From f006d25a15216a483cec71e886786874f66f9452 Mon Sep 17 00:00:00 2001 From: Han Pingtian Date: Thu, 6 Jan 2011 17:39:22 +0800 Subject: [PATCH 01/28] perf tools: Fix buffer overflow error when specifying all tracepoints I found when specifying all tracepoints with -e to one of subcommand, such as 'stat', the program will trigger a buffer overflow error, like this: *** buffer overflow detected ***: ./perf terminated ======= Backtrace: ========= /lib64/libc.so.6(__fortify_fail+0x37)[0x382cefb2c7] .... The tracepoints are separated by comma, something like this: $ perf stat -a -e `perf list |grep Tracepoint|awk -F'[' '{gsub(/[[:space:]]+/,"",$1);array[FNR]=$1}END{outputs=array[1];for (i=2;i<=FNR;i++){ outputs=outputs "," array[i];};print outputs}'` The root reason of this problem is that store_event_type() is called for all events, and will overflow the 'filename' at: strncat(filename, orgname, strlen(orgname)); This patch fixes it by calling store_event_type() only when the event name has been found. LKML-Reference: <20110106093922.GB6713@hpt.nay.redhat.com> Signed-off-by: Han Pingtian Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/parse-events.c | 61 +++++++++++++++++----------------- 1 file changed, 30 insertions(+), 31 deletions(-) diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 649083f27e08..917a0ca521c1 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -490,6 +490,31 @@ parse_multiple_tracepoint_event(char *sys_name, const char *evt_exp, return EVT_HANDLED_ALL; } +static int store_event_type(const char *orgname) +{ + char filename[PATH_MAX], *c; + FILE *file; + int id, n; + + sprintf(filename, "%s/", debugfs_path); + strncat(filename, orgname, strlen(orgname)); + strcat(filename, "/id"); + + c = strchr(filename, ':'); + if (c) + *c = '/'; + + file = fopen(filename, "r"); + if (!file) + return 0; + n = fscanf(file, "%i", &id); + fclose(file); + if (n < 1) { + pr_err("cannot store event ID\n"); + return -EINVAL; + } + return perf_header__push_event(id, orgname); +} static enum event_result parse_tracepoint_event(const char **strp, struct perf_event_attr *attr) @@ -533,9 +558,13 @@ static enum event_result parse_tracepoint_event(const char **strp, *strp += strlen(sys_name) + evt_length; return parse_multiple_tracepoint_event(sys_name, evt_name, flags); - } else + } else { + if (store_event_type(evt_name) < 0) + return EVT_FAILED; + return parse_single_tracepoint_event(sys_name, evt_name, evt_length, attr, strp); + } } static enum event_result @@ -778,41 +807,11 @@ modifier: return ret; } -static int store_event_type(const char *orgname) -{ - char filename[PATH_MAX], *c; - FILE *file; - int id, n; - - sprintf(filename, "%s/", debugfs_path); - strncat(filename, orgname, strlen(orgname)); - strcat(filename, "/id"); - - c = strchr(filename, ':'); - if (c) - *c = '/'; - - file = fopen(filename, "r"); - if (!file) - return 0; - n = fscanf(file, "%i", &id); - fclose(file); - if (n < 1) { - pr_err("cannot store event ID\n"); - return -EINVAL; - } - return perf_header__push_event(id, orgname); -} - int parse_events(const struct option *opt __used, const char *str, int unset __used) { struct perf_event_attr attr; enum event_result ret; - if (strchr(str, ':')) - if (store_event_type(str) < 0) - return -1; - for (;;) { memset(&attr, 0, sizeof(attr)); ret = parse_event_symbols(&str, &attr); From 6b01f2c4f6188da50d8fe094e369a9c0390424ab Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Thu, 6 Jan 2011 15:51:53 +0100 Subject: [PATCH 02/28] perf tools: Build with frame pointer It seems that some gcc versions build by default with frame pointers and some others omit them. Just build the tools with frame pointers as the callchains can be an important part of the perf workflow. Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Stephane Eranian LKML-Reference: <1294325513-14276-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Makefile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/Makefile b/tools/perf/Makefile index 1b9b13ee2a72..2b5387d53ba5 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -227,7 +227,7 @@ ifndef PERF_DEBUG CFLAGS_OPTIMIZE = -O6 endif -CFLAGS = -ggdb3 -Wall -Wextra -std=gnu99 -Werror $(CFLAGS_OPTIMIZE) -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS) $(EXTRA_CFLAGS) +CFLAGS = -fno-omit-frame-pointer -ggdb3 -Wall -Wextra -std=gnu99 -Werror $(CFLAGS_OPTIMIZE) -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS) $(EXTRA_CFLAGS) EXTLIBS = -lpthread -lrt -lelf -lm ALL_CFLAGS = $(CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 ALL_LDFLAGS = $(LDFLAGS) From 23a2f3ab46596d9fd0b0e592d2101bea90970594 Mon Sep 17 00:00:00 2001 From: Lin Ming Date: Fri, 7 Jan 2011 11:11:09 +0800 Subject: [PATCH 03/28] 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/builtin-stat.c | 3 +-- tools/perf/builtin-test.c | 6 +++++- tools/perf/util/evsel.c | 5 ++--- tools/perf/util/evsel.h | 2 +- tools/perf/util/parse-events.c | 13 ++++++++++--- 5 files changed, 19 insertions(+), 10 deletions(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 02b2d8013a61..2dfcb613e66b 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -683,8 +683,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used) nr_counters = ARRAY_SIZE(default_attrs); for (c = 0; c < ARRAY_SIZE(default_attrs); ++c) { - pos = perf_evsel__new(default_attrs[c].type, - default_attrs[c].config, + pos = perf_evsel__new(&default_attrs[c], nr_counters); if (pos == NULL) goto out; diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 1c984342a579..e12753f976a1 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -264,6 +264,7 @@ static int test__open_syscall_event(void) int err = -1, fd; struct thread_map *threads; struct perf_evsel *evsel; + struct perf_event_attr attr; unsigned int nr_open_calls = 111, i; int id = trace_event__id("sys_enter_open"); @@ -278,7 +279,10 @@ static int test__open_syscall_event(void) return -1; } - evsel = perf_evsel__new(PERF_TYPE_TRACEPOINT, id, 0); + memset(&attr, 0, sizeof(attr)); + attr.type = PERF_TYPE_TRACEPOINT; + attr.config = id; + evsel = perf_evsel__new(&attr, 0); if (evsel == NULL) { pr_debug("perf_evsel__new\n"); goto out_thread_map_delete; diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index c95267e63c5b..1a5591d7a245 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); } diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index a0ccd69c3fc2..b2d755fe88a5 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -37,7 +37,7 @@ struct perf_evsel { struct cpu_map; struct thread_map; -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); void perf_evsel__delete(struct perf_evsel *evsel); int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads); diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 917a0ca521c1..5cb6f4bde905 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -823,7 +823,7 @@ int parse_events(const struct option *opt __used, const char *str, int unset __u if (ret != EVT_HANDLED_ALL) { struct perf_evsel *evsel; - evsel = perf_evsel__new(attr.type, attr.config, + evsel = perf_evsel__new(&attr, nr_counters); if (evsel == NULL) return -1; @@ -1013,8 +1013,15 @@ void print_events(void) int perf_evsel_list__create_default(void) { - struct perf_evsel *evsel = perf_evsel__new(PERF_TYPE_HARDWARE, - PERF_COUNT_HW_CPU_CYCLES, 0); + struct perf_evsel *evsel; + struct perf_event_attr attr; + + memset(&attr, 0, sizeof(attr)); + attr.type = PERF_TYPE_HARDWARE; + attr.config = PERF_COUNT_HW_CPU_CYCLES; + + evsel = perf_evsel__new(&attr, 0); + if (evsel == NULL) return -ENOMEM; From 0b3fcf178deefd7b64154c2c0760a2c63df0b74f Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Mon, 3 Jan 2011 18:20:01 +0200 Subject: [PATCH 04/28] perf_events: Move code around to prepare for cgroup In particular this patch move perf_event_exit_task() before cgroup_exit() to allow for cgroup support. The cgroup_exit() function detaches the cgroups attached to a task. Other movements include hoisting some definitions and inlines at the top of perf_event.c Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra LKML-Reference: <4d22058b.cdace30a.4657.ffff95b1@mx.google.com> Signed-off-by: Ingo Molnar --- kernel/exit.c | 14 +++++++++----- kernel/perf_event.c | 28 +++++++++++++++++----------- 2 files changed, 26 insertions(+), 16 deletions(-) diff --git a/kernel/exit.c b/kernel/exit.c index 676149a4ac5f..8cb89045ecf3 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -994,6 +994,15 @@ NORET_TYPE void do_exit(long code) exit_fs(tsk); check_stack_usage(); exit_thread(); + + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + * + * because of cgroup mode, must be called before cgroup_exit() + */ + perf_event_exit_task(tsk); + cgroup_exit(tsk, 1); if (group_dead) @@ -1007,11 +1016,6 @@ NORET_TYPE void do_exit(long code) * FIXME: do that only when needed, using sched_exit tracepoint */ flush_ptrace_hw_breakpoint(tsk); - /* - * Flush inherited counters to the parent - before the parent - * gets woken up by child-exit notifications. - */ - perf_event_exit_task(tsk); exit_notify(tsk, group_dead); #ifdef CONFIG_NUMA diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 11847bf1e8cc..2c14e3afdf0d 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -38,6 +38,12 @@ #include +enum event_type_t { + EVENT_FLEXIBLE = 0x1, + EVENT_PINNED = 0x2, + EVENT_ALL = EVENT_FLEXIBLE | EVENT_PINNED, +}; + atomic_t perf_task_events __read_mostly; static atomic_t nr_mmap_events __read_mostly; static atomic_t nr_comm_events __read_mostly; @@ -65,6 +71,12 @@ int sysctl_perf_event_sample_rate __read_mostly = 100000; static atomic64_t perf_event_id; +static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx, + enum event_type_t event_type); + +static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx, + enum event_type_t event_type); + void __weak perf_event_print_debug(void) { } extern __weak const char *perf_pmu_name(void) @@ -72,6 +84,11 @@ extern __weak const char *perf_pmu_name(void) return "pmu"; } +static inline u64 perf_clock(void) +{ + return local_clock(); +} + void perf_pmu_disable(struct pmu *pmu) { int *count = this_cpu_ptr(pmu->pmu_disable_count); @@ -240,11 +257,6 @@ static void perf_unpin_context(struct perf_event_context *ctx) put_ctx(ctx); } -static inline u64 perf_clock(void) -{ - return local_clock(); -} - /* * Update the record of the current time in a context. */ @@ -1193,12 +1205,6 @@ static int perf_event_refresh(struct perf_event *event, int refresh) return 0; } -enum event_type_t { - EVENT_FLEXIBLE = 0x1, - EVENT_PINNED = 0x2, - EVENT_ALL = EVENT_FLEXIBLE | EVENT_PINNED, -}; - static void ctx_sched_out(struct perf_event_context *ctx, struct perf_cpu_context *cpuctx, enum event_type_t event_type) From 5632ab12e9e1fcd7e94058567e181d8f35e83798 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Mon, 3 Jan 2011 18:20:01 +0200 Subject: [PATCH 05/28] perf_events: Generalize use of event_filter_match() Replace all occurrences of: event->cpu != -1 && event->cpu == smp_processor_id() by a call to: event_filter_match(event) This makes the code more consistent and will make the cgroup patch smaller. Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra LKML-Reference: <4d220593.2308e30a.48c5.ffff8ae9@mx.google.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 2c14e3afdf0d..dcdb19ed83a6 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -949,7 +949,7 @@ static void __perf_install_in_context(void *info) add_event_to_ctx(event, ctx); - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) goto unlock; /* @@ -1094,7 +1094,7 @@ static void __perf_event_enable(void *info) goto unlock; __perf_event_mark_enabled(event, ctx); - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) goto unlock; /* @@ -1441,7 +1441,7 @@ ctx_pinned_sched_in(struct perf_event_context *ctx, list_for_each_entry(event, &ctx->pinned_groups, group_entry) { if (event->state <= PERF_EVENT_STATE_OFF) continue; - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) continue; if (group_can_go_on(event, cpuctx, 1)) @@ -1473,7 +1473,7 @@ ctx_flexible_sched_in(struct perf_event_context *ctx, * Listen to the 'cpu' scheduling filter constraint * of events: */ - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) continue; if (group_can_go_on(event, cpuctx, can_add_hw)) { @@ -1700,7 +1700,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx, u64 period) if (event->state != PERF_EVENT_STATE_ACTIVE) continue; - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) continue; hwc = &event->hw; @@ -3899,7 +3899,7 @@ static int perf_event_task_match(struct perf_event *event) if (event->state < PERF_EVENT_STATE_INACTIVE) return 0; - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) return 0; if (event->attr.comm || event->attr.mmap || @@ -4036,7 +4036,7 @@ static int perf_event_comm_match(struct perf_event *event) if (event->state < PERF_EVENT_STATE_INACTIVE) return 0; - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) return 0; if (event->attr.comm) @@ -4184,7 +4184,7 @@ static int perf_event_mmap_match(struct perf_event *event, if (event->state < PERF_EVENT_STATE_INACTIVE) return 0; - if (event->cpu != -1 && event->cpu != smp_processor_id()) + if (!event_filter_match(event)) return 0; if ((!executable && event->attr.mmap_data) || From 4158755d3136f4cb05c1a8a260e9c06f93baeb48 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Mon, 3 Jan 2011 18:20:01 +0200 Subject: [PATCH 06/28] perf_events: Add perf_event_time() Adds perf_event_time() to try and centralize access to event timing and in particular ctx->time. Prepares for cgroup support. Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra LKML-Reference: <4d22059c.122ae30a.5e0e.ffff8b8b@mx.google.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 38 ++++++++++++++++++++++++-------------- 1 file changed, 24 insertions(+), 14 deletions(-) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index dcdb19ed83a6..b782b7a79f00 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -268,6 +268,12 @@ static void update_context_time(struct perf_event_context *ctx) ctx->timestamp = now; } +static u64 perf_event_time(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + return ctx ? ctx->time : 0; +} + /* * Update the total_time_enabled and total_time_running fields for a event. */ @@ -281,7 +287,7 @@ static void update_event_times(struct perf_event *event) return; if (ctx->is_active) - run_end = ctx->time; + run_end = perf_event_time(event); else run_end = event->tstamp_stopped; @@ -290,7 +296,7 @@ static void update_event_times(struct perf_event *event) if (event->state == PERF_EVENT_STATE_INACTIVE) run_end = event->tstamp_stopped; else - run_end = ctx->time; + run_end = perf_event_time(event); event->total_time_running = run_end - event->tstamp_running; } @@ -546,6 +552,7 @@ event_sched_out(struct perf_event *event, struct perf_cpu_context *cpuctx, struct perf_event_context *ctx) { + u64 tstamp = perf_event_time(event); u64 delta; /* * An event which could not be activated because of @@ -557,7 +564,7 @@ event_sched_out(struct perf_event *event, && !event_filter_match(event)) { delta = ctx->time - event->tstamp_stopped; event->tstamp_running += delta; - event->tstamp_stopped = ctx->time; + event->tstamp_stopped = tstamp; } if (event->state != PERF_EVENT_STATE_ACTIVE) @@ -568,7 +575,7 @@ event_sched_out(struct perf_event *event, event->pending_disable = 0; event->state = PERF_EVENT_STATE_OFF; } - event->tstamp_stopped = ctx->time; + event->tstamp_stopped = tstamp; event->pmu->del(event, 0); event->oncpu = -1; @@ -780,6 +787,8 @@ event_sched_in(struct perf_event *event, struct perf_cpu_context *cpuctx, struct perf_event_context *ctx) { + u64 tstamp = perf_event_time(event); + if (event->state <= PERF_EVENT_STATE_OFF) return 0; @@ -796,9 +805,9 @@ event_sched_in(struct perf_event *event, return -EAGAIN; } - event->tstamp_running += ctx->time - event->tstamp_stopped; + event->tstamp_running += tstamp - event->tstamp_stopped; - event->shadow_ctx_time = ctx->time - ctx->timestamp; + event->shadow_ctx_time = tstamp - ctx->timestamp; if (!is_software_event(event)) cpuctx->active_oncpu++; @@ -910,11 +919,13 @@ static int group_can_go_on(struct perf_event *event, static void add_event_to_ctx(struct perf_event *event, struct perf_event_context *ctx) { + u64 tstamp = perf_event_time(event); + list_add_event(event, ctx); perf_group_attach(event); - event->tstamp_enabled = ctx->time; - event->tstamp_running = ctx->time; - event->tstamp_stopped = ctx->time; + event->tstamp_enabled = tstamp; + event->tstamp_running = tstamp; + event->tstamp_stopped = tstamp; } /* @@ -1054,14 +1065,13 @@ static void __perf_event_mark_enabled(struct perf_event *event, struct perf_event_context *ctx) { struct perf_event *sub; + u64 tstamp = perf_event_time(event); event->state = PERF_EVENT_STATE_INACTIVE; - event->tstamp_enabled = ctx->time - event->total_time_enabled; + event->tstamp_enabled = tstamp - event->total_time_enabled; list_for_each_entry(sub, &event->sibling_list, group_entry) { - if (sub->state >= PERF_EVENT_STATE_INACTIVE) { - sub->tstamp_enabled = - ctx->time - sub->total_time_enabled; - } + if (sub->state >= PERF_EVENT_STATE_INACTIVE) + sub->tstamp_enabled = tstamp - sub->total_time_enabled; } } From 1c7b74d46fed530cca22a9a54140cdac2815c797 Mon Sep 17 00:00:00 2001 From: Huang Ying Date: Thu, 6 Jan 2011 16:18:47 -0500 Subject: [PATCH 07/28] x86, NMI: Add NMI symbol constants and rename memory parity to PCI SERR Replace the NMI related magic numbers with symbol constants. Memory parity error is only valid for IBM PC-AT, newer machine use bit 7 (0x80) of 0x61 port for PCI SERR. While memory error is usually reported via MCE. So corresponding function name and kernel log string is changed. But on some machines, PCI SERR line is still used to report memory errors. This is used by EDAC, so corresponding EDAC call is reserved. Signed-off-by: Huang Ying Signed-off-by: Don Zickus Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-2-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/mach_traps.h | 12 +++++++- arch/x86/kernel/traps.c | 51 ++++++++++++++++--------------- 2 files changed, 37 insertions(+), 26 deletions(-) diff --git a/arch/x86/include/asm/mach_traps.h b/arch/x86/include/asm/mach_traps.h index f7920601e472..72a8b52e7dfd 100644 --- a/arch/x86/include/asm/mach_traps.h +++ b/arch/x86/include/asm/mach_traps.h @@ -7,9 +7,19 @@ #include +#define NMI_REASON_PORT 0x61 + +#define NMI_REASON_SERR 0x80 +#define NMI_REASON_IOCHK 0x40 +#define NMI_REASON_MASK (NMI_REASON_SERR | NMI_REASON_IOCHK) + +#define NMI_REASON_CLEAR_SERR 0x04 +#define NMI_REASON_CLEAR_IOCHK 0x08 +#define NMI_REASON_CLEAR_MASK 0x0f + static inline unsigned char get_nmi_reason(void) { - return inb(0x61); + return inb(NMI_REASON_PORT); } static inline void reassert_nmi(void) diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c index c76aaca5694d..c7fd1cea0374 100644 --- a/arch/x86/kernel/traps.c +++ b/arch/x86/kernel/traps.c @@ -310,15 +310,15 @@ static int __init setup_unknown_nmi_panic(char *str) __setup("unknown_nmi_panic", setup_unknown_nmi_panic); static notrace __kprobes void -mem_parity_error(unsigned char reason, struct pt_regs *regs) +pci_serr_error(unsigned char reason, struct pt_regs *regs) { - printk(KERN_EMERG - "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", - reason, smp_processor_id()); - - printk(KERN_EMERG - "You have some hardware problem, likely on the PCI bus.\n"); + pr_emerg("NMI: PCI system error (SERR) for reason %02x on CPU %d.\n", + reason, smp_processor_id()); + /* + * On some machines, PCI SERR line is used to report memory + * errors. EDAC makes use of it. + */ #if defined(CONFIG_EDAC) if (edac_handler_set()) { edac_atomic_assert_error(); @@ -329,11 +329,11 @@ mem_parity_error(unsigned char reason, struct pt_regs *regs) if (panic_on_unrecovered_nmi) panic("NMI: Not continuing"); - printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); + pr_emerg("Dazed and confused, but trying to continue\n"); - /* Clear and disable the memory parity error line. */ - reason = (reason & 0xf) | 4; - outb(reason, 0x61); + /* Clear and disable the PCI SERR error line. */ + reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_SERR; + outb(reason, NMI_REASON_PORT); } static notrace __kprobes void @@ -341,15 +341,17 @@ io_check_error(unsigned char reason, struct pt_regs *regs) { unsigned long i; - printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n"); + pr_emerg( + "NMI: IOCK error (debug interrupt?) for reason %02x on CPU %d.\n", + reason, smp_processor_id()); show_registers(regs); if (panic_on_io_nmi) panic("NMI IOCK error: Not continuing"); /* Re-enable the IOCK line, wait for a few seconds */ - reason = (reason & 0xf) | 8; - outb(reason, 0x61); + reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK; + outb(reason, NMI_REASON_PORT); i = 20000; while (--i) { @@ -357,8 +359,8 @@ io_check_error(unsigned char reason, struct pt_regs *regs) udelay(100); } - reason &= ~8; - outb(reason, 0x61); + reason &= ~NMI_REASON_CLEAR_IOCHK; + outb(reason, NMI_REASON_PORT); } static notrace __kprobes void @@ -377,15 +379,14 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs) return; } #endif - printk(KERN_EMERG - "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", - reason, smp_processor_id()); + pr_emerg("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", + reason, smp_processor_id()); - printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n"); + pr_emerg("Do you have a strange power saving mode enabled?\n"); if (unknown_nmi_panic || panic_on_unrecovered_nmi) panic("NMI: Not continuing"); - printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); + pr_emerg("Dazed and confused, but trying to continue\n"); } static notrace __kprobes void default_do_nmi(struct pt_regs *regs) @@ -399,7 +400,7 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) if (!cpu) reason = get_nmi_reason(); - if (!(reason & 0xc0)) { + if (!(reason & NMI_REASON_MASK)) { if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) == NOTIFY_STOP) return; @@ -417,9 +418,9 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) return; /* AK: following checks seem to be broken on modern chipsets. FIXME */ - if (reason & 0x80) - mem_parity_error(reason, regs); - if (reason & 0x40) + if (reason & NMI_REASON_SERR) + pci_serr_error(reason, regs); + if (reason & NMI_REASON_IOCHK) io_check_error(reason, regs); #ifdef CONFIG_X86_32 /* From 673a6092ce5f5bec45619b7a7f89cfcf8bcf3c41 Mon Sep 17 00:00:00 2001 From: Don Zickus Date: Thu, 6 Jan 2011 16:18:48 -0500 Subject: [PATCH 08/28] x86: Convert some devices to use DIE_NMIUNKNOWN They are a handful of places in the code that register a die_notifier as a catch all in case no claims the NMI. Unfortunately, they trigger on events like DIE_NMI and DIE_NMI_IPI, which depending on when they registered may collide with other handlers that have the ability to determine if the NMI is theirs or not. The function unknown_nmi_error() makes one last effort to walk the die_chain when no one else has claimed the NMI before spitting out messages that the NMI is unknown. This is a better spot for these devices to execute any code without colliding with the other handlers. The two drivers modified are only compiled on x86 arches I believe, so they shouldn't be affected by other arches that may not have DIE_NMIUNKNOWN defined. Signed-off-by: Don Zickus Cc: Russ Anderson Cc: Corey Minyard Cc: openipmi-developer@lists.sourceforge.net Cc: dann frazier Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-3-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic/x2apic_uv_x.c | 2 +- drivers/char/ipmi/ipmi_watchdog.c | 2 +- drivers/watchdog/hpwdt.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/arch/x86/kernel/apic/x2apic_uv_x.c b/arch/x86/kernel/apic/x2apic_uv_x.c index c1c52c341f40..927902d90fe6 100644 --- a/arch/x86/kernel/apic/x2apic_uv_x.c +++ b/arch/x86/kernel/apic/x2apic_uv_x.c @@ -639,7 +639,7 @@ void __cpuinit uv_cpu_init(void) */ int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data) { - if (reason != DIE_NMI_IPI) + if (reason != DIE_NMIUNKNOWN) return NOTIFY_OK; if (in_crash_kexec) diff --git a/drivers/char/ipmi/ipmi_watchdog.c b/drivers/char/ipmi/ipmi_watchdog.c index f4d334f2536e..320668f4c3aa 100644 --- a/drivers/char/ipmi/ipmi_watchdog.c +++ b/drivers/char/ipmi/ipmi_watchdog.c @@ -1081,7 +1081,7 @@ ipmi_nmi(struct notifier_block *self, unsigned long val, void *data) { struct die_args *args = data; - if (val != DIE_NMI) + if (val != DIE_NMIUNKNOWN) return NOTIFY_OK; /* Hack, if it's a memory or I/O error, ignore it. */ diff --git a/drivers/watchdog/hpwdt.c b/drivers/watchdog/hpwdt.c index dea7b5bf6e2c..24b966d5061a 100644 --- a/drivers/watchdog/hpwdt.c +++ b/drivers/watchdog/hpwdt.c @@ -469,7 +469,7 @@ static int hpwdt_pretimeout(struct notifier_block *nb, unsigned long ulReason, unsigned long rom_pl; static int die_nmi_called; - if (ulReason != DIE_NMI && ulReason != DIE_NMI_IPI) + if (ulReason != DIE_NMIUNKNOWN) goto out; if (!hpwdt_nmi_decoding) From 166d751479c6d4e5b17dfc1f204a9c4397c9b3f1 Mon Sep 17 00:00:00 2001 From: Don Zickus Date: Thu, 6 Jan 2011 16:18:49 -0500 Subject: [PATCH 09/28] x86, NMI: Add priorities to handlers In order to consolidate the NMI die_chain events, we need to setup the priorities for the die notifiers. I started by defining a bunch of common priorities that can be used by the notifier blocks. Then I modified the notifier blocks to use the newly created priorities. Now that the priorities are straightened out, it should be easier to remove the event DIE_NMI_IPI. Signed-off-by: Don Zickus Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-4-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/nmi.h | 20 ++++++++++++++++++++ arch/x86/kernel/apic/hw_nmi.c | 2 +- arch/x86/kernel/cpu/mcheck/mce-inject.c | 2 +- arch/x86/kernel/cpu/perf_event.c | 2 +- arch/x86/kernel/kgdb.c | 3 ++- arch/x86/kernel/reboot.c | 2 ++ arch/x86/oprofile/nmi_int.c | 2 +- arch/x86/oprofile/nmi_timer_int.c | 2 +- 8 files changed, 29 insertions(+), 6 deletions(-) diff --git a/arch/x86/include/asm/nmi.h b/arch/x86/include/asm/nmi.h index c4021b953510..c76f5b92b840 100644 --- a/arch/x86/include/asm/nmi.h +++ b/arch/x86/include/asm/nmi.h @@ -23,6 +23,26 @@ void arch_trigger_all_cpu_backtrace(void); #define arch_trigger_all_cpu_backtrace arch_trigger_all_cpu_backtrace #endif +/* + * Define some priorities for the nmi notifier call chain. + * + * Create a local nmi bit that has a higher priority than + * external nmis, because the local ones are more frequent. + * + * Also setup some default high/normal/low settings for + * subsystems to registers with. Using 4 bits to seperate + * the priorities. This can go alot higher if needed be. + */ + +#define NMI_LOCAL_SHIFT 16 /* randomly picked */ +#define NMI_LOCAL_BIT (1ULL << NMI_LOCAL_SHIFT) +#define NMI_HIGH_PRIOR (1ULL << 8) +#define NMI_NORMAL_PRIOR (1ULL << 4) +#define NMI_LOW_PRIOR (1ULL << 0) +#define NMI_LOCAL_HIGH_PRIOR (NMI_LOCAL_BIT | NMI_HIGH_PRIOR) +#define NMI_LOCAL_NORMAL_PRIOR (NMI_LOCAL_BIT | NMI_NORMAL_PRIOR) +#define NMI_LOCAL_LOW_PRIOR (NMI_LOCAL_BIT | NMI_LOW_PRIOR) + void stop_nmi(void); void restart_nmi(void); diff --git a/arch/x86/kernel/apic/hw_nmi.c b/arch/x86/kernel/apic/hw_nmi.c index 72ec29e1ae06..8bc49f1ac7bc 100644 --- a/arch/x86/kernel/apic/hw_nmi.c +++ b/arch/x86/kernel/apic/hw_nmi.c @@ -96,7 +96,7 @@ arch_trigger_all_cpu_backtrace_handler(struct notifier_block *self, static __read_mostly struct notifier_block backtrace_notifier = { .notifier_call = arch_trigger_all_cpu_backtrace_handler, .next = NULL, - .priority = 1 + .priority = NMI_LOCAL_LOW_PRIOR, }; static int __init register_trigger_all_cpu_backtrace(void) diff --git a/arch/x86/kernel/cpu/mcheck/mce-inject.c b/arch/x86/kernel/cpu/mcheck/mce-inject.c index e7dbde7bfedb..59546c1219f9 100644 --- a/arch/x86/kernel/cpu/mcheck/mce-inject.c +++ b/arch/x86/kernel/cpu/mcheck/mce-inject.c @@ -95,7 +95,7 @@ static int mce_raise_notify(struct notifier_block *self, static struct notifier_block mce_raise_nb = { .notifier_call = mce_raise_notify, - .priority = 1000, + .priority = NMI_LOCAL_NORMAL_PRIOR, }; /* Inject mce on current CPU */ diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 0a360d146596..5e14b5e5fb81 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -1318,7 +1318,7 @@ perf_event_nmi_handler(struct notifier_block *self, static __read_mostly struct notifier_block perf_event_nmi_notifier = { .notifier_call = perf_event_nmi_handler, .next = NULL, - .priority = 1 + .priority = NMI_LOCAL_LOW_PRIOR, }; static struct event_constraint unconstrained; diff --git a/arch/x86/kernel/kgdb.c b/arch/x86/kernel/kgdb.c index cd21b654dec6..d43c84183d8f 100644 --- a/arch/x86/kernel/kgdb.c +++ b/arch/x86/kernel/kgdb.c @@ -48,6 +48,7 @@ #include #include #include +#include struct dbg_reg_def_t dbg_reg_def[DBG_MAX_REG_NUM] = { @@ -606,7 +607,7 @@ static struct notifier_block kgdb_notifier = { /* * Lowest-prio notifier priority, we want to be notified last: */ - .priority = -INT_MAX, + .priority = NMI_LOCAL_LOW_PRIOR, }; /** diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c index c495aa8d4815..9c1c83e4a742 100644 --- a/arch/x86/kernel/reboot.c +++ b/arch/x86/kernel/reboot.c @@ -778,6 +778,8 @@ static void smp_send_nmi_allbutself(void) static struct notifier_block crash_nmi_nb = { .notifier_call = crash_nmi_callback, + /* we want to be the first one called */ + .priority = NMI_LOCAL_HIGH_PRIOR+1, }; /* Halt all other CPUs, calling the specified function on each of them diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c index 358c8b9c96a7..6e84ea42085a 100644 --- a/arch/x86/oprofile/nmi_int.c +++ b/arch/x86/oprofile/nmi_int.c @@ -361,7 +361,7 @@ static void nmi_cpu_setup(void *dummy) static struct notifier_block profile_exceptions_nb = { .notifier_call = profile_exceptions_notify, .next = NULL, - .priority = 2 + .priority = NMI_LOCAL_LOW_PRIOR, }; static void nmi_cpu_restore_registers(struct op_msrs *msrs) diff --git a/arch/x86/oprofile/nmi_timer_int.c b/arch/x86/oprofile/nmi_timer_int.c index 0636dd93cef8..720bf5a53c51 100644 --- a/arch/x86/oprofile/nmi_timer_int.c +++ b/arch/x86/oprofile/nmi_timer_int.c @@ -38,7 +38,7 @@ static int profile_timer_exceptions_notify(struct notifier_block *self, static struct notifier_block profile_timer_exceptions_nb = { .notifier_call = profile_timer_exceptions_notify, .next = NULL, - .priority = 0 + .priority = NMI_LOW_PRIOR, }; static int timer_start(void) From c410b8307702c1e1f35be3fd868ad18e4ba0410f Mon Sep 17 00:00:00 2001 From: Don Zickus Date: Thu, 6 Jan 2011 16:18:50 -0500 Subject: [PATCH 10/28] x86, NMI: Remove DIE_NMI_IPI With priorities in place and no one really understanding the difference between DIE_NMI and DIE_NMI_IPI, just remove DIE_NMI_IPI and convert everyone to DIE_NMI. This also simplifies default_do_nmi() a little bit. Instead of calling the die_notifier in both the if and else part, just pull it out and call it before the if-statement. This has the side benefit of avoiding a call to the ioport to see if there is an external NMI sitting around until after the (more frequent) internal NMIs are dealt with. Patch-Inspired-by: Huang Ying Signed-off-by: Don Zickus Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-5-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/kdebug.h | 1 - arch/x86/kernel/apic/hw_nmi.c | 1 - arch/x86/kernel/cpu/mcheck/mce-inject.c | 3 ++- arch/x86/kernel/cpu/perf_event.c | 1 - arch/x86/kernel/kgdb.c | 4 ---- arch/x86/kernel/reboot.c | 3 ++- arch/x86/kernel/traps.c | 19 ++++++++----------- arch/x86/oprofile/nmi_int.c | 1 - 8 files changed, 12 insertions(+), 21 deletions(-) diff --git a/arch/x86/include/asm/kdebug.h b/arch/x86/include/asm/kdebug.h index f23eb2528464..ca242d35e873 100644 --- a/arch/x86/include/asm/kdebug.h +++ b/arch/x86/include/asm/kdebug.h @@ -18,7 +18,6 @@ enum die_val { DIE_TRAP, DIE_GPF, DIE_CALL, - DIE_NMI_IPI, DIE_PAGE_FAULT, DIE_NMIUNKNOWN, }; diff --git a/arch/x86/kernel/apic/hw_nmi.c b/arch/x86/kernel/apic/hw_nmi.c index 8bc49f1ac7bc..79fd43ca6f96 100644 --- a/arch/x86/kernel/apic/hw_nmi.c +++ b/arch/x86/kernel/apic/hw_nmi.c @@ -68,7 +68,6 @@ arch_trigger_all_cpu_backtrace_handler(struct notifier_block *self, switch (cmd) { case DIE_NMI: - case DIE_NMI_IPI: break; default: diff --git a/arch/x86/kernel/cpu/mcheck/mce-inject.c b/arch/x86/kernel/cpu/mcheck/mce-inject.c index 59546c1219f9..a77971979564 100644 --- a/arch/x86/kernel/cpu/mcheck/mce-inject.c +++ b/arch/x86/kernel/cpu/mcheck/mce-inject.c @@ -25,6 +25,7 @@ #include #include #include +#include /* Update fake mce registers on current CPU. */ static void inject_mce(struct mce *m) @@ -83,7 +84,7 @@ static int mce_raise_notify(struct notifier_block *self, struct die_args *args = (struct die_args *)data; int cpu = smp_processor_id(); struct mce *m = &__get_cpu_var(injectm); - if (val != DIE_NMI_IPI || !cpumask_test_cpu(cpu, mce_inject_cpumask)) + if (val != DIE_NMI || !cpumask_test_cpu(cpu, mce_inject_cpumask)) return NOTIFY_DONE; cpumask_clear_cpu(cpu, mce_inject_cpumask); if (m->inject_flags & MCJ_EXCEPTION) diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 5e14b5e5fb81..c71bae43a51b 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -1268,7 +1268,6 @@ perf_event_nmi_handler(struct notifier_block *self, switch (cmd) { case DIE_NMI: - case DIE_NMI_IPI: break; case DIE_NMIUNKNOWN: this_nmi = percpu_read(irq_stat.__nmi_count); diff --git a/arch/x86/kernel/kgdb.c b/arch/x86/kernel/kgdb.c index d43c84183d8f..a4130005028a 100644 --- a/arch/x86/kernel/kgdb.c +++ b/arch/x86/kernel/kgdb.c @@ -526,10 +526,6 @@ static int __kgdb_notify(struct die_args *args, unsigned long cmd) } return NOTIFY_DONE; - case DIE_NMI_IPI: - /* Just ignore, we will handle the roundup on DIE_NMI. */ - return NOTIFY_DONE; - case DIE_NMIUNKNOWN: if (was_in_debug_nmi[raw_smp_processor_id()]) { was_in_debug_nmi[raw_smp_processor_id()] = 0; diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c index 9c1c83e4a742..fc7aae1e2bc7 100644 --- a/arch/x86/kernel/reboot.c +++ b/arch/x86/kernel/reboot.c @@ -18,6 +18,7 @@ #include #include #include +#include #ifdef CONFIG_X86_32 # include @@ -747,7 +748,7 @@ static int crash_nmi_callback(struct notifier_block *self, { int cpu; - if (val != DIE_NMI_IPI) + if (val != DIE_NMI) return NOTIFY_OK; cpu = raw_smp_processor_id(); diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c index c7fd1cea0374..23f6ac05d04c 100644 --- a/arch/x86/kernel/traps.c +++ b/arch/x86/kernel/traps.c @@ -394,6 +394,14 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) unsigned char reason = 0; int cpu; + /* + * CPU-specific NMI must be processed before non-CPU-specific + * NMI, otherwise we may lose it, because the CPU-specific + * NMI can not be detected/processed on other CPUs. + */ + if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP) + return; + cpu = smp_processor_id(); /* Only the BSP gets external NMIs from the system. */ @@ -401,21 +409,10 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) reason = get_nmi_reason(); if (!(reason & NMI_REASON_MASK)) { - if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) - == NOTIFY_STOP) - return; - -#ifdef CONFIG_X86_LOCAL_APIC - if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) - == NOTIFY_STOP) - return; -#endif unknown_nmi_error(reason, regs); return; } - if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) - return; /* AK: following checks seem to be broken on modern chipsets. FIXME */ if (reason & NMI_REASON_SERR) diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c index 6e84ea42085a..e77ea0b566e0 100644 --- a/arch/x86/oprofile/nmi_int.c +++ b/arch/x86/oprofile/nmi_int.c @@ -65,7 +65,6 @@ static int profile_exceptions_notify(struct notifier_block *self, switch (val) { case DIE_NMI: - case DIE_NMI_IPI: if (ctr_running) model->check_ctrs(args->regs, &__get_cpu_var(cpu_msrs)); else if (!nmi_enabled) From ab846f13f69fa64f8ed69ce0c3e239e075910d23 Mon Sep 17 00:00:00 2001 From: Don Zickus Date: Thu, 6 Jan 2011 16:18:51 -0500 Subject: [PATCH 11/28] x86, NMI: Allow NMI reason io port (0x61) to be processed on any CPU In original NMI handler, NMI reason io port (0x61) is only processed on BSP. This makes it impossible to hot-remove BSP. To solve the issue, a raw spinlock is used to allow the port to be processed on any CPU. Originally-by: Huang Ying Signed-off-by: Don Zickus Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-6-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/traps.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c index 23f6ac05d04c..613b3d284a89 100644 --- a/arch/x86/kernel/traps.c +++ b/arch/x86/kernel/traps.c @@ -84,6 +84,11 @@ EXPORT_SYMBOL_GPL(used_vectors); static int ignore_nmis; int unknown_nmi_panic; +/* + * Prevent NMI reason port (0x61) being accessed simultaneously, can + * only be used in NMI handler. + */ +static DEFINE_RAW_SPINLOCK(nmi_reason_lock); static inline void conditional_sti(struct pt_regs *regs) { @@ -392,7 +397,6 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs) static notrace __kprobes void default_do_nmi(struct pt_regs *regs) { unsigned char reason = 0; - int cpu; /* * CPU-specific NMI must be processed before non-CPU-specific @@ -402,13 +406,12 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP) return; - cpu = smp_processor_id(); - - /* Only the BSP gets external NMIs from the system. */ - if (!cpu) - reason = get_nmi_reason(); + /* Non-CPU-specific NMI: NMI sources can be processed on any CPU */ + raw_spin_lock(&nmi_reason_lock); + reason = get_nmi_reason(); if (!(reason & NMI_REASON_MASK)) { + raw_spin_unlock(&nmi_reason_lock); unknown_nmi_error(reason, regs); return; @@ -426,6 +429,7 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) */ reassert_nmi(); #endif + raw_spin_unlock(&nmi_reason_lock); } dotraplinkage notrace __kprobes void From f2fd43954abc058586e95d4eb91e7a5477070704 Mon Sep 17 00:00:00 2001 From: Don Zickus Date: Thu, 6 Jan 2011 16:18:52 -0500 Subject: [PATCH 12/28] x86, NMI: Clean-up default_do_nmi() Just re-arrange the code a bit to make it easier to follow what is going on. Basically un-negating the if-statement and swapping the code inside the if-statement with code outside. No functional changes. Originally-by: Huang Ying Signed-off-by: Don Zickus Signed-off-by: Peter Zijlstra LKML-Reference: <1294348732-15030-7-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/traps.c | 30 ++++++++++++++---------------- 1 file changed, 14 insertions(+), 16 deletions(-) diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c index 613b3d284a89..b9b67166f9de 100644 --- a/arch/x86/kernel/traps.c +++ b/arch/x86/kernel/traps.c @@ -410,26 +410,24 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs) raw_spin_lock(&nmi_reason_lock); reason = get_nmi_reason(); - if (!(reason & NMI_REASON_MASK)) { + if (reason & NMI_REASON_MASK) { + if (reason & NMI_REASON_SERR) + pci_serr_error(reason, regs); + else if (reason & NMI_REASON_IOCHK) + io_check_error(reason, regs); +#ifdef CONFIG_X86_32 + /* + * Reassert NMI in case it became active + * meanwhile as it's edge-triggered: + */ + reassert_nmi(); +#endif raw_spin_unlock(&nmi_reason_lock); - unknown_nmi_error(reason, regs); - return; } - - /* AK: following checks seem to be broken on modern chipsets. FIXME */ - if (reason & NMI_REASON_SERR) - pci_serr_error(reason, regs); - if (reason & NMI_REASON_IOCHK) - io_check_error(reason, regs); -#ifdef CONFIG_X86_32 - /* - * Reassert NMI in case it became active meanwhile - * as it's edge-triggered: - */ - reassert_nmi(); -#endif raw_spin_unlock(&nmi_reason_lock); + + unknown_nmi_error(reason, regs); } dotraplinkage notrace __kprobes void From 39a6eebda253aa34d659ca9436e3c32ef60473f1 Mon Sep 17 00:00:00 2001 From: Rakib Mullick Date: Fri, 7 Jan 2011 16:59:49 +0100 Subject: [PATCH 13/28] x86, dumpstack: Fix unused variable warning In dump_stack function, bp isn't used anymore, which is introduced by commit 9c0729dc8062bed96189bd14ac6d4920f3958743. This patch removes bp completely. Signed-off-by: Rakib Mullick Cc: Soeren Sandmann Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: Ingo Molnar Cc: H. Peter Anvin LKML-Reference: Signed-off-by: Frederic Weisbecker --- arch/x86/kernel/dumpstack.c | 6 ------ 1 file changed, 6 deletions(-) diff --git a/arch/x86/kernel/dumpstack.c b/arch/x86/kernel/dumpstack.c index 8474c998cbd4..d6fb146c0d8b 100644 --- a/arch/x86/kernel/dumpstack.c +++ b/arch/x86/kernel/dumpstack.c @@ -197,14 +197,8 @@ void show_stack(struct task_struct *task, unsigned long *sp) */ void dump_stack(void) { - unsigned long bp = 0; unsigned long stack; -#ifdef CONFIG_FRAME_POINTER - if (!bp) - get_bp(bp); -#endif - printk("Pid: %d, comm: %.20s %s %s %.*s\n", current->pid, current->comm, print_tainted(), init_utsname()->release, From 625dbc3b8acbefefefe27e1d7bbc6e53eb4f3f2d Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Thu, 6 Jan 2011 15:22:47 +0100 Subject: [PATCH 14/28] x86: Save rbp in pt_regs on irq entry From the x86_64 low level interrupt handlers, the frame pointer is saved right after the partial pt_regs frame. rbp is not supposed to be part of the irq partial saved registers, but it only requires to extend the pt_regs frame by 8 bytes to do so, plus a tiny stack offset fixup on irq exit. This changes a bit the semantics or get_irq_entry() that is supposed to provide only the value of caller saved registers and the cpu saved frame. However it's a win for unwinders that can walk through stack frames on top of get_irq_regs() snapshots. A noticeable impact is that it makes perf events cpu-clock and task-clock events based callchains working on x86_64. Let's then save rbp into the irq pt_regs. As a result with: perf record -e cpu-clock perf bench sched messaging perf report --stdio Before: 20.94% perf [kernel.kallsyms] [k] lock_acquire | --- lock_acquire | |--44.01%-- __write_nocancel | |--43.18%-- __read | |--6.08%-- fork | create_worker | |--0.88%-- _dl_fixup | |--0.65%-- do_lookup_x | |--0.53%-- __GI___libc_read --4.67%-- [...] After: 19.23% perf [kernel.kallsyms] [k] __lock_acquire | --- __lock_acquire | |--97.74%-- lock_acquire | | | |--21.82%-- _raw_spin_lock | | | | | |--37.26%-- unix_stream_recvmsg | | | sock_aio_read | | | do_sync_read | | | vfs_read | | | sys_read | | | system_call | | | __read | | | | | |--24.09%-- unix_stream_sendmsg | | | sock_aio_write | | | do_sync_write | | | vfs_write | | | sys_write | | | system_call | | | __write_nocancel v2: Fix cfi annotations. Reported-by: Soeren Sandmann Pedersen Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Thomas Gleixner Cc: H. Peter Anvin Cc: Arnaldo Carvalho de Melo Cc: Stephane Eranian Cc: Jan Beulich --- arch/x86/kernel/entry_64.S | 34 ++++++++++++++++++++++------------ 1 file changed, 22 insertions(+), 12 deletions(-) diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S index e3ba417e8697..d3b895f375d3 100644 --- a/arch/x86/kernel/entry_64.S +++ b/arch/x86/kernel/entry_64.S @@ -299,17 +299,21 @@ ENDPROC(native_usergs_sysret64) ENTRY(save_args) XCPT_FRAME cld - movq_cfi rdi, RDI+16-ARGOFFSET - movq_cfi rsi, RSI+16-ARGOFFSET - movq_cfi rdx, RDX+16-ARGOFFSET - movq_cfi rcx, RCX+16-ARGOFFSET - movq_cfi rax, RAX+16-ARGOFFSET - movq_cfi r8, R8+16-ARGOFFSET - movq_cfi r9, R9+16-ARGOFFSET - movq_cfi r10, R10+16-ARGOFFSET - movq_cfi r11, R11+16-ARGOFFSET + /* + * start from rbp in pt_regs and jump over + * return address. + */ + movq_cfi rdi, RDI+8-RBP + movq_cfi rsi, RSI+8-RBP + movq_cfi rdx, RDX+8-RBP + movq_cfi rcx, RCX+8-RBP + movq_cfi rax, RAX+8-RBP + movq_cfi r8, R8+8-RBP + movq_cfi r9, R9+8-RBP + movq_cfi r10, R10+8-RBP + movq_cfi r11, R11+8-RBP - leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */ + leaq -RBP+8(%rsp),%rdi /* arg1 for handler */ movq_cfi rbp, 8 /* push %rbp */ leaq 8(%rsp), %rbp /* mov %rsp, %ebp */ testl $3, CS(%rdi) @@ -782,8 +786,9 @@ END(interrupt) /* 0(%rsp): ~(interrupt number) */ .macro interrupt func - subq $ORIG_RAX-ARGOFFSET+8, %rsp - CFI_ADJUST_CFA_OFFSET ORIG_RAX-ARGOFFSET+8 + /* reserve pt_regs for scratch regs and rbp */ + subq $ORIG_RAX-RBP, %rsp + CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP call save_args PARTIAL_FRAME 0 call \func @@ -808,9 +813,14 @@ ret_from_intr: TRACE_IRQS_OFF decl PER_CPU_VAR(irq_count) leaveq + CFI_RESTORE rbp CFI_DEF_CFA_REGISTER rsp CFI_ADJUST_CFA_OFFSET -8 + + /* we did not save rbx, restore only from ARGOFFSET */ + addq $8, %rsp + CFI_ADJUST_CFA_OFFSET -8 exit_intr: GET_THREAD_INFO(%rcx) testl $3,CS-ARGOFFSET(%rsp) From 3a9f987b3141f086de27832514aad9f50a53f754 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Fri, 7 Jan 2011 15:40:10 -0500 Subject: [PATCH 15/28] tracing: Include module.h in define_trace.h While doing some developing, Peter Zijlstra and I have found that if a CREATE_TRACE_POINTS include is done before module.h is included, it can break the build. We have been lucky so far that this has not broke the build since module.h is included in almost everything. Reported-by: Peter Zijlstra Signed-off-by: Steven Rostedt --- include/trace/define_trace.h | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/include/trace/define_trace.h b/include/trace/define_trace.h index b0b4eb24d592..da39b22636f7 100644 --- a/include/trace/define_trace.h +++ b/include/trace/define_trace.h @@ -21,6 +21,16 @@ #undef CREATE_TRACE_POINTS #include +/* + * module.h includes tracepoints, and because ftrace.h + * pulls in module.h: + * trace/ftrace.h -> linux/ftrace_event.h -> linux/perf_event.h -> + * linux/ftrace.h -> linux/module.h + * we must include module.h here before we play with any of + * the TRACE_EVENT() macros, otherwise the tracepoints included + * by module.h may break the build. + */ +#include #undef TRACE_EVENT #define TRACE_EVENT(name, proto, args, tstruct, assign, print) \ From ec6e7c3ae39cb1dc279b5274aaaadd09ff8e224b Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Thu, 6 Jan 2011 13:45:32 -0500 Subject: [PATCH 16/28] tracing/trivial: Add missing comma in TRACE_EVENT comment Add missing comma within the TRACE_EVENT() example in tracepoint.h. Signed-off-by: Mathieu Desnoyers LKML-Reference: <20110106184532.GA2526@Krystal> CC: Frederic Weisbecker CC: Ingo Molnar CC: Thomas Gleixner Signed-off-by: Steven Rostedt --- include/linux/tracepoint.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h index d3e4f87e95c0..899103c8f6f2 100644 --- a/include/linux/tracepoint.h +++ b/include/linux/tracepoint.h @@ -326,7 +326,7 @@ do_trace: \ * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); * __entry->next_pid = next->pid; * __entry->next_prio = next->prio; - * ) + * ), * * * * * Formatted output of a trace record via TP_printk(). From 881fe4bdcdc899674524e30a76c76d298b447008 Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Thu, 6 Jan 2011 12:53:19 -0500 Subject: [PATCH 17/28] tracing: remove duplicate null-pointer check in skb tracepoint The check for NULL skb in the kfree_skb trace event is a duplicate from the check already done in its only caller, kfree_skb(). Remove this duplicate check. Signed-off-by: Mathieu Desnoyers LKML-Reference: <20110106175319.GA30610@Krystal> Acked-by: Neil Horman Acked-by: David S. Miller CC: Steven Rostedt CC: Frederic Weisbecker CC: Ingo Molnar CC: Thomas Gleixner CC: Zhaolei Signed-off-by: Steven Rostedt --- include/trace/events/skb.h | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/include/trace/events/skb.h b/include/trace/events/skb.h index 75ce9d500d8e..f10293c41b1e 100644 --- a/include/trace/events/skb.h +++ b/include/trace/events/skb.h @@ -25,9 +25,7 @@ TRACE_EVENT(kfree_skb, TP_fast_assign( __entry->skbaddr = skb; - if (skb) { - __entry->protocol = ntohs(skb->protocol); - } + __entry->protocol = ntohs(skb->protocol); __entry->location = location; ), From bd1c8b22b7b81c6f6c4f5c19cb2387da3d02fb0f Mon Sep 17 00:00:00 2001 From: Lai Jiangshan Date: Tue, 4 Jan 2011 16:06:09 +0800 Subject: [PATCH 18/28] tracepoint: Add __rcu annotation Add __rcu annotation to : (struct tracepoint)->funcs Acked-by: Mathieu Desnoyers Signed-off-by: Lai Jiangshan LKML-Reference: <4D22D4F1.50505@cn.fujitsu.com> Signed-off-by: Steven Rostedt --- include/linux/tracepoint.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h index 899103c8f6f2..c6814616653b 100644 --- a/include/linux/tracepoint.h +++ b/include/linux/tracepoint.h @@ -32,7 +32,7 @@ struct tracepoint { int state; /* State. */ void (*regfunc)(void); void (*unregfunc)(void); - struct tracepoint_func *funcs; + struct tracepoint_func __rcu *funcs; } __attribute__((aligned(32))); /* * Aligned on 32 bytes because it is * globally visible and gcc happily From 1dbd1951f39e13da579ffe879cce19586d0462de Mon Sep 17 00:00:00 2001 From: Li Zefan Date: Thu, 9 Dec 2010 15:47:56 +0800 Subject: [PATCH 19/28] tracing: Fix preempt count leak While running my ftrace stress test, this showed up: BUG: sleeping function called from invalid context at mm/mmap.c:233 ... note: cat[3293] exited with preempt_count 1 The bug was introduced by commit 91e86e560d0b3ce4c5fc64fd2bbb99f856a30a4e ("tracing: Fix recursive user stack trace") Cc: Signed-off-by: Li Zefan LKML-Reference: <4D0089AC.1020802@cn.fujitsu.com> Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index f8cf959bad45..dc53ecb80589 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -1313,12 +1313,10 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc) __this_cpu_inc(user_stack_count); - - event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK, sizeof(*entry), flags, pc); if (!event) - return; + goto out_drop_count; entry = ring_buffer_event_data(event); entry->tgid = current->tgid; @@ -1333,8 +1331,8 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc) if (!filter_check_discard(call, entry, buffer, event)) ring_buffer_unlock_commit(buffer, event); + out_drop_count: __this_cpu_dec(user_stack_count); - out: preempt_enable(); } From 870915e047a2da695be118d32dd5a900f0c3e933 Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Thu, 28 Oct 2010 11:31:17 -0400 Subject: [PATCH 20/28] tracing: Fix TRACE_EVENT power tracepoint creation DEFINE_TRACE should also exist when CONFIG_EVENT_TRACING=n. Otherwise, setting only TRACEPOINTS=y is broken. Acked-by: Arjan van de Ven Signed-off-by: Mathieu Desnoyers LKML-Reference: <20101028153117.GA4051@Krystal> Signed-off-by: Steven Rostedt --- kernel/Makefile | 1 + kernel/trace/Makefile | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/kernel/Makefile b/kernel/Makefile index 0b5ff083fa22..e0f2831634b4 100644 --- a/kernel/Makefile +++ b/kernel/Makefile @@ -100,6 +100,7 @@ obj-$(CONFIG_FUNCTION_TRACER) += trace/ obj-$(CONFIG_TRACING) += trace/ obj-$(CONFIG_X86_DS) += trace/ obj-$(CONFIG_RING_BUFFER) += trace/ +obj-$(CONFIG_TRACEPOINTS) += trace/ obj-$(CONFIG_SMP) += sched_cpupri.o obj-$(CONFIG_IRQ_WORK) += irq_work.o obj-$(CONFIG_PERF_EVENTS) += perf_event.o diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile index 53f338190b26..761c510a06c5 100644 --- a/kernel/trace/Makefile +++ b/kernel/trace/Makefile @@ -52,7 +52,7 @@ obj-$(CONFIG_EVENT_TRACING) += trace_event_perf.o endif obj-$(CONFIG_EVENT_TRACING) += trace_events_filter.o obj-$(CONFIG_KPROBE_EVENT) += trace_kprobe.o -obj-$(CONFIG_EVENT_TRACING) += power-traces.o +obj-$(CONFIG_TRACEPOINTS) += power-traces.o ifeq ($(CONFIG_TRACING),y) obj-$(CONFIG_KGDB_KDB) += trace_kdb.o endif From 2d75af2f2a7a6103a6d539a492fe81deacabde44 Mon Sep 17 00:00:00 2001 From: Jason Baron Date: Fri, 7 Jan 2011 13:36:58 -0500 Subject: [PATCH 21/28] dynamic debug: Fix build issue with older gcc On older gcc (3.3) dynamic debug fails to compile: include/net/inet_connection_sock.h: In function `inet_csk_reset_xmit_timer': include/net/inet_connection_sock.h:236: error: duplicate label declaration `do_printk' include/net/inet_connection_sock.h:219: error: this is a previous declaration include/net/inet_connection_sock.h:236: error: duplicate label declaration `out' include/net/inet_connection_sock.h:219: error: this is a previous declaration include/net/inet_connection_sock.h:236: error: duplicate label `do_printk' include/net/inet_connection_sock.h:236: error: duplicate label `out' Fix, by reverting the usage of JUMP_LABEL() in dynamic debug for now. Cc: Reported-by: Tetsuo Handa Tested-by: Tetsuo Handa Signed-off-by: Jason Baron Signed-off-by: Steven Rostedt --- include/linux/dynamic_debug.h | 18 ++++-------------- lib/dynamic_debug.c | 9 ++++----- 2 files changed, 8 insertions(+), 19 deletions(-) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index a90b3892074a..1c70028f81f9 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -44,34 +44,24 @@ int ddebug_add_module(struct _ddebug *tab, unsigned int n, extern int ddebug_remove_module(const char *mod_name); #define dynamic_pr_debug(fmt, ...) do { \ - __label__ do_printk; \ - __label__ out; \ static struct _ddebug descriptor \ __used \ __attribute__((section("__verbose"), aligned(8))) = \ { KBUILD_MODNAME, __func__, __FILE__, fmt, __LINE__, \ _DPRINTK_FLAGS_DEFAULT }; \ - JUMP_LABEL(&descriptor.enabled, do_printk); \ - goto out; \ -do_printk: \ - printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__); \ -out: ; \ + if (unlikely(descriptor.enabled)) \ + printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__); \ } while (0) #define dynamic_dev_dbg(dev, fmt, ...) do { \ - __label__ do_printk; \ - __label__ out; \ static struct _ddebug descriptor \ __used \ __attribute__((section("__verbose"), aligned(8))) = \ { KBUILD_MODNAME, __func__, __FILE__, fmt, __LINE__, \ _DPRINTK_FLAGS_DEFAULT }; \ - JUMP_LABEL(&descriptor.enabled, do_printk); \ - goto out; \ -do_printk: \ - dev_printk(KERN_DEBUG, dev, fmt, ##__VA_ARGS__); \ -out: ; \ + if (unlikely(descriptor.enabled)) \ + dev_printk(KERN_DEBUG, dev, fmt, ##__VA_ARGS__); \ } while (0) #else diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index 3094318bfea7..b335acb43be2 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -141,11 +141,10 @@ static void ddebug_change(const struct ddebug_query *query, else if (!dp->flags) dt->num_enabled++; dp->flags = newflags; - if (newflags) { - jump_label_enable(&dp->enabled); - } else { - jump_label_disable(&dp->enabled); - } + if (newflags) + dp->enabled = 1; + else + dp->enabled = 0; if (verbose) printk(KERN_INFO "ddebug: changed %s:%d [%s]%s %s\n", From 047a3772feaae8e43d81d790f3d3f80dae8ae676 Mon Sep 17 00:00:00 2001 From: Cyrill Gorcunov Date: Fri, 7 Jan 2011 21:42:06 +0300 Subject: [PATCH 22/28] perf, x86: P4 PMU - Fix unflagged overflows handling Don found that P4 PMU reads CCCR register instead of counter itself (in attempt to catch unflagged event) this makes P4 NMI handler to consume all NMIs it observes. So the other NMI users such as kgdb simply have no chance to get NMI on their hands. Side note: at moment there is no way to run nmi-watchdog together with perf tool. This is because both 'perf top' and nmi-watchdog use same event. So while nmi-watchdog reserves one event/counter for own needs there is no room for perf tool left (there is a way to disable nmi-watchdog on boot of course). Ming has tested this patch with the following results | 1. watchdog disabled | | kgdb tests on boot OK | perf works OK | | 2. watchdog enabled, without patch perf-x86-p4-nmi-4 | | kgdb tests on boot hang | | 3. watchdog enabled, without patch perf-x86-p4-nmi-4 and do not run kgdb | tests on boot | | "perf top" partialy works | cpu-cycles no | instructions yes | cache-references no | cache-misses no | branch-instructions no | branch-misses yes | bus-cycles no | | 4. watchdog enabled, with patch perf-x86-p4-nmi-4 applied | | kgdb tests on boot OK | perf does not work, NMI "Dazed and confused" messages show up | Which means we still have problems with p4 box due to 'unknown' nmi happens but at least it should fix kgdb test cases. Reported-by: Jason Wessel Reported-by: Don Zickus Signed-off-by: Cyrill Gorcunov Acked-by: Don Zickus Acked-by: Lin Ming Cc: Stephane Eranian Cc: Peter Zijlstra LKML-Reference: <4D275E7E.3040903@gmail.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/perf_event_p4.h | 3 +++ arch/x86/kernel/cpu/perf_event_p4.c | 28 +++++++++++++++------------- 2 files changed, 18 insertions(+), 13 deletions(-) diff --git a/arch/x86/include/asm/perf_event_p4.h b/arch/x86/include/asm/perf_event_p4.h index 295e2ff18a6a..e2f6a99f14ab 100644 --- a/arch/x86/include/asm/perf_event_p4.h +++ b/arch/x86/include/asm/perf_event_p4.h @@ -20,6 +20,9 @@ #define ARCH_P4_MAX_ESCR (ARCH_P4_TOTAL_ESCR - ARCH_P4_RESERVED_ESCR) #define ARCH_P4_MAX_CCCR (18) +#define ARCH_P4_CNTRVAL_BITS (40) +#define ARCH_P4_CNTRVAL_MASK ((1ULL << ARCH_P4_CNTRVAL_BITS) - 1) + #define P4_ESCR_EVENT_MASK 0x7e000000U #define P4_ESCR_EVENT_SHIFT 25 #define P4_ESCR_EVENTMASK_MASK 0x01fffe00U diff --git a/arch/x86/kernel/cpu/perf_event_p4.c b/arch/x86/kernel/cpu/perf_event_p4.c index 81400b93e694..e56b9bfbabd1 100644 --- a/arch/x86/kernel/cpu/perf_event_p4.c +++ b/arch/x86/kernel/cpu/perf_event_p4.c @@ -753,19 +753,21 @@ out: static inline int p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc) { - int overflow = 0; - u32 low, high; + u64 v; - rdmsr(hwc->config_base + hwc->idx, low, high); - - /* we need to check high bit for unflagged overflows */ - if ((low & P4_CCCR_OVF) || !(high & (1 << 31))) { - overflow = 1; - (void)checking_wrmsrl(hwc->config_base + hwc->idx, - ((u64)low) & ~P4_CCCR_OVF); + /* an official way for overflow indication */ + rdmsrl(hwc->config_base + hwc->idx, v); + if (v & P4_CCCR_OVF) { + wrmsrl(hwc->config_base + hwc->idx, v & ~P4_CCCR_OVF); + return 1; } - return overflow; + /* it might be unflagged overflow */ + rdmsrl(hwc->event_base + hwc->idx, v); + if (!(v & ARCH_P4_CNTRVAL_MASK)) + return 1; + + return 0; } static void p4_pmu_disable_pebs(void) @@ -1152,9 +1154,9 @@ static __initconst const struct x86_pmu p4_pmu = { */ .num_counters = ARCH_P4_MAX_CCCR, .apic = 1, - .cntval_bits = 40, - .cntval_mask = (1ULL << 40) - 1, - .max_period = (1ULL << 39) - 1, + .cntval_bits = ARCH_P4_CNTRVAL_BITS, + .cntval_mask = ARCH_P4_CNTRVAL_MASK, + .max_period = (1ULL << (ARCH_P4_CNTRVAL_BITS - 1)) - 1, .hw_config = p4_hw_config, .schedule_events = p4_pmu_schedule_events, /* From e462dc553ea5e09d4713e7c35a11ed331dc6f369 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 10 Jan 2011 10:48:47 -0200 Subject: [PATCH 23/28] perf sched: Fix allocation result check Bug introduced in ce47dc56. Reported-by: Mike Galbraith Cc: Chris Samuel 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/builtin-sched.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c index 7a4ebeb8b016..54024d2a017e 100644 --- a/tools/perf/builtin-sched.c +++ b/tools/perf/builtin-sched.c @@ -1861,7 +1861,7 @@ static int __cmd_record(int argc, const char **argv) rec_argc = ARRAY_SIZE(record_args) + argc - 1; rec_argv = calloc(rec_argc + 1, sizeof(char *)); - if (rec_argv) + if (rec_argv == NULL) return -ENOMEM; for (i = 0; i < ARRAY_SIZE(record_args); i++) From 5a3446bc648d86a99a895e93d28d7df00bbbfb3c Mon Sep 17 00:00:00 2001 From: David Ahern Date: Sun, 9 Jan 2011 11:21:12 -0700 Subject: [PATCH 24/28] perf stat: better error message for unsupported events For unsupported events (e.g., H/W events when running in a VM) perf stat currently fails with the error message: Error: open_counter returned with 2 (No such file or directory). /bin/dmesg may provide additional information. Fatal: Not all events could be opened. dmesg is of no help and it is not clear as to why it fails to open the counter. This patch changes the error message to Error: cache-misses event is not supported. Fatal: Not all events could be opened. Cc: Ingo Molnar Cc: Paul Mackerras Cc: a.p.zijlstra@chello.nl LPU-Reference: <1294597272-17335-1-git-send-email-daahern@cisco.com> Signed-off-by: David Ahern Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 2dfcb613e66b..c385a63ebfd1 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -316,6 +316,8 @@ static int run_perf_stat(int argc __used, const char **argv) "\t Consider tweaking" " /proc/sys/kernel/perf_event_paranoid or running as root.", system_wide ? "system-wide " : ""); + } else if (errno == ENOENT) { + error("%s event is not supported. ", event_name(counter)); } else { error("open_counter returned with %d (%s). " "/bin/dmesg may provide additional information.\n", From aa7bc7ef73efc46d7c3a0e185eefaf85744aec98 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 10 Jan 2011 13:36:24 -0200 Subject: [PATCH 25/28] perf tools: Emit clearer message for sys_perf_event_open ENOENT return Improve sys_perf_event_open ENOENT return handling in top and record, just like 5a3446b does for stat. Cc: David Ahern 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/builtin-record.c | 3 +++ tools/perf/builtin-top.c | 2 ++ 2 files changed, 5 insertions(+) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 7bc049035484..7069bd3e90b3 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -331,6 +331,9 @@ try_again: else if (err == ENODEV && cpu_list) { die("No such device - did you specify" " an out-of-range profile CPU?\n"); + } else if (err == ENOENT) { + die("%s event is not supported. ", + event_name(evsel)); } else if (err == EINVAL && sample_id_all_avail) { /* * Old kernel, no attr->sample_id_type_all field diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 1e67ab9c7ebc..6ce4042421bd 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1247,6 +1247,8 @@ try_again: die("Permission error - are you root?\n" "\t Consider tweaking" " /proc/sys/kernel/perf_event_paranoid.\n"); + if (err == ENOENT) + die("%s event is not supported. ", event_name(evsel)); /* * If it's cycles then fall back to hrtimer * based cpu-clock-tick sw counter, which From 12f7e0364375ba1ba55abcc5ac082b68fb526c80 Mon Sep 17 00:00:00 2001 From: Jiri Pirko Date: Mon, 10 Jan 2011 14:14:23 -0200 Subject: [PATCH 26/28] perf sched: Use PTHREAD_STACK_MIN to avoid pthread_attr_setstacksize() fail on ppc64: /usr/include/bits/local_lim.h:#define PTHREAD_STACK_MIN 131072 therefore following set of commands: gives: perf.2.6.37test: builtin-sched.c:493: create_tasks: Assertion `!(err)' failed. So make sure we do not set stack size lower than PTHREAD_STACK_MIN. Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <20110110160417.GB2685@psychotron.brq.redhat.com> Signed-off-by: Jiri Pirko Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-sched.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c index 54024d2a017e..abd4b8497bc4 100644 --- a/tools/perf/builtin-sched.c +++ b/tools/perf/builtin-sched.c @@ -489,7 +489,8 @@ static void create_tasks(void) err = pthread_attr_init(&attr); BUG_ON(err); - err = pthread_attr_setstacksize(&attr, (size_t)(16*1024)); + err = pthread_attr_setstacksize(&attr, + (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); BUG_ON(err); err = pthread_mutex_lock(&start_work_mutex); BUG_ON(err); From 0252208eb52f6fe8731a47804eddc7ba93f60a87 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 4 Jan 2011 11:55:27 -0200 Subject: [PATCH 27/28] 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/builtin-test.c | 110 ++++++++++++++++++++++++++++++++++++++ tools/perf/util/evsel.c | 96 +++++++++++++++++++-------------- 2 files changed, 166 insertions(+), 40 deletions(-) diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index e12753f976a1..ed5696198d3d 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -234,6 +234,7 @@ out: return err; } +#include "util/cpumap.h" #include "util/evsel.h" #include @@ -321,6 +322,111 @@ out_thread_map_delete: return err; } +#include + +static int test__open_syscall_event_on_all_cpus(void) +{ + int err = -1, fd, cpu; + struct thread_map *threads; + struct cpu_map *cpus; + struct perf_evsel *evsel; + struct perf_event_attr attr; + unsigned int nr_open_calls = 111, i; + cpu_set_t *cpu_set; + size_t cpu_set_size; + int id = trace_event__id("sys_enter_open"); + + if (id < 0) { + pr_debug("is debugfs mounted on /sys/kernel/debug?\n"); + return -1; + } + + threads = thread_map__new(-1, getpid()); + if (threads == NULL) { + pr_debug("thread_map__new\n"); + return -1; + } + + cpus = cpu_map__new(NULL); + if (threads == NULL) { + pr_debug("thread_map__new\n"); + return -1; + } + + cpu_set = CPU_ALLOC(cpus->nr); + + if (cpu_set == NULL) + goto out_thread_map_delete; + + cpu_set_size = CPU_ALLOC_SIZE(cpus->nr); + CPU_ZERO_S(cpu_set_size, cpu_set); + + memset(&attr, 0, sizeof(attr)); + attr.type = PERF_TYPE_TRACEPOINT; + attr.config = id; + evsel = perf_evsel__new(&attr, 0); + if (evsel == NULL) { + pr_debug("perf_evsel__new\n"); + goto out_cpu_free; + } + + if (perf_evsel__open(evsel, cpus, threads) < 0) { + pr_debug("failed to open counter: %s, " + "tweak /proc/sys/kernel/perf_event_paranoid?\n", + strerror(errno)); + goto out_evsel_delete; + } + + for (cpu = 0; cpu < cpus->nr; ++cpu) { + unsigned int ncalls = nr_open_calls + cpu; + + CPU_SET(cpu, cpu_set); + sched_setaffinity(0, cpu_set_size, cpu_set); + for (i = 0; i < ncalls; ++i) { + fd = open("/etc/passwd", O_RDONLY); + close(fd); + } + CPU_CLR(cpu, cpu_set); + } + + /* + * Here we need to explicitely preallocate the counts, as if + * we use the auto allocation it will allocate just for 1 cpu, + * as we start by cpu 0. + */ + if (perf_evsel__alloc_counts(evsel, cpus->nr) < 0) { + pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus->nr); + goto out_close_fd; + } + + for (cpu = 0; cpu < cpus->nr; ++cpu) { + unsigned int expected; + + if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) { + pr_debug("perf_evsel__open_read_on_cpu\n"); + goto out_close_fd; + } + + expected = nr_open_calls + cpu; + if (evsel->counts->cpu[cpu].val != expected) { + pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %Ld\n", + expected, cpu, evsel->counts->cpu[cpu].val); + goto out_close_fd; + } + } + + err = 0; +out_close_fd: + perf_evsel__close_fd(evsel, 1, threads->nr); +out_evsel_delete: + perf_evsel__delete(evsel); +out_cpu_free: + CPU_FREE(cpu_set); +out_thread_map_delete: + thread_map__delete(threads); + return err; +} + static struct test { const char *desc; int (*func)(void); @@ -333,6 +439,10 @@ static struct test { .desc = "detect open syscall event", .func = test__open_syscall_event, }, + { + .desc = "detect open syscall event on all cpus", + .func = test__open_syscall_event_on_all_cpus, + }, { .func = NULL, }, diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 1a5591d7a245..f5cfed60af98 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; } +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) +{ + + if (cpus == NULL) { + /* Work around old compiler warnings about strict aliasing */ + cpus = &empty_cpu_map.map; + } + + if (threads == NULL) + threads = &empty_thread_map.map; + + return __perf_evsel__open(evsel, cpus, threads); +} + +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) +{ + return __perf_evsel__open(evsel, cpus, &empty_thread_map.map); +} + int perf_evsel__open_per_thread(struct perf_evsel *evsel, 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; - } - - 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, - struct cpu_map *cpus, struct thread_map *threads) -{ - if (threads == NULL) - return perf_evsel__open_per_cpu(evsel, cpus); - - return perf_evsel__open_per_thread(evsel, threads); + return __perf_evsel__open(evsel, &empty_cpu_map.map, threads); } From 3d03e2ea74103a50c23d6ab1906cf73399c0dafb Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 10 Jan 2011 21:37:57 -0200 Subject: [PATCH 28/28] perf session: Fix infinite loop in __perf_session__process_events In this if statement: if (head + event->header.size >= mmap_size) { if (mmaps[map_idx]) { munmap(mmaps[map_idx], mmap_size); mmaps[map_idx] = NULL; } page_offset = page_size * (head / page_size); file_offset += page_offset; head -= page_offset; goto remap; } With, for instance, these values: head=2992 event->header.size=48 mmap_size=3040 We end up endlessly looping back to remap. Off by one. Problem introduced in 55b4462. Reported-by: Linus Torvalds Reported-by: Ingo Molnar Reported-by: David Ahern Bisected-by: David Ahern Tested-by: David Ahern Cc: David Ahern Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/session.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c index 6fb4694d05fa..313dac2d94ce 100644 --- a/tools/perf/util/session.c +++ b/tools/perf/util/session.c @@ -1007,7 +1007,7 @@ more: if (size == 0) size = 8; - if (head + event->header.size >= mmap_size) { + if (head + event->header.size > mmap_size) { if (mmaps[map_idx]) { munmap(mmaps[map_idx], mmap_size); mmaps[map_idx] = NULL;