Root/tools/perf/builtin-test.c

1/*
2 * builtin-test.c
3 *
4 * Builtin regression testing command: ever growing number of sanity tests
5 */
6#include "builtin.h"
7
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"
16#include "util/pmu.h"
17#include "../../include/linux/hw_breakpoint.h"
18
19#include <sys/mman.h>
20
21static int vmlinux_matches_kallsyms_filter(struct map *map __used, struct symbol *sym)
22{
23    bool *visited = symbol__priv(sym);
24    *visited = true;
25    return 0;
26}
27
28static int test__vmlinux_matches_kallsyms(void)
29{
30    int err = -1;
31    struct rb_node *nd;
32    struct symbol *sym;
33    struct map *kallsyms_map, *vmlinux_map;
34    struct machine kallsyms, vmlinux;
35    enum map_type type = MAP__FUNCTION;
36    long page_size = sysconf(_SC_PAGE_SIZE);
37    struct ref_reloc_sym ref_reloc_sym = { .name = "_stext", };
38
39    /*
40     * Step 1:
41     *
42     * Init the machines that will hold kernel, modules obtained from
43     * both vmlinux + .ko files and from /proc/kallsyms split by modules.
44     */
45    machine__init(&kallsyms, "", HOST_KERNEL_ID);
46    machine__init(&vmlinux, "", HOST_KERNEL_ID);
47
48    /*
49     * Step 2:
50     *
51     * Create the kernel maps for kallsyms and the DSO where we will then
52     * load /proc/kallsyms. Also create the modules maps from /proc/modules
53     * and find the .ko files that match them in /lib/modules/`uname -r`/.
54     */
55    if (machine__create_kernel_maps(&kallsyms) < 0) {
56        pr_debug("machine__create_kernel_maps ");
57        return -1;
58    }
59
60    /*
61     * Step 3:
62     *
63     * Load and split /proc/kallsyms into multiple maps, one per module.
64     */
65    if (machine__load_kallsyms(&kallsyms, "/proc/kallsyms", type, NULL) <= 0) {
66        pr_debug("dso__load_kallsyms ");
67        goto out;
68    }
69
70    /*
71     * Step 4:
72     *
73     * kallsyms will be internally on demand sorted by name so that we can
74     * find the reference relocation * symbol, i.e. the symbol we will use
75     * to see if the running kernel was relocated by checking if it has the
76     * same value in the vmlinux file we load.
77     */
78    kallsyms_map = machine__kernel_map(&kallsyms, type);
79
80    sym = map__find_symbol_by_name(kallsyms_map, ref_reloc_sym.name, NULL);
81    if (sym == NULL) {
82        pr_debug("dso__find_symbol_by_name ");
83        goto out;
84    }
85
86    ref_reloc_sym.addr = sym->start;
87
88    /*
89     * Step 5:
90     *
91     * Now repeat step 2, this time for the vmlinux file we'll auto-locate.
92     */
93    if (machine__create_kernel_maps(&vmlinux) < 0) {
94        pr_debug("machine__create_kernel_maps ");
95        goto out;
96    }
97
98    vmlinux_map = machine__kernel_map(&vmlinux, type);
99    map__kmap(vmlinux_map)->ref_reloc_sym = &ref_reloc_sym;
100
101    /*
102     * Step 6:
103     *
104     * Locate a vmlinux file in the vmlinux path that has a buildid that
105     * matches the one of the running kernel.
106     *
107     * While doing that look if we find the ref reloc symbol, if we find it
108     * we'll have its ref_reloc_symbol.unrelocated_addr and then
109     * maps__reloc_vmlinux will notice and set proper ->[un]map_ip routines
110     * to fixup the symbols.
111     */
112    if (machine__load_vmlinux_path(&vmlinux, type,
113                       vmlinux_matches_kallsyms_filter) <= 0) {
114        pr_debug("machine__load_vmlinux_path ");
115        goto out;
116    }
117
118    err = 0;
119    /*
120     * Step 7:
121     *
122     * Now look at the symbols in the vmlinux DSO and check if we find all of them
123     * in the kallsyms dso. For the ones that are in both, check its names and
124     * end addresses too.
125     */
126    for (nd = rb_first(&vmlinux_map->dso->symbols[type]); nd; nd = rb_next(nd)) {
127        struct symbol *pair, *first_pair;
128        bool backwards = true;
129
130        sym = rb_entry(nd, struct symbol, rb_node);
131
132        if (sym->start == sym->end)
133            continue;
134
135        first_pair = machine__find_kernel_symbol(&kallsyms, type, sym->start, NULL, NULL);
136        pair = first_pair;
137
138        if (pair && pair->start == sym->start) {
139next_pair:
140            if (strcmp(sym->name, pair->name) == 0) {
141                /*
142                 * kallsyms don't have the symbol end, so we
143                 * set that by using the next symbol start - 1,
144                 * in some cases we get this up to a page
145                 * wrong, trace_kmalloc when I was developing
146                 * this code was one such example, 2106 bytes
147                 * off the real size. More than that and we
148                 * _really_ have a problem.
149                 */
150                s64 skew = sym->end - pair->end;
151                if (llabs(skew) < page_size)
152                    continue;
153
154                pr_debug("%#" PRIx64 ": diff end addr for %s v: %#" PRIx64 " k: %#" PRIx64 "\n",
155                     sym->start, sym->name, sym->end, pair->end);
156            } else {
157                struct rb_node *nnd;
158detour:
159                nnd = backwards ? rb_prev(&pair->rb_node) :
160                          rb_next(&pair->rb_node);
161                if (nnd) {
162                    struct symbol *next = rb_entry(nnd, struct symbol, rb_node);
163
164                    if (next->start == sym->start) {
165                        pair = next;
166                        goto next_pair;
167                    }
168                }
169
170                if (backwards) {
171                    backwards = false;
172                    pair = first_pair;
173                    goto detour;
174                }
175
176                pr_debug("%#" PRIx64 ": diff name v: %s k: %s\n",
177                     sym->start, sym->name, pair->name);
178            }
179        } else
180            pr_debug("%#" PRIx64 ": %s not on kallsyms\n", sym->start, sym->name);
181
182        err = -1;
183    }
184
185    if (!verbose)
186        goto out;
187
188    pr_info("Maps only in vmlinux:\n");
189
190    for (nd = rb_first(&vmlinux.kmaps.maps[type]); nd; nd = rb_next(nd)) {
191        struct map *pos = rb_entry(nd, struct map, rb_node), *pair;
192        /*
193         * If it is the kernel, kallsyms is always "[kernel.kallsyms]", while
194         * the kernel will have the path for the vmlinux file being used,
195         * so use the short name, less descriptive but the same ("[kernel]" in
196         * both cases.
197         */
198        pair = map_groups__find_by_name(&kallsyms.kmaps, type,
199                        (pos->dso->kernel ?
200                            pos->dso->short_name :
201                            pos->dso->name));
202        if (pair)
203            pair->priv = 1;
204        else
205            map__fprintf(pos, stderr);
206    }
207
208    pr_info("Maps in vmlinux with a different name in kallsyms:\n");
209
210    for (nd = rb_first(&vmlinux.kmaps.maps[type]); nd; nd = rb_next(nd)) {
211        struct map *pos = rb_entry(nd, struct map, rb_node), *pair;
212
213        pair = map_groups__find(&kallsyms.kmaps, type, pos->start);
214        if (pair == NULL || pair->priv)
215            continue;
216
217        if (pair->start == pos->start) {
218            pair->priv = 1;
219            pr_info(" %" PRIx64 "-%" PRIx64 " %" PRIx64 " %s in kallsyms as",
220                pos->start, pos->end, pos->pgoff, pos->dso->name);
221            if (pos->pgoff != pair->pgoff || pos->end != pair->end)
222                pr_info(": \n*%" PRIx64 "-%" PRIx64 " %" PRIx64 "",
223                    pair->start, pair->end, pair->pgoff);
224            pr_info(" %s\n", pair->dso->name);
225            pair->priv = 1;
226        }
227    }
228
229    pr_info("Maps only in kallsyms:\n");
230
231    for (nd = rb_first(&kallsyms.kmaps.maps[type]);
232         nd; nd = rb_next(nd)) {
233        struct map *pos = rb_entry(nd, struct map, rb_node);
234
235        if (!pos->priv)
236            map__fprintf(pos, stderr);
237    }
238out:
239    return err;
240}
241
242#include "util/cpumap.h"
243#include "util/evsel.h"
244#include <sys/types.h>
245
246static int trace_event__id(const char *evname)
247{
248    char *filename;
249    int err = -1, fd;
250
251    if (asprintf(&filename,
252             "%s/syscalls/%s/id",
253             tracing_events_path, evname) < 0)
254        return -1;
255
256    fd = open(filename, O_RDONLY);
257    if (fd >= 0) {
258        char id[16];
259        if (read(fd, id, sizeof(id)) > 0)
260            err = atoi(id);
261        close(fd);
262    }
263
264    free(filename);
265    return err;
266}
267
268static int test__open_syscall_event(void)
269{
270    int err = -1, fd;
271    struct thread_map *threads;
272    struct perf_evsel *evsel;
273    struct perf_event_attr attr;
274    unsigned int nr_open_calls = 111, i;
275    int id = trace_event__id("sys_enter_open");
276
277    if (id < 0) {
278        pr_debug("is debugfs mounted on /sys/kernel/debug?\n");
279        return -1;
280    }
281
282    threads = thread_map__new(-1, getpid(), UINT_MAX);
283    if (threads == NULL) {
284        pr_debug("thread_map__new\n");
285        return -1;
286    }
287
288    memset(&attr, 0, sizeof(attr));
289    attr.type = PERF_TYPE_TRACEPOINT;
290    attr.config = id;
291    evsel = perf_evsel__new(&attr, 0);
292    if (evsel == NULL) {
293        pr_debug("perf_evsel__new\n");
294        goto out_thread_map_delete;
295    }
296
297    if (perf_evsel__open_per_thread(evsel, threads, false, NULL) < 0) {
298        pr_debug("failed to open counter: %s, "
299             "tweak /proc/sys/kernel/perf_event_paranoid?\n",
300             strerror(errno));
301        goto out_evsel_delete;
302    }
303
304    for (i = 0; i < nr_open_calls; ++i) {
305        fd = open("/etc/passwd", O_RDONLY);
306        close(fd);
307    }
308
309    if (perf_evsel__read_on_cpu(evsel, 0, 0) < 0) {
310        pr_debug("perf_evsel__read_on_cpu\n");
311        goto out_close_fd;
312    }
313
314    if (evsel->counts->cpu[0].val != nr_open_calls) {
315        pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls, got %" PRIu64 "\n",
316             nr_open_calls, evsel->counts->cpu[0].val);
317        goto out_close_fd;
318    }
319    
320    err = 0;
321out_close_fd:
322    perf_evsel__close_fd(evsel, 1, threads->nr);
323out_evsel_delete:
324    perf_evsel__delete(evsel);
325out_thread_map_delete:
326    thread_map__delete(threads);
327    return err;
328}
329
330#include <sched.h>
331
332static int test__open_syscall_event_on_all_cpus(void)
333{
334    int err = -1, fd, cpu;
335    struct thread_map *threads;
336    struct cpu_map *cpus;
337    struct perf_evsel *evsel;
338    struct perf_event_attr attr;
339    unsigned int nr_open_calls = 111, i;
340    cpu_set_t cpu_set;
341    int id = trace_event__id("sys_enter_open");
342
343    if (id < 0) {
344        pr_debug("is debugfs mounted on /sys/kernel/debug?\n");
345        return -1;
346    }
347
348    threads = thread_map__new(-1, getpid(), UINT_MAX);
349    if (threads == NULL) {
350        pr_debug("thread_map__new\n");
351        return -1;
352    }
353
354    cpus = cpu_map__new(NULL);
355    if (cpus == NULL) {
356        pr_debug("cpu_map__new\n");
357        goto out_thread_map_delete;
358    }
359
360
361    CPU_ZERO(&cpu_set);
362
363    memset(&attr, 0, sizeof(attr));
364    attr.type = PERF_TYPE_TRACEPOINT;
365    attr.config = id;
366    evsel = perf_evsel__new(&attr, 0);
367    if (evsel == NULL) {
368        pr_debug("perf_evsel__new\n");
369        goto out_thread_map_delete;
370    }
371
372    if (perf_evsel__open(evsel, cpus, threads, false, NULL) < 0) {
373        pr_debug("failed to open counter: %s, "
374             "tweak /proc/sys/kernel/perf_event_paranoid?\n",
375             strerror(errno));
376        goto out_evsel_delete;
377    }
378
379    for (cpu = 0; cpu < cpus->nr; ++cpu) {
380        unsigned int ncalls = nr_open_calls + cpu;
381        /*
382         * XXX eventually lift this restriction in a way that
383         * keeps perf building on older glibc installations
384         * without CPU_ALLOC. 1024 cpus in 2010 still seems
385         * a reasonable upper limit tho :-)
386         */
387        if (cpus->map[cpu] >= CPU_SETSIZE) {
388            pr_debug("Ignoring CPU %d\n", cpus->map[cpu]);
389            continue;
390        }
391
392        CPU_SET(cpus->map[cpu], &cpu_set);
393        if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) {
394            pr_debug("sched_setaffinity() failed on CPU %d: %s ",
395                 cpus->map[cpu],
396                 strerror(errno));
397            goto out_close_fd;
398        }
399        for (i = 0; i < ncalls; ++i) {
400            fd = open("/etc/passwd", O_RDONLY);
401            close(fd);
402        }
403        CPU_CLR(cpus->map[cpu], &cpu_set);
404    }
405
406    /*
407     * Here we need to explicitely preallocate the counts, as if
408     * we use the auto allocation it will allocate just for 1 cpu,
409     * as we start by cpu 0.
410     */
411    if (perf_evsel__alloc_counts(evsel, cpus->nr) < 0) {
412        pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus->nr);
413        goto out_close_fd;
414    }
415
416    err = 0;
417
418    for (cpu = 0; cpu < cpus->nr; ++cpu) {
419        unsigned int expected;
420
421        if (cpus->map[cpu] >= CPU_SETSIZE)
422            continue;
423
424        if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) {
425            pr_debug("perf_evsel__read_on_cpu\n");
426            err = -1;
427            break;
428        }
429
430        expected = nr_open_calls + cpu;
431        if (evsel->counts->cpu[cpu].val != expected) {
432            pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %" PRIu64 "\n",
433                 expected, cpus->map[cpu], evsel->counts->cpu[cpu].val);
434            err = -1;
435        }
436    }
437
438out_close_fd:
439    perf_evsel__close_fd(evsel, 1, threads->nr);
440out_evsel_delete:
441    perf_evsel__delete(evsel);
442out_thread_map_delete:
443    thread_map__delete(threads);
444    return err;
445}
446
447/*
448 * This test will generate random numbers of calls to some getpid syscalls,
449 * then establish an mmap for a group of events that are created to monitor
450 * the syscalls.
451 *
452 * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
453 * sample.id field to map back to its respective perf_evsel instance.
454 *
455 * Then it checks if the number of syscalls reported as perf events by
456 * the kernel corresponds to the number of syscalls made.
457 */
458static int test__basic_mmap(void)
459{
460    int err = -1;
461    union perf_event *event;
462    struct thread_map *threads;
463    struct cpu_map *cpus;
464    struct perf_evlist *evlist;
465    struct perf_event_attr attr = {
466        .type = PERF_TYPE_TRACEPOINT,
467        .read_format = PERF_FORMAT_ID,
468        .sample_type = PERF_SAMPLE_ID,
469        .watermark = 0,
470    };
471    cpu_set_t cpu_set;
472    const char *syscall_names[] = { "getsid", "getppid", "getpgrp",
473                    "getpgid", };
474    pid_t (*syscalls[])(void) = { (void *)getsid, getppid, getpgrp,
475                      (void*)getpgid };
476#define nsyscalls ARRAY_SIZE(syscall_names)
477    int ids[nsyscalls];
478    unsigned int nr_events[nsyscalls],
479             expected_nr_events[nsyscalls], i, j;
480    struct perf_evsel *evsels[nsyscalls], *evsel;
481    int sample_size = __perf_evsel__sample_size(attr.sample_type);
482
483    for (i = 0; i < nsyscalls; ++i) {
484        char name[64];
485
486        snprintf(name, sizeof(name), "sys_enter_%s", syscall_names[i]);
487        ids[i] = trace_event__id(name);
488        if (ids[i] < 0) {
489            pr_debug("Is debugfs mounted on /sys/kernel/debug?\n");
490            return -1;
491        }
492        nr_events[i] = 0;
493        expected_nr_events[i] = random() % 257;
494    }
495
496    threads = thread_map__new(-1, getpid(), UINT_MAX);
497    if (threads == NULL) {
498        pr_debug("thread_map__new\n");
499        return -1;
500    }
501
502    cpus = cpu_map__new(NULL);
503    if (cpus == NULL) {
504        pr_debug("cpu_map__new\n");
505        goto out_free_threads;
506    }
507
508    CPU_ZERO(&cpu_set);
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));
514        goto out_free_cpus;
515    }
516
517    evlist = perf_evlist__new(cpus, threads);
518    if (evlist == NULL) {
519        pr_debug("perf_evlist__new\n");
520        goto out_free_cpus;
521    }
522
523    /* anonymous union fields, can't be initialized above */
524    attr.wakeup_events = 1;
525    attr.sample_period = 1;
526
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;
533        }
534
535        perf_evlist__add(evlist, evsels[i]);
536
537        if (perf_evsel__open(evsels[i], cpus, threads, false, NULL) < 0) {
538            pr_debug("failed to open counter: %s, "
539                 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
540                 strerror(errno));
541            goto out_close_fd;
542        }
543    }
544
545    if (perf_evlist__mmap(evlist, 128, true) < 0) {
546        pr_debug("failed to mmap events: %d (%s)\n", errno,
547             strerror(errno));
548        goto out_close_fd;
549    }
550
551    for (i = 0; i < nsyscalls; ++i)
552        for (j = 0; j < expected_nr_events[i]; ++j) {
553            int foo = syscalls[i]();
554            ++foo;
555        }
556
557    while ((event = perf_evlist__mmap_read(evlist, 0)) != NULL) {
558        struct perf_sample sample;
559
560        if (event->header.type != PERF_RECORD_SAMPLE) {
561            pr_debug("unexpected %s event\n",
562                 perf_event__name(event->header.type));
563            goto out_munmap;
564        }
565
566        err = perf_event__parse_sample(event, attr.sample_type, sample_size,
567                           false, &sample, false);
568        if (err) {
569            pr_err("Can't parse sample, err = %d\n", err);
570            goto out_munmap;
571        }
572
573        evsel = perf_evlist__id2evsel(evlist, sample.id);
574        if (evsel == NULL) {
575            pr_debug("event with id %" PRIu64
576                 " doesn't map to an evsel\n", sample.id);
577            goto out_munmap;
578        }
579        nr_events[evsel->idx]++;
580    }
581
582    list_for_each_entry(evsel, &evlist->entries, node) {
583        if (nr_events[evsel->idx] != expected_nr_events[evsel->idx]) {
584            pr_debug("expected %d %s events, got %d\n",
585                 expected_nr_events[evsel->idx],
586                 event_name(evsel), nr_events[evsel->idx]);
587            goto out_munmap;
588        }
589    }
590
591    err = 0;
592out_munmap:
593    perf_evlist__munmap(evlist);
594out_close_fd:
595    for (i = 0; i < nsyscalls; ++i)
596        perf_evsel__close_fd(evsels[i], 1, threads->nr);
597out_free_evlist:
598    perf_evlist__delete(evlist);
599out_free_cpus:
600    cpu_map__delete(cpus);
601out_free_threads:
602    thread_map__delete(threads);
603    return err;
604#undef nsyscalls
605}
606
607#define TEST_ASSERT_VAL(text, cond) \
608do { \
609    if (!(cond)) { \
610        pr_debug("FAILED %s:%d %s\n", __FILE__, __LINE__, text); \
611        return -1; \
612    } \
613} while (0)
614
615static int test__checkevent_tracepoint(struct perf_evlist *evlist)
616{
617    struct perf_evsel *evsel = list_entry(evlist->entries.next,
618                          struct perf_evsel, node);
619
620    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
621    TEST_ASSERT_VAL("wrong type", PERF_TYPE_TRACEPOINT == evsel->attr.type);
622    TEST_ASSERT_VAL("wrong sample_type",
623        (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU) ==
624        evsel->attr.sample_type);
625    TEST_ASSERT_VAL("wrong sample_period", 1 == evsel->attr.sample_period);
626    return 0;
627}
628
629static int test__checkevent_tracepoint_multi(struct perf_evlist *evlist)
630{
631    struct perf_evsel *evsel;
632
633    TEST_ASSERT_VAL("wrong number of entries", evlist->nr_entries > 1);
634
635    list_for_each_entry(evsel, &evlist->entries, node) {
636        TEST_ASSERT_VAL("wrong type",
637            PERF_TYPE_TRACEPOINT == evsel->attr.type);
638        TEST_ASSERT_VAL("wrong sample_type",
639            (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU)
640            == evsel->attr.sample_type);
641        TEST_ASSERT_VAL("wrong sample_period",
642            1 == evsel->attr.sample_period);
643    }
644    return 0;
645}
646
647static int test__checkevent_raw(struct perf_evlist *evlist)
648{
649    struct perf_evsel *evsel = list_entry(evlist->entries.next,
650                          struct perf_evsel, node);
651
652    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
653    TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type);
654    TEST_ASSERT_VAL("wrong config", 0x1a == evsel->attr.config);
655    return 0;
656}
657
658static int test__checkevent_numeric(struct perf_evlist *evlist)
659{
660    struct perf_evsel *evsel = list_entry(evlist->entries.next,
661                          struct perf_evsel, node);
662
663    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
664    TEST_ASSERT_VAL("wrong type", 1 == evsel->attr.type);
665    TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config);
666    return 0;
667}
668
669static int test__checkevent_symbolic_name(struct perf_evlist *evlist)
670{
671    struct perf_evsel *evsel = list_entry(evlist->entries.next,
672                          struct perf_evsel, node);
673
674    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
675    TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE == evsel->attr.type);
676    TEST_ASSERT_VAL("wrong config",
677            PERF_COUNT_HW_INSTRUCTIONS == evsel->attr.config);
678    return 0;
679}
680
681static int test__checkevent_symbolic_name_config(struct perf_evlist *evlist)
682{
683    struct perf_evsel *evsel = list_entry(evlist->entries.next,
684                          struct perf_evsel, node);
685
686    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
687    TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE == evsel->attr.type);
688    TEST_ASSERT_VAL("wrong config",
689            PERF_COUNT_HW_CPU_CYCLES == evsel->attr.config);
690    TEST_ASSERT_VAL("wrong period",
691            100000 == evsel->attr.sample_period);
692    TEST_ASSERT_VAL("wrong config1",
693            0 == evsel->attr.config1);
694    TEST_ASSERT_VAL("wrong config2",
695            1 == evsel->attr.config2);
696    return 0;
697}
698
699static int test__checkevent_symbolic_alias(struct perf_evlist *evlist)
700{
701    struct perf_evsel *evsel = list_entry(evlist->entries.next,
702                          struct perf_evsel, node);
703
704    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
705    TEST_ASSERT_VAL("wrong type", PERF_TYPE_SOFTWARE == evsel->attr.type);
706    TEST_ASSERT_VAL("wrong config",
707            PERF_COUNT_SW_PAGE_FAULTS == evsel->attr.config);
708    return 0;
709}
710
711static int test__checkevent_genhw(struct perf_evlist *evlist)
712{
713    struct perf_evsel *evsel = list_entry(evlist->entries.next,
714                          struct perf_evsel, node);
715
716    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
717    TEST_ASSERT_VAL("wrong type", PERF_TYPE_HW_CACHE == evsel->attr.type);
718    TEST_ASSERT_VAL("wrong config", (1 << 16) == evsel->attr.config);
719    return 0;
720}
721
722static int test__checkevent_breakpoint(struct perf_evlist *evlist)
723{
724    struct perf_evsel *evsel = list_entry(evlist->entries.next,
725                          struct perf_evsel, node);
726
727    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
728    TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type);
729    TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
730    TEST_ASSERT_VAL("wrong bp_type", (HW_BREAKPOINT_R | HW_BREAKPOINT_W) ==
731                     evsel->attr.bp_type);
732    TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_4 ==
733                    evsel->attr.bp_len);
734    return 0;
735}
736
737static int test__checkevent_breakpoint_x(struct perf_evlist *evlist)
738{
739    struct perf_evsel *evsel = list_entry(evlist->entries.next,
740                          struct perf_evsel, node);
741
742    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
743    TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type);
744    TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
745    TEST_ASSERT_VAL("wrong bp_type",
746            HW_BREAKPOINT_X == evsel->attr.bp_type);
747    TEST_ASSERT_VAL("wrong bp_len", sizeof(long) == evsel->attr.bp_len);
748    return 0;
749}
750
751static int test__checkevent_breakpoint_r(struct perf_evlist *evlist)
752{
753    struct perf_evsel *evsel = list_entry(evlist->entries.next,
754                          struct perf_evsel, node);
755
756    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
757    TEST_ASSERT_VAL("wrong type",
758            PERF_TYPE_BREAKPOINT == evsel->attr.type);
759    TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
760    TEST_ASSERT_VAL("wrong bp_type",
761            HW_BREAKPOINT_R == evsel->attr.bp_type);
762    TEST_ASSERT_VAL("wrong bp_len",
763            HW_BREAKPOINT_LEN_4 == evsel->attr.bp_len);
764    return 0;
765}
766
767static int test__checkevent_breakpoint_w(struct perf_evlist *evlist)
768{
769    struct perf_evsel *evsel = list_entry(evlist->entries.next,
770                          struct perf_evsel, node);
771
772    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
773    TEST_ASSERT_VAL("wrong type",
774            PERF_TYPE_BREAKPOINT == evsel->attr.type);
775    TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
776    TEST_ASSERT_VAL("wrong bp_type",
777            HW_BREAKPOINT_W == evsel->attr.bp_type);
778    TEST_ASSERT_VAL("wrong bp_len",
779            HW_BREAKPOINT_LEN_4 == evsel->attr.bp_len);
780    return 0;
781}
782
783static int test__checkevent_tracepoint_modifier(struct perf_evlist *evlist)
784{
785    struct perf_evsel *evsel = list_entry(evlist->entries.next,
786                          struct perf_evsel, node);
787
788    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
789    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
790    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
791    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
792
793    return test__checkevent_tracepoint(evlist);
794}
795
796static int
797test__checkevent_tracepoint_multi_modifier(struct perf_evlist *evlist)
798{
799    struct perf_evsel *evsel;
800
801    TEST_ASSERT_VAL("wrong number of entries", evlist->nr_entries > 1);
802
803    list_for_each_entry(evsel, &evlist->entries, node) {
804        TEST_ASSERT_VAL("wrong exclude_user",
805                !evsel->attr.exclude_user);
806        TEST_ASSERT_VAL("wrong exclude_kernel",
807                evsel->attr.exclude_kernel);
808        TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
809        TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
810    }
811
812    return test__checkevent_tracepoint_multi(evlist);
813}
814
815static int test__checkevent_raw_modifier(struct perf_evlist *evlist)
816{
817    struct perf_evsel *evsel = list_entry(evlist->entries.next,
818                          struct perf_evsel, node);
819
820    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
821    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
822    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
823    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
824
825    return test__checkevent_raw(evlist);
826}
827
828static int test__checkevent_numeric_modifier(struct perf_evlist *evlist)
829{
830    struct perf_evsel *evsel = list_entry(evlist->entries.next,
831                          struct perf_evsel, node);
832
833    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
834    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
835    TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv);
836    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
837
838    return test__checkevent_numeric(evlist);
839}
840
841static int test__checkevent_symbolic_name_modifier(struct perf_evlist *evlist)
842{
843    struct perf_evsel *evsel = list_entry(evlist->entries.next,
844                          struct perf_evsel, node);
845
846    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
847    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
848    TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv);
849    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
850
851    return test__checkevent_symbolic_name(evlist);
852}
853
854static int test__checkevent_exclude_host_modifier(struct perf_evlist *evlist)
855{
856    struct perf_evsel *evsel = list_entry(evlist->entries.next,
857                          struct perf_evsel, node);
858
859    TEST_ASSERT_VAL("wrong exclude guest", !evsel->attr.exclude_guest);
860    TEST_ASSERT_VAL("wrong exclude host", evsel->attr.exclude_host);
861
862    return test__checkevent_symbolic_name(evlist);
863}
864
865static int test__checkevent_exclude_guest_modifier(struct perf_evlist *evlist)
866{
867    struct perf_evsel *evsel = list_entry(evlist->entries.next,
868                          struct perf_evsel, node);
869
870    TEST_ASSERT_VAL("wrong exclude guest", evsel->attr.exclude_guest);
871    TEST_ASSERT_VAL("wrong exclude host", !evsel->attr.exclude_host);
872
873    return test__checkevent_symbolic_name(evlist);
874}
875
876static int test__checkevent_symbolic_alias_modifier(struct perf_evlist *evlist)
877{
878    struct perf_evsel *evsel = list_entry(evlist->entries.next,
879                          struct perf_evsel, node);
880
881    TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
882    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
883    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
884    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
885
886    return test__checkevent_symbolic_alias(evlist);
887}
888
889static int test__checkevent_genhw_modifier(struct perf_evlist *evlist)
890{
891    struct perf_evsel *evsel = list_entry(evlist->entries.next,
892                          struct perf_evsel, node);
893
894    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
895    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
896    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
897    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
898
899    return test__checkevent_genhw(evlist);
900}
901
902static int test__checkevent_breakpoint_modifier(struct perf_evlist *evlist)
903{
904    struct perf_evsel *evsel = list_entry(evlist->entries.next,
905                          struct perf_evsel, node);
906
907    TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
908    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
909    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
910    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
911
912    return test__checkevent_breakpoint(evlist);
913}
914
915static int test__checkevent_breakpoint_x_modifier(struct perf_evlist *evlist)
916{
917    struct perf_evsel *evsel = list_entry(evlist->entries.next,
918                          struct perf_evsel, node);
919
920    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
921    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
922    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
923    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
924
925    return test__checkevent_breakpoint_x(evlist);
926}
927
928static int test__checkevent_breakpoint_r_modifier(struct perf_evlist *evlist)
929{
930    struct perf_evsel *evsel = list_entry(evlist->entries.next,
931                          struct perf_evsel, node);
932
933    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
934    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
935    TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv);
936    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
937
938    return test__checkevent_breakpoint_r(evlist);
939}
940
941static int test__checkevent_breakpoint_w_modifier(struct perf_evlist *evlist)
942{
943    struct perf_evsel *evsel = list_entry(evlist->entries.next,
944                          struct perf_evsel, node);
945
946    TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
947    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
948    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
949    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
950
951    return test__checkevent_breakpoint_w(evlist);
952}
953
954static int test__checkevent_pmu(struct perf_evlist *evlist)
955{
956
957    struct perf_evsel *evsel = list_entry(evlist->entries.next,
958                          struct perf_evsel, node);
959
960    TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
961    TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type);
962    TEST_ASSERT_VAL("wrong config", 10 == evsel->attr.config);
963    TEST_ASSERT_VAL("wrong config1", 1 == evsel->attr.config1);
964    TEST_ASSERT_VAL("wrong config2", 3 == evsel->attr.config2);
965    TEST_ASSERT_VAL("wrong period", 1000 == evsel->attr.sample_period);
966
967    return 0;
968}
969
970static int test__checkevent_list(struct perf_evlist *evlist)
971{
972    struct perf_evsel *evsel;
973
974    TEST_ASSERT_VAL("wrong number of entries", 3 == evlist->nr_entries);
975
976    /* r1 */
977    evsel = list_entry(evlist->entries.next, struct perf_evsel, node);
978    TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type);
979    TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config);
980    TEST_ASSERT_VAL("wrong config1", 0 == evsel->attr.config1);
981    TEST_ASSERT_VAL("wrong config2", 0 == evsel->attr.config2);
982    TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
983    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
984    TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv);
985    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
986
987    /* syscalls:sys_enter_open:k */
988    evsel = list_entry(evsel->node.next, struct perf_evsel, node);
989    TEST_ASSERT_VAL("wrong type", PERF_TYPE_TRACEPOINT == evsel->attr.type);
990    TEST_ASSERT_VAL("wrong sample_type",
991        (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU) ==
992        evsel->attr.sample_type);
993    TEST_ASSERT_VAL("wrong sample_period", 1 == evsel->attr.sample_period);
994    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
995    TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
996    TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
997    TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
998
999    /* 1:1:hp */
1000    evsel = list_entry(evsel->node.next, struct perf_evsel, node);
1001    TEST_ASSERT_VAL("wrong type", 1 == evsel->attr.type);
1002    TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config);
1003    TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
1004    TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
1005    TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv);
1006    TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip);
1007
1008    return 0;
1009}
1010
1011static struct test__event_st {
1012    const char *name;
1013    __u32 type;
1014    int (*check)(struct perf_evlist *evlist);
1015} test__events[] = {
1016    {
1017        .name = "syscalls:sys_enter_open",
1018        .check = test__checkevent_tracepoint,
1019    },
1020    {
1021        .name = "syscalls:*",
1022        .check = test__checkevent_tracepoint_multi,
1023    },
1024    {
1025        .name = "r1a",
1026        .check = test__checkevent_raw,
1027    },
1028    {
1029        .name = "1:1",
1030        .check = test__checkevent_numeric,
1031    },
1032    {
1033        .name = "instructions",
1034        .check = test__checkevent_symbolic_name,
1035    },
1036    {
1037        .name = "cycles/period=100000,config2/",
1038        .check = test__checkevent_symbolic_name_config,
1039    },
1040    {
1041        .name = "faults",
1042        .check = test__checkevent_symbolic_alias,
1043    },
1044    {
1045        .name = "L1-dcache-load-miss",
1046        .check = test__checkevent_genhw,
1047    },
1048    {
1049        .name = "mem:0",
1050        .check = test__checkevent_breakpoint,
1051    },
1052    {
1053        .name = "mem:0:x",
1054        .check = test__checkevent_breakpoint_x,
1055    },
1056    {
1057        .name = "mem:0:r",
1058        .check = test__checkevent_breakpoint_r,
1059    },
1060    {
1061        .name = "mem:0:w",
1062        .check = test__checkevent_breakpoint_w,
1063    },
1064    {
1065        .name = "syscalls:sys_enter_open:k",
1066        .check = test__checkevent_tracepoint_modifier,
1067    },
1068    {
1069        .name = "syscalls:*:u",
1070        .check = test__checkevent_tracepoint_multi_modifier,
1071    },
1072    {
1073        .name = "r1a:kp",
1074        .check = test__checkevent_raw_modifier,
1075    },
1076    {
1077        .name = "1:1:hp",
1078        .check = test__checkevent_numeric_modifier,
1079    },
1080    {
1081        .name = "instructions:h",
1082        .check = test__checkevent_symbolic_name_modifier,
1083    },
1084    {
1085        .name = "faults:u",
1086        .check = test__checkevent_symbolic_alias_modifier,
1087    },
1088    {
1089        .name = "L1-dcache-load-miss:kp",
1090        .check = test__checkevent_genhw_modifier,
1091    },
1092    {
1093        .name = "mem:0:u",
1094        .check = test__checkevent_breakpoint_modifier,
1095    },
1096    {
1097        .name = "mem:0:x:k",
1098        .check = test__checkevent_breakpoint_x_modifier,
1099    },
1100    {
1101        .name = "mem:0:r:hp",
1102        .check = test__checkevent_breakpoint_r_modifier,
1103    },
1104    {
1105        .name = "mem:0:w:up",
1106        .check = test__checkevent_breakpoint_w_modifier,
1107    },
1108    {
1109        .name = "cpu/config=10,config1,config2=3,period=1000/u",
1110        .check = test__checkevent_pmu,
1111    },
1112    {
1113        .name = "r1,syscalls:sys_enter_open:k,1:1:hp",
1114        .check = test__checkevent_list,
1115    },
1116    {
1117        .name = "instructions:G",
1118        .check = test__checkevent_exclude_host_modifier,
1119    },
1120    {
1121        .name = "instructions:H",
1122        .check = test__checkevent_exclude_guest_modifier,
1123    },
1124};
1125
1126#define TEST__EVENTS_CNT (sizeof(test__events) / sizeof(struct test__event_st))
1127
1128static int test__parse_events(void)
1129{
1130    struct perf_evlist *evlist;
1131    u_int i;
1132    int ret = 0;
1133
1134    for (i = 0; i < TEST__EVENTS_CNT; i++) {
1135        struct test__event_st *e = &test__events[i];
1136
1137        evlist = perf_evlist__new(NULL, NULL);
1138        if (evlist == NULL)
1139            break;
1140
1141        ret = parse_events(evlist, e->name, 0);
1142        if (ret) {
1143            pr_debug("failed to parse event '%s', err %d\n",
1144                 e->name, ret);
1145            break;
1146        }
1147
1148        ret = e->check(evlist);
1149        perf_evlist__delete(evlist);
1150        if (ret)
1151            break;
1152    }
1153
1154    return ret;
1155}
1156
1157static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t **maskp,
1158                     size_t *sizep)
1159{
1160    cpu_set_t *mask;
1161    size_t size;
1162    int i, cpu = -1, nrcpus = 1024;
1163realloc:
1164    mask = CPU_ALLOC(nrcpus);
1165    size = CPU_ALLOC_SIZE(nrcpus);
1166    CPU_ZERO_S(size, mask);
1167
1168    if (sched_getaffinity(pid, size, mask) == -1) {
1169        CPU_FREE(mask);
1170        if (errno == EINVAL && nrcpus < (1024 << 8)) {
1171            nrcpus = nrcpus << 2;
1172            goto realloc;
1173        }
1174        perror("sched_getaffinity");
1175            return -1;
1176    }
1177
1178    for (i = 0; i < nrcpus; i++) {
1179        if (CPU_ISSET_S(i, size, mask)) {
1180            if (cpu == -1) {
1181                cpu = i;
1182                *maskp = mask;
1183                *sizep = size;
1184            } else
1185                CPU_CLR_S(i, size, mask);
1186        }
1187    }
1188
1189    if (cpu == -1)
1190        CPU_FREE(mask);
1191
1192    return cpu;
1193}
1194
1195static int test__PERF_RECORD(void)
1196{
1197    struct perf_record_opts opts = {
1198        .no_delay = true,
1199        .freq = 10,
1200        .mmap_pages = 256,
1201    };
1202    cpu_set_t *cpu_mask = NULL;
1203    size_t cpu_mask_size = 0;
1204    struct perf_evlist *evlist = perf_evlist__new(NULL, NULL);
1205    struct perf_evsel *evsel;
1206    struct perf_sample sample;
1207    const char *cmd = "sleep";
1208    const char *argv[] = { cmd, "1", NULL, };
1209    char *bname;
1210    u64 sample_type, prev_time = 0;
1211    bool found_cmd_mmap = false,
1212         found_libc_mmap = false,
1213         found_vdso_mmap = false,
1214         found_ld_mmap = false;
1215    int err = -1, errs = 0, i, wakeups = 0, sample_size;
1216    u32 cpu;
1217    int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, };
1218
1219    if (evlist == NULL || argv == NULL) {
1220        pr_debug("Not enough memory to create evlist\n");
1221        goto out;
1222    }
1223
1224    /*
1225     * We need at least one evsel in the evlist, use the default
1226     * one: "cycles".
1227     */
1228    err = perf_evlist__add_default(evlist);
1229    if (err < 0) {
1230        pr_debug("Not enough memory to create evsel\n");
1231        goto out_delete_evlist;
1232    }
1233
1234    /*
1235     * Create maps of threads and cpus to monitor. In this case
1236     * we start with all threads and cpus (-1, -1) but then in
1237     * perf_evlist__prepare_workload we'll fill in the only thread
1238     * we're monitoring, the one forked there.
1239     */
1240    err = perf_evlist__create_maps(evlist, opts.target_pid,
1241                       opts.target_tid, UINT_MAX, opts.cpu_list);
1242    if (err < 0) {
1243        pr_debug("Not enough memory to create thread/cpu maps\n");
1244        goto out_delete_evlist;
1245    }
1246
1247    /*
1248     * Prepare the workload in argv[] to run, it'll fork it, and then wait
1249     * for perf_evlist__start_workload() to exec it. This is done this way
1250     * so that we have time to open the evlist (calling sys_perf_event_open
1251     * on all the fds) and then mmap them.
1252     */
1253    err = perf_evlist__prepare_workload(evlist, &opts, argv);
1254    if (err < 0) {
1255        pr_debug("Couldn't run the workload!\n");
1256        goto out_delete_evlist;
1257    }
1258
1259    /*
1260     * Config the evsels, setting attr->comm on the first one, etc.
1261     */
1262    evsel = list_entry(evlist->entries.next, struct perf_evsel, node);
1263    evsel->attr.sample_type |= PERF_SAMPLE_CPU;
1264    evsel->attr.sample_type |= PERF_SAMPLE_TID;
1265    evsel->attr.sample_type |= PERF_SAMPLE_TIME;
1266    perf_evlist__config_attrs(evlist, &opts);
1267
1268    err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask,
1269                        &cpu_mask_size);
1270    if (err < 0) {
1271        pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
1272        goto out_delete_evlist;
1273    }
1274
1275    cpu = err;
1276
1277    /*
1278     * So that we can check perf_sample.cpu on all the samples.
1279     */
1280    if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, cpu_mask) < 0) {
1281        pr_debug("sched_setaffinity: %s\n", strerror(errno));
1282        goto out_free_cpu_mask;
1283    }
1284
1285    /*
1286     * Call sys_perf_event_open on all the fds on all the evsels,
1287     * grouping them if asked to.
1288     */
1289    err = perf_evlist__open(evlist, opts.group);
1290    if (err < 0) {
1291        pr_debug("perf_evlist__open: %s\n", strerror(errno));
1292        goto out_delete_evlist;
1293    }
1294
1295    /*
1296     * mmap the first fd on a given CPU and ask for events for the other
1297     * fds in the same CPU to be injected in the same mmap ring buffer
1298     * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
1299     */
1300    err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
1301    if (err < 0) {
1302        pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
1303        goto out_delete_evlist;
1304    }
1305
1306    /*
1307     * We'll need these two to parse the PERF_SAMPLE_* fields in each
1308     * event.
1309     */
1310    sample_type = perf_evlist__sample_type(evlist);
1311    sample_size = __perf_evsel__sample_size(sample_type);
1312
1313    /*
1314     * Now that all is properly set up, enable the events, they will
1315     * count just on workload.pid, which will start...
1316     */
1317    perf_evlist__enable(evlist);
1318
1319    /*
1320     * Now!
1321     */
1322    perf_evlist__start_workload(evlist);
1323
1324    while (1) {
1325        int before = total_events;
1326
1327        for (i = 0; i < evlist->nr_mmaps; i++) {
1328            union perf_event *event;
1329
1330            while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
1331                const u32 type = event->header.type;
1332                const char *name = perf_event__name(type);
1333
1334                ++total_events;
1335                if (type < PERF_RECORD_MAX)
1336                    nr_events[type]++;
1337
1338                err = perf_event__parse_sample(event, sample_type,
1339                                   sample_size, true,
1340                                   &sample, false);
1341                if (err < 0) {
1342                    if (verbose)
1343                        perf_event__fprintf(event, stderr);
1344                    pr_debug("Couldn't parse sample\n");
1345                    goto out_err;
1346                }
1347
1348                if (verbose) {
1349                    pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
1350                    perf_event__fprintf(event, stderr);
1351                }
1352
1353                if (prev_time > sample.time) {
1354                    pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
1355                         name, prev_time, sample.time);
1356                    ++errs;
1357                }
1358
1359                prev_time = sample.time;
1360
1361                if (sample.cpu != cpu) {
1362                    pr_debug("%s with unexpected cpu, expected %d, got %d\n",
1363                         name, cpu, sample.cpu);
1364                    ++errs;
1365                }
1366
1367                if ((pid_t)sample.pid != evlist->workload.pid) {
1368                    pr_debug("%s with unexpected pid, expected %d, got %d\n",
1369                         name, evlist->workload.pid, sample.pid);
1370                    ++errs;
1371                }
1372
1373                if ((pid_t)sample.tid != evlist->workload.pid) {
1374                    pr_debug("%s with unexpected tid, expected %d, got %d\n",
1375                         name, evlist->workload.pid, sample.tid);
1376                    ++errs;
1377                }
1378
1379                if ((type == PERF_RECORD_COMM ||
1380                     type == PERF_RECORD_MMAP ||
1381                     type == PERF_RECORD_FORK ||
1382                     type == PERF_RECORD_EXIT) &&
1383                     (pid_t)event->comm.pid != evlist->workload.pid) {
1384                    pr_debug("%s with unexpected pid/tid\n", name);
1385                    ++errs;
1386                }
1387
1388                if ((type == PERF_RECORD_COMM ||
1389                     type == PERF_RECORD_MMAP) &&
1390                     event->comm.pid != event->comm.tid) {
1391                    pr_debug("%s with different pid/tid!\n", name);
1392                    ++errs;
1393                }
1394
1395                switch (type) {
1396                case PERF_RECORD_COMM:
1397                    if (strcmp(event->comm.comm, cmd)) {
1398                        pr_debug("%s with unexpected comm!\n", name);
1399                        ++errs;
1400                    }
1401                    break;
1402                case PERF_RECORD_EXIT:
1403                    goto found_exit;
1404                case PERF_RECORD_MMAP:
1405                    bname = strrchr(event->mmap.filename, '/');
1406                    if (bname != NULL) {
1407                        if (!found_cmd_mmap)
1408                            found_cmd_mmap = !strcmp(bname + 1, cmd);
1409                        if (!found_libc_mmap)
1410                            found_libc_mmap = !strncmp(bname + 1, "libc", 4);
1411                        if (!found_ld_mmap)
1412                            found_ld_mmap = !strncmp(bname + 1, "ld", 2);
1413                    } else if (!found_vdso_mmap)
1414                        found_vdso_mmap = !strcmp(event->mmap.filename, "[vdso]");
1415                    break;
1416
1417                case PERF_RECORD_SAMPLE:
1418                    /* Just ignore samples for now */
1419                    break;
1420                default:
1421                    pr_debug("Unexpected perf_event->header.type %d!\n",
1422                         type);
1423                    ++errs;
1424                }
1425            }
1426        }
1427
1428        /*
1429         * We don't use poll here because at least at 3.1 times the
1430         * PERF_RECORD_{!SAMPLE} events don't honour
1431         * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
1432         */
1433        if (total_events == before && false)
1434            poll(evlist->pollfd, evlist->nr_fds, -1);
1435
1436        sleep(1);
1437        if (++wakeups > 5) {
1438            pr_debug("No PERF_RECORD_EXIT event!\n");
1439            break;
1440        }
1441    }
1442
1443found_exit:
1444    if (nr_events[PERF_RECORD_COMM] > 1) {
1445        pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
1446        ++errs;
1447    }
1448
1449    if (nr_events[PERF_RECORD_COMM] == 0) {
1450        pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
1451        ++errs;
1452    }
1453
1454    if (!found_cmd_mmap) {
1455        pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
1456        ++errs;
1457    }
1458
1459    if (!found_libc_mmap) {
1460        pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
1461        ++errs;
1462    }
1463
1464    if (!found_ld_mmap) {
1465        pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
1466        ++errs;
1467    }
1468
1469    if (!found_vdso_mmap) {
1470        pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
1471        ++errs;
1472    }
1473out_err:
1474    perf_evlist__munmap(evlist);
1475out_free_cpu_mask:
1476    CPU_FREE(cpu_mask);
1477out_delete_evlist:
1478    perf_evlist__delete(evlist);
1479out:
1480    return (err < 0 || errs > 0) ? -1 : 0;
1481}
1482
1483
1484#if defined(__x86_64__) || defined(__i386__)
1485
1486#define barrier() asm volatile("" ::: "memory")
1487
1488static u64 rdpmc(unsigned int counter)
1489{
1490    unsigned int low, high;
1491
1492    asm volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (counter));
1493
1494    return low | ((u64)high) << 32;
1495}
1496
1497static u64 rdtsc(void)
1498{
1499    unsigned int low, high;
1500
1501    asm volatile("rdtsc" : "=a" (low), "=d" (high));
1502
1503    return low | ((u64)high) << 32;
1504}
1505
1506static u64 mmap_read_self(void *addr)
1507{
1508    struct perf_event_mmap_page *pc = addr;
1509    u32 seq, idx, time_mult = 0, time_shift = 0;
1510    u64 count, cyc = 0, time_offset = 0, enabled, running, delta;
1511
1512    do {
1513        seq = pc->lock;
1514        barrier();
1515
1516        enabled = pc->time_enabled;
1517        running = pc->time_running;
1518
1519        if (enabled != running) {
1520            cyc = rdtsc();
1521            time_mult = pc->time_mult;
1522            time_shift = pc->time_shift;
1523            time_offset = pc->time_offset;
1524        }
1525
1526        idx = pc->index;
1527        count = pc->offset;
1528        if (idx)
1529            count += rdpmc(idx - 1);
1530
1531        barrier();
1532    } while (pc->lock != seq);
1533
1534    if (enabled != running) {
1535        u64 quot, rem;
1536
1537        quot = (cyc >> time_shift);
1538        rem = cyc & ((1 << time_shift) - 1);
1539        delta = time_offset + quot * time_mult +
1540            ((rem * time_mult) >> time_shift);
1541
1542        enabled += delta;
1543        if (idx)
1544            running += delta;
1545
1546        quot = count / running;
1547        rem = count % running;
1548        count = quot * enabled + (rem * enabled) / running;
1549    }
1550
1551    return count;
1552}
1553
1554/*
1555 * If the RDPMC instruction faults then signal this back to the test parent task:
1556 */
1557static void segfault_handler(int sig __used, siginfo_t *info __used, void *uc __used)
1558{
1559    exit(-1);
1560}
1561
1562static int __test__rdpmc(void)
1563{
1564    long page_size = sysconf(_SC_PAGE_SIZE);
1565    volatile int tmp = 0;
1566    u64 i, loops = 1000;
1567    int n;
1568    int fd;
1569    void *addr;
1570    struct perf_event_attr attr = {
1571        .type = PERF_TYPE_HARDWARE,
1572        .config = PERF_COUNT_HW_INSTRUCTIONS,
1573        .exclude_kernel = 1,
1574    };
1575    u64 delta_sum = 0;
1576        struct sigaction sa;
1577
1578    sigfillset(&sa.sa_mask);
1579    sa.sa_sigaction = segfault_handler;
1580    sigaction(SIGSEGV, &sa, NULL);
1581
1582    fprintf(stderr, "\n\n");
1583
1584    fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
1585    if (fd < 0) {
1586        die("Error: sys_perf_event_open() syscall returned "
1587            "with %d (%s)\n", fd, strerror(errno));
1588    }
1589
1590    addr = mmap(NULL, page_size, PROT_READ, MAP_SHARED, fd, 0);
1591    if (addr == (void *)(-1)) {
1592        die("Error: mmap() syscall returned "
1593            "with (%s)\n", strerror(errno));
1594    }
1595
1596    for (n = 0; n < 6; n++) {
1597        u64 stamp, now, delta;
1598
1599        stamp = mmap_read_self(addr);
1600
1601        for (i = 0; i < loops; i++)
1602            tmp++;
1603
1604        now = mmap_read_self(addr);
1605        loops *= 10;
1606
1607        delta = now - stamp;
1608        fprintf(stderr, "%14d: %14Lu\n", n, (long long)delta);
1609
1610        delta_sum += delta;
1611    }
1612
1613    munmap(addr, page_size);
1614    close(fd);
1615
1616    fprintf(stderr, " ");
1617
1618    if (!delta_sum)
1619        return -1;
1620
1621    return 0;
1622}
1623
1624static int test__rdpmc(void)
1625{
1626    int status = 0;
1627    int wret = 0;
1628    int ret;
1629    int pid;
1630
1631    pid = fork();
1632    if (pid < 0)
1633        return -1;
1634
1635    if (!pid) {
1636        ret = __test__rdpmc();
1637
1638        exit(ret);
1639    }
1640
1641    wret = waitpid(pid, &status, 0);
1642    if (wret < 0 || status)
1643        return -1;
1644
1645    return 0;
1646}
1647
1648#endif
1649
1650static int test__perf_pmu(void)
1651{
1652    return perf_pmu__test();
1653}
1654
1655static struct test {
1656    const char *desc;
1657    int (*func)(void);
1658} tests[] = {
1659    {
1660        .desc = "vmlinux symtab matches kallsyms",
1661        .func = test__vmlinux_matches_kallsyms,
1662    },
1663    {
1664        .desc = "detect open syscall event",
1665        .func = test__open_syscall_event,
1666    },
1667    {
1668        .desc = "detect open syscall event on all cpus",
1669        .func = test__open_syscall_event_on_all_cpus,
1670    },
1671    {
1672        .desc = "read samples using the mmap interface",
1673        .func = test__basic_mmap,
1674    },
1675    {
1676        .desc = "parse events tests",
1677        .func = test__parse_events,
1678    },
1679#if defined(__x86_64__) || defined(__i386__)
1680    {
1681        .desc = "x86 rdpmc test",
1682        .func = test__rdpmc,
1683    },
1684#endif
1685    {
1686        .desc = "Validate PERF_RECORD_* events & perf_sample fields",
1687        .func = test__PERF_RECORD,
1688    },
1689    {
1690        .desc = "Test perf pmu format parsing",
1691        .func = test__perf_pmu,
1692    },
1693    {
1694        .func = NULL,
1695    },
1696};
1697
1698static bool perf_test__matches(int curr, int argc, const char *argv[])
1699{
1700    int i;
1701
1702    if (argc == 0)
1703        return true;
1704
1705    for (i = 0; i < argc; ++i) {
1706        char *end;
1707        long nr = strtoul(argv[i], &end, 10);
1708
1709        if (*end == '\0') {
1710            if (nr == curr + 1)
1711                return true;
1712            continue;
1713        }
1714
1715        if (strstr(tests[curr].desc, argv[i]))
1716            return true;
1717    }
1718
1719    return false;
1720}
1721
1722static int __cmd_test(int argc, const char *argv[])
1723{
1724    int i = 0;
1725
1726    while (tests[i].func) {
1727        int curr = i++, err;
1728
1729        if (!perf_test__matches(curr, argc, argv))
1730            continue;
1731
1732        pr_info("%2d: %s:", i, tests[curr].desc);
1733        pr_debug("\n--- start ---\n");
1734        err = tests[curr].func();
1735        pr_debug("---- end ----\n%s:", tests[curr].desc);
1736        pr_info(" %s\n", err ? "FAILED!\n" : "Ok");
1737    }
1738
1739    return 0;
1740}
1741
1742static int perf_test__list(int argc, const char **argv)
1743{
1744    int i = 0;
1745
1746    while (tests[i].func) {
1747        int curr = i++;
1748
1749        if (argc > 1 && !strstr(tests[curr].desc, argv[1]))
1750            continue;
1751
1752        pr_info("%2d: %s\n", i, tests[curr].desc);
1753    }
1754
1755    return 0;
1756}
1757
1758int cmd_test(int argc, const char **argv, const char *prefix __used)
1759{
1760    const char * const test_usage[] = {
1761    "perf test [<options>] [{list <test-name-fragment>|[<test-name-fragments>|<test-numbers>]}]",
1762    NULL,
1763    };
1764    const struct option test_options[] = {
1765    OPT_INCR('v', "verbose", &verbose,
1766            "be more verbose (show symbol address, etc)"),
1767    OPT_END()
1768    };
1769
1770    argc = parse_options(argc, argv, test_options, test_usage, 0);
1771    if (argc >= 1 && !strcmp(argv[0], "list"))
1772        return perf_test__list(argc, argv);
1773
1774    symbol_conf.priv_size = sizeof(int);
1775    symbol_conf.sort_by_name = true;
1776    symbol_conf.try_vmlinux_path = true;
1777
1778    if (symbol__init() < 0)
1779        return -1;
1780
1781    return __cmd_test(argc, argv);
1782}
1783

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive