From 69aad6f1ee69546dea8535ab8f3da9f445d57328 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 3 Jan 2011 16:39:04 -0200 Subject: perf tools: Introduce event selectors Out of ad-hoc code and global arrays with hard coded sizes. This is the first step on having a library that will be first used on regression tests in the 'perf test' tool. [acme@felicio linux]$ size /tmp/perf.before text data bss dec hex filename 1273776 97384 5104416 6475576 62cf38 /tmp/perf.before [acme@felicio linux]$ size /tmp/perf.new text data bss dec hex filename 1275422 97416 1392416 2765254 2a31c6 /tmp/perf.new 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/evsel.c | 35 +++++++++++++++++++++++++++++++++++ 1 file changed, 35 insertions(+) create mode 100644 tools/perf/util/evsel.c (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c new file mode 100644 index 0000000..6539ec9 --- /dev/null +++ b/tools/perf/util/evsel.c @@ -0,0 +1,35 @@ +#include "evsel.h" +#include "util.h" + +struct perf_evsel *perf_evsel__new(u32 type, u64 config, int idx) +{ + struct perf_evsel *evsel = zalloc(sizeof(*evsel)); + + if (evsel != NULL) { + evsel->idx = idx; + evsel->attr.type = type; + evsel->attr.config = config; + INIT_LIST_HEAD(&evsel->node); + } + + return evsel; +} + +int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads) +{ + evsel->fd = xyarray__new(ncpus, nthreads, sizeof(int)); + return evsel->fd != NULL ? 0 : -ENOMEM; +} + +void perf_evsel__free_fd(struct perf_evsel *evsel) +{ + xyarray__delete(evsel->fd); + evsel->fd = NULL; +} + +void perf_evsel__delete(struct perf_evsel *evsel) +{ + assert(list_empty(&evsel->node)); + xyarray__delete(evsel->fd); + free(evsel); +} -- cgit v1.1 From c52b12ed2511e6c031a0295fd903ea72b93701fb Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 3 Jan 2011 17:45:52 -0200 Subject: perf evsel: Steal the counter reading routines from stat Making them hopefully generic enough to be used in 'perf test', well see. 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/evsel.c | 88 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 88 insertions(+) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 6539ec9..3f5de51 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1,6 +1,8 @@ #include "evsel.h" #include "util.h" +#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) + struct perf_evsel *perf_evsel__new(u32 type, u64 config, int idx) { struct perf_evsel *evsel = zalloc(sizeof(*evsel)); @@ -21,15 +23,101 @@ int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads) return evsel->fd != NULL ? 0 : -ENOMEM; } +int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus) +{ + evsel->counts = zalloc((sizeof(*evsel->counts) + + (ncpus * sizeof(struct perf_counts_values)))); + return evsel->counts != NULL ? 0 : -ENOMEM; +} + void perf_evsel__free_fd(struct perf_evsel *evsel) { xyarray__delete(evsel->fd); evsel->fd = NULL; } +void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) +{ + int cpu, thread; + + for (cpu = 0; cpu < ncpus; cpu++) + for (thread = 0; thread < nthreads; ++thread) { + close(FD(evsel, cpu, thread)); + FD(evsel, cpu, thread) = -1; + } +} + void perf_evsel__delete(struct perf_evsel *evsel) { assert(list_empty(&evsel->node)); xyarray__delete(evsel->fd); free(evsel); } + +int __perf_evsel__read_on_cpu(struct perf_evsel *evsel, + int cpu, int thread, bool scale) +{ + struct perf_counts_values count; + size_t nv = scale ? 3 : 1; + + if (FD(evsel, cpu, thread) < 0) + return -EINVAL; + + if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0) + return -errno; + + if (scale) { + if (count.run == 0) + count.val = 0; + else if (count.run < count.ena) + count.val = (u64)((double)count.val * count.ena / count.run + 0.5); + } else + count.ena = count.run = 0; + + evsel->counts->cpu[cpu] = count; + return 0; +} + +int __perf_evsel__read(struct perf_evsel *evsel, + int ncpus, int nthreads, bool scale) +{ + size_t nv = scale ? 3 : 1; + int cpu, thread; + struct perf_counts_values *aggr = &evsel->counts->aggr, count; + + aggr->val = 0; + + for (cpu = 0; cpu < ncpus; cpu++) { + for (thread = 0; thread < nthreads; thread++) { + if (FD(evsel, cpu, thread) < 0) + continue; + + if (readn(FD(evsel, cpu, thread), + &count, nv * sizeof(u64)) < 0) + return -errno; + + aggr->val += count.val; + if (scale) { + aggr->ena += count.ena; + aggr->run += count.run; + } + } + } + + evsel->counts->scaled = 0; + if (scale) { + if (aggr->run == 0) { + evsel->counts->scaled = -1; + aggr->val = 0; + return 0; + } + + if (aggr->run < aggr->ena) { + evsel->counts->scaled = 1; + aggr->val = (u64)((double)aggr->val * aggr->ena / aggr->run + 0.5); + } + } else + aggr->ena = aggr->run = 0; + + return 0; +} -- cgit v1.1 From 48290609c0d265f5dac0fca6fd4e3c5732542f67 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 3 Jan 2011 17:48:12 -0200 Subject: perf evsel: Introduce per cpu and per thread open helpers Abstracting away the loops needed to create the various event fd handlers. The users have to pass a confiruged perf->evsel.attr field, which is already usable after perf_evsel__new (constructor) time, using defaults. Comes out of the ad-hoc routines in builtin-stat, that now uses it. Fixed a small silly bug where we were die()ing before killing our children, dysfunctional family this one 8-) 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/evsel.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 52 insertions(+) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 3f5de51..e62cc5e 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1,4 +1,5 @@ #include "evsel.h" +#include "../perf.h" #include "util.h" #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) @@ -121,3 +122,54 @@ int __perf_evsel__read(struct perf_evsel *evsel, return 0; } + +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, int ncpus, int *cpu_map) +{ + int cpu; + + for (cpu = 0; cpu < ncpus; cpu++) { + FD(evsel, cpu, 0) = sys_perf_event_open(&evsel->attr, -1, + cpu_map[cpu], -1, 0); + if (FD(evsel, cpu, 0) < 0) + goto out_close; + } + + return 0; + +out_close: + while (--cpu >= 0) { + close(FD(evsel, cpu, 0)); + FD(evsel, cpu, 0) = -1; + } + return -1; +} + +int perf_evsel__open_per_thread(struct perf_evsel *evsel, int nthreads, int *thread_map) +{ + int thread; + + for (thread = 0; thread < nthreads; thread++) { + FD(evsel, 0, thread) = sys_perf_event_open(&evsel->attr, + thread_map[thread], -1, -1, 0); + if (FD(evsel, 0, thread) < 0) + goto out_close; + } + + return 0; + +out_close: + while (--thread >= 0) { + close(FD(evsel, 0, thread)); + FD(evsel, 0, thread) = -1; + } + return -1; +} + +int perf_evsel__open(struct perf_evsel *evsel, int ncpus, int nthreads, + int *cpu_map, int *thread_map) +{ + if (nthreads < 0) + return perf_evsel__open_per_cpu(evsel, ncpus, cpu_map); + + return perf_evsel__open_per_thread(evsel, nthreads, thread_map); +} -- cgit v1.1 From 86bd5e8603b00b06189328c6d7034d2dc434d6bb Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 3 Jan 2011 23:09:46 -0200 Subject: perf evsel: Use {cpu,thread}_map to shorten list of parameters 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/evsel.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index e62cc5e..e44be52 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1,6 +1,8 @@ #include "evsel.h" #include "../perf.h" #include "util.h" +#include "cpumap.h" +#include "thread.h" #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) @@ -123,13 +125,13 @@ int __perf_evsel__read(struct perf_evsel *evsel, return 0; } -int perf_evsel__open_per_cpu(struct perf_evsel *evsel, int ncpus, int *cpu_map) +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) { int cpu; - for (cpu = 0; cpu < ncpus; cpu++) { + for (cpu = 0; cpu < cpus->nr; cpu++) { FD(evsel, cpu, 0) = sys_perf_event_open(&evsel->attr, -1, - cpu_map[cpu], -1, 0); + cpus->map[cpu], -1, 0); if (FD(evsel, cpu, 0) < 0) goto out_close; } @@ -144,13 +146,13 @@ out_close: return -1; } -int perf_evsel__open_per_thread(struct perf_evsel *evsel, int nthreads, int *thread_map) +int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads) { int thread; - for (thread = 0; thread < nthreads; thread++) { + for (thread = 0; thread < threads->nr; thread++) { FD(evsel, 0, thread) = sys_perf_event_open(&evsel->attr, - thread_map[thread], -1, -1, 0); + threads->map[thread], -1, -1, 0); if (FD(evsel, 0, thread) < 0) goto out_close; } @@ -165,11 +167,11 @@ out_close: return -1; } -int perf_evsel__open(struct perf_evsel *evsel, int ncpus, int nthreads, - int *cpu_map, int *thread_map) +int perf_evsel__open(struct perf_evsel *evsel, + struct cpu_map *cpus, struct thread_map *threads) { - if (nthreads < 0) - return perf_evsel__open_per_cpu(evsel, ncpus, cpu_map); + if (threads == NULL) + return perf_evsel__open_per_cpu(evsel, cpus); - return perf_evsel__open_per_thread(evsel, nthreads, thread_map); + return perf_evsel__open_per_thread(evsel, threads); } -- cgit v1.1 From 4eed11d5e24540dc133003b6e8f904cb747ac4bb Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 4 Jan 2011 00:13:17 -0200 Subject: perf evsel: Auto allocate resources needed for some methods While writing the first user of the routines created from the ad-hoc routines in the existing builtins I noticed that the resulting set of calls was too long, reduce it by doing some best effort allocations. Tools that need to operate on multiple threads and cpus should pre-allocate enough resources by explicitely calling the perf_evsel__alloc_{fd,counters} methods. 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/evsel.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index e44be52..c95267e 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -66,6 +66,9 @@ int __perf_evsel__read_on_cpu(struct perf_evsel *evsel, if (FD(evsel, cpu, thread) < 0) return -EINVAL; + if (evsel->counts == NULL && perf_evsel__alloc_counts(evsel, cpu + 1) < 0) + return -ENOMEM; + if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0) return -errno; @@ -129,6 +132,9 @@ int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) { int cpu; + if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, cpus->nr, 1) < 0) + return -1; + for (cpu = 0; cpu < cpus->nr; cpu++) { FD(evsel, cpu, 0) = sys_perf_event_open(&evsel->attr, -1, cpus->map[cpu], -1, 0); @@ -150,6 +156,9 @@ int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *thr { int thread; + if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, 1, threads->nr)) + return -1; + for (thread = 0; thread < threads->nr; thread++) { FD(evsel, 0, thread) = sys_perf_event_open(&evsel->attr, threads->map[thread], -1, -1, 0); -- cgit v1.1 From 23a2f3ab46596d9fd0b0e592d2101bea90970594 Mon Sep 17 00:00:00 2001 From: Lin Ming Date: Fri, 7 Jan 2011 11:11:09 +0800 Subject: perf tools: Pass whole attr to event selectors Since commit 69aad6f1(perf tools: Introduce event selectors), only perf_event_attr::type and ::config are passed to event selector, which makes perf tool not work correctly. For example, PEBS does not work because perf_event_attr::precise_ip is not passed to the syscall. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Peter Zijlstra LKML-Reference: <1294369869.20563.19.camel@minggr.sh.intel.com> Signed-off-by: Lin Ming Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evsel.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index c95267e..1a5591d 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -6,14 +6,13 @@ #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) -struct perf_evsel *perf_evsel__new(u32 type, u64 config, int idx) +struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) { struct perf_evsel *evsel = zalloc(sizeof(*evsel)); if (evsel != NULL) { evsel->idx = idx; - evsel->attr.type = type; - evsel->attr.config = config; + evsel->attr = *attr; INIT_LIST_HEAD(&evsel->node); } -- cgit v1.1 From 0252208eb52f6fe8731a47804eddc7ba93f60a87 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 4 Jan 2011 11:55:27 -0200 Subject: perf evsel: Support perf_evsel__open(cpus > 1 && threads > 1) And a test for it: [acme@felicio linux]$ perf test 1: vmlinux symtab matches kallsyms: Ok 2: detect open syscall event: Ok 3: detect open syscall event on all cpus: Ok [acme@felicio linux]$ Translating C the test does: 1. generates different number of open syscalls on each CPU by using sched_setaffinity 2. Verifies that the expected number of events is generated on each CPU It works as expected. 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/evsel.c | 82 +++++++++++++++++++++++++++++-------------------- 1 file changed, 49 insertions(+), 33 deletions(-) (limited to 'tools/perf/util/evsel.c') diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 1a5591d..f5cfed6 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -127,59 +127,75 @@ int __perf_evsel__read(struct perf_evsel *evsel, return 0; } -int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) +static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, + struct thread_map *threads) { - int cpu; + int cpu, thread; - if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, cpus->nr, 1) < 0) + if (evsel->fd == NULL && + perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0) return -1; for (cpu = 0; cpu < cpus->nr; cpu++) { - FD(evsel, cpu, 0) = sys_perf_event_open(&evsel->attr, -1, - cpus->map[cpu], -1, 0); - if (FD(evsel, cpu, 0) < 0) - goto out_close; + for (thread = 0; thread < threads->nr; thread++) { + FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, + threads->map[thread], + cpus->map[cpu], -1, 0); + if (FD(evsel, cpu, thread) < 0) + goto out_close; + } } return 0; out_close: - while (--cpu >= 0) { - close(FD(evsel, cpu, 0)); - FD(evsel, cpu, 0) = -1; - } + do { + while (--thread >= 0) { + close(FD(evsel, cpu, thread)); + FD(evsel, cpu, thread) = -1; + } + thread = threads->nr; + } while (--cpu >= 0); return -1; } -int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads) +static struct { + struct cpu_map map; + int cpus[1]; +} empty_cpu_map = { + .map.nr = 1, + .cpus = { -1, }, +}; + +static struct { + struct thread_map map; + int threads[1]; +} empty_thread_map = { + .map.nr = 1, + .threads = { -1, }, +}; + +int perf_evsel__open(struct perf_evsel *evsel, + struct cpu_map *cpus, struct thread_map *threads) { - int thread; - - if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, 1, threads->nr)) - return -1; - for (thread = 0; thread < threads->nr; thread++) { - FD(evsel, 0, thread) = sys_perf_event_open(&evsel->attr, - threads->map[thread], -1, -1, 0); - if (FD(evsel, 0, thread) < 0) - goto out_close; + if (cpus == NULL) { + /* Work around old compiler warnings about strict aliasing */ + cpus = &empty_cpu_map.map; } - return 0; + if (threads == NULL) + threads = &empty_thread_map.map; -out_close: - while (--thread >= 0) { - close(FD(evsel, 0, thread)); - FD(evsel, 0, thread) = -1; - } - return -1; + return __perf_evsel__open(evsel, cpus, threads); } -int perf_evsel__open(struct perf_evsel *evsel, - struct cpu_map *cpus, struct thread_map *threads) +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) { - if (threads == NULL) - return perf_evsel__open_per_cpu(evsel, cpus); + return __perf_evsel__open(evsel, cpus, &empty_thread_map.map); +} - return perf_evsel__open_per_thread(evsel, threads); +int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads) +{ + return __perf_evsel__open(evsel, &empty_cpu_map.map, threads); } -- cgit v1.1