;
}
ret += err;
}
return ret;
}
int parse_synth_opt(char *synth)
{
char *p, *q;
int ret = 0;
if (synth == NULL)
return -1;
for (q = synth; (p = strsep(&q, ",")); p = q) {
if (!strcasecmp(p, "no") || !strcasecmp(p, "none"))
return 0;
if (!strcasecmp(p, "all"))
return PERF_SYNTH_ALL;
if (!strcasecmp(p, "task"))
ret |= PERF_SYNTH_TASK;
else if (!strcasecmp(p, "mmap"))
ret |= PERF_SYNTH_TASK | PERF_SYNTH_MMAP;
else if (!strcasecmp(p, "cgroup"))
ret |= PERF_SYNTH_CGROUP;
else
return -1;
}
return ret;
}