4 * Builtin regression testing command: ever growing number of sanity tests
8 #include "util/cache.h"
9 #include "util/debug.h"
10 #include "util/debugfs.h"
11 #include "util/evlist.h"
12 #include "util/parse-options.h"
13 #include "util/parse-events.h"
14 #include "util/symbol.h"
15 #include "util/thread_map.h"
17 #include "../../include/linux/hw_breakpoint.h"
21 static int vmlinux_matches_kallsyms_filter(struct map
*map __maybe_unused
,
24 bool *visited
= symbol__priv(sym
);
29 static int test__vmlinux_matches_kallsyms(void)
34 struct map
*kallsyms_map
, *vmlinux_map
;
35 struct machine kallsyms
, vmlinux
;
36 enum map_type type
= MAP__FUNCTION
;
37 long page_size
= sysconf(_SC_PAGE_SIZE
);
38 struct ref_reloc_sym ref_reloc_sym
= { .name
= "_stext", };
43 * Init the machines that will hold kernel, modules obtained from
44 * both vmlinux + .ko files and from /proc/kallsyms split by modules.
46 machine__init(&kallsyms
, "", HOST_KERNEL_ID
);
47 machine__init(&vmlinux
, "", HOST_KERNEL_ID
);
52 * Create the kernel maps for kallsyms and the DSO where we will then
53 * load /proc/kallsyms. Also create the modules maps from /proc/modules
54 * and find the .ko files that match them in /lib/modules/`uname -r`/.
56 if (machine__create_kernel_maps(&kallsyms
) < 0) {
57 pr_debug("machine__create_kernel_maps ");
64 * Load and split /proc/kallsyms into multiple maps, one per module.
66 if (machine__load_kallsyms(&kallsyms
, "/proc/kallsyms", type
, NULL
) <= 0) {
67 pr_debug("dso__load_kallsyms ");
74 * kallsyms will be internally on demand sorted by name so that we can
75 * find the reference relocation * symbol, i.e. the symbol we will use
76 * to see if the running kernel was relocated by checking if it has the
77 * same value in the vmlinux file we load.
79 kallsyms_map
= machine__kernel_map(&kallsyms
, type
);
81 sym
= map__find_symbol_by_name(kallsyms_map
, ref_reloc_sym
.name
, NULL
);
83 pr_debug("dso__find_symbol_by_name ");
87 ref_reloc_sym
.addr
= sym
->start
;
92 * Now repeat step 2, this time for the vmlinux file we'll auto-locate.
94 if (machine__create_kernel_maps(&vmlinux
) < 0) {
95 pr_debug("machine__create_kernel_maps ");
99 vmlinux_map
= machine__kernel_map(&vmlinux
, type
);
100 map__kmap(vmlinux_map
)->ref_reloc_sym
= &ref_reloc_sym
;
105 * Locate a vmlinux file in the vmlinux path that has a buildid that
106 * matches the one of the running kernel.
108 * While doing that look if we find the ref reloc symbol, if we find it
109 * we'll have its ref_reloc_symbol.unrelocated_addr and then
110 * maps__reloc_vmlinux will notice and set proper ->[un]map_ip routines
111 * to fixup the symbols.
113 if (machine__load_vmlinux_path(&vmlinux
, type
,
114 vmlinux_matches_kallsyms_filter
) <= 0) {
115 pr_debug("machine__load_vmlinux_path ");
123 * Now look at the symbols in the vmlinux DSO and check if we find all of them
124 * in the kallsyms dso. For the ones that are in both, check its names and
127 for (nd
= rb_first(&vmlinux_map
->dso
->symbols
[type
]); nd
; nd
= rb_next(nd
)) {
128 struct symbol
*pair
, *first_pair
;
129 bool backwards
= true;
131 sym
= rb_entry(nd
, struct symbol
, rb_node
);
133 if (sym
->start
== sym
->end
)
136 first_pair
= machine__find_kernel_symbol(&kallsyms
, type
, sym
->start
, NULL
, NULL
);
139 if (pair
&& pair
->start
== sym
->start
) {
141 if (strcmp(sym
->name
, pair
->name
) == 0) {
143 * kallsyms don't have the symbol end, so we
144 * set that by using the next symbol start - 1,
145 * in some cases we get this up to a page
146 * wrong, trace_kmalloc when I was developing
147 * this code was one such example, 2106 bytes
148 * off the real size. More than that and we
149 * _really_ have a problem.
151 s64 skew
= sym
->end
- pair
->end
;
152 if (llabs(skew
) < page_size
)
155 pr_debug("%#" PRIx64
": diff end addr for %s v: %#" PRIx64
" k: %#" PRIx64
"\n",
156 sym
->start
, sym
->name
, sym
->end
, pair
->end
);
160 nnd
= backwards
? rb_prev(&pair
->rb_node
) :
161 rb_next(&pair
->rb_node
);
163 struct symbol
*next
= rb_entry(nnd
, struct symbol
, rb_node
);
165 if (next
->start
== sym
->start
) {
177 pr_debug("%#" PRIx64
": diff name v: %s k: %s\n",
178 sym
->start
, sym
->name
, pair
->name
);
181 pr_debug("%#" PRIx64
": %s not on kallsyms\n", sym
->start
, sym
->name
);
189 pr_info("Maps only in vmlinux:\n");
191 for (nd
= rb_first(&vmlinux
.kmaps
.maps
[type
]); nd
; nd
= rb_next(nd
)) {
192 struct map
*pos
= rb_entry(nd
, struct map
, rb_node
), *pair
;
194 * If it is the kernel, kallsyms is always "[kernel.kallsyms]", while
195 * the kernel will have the path for the vmlinux file being used,
196 * so use the short name, less descriptive but the same ("[kernel]" in
199 pair
= map_groups__find_by_name(&kallsyms
.kmaps
, type
,
201 pos
->dso
->short_name
:
206 map__fprintf(pos
, stderr
);
209 pr_info("Maps in vmlinux with a different name in kallsyms:\n");
211 for (nd
= rb_first(&vmlinux
.kmaps
.maps
[type
]); nd
; nd
= rb_next(nd
)) {
212 struct map
*pos
= rb_entry(nd
, struct map
, rb_node
), *pair
;
214 pair
= map_groups__find(&kallsyms
.kmaps
, type
, pos
->start
);
215 if (pair
== NULL
|| pair
->priv
)
218 if (pair
->start
== pos
->start
) {
220 pr_info(" %" PRIx64
"-%" PRIx64
" %" PRIx64
" %s in kallsyms as",
221 pos
->start
, pos
->end
, pos
->pgoff
, pos
->dso
->name
);
222 if (pos
->pgoff
!= pair
->pgoff
|| pos
->end
!= pair
->end
)
223 pr_info(": \n*%" PRIx64
"-%" PRIx64
" %" PRIx64
"",
224 pair
->start
, pair
->end
, pair
->pgoff
);
225 pr_info(" %s\n", pair
->dso
->name
);
230 pr_info("Maps only in kallsyms:\n");
232 for (nd
= rb_first(&kallsyms
.kmaps
.maps
[type
]);
233 nd
; nd
= rb_next(nd
)) {
234 struct map
*pos
= rb_entry(nd
, struct map
, rb_node
);
237 map__fprintf(pos
, stderr
);
243 #include "util/cpumap.h"
244 #include "util/evsel.h"
245 #include <sys/types.h>
247 static int trace_event__id(const char *evname
)
252 if (asprintf(&filename
,
254 tracing_events_path
, evname
) < 0)
257 fd
= open(filename
, O_RDONLY
);
260 if (read(fd
, id
, sizeof(id
)) > 0)
269 static int test__open_syscall_event(void)
272 struct thread_map
*threads
;
273 struct perf_evsel
*evsel
;
274 struct perf_event_attr attr
;
275 unsigned int nr_open_calls
= 111, i
;
276 int id
= trace_event__id("sys_enter_open");
279 pr_debug("is debugfs mounted on /sys/kernel/debug?\n");
283 threads
= thread_map__new(-1, getpid(), UINT_MAX
);
284 if (threads
== NULL
) {
285 pr_debug("thread_map__new\n");
289 memset(&attr
, 0, sizeof(attr
));
290 attr
.type
= PERF_TYPE_TRACEPOINT
;
292 evsel
= perf_evsel__new(&attr
, 0);
294 pr_debug("perf_evsel__new\n");
295 goto out_thread_map_delete
;
298 if (perf_evsel__open_per_thread(evsel
, threads
) < 0) {
299 pr_debug("failed to open counter: %s, "
300 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
302 goto out_evsel_delete
;
305 for (i
= 0; i
< nr_open_calls
; ++i
) {
306 fd
= open("/etc/passwd", O_RDONLY
);
310 if (perf_evsel__read_on_cpu(evsel
, 0, 0) < 0) {
311 pr_debug("perf_evsel__read_on_cpu\n");
315 if (evsel
->counts
->cpu
[0].val
!= nr_open_calls
) {
316 pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls, got %" PRIu64
"\n",
317 nr_open_calls
, evsel
->counts
->cpu
[0].val
);
323 perf_evsel__close_fd(evsel
, 1, threads
->nr
);
325 perf_evsel__delete(evsel
);
326 out_thread_map_delete
:
327 thread_map__delete(threads
);
333 static int test__open_syscall_event_on_all_cpus(void)
335 int err
= -1, fd
, cpu
;
336 struct thread_map
*threads
;
337 struct cpu_map
*cpus
;
338 struct perf_evsel
*evsel
;
339 struct perf_event_attr attr
;
340 unsigned int nr_open_calls
= 111, i
;
342 int id
= trace_event__id("sys_enter_open");
345 pr_debug("is debugfs mounted on /sys/kernel/debug?\n");
349 threads
= thread_map__new(-1, getpid(), UINT_MAX
);
350 if (threads
== NULL
) {
351 pr_debug("thread_map__new\n");
355 cpus
= cpu_map__new(NULL
);
357 pr_debug("cpu_map__new\n");
358 goto out_thread_map_delete
;
364 memset(&attr
, 0, sizeof(attr
));
365 attr
.type
= PERF_TYPE_TRACEPOINT
;
367 evsel
= perf_evsel__new(&attr
, 0);
369 pr_debug("perf_evsel__new\n");
370 goto out_thread_map_delete
;
373 if (perf_evsel__open(evsel
, cpus
, threads
) < 0) {
374 pr_debug("failed to open counter: %s, "
375 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
377 goto out_evsel_delete
;
380 for (cpu
= 0; cpu
< cpus
->nr
; ++cpu
) {
381 unsigned int ncalls
= nr_open_calls
+ cpu
;
383 * XXX eventually lift this restriction in a way that
384 * keeps perf building on older glibc installations
385 * without CPU_ALLOC. 1024 cpus in 2010 still seems
386 * a reasonable upper limit tho :-)
388 if (cpus
->map
[cpu
] >= CPU_SETSIZE
) {
389 pr_debug("Ignoring CPU %d\n", cpus
->map
[cpu
]);
393 CPU_SET(cpus
->map
[cpu
], &cpu_set
);
394 if (sched_setaffinity(0, sizeof(cpu_set
), &cpu_set
) < 0) {
395 pr_debug("sched_setaffinity() failed on CPU %d: %s ",
400 for (i
= 0; i
< ncalls
; ++i
) {
401 fd
= open("/etc/passwd", O_RDONLY
);
404 CPU_CLR(cpus
->map
[cpu
], &cpu_set
);
408 * Here we need to explicitely preallocate the counts, as if
409 * we use the auto allocation it will allocate just for 1 cpu,
410 * as we start by cpu 0.
412 if (perf_evsel__alloc_counts(evsel
, cpus
->nr
) < 0) {
413 pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus
->nr
);
419 for (cpu
= 0; cpu
< cpus
->nr
; ++cpu
) {
420 unsigned int expected
;
422 if (cpus
->map
[cpu
] >= CPU_SETSIZE
)
425 if (perf_evsel__read_on_cpu(evsel
, cpu
, 0) < 0) {
426 pr_debug("perf_evsel__read_on_cpu\n");
431 expected
= nr_open_calls
+ cpu
;
432 if (evsel
->counts
->cpu
[cpu
].val
!= expected
) {
433 pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %" PRIu64
"\n",
434 expected
, cpus
->map
[cpu
], evsel
->counts
->cpu
[cpu
].val
);
440 perf_evsel__close_fd(evsel
, 1, threads
->nr
);
442 perf_evsel__delete(evsel
);
443 out_thread_map_delete
:
444 thread_map__delete(threads
);
449 * This test will generate random numbers of calls to some getpid syscalls,
450 * then establish an mmap for a group of events that are created to monitor
453 * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
454 * sample.id field to map back to its respective perf_evsel instance.
456 * Then it checks if the number of syscalls reported as perf events by
457 * the kernel corresponds to the number of syscalls made.
459 static int test__basic_mmap(void)
462 union perf_event
*event
;
463 struct thread_map
*threads
;
464 struct cpu_map
*cpus
;
465 struct perf_evlist
*evlist
;
466 struct perf_event_attr attr
= {
467 .type
= PERF_TYPE_TRACEPOINT
,
468 .read_format
= PERF_FORMAT_ID
,
469 .sample_type
= PERF_SAMPLE_ID
,
473 const char *syscall_names
[] = { "getsid", "getppid", "getpgrp",
475 pid_t (*syscalls
[])(void) = { (void *)getsid
, getppid
, getpgrp
,
477 #define nsyscalls ARRAY_SIZE(syscall_names)
479 unsigned int nr_events
[nsyscalls
],
480 expected_nr_events
[nsyscalls
], i
, j
;
481 struct perf_evsel
*evsels
[nsyscalls
], *evsel
;
483 for (i
= 0; i
< nsyscalls
; ++i
) {
486 snprintf(name
, sizeof(name
), "sys_enter_%s", syscall_names
[i
]);
487 ids
[i
] = trace_event__id(name
);
489 pr_debug("Is debugfs mounted on /sys/kernel/debug?\n");
493 expected_nr_events
[i
] = random() % 257;
496 threads
= thread_map__new(-1, getpid(), UINT_MAX
);
497 if (threads
== NULL
) {
498 pr_debug("thread_map__new\n");
502 cpus
= cpu_map__new(NULL
);
504 pr_debug("cpu_map__new\n");
505 goto out_free_threads
;
509 CPU_SET(cpus
->map
[0], &cpu_set
);
510 sched_setaffinity(0, sizeof(cpu_set
), &cpu_set
);
511 if (sched_setaffinity(0, sizeof(cpu_set
), &cpu_set
) < 0) {
512 pr_debug("sched_setaffinity() failed on CPU %d: %s ",
513 cpus
->map
[0], strerror(errno
));
517 evlist
= perf_evlist__new(cpus
, threads
);
518 if (evlist
== NULL
) {
519 pr_debug("perf_evlist__new\n");
523 /* anonymous union fields, can't be initialized above */
524 attr
.wakeup_events
= 1;
525 attr
.sample_period
= 1;
527 for (i
= 0; i
< nsyscalls
; ++i
) {
528 attr
.config
= ids
[i
];
529 evsels
[i
] = perf_evsel__new(&attr
, i
);
530 if (evsels
[i
] == NULL
) {
531 pr_debug("perf_evsel__new\n");
532 goto out_free_evlist
;
535 perf_evlist__add(evlist
, evsels
[i
]);
537 if (perf_evsel__open(evsels
[i
], cpus
, threads
) < 0) {
538 pr_debug("failed to open counter: %s, "
539 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
545 if (perf_evlist__mmap(evlist
, 128, true) < 0) {
546 pr_debug("failed to mmap events: %d (%s)\n", errno
,
551 for (i
= 0; i
< nsyscalls
; ++i
)
552 for (j
= 0; j
< expected_nr_events
[i
]; ++j
) {
553 int foo
= syscalls
[i
]();
557 while ((event
= perf_evlist__mmap_read(evlist
, 0)) != NULL
) {
558 struct perf_sample sample
;
560 if (event
->header
.type
!= PERF_RECORD_SAMPLE
) {
561 pr_debug("unexpected %s event\n",
562 perf_event__name(event
->header
.type
));
566 err
= perf_evlist__parse_sample(evlist
, event
, &sample
, false);
568 pr_err("Can't parse sample, err = %d\n", err
);
572 evsel
= perf_evlist__id2evsel(evlist
, sample
.id
);
574 pr_debug("event with id %" PRIu64
575 " doesn't map to an evsel\n", sample
.id
);
578 nr_events
[evsel
->idx
]++;
581 list_for_each_entry(evsel
, &evlist
->entries
, node
) {
582 if (nr_events
[evsel
->idx
] != expected_nr_events
[evsel
->idx
]) {
583 pr_debug("expected %d %s events, got %d\n",
584 expected_nr_events
[evsel
->idx
],
585 perf_evsel__name(evsel
), nr_events
[evsel
->idx
]);
592 perf_evlist__munmap(evlist
);
594 for (i
= 0; i
< nsyscalls
; ++i
)
595 perf_evsel__close_fd(evsels
[i
], 1, threads
->nr
);
597 perf_evlist__delete(evlist
);
599 cpu_map__delete(cpus
);
601 thread_map__delete(threads
);
606 static int sched__get_first_possible_cpu(pid_t pid
, cpu_set_t
**maskp
,
611 int i
, cpu
= -1, nrcpus
= 1024;
613 mask
= CPU_ALLOC(nrcpus
);
614 size
= CPU_ALLOC_SIZE(nrcpus
);
615 CPU_ZERO_S(size
, mask
);
617 if (sched_getaffinity(pid
, size
, mask
) == -1) {
619 if (errno
== EINVAL
&& nrcpus
< (1024 << 8)) {
620 nrcpus
= nrcpus
<< 2;
623 perror("sched_getaffinity");
627 for (i
= 0; i
< nrcpus
; i
++) {
628 if (CPU_ISSET_S(i
, size
, mask
)) {
634 CPU_CLR_S(i
, size
, mask
);
644 static int test__PERF_RECORD(void)
646 struct perf_record_opts opts
= {
655 cpu_set_t
*cpu_mask
= NULL
;
656 size_t cpu_mask_size
= 0;
657 struct perf_evlist
*evlist
= perf_evlist__new(NULL
, NULL
);
658 struct perf_evsel
*evsel
;
659 struct perf_sample sample
;
660 const char *cmd
= "sleep";
661 const char *argv
[] = { cmd
, "1", NULL
, };
664 bool found_cmd_mmap
= false,
665 found_libc_mmap
= false,
666 found_vdso_mmap
= false,
667 found_ld_mmap
= false;
668 int err
= -1, errs
= 0, i
, wakeups
= 0;
670 int total_events
= 0, nr_events
[PERF_RECORD_MAX
] = { 0, };
672 if (evlist
== NULL
|| argv
== NULL
) {
673 pr_debug("Not enough memory to create evlist\n");
678 * We need at least one evsel in the evlist, use the default
681 err
= perf_evlist__add_default(evlist
);
683 pr_debug("Not enough memory to create evsel\n");
684 goto out_delete_evlist
;
688 * Create maps of threads and cpus to monitor. In this case
689 * we start with all threads and cpus (-1, -1) but then in
690 * perf_evlist__prepare_workload we'll fill in the only thread
691 * we're monitoring, the one forked there.
693 err
= perf_evlist__create_maps(evlist
, &opts
.target
);
695 pr_debug("Not enough memory to create thread/cpu maps\n");
696 goto out_delete_evlist
;
700 * Prepare the workload in argv[] to run, it'll fork it, and then wait
701 * for perf_evlist__start_workload() to exec it. This is done this way
702 * so that we have time to open the evlist (calling sys_perf_event_open
703 * on all the fds) and then mmap them.
705 err
= perf_evlist__prepare_workload(evlist
, &opts
, argv
);
707 pr_debug("Couldn't run the workload!\n");
708 goto out_delete_evlist
;
712 * Config the evsels, setting attr->comm on the first one, etc.
714 evsel
= perf_evlist__first(evlist
);
715 evsel
->attr
.sample_type
|= PERF_SAMPLE_CPU
;
716 evsel
->attr
.sample_type
|= PERF_SAMPLE_TID
;
717 evsel
->attr
.sample_type
|= PERF_SAMPLE_TIME
;
718 perf_evlist__config_attrs(evlist
, &opts
);
720 err
= sched__get_first_possible_cpu(evlist
->workload
.pid
, &cpu_mask
,
723 pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno
));
724 goto out_delete_evlist
;
730 * So that we can check perf_sample.cpu on all the samples.
732 if (sched_setaffinity(evlist
->workload
.pid
, cpu_mask_size
, cpu_mask
) < 0) {
733 pr_debug("sched_setaffinity: %s\n", strerror(errno
));
734 goto out_free_cpu_mask
;
738 * Call sys_perf_event_open on all the fds on all the evsels,
739 * grouping them if asked to.
741 err
= perf_evlist__open(evlist
);
743 pr_debug("perf_evlist__open: %s\n", strerror(errno
));
744 goto out_delete_evlist
;
748 * mmap the first fd on a given CPU and ask for events for the other
749 * fds in the same CPU to be injected in the same mmap ring buffer
750 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
752 err
= perf_evlist__mmap(evlist
, opts
.mmap_pages
, false);
754 pr_debug("perf_evlist__mmap: %s\n", strerror(errno
));
755 goto out_delete_evlist
;
759 * Now that all is properly set up, enable the events, they will
760 * count just on workload.pid, which will start...
762 perf_evlist__enable(evlist
);
767 perf_evlist__start_workload(evlist
);
770 int before
= total_events
;
772 for (i
= 0; i
< evlist
->nr_mmaps
; i
++) {
773 union perf_event
*event
;
775 while ((event
= perf_evlist__mmap_read(evlist
, i
)) != NULL
) {
776 const u32 type
= event
->header
.type
;
777 const char *name
= perf_event__name(type
);
780 if (type
< PERF_RECORD_MAX
)
783 err
= perf_evlist__parse_sample(evlist
, event
, &sample
, false);
786 perf_event__fprintf(event
, stderr
);
787 pr_debug("Couldn't parse sample\n");
792 pr_info("%" PRIu64
" %d ", sample
.time
, sample
.cpu
);
793 perf_event__fprintf(event
, stderr
);
796 if (prev_time
> sample
.time
) {
797 pr_debug("%s going backwards in time, prev=%" PRIu64
", curr=%" PRIu64
"\n",
798 name
, prev_time
, sample
.time
);
802 prev_time
= sample
.time
;
804 if (sample
.cpu
!= cpu
) {
805 pr_debug("%s with unexpected cpu, expected %d, got %d\n",
806 name
, cpu
, sample
.cpu
);
810 if ((pid_t
)sample
.pid
!= evlist
->workload
.pid
) {
811 pr_debug("%s with unexpected pid, expected %d, got %d\n",
812 name
, evlist
->workload
.pid
, sample
.pid
);
816 if ((pid_t
)sample
.tid
!= evlist
->workload
.pid
) {
817 pr_debug("%s with unexpected tid, expected %d, got %d\n",
818 name
, evlist
->workload
.pid
, sample
.tid
);
822 if ((type
== PERF_RECORD_COMM
||
823 type
== PERF_RECORD_MMAP
||
824 type
== PERF_RECORD_FORK
||
825 type
== PERF_RECORD_EXIT
) &&
826 (pid_t
)event
->comm
.pid
!= evlist
->workload
.pid
) {
827 pr_debug("%s with unexpected pid/tid\n", name
);
831 if ((type
== PERF_RECORD_COMM
||
832 type
== PERF_RECORD_MMAP
) &&
833 event
->comm
.pid
!= event
->comm
.tid
) {
834 pr_debug("%s with different pid/tid!\n", name
);
839 case PERF_RECORD_COMM
:
840 if (strcmp(event
->comm
.comm
, cmd
)) {
841 pr_debug("%s with unexpected comm!\n", name
);
845 case PERF_RECORD_EXIT
:
847 case PERF_RECORD_MMAP
:
848 bname
= strrchr(event
->mmap
.filename
, '/');
851 found_cmd_mmap
= !strcmp(bname
+ 1, cmd
);
852 if (!found_libc_mmap
)
853 found_libc_mmap
= !strncmp(bname
+ 1, "libc", 4);
855 found_ld_mmap
= !strncmp(bname
+ 1, "ld", 2);
856 } else if (!found_vdso_mmap
)
857 found_vdso_mmap
= !strcmp(event
->mmap
.filename
, "[vdso]");
860 case PERF_RECORD_SAMPLE
:
861 /* Just ignore samples for now */
864 pr_debug("Unexpected perf_event->header.type %d!\n",
872 * We don't use poll here because at least at 3.1 times the
873 * PERF_RECORD_{!SAMPLE} events don't honour
874 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
876 if (total_events
== before
&& false)
877 poll(evlist
->pollfd
, evlist
->nr_fds
, -1);
881 pr_debug("No PERF_RECORD_EXIT event!\n");
887 if (nr_events
[PERF_RECORD_COMM
] > 1) {
888 pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
892 if (nr_events
[PERF_RECORD_COMM
] == 0) {
893 pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd
);
897 if (!found_cmd_mmap
) {
898 pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd
);
902 if (!found_libc_mmap
) {
903 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
907 if (!found_ld_mmap
) {
908 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
912 if (!found_vdso_mmap
) {
913 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
917 perf_evlist__munmap(evlist
);
921 perf_evlist__delete(evlist
);
923 return (err
< 0 || errs
> 0) ? -1 : 0;
927 #if defined(__x86_64__) || defined(__i386__)
929 #define barrier() asm volatile("" ::: "memory")
931 static u64
rdpmc(unsigned int counter
)
933 unsigned int low
, high
;
935 asm volatile("rdpmc" : "=a" (low
), "=d" (high
) : "c" (counter
));
937 return low
| ((u64
)high
) << 32;
940 static u64
rdtsc(void)
942 unsigned int low
, high
;
944 asm volatile("rdtsc" : "=a" (low
), "=d" (high
));
946 return low
| ((u64
)high
) << 32;
949 static u64
mmap_read_self(void *addr
)
951 struct perf_event_mmap_page
*pc
= addr
;
952 u32 seq
, idx
, time_mult
= 0, time_shift
= 0;
953 u64 count
, cyc
= 0, time_offset
= 0, enabled
, running
, delta
;
959 enabled
= pc
->time_enabled
;
960 running
= pc
->time_running
;
962 if (enabled
!= running
) {
964 time_mult
= pc
->time_mult
;
965 time_shift
= pc
->time_shift
;
966 time_offset
= pc
->time_offset
;
972 count
+= rdpmc(idx
- 1);
975 } while (pc
->lock
!= seq
);
977 if (enabled
!= running
) {
980 quot
= (cyc
>> time_shift
);
981 rem
= cyc
& ((1 << time_shift
) - 1);
982 delta
= time_offset
+ quot
* time_mult
+
983 ((rem
* time_mult
) >> time_shift
);
989 quot
= count
/ running
;
990 rem
= count
% running
;
991 count
= quot
* enabled
+ (rem
* enabled
) / running
;
998 * If the RDPMC instruction faults then signal this back to the test parent task:
1000 static void segfault_handler(int sig __maybe_unused
,
1001 siginfo_t
*info __maybe_unused
,
1002 void *uc __maybe_unused
)
1007 static int __test__rdpmc(void)
1009 long page_size
= sysconf(_SC_PAGE_SIZE
);
1010 volatile int tmp
= 0;
1011 u64 i
, loops
= 1000;
1015 struct perf_event_attr attr
= {
1016 .type
= PERF_TYPE_HARDWARE
,
1017 .config
= PERF_COUNT_HW_INSTRUCTIONS
,
1018 .exclude_kernel
= 1,
1021 struct sigaction sa
;
1023 sigfillset(&sa
.sa_mask
);
1024 sa
.sa_sigaction
= segfault_handler
;
1025 sigaction(SIGSEGV
, &sa
, NULL
);
1027 fd
= sys_perf_event_open(&attr
, 0, -1, -1, 0);
1029 pr_debug("Error: sys_perf_event_open() syscall returned "
1030 "with %d (%s)\n", fd
, strerror(errno
));
1034 addr
= mmap(NULL
, page_size
, PROT_READ
, MAP_SHARED
, fd
, 0);
1035 if (addr
== (void *)(-1)) {
1036 pr_debug("Error: mmap() syscall returned with (%s)\n",
1041 for (n
= 0; n
< 6; n
++) {
1042 u64 stamp
, now
, delta
;
1044 stamp
= mmap_read_self(addr
);
1046 for (i
= 0; i
< loops
; i
++)
1049 now
= mmap_read_self(addr
);
1052 delta
= now
- stamp
;
1053 pr_debug("%14d: %14Lu\n", n
, (long long)delta
);
1058 munmap(addr
, page_size
);
1069 static int test__rdpmc(void)
1081 ret
= __test__rdpmc();
1086 wret
= waitpid(pid
, &status
, 0);
1087 if (wret
< 0 || status
)
1095 static int test__perf_pmu(void)
1097 return perf_pmu__test();
1100 static int perf_evsel__roundtrip_cache_name_test(void)
1103 int type
, op
, err
= 0, ret
= 0, i
, idx
;
1104 struct perf_evsel
*evsel
;
1105 struct perf_evlist
*evlist
= perf_evlist__new(NULL
, NULL
);
1110 for (type
= 0; type
< PERF_COUNT_HW_CACHE_MAX
; type
++) {
1111 for (op
= 0; op
< PERF_COUNT_HW_CACHE_OP_MAX
; op
++) {
1112 /* skip invalid cache type */
1113 if (!perf_evsel__is_cache_op_valid(type
, op
))
1116 for (i
= 0; i
< PERF_COUNT_HW_CACHE_RESULT_MAX
; i
++) {
1117 __perf_evsel__hw_cache_type_op_res_name(type
, op
, i
,
1118 name
, sizeof(name
));
1119 err
= parse_events(evlist
, name
, 0);
1127 evsel
= perf_evlist__first(evlist
);
1129 for (type
= 0; type
< PERF_COUNT_HW_CACHE_MAX
; type
++) {
1130 for (op
= 0; op
< PERF_COUNT_HW_CACHE_OP_MAX
; op
++) {
1131 /* skip invalid cache type */
1132 if (!perf_evsel__is_cache_op_valid(type
, op
))
1135 for (i
= 0; i
< PERF_COUNT_HW_CACHE_RESULT_MAX
; i
++) {
1136 __perf_evsel__hw_cache_type_op_res_name(type
, op
, i
,
1137 name
, sizeof(name
));
1138 if (evsel
->idx
!= idx
)
1143 if (strcmp(perf_evsel__name(evsel
), name
)) {
1144 pr_debug("%s != %s\n", perf_evsel__name(evsel
), name
);
1148 evsel
= perf_evsel__next(evsel
);
1153 perf_evlist__delete(evlist
);
1157 static int __perf_evsel__name_array_test(const char *names
[], int nr_names
)
1160 struct perf_evsel
*evsel
;
1161 struct perf_evlist
*evlist
= perf_evlist__new(NULL
, NULL
);
1166 for (i
= 0; i
< nr_names
; ++i
) {
1167 err
= parse_events(evlist
, names
[i
], 0);
1169 pr_debug("failed to parse event '%s', err %d\n",
1171 goto out_delete_evlist
;
1176 list_for_each_entry(evsel
, &evlist
->entries
, node
) {
1177 if (strcmp(perf_evsel__name(evsel
), names
[evsel
->idx
])) {
1179 pr_debug("%s != %s\n", perf_evsel__name(evsel
), names
[evsel
->idx
]);
1184 perf_evlist__delete(evlist
);
1188 #define perf_evsel__name_array_test(names) \
1189 __perf_evsel__name_array_test(names, ARRAY_SIZE(names))
1191 static int perf_evsel__roundtrip_name_test(void)
1193 int err
= 0, ret
= 0;
1195 err
= perf_evsel__name_array_test(perf_evsel__hw_names
);
1199 err
= perf_evsel__name_array_test(perf_evsel__sw_names
);
1203 err
= perf_evsel__roundtrip_cache_name_test();
1210 static struct test
{
1215 .desc
= "vmlinux symtab matches kallsyms",
1216 .func
= test__vmlinux_matches_kallsyms
,
1219 .desc
= "detect open syscall event",
1220 .func
= test__open_syscall_event
,
1223 .desc
= "detect open syscall event on all cpus",
1224 .func
= test__open_syscall_event_on_all_cpus
,
1227 .desc
= "read samples using the mmap interface",
1228 .func
= test__basic_mmap
,
1231 .desc
= "parse events tests",
1232 .func
= parse_events__test
,
1234 #if defined(__x86_64__) || defined(__i386__)
1236 .desc
= "x86 rdpmc test",
1237 .func
= test__rdpmc
,
1241 .desc
= "Validate PERF_RECORD_* events & perf_sample fields",
1242 .func
= test__PERF_RECORD
,
1245 .desc
= "Test perf pmu format parsing",
1246 .func
= test__perf_pmu
,
1249 .desc
= "Test dso data interface",
1250 .func
= dso__test_data
,
1253 .desc
= "roundtrip evsel->name check",
1254 .func
= perf_evsel__roundtrip_name_test
,
1261 static bool perf_test__matches(int curr
, int argc
, const char *argv
[])
1268 for (i
= 0; i
< argc
; ++i
) {
1270 long nr
= strtoul(argv
[i
], &end
, 10);
1278 if (strstr(tests
[curr
].desc
, argv
[i
]))
1285 static int __cmd_test(int argc
, const char *argv
[])
1289 while (tests
[i
].func
) {
1290 int curr
= i
++, err
;
1292 if (!perf_test__matches(curr
, argc
, argv
))
1295 pr_info("%2d: %s:", i
, tests
[curr
].desc
);
1296 pr_debug("\n--- start ---\n");
1297 err
= tests
[curr
].func();
1298 pr_debug("---- end ----\n%s:", tests
[curr
].desc
);
1299 pr_info(" %s\n", err
? "FAILED!\n" : "Ok");
1305 static int perf_test__list(int argc
, const char **argv
)
1309 while (tests
[i
].func
) {
1312 if (argc
> 1 && !strstr(tests
[curr
].desc
, argv
[1]))
1315 pr_info("%2d: %s\n", i
, tests
[curr
].desc
);
1321 int cmd_test(int argc
, const char **argv
, const char *prefix __maybe_unused
)
1323 const char * const test_usage
[] = {
1324 "perf test [<options>] [{list <test-name-fragment>|[<test-name-fragments>|<test-numbers>]}]",
1327 const struct option test_options
[] = {
1328 OPT_INCR('v', "verbose", &verbose
,
1329 "be more verbose (show symbol address, etc)"),
1333 argc
= parse_options(argc
, argv
, test_options
, test_usage
, 0);
1334 if (argc
>= 1 && !strcmp(argv
[0], "list"))
1335 return perf_test__list(argc
, argv
);
1337 symbol_conf
.priv_size
= sizeof(int);
1338 symbol_conf
.sort_by_name
= true;
1339 symbol_conf
.try_vmlinux_path
= true;
1341 if (symbol__init() < 0)
1344 return __cmd_test(argc
, argv
);