From 69fef0d2e2c2c049ef4207a52e78b50d527bd85a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 4 Mar 2010 13:57:24 +0100 Subject: perf: Add attr->precise support to raw event parsing Minimal userspace interface to the new 'precise' events flag. Can be used like "perf top -e r00c0p" which will use PEBS to sample retired instructions. Signed-off-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: paulus@samba.org Cc: eranian@google.com Cc: robert.richter@amd.com Cc: fweisbec@gmail.com LKML-Reference: <20100304140100.468665803@chello.nl> Signed-off-by: Ingo Molnar --- tools/perf/util/parse-events.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 05d0c5c2030..a2014459125 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -656,6 +656,10 @@ parse_raw_event(const char **strp, struct perf_event_attr *attr) return EVT_FAILED; n = hex2u64(str + 1, &config); if (n > 0) { + if (str[n+1] == 'p') { + attr->precise = 1; + n++; + } *strp = str + n + 1; attr->type = PERF_TYPE_RAW; attr->config = config; -- cgit v1.2.3-70-g09d2 From 5aab621b7bf024608f0c089e21656e7fe875a150 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 25 Mar 2010 19:59:00 -0300 Subject: perf symbols: Move hex2u64 and strxfrchar to symbol.c MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Mostly used in symbol.c so move them there to reduce the number of files needed to use the symbol system. Also do some header adjustments with the same intent. Signed-off-by: Arnaldo Carvalho de Melo Cc: Frédéric Weisbecker Cc: Mike Galbraith Cc: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <1269557941-15617-5-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar --- tools/perf/util/parse-events.c | 1 + tools/perf/util/string.c | 43 --------------------- tools/perf/util/string.h | 2 - tools/perf/util/symbol.c | 85 +++++++++++++++++++++++++++++++++--------- tools/perf/util/symbol.h | 10 ++++- 5 files changed, 76 insertions(+), 65 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index a2014459125..435781e0c20 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -5,6 +5,7 @@ #include "parse-events.h" #include "exec_cmd.h" #include "string.h" +#include "symbol.h" #include "cache.h" #include "header.h" #include "debugfs.h" diff --git a/tools/perf/util/string.c b/tools/perf/util/string.c index a175949ed21..d4389242cfd 100644 --- a/tools/perf/util/string.c +++ b/tools/perf/util/string.c @@ -1,49 +1,6 @@ #include "string.h" #include "util.h" -static int hex(char ch) -{ - if ((ch >= '0') && (ch <= '9')) - return ch - '0'; - if ((ch >= 'a') && (ch <= 'f')) - return ch - 'a' + 10; - if ((ch >= 'A') && (ch <= 'F')) - return ch - 'A' + 10; - return -1; -} - -/* - * While we find nice hex chars, build a long_val. - * Return number of chars processed. - */ -int hex2u64(const char *ptr, u64 *long_val) -{ - const char *p = ptr; - *long_val = 0; - - while (*p) { - const int hex_val = hex(*p); - - if (hex_val < 0) - break; - - *long_val = (*long_val << 4) | hex_val; - p++; - } - - return p - ptr; -} - -char *strxfrchar(char *s, char from, char to) -{ - char *p = s; - - while ((p = strchr(p, from)) != NULL) - *p++ = to; - - return s; -} - #define K 1024LL /* * perf_atoll() diff --git a/tools/perf/util/string.h b/tools/perf/util/string.h index 542e44de371..70058241666 100644 --- a/tools/perf/util/string.h +++ b/tools/perf/util/string.h @@ -4,8 +4,6 @@ #include #include "types.h" -int hex2u64(const char *ptr, u64 *val); -char *strxfrchar(char *s, char from, char to); s64 perf_atoll(const char *str); char **argv_split(const char *str, int *argcp); void argv_free(char **argv); diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c index 3eb9de4baef..f3d4151e46a 100644 --- a/tools/perf/util/symbol.c +++ b/tools/perf/util/symbol.c @@ -1,13 +1,19 @@ -#include "util.h" -#include "../perf.h" -#include "sort.h" -#include "string.h" +#define _GNU_SOURCE +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include #include "symbol.h" -#include "thread.h" +#include "strlist.h" -#include "debug.h" - -#include #include #include #include @@ -114,8 +120,8 @@ static void map_groups__fixup_end(struct map_groups *self) static struct symbol *symbol__new(u64 start, u64 len, const char *name) { size_t namelen = strlen(name) + 1; - struct symbol *self = zalloc(symbol_conf.priv_size + - sizeof(*self) + namelen); + struct symbol *self = calloc(1, (symbol_conf.priv_size + + sizeof(*self) + namelen)); if (self == NULL) return NULL; @@ -166,7 +172,7 @@ static void dso__set_basename(struct dso *self) struct dso *dso__new(const char *name) { - struct dso *self = zalloc(sizeof(*self) + strlen(name) + 1); + struct dso *self = calloc(1, sizeof(*self) + strlen(name) + 1); if (self != NULL) { int i; @@ -1382,13 +1388,13 @@ static int dso__kernel_module_get_build_id(struct dso *self) return 0; } -static int map_groups__set_modules_path_dir(struct map_groups *self, char *dirname) +static int map_groups__set_modules_path_dir(struct map_groups *self, char *dir_name) { struct dirent *dent; - DIR *dir = opendir(dirname); + DIR *dir = opendir(dir_name); if (!dir) { - pr_debug("%s: cannot open %s dir\n", __func__, dirname); + pr_debug("%s: cannot open %s dir\n", __func__, dir_name); return -1; } @@ -1401,7 +1407,7 @@ static int map_groups__set_modules_path_dir(struct map_groups *self, char *dirna continue; snprintf(path, sizeof(path), "%s/%s", - dirname, dent->d_name); + dir_name, dent->d_name); if (map_groups__set_modules_path_dir(self, path) < 0) goto failure; } else { @@ -1421,7 +1427,7 @@ static int map_groups__set_modules_path_dir(struct map_groups *self, char *dirna continue; snprintf(path, sizeof(path), "%s/%s", - dirname, dent->d_name); + dir_name, dent->d_name); long_name = strdup(path); if (long_name == NULL) @@ -1458,8 +1464,8 @@ static int map_groups__set_modules_path(struct map_groups *self) */ static struct map *map__new2(u64 start, struct dso *dso, enum map_type type) { - struct map *self = zalloc(sizeof(*self) + - (dso->kernel ? sizeof(struct kmap) : 0)); + struct map *self = calloc(1, (sizeof(*self) + + (dso->kernel ? sizeof(struct kmap) : 0))); if (self != NULL) { /* * ->end will be filled after we load all the symbols @@ -1963,3 +1969,46 @@ int map_groups__create_kernel_maps(struct map_groups *self, map_groups__fixup_end(self); return 0; } + +static int hex(char ch) +{ + if ((ch >= '0') && (ch <= '9')) + return ch - '0'; + if ((ch >= 'a') && (ch <= 'f')) + return ch - 'a' + 10; + if ((ch >= 'A') && (ch <= 'F')) + return ch - 'A' + 10; + return -1; +} + +/* + * While we find nice hex chars, build a long_val. + * Return number of chars processed. + */ +int hex2u64(const char *ptr, u64 *long_val) +{ + const char *p = ptr; + *long_val = 0; + + while (*p) { + const int hex_val = hex(*p); + + if (hex_val < 0) + break; + + *long_val = (*long_val << 4) | hex_val; + p++; + } + + return p - ptr; +} + +char *strxfrchar(char *s, char from, char to) +{ + char *p = s; + + while ((p = strchr(p, from)) != NULL) + *p++ = to; + + return s; +} diff --git a/tools/perf/util/symbol.h b/tools/perf/util/symbol.h index a4a894b8ea0..757fae3f5ee 100644 --- a/tools/perf/util/symbol.h +++ b/tools/perf/util/symbol.h @@ -3,10 +3,11 @@ #include #include -#include "types.h" +#include +#include "map.h" #include #include -#include "event.h" +#include #define DEBUG_CACHE_DIR ".debug" @@ -29,6 +30,9 @@ static inline char *bfd_demangle(void __used *v, const char __used *c, #endif #endif +int hex2u64(const char *ptr, u64 *val); +char *strxfrchar(char *s, char from, char to); + /* * libelf 0.8.x and earlier do not support ELF_C_READ_MMAP; * for newer versions we can use mmap to reduce memory usage: @@ -44,6 +48,8 @@ static inline char *bfd_demangle(void __used *v, const char __used *c, #define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */ #endif +#define BUILD_ID_SIZE 20 + struct symbol { struct rb_node rb_node; u64 start; -- cgit v1.2.3-70-g09d2 From bdef3b02ceeb97f5f67fcfa6dff13c4e70b34fb7 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 14 Apr 2010 20:05:17 +0200 Subject: perf: Always record tracepoints raw samples from perf record Trace events are mostly used for tracing rather than simple counting. Don't bother anymore with adding -R when using them, just record raw samples of trace events every time. Signed-off-by: Frederic Weisbecker Cc: Arnaldo Carvalho de Melo Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Ingo Molnar Cc: Thomas Gleixner --- tools/perf/Documentation/perf-record.txt | 2 +- tools/perf/util/parse-events.c | 14 ++++---------- 2 files changed, 5 insertions(+), 11 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt index b29bd2db6a4..020d871c793 100644 --- a/tools/perf/Documentation/perf-record.txt +++ b/tools/perf/Documentation/perf-record.txt @@ -101,7 +101,7 @@ OPTIONS -R:: --raw-samples:: -Collect raw sample records from all opened counters (typically for tracepoint counters). +Collect raw sample records from all opened counters (default for tracepoint counters). SEE ALSO -------- diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 435781e0c20..880070c02fd 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -410,7 +410,6 @@ static enum event_result parse_single_tracepoint_event(char *sys_name, const char *evt_name, unsigned int evt_length, - char *flags, struct perf_event_attr *attr, const char **strp) { @@ -419,13 +418,9 @@ parse_single_tracepoint_event(char *sys_name, u64 id; int fd; - if (flags) { - if (!strncmp(flags, "record", strlen(flags))) { - attr->sample_type |= PERF_SAMPLE_RAW; - attr->sample_type |= PERF_SAMPLE_TIME; - attr->sample_type |= PERF_SAMPLE_CPU; - } - } + attr->sample_type |= PERF_SAMPLE_RAW; + attr->sample_type |= PERF_SAMPLE_TIME; + attr->sample_type |= PERF_SAMPLE_CPU; snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path, sys_name, evt_name); @@ -533,8 +528,7 @@ static enum event_result parse_tracepoint_event(const char **strp, flags); } else return parse_single_tracepoint_event(sys_name, evt_name, - evt_length, flags, - attr, strp); + evt_length, attr, strp); } static enum event_result -- cgit v1.2.3-70-g09d2 From f92128193094c288bc315db1694fafeaeb7ee1d0 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 14 Apr 2010 22:09:02 +0200 Subject: perf: Make the trace events sample period default to 1 Trace events are mostly used for tracing and then require not to be lost when possible. As opposite to hardware events that really require to trigger after a given sample period, trace events mostly need to trigger everytime. It is a frustrating experience to trace with perf and realize we lost a lot of events because we forgot the "-c 1" option. Then default sample_period to 1 for trace events but let the user override it. Signed-off-by: Frederic Weisbecker Cc: Arnaldo Carvalho de Melo Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Ingo Molnar Cc: Thomas Gleixner --- tools/perf/builtin-record.c | 36 ++++++++++++++++++++++-------------- tools/perf/util/parse-events.c | 2 ++ 2 files changed, 24 insertions(+), 14 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index dcda8993082..ca2affc9233 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -33,11 +33,13 @@ enum write_mode_t { static int *fd[MAX_NR_CPUS][MAX_COUNTERS]; +static unsigned int user_interval = UINT_MAX; static long default_interval = 0; static int nr_cpus = 0; static unsigned int page_size; static unsigned int mmap_pages = 128; +static unsigned int user_freq = UINT_MAX; static int freq = 1000; static int output; static const char *output_name = "perf.data"; @@ -255,10 +257,19 @@ static void create_counter(int counter, int cpu) if (nr_counters > 1) attr->sample_type |= PERF_SAMPLE_ID; - if (freq) { - attr->sample_type |= PERF_SAMPLE_PERIOD; - attr->freq = 1; - attr->sample_freq = freq; + /* + * We default some events to a 1 default interval. But keep + * it a weak assumption overridable by the user. + */ + if (!attr->sample_period || (user_freq != UINT_MAX && + user_interval != UINT_MAX)) { + if (freq) { + attr->sample_type |= PERF_SAMPLE_PERIOD; + attr->freq = 1; + attr->sample_freq = freq; + } else { + attr->sample_period = default_interval; + } } if (no_samples) @@ -689,13 +700,13 @@ static const struct option options[] = { "CPU to profile on"), OPT_BOOLEAN('f', "force", &force, "overwrite existing data file (deprecated)"), - OPT_LONG('c', "count", &default_interval, + OPT_LONG('c', "count", &user_interval, "event period to sample"), OPT_STRING('o', "output", &output_name, "file", "output file name"), OPT_BOOLEAN('i', "inherit", &inherit, "child tasks inherit counters"), - OPT_INTEGER('F', "freq", &freq, + OPT_INTEGER('F', "freq", &user_freq, "profile at this frequency"), OPT_INTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), @@ -716,7 +727,6 @@ static const struct option options[] = { int cmd_record(int argc, const char **argv, const char *prefix __used) { - int counter; int i,j; argc = parse_options(argc, argv, options, record_usage, @@ -774,6 +784,11 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) if (!event_array) return -ENOMEM; + if (user_interval != UINT_MAX) + default_interval = user_interval; + if (user_freq != UINT_MAX) + freq = user_freq; + /* * User specified count overrides default frequency. */ @@ -786,12 +801,5 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) exit(EXIT_FAILURE); } - for (counter = 0; counter < nr_counters; counter++) { - if (attrs[counter].sample_period) - continue; - - attrs[counter].sample_period = default_interval; - } - return __cmd_record(argc, argv); } diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 880070c02fd..3b4ec679756 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -422,6 +422,8 @@ parse_single_tracepoint_event(char *sys_name, attr->sample_type |= PERF_SAMPLE_TIME; attr->sample_type |= PERF_SAMPLE_CPU; + attr->sample_period = 1; + snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path, sys_name, evt_name); -- cgit v1.2.3-70-g09d2 From 5710fcad7c367adefe5634dc998f1f88780a8457 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Wed, 21 Apr 2010 18:06:01 +0200 Subject: perf: Fix initialization bug in parse_single_tracepoint_event() The parse_single_tracepoint_event() was setting some attributes before it validated the event was indeed a tracepoint event. This caused problems with other initialization routines like in the builtin-top.c module whereby sample_period is not set if not 0. Signed-off-by: Stephane Eranian Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Ingo Molnar LKML-Reference: <4bcf232b.698fd80a.6fbe.ffffb737@mx.google.com> Signed-off-by: Frederic Weisbecker --- tools/perf/util/parse-events.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 3b4ec679756..600d3271425 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -418,12 +418,6 @@ parse_single_tracepoint_event(char *sys_name, u64 id; int fd; - attr->sample_type |= PERF_SAMPLE_RAW; - attr->sample_type |= PERF_SAMPLE_TIME; - attr->sample_type |= PERF_SAMPLE_CPU; - - attr->sample_period = 1; - snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path, sys_name, evt_name); @@ -442,6 +436,13 @@ parse_single_tracepoint_event(char *sys_name, attr->type = PERF_TYPE_TRACEPOINT; *strp = evt_name + evt_length; + attr->sample_type |= PERF_SAMPLE_RAW; + attr->sample_type |= PERF_SAMPLE_TIME; + attr->sample_type |= PERF_SAMPLE_CPU; + + attr->sample_period = 1; + + return EVT_HANDLED; } -- cgit v1.2.3-70-g09d2 From 9e32a3cb0661a6a30e0fd2b77ce85293805e6337 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 5 May 2010 11:20:05 -0300 Subject: perf list: Add explanation about raw hardware event descriptors MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Using explanation given by Ingo Molnar in the oprofile mailing list. Suggested-by: Nick Black Cc: Frédéric Weisbecker Cc: Mike Galbraith Cc: Nick Black Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Documentation/perf-list.txt | 18 ++++++++++++++++++ tools/perf/util/parse-events.c | 2 +- 2 files changed, 19 insertions(+), 1 deletion(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/Documentation/perf-list.txt b/tools/perf/Documentation/perf-list.txt index 8290b942266..ad765e0b886 100644 --- a/tools/perf/Documentation/perf-list.txt +++ b/tools/perf/Documentation/perf-list.txt @@ -15,6 +15,24 @@ DESCRIPTION This command displays the symbolic event types which can be selected in the various perf commands with the -e option. +RAW HARDWARE EVENT DESCRIPTOR +----------------------------- +Even when an event is not available in a symbolic form within perf right now, +it can be encoded as , for instance, if the Intel docs +describe an event as: + + Event Umask Event Mask + Num. Value Mnemonic Description Comment + + A8H 01H LSD.UOPS Counts the number of micro-ops Use cmask=1 and + delivered by loop stream detector invert to count + cycles + +raw encoding of 0x1A8 can be used: + + perf stat -e r1a8 -a sleep 1 + perf record -e r1a8 ... + OPTIONS ------- None diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 600d3271425..bc8b7e61420 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -936,7 +936,7 @@ void print_events(void) printf("\n"); printf(" %-42s [%s]\n", - "rNNN", event_type_descriptors[PERF_TYPE_RAW]); + "rNNN (NNN=)", event_type_descriptors[PERF_TYPE_RAW]); printf("\n"); printf(" %-42s [%s]\n", -- cgit v1.2.3-70-g09d2 From ab608344bcbde4f55ec4cd911b686b0ce3eae076 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 8 Apr 2010 23:03:20 +0200 Subject: perf, x86: Improve the PEBS ABI Rename perf_event_attr::precise to perf_event_attr::precise_ip and widen it to 2 bits. This new field describes the required precision of the PERF_SAMPLE_IP field: 0 - SAMPLE_IP can have arbitrary skid 1 - SAMPLE_IP must have constant skid 2 - SAMPLE_IP requested to have 0 skid 3 - SAMPLE_IP must have 0 skid And modify the Intel PEBS code accordingly. The PEBS implementation now supports up to precise_ip == 2, where we perform the IP fixup. Also s/PERF_RECORD_MISC_EXACT/&_IP/ to clarify its meaning, this bit should be set for each PERF_SAMPLE_IP field known to match the actual instruction triggering the event. This new scheme allows for a PEBS mode that uses the buffer for more than a single event. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Stephane Eranian LKML-Reference: Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.c | 17 ++++++++++++++++- arch/x86/kernel/cpu/perf_event_intel.c | 4 ++-- arch/x86/kernel/cpu/perf_event_intel_ds.c | 12 ++++++------ include/linux/perf_event.h | 23 +++++++++++++++++++---- tools/perf/builtin-top.c | 2 +- tools/perf/util/parse-events.c | 25 ++++++++++++++++--------- 6 files changed, 60 insertions(+), 23 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 4a3f1f2b9b9..27fa9eeed02 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -488,6 +488,21 @@ static int x86_setup_perfctr(struct perf_event *event) static int x86_pmu_hw_config(struct perf_event *event) { + if (event->attr.precise_ip) { + int precise = 0; + + /* Support for constant skid */ + if (x86_pmu.pebs) + precise++; + + /* Support for IP fixup */ + if (x86_pmu.lbr_nr) + precise++; + + if (event->attr.precise_ip > precise) + return -EOPNOTSUPP; + } + /* * Generate PMC IRQs: * (keep 'enabled' bit clear for now) @@ -1780,7 +1795,7 @@ unsigned long perf_misc_flags(struct pt_regs *regs) } if (regs->flags & PERF_EFLAGS_EXACT) - misc |= PERF_RECORD_MISC_EXACT; + misc |= PERF_RECORD_MISC_EXACT_IP; return misc; } diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c index a4b56ac425c..fdbc652d3fe 100644 --- a/arch/x86/kernel/cpu/perf_event_intel.c +++ b/arch/x86/kernel/cpu/perf_event_intel.c @@ -563,7 +563,7 @@ static void intel_pmu_disable_event(struct perf_event *event) x86_pmu_disable_event(event); - if (unlikely(event->attr.precise)) + if (unlikely(event->attr.precise_ip)) intel_pmu_pebs_disable(event); } @@ -615,7 +615,7 @@ static void intel_pmu_enable_event(struct perf_event *event) return; } - if (unlikely(event->attr.precise)) + if (unlikely(event->attr.precise_ip)) intel_pmu_pebs_enable(event); __x86_pmu_enable_event(hwc, ARCH_PERFMON_EVENTSEL_ENABLE); diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c index 35056f715e9..18018d1311c 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_ds.c +++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c @@ -307,7 +307,7 @@ intel_pebs_constraints(struct perf_event *event) { struct event_constraint *c; - if (!event->attr.precise) + if (!event->attr.precise_ip) return NULL; if (x86_pmu.pebs_constraints) { @@ -330,7 +330,7 @@ static void intel_pmu_pebs_enable(struct perf_event *event) cpuc->pebs_enabled |= 1ULL << hwc->idx; WARN_ON_ONCE(cpuc->enabled); - if (x86_pmu.intel_cap.pebs_trap) + if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1) intel_pmu_lbr_enable(event); } @@ -345,7 +345,7 @@ static void intel_pmu_pebs_disable(struct perf_event *event) hwc->config |= ARCH_PERFMON_EVENTSEL_INT; - if (x86_pmu.intel_cap.pebs_trap) + if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1) intel_pmu_lbr_disable(event); } @@ -485,7 +485,7 @@ static void __intel_pmu_pebs_event(struct perf_event *event, regs.bp = pebs->bp; regs.sp = pebs->sp; - if (intel_pmu_pebs_fixup_ip(regs)) + if (event->attr.precise_ip > 1 && intel_pmu_pebs_fixup_ip(®s)) regs.flags |= PERF_EFLAGS_EXACT; else regs.flags &= ~PERF_EFLAGS_EXACT; @@ -518,7 +518,7 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs) WARN_ON_ONCE(!event); - if (!event->attr.precise) + if (!event->attr.precise_ip) return; n = top - at; @@ -570,7 +570,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs) WARN_ON_ONCE(!event); - if (!event->attr.precise) + if (!event->attr.precise_ip) continue; if (__test_and_set_bit(bit, (unsigned long *)&status)) diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 6be4a0f9137..23cd0057a68 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -203,9 +203,19 @@ struct perf_event_attr { enable_on_exec : 1, /* next exec enables */ task : 1, /* trace fork/exit */ watermark : 1, /* wakeup_watermark */ - precise : 1, /* OoO invariant counter */ - - __reserved_1 : 48; + /* + * precise_ip: + * + * 0 - SAMPLE_IP can have arbitrary skid + * 1 - SAMPLE_IP must have constant skid + * 2 - SAMPLE_IP requested to have 0 skid + * 3 - SAMPLE_IP must have 0 skid + * + * See also PERF_RECORD_MISC_EXACT_IP + */ + precise_ip : 2, /* skid constraint */ + + __reserved_1 : 47; union { __u32 wakeup_events; /* wakeup every n events */ @@ -296,7 +306,12 @@ struct perf_event_mmap_page { #define PERF_RECORD_MISC_GUEST_KERNEL (4 << 0) #define PERF_RECORD_MISC_GUEST_USER (5 << 0) -#define PERF_RECORD_MISC_EXACT (1 << 14) +/* + * Indicates that the content of PERF_SAMPLE_IP points to + * the actual instruction that triggered the event. See also + * perf_event_attr::precise_ip. + */ +#define PERF_RECORD_MISC_EXACT_IP (1 << 14) /* * Reserve the last bit to indicate some extended misc field */ diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 3de397764cb..ed9b5b6905f 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1021,7 +1021,7 @@ static void event__process_sample(const event_t *self, return; } - if (self->header.misc & PERF_RECORD_MISC_EXACT) + if (self->header.misc & PERF_RECORD_MISC_EXACT_IP) exact_samples++; if (event__preprocess_sample(self, session, &al, symbol_filter) < 0 || diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index bc8b7e61420..ae7f5917935 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -654,10 +654,6 @@ parse_raw_event(const char **strp, struct perf_event_attr *attr) return EVT_FAILED; n = hex2u64(str + 1, &config); if (n > 0) { - if (str[n+1] == 'p') { - attr->precise = 1; - n++; - } *strp = str + n + 1; attr->type = PERF_TYPE_RAW; attr->config = config; @@ -692,19 +688,29 @@ static enum event_result parse_event_modifier(const char **strp, struct perf_event_attr *attr) { const char *str = *strp; - int eu = 1, ek = 1, eh = 1; + int exclude = 0; + int eu = 0, ek = 0, eh = 0, precise = 0; if (*str++ != ':') return 0; while (*str) { - if (*str == 'u') + if (*str == 'u') { + if (!exclude) + exclude = eu = ek = eh = 1; eu = 0; - else if (*str == 'k') + } else if (*str == 'k') { + if (!exclude) + exclude = eu = ek = eh = 1; ek = 0; - else if (*str == 'h') + } else if (*str == 'h') { + if (!exclude) + exclude = eu = ek = eh = 1; eh = 0; - else + } else if (*str == 'p') { + precise++; + } else break; + ++str; } if (str >= *strp + 2) { @@ -712,6 +718,7 @@ parse_event_modifier(const char **strp, struct perf_event_attr *attr) attr->exclude_user = eu; attr->exclude_kernel = ek; attr->exclude_hv = eh; + attr->precise_ip = precise; return 1; } return 0; -- cgit v1.2.3-70-g09d2 From 1cf4a0632c24ea61162ed819bde358bc94c55510 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 7 May 2010 14:07:05 -0300 Subject: perf list: Improve the raw hw event descriptor documentation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It was x86 specific and imcomplete at that, improve the situation by making it clear where the example provided applies and by adding the URLs for the Intel and AMD manuals where this is discussed in depth. Acked-by: Robert Richter Cc: Cyrill Gorcunov Cc: Frédéric Weisbecker Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Tom Zanussi Cc: Robert Richter Reported-by: Robert Richter Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Documentation/perf-list.txt | 19 ++++++++++++++++--- tools/perf/util/parse-events.c | 3 ++- 2 files changed, 18 insertions(+), 4 deletions(-) (limited to 'tools/perf/util/parse-events.c') diff --git a/tools/perf/Documentation/perf-list.txt b/tools/perf/Documentation/perf-list.txt index ad765e0b886..43e3dd284b9 100644 --- a/tools/perf/Documentation/perf-list.txt +++ b/tools/perf/Documentation/perf-list.txt @@ -18,8 +18,16 @@ various perf commands with the -e option. RAW HARDWARE EVENT DESCRIPTOR ----------------------------- Even when an event is not available in a symbolic form within perf right now, -it can be encoded as , for instance, if the Intel docs -describe an event as: +it can be encoded in a per processor specific way. + +For instance For x86 CPUs NNN represents the raw register encoding with the +layout of IA32_PERFEVTSELx MSRs (see [Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide] Figure 30-1 Layout +of IA32_PERFEVTSELx MSRs) or AMD's PerfEvtSeln (see [AMD64 Architecture Programmer’s Manual Volume 2: System Programming], Page 344, +Figure 13-7 Performance Event-Select Register (PerfEvtSeln)). + +Example: + +If the Intel docs for a QM720 Core i7 describe an event as: Event Umask Event Mask Num. Value Mnemonic Description Comment @@ -33,6 +41,9 @@ raw encoding of 0x1A8 can be used: perf stat -e r1a8 -a sleep 1 perf record -e r1a8 ... +You should refer to the processor specific documentation for getting these +details. Some of them are referenced in the SEE ALSO section below. + OPTIONS ------- None @@ -40,4 +51,6 @@ None SEE ALSO -------- linkperf:perf-stat[1], linkperf:perf-top[1], -linkperf:perf-record[1] +linkperf:perf-record[1], +http://www.intel.com/Assets/PDF/manual/253669.pdf[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide], +http://support.amd.com/us/Processor_TechDocs/24593.pdf[AMD64 Architecture Programmer’s Manual Volume 2: System Programming] diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index bc8b7e61420..7b24b5dee8b 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -936,7 +936,8 @@ void print_events(void) printf("\n"); printf(" %-42s [%s]\n", - "rNNN (NNN=)", event_type_descriptors[PERF_TYPE_RAW]); + "rNNN (see 'perf list --help' on how to encode it)", + event_type_descriptors[PERF_TYPE_RAW]); printf("\n"); printf(" %-42s [%s]\n", -- cgit v1.2.3-70-g09d2