From 2721e72dd10f71a3ba90f59781becf02638aa0d9 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Tue, 12 Mar 2013 11:32:32 -0400 Subject: [PATCH 01/18] tracing: Fix race in snapshot swapping Although the swap is wrapped with a spin_lock, the assignment of the temp buffer used to swap is not within that lock. It needs to be moved into that lock, otherwise two swaps happening on two different CPUs, can end up using the wrong temp buffer to assign in the swap. Luckily, all current callers of the swap function appear to have their own locks. But in case something is added that allows two different callers to call the swap, then there's a chance that this race can trigger and corrupt the buffers. New code is coming soon that will allow for this race to trigger. I've Cc'd stable, so this bug will not show up if someone backports one of the changes that can trigger this bug. Cc: stable@vger.kernel.org Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 1f835a83cb2c..53df2839bb93 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -704,7 +704,7 @@ __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) { - struct ring_buffer *buf = tr->buffer; + struct ring_buffer *buf; if (trace_stop_count) return; @@ -719,6 +719,7 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) arch_spin_lock(&ftrace_max_lock); + buf = tr->buffer; tr->buffer = max_tr.buffer; max_tr.buffer = buf; From 1e825efb571a87d039cb3fe7ca833a25f11c7cb9 Mon Sep 17 00:00:00 2001 From: Michael Ellerman Date: Tue, 26 Feb 2013 16:02:02 +1100 Subject: [PATCH 02/18] perf annotate: Fix build with NO_NEWT=1 Commit 18c9e5c "Make it to be able to skip unannotatable symbols" broke the build with NO_NEWT=1: CC builtin-annotate.o builtin-annotate.c: In function 'hists__find_annotations': builtin-annotate.c:161:4: error: duplicate case value builtin-annotate.c:154:4: error: previously used here make: *** [builtin-annotate.o] Error 1 This is because without NEWT support K_LEFT is #defined to -1 in utils/hist.h Fix it by shifting the K_LEFT/K_RIGHT #defines out of the likely range of error values. Signed-off-by: Michael Ellerman Cc: Feng Tang Cc: Namhyung Kim Cc: Namhyung Kim Link: http://lkml.kernel.org/r/1361854923-1814-1-git-send-email-michael@ellerman.id.au Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/hist.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h index 38624686ee9a..609a115e35cf 100644 --- a/tools/perf/util/hist.h +++ b/tools/perf/util/hist.h @@ -208,8 +208,8 @@ static inline int script_browse(const char *script_opt __maybe_unused) return 0; } -#define K_LEFT -1 -#define K_RIGHT -2 +#define K_LEFT -1000 +#define K_RIGHT -2000 #endif #ifdef GTK2_SUPPORT From 5f7439e07822942f32b9e0a66f4a3cc9c3e29e67 Mon Sep 17 00:00:00 2001 From: Michael Ellerman Date: Tue, 26 Feb 2013 16:02:03 +1100 Subject: [PATCH 03/18] perf report: Fix build with NO_NEWT=1 Commit ad0de09 "Enable the runtime switching of perf data file" broke the build with NO_NEWT=1: CC builtin-report.o builtin-report.c: In function '__cmd_report': builtin-report.c:479:15: error: 'K_SWITCH_INPUT_DATA' undeclared (first use in this function) builtin-report.c:479:15: note: each undeclared identifier is reported only once for each function it appears in builtin-report.c: In function 'cmd_report': builtin-report.c:823:13: error: 'K_SWITCH_INPUT_DATA' undeclared (first use in this function) make: *** [builtin-report.o] Error 1 Fix it by adding a dummy definition of K_SWITCH_INPUT_DATA. Signed-off-by: Michael Ellerman Cc: Feng Tang Cc: Namhyung Kim Cc: Namhyung Kim Link: http://lkml.kernel.org/r/1361854923-1814-2-git-send-email-michael@ellerman.id.au Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/hist.h | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h index 609a115e35cf..226a4ae2f936 100644 --- a/tools/perf/util/hist.h +++ b/tools/perf/util/hist.h @@ -210,6 +210,7 @@ static inline int script_browse(const char *script_opt __maybe_unused) #define K_LEFT -1000 #define K_RIGHT -2000 +#define K_SWITCH_INPUT_DATA -3000 #endif #ifdef GTK2_SUPPORT From d2f32479e5526a1ab3b4e43910fcb279871524ce Mon Sep 17 00:00:00 2001 From: Marcin Slusarz Date: Sun, 17 Feb 2013 16:03:36 +0100 Subject: [PATCH 04/18] perf tools: check if -DFORTIFY_SOURCE=2 is allowed It seems gcc (4.7.2) defines _FORTIFY_SOURCE internally and becomes confused when it sees another definition in flags. For me, build failed like this: CHK glibc Makefile:548: *** No gnu/libc-version.h found, please install glibc-dev[el]/glibc-static. Stop. and only with V=1 it printed: :0:0: error: "_FORTIFY_SOURCE" redefined [-Werror] :1:0: note: this is the location of the previous definition Signed-off-by: Marcin Slusarz Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1361113416-8662-1-git-send-email-marcin.slusarz@gmail.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Makefile | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/tools/perf/Makefile b/tools/perf/Makefile index a2108ca1cc17..bb74c79cd16e 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -95,7 +95,7 @@ ifeq ("$(origin DEBUG)", "command line") PERF_DEBUG = $(DEBUG) endif ifndef PERF_DEBUG - CFLAGS_OPTIMIZE = -O6 -D_FORTIFY_SOURCE=2 + CFLAGS_OPTIMIZE = -O6 endif ifdef PARSER_DEBUG @@ -180,6 +180,12 @@ ifeq ($(call try-cc,$(SOURCE_HELLO),$(CFLAGS) -Werror -Wvolatile-register-var,-W CFLAGS := $(CFLAGS) -Wvolatile-register-var endif +ifndef PERF_DEBUG + ifeq ($(call try-cc,$(SOURCE_HELLO),$(CFLAGS) -D_FORTIFY_SOURCE=2,-D_FORTIFY_SOURCE=2),y) + CFLAGS := $(CFLAGS) -D_FORTIFY_SOURCE=2 + endif +endif + ### --- END CONFIGURATION SECTION --- ifeq ($(srctree),) From e4dd45fe7add0de09e0e5b2b511732d9c86b6ee7 Mon Sep 17 00:00:00 2001 From: Jiri Olsa Date: Mon, 25 Feb 2013 10:52:48 +0100 Subject: [PATCH 05/18] perf record: Fix -C option Currently the -C option does not work for record command, because of the targets mismatch when synthesizing threads. Fixing this by using proper target interface for the synthesize decision. Signed-off-by: Jiri Olsa Reported-by: Oleg Nesterov Cc: Corey Ashford Cc: David Ahern Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Namhyung Kim Cc: Oleg Nesterov Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1361785972-7431-2-git-send-email-jolsa@redhat.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 774c90713a53..f1a939ebc19c 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -573,13 +573,15 @@ static int __cmd_record(struct perf_record *rec, int argc, const char **argv) perf_event__synthesize_guest_os, tool); } - if (!opts->target.system_wide) + if (perf_target__has_task(&opts->target)) err = perf_event__synthesize_thread_map(tool, evsel_list->threads, process_synthesized_event, machine); - else + else if (perf_target__has_cpu(&opts->target)) err = perf_event__synthesize_threads(tool, process_synthesized_event, machine); + else /* command specified */ + err = 0; if (err != 0) goto out_delete_session; From b9e8c37220c80e78289a1e87b50c09418eb59a7e Mon Sep 17 00:00:00 2001 From: Jack Mitchell Date: Fri, 8 Mar 2013 11:21:52 +0000 Subject: [PATCH 06/18] libtraceevent: Remove hard coded include to /usr/local/include in Makefile having /usr/local/include hardcoded into the makefile is not necessary as this is automatically included by GCC. It also infects cross-compile builds with the host systems includes. Signed-off-by: Jack Mitchell Acked-by: Namhyung Kim Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1362741712-21308-1-git-send-email-ml@communistcode.co.uk Signed-off-by: Arnaldo Carvalho de Melo --- tools/lib/traceevent/Makefile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/lib/traceevent/Makefile b/tools/lib/traceevent/Makefile index a20e32033431..0b0a90787db6 100644 --- a/tools/lib/traceevent/Makefile +++ b/tools/lib/traceevent/Makefile @@ -122,7 +122,7 @@ export Q VERBOSE EVENT_PARSE_VERSION = $(EP_VERSION).$(EP_PATCHLEVEL).$(EP_EXTRAVERSION) -INCLUDES = -I. -I/usr/local/include $(CONFIG_INCLUDES) +INCLUDES = -I. $(CONFIG_INCLUDES) # Set compile option CFLAGS if not set elsewhere CFLAGS ?= -g -Wall From 79146a69c8bc3f28e51c5267633abc6babf47a31 Mon Sep 17 00:00:00 2001 From: Ananth N Mavinakayanahalli Date: Tue, 12 Mar 2013 14:32:17 +0530 Subject: [PATCH 07/18] perf probe: Fix segfault Fix segfault in perf probe due to a bug introduced by commit d8639f068 (perf tools: Stop using 'self' in strlist). Signed-off-by: Ananth N Mavinakayanahalli Acked-by: Srikar Dronamraju Cc: Srikar Dronamraju Link: http://lkml.kernel.org/r/20130312090217.GC4668@in.ibm.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/strlist.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/util/strlist.c b/tools/perf/util/strlist.c index 55433aa42c8f..eabdce0a2daa 100644 --- a/tools/perf/util/strlist.c +++ b/tools/perf/util/strlist.c @@ -143,7 +143,7 @@ struct strlist *strlist__new(bool dupstr, const char *list) slist->rblist.node_delete = strlist__node_delete; slist->dupstr = dupstr; - if (slist && strlist__parse_list(slist, list) != 0) + if (list && strlist__parse_list(slist, list) != 0) goto out_error; } From 3bf7b07ece6e00747602938f68c1db8001b9925f Mon Sep 17 00:00:00 2001 From: Sukadev Bhattiprolu Date: Tue, 5 Mar 2013 21:48:26 -0800 Subject: [PATCH 08/18] perf/POWER7: Create a sysfs format entry for Power7 events Create a sysfs entry, '/sys/bus/event_source/devices/cpu/format/event' which describes the format of the POWER7 PMU events. This code is based on corresponding code in x86. Changelog[v4]: [Michael Ellerman, Paul Mckerras] The event format is different for other POWER cpus. So move the code to POWER7-specific, power7-pmu.c Also, the POWER7 format uses bits 0-19 not 0-20. Changelog[v2]: [Jiri Osla] Use PMU_FORMAT_ATTR rather than duplicating code. Signed-off-by: Sukadev Bhattiprolu Acked-by: Paul Mackerras Tested-by: Michael Ellerman Cc: Andi Kleen Cc: Anton Blanchard Cc: Ingo Molnar Cc: Jiri Olsa Cc: Michael Ellerman Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Robert Richter Cc: Stephane Eranian Cc: benh@kernel.crashing.org Cc: linuxppc-dev@ozlabs.org Link: http://lkml.kernel.org/r/20130306054826.GA14627@us.ibm.com Signed-off-by: Arnaldo Carvalho de Melo --- arch/powerpc/perf/power7-pmu.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/arch/powerpc/perf/power7-pmu.c b/arch/powerpc/perf/power7-pmu.c index b554879bd31e..3c475d6267c7 100644 --- a/arch/powerpc/perf/power7-pmu.c +++ b/arch/powerpc/perf/power7-pmu.c @@ -420,7 +420,20 @@ static struct attribute_group power7_pmu_events_group = { .attrs = power7_events_attr, }; +PMU_FORMAT_ATTR(event, "config:0-19"); + +static struct attribute *power7_pmu_format_attr[] = { + &format_attr_event.attr, + NULL, +}; + +struct attribute_group power7_pmu_format_group = { + .name = "format", + .attrs = power7_pmu_format_attr, +}; + static const struct attribute_group *power7_pmu_attr_groups[] = { + &power7_pmu_format_group, &power7_pmu_events_group, NULL, }; From 740466bc89ad8bd5afcc8de220f715f62b21e365 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Wed, 13 Mar 2013 11:15:19 -0400 Subject: [PATCH 09/18] tracing: Fix free of probe entry by calling call_rcu_sched() Because function tracing is very invasive, and can even trace calls to rcu_read_lock(), RCU access in function tracing is done with preempt_disable_notrace(). This requires a synchronize_sched() for updates and not a synchronize_rcu(). Function probes (traceon, traceoff, etc) must be freed after a synchronize_sched() after its entry has been removed from the hash. But call_rcu() is used. Fix this by using call_rcu_sched(). Also fix the usage to use hlist_del_rcu() instead of hlist_del(). Cc: stable@vger.kernel.org Cc: Paul McKenney Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 98ca94a41819..e6effd0c40a9 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -3108,8 +3108,8 @@ __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops, continue; } - hlist_del(&entry->node); - call_rcu(&entry->rcu, ftrace_free_entry_rcu); + hlist_del_rcu(&entry->node); + call_rcu_sched(&entry->rcu, ftrace_free_entry_rcu); } } __disable_ftrace_function_probe(); From d1398ccfec56e54010476efd6a316427d29045a6 Mon Sep 17 00:00:00 2001 From: Vinson Lee Date: Wed, 13 Mar 2013 15:34:24 -0700 Subject: [PATCH 10/18] perf tools: Fix LIBNUMA build with glibc 2.12 and older. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The tokens MADV_HUGEPAGE and MADV_NOHUGEPAGE are not available with glibc 2.12 and older. Define these tokens if they are not already defined. This patch fixes these build errors with older versions of glibc. CC bench/numa.o bench/numa.c: In function ‘alloc_data’: bench/numa.c:334: error: ‘MADV_HUGEPAGE’ undeclared (first use in this function) bench/numa.c:334: error: (Each undeclared identifier is reported only once bench/numa.c:334: error: for each function it appears in.) bench/numa.c:341: error: ‘MADV_NOHUGEPAGE’ undeclared (first use in this function) make: *** [bench/numa.o] Error 1 Signed-off-by: Vinson Lee Acked-by: Ingo Molnar Cc: Ingo Molnar Cc: Irina Tirdea Cc: Paul Mackerras Cc: Pekka Enberg Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1363214064-4671-2-git-send-email-vlee@twitter.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/bench/bench.h | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h index a5223e6a7b43..0fdc85269c4d 100644 --- a/tools/perf/bench/bench.h +++ b/tools/perf/bench/bench.h @@ -1,6 +1,30 @@ #ifndef BENCH_H #define BENCH_H +/* + * The madvise transparent hugepage constants were added in glibc + * 2.13. For compatibility with older versions of glibc, define these + * tokens if they are not already defined. + * + * PA-RISC uses different madvise values from other architectures and + * needs to be special-cased. + */ +#ifdef __hppa__ +# ifndef MADV_HUGEPAGE +# define MADV_HUGEPAGE 67 +# endif +# ifndef MADV_NOHUGEPAGE +# define MADV_NOHUGEPAGE 68 +# endif +#else +# ifndef MADV_HUGEPAGE +# define MADV_HUGEPAGE 14 +# endif +# ifndef MADV_NOHUGEPAGE +# define MADV_NOHUGEPAGE 15 +# endif +#endif + extern int bench_numa(int argc, const char **argv, const char *prefix); extern int bench_sched_messaging(int argc, const char **argv, const char *prefix); extern int bench_sched_pipe(int argc, const char **argv, const char *prefix); From 69d34da2984c95b33ea21518227e1f9470f11d95 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Thu, 14 Mar 2013 13:50:56 -0400 Subject: [PATCH 11/18] tracing: Protect tracer flags with trace_types_lock Seems that the tracer flags have never been protected from synchronous writes. Luckily, admins don't usually modify the tracing flags via two different tasks. But if scripts were to be used to modify them, then they could get corrupted. Move the trace_types_lock that protects against tracers changing to also protect the flags being set. Cc: stable@vger.kernel.org Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 53df2839bb93..00daf5f8c50b 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2916,6 +2916,8 @@ static int trace_set_options(char *option) cmp += 2; } + mutex_lock(&trace_types_lock); + for (i = 0; trace_options[i]; i++) { if (strcmp(cmp, trace_options[i]) == 0) { set_tracer_flags(1 << i, !neg); @@ -2924,11 +2926,10 @@ static int trace_set_options(char *option) } /* If no option could be set, test the specific tracer options */ - if (!trace_options[i]) { - mutex_lock(&trace_types_lock); + if (!trace_options[i]) ret = set_tracer_option(current_trace, cmp, neg); - mutex_unlock(&trace_types_lock); - } + + mutex_unlock(&trace_types_lock); return ret; } @@ -4781,7 +4782,10 @@ trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, if (val != 0 && val != 1) return -EINVAL; + + mutex_lock(&trace_types_lock); set_tracer_flags(1 << index, val); + mutex_unlock(&trace_types_lock); *ppos += cnt; From 80902822658aab18330569587cdb69ac1dfdcea8 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Thu, 14 Mar 2013 14:20:54 -0400 Subject: [PATCH 12/18] tracing: Keep overwrite in sync between regular and snapshot buffers Changing the overwrite mode for the ring buffer via the trace option only sets the normal buffer. But the snapshot buffer could swap with it, and then the snapshot would be in non overwrite mode and the normal buffer would be in overwrite mode, even though the option flag states otherwise. Keep the two buffers overwrite modes in sync. Cc: stable@vger.kernel.org Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 00daf5f8c50b..02debabe9ed4 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2895,8 +2895,12 @@ static void set_tracer_flags(unsigned int mask, int enabled) if (mask == TRACE_ITER_RECORD_CMD) trace_event_enable_cmd_record(enabled); - if (mask == TRACE_ITER_OVERWRITE) + if (mask == TRACE_ITER_OVERWRITE) { ring_buffer_change_overwrite(global_trace.buffer, enabled); +#ifdef CONFIG_TRACER_MAX_TRACE + ring_buffer_change_overwrite(max_tr.buffer, enabled); +#endif + } if (mask == TRACE_ITER_PRINTK) trace_printk_start_stop_comm(enabled); From 613f04a0f51e6e68ac6fe571ab79da3c0a5eb4da Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Thu, 14 Mar 2013 15:03:53 -0400 Subject: [PATCH 13/18] tracing: Prevent buffer overwrite disabled for latency tracers The latency tracers require the buffers to be in overwrite mode, otherwise they get screwed up. Force the buffers to stay in overwrite mode when latency tracers are enabled. Added a flag_changed() method to the tracer structure to allow the tracers to see what flags are being changed, and also be able to prevent the change from happing. Cc: stable@vger.kernel.org Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 38 ++++++++++++++++++++++++++----- kernel/trace/trace.h | 6 +++++ kernel/trace/trace_irqsoff.c | 19 ++++++++++++---- kernel/trace/trace_sched_wakeup.c | 18 +++++++++++---- 4 files changed, 65 insertions(+), 16 deletions(-) diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 02debabe9ed4..4f1dade56981 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2881,11 +2881,25 @@ static int set_tracer_option(struct tracer *trace, char *cmp, int neg) return -EINVAL; } -static void set_tracer_flags(unsigned int mask, int enabled) +/* Some tracers require overwrite to stay enabled */ +int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set) +{ + if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set) + return -1; + + return 0; +} + +int set_tracer_flag(unsigned int mask, int enabled) { /* do nothing if flag is already set */ if (!!(trace_flags & mask) == !!enabled) - return; + return 0; + + /* Give the tracer a chance to approve the change */ + if (current_trace->flag_changed) + if (current_trace->flag_changed(current_trace, mask, !!enabled)) + return -EINVAL; if (enabled) trace_flags |= mask; @@ -2904,13 +2918,15 @@ static void set_tracer_flags(unsigned int mask, int enabled) if (mask == TRACE_ITER_PRINTK) trace_printk_start_stop_comm(enabled); + + return 0; } static int trace_set_options(char *option) { char *cmp; int neg = 0; - int ret = 0; + int ret = -ENODEV; int i; cmp = strstrip(option); @@ -2924,7 +2940,7 @@ static int trace_set_options(char *option) for (i = 0; trace_options[i]; i++) { if (strcmp(cmp, trace_options[i]) == 0) { - set_tracer_flags(1 << i, !neg); + ret = set_tracer_flag(1 << i, !neg); break; } } @@ -2943,6 +2959,7 @@ tracing_trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos) { char buf[64]; + int ret; if (cnt >= sizeof(buf)) return -EINVAL; @@ -2952,7 +2969,9 @@ tracing_trace_options_write(struct file *filp, const char __user *ubuf, buf[cnt] = 0; - trace_set_options(buf); + ret = trace_set_options(buf); + if (ret < 0) + return ret; *ppos += cnt; @@ -3256,6 +3275,9 @@ static int tracing_set_tracer(const char *buf) goto out; trace_branch_disable(); + + current_trace->enabled = false; + if (current_trace->reset) current_trace->reset(tr); @@ -3300,6 +3322,7 @@ static int tracing_set_tracer(const char *buf) } current_trace = t; + current_trace->enabled = true; trace_branch_enable(tr); out: mutex_unlock(&trace_types_lock); @@ -4788,9 +4811,12 @@ trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, return -EINVAL; mutex_lock(&trace_types_lock); - set_tracer_flags(1 << index, val); + ret = set_tracer_flag(1 << index, val); mutex_unlock(&trace_types_lock); + if (ret < 0) + return ret; + *ppos += cnt; return cnt; diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index 57d7e5397d56..2081971367ea 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -283,11 +283,15 @@ struct tracer { enum print_line_t (*print_line)(struct trace_iterator *iter); /* If you handled the flag setting, return 0 */ int (*set_flag)(u32 old_flags, u32 bit, int set); + /* Return 0 if OK with change, else return non-zero */ + int (*flag_changed)(struct tracer *tracer, + u32 mask, int set); struct tracer *next; struct tracer_flags *flags; bool print_max; bool use_max_tr; bool allocated_snapshot; + bool enabled; }; @@ -943,6 +947,8 @@ extern const char *__stop___trace_bprintk_fmt[]; void trace_printk_init_buffers(void); void trace_printk_start_comm(void); +int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set); +int set_tracer_flag(unsigned int mask, int enabled); #undef FTRACE_ENTRY #define FTRACE_ENTRY(call, struct_name, id, tstruct, print, filter) \ diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c index 713a2cac4881..443b25b43b4f 100644 --- a/kernel/trace/trace_irqsoff.c +++ b/kernel/trace/trace_irqsoff.c @@ -32,7 +32,7 @@ enum { static int trace_type __read_mostly; -static int save_lat_flag; +static int save_flags; static void stop_irqsoff_tracer(struct trace_array *tr, int graph); static int start_irqsoff_tracer(struct trace_array *tr, int graph); @@ -558,8 +558,11 @@ static void stop_irqsoff_tracer(struct trace_array *tr, int graph) static void __irqsoff_tracer_init(struct trace_array *tr) { - save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT; - trace_flags |= TRACE_ITER_LATENCY_FMT; + save_flags = trace_flags; + + /* non overwrite screws up the latency tracers */ + set_tracer_flag(TRACE_ITER_OVERWRITE, 1); + set_tracer_flag(TRACE_ITER_LATENCY_FMT, 1); tracing_max_latency = 0; irqsoff_trace = tr; @@ -573,10 +576,13 @@ static void __irqsoff_tracer_init(struct trace_array *tr) static void irqsoff_tracer_reset(struct trace_array *tr) { + int lat_flag = save_flags & TRACE_ITER_LATENCY_FMT; + int overwrite_flag = save_flags & TRACE_ITER_OVERWRITE; + stop_irqsoff_tracer(tr, is_graph()); - if (!save_lat_flag) - trace_flags &= ~TRACE_ITER_LATENCY_FMT; + set_tracer_flag(TRACE_ITER_LATENCY_FMT, lat_flag); + set_tracer_flag(TRACE_ITER_OVERWRITE, overwrite_flag); } static void irqsoff_tracer_start(struct trace_array *tr) @@ -609,6 +615,7 @@ static struct tracer irqsoff_tracer __read_mostly = .print_line = irqsoff_print_line, .flags = &tracer_flags, .set_flag = irqsoff_set_flag, + .flag_changed = trace_keep_overwrite, #ifdef CONFIG_FTRACE_SELFTEST .selftest = trace_selftest_startup_irqsoff, #endif @@ -642,6 +649,7 @@ static struct tracer preemptoff_tracer __read_mostly = .print_line = irqsoff_print_line, .flags = &tracer_flags, .set_flag = irqsoff_set_flag, + .flag_changed = trace_keep_overwrite, #ifdef CONFIG_FTRACE_SELFTEST .selftest = trace_selftest_startup_preemptoff, #endif @@ -677,6 +685,7 @@ static struct tracer preemptirqsoff_tracer __read_mostly = .print_line = irqsoff_print_line, .flags = &tracer_flags, .set_flag = irqsoff_set_flag, + .flag_changed = trace_keep_overwrite, #ifdef CONFIG_FTRACE_SELFTEST .selftest = trace_selftest_startup_preemptirqsoff, #endif diff --git a/kernel/trace/trace_sched_wakeup.c b/kernel/trace/trace_sched_wakeup.c index 75aa97fbe1a1..fde652c9a511 100644 --- a/kernel/trace/trace_sched_wakeup.c +++ b/kernel/trace/trace_sched_wakeup.c @@ -36,7 +36,7 @@ static void __wakeup_reset(struct trace_array *tr); static int wakeup_graph_entry(struct ftrace_graph_ent *trace); static void wakeup_graph_return(struct ftrace_graph_ret *trace); -static int save_lat_flag; +static int save_flags; #define TRACE_DISPLAY_GRAPH 1 @@ -540,8 +540,11 @@ static void stop_wakeup_tracer(struct trace_array *tr) static int __wakeup_tracer_init(struct trace_array *tr) { - save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT; - trace_flags |= TRACE_ITER_LATENCY_FMT; + save_flags = trace_flags; + + /* non overwrite screws up the latency tracers */ + set_tracer_flag(TRACE_ITER_OVERWRITE, 1); + set_tracer_flag(TRACE_ITER_LATENCY_FMT, 1); tracing_max_latency = 0; wakeup_trace = tr; @@ -563,12 +566,15 @@ static int wakeup_rt_tracer_init(struct trace_array *tr) static void wakeup_tracer_reset(struct trace_array *tr) { + int lat_flag = save_flags & TRACE_ITER_LATENCY_FMT; + int overwrite_flag = save_flags & TRACE_ITER_OVERWRITE; + stop_wakeup_tracer(tr); /* make sure we put back any tasks we are tracing */ wakeup_reset(tr); - if (!save_lat_flag) - trace_flags &= ~TRACE_ITER_LATENCY_FMT; + set_tracer_flag(TRACE_ITER_LATENCY_FMT, lat_flag); + set_tracer_flag(TRACE_ITER_OVERWRITE, overwrite_flag); } static void wakeup_tracer_start(struct trace_array *tr) @@ -594,6 +600,7 @@ static struct tracer wakeup_tracer __read_mostly = .print_line = wakeup_print_line, .flags = &tracer_flags, .set_flag = wakeup_set_flag, + .flag_changed = trace_keep_overwrite, #ifdef CONFIG_FTRACE_SELFTEST .selftest = trace_selftest_startup_wakeup, #endif @@ -615,6 +622,7 @@ static struct tracer wakeup_rt_tracer __read_mostly = .print_line = wakeup_print_line, .flags = &tracer_flags, .set_flag = wakeup_set_flag, + .flag_changed = trace_keep_overwrite, #ifdef CONFIG_FTRACE_SELFTEST .selftest = trace_selftest_startup_wakeup, #endif From 778141e3cf0bf29f91cd3cb5c314ea477b9402a7 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Mon, 18 Mar 2013 11:41:46 +0900 Subject: [PATCH 14/18] perf: Reset hwc->last_period on sw clock events When cpu/task clock events are initialized, their sampling frequencies are converted to have a fixed value. However it missed to update the hwc->last_period which was set to 1 for initial sampling frequency calibration. Because this hwc->last_period value is used as a period in perf_swevent_ hrtime(), every recorded sample will have an incorrected period of 1. $ perf record -e task-clock noploop 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.158 MB perf.data (~6919 samples) ] $ perf report -n --show-total-period --stdio # Samples: 4K of event 'task-clock' # Event count (approx.): 4000 # # Overhead Samples Period Command Shared Object Symbol # ........ ............ ............ ....... ............. .................. # 99.95% 3998 3998 noploop noploop [.] main 0.03% 1 1 noploop libc-2.15.so [.] init_cacheinfo 0.03% 1 1 noploop ld-2.15.so [.] open_verify Note that it doesn't affect the non-sampling event so that the perf stat still gets correct value with or without this patch. $ perf stat -e task-clock noploop 1 Performance counter stats for 'noploop 1': 1000.272525 task-clock # 1.000 CPUs utilized 1.000560605 seconds time elapsed Signed-off-by: Namhyung Kim Cc: Arnaldo Carvalho de Melo Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1363574507-18808-1-git-send-email-namhyung@kernel.org Signed-off-by: Ingo Molnar --- kernel/events/core.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/events/core.c b/kernel/events/core.c index b0cd86501c30..fa79c377d65d 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5647,6 +5647,7 @@ static void perf_swevent_init_hrtimer(struct perf_event *event) event->attr.sample_period = NSEC_PER_SEC / freq; hwc->sample_period = event->attr.sample_period; local64_set(&hwc->period_left, hwc->sample_period); + hwc->last_period = hwc->sample_period; event->attr.freq = 0; } } From d610d98b5de6860feb21539726e9af7c9094151c Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Fri, 15 Mar 2013 16:27:13 +0900 Subject: [PATCH 15/18] perf: Generate EXIT event only once per task context perf_event_task_event() iterates pmu list and generate events for each eligible pmu context. But if task_event has task_ctx like in EXIT it'll generate events even though the pmu doesn't have an eligible one. Fix it by moving the code to proper places. Before this patch: $ perf record -n true [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.006 MB perf.data (~248 samples) ] $ perf report -D | tail Aggregated stats: TOTAL events: 73 MMAP events: 67 COMM events: 2 EXIT events: 4 cycles stats: TOTAL events: 73 MMAP events: 67 COMM events: 2 EXIT events: 4 After this patch: $ perf report -D | tail Aggregated stats: TOTAL events: 70 MMAP events: 67 COMM events: 2 EXIT events: 1 cycles stats: TOTAL events: 70 MMAP events: 67 COMM events: 2 EXIT events: 1 Signed-off-by: Namhyung Kim Cc: Arnaldo Carvalho de Melo Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1363332433-7637-1-git-send-email-namhyung@kernel.org Signed-off-by: Ingo Molnar --- kernel/events/core.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/kernel/events/core.c b/kernel/events/core.c index fa79c377d65d..59412d037eed 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -4434,12 +4434,15 @@ static void perf_event_task_event(struct perf_task_event *task_event) if (ctxn < 0) goto next; ctx = rcu_dereference(current->perf_event_ctxp[ctxn]); + if (ctx) + perf_event_task_ctx(ctx, task_event); } - if (ctx) - perf_event_task_ctx(ctx, task_event); next: put_cpu_ptr(pmu->pmu_cpu_context); } + if (task_event->task_ctx) + perf_event_task_ctx(task_event->task_ctx, task_event); + rcu_read_unlock(); } From 65c10553552b487a71bf5e4676743435046fae6f Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Thu, 14 Mar 2013 20:52:30 +0900 Subject: [PATCH 16/18] kprobes: Make hash_64() as always inlined Because hash_64() is called from the get_kprobe() inside int3 handler, kernel causes int3 recursion and crashes if kprobes user puts a probe on it. Usually hash_64() is inlined into caller function, but in some cases, it has instances by gcc's interprocedural constant propagation. This patch uses __always_inline instead of inline to prevent gcc from doing such things. Reported-by: Timo Juhani Lindfors Signed-off-by: Masami Hiramatsu Acked-by: Ananth N Mavinakayanahalli Cc: Pavel Emelyanov Cc: Jiri Kosina Cc: Nadia Yvette Chambers Cc: yrl.pp-manager.tt@hitachi.com Cc: David S. Miller Cc: Linus Torvalds Link: http://lkml.kernel.org/r/20130314115230.19690.39387.stgit@mhiramat-M0-7522 Signed-off-by: Ingo Molnar --- include/linux/hash.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/include/linux/hash.h b/include/linux/hash.h index 61c97ae22e01..f09a0ae4d858 100644 --- a/include/linux/hash.h +++ b/include/linux/hash.h @@ -15,6 +15,7 @@ */ #include +#include /* 2^31 + 2^29 - 2^25 + 2^22 - 2^19 - 2^16 + 1 */ #define GOLDEN_RATIO_PRIME_32 0x9e370001UL @@ -31,7 +32,7 @@ #error Wordsize not 32 or 64 #endif -static inline u64 hash_64(u64 val, unsigned int bits) +static __always_inline u64 hash_64(u64 val, unsigned int bits) { u64 hash = val; From 9a556ab998071457e79b319f2527642dd6e50617 Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Thu, 14 Mar 2013 20:52:43 +0900 Subject: [PATCH 17/18] kprobes/x86: Check Interrupt Flag modifier when registering probe Currently kprobes check whether the copied instruction modifies IF (interrupt flag) on each probe hit. This results not only in introducing overhead but also involving inat_get_opcode_attribute into the kprobes hot path, and it can cause an infinite recursive call (and kernel panic in the end). Actually, since the copied instruction itself can never be modified on the buffer, it is needless to analyze the instruction on every probe hit. To fix this issue, we check it only once when registering probe and store the result on ainsn->if_modifier. Reported-by: Timo Juhani Lindfors Signed-off-by: Masami Hiramatsu Acked-by: Ananth N Mavinakayanahalli Cc: yrl.pp-manager.tt@hitachi.com Cc: Steven Rostedt Cc: David S. Miller Cc: Linus Torvalds Link: http://lkml.kernel.org/r/20130314115242.19690.33573.stgit@mhiramat-M0-7522 Signed-off-by: Ingo Molnar --- arch/x86/include/asm/kprobes.h | 1 + arch/x86/kernel/kprobes/core.c | 5 ++++- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/arch/x86/include/asm/kprobes.h b/arch/x86/include/asm/kprobes.h index d3ddd17405d0..5a6d2873f80e 100644 --- a/arch/x86/include/asm/kprobes.h +++ b/arch/x86/include/asm/kprobes.h @@ -77,6 +77,7 @@ struct arch_specific_insn { * a post_handler or break_handler). */ int boostable; + bool if_modifier; }; struct arch_optimized_insn { diff --git a/arch/x86/kernel/kprobes/core.c b/arch/x86/kernel/kprobes/core.c index 3f06e6149981..7bfe318d3d8a 100644 --- a/arch/x86/kernel/kprobes/core.c +++ b/arch/x86/kernel/kprobes/core.c @@ -375,6 +375,9 @@ static void __kprobes arch_copy_kprobe(struct kprobe *p) else p->ainsn.boostable = -1; + /* Check whether the instruction modifies Interrupt Flag or not */ + p->ainsn.if_modifier = is_IF_modifier(p->ainsn.insn); + /* Also, displacement change doesn't affect the first byte */ p->opcode = p->ainsn.insn[0]; } @@ -434,7 +437,7 @@ static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs, __this_cpu_write(current_kprobe, p); kcb->kprobe_saved_flags = kcb->kprobe_old_flags = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF)); - if (is_IF_modifier(p->ainsn.insn)) + if (p->ainsn.if_modifier) kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF; } From fd4a5aef002bb57e8a35ed34d8a878034b9bde94 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Sun, 17 Mar 2013 14:49:57 +0100 Subject: [PATCH 18/18] perf/x86: Add SNB/SNB-EP scheduling constraints for cycle_activity event Add scheduling constraints for SNB/SNB-EP CYCLE_ACTIVITY event as defined by SDM Jan 2013 edition. The STALLS umasks are combinations with the NO_DISPATCH umask. Signed-off-by: Stephane Eranian Cc: peterz@infradead.org Cc: ak@linux.intel.com Cc: jolsa@redhat.com Link: http://lkml.kernel.org/r/20130317134957.GA8550@quad Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event_intel.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c index 529c8931fc02..dab7580c47ae 100644 --- a/arch/x86/kernel/cpu/perf_event_intel.c +++ b/arch/x86/kernel/cpu/perf_event_intel.c @@ -101,6 +101,10 @@ static struct event_constraint intel_snb_event_constraints[] __read_mostly = FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */ FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */ FIXED_EVENT_CONSTRAINT(0x0300, 2), /* CPU_CLK_UNHALTED.REF */ + INTEL_UEVENT_CONSTRAINT(0x04a3, 0xf), /* CYCLE_ACTIVITY.CYCLES_NO_DISPATCH */ + INTEL_UEVENT_CONSTRAINT(0x05a3, 0xf), /* CYCLE_ACTIVITY.STALLS_L2_PENDING */ + INTEL_UEVENT_CONSTRAINT(0x02a3, 0x4), /* CYCLE_ACTIVITY.CYCLES_L1D_PENDING */ + INTEL_UEVENT_CONSTRAINT(0x06a3, 0x4), /* CYCLE_ACTIVITY.STALLS_L1D_PENDING */ INTEL_EVENT_CONSTRAINT(0x48, 0x4), /* L1D_PEND_MISS.PENDING */ INTEL_UEVENT_CONSTRAINT(0x01c0, 0x2), /* INST_RETIRED.PREC_DIST */ INTEL_EVENT_CONSTRAINT(0xcd, 0x8), /* MEM_TRANS_RETIRED.LOAD_LATENCY */