From 361c99a661a78ed22264649440e87fe4fe8da1f2 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 11 Jan 2011 20:56:53 -0200 Subject: perf evsel: Introduce perf_evlist Killing two more perf wide global variables: nr_counters and evsel_list as a list_head. There are more operations that will need more fields in perf_evlist, like the pollfd for polling all the fds in a list of evsel instances. Use option->value to pass the evsel_list to parse_{events,filters}. LKML-Reference: Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 53 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 53 insertions(+) create mode 100644 tools/perf/util/evlist.c (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c new file mode 100644 index 000000000000..7b4faec23737 --- /dev/null +++ b/tools/perf/util/evlist.c @@ -0,0 +1,53 @@ +#include "evlist.h" +#include "evsel.h" +#include "util.h" + +struct perf_evlist *perf_evlist__new(void) +{ + struct perf_evlist *evlist = zalloc(sizeof(*evlist)); + + if (evlist != NULL) { + INIT_LIST_HEAD(&evlist->entries); + } + + return evlist; +} + +static void perf_evlist__purge(struct perf_evlist *evlist) +{ + struct perf_evsel *pos, *n; + + list_for_each_entry_safe(pos, n, &evlist->entries, node) { + list_del_init(&pos->node); + perf_evsel__delete(pos); + } + + evlist->nr_entries = 0; +} + +void perf_evlist__delete(struct perf_evlist *evlist) +{ + perf_evlist__purge(evlist); + free(evlist); +} + +void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry) +{ + list_add_tail(&entry->node, &evlist->entries); + ++evlist->nr_entries; +} + +int perf_evlist__add_default(struct perf_evlist *evlist) +{ + struct perf_event_attr attr = { + .type = PERF_TYPE_HARDWARE, + .config = PERF_COUNT_HW_CPU_CYCLES, + }; + struct perf_evsel *evsel = perf_evsel__new(&attr, 0); + + if (evsel == NULL) + return -ENOMEM; + + perf_evlist__add(evlist, evsel); + return 0; +} -- cgit v1.2.2 From 5c581041cf97aa7980b442de81ddea8273d6dcde Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 11 Jan 2011 22:30:02 -0200 Subject: perf evlist: Adopt the pollfd array Allocating just the space needed for nr_cpus * nr_threads * nr_evsels, not the MAX_NR_CPUS and counters. LKML-Reference: Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 7b4faec23737..2abf949259d0 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -1,3 +1,4 @@ +#include #include "evlist.h" #include "evsel.h" #include "util.h" @@ -28,6 +29,7 @@ static void perf_evlist__purge(struct perf_evlist *evlist) void perf_evlist__delete(struct perf_evlist *evlist) { perf_evlist__purge(evlist); + free(evlist->pollfd); free(evlist); } @@ -51,3 +53,10 @@ int perf_evlist__add_default(struct perf_evlist *evlist) perf_evlist__add(evlist, evsel); return 0; } + +int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads) +{ + int nfds = ncpus * nthreads * evlist->nr_entries; + evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); + return evlist->pollfd != NULL ? 0 : -ENOMEM; +} -- cgit v1.2.2 From 70082dd92c4b288bd723a77897e2b555f0e63113 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 17:03:24 -0200 Subject: perf evsel: Introduce mmap support Out of the code in 'perf top'. Record is next in line. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 2abf949259d0..6d4129214ee8 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -60,3 +60,11 @@ int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthread evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); return evlist->pollfd != NULL ? 0 : -ENOMEM; } + +void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) +{ + fcntl(fd, F_SETFL, O_NONBLOCK); + evlist->pollfd[evlist->nr_fds].fd = fd; + evlist->pollfd[evlist->nr_fds].events = POLLIN; + evlist->nr_fds++; +} -- cgit v1.2.2 From 70db7533caef02350ec8d6852e589491bca3a951 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 22:39:13 -0200 Subject: perf evlist: Move the mmap array from perf_evsel Adopting the new model used in 'perf record', where we don't have a map per thread per cpu, instead we have an mmap per cpu, established on the first fd for that cpu and ask the kernel using the PERF_EVENT_IOC_SET_OUTPUT ioctl to send events for the other fds on that cpu for the one with the mmap. The methods moved from perf_evsel to perf_evlist, but for easing review they were modified in place, in evsel.c, the next patch will move the migrated methods to evlist.c. With this 'perf top' now uses the same mmap model used by 'perf record' and the next patches will make 'perf record' use these new routines, establishing a common codebase for both tools. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 6d4129214ee8..deb82a4fc312 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -3,11 +3,18 @@ #include "evsel.h" #include "util.h" +#include +#include + struct perf_evlist *perf_evlist__new(void) { struct perf_evlist *evlist = zalloc(sizeof(*evlist)); if (evlist != NULL) { + int i; + + for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) + INIT_HLIST_HEAD(&evlist->heads[i]); INIT_LIST_HEAD(&evlist->entries); } @@ -29,6 +36,7 @@ static void perf_evlist__purge(struct perf_evlist *evlist) void perf_evlist__delete(struct perf_evlist *evlist) { perf_evlist__purge(evlist); + free(evlist->mmap); free(evlist->pollfd); free(evlist); } @@ -68,3 +76,22 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) evlist->pollfd[evlist->nr_fds].events = POLLIN; evlist->nr_fds++; } + +struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) +{ + struct hlist_head *head; + struct hlist_node *pos; + struct perf_sample_id *sid; + int hash; + + if (evlist->nr_entries == 1) + return list_entry(evlist->entries.next, struct perf_evsel, node); + + hash = hash_64(id, PERF_EVLIST__HLIST_BITS); + head = &evlist->heads[hash]; + + hlist_for_each_entry(sid, pos, head, node) + if (sid->id == id) + return sid->evsel; + return NULL; +} -- cgit v1.2.2 From 04391debc3e1195222a4dbb162ace6542dd89c1c Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sat, 15 Jan 2011 10:40:59 -0200 Subject: perf evlist: Steal mmap reading routine from 'perf top' Will be used in the upcoming 'perf test' entry for the evlist mmap routines. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 62 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index deb82a4fc312..4b3b84cd71a1 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -95,3 +95,65 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) return sid->evsel; return NULL; } + +event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) +{ + /* XXX Move this to perf.c, making it generally available */ + unsigned int page_size = sysconf(_SC_PAGE_SIZE); + struct perf_mmap *md = &evlist->mmap[cpu]; + unsigned int head = perf_mmap__read_head(md); + unsigned int old = md->prev; + unsigned char *data = md->base + page_size; + event_t *event = NULL; + int diff; + + /* + * If we're further behind than half the buffer, there's a chance + * the writer will bite our tail and mess up the samples under us. + * + * If we somehow ended up ahead of the head, we got messed up. + * + * In either case, truncate and restart at head. + */ + diff = head - old; + if (diff > md->mask / 2 || diff < 0) { + fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); + + /* + * head points to a known good entry, start there. + */ + old = head; + } + + if (old != head) { + size_t size; + + event = (event_t *)&data[old & md->mask]; + size = event->header.size; + + /* + * Event straddles the mmap boundary -- header should always + * be inside due to u64 alignment of output. + */ + if ((old & md->mask) + size != ((old + size) & md->mask)) { + unsigned int offset = old; + unsigned int len = min(sizeof(*event), size), cpy; + void *dst = &evlist->event_copy; + + do { + cpy = min(md->mask + 1 - (offset & md->mask), len); + memcpy(dst, &data[offset & md->mask], cpy); + offset += cpy; + dst += cpy; + len -= cpy; + } while (len); + + event = &evlist->event_copy; + } + + old += size; + } + + md->prev = old; + return event; +} -- cgit v1.2.2 From ef1d1af28ca37fdbc2745da040529cd2953c1af5 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 18 Jan 2011 21:41:45 -0200 Subject: perf evsel: Introduce perf_evsel__{in,ex}it Out of the {con,des}structor, as in interpreted language bindings we will need to go back from the wrapper object to the real thing. In that case using container_of will save us to have an extra pointer in the perf_evsel struct. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 29 ++++++++++++++++++++--------- 1 file changed, 20 insertions(+), 9 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 4b3b84cd71a1..df0610e9c61b 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -6,17 +6,21 @@ #include #include +void perf_evlist__init(struct perf_evlist *evlist) +{ + int i; + + for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) + INIT_HLIST_HEAD(&evlist->heads[i]); + INIT_LIST_HEAD(&evlist->entries); +} + struct perf_evlist *perf_evlist__new(void) { struct perf_evlist *evlist = zalloc(sizeof(*evlist)); - if (evlist != NULL) { - int i; - - for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) - INIT_HLIST_HEAD(&evlist->heads[i]); - INIT_LIST_HEAD(&evlist->entries); - } + if (evlist != NULL) + perf_evlist__init(evlist); return evlist; } @@ -33,11 +37,18 @@ static void perf_evlist__purge(struct perf_evlist *evlist) evlist->nr_entries = 0; } -void perf_evlist__delete(struct perf_evlist *evlist) +void perf_evlist__exit(struct perf_evlist *evlist) { - perf_evlist__purge(evlist); free(evlist->mmap); free(evlist->pollfd); + evlist->mmap = NULL; + evlist->pollfd = NULL; +} + +void perf_evlist__delete(struct perf_evlist *evlist) +{ + perf_evlist__purge(evlist); + perf_evlist__exit(evlist); free(evlist); } -- cgit v1.2.2 From 7bb41152b9be7e31f10d8919bce5034135525d9d Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sat, 29 Jan 2011 09:08:13 -0200 Subject: perf evlist: Support non overwrite mode in perf_evlist__read_on_cpu I.e. stash the overwrite mode in struct perf_evlist and act accordingly in perf_evlist__read_on_cpu, not checking for overwrites and touching the tail after consuming one event, like perf record does, for instance. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 35 ++++++++++++++++++++--------------- 1 file changed, 20 insertions(+), 15 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index df0610e9c61b..b498eecbe850 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -116,24 +116,25 @@ event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) unsigned int old = md->prev; unsigned char *data = md->base + page_size; event_t *event = NULL; - int diff; - - /* - * If we're further behind than half the buffer, there's a chance - * the writer will bite our tail and mess up the samples under us. - * - * If we somehow ended up ahead of the head, we got messed up. - * - * In either case, truncate and restart at head. - */ - diff = head - old; - if (diff > md->mask / 2 || diff < 0) { - fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); + if (evlist->overwrite) { /* - * head points to a known good entry, start there. + * If we're further behind than half the buffer, there's a chance + * the writer will bite our tail and mess up the samples under us. + * + * If we somehow ended up ahead of the head, we got messed up. + * + * In either case, truncate and restart at head. */ - old = head; + int diff = head - old; + if (diff > md->mask / 2 || diff < 0) { + fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); + + /* + * head points to a known good entry, start there. + */ + old = head; + } } if (old != head) { @@ -166,5 +167,9 @@ event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) } md->prev = old; + + if (!evlist->overwrite) + perf_mmap__write_tail(md, old); + return event; } -- cgit v1.2.2 From 8115d60c323dd9931b95221c0a392aeddc1d6ef3 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sat, 29 Jan 2011 14:01:45 -0200 Subject: perf tools: Kill event_t typedef, use 'union perf_event' instead And move the event_t methods to the perf_event__ too. No code changes, just namespace consistency. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index b498eecbe850..917fc18d0bed 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -107,7 +107,7 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) return NULL; } -event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) +union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) { /* XXX Move this to perf.c, making it generally available */ unsigned int page_size = sysconf(_SC_PAGE_SIZE); @@ -115,7 +115,7 @@ event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) unsigned int head = perf_mmap__read_head(md); unsigned int old = md->prev; unsigned char *data = md->base + page_size; - event_t *event = NULL; + union perf_event *event = NULL; if (evlist->overwrite) { /* @@ -140,7 +140,7 @@ event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) if (old != head) { size_t size; - event = (event_t *)&data[old & md->mask]; + event = (union perf_event *)&data[old & md->mask]; size = event->header.size; /* -- cgit v1.2.2 From f8a9530939ed87b9a1b1a038b90e355098b679a2 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sun, 30 Jan 2011 10:46:46 -0200 Subject: perf evlist: Move evlist methods to evlist.c They were on evsel.c because they came from refactoring existing evsel methods, so, to make reviewing the changes easier, I kept it there, now its a plain move. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 142 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 142 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 917fc18d0bed..dcd59328bb49 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -1,11 +1,26 @@ +/* + * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo + * + * Parts came from builtin-{top,stat,record}.c, see those files for further + * copyright notes. + * + * Released under the GPL v2. (and only v2, not any later version) + */ #include +#include "cpumap.h" +#include "thread_map.h" #include "evlist.h" #include "evsel.h" #include "util.h" +#include + #include #include +#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) +#define SID(e, x, y) xyarray__entry(e->id, x, y) + void perf_evlist__init(struct perf_evlist *evlist) { int i; @@ -88,6 +103,30 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) evlist->nr_fds++; } +static int perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, + int cpu, int thread, int fd) +{ + struct perf_sample_id *sid; + u64 read_data[4] = { 0, }; + int hash, id_idx = 1; /* The first entry is the counter value */ + + if (!(evsel->attr.read_format & PERF_FORMAT_ID) || + read(fd, &read_data, sizeof(read_data)) == -1) + return -1; + + if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + ++id_idx; + if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + ++id_idx; + + sid = SID(evsel, cpu, thread); + sid->id = read_data[id_idx]; + sid->evsel = evsel; + hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); + hlist_add_head(&sid->node, &evlist->heads[hash]); + return 0; +} + struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) { struct hlist_head *head; @@ -173,3 +212,106 @@ union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) return event; } + +void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus) +{ + int cpu; + + for (cpu = 0; cpu < ncpus; cpu++) { + if (evlist->mmap[cpu].base != NULL) { + munmap(evlist->mmap[cpu].base, evlist->mmap_len); + evlist->mmap[cpu].base = NULL; + } + } +} + +int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus) +{ + evlist->mmap = zalloc(ncpus * sizeof(struct perf_mmap)); + return evlist->mmap != NULL ? 0 : -ENOMEM; +} + +static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot, + int mask, int fd) +{ + evlist->mmap[cpu].prev = 0; + evlist->mmap[cpu].mask = mask; + evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot, + MAP_SHARED, fd, 0); + if (evlist->mmap[cpu].base == MAP_FAILED) + return -1; + + perf_evlist__add_pollfd(evlist, fd); + return 0; +} + +/** perf_evlist__mmap - Create per cpu maps to receive events + * + * @evlist - list of events + * @cpus - cpu map being monitored + * @threads - threads map being monitored + * @pages - map length in pages + * @overwrite - overwrite older events? + * + * If overwrite is false the user needs to signal event consuption using: + * + * struct perf_mmap *m = &evlist->mmap[cpu]; + * unsigned int head = perf_mmap__read_head(m); + * + * perf_mmap__write_tail(m, head) + */ +int perf_evlist__mmap(struct perf_evlist *evlist, struct cpu_map *cpus, + struct thread_map *threads, int pages, bool overwrite) +{ + unsigned int page_size = sysconf(_SC_PAGE_SIZE); + int mask = pages * page_size - 1, cpu; + struct perf_evsel *first_evsel, *evsel; + int thread, prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); + + if (evlist->mmap == NULL && + perf_evlist__alloc_mmap(evlist, cpus->nr) < 0) + return -ENOMEM; + + if (evlist->pollfd == NULL && + perf_evlist__alloc_pollfd(evlist, cpus->nr, threads->nr) < 0) + return -ENOMEM; + + evlist->overwrite = overwrite; + evlist->mmap_len = (pages + 1) * page_size; + first_evsel = list_entry(evlist->entries.next, struct perf_evsel, node); + + list_for_each_entry(evsel, &evlist->entries, node) { + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + evsel->id == NULL && + perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) + return -ENOMEM; + + for (cpu = 0; cpu < cpus->nr; cpu++) { + for (thread = 0; thread < threads->nr; thread++) { + int fd = FD(evsel, cpu, thread); + + if (evsel->idx || thread) { + if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, + FD(first_evsel, cpu, 0)) != 0) + goto out_unmap; + } else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0) + goto out_unmap; + + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + perf_evlist__id_hash(evlist, evsel, cpu, thread, fd) < 0) + goto out_unmap; + } + } + } + + return 0; + +out_unmap: + for (cpu = 0; cpu < cpus->nr; cpu++) { + if (evlist->mmap[cpu].base != NULL) { + munmap(evlist->mmap[cpu].base, evlist->mmap_len); + evlist->mmap[cpu].base = NULL; + } + } + return -1; +} -- cgit v1.2.2 From 7e2ed097538c57ff5268e9a6bced7c0b885809c8 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sun, 30 Jan 2011 11:59:43 -0200 Subject: perf evlist: Store pointer to the cpu and thread maps So that we don't have to pass it around to the several methods that needs it, simplifying usage. There is one case where we don't have the thread/cpu map in advance, which is in the parsing routines used by top, stat, record, that we have to wait till all options are parsed to know if a cpu or thread list was passed to then create those maps. For that case consolidate the cpu and thread map creation via perf_evlist__create_maps() out of the code in top and record, while also providing a perf_evlist__set_maps() for cases where multiple evlists share maps or for when maps that represent CPU sockets, for instance, get crafted out of topology information or subsets of threads in a particular application are to be monitored, providing more granularity in specifying which cpus and threads to monitor. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 67 ++++++++++++++++++++++++++++++++++++------------ 1 file changed, 50 insertions(+), 17 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index dcd59328bb49..95b21fece2ce 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -21,21 +21,24 @@ #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) #define SID(e, x, y) xyarray__entry(e->id, x, y) -void perf_evlist__init(struct perf_evlist *evlist) +void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus, + struct thread_map *threads) { int i; for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) INIT_HLIST_HEAD(&evlist->heads[i]); INIT_LIST_HEAD(&evlist->entries); + perf_evlist__set_maps(evlist, cpus, threads); } -struct perf_evlist *perf_evlist__new(void) +struct perf_evlist *perf_evlist__new(struct cpu_map *cpus, + struct thread_map *threads) { struct perf_evlist *evlist = zalloc(sizeof(*evlist)); if (evlist != NULL) - perf_evlist__init(evlist); + perf_evlist__init(evlist, cpus, threads); return evlist; } @@ -88,9 +91,9 @@ int perf_evlist__add_default(struct perf_evlist *evlist) return 0; } -int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads) +int perf_evlist__alloc_pollfd(struct perf_evlist *evlist) { - int nfds = ncpus * nthreads * evlist->nr_entries; + int nfds = evlist->cpus->nr * evlist->threads->nr * evlist->nr_entries; evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); return evlist->pollfd != NULL ? 0 : -ENOMEM; } @@ -213,11 +216,11 @@ union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) return event; } -void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus) +void perf_evlist__munmap(struct perf_evlist *evlist) { int cpu; - for (cpu = 0; cpu < ncpus; cpu++) { + for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { if (evlist->mmap[cpu].base != NULL) { munmap(evlist->mmap[cpu].base, evlist->mmap_len); evlist->mmap[cpu].base = NULL; @@ -225,9 +228,9 @@ void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus) } } -int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus) +int perf_evlist__alloc_mmap(struct perf_evlist *evlist) { - evlist->mmap = zalloc(ncpus * sizeof(struct perf_mmap)); + evlist->mmap = zalloc(evlist->cpus->nr * sizeof(struct perf_mmap)); return evlist->mmap != NULL ? 0 : -ENOMEM; } @@ -248,8 +251,6 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot, /** perf_evlist__mmap - Create per cpu maps to receive events * * @evlist - list of events - * @cpus - cpu map being monitored - * @threads - threads map being monitored * @pages - map length in pages * @overwrite - overwrite older events? * @@ -259,21 +260,22 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot, * unsigned int head = perf_mmap__read_head(m); * * perf_mmap__write_tail(m, head) + * + * Using perf_evlist__read_on_cpu does this automatically. */ -int perf_evlist__mmap(struct perf_evlist *evlist, struct cpu_map *cpus, - struct thread_map *threads, int pages, bool overwrite) +int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) { unsigned int page_size = sysconf(_SC_PAGE_SIZE); int mask = pages * page_size - 1, cpu; struct perf_evsel *first_evsel, *evsel; + const struct cpu_map *cpus = evlist->cpus; + const struct thread_map *threads = evlist->threads; int thread, prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); - if (evlist->mmap == NULL && - perf_evlist__alloc_mmap(evlist, cpus->nr) < 0) + if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0) return -ENOMEM; - if (evlist->pollfd == NULL && - perf_evlist__alloc_pollfd(evlist, cpus->nr, threads->nr) < 0) + if (evlist->pollfd == NULL && perf_evlist__alloc_pollfd(evlist) < 0) return -ENOMEM; evlist->overwrite = overwrite; @@ -315,3 +317,34 @@ out_unmap: } return -1; } + +int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, + pid_t target_tid, const char *cpu_list) +{ + evlist->threads = thread_map__new(target_pid, target_tid); + + if (evlist->threads == NULL) + return -1; + + if (target_tid != -1) + evlist->cpus = cpu_map__dummy_new(); + else + evlist->cpus = cpu_map__new(cpu_list); + + if (evlist->cpus == NULL) + goto out_delete_threads; + + return 0; + +out_delete_threads: + thread_map__delete(evlist->threads); + return -1; +} + +void perf_evlist__delete_maps(struct perf_evlist *evlist) +{ + cpu_map__delete(evlist->cpus); + thread_map__delete(evlist->threads); + evlist->cpus = NULL; + evlist->threads = NULL; +} -- cgit v1.2.2 From 0a10247914a5cad3caf7ef8a255c54c4d3ed2062 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Sat, 26 Feb 2011 04:51:54 +0100 Subject: perf: Set filters before mmaping events We currently set the filters after we mmap the events, this is a race that let undesired events record themselves in the buffer before we had the time to set the filters. So set the filters before they can be recorded. That also librarizes the filters setting so that filtering can be done more easily from other tools than perf record later. Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Cc: Arnaldo Carvalho de Melo Cc: Steven Rostedt --- tools/perf/util/evlist.c | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 95b21fece2ce..030ae7f05e03 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -348,3 +348,31 @@ void perf_evlist__delete_maps(struct perf_evlist *evlist) evlist->cpus = NULL; evlist->threads = NULL; } + +int perf_evlist__set_filters(struct perf_evlist *evlist) +{ + const struct thread_map *threads = evlist->threads; + const struct cpu_map *cpus = evlist->cpus; + struct perf_evsel *evsel; + char *filter; + int thread; + int cpu; + int err; + int fd; + + list_for_each_entry(evsel, &evlist->entries, node) { + filter = evsel->filter; + if (!filter) + continue; + for (cpu = 0; cpu < cpus->nr; cpu++) { + for (thread = 0; thread < threads->nr; thread++) { + fd = FD(evsel, cpu, thread); + err = ioctl(fd, PERF_EVENT_IOC_SET_FILTER, filter); + if (err) + return err; + } + } + } + + return 0; +} -- cgit v1.2.2 From 3d3b5e95997208067c963923db90ed1517565d14 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 4 Mar 2011 22:29:39 -0300 Subject: perf evlist: Split perf_evlist__id_hash The previous situation was to receive an fd from where to read the event ID. Spin off a routine for when we have the ID handy, not having to read it from some fd. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 28 ++++++++++++++++++---------- 1 file changed, 18 insertions(+), 10 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 030ae7f05e03..190c64c6e266 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -106,12 +106,24 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) evlist->nr_fds++; } -static int perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, - int cpu, int thread, int fd) +void perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, + int cpu, int thread, u64 id) +{ + int hash; + struct perf_sample_id *sid = SID(evsel, cpu, thread); + + sid->id = id; + sid->evsel = evsel; + hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); + hlist_add_head(&sid->node, &evlist->heads[hash]); +} + +static int perf_evlist__id_hash_fd(struct perf_evlist *evlist, + struct perf_evsel *evsel, + int cpu, int thread, int fd) { - struct perf_sample_id *sid; u64 read_data[4] = { 0, }; - int hash, id_idx = 1; /* The first entry is the counter value */ + int id_idx = 1; /* The first entry is the counter value */ if (!(evsel->attr.read_format & PERF_FORMAT_ID) || read(fd, &read_data, sizeof(read_data)) == -1) @@ -122,11 +134,7 @@ static int perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *e if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) ++id_idx; - sid = SID(evsel, cpu, thread); - sid->id = read_data[id_idx]; - sid->evsel = evsel; - hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); - hlist_add_head(&sid->node, &evlist->heads[hash]); + perf_evlist__id_hash(evlist, evsel, cpu, thread, read_data[id_idx]); return 0; } @@ -300,7 +308,7 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) goto out_unmap; if ((evsel->attr.read_format & PERF_FORMAT_ID) && - perf_evlist__id_hash(evlist, evsel, cpu, thread, fd) < 0) + perf_evlist__id_hash_fd(evlist, evsel, cpu, thread, fd) < 0) goto out_unmap; } } -- cgit v1.2.2 From a91e5431d54f5359fccb5ec2512f252eb217707e Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 10 Mar 2011 11:15:54 -0300 Subject: perf session: Use evlist/evsel for managing perf.data attributes So that we can reuse things like the id to attr lookup routine (perf_evlist__id2evsel) that uses a hash table instead of the linear lookup done in the older perf_header_attr routines, etc. Also to make evsels/evlist more pervasive an API, simplyfing using the emerging perf lib. cc: Arun Sharma Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 26 +++++++++++++++++--------- 1 file changed, 17 insertions(+), 9 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 190c64c6e266..d852cefa20de 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -19,7 +19,7 @@ #include #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) -#define SID(e, x, y) xyarray__entry(e->id, x, y) +#define SID(e, x, y) xyarray__entry(e->sample_id, x, y) void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus, struct thread_map *threads) @@ -106,8 +106,9 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) evlist->nr_fds++; } -void perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, - int cpu, int thread, u64 id) +static void perf_evlist__id_hash(struct perf_evlist *evlist, + struct perf_evsel *evsel, + int cpu, int thread, u64 id) { int hash; struct perf_sample_id *sid = SID(evsel, cpu, thread); @@ -118,9 +119,16 @@ void perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, hlist_add_head(&sid->node, &evlist->heads[hash]); } -static int perf_evlist__id_hash_fd(struct perf_evlist *evlist, - struct perf_evsel *evsel, - int cpu, int thread, int fd) +void perf_evlist__id_add(struct perf_evlist *evlist, struct perf_evsel *evsel, + int cpu, int thread, u64 id) +{ + perf_evlist__id_hash(evlist, evsel, cpu, thread, id); + evsel->id[evsel->ids++] = id; +} + +static int perf_evlist__id_add_fd(struct perf_evlist *evlist, + struct perf_evsel *evsel, + int cpu, int thread, int fd) { u64 read_data[4] = { 0, }; int id_idx = 1; /* The first entry is the counter value */ @@ -134,7 +142,7 @@ static int perf_evlist__id_hash_fd(struct perf_evlist *evlist, if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) ++id_idx; - perf_evlist__id_hash(evlist, evsel, cpu, thread, read_data[id_idx]); + perf_evlist__id_add(evlist, evsel, cpu, thread, read_data[id_idx]); return 0; } @@ -292,7 +300,7 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) list_for_each_entry(evsel, &evlist->entries, node) { if ((evsel->attr.read_format & PERF_FORMAT_ID) && - evsel->id == NULL && + evsel->sample_id == NULL && perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) return -ENOMEM; @@ -308,7 +316,7 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) goto out_unmap; if ((evsel->attr.read_format & PERF_FORMAT_ID) && - perf_evlist__id_hash_fd(evlist, evsel, cpu, thread, fd) < 0) + perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) goto out_unmap; } } -- cgit v1.2.2 From 5d2cd90922c778908bd0cd669e572a5b5eafd737 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 14 Apr 2011 11:20:14 -0300 Subject: perf evsel: Fix use of inherit perf stat doesn't mmap and its perfectly fine for it to use task-bound counters with inheritance. So set the attr.inherit on the caller and leave the syscall itself to validate it. When the mmap fails perf_evlist__mmap will just emit a warning if this is the failure reason. Reported-by: Peter Zijlstra Acked-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Link: http://lkml.kernel.org/r/20110414170121.GC3229@ghostprotocols.net Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index d852cefa20de..45da8d186b49 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -12,6 +12,7 @@ #include "evlist.h" #include "evsel.h" #include "util.h" +#include "debug.h" #include @@ -250,15 +251,19 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist) return evlist->mmap != NULL ? 0 : -ENOMEM; } -static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot, - int mask, int fd) +static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *evsel, + int cpu, int prot, int mask, int fd) { evlist->mmap[cpu].prev = 0; evlist->mmap[cpu].mask = mask; evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot, MAP_SHARED, fd, 0); - if (evlist->mmap[cpu].base == MAP_FAILED) + if (evlist->mmap[cpu].base == MAP_FAILED) { + if (evlist->cpus->map[cpu] == -1 && evsel->attr.inherit) + ui__warning("Inherit is not allowed on per-task " + "events using mmap.\n"); return -1; + } perf_evlist__add_pollfd(evlist, fd); return 0; @@ -312,7 +317,8 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, FD(first_evsel, cpu, 0)) != 0) goto out_unmap; - } else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0) + } else if (__perf_evlist__mmap(evlist, evsel, cpu, + prot, mask, fd) < 0) goto out_unmap; if ((evsel->attr.read_format & PERF_FORMAT_ID) && -- cgit v1.2.2 From b90194181988063266f3da0b7bf3e57268c627c8 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 25 Apr 2011 16:25:20 -0300 Subject: perf tools: Honour the cpu list parameter when also monitoring a thread list The perf_evlist__create_maps was discarding the --cpu parameter when a --pid or --tid was specified, fix that. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 45da8d186b49..1884a7c7eb8f 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -348,7 +348,7 @@ int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, if (evlist->threads == NULL) return -1; - if (target_tid != -1) + if (cpu_list == NULL && target_tid != -1) evlist->cpus = cpu_map__dummy_new(); else evlist->cpus = cpu_map__new(cpu_list); -- cgit v1.2.2 From aece948f5ddd70d70df2f35855c706ef9a4f62e2 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sun, 15 May 2011 09:39:00 -0300 Subject: perf evlist: Fix per thread mmap setup The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using --pid when monitoring multithreaded apps, as we can only share a ring buffer for events on the same thread if not doing per cpu. Fix it by using per thread ring buffers. Tested with: [root@felicio ~]# tuna -t 26131 -CP | nl 1 thread ctxt_switches 2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd 3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse 4 642 OTHER 0 0,1 14688 0 chromium-browse 5 26148 OTHER 0 0,1 713602 115479 chromium-browse 6 26149 OTHER 0 0,1 801958 2262 chromium-browse 7 26150 OTHER 0 0,1 1271128 248 chromium-browse 8 26151 OTHER 0 0,1 3 0 chromium-browse 9 27049 OTHER 0 0,1 36796 9 chromium-browse 10 618 OTHER 0 0,1 14711 0 chromium-browse 11 661 OTHER 0 0,1 14593 0 chromium-browse 12 29048 OTHER 0 0,1 28125 0 chromium-browse 13 26143 OTHER 0 0,1 2202789 781 chromium-browse [root@felicio ~]# So 11 threads under pid 26131, then: [root@felicio ~]# perf record -F 50000 --pid 26131 [root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl 1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] [root@felicio ~]# 11 mmaps, one per thread since we didn't specify any CPU list, so we need one mmap per thread and: [root@felicio ~]# perf record -F 50000 --pid 26131 ^M ^C[ perf record: Woken up 79 times to write data ] [ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ] [root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl 1 371310 26131 2 96516 26148 3 95694 26149 4 95203 26150 5 7291 26143 6 87 27049 7 76 661 8 60 29048 9 47 618 10 43 642 [root@felicio ~]# Ok, one of the threads, 26151 was quiescent, so no samples there, but all the others are there. Then, if I specify one CPU: [root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1 ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ] [root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl 1 8444 26131 2 2584 26149 3 2518 26148 4 2324 26150 5 123 26143 6 9 661 7 9 29048 [root@felicio ~]# This machine has two cores, so fewer threads appeared on the radar, and: [root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl 1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] [root@felicio ~]# Just one mmap, as now we can use just one per-cpu buffer instead of the per-thread needed in the previous case. For global profiling: [root@felicio ~]# perf record -F 50000 -a ^C[ perf record: Woken up 26 times to write data ] [ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ] [root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl 1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] 2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] [root@felicio ~]# It uses per-cpu buffers. For just one thread: [root@felicio ~]# perf record -F 50000 --tid 26148 ^C[ perf record: Woken up 2 times to write data ] [ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ] [root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl 1 9969 26148 [root@felicio ~]# [root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl 1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event] [root@felicio ~]# Tested-by: David Ahern Tested-by: Lin Ming Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 151 +++++++++++++++++++++++++++++++++-------------- 1 file changed, 106 insertions(+), 45 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 1884a7c7eb8f..23eb22b05d27 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -166,11 +166,11 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) return NULL; } -union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) +union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx) { /* XXX Move this to perf.c, making it generally available */ unsigned int page_size = sysconf(_SC_PAGE_SIZE); - struct perf_mmap *md = &evlist->mmap[cpu]; + struct perf_mmap *md = &evlist->mmap[idx]; unsigned int head = perf_mmap__read_head(md); unsigned int old = md->prev; unsigned char *data = md->base + page_size; @@ -235,31 +235,37 @@ union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) void perf_evlist__munmap(struct perf_evlist *evlist) { - int cpu; + int i; - for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { - if (evlist->mmap[cpu].base != NULL) { - munmap(evlist->mmap[cpu].base, evlist->mmap_len); - evlist->mmap[cpu].base = NULL; + for (i = 0; i < evlist->nr_mmaps; i++) { + if (evlist->mmap[i].base != NULL) { + munmap(evlist->mmap[i].base, evlist->mmap_len); + evlist->mmap[i].base = NULL; } } + + free(evlist->mmap); + evlist->mmap = NULL; } int perf_evlist__alloc_mmap(struct perf_evlist *evlist) { - evlist->mmap = zalloc(evlist->cpus->nr * sizeof(struct perf_mmap)); + evlist->nr_mmaps = evlist->cpus->nr; + if (evlist->cpus->map[0] == -1) + evlist->nr_mmaps = evlist->threads->nr; + evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap)); return evlist->mmap != NULL ? 0 : -ENOMEM; } static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *evsel, - int cpu, int prot, int mask, int fd) + int idx, int prot, int mask, int fd) { - evlist->mmap[cpu].prev = 0; - evlist->mmap[cpu].mask = mask; - evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot, + evlist->mmap[idx].prev = 0; + evlist->mmap[idx].mask = mask; + evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot, MAP_SHARED, fd, 0); - if (evlist->mmap[cpu].base == MAP_FAILED) { - if (evlist->cpus->map[cpu] == -1 && evsel->attr.inherit) + if (evlist->mmap[idx].base == MAP_FAILED) { + if (evlist->cpus->map[idx] == -1 && evsel->attr.inherit) ui__warning("Inherit is not allowed on per-task " "events using mmap.\n"); return -1; @@ -269,6 +275,86 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *ev return 0; } +static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mask) +{ + struct perf_evsel *evsel; + int cpu, thread; + + for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { + int output = -1; + + for (thread = 0; thread < evlist->threads->nr; thread++) { + list_for_each_entry(evsel, &evlist->entries, node) { + int fd = FD(evsel, cpu, thread); + + if (output == -1) { + output = fd; + if (__perf_evlist__mmap(evlist, evsel, cpu, + prot, mask, output) < 0) + goto out_unmap; + } else { + if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) + goto out_unmap; + } + + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) + goto out_unmap; + } + } + } + + return 0; + +out_unmap: + for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { + if (evlist->mmap[cpu].base != NULL) { + munmap(evlist->mmap[cpu].base, evlist->mmap_len); + evlist->mmap[cpu].base = NULL; + } + } + return -1; +} + +static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, int mask) +{ + struct perf_evsel *evsel; + int thread; + + for (thread = 0; thread < evlist->threads->nr; thread++) { + int output = -1; + + list_for_each_entry(evsel, &evlist->entries, node) { + int fd = FD(evsel, 0, thread); + + if (output == -1) { + output = fd; + if (__perf_evlist__mmap(evlist, evsel, thread, + prot, mask, output) < 0) + goto out_unmap; + } else { + if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) + goto out_unmap; + } + + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + perf_evlist__id_add_fd(evlist, evsel, 0, thread, fd) < 0) + goto out_unmap; + } + } + + return 0; + +out_unmap: + for (thread = 0; thread < evlist->threads->nr; thread++) { + if (evlist->mmap[thread].base != NULL) { + munmap(evlist->mmap[thread].base, evlist->mmap_len); + evlist->mmap[thread].base = NULL; + } + } + return -1; +} + /** perf_evlist__mmap - Create per cpu maps to receive events * * @evlist - list of events @@ -287,11 +373,11 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *ev int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) { unsigned int page_size = sysconf(_SC_PAGE_SIZE); - int mask = pages * page_size - 1, cpu; - struct perf_evsel *first_evsel, *evsel; + int mask = pages * page_size - 1; + struct perf_evsel *evsel; const struct cpu_map *cpus = evlist->cpus; const struct thread_map *threads = evlist->threads; - int thread, prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); + int prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0) return -ENOMEM; @@ -301,43 +387,18 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) evlist->overwrite = overwrite; evlist->mmap_len = (pages + 1) * page_size; - first_evsel = list_entry(evlist->entries.next, struct perf_evsel, node); list_for_each_entry(evsel, &evlist->entries, node) { if ((evsel->attr.read_format & PERF_FORMAT_ID) && evsel->sample_id == NULL && perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) return -ENOMEM; - - for (cpu = 0; cpu < cpus->nr; cpu++) { - for (thread = 0; thread < threads->nr; thread++) { - int fd = FD(evsel, cpu, thread); - - if (evsel->idx || thread) { - if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, - FD(first_evsel, cpu, 0)) != 0) - goto out_unmap; - } else if (__perf_evlist__mmap(evlist, evsel, cpu, - prot, mask, fd) < 0) - goto out_unmap; - - if ((evsel->attr.read_format & PERF_FORMAT_ID) && - perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) - goto out_unmap; - } - } } - return 0; + if (evlist->cpus->map[0] == -1) + return perf_evlist__mmap_per_thread(evlist, prot, mask); -out_unmap: - for (cpu = 0; cpu < cpus->nr; cpu++) { - if (evlist->mmap[cpu].base != NULL) { - munmap(evlist->mmap[cpu].base, evlist->mmap_len); - evlist->mmap[cpu].base = NULL; - } - } - return -1; + return perf_evlist__mmap_per_cpu(evlist, prot, mask); } int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, -- cgit v1.2.2 From 74429964d8e29c0107fa6e9cdf35b8f33f57405d Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Sat, 21 May 2011 17:49:00 +0200 Subject: perf tools: Move evlist sample helpers to evlist area These APIs should belong to evlist.c as they may not be exclusively tied to the headers. Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Stephane Eranian entries, node) { + if (!type) + type = pos->attr.sample_type; + else if (type != pos->attr.sample_type) + die("non matching sample_type"); + } + + return type; +} + +bool perf_evlist__sample_id_all(const struct perf_evlist *evlist) +{ + bool value = false, first = true; + struct perf_evsel *pos; + + list_for_each_entry(pos, &evlist->entries, node) { + if (first) { + value = pos->attr.sample_id_all; + first = false; + } else if (value != pos->attr.sample_id_all) + die("non matching sample_id_all"); + } + + return value; +} -- cgit v1.2.2 From d21cc9f67d689effbfd24bac878bc2c057de8c46 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 2 Jun 2011 10:39:43 -0300 Subject: perf evlist: Remove dependency on debug routines So far we avoided having to link debug.o in the python binding, keep it that way by not using ui__warning() in evlist.c. Cc: Frederic Weisbecker Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-4wtew8hd3g7ejnlehtspys2t@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 13 ++++--------- 1 file changed, 4 insertions(+), 9 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 50aa34879c33..04c8a60075b3 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -12,7 +12,6 @@ #include "evlist.h" #include "evsel.h" #include "util.h" -#include "debug.h" #include @@ -257,19 +256,15 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist) return evlist->mmap != NULL ? 0 : -ENOMEM; } -static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *evsel, +static int __perf_evlist__mmap(struct perf_evlist *evlist, int idx, int prot, int mask, int fd) { evlist->mmap[idx].prev = 0; evlist->mmap[idx].mask = mask; evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot, MAP_SHARED, fd, 0); - if (evlist->mmap[idx].base == MAP_FAILED) { - if (evlist->cpus->map[idx] == -1 && evsel->attr.inherit) - ui__warning("Inherit is not allowed on per-task " - "events using mmap.\n"); + if (evlist->mmap[idx].base == MAP_FAILED) return -1; - } perf_evlist__add_pollfd(evlist, fd); return 0; @@ -289,7 +284,7 @@ static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int m if (output == -1) { output = fd; - if (__perf_evlist__mmap(evlist, evsel, cpu, + if (__perf_evlist__mmap(evlist, cpu, prot, mask, output) < 0) goto out_unmap; } else { @@ -329,7 +324,7 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, in if (output == -1) { output = fd; - if (__perf_evlist__mmap(evlist, evsel, thread, + if (__perf_evlist__mmap(evlist, thread, prot, mask, output) < 0) goto out_unmap; } else { -- cgit v1.2.2 From 56722381b8506733852c44dacf6d7bc5f90aedaf Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 2 Jun 2011 11:04:54 -0300 Subject: perf evlist: Don't die if sample_{id_all|type} is invalid Fixes two more cases where the python binding would not load: . Not finding die(), which it shouldn't anyway, not good to just stop the world because some particular perf.data file is invalid, just propagate the error to the caller. . Not finding perf_sample_size: fix it by moving it from event.c to evsel, where it belongs, as most cases are moving to operate on an evsel object.o One of the fixed problems: [root@emilia ~]# python >>> import perf Traceback (most recent call last): File "", line 1, in ImportError: /home/acme/git/build/perf/python/perf.so: undefined symbol: perf_sample_size >>> [root@emilia ~]# Cc: Frederic Weisbecker Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-1hkj7b2cvgbfnoizsekjb6c9@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 55 ++++++++++++++++++++++++++++++------------------ 1 file changed, 34 insertions(+), 21 deletions(-) (limited to 'tools/perf/util/evlist.c') diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 04c8a60075b3..b021ea9265c3 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -455,33 +455,46 @@ int perf_evlist__set_filters(struct perf_evlist *evlist) return 0; } -u64 perf_evlist__sample_type(struct perf_evlist *evlist) +bool perf_evlist__valid_sample_type(const struct perf_evlist *evlist) { - struct perf_evsel *pos; - u64 type = 0; - - list_for_each_entry(pos, &evlist->entries, node) { - if (!type) - type = pos->attr.sample_type; - else if (type != pos->attr.sample_type) - die("non matching sample_type"); + struct perf_evsel *pos, *first; + + pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); + + list_for_each_entry_continue(pos, &evlist->entries, node) { + if (first->attr.sample_type != pos->attr.sample_type) + return false; } - return type; + return true; } -bool perf_evlist__sample_id_all(const struct perf_evlist *evlist) +u64 perf_evlist__sample_type(const struct perf_evlist *evlist) +{ + struct perf_evsel *first; + + first = list_entry(evlist->entries.next, struct perf_evsel, node); + return first->attr.sample_type; +} + +bool perf_evlist__valid_sample_id_all(const struct perf_evlist *evlist) { - bool value = false, first = true; - struct perf_evsel *pos; - - list_for_each_entry(pos, &evlist->entries, node) { - if (first) { - value = pos->attr.sample_id_all; - first = false; - } else if (value != pos->attr.sample_id_all) - die("non matching sample_id_all"); + struct perf_evsel *pos, *first; + + pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); + + list_for_each_entry_continue(pos, &evlist->entries, node) { + if (first->attr.sample_id_all != pos->attr.sample_id_all) + return false; } - return value; + return true; +} + +bool perf_evlist__sample_id_all(const struct perf_evlist *evlist) +{ + struct perf_evsel *first; + + first = list_entry(evlist->entries.next, struct perf_evsel, node); + return first->attr.sample_id_all; } -- cgit v1.2.2