From 6fb2915df7f0747d9044da9dbff5b46dc2e20830 Mon Sep 17 00:00:00 2001 From: Li Zefan Date: Thu, 15 Oct 2009 11:21:42 +0800 Subject: tracing/profile: Add filter support - Add an ioctl to allocate a filter for a perf event. - Free the filter when the associated perf event is to be freed. - Do the filtering in perf_swevent_match(). Signed-off-by: Li Zefan Acked-by: Peter Zijlstra Acked-by: Frederic Weisbecker Cc: Steven Rostedt Cc: Tom Zanussi LKML-Reference: <4AD69546.8050401@cn.fujitsu.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 73 insertions(+), 7 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 9d0b5c665883..12b5ec39bf97 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -28,6 +28,7 @@ #include #include #include +#include #include @@ -1658,6 +1659,8 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) return ERR_PTR(err); } +static void perf_event_free_filter(struct perf_event *event); + static void free_event_rcu(struct rcu_head *head) { struct perf_event *event; @@ -1665,6 +1668,7 @@ static void free_event_rcu(struct rcu_head *head) event = container_of(head, struct perf_event, rcu_head); if (event->ns) put_pid_ns(event->ns); + perf_event_free_filter(event); kfree(event); } @@ -1974,7 +1978,8 @@ unlock: return ret; } -int perf_event_set_output(struct perf_event *event, int output_fd); +static int perf_event_set_output(struct perf_event *event, int output_fd); +static int perf_event_set_filter(struct perf_event *event, void __user *arg); static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { @@ -2002,6 +2007,9 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case PERF_EVENT_IOC_SET_OUTPUT: return perf_event_set_output(event, arg); + case PERF_EVENT_IOC_SET_FILTER: + return perf_event_set_filter(event, (void __user *)arg); + default: return -ENOTTY; } @@ -3806,9 +3814,14 @@ static int perf_swevent_is_counting(struct perf_event *event) return 1; } +static int perf_tp_event_match(struct perf_event *event, + struct perf_sample_data *data); + static int perf_swevent_match(struct perf_event *event, enum perf_type_id type, - u32 event_id, struct pt_regs *regs) + u32 event_id, + struct perf_sample_data *data, + struct pt_regs *regs) { if (!perf_swevent_is_counting(event)) return 0; @@ -3826,6 +3839,10 @@ static int perf_swevent_match(struct perf_event *event, return 0; } + if (event->attr.type == PERF_TYPE_TRACEPOINT && + !perf_tp_event_match(event, data)) + return 0; + return 1; } @@ -3842,7 +3859,7 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { - if (perf_swevent_match(event, type, event_id, regs)) + if (perf_swevent_match(event, type, event_id, data, regs)) perf_swevent_add(event, nr, nmi, data, regs); } rcu_read_unlock(); @@ -4086,6 +4103,7 @@ static const struct pmu perf_ops_task_clock = { }; #ifdef CONFIG_EVENT_PROFILE + void perf_tp_event(int event_id, u64 addr, u64 count, void *record, int entry_size) { @@ -4109,8 +4127,15 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record, } EXPORT_SYMBOL_GPL(perf_tp_event); -extern int ftrace_profile_enable(int); -extern void ftrace_profile_disable(int); +static int perf_tp_event_match(struct perf_event *event, + struct perf_sample_data *data) +{ + void *record = data->raw->data; + + if (likely(!event->filter) || filter_match_preds(event->filter, record)) + return 1; + return 0; +} static void tp_perf_event_destroy(struct perf_event *event) { @@ -4135,12 +4160,53 @@ static const struct pmu *tp_perf_event_init(struct perf_event *event) return &perf_ops_generic; } + +static int perf_event_set_filter(struct perf_event *event, void __user *arg) +{ + char *filter_str; + int ret; + + if (event->attr.type != PERF_TYPE_TRACEPOINT) + return -EINVAL; + + filter_str = strndup_user(arg, PAGE_SIZE); + if (IS_ERR(filter_str)) + return PTR_ERR(filter_str); + + ret = ftrace_profile_set_filter(event, event->attr.config, filter_str); + + kfree(filter_str); + return ret; +} + +static void perf_event_free_filter(struct perf_event *event) +{ + ftrace_profile_free_filter(event); +} + #else + +static int perf_tp_event_match(struct perf_event *event, + struct perf_sample_data *data) +{ + return 1; +} + static const struct pmu *tp_perf_event_init(struct perf_event *event) { return NULL; } -#endif + +static int perf_event_set_filter(struct perf_event *event, void __user *arg) +{ + return -ENOENT; +} + +static void perf_event_free_filter(struct perf_event *event) +{ +} + +#endif /* CONFIG_EVENT_PROFILE */ atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; @@ -4394,7 +4460,7 @@ err_size: goto out; } -int perf_event_set_output(struct perf_event *event, int output_fd) +static int perf_event_set_output(struct perf_event *event, int output_fd) { struct perf_event *output_event = NULL; struct file *output_file = NULL; -- cgit v1.2.3 From f7d7986060b2890fc26db6ab5203efbd33aa2497 Mon Sep 17 00:00:00 2001 From: Anton Blanchard Date: Sun, 18 Oct 2009 01:09:29 +0000 Subject: perf_event: Add alignment-faults and emulation-faults software events Add two more software events that are common to many cpus. Alignment faults: When a load or store is not aligned properly. Emulation faults: When an instruction is emulated in software. Both cause a very significant slowdown (100x or worse), so identifying and fixing them is very important. Signed-off-by: Anton Blanchard Signed-off-by: Paul Mackerras --- include/linux/perf_counter.h | 2 ++ include/linux/perf_event.h | 2 ++ kernel/perf_event.c | 2 ++ tools/perf/design.txt | 2 ++ tools/perf/util/parse-events.c | 4 ++++ 5 files changed, 12 insertions(+) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 7b7fbf433cff..d6b95d1e79f0 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -106,6 +106,8 @@ enum perf_sw_ids { PERF_COUNT_SW_CPU_MIGRATIONS = 4, PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, + PERF_COUNT_SW_ALIGNMENT_FAULTS = 7, + PERF_COUNT_SW_EMULATION_FAULTS = 8, PERF_COUNT_SW_MAX, /* non-ABI */ }; diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 2e6d95f97419..a33707a3a788 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -102,6 +102,8 @@ enum perf_sw_ids { PERF_COUNT_SW_CPU_MIGRATIONS = 4, PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, + PERF_COUNT_SW_ALIGNMENT_FAULTS = 7, + PERF_COUNT_SW_EMULATION_FAULTS = 8, PERF_COUNT_SW_MAX, /* non-ABI */ }; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 9d0b5c665883..0683b33cbb28 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4186,6 +4186,8 @@ static const struct pmu *sw_perf_event_init(struct perf_event *event) case PERF_COUNT_SW_PAGE_FAULTS_MAJ: case PERF_COUNT_SW_CONTEXT_SWITCHES: case PERF_COUNT_SW_CPU_MIGRATIONS: + case PERF_COUNT_SW_ALIGNMENT_FAULTS: + case PERF_COUNT_SW_EMULATION_FAULTS: if (!event->parent) { atomic_inc(&perf_swevent_enabled[event_id]); event->destroy = sw_perf_event_destroy; diff --git a/tools/perf/design.txt b/tools/perf/design.txt index fdd42a824c98..f000c30877ac 100644 --- a/tools/perf/design.txt +++ b/tools/perf/design.txt @@ -137,6 +137,8 @@ enum sw_event_ids { PERF_COUNT_SW_CPU_MIGRATIONS = 4, PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, + PERF_COUNT_SW_ALIGNMENT_FAULTS = 7, + PERF_COUNT_SW_EMULATION_FAULTS = 8, }; Counters of the type PERF_TYPE_TRACEPOINT are available when the ftrace event diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 8cfb48cbbea0..34bd84423933 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -46,6 +46,8 @@ static struct event_symbol event_symbols[] = { { CSW(PAGE_FAULTS_MAJ), "major-faults", "" }, { CSW(CONTEXT_SWITCHES), "context-switches", "cs" }, { CSW(CPU_MIGRATIONS), "cpu-migrations", "migrations" }, + { CSW(ALIGNMENT_FAULTS), "alignment-faults", "" }, + { CSW(EMULATION_FAULTS), "emulation-faults", "" }, }; #define __PERF_EVENT_FIELD(config, name) \ @@ -74,6 +76,8 @@ static const char *sw_event_names[] = { "CPU-migrations", "minor-faults", "major-faults", + "alignment-faults", + "emulation-faults", }; #define MAX_ALIASES 8 -- cgit v1.2.3 From fb0459d75c1d0a4ba3cafdd2c754e7486968a676 Mon Sep 17 00:00:00 2001 From: Arjan van de Ven Date: Fri, 25 Sep 2009 12:25:56 +0200 Subject: perf/core: Provide a kernel-internal interface to get to performance counters There are reasons for kernel code to ask for, and use, performance counters. For example, in CPU freq governors this tends to be a good idea, but there are other examples possible as well of course. This patch adds the needed bits to do enable this functionality; they have been tested in an experimental cpufreq driver that I'm working on, and the changes are all that I needed to access counters properly. [fweisbec@gmail.com: added pid to perf_event_create_kernel_counter so that we can profile a particular task too TODO: Have a better error reporting, don't just return NULL in fail case.] v2: Remove the wrong comment about the fact perf_event_create_kernel_counter must be called from a kernel thread. Signed-off-by: Arjan van de Ven Acked-by: Peter Zijlstra Cc: "K.Prasad" Cc: Alan Stern Cc: Arnaldo Carvalho de Melo Cc: Steven Rostedt Cc: Ingo Molnar Cc: Jan Kiszka Cc: Jiri Slaby Cc: Li Zefan Cc: Avi Kivity Cc: Paul Mackerras Cc: Mike Galbraith Cc: Masami Hiramatsu Cc: Paul Mundt Cc: Jan Kiszka Cc: Avi Kivity LKML-Reference: <20090925122556.2f8bd939@infradead.org> Signed-off-by: Frederic Weisbecker --- include/linux/perf_event.h | 6 ++++ kernel/perf_event.c | 75 +++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 80 insertions(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index df9d964c15fc..fa151d49a2ee 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -744,6 +744,12 @@ extern int hw_perf_group_sched_in(struct perf_event *group_leader, struct perf_cpu_context *cpuctx, struct perf_event_context *ctx, int cpu); extern void perf_event_update_userpage(struct perf_event *event); +extern int perf_event_release_kernel(struct perf_event *event); +extern struct perf_event * +perf_event_create_kernel_counter(struct perf_event_attr *attr, + int cpu, + pid_t pid); +extern u64 perf_event_read_value(struct perf_event *event); struct perf_sample_data { u64 type; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 12b5ec39bf97..02d4ff041b01 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1725,6 +1725,26 @@ static int perf_release(struct inode *inode, struct file *file) return 0; } +int perf_event_release_kernel(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + + WARN_ON_ONCE(ctx->parent_ctx); + mutex_lock(&ctx->mutex); + perf_event_remove_from_context(event); + mutex_unlock(&ctx->mutex); + + mutex_lock(&event->owner->perf_event_mutex); + list_del_init(&event->owner_entry); + mutex_unlock(&event->owner->perf_event_mutex); + put_task_struct(event->owner); + + free_event(event); + + return 0; +} +EXPORT_SYMBOL_GPL(perf_event_release_kernel); + static int perf_event_read_size(struct perf_event *event) { int entry = sizeof(u64); /* value */ @@ -1750,7 +1770,7 @@ static int perf_event_read_size(struct perf_event *event) return size; } -static u64 perf_event_read_value(struct perf_event *event) +u64 perf_event_read_value(struct perf_event *event) { struct perf_event *child; u64 total = 0; @@ -1761,6 +1781,7 @@ static u64 perf_event_read_value(struct perf_event *event) return total; } +EXPORT_SYMBOL_GPL(perf_event_read_value); static int perf_event_read_entry(struct perf_event *event, u64 read_format, char __user *buf) @@ -4638,6 +4659,58 @@ err_put_context: return err; } +/** + * perf_event_create_kernel_counter + * + * @attr: attributes of the counter to create + * @cpu: cpu in which the counter is bound + * @pid: task to profile + */ +struct perf_event * +perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, + pid_t pid) +{ + struct perf_event *event; + struct perf_event_context *ctx; + int err; + + /* + * Get the target context (task or percpu): + */ + + ctx = find_get_context(pid, cpu); + if (IS_ERR(ctx)) + return NULL ; + + event = perf_event_alloc(attr, cpu, ctx, NULL, + NULL, GFP_KERNEL); + err = PTR_ERR(event); + if (IS_ERR(event)) + goto err_put_context; + + event->filp = NULL; + WARN_ON_ONCE(ctx->parent_ctx); + mutex_lock(&ctx->mutex); + perf_install_in_context(ctx, event, cpu); + ++ctx->generation; + mutex_unlock(&ctx->mutex); + + event->owner = current; + get_task_struct(current); + mutex_lock(¤t->perf_event_mutex); + list_add_tail(&event->owner_entry, ¤t->perf_event_list); + mutex_unlock(¤t->perf_event_mutex); + + return event; + +err_put_context: + if (err < 0) + put_ctx(ctx); + + return NULL; +} +EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter); + /* * inherit a event from parent task to child task: */ -- cgit v1.2.3 From 97eaf5300b9d0cd99c310bf8c4a0f2f3296d88a3 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Sun, 18 Oct 2009 15:33:50 +0200 Subject: perf/core: Add a callback to perf events A simple callback in a perf event can be used for multiple purposes. For example it is useful for triggered based events like hardware breakpoints that need a callback to dispatch a triggered breakpoint event. v2: Simplify a bit the callback attribution as suggested by Paul Mackerras Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: "K.Prasad" Cc: Alan Stern Cc: Arnaldo Carvalho de Melo Cc: Steven Rostedt Cc: Ingo Molnar Cc: Paul Mackerras Cc: Mike Galbraith Cc: Paul Mundt --- include/linux/perf_event.h | 7 ++++++- kernel/perf_event.c | 14 ++++++++++---- 2 files changed, 16 insertions(+), 5 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index fa151d49a2ee..8d54e6d25eeb 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -544,6 +544,8 @@ struct perf_pending_entry { void (*func)(struct perf_pending_entry *); }; +typedef void (*perf_callback_t)(struct perf_event *, void *); + /** * struct perf_event - performance event kernel representation: */ @@ -639,6 +641,8 @@ struct perf_event { struct event_filter *filter; #endif + perf_callback_t callback; + #endif /* CONFIG_PERF_EVENTS */ }; @@ -748,7 +752,8 @@ extern int perf_event_release_kernel(struct perf_event *event); extern struct perf_event * perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, - pid_t pid); + pid_t pid, + perf_callback_t callback); extern u64 perf_event_read_value(struct perf_event *event); struct perf_sample_data { diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 02d4ff041b01..5087125e2a00 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4293,6 +4293,7 @@ perf_event_alloc(struct perf_event_attr *attr, struct perf_event_context *ctx, struct perf_event *group_leader, struct perf_event *parent_event, + perf_callback_t callback, gfp_t gfpflags) { const struct pmu *pmu; @@ -4335,6 +4336,11 @@ perf_event_alloc(struct perf_event_attr *attr, event->state = PERF_EVENT_STATE_INACTIVE; + if (!callback && parent_event) + callback = parent_event->callback; + + event->callback = callback; + if (attr->disabled) event->state = PERF_EVENT_STATE_OFF; @@ -4611,7 +4617,7 @@ SYSCALL_DEFINE5(perf_event_open, } event = perf_event_alloc(&attr, cpu, ctx, group_leader, - NULL, GFP_KERNEL); + NULL, NULL, GFP_KERNEL); err = PTR_ERR(event); if (IS_ERR(event)) goto err_put_context; @@ -4668,7 +4674,7 @@ err_put_context: */ struct perf_event * perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, - pid_t pid) + pid_t pid, perf_callback_t callback) { struct perf_event *event; struct perf_event_context *ctx; @@ -4683,7 +4689,7 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, return NULL ; event = perf_event_alloc(attr, cpu, ctx, NULL, - NULL, GFP_KERNEL); + NULL, callback, GFP_KERNEL); err = PTR_ERR(event); if (IS_ERR(event)) goto err_put_context; @@ -4736,7 +4742,7 @@ inherit_event(struct perf_event *parent_event, child_event = perf_event_alloc(&parent_event->attr, parent_event->cpu, child_ctx, group_leader, parent_event, - GFP_KERNEL); + NULL, GFP_KERNEL); if (IS_ERR(child_event)) return child_event; get_ctx(child_ctx); -- cgit v1.2.3 From 24f1e32c60c45c89a997c73395b69c8af6f0a84e Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 9 Sep 2009 19:22:48 +0200 Subject: hw-breakpoints: Rewrite the hw-breakpoints layer on top of perf events This patch rebase the implementation of the breakpoints API on top of perf events instances. Each breakpoints are now perf events that handle the register scheduling, thread/cpu attachment, etc.. The new layering is now made as follows: ptrace kgdb ftrace perf syscall \ | / / \ | / / / Core breakpoint API / / | / | / Breakpoints perf events | | Breakpoints PMU ---- Debug Register constraints handling (Part of core breakpoint API) | | Hardware debug registers Reasons of this rewrite: - Use the centralized/optimized pmu registers scheduling, implying an easier arch integration - More powerful register handling: perf attributes (pinned/flexible events, exclusive/non-exclusive, tunable period, etc...) Impact: - New perf ABI: the hardware breakpoints counters - Ptrace breakpoints setting remains tricky and still needs some per thread breakpoints references. Todo (in the order): - Support breakpoints perf counter events for perf tools (ie: implement perf_bpcounter_event()) - Support from perf tools Changes in v2: - Follow the perf "event " rename - The ptrace regression have been fixed (ptrace breakpoint perf events weren't released when a task ended) - Drop the struct hw_breakpoint and store generic fields in perf_event_attr. - Separate core and arch specific headers, drop asm-generic/hw_breakpoint.h and create linux/hw_breakpoint.h - Use new generic len/type for breakpoint - Handle off case: when breakpoints api is not supported by an arch Changes in v3: - Fix broken CONFIG_KVM, we need to propagate the breakpoint api changes to kvm when we exit the guest and restore the bp registers to the host. Changes in v4: - Drop the hw_breakpoint_restore() stub as it is only used by KVM - EXPORT_SYMBOL_GPL hw_breakpoint_restore() as KVM can be built as a module - Restore the breakpoints unconditionally on kvm guest exit: TIF_DEBUG_THREAD doesn't anymore cover every cases of running breakpoints and vcpu->arch.switch_db_regs might not always be set when the guest used debug registers. (Waiting for a reliable optimization) Changes in v5: - Split-up the asm-generic/hw-breakpoint.h moving to linux/hw_breakpoint.h into a separate patch - Optimize the breakpoints restoring while switching from kvm guest to host. We only want to restore the state if we have active breakpoints to the host, otherwise we don't care about messed-up address registers. - Add asm/hw_breakpoint.h to Kbuild - Fix bad breakpoint type in trace_selftest.c Changes in v6: - Fix wrong header inclusion in trace.h (triggered a build error with CONFIG_FTRACE_SELFTEST Signed-off-by: Frederic Weisbecker Cc: Prasad Cc: Alan Stern Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Steven Rostedt Cc: Ingo Molnar Cc: Jan Kiszka Cc: Jiri Slaby Cc: Li Zefan Cc: Avi Kivity Cc: Paul Mackerras Cc: Mike Galbraith Cc: Masami Hiramatsu Cc: Paul Mundt --- arch/Kconfig | 3 + arch/x86/include/asm/Kbuild | 1 + arch/x86/include/asm/debugreg.h | 11 +- arch/x86/include/asm/hw_breakpoint.h | 58 +++-- arch/x86/include/asm/processor.h | 12 +- arch/x86/kernel/hw_breakpoint.c | 391 +++++++++++++++++++++----------- arch/x86/kernel/process.c | 7 +- arch/x86/kernel/process_32.c | 26 +-- arch/x86/kernel/process_64.c | 26 +-- arch/x86/kernel/ptrace.c | 182 ++++++++++----- arch/x86/kernel/smpboot.c | 3 - arch/x86/kvm/x86.c | 18 +- arch/x86/power/cpu.c | 6 - include/linux/hw_breakpoint.h | 243 ++++++++++---------- include/linux/perf_event.h | 26 ++- kernel/exit.c | 5 + kernel/hw_breakpoint.c | 424 ++++++++++++++--------------------- kernel/perf_event.c | 53 ++++- kernel/trace/trace.h | 5 +- kernel/trace/trace_entries.h | 6 +- kernel/trace/trace_ksym.c | 126 +++++------ kernel/trace/trace_selftest.c | 3 +- 22 files changed, 885 insertions(+), 750 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/arch/Kconfig b/arch/Kconfig index acb664397945..eef3bbb97075 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -128,6 +128,9 @@ config HAVE_DEFAULT_NO_SPIN_MUTEXES config HAVE_HW_BREAKPOINT bool + depends on HAVE_PERF_EVENTS + select ANON_INODES + select PERF_EVENTS source "kernel/gcov/Kconfig" diff --git a/arch/x86/include/asm/Kbuild b/arch/x86/include/asm/Kbuild index 4a8e80cdcfa5..9f828f87ca35 100644 --- a/arch/x86/include/asm/Kbuild +++ b/arch/x86/include/asm/Kbuild @@ -10,6 +10,7 @@ header-y += ptrace-abi.h header-y += sigcontext32.h header-y += ucontext.h header-y += processor-flags.h +header-y += hw_breakpoint.h unifdef-y += e820.h unifdef-y += ist.h diff --git a/arch/x86/include/asm/debugreg.h b/arch/x86/include/asm/debugreg.h index 23439fbb1d0e..9a3333c91f9a 100644 --- a/arch/x86/include/asm/debugreg.h +++ b/arch/x86/include/asm/debugreg.h @@ -75,13 +75,8 @@ */ #ifdef __KERNEL__ -/* For process management */ -extern void flush_thread_hw_breakpoint(struct task_struct *tsk); -extern int copy_thread_hw_breakpoint(struct task_struct *tsk, - struct task_struct *child, unsigned long clone_flags); +DECLARE_PER_CPU(unsigned long, dr7); -/* For CPU management */ -extern void load_debug_registers(void); static inline void hw_breakpoint_disable(void) { /* Zero the control register for HW Breakpoint */ @@ -94,6 +89,10 @@ static inline void hw_breakpoint_disable(void) set_debugreg(0UL, 3); } +#ifdef CONFIG_KVM +extern void hw_breakpoint_restore(void); +#endif + #endif /* __KERNEL__ */ #endif /* _ASM_X86_DEBUGREG_H */ diff --git a/arch/x86/include/asm/hw_breakpoint.h b/arch/x86/include/asm/hw_breakpoint.h index 3cfca8e2b5f6..0675a7c4c20e 100644 --- a/arch/x86/include/asm/hw_breakpoint.h +++ b/arch/x86/include/asm/hw_breakpoint.h @@ -4,6 +4,11 @@ #ifdef __KERNEL__ #define __ARCH_HW_BREAKPOINT_H +/* + * The name should probably be something dealt in + * a higher level. While dealing with the user + * (display/resolving) + */ struct arch_hw_breakpoint { char *name; /* Contains name of the symbol to set bkpt */ unsigned long address; @@ -12,44 +17,57 @@ struct arch_hw_breakpoint { }; #include -#include +#include +#include /* Available HW breakpoint length encodings */ -#define HW_BREAKPOINT_LEN_1 0x40 -#define HW_BREAKPOINT_LEN_2 0x44 -#define HW_BREAKPOINT_LEN_4 0x4c -#define HW_BREAKPOINT_LEN_EXECUTE 0x40 +#define X86_BREAKPOINT_LEN_1 0x40 +#define X86_BREAKPOINT_LEN_2 0x44 +#define X86_BREAKPOINT_LEN_4 0x4c +#define X86_BREAKPOINT_LEN_EXECUTE 0x40 #ifdef CONFIG_X86_64 -#define HW_BREAKPOINT_LEN_8 0x48 +#define X86_BREAKPOINT_LEN_8 0x48 #endif /* Available HW breakpoint type encodings */ /* trigger on instruction execute */ -#define HW_BREAKPOINT_EXECUTE 0x80 +#define X86_BREAKPOINT_EXECUTE 0x80 /* trigger on memory write */ -#define HW_BREAKPOINT_WRITE 0x81 +#define X86_BREAKPOINT_WRITE 0x81 /* trigger on memory read or write */ -#define HW_BREAKPOINT_RW 0x83 +#define X86_BREAKPOINT_RW 0x83 /* Total number of available HW breakpoint registers */ #define HBP_NUM 4 -extern struct hw_breakpoint *hbp_kernel[HBP_NUM]; -DECLARE_PER_CPU(struct hw_breakpoint*, this_hbp_kernel[HBP_NUM]); -extern unsigned int hbp_user_refcount[HBP_NUM]; +struct perf_event; +struct pmu; -extern void arch_install_thread_hw_breakpoint(struct task_struct *tsk); -extern void arch_uninstall_thread_hw_breakpoint(void); extern int arch_check_va_in_userspace(unsigned long va, u8 hbp_len); -extern int arch_validate_hwbkpt_settings(struct hw_breakpoint *bp, - struct task_struct *tsk); -extern void arch_update_user_hw_breakpoint(int pos, struct task_struct *tsk); -extern void arch_flush_thread_hw_breakpoint(struct task_struct *tsk); -extern void arch_update_kernel_hw_breakpoint(void *); +extern int arch_validate_hwbkpt_settings(struct perf_event *bp, + struct task_struct *tsk); extern int hw_breakpoint_exceptions_notify(struct notifier_block *unused, - unsigned long val, void *data); + unsigned long val, void *data); + + +int arch_install_hw_breakpoint(struct perf_event *bp); +void arch_uninstall_hw_breakpoint(struct perf_event *bp); +void hw_breakpoint_pmu_read(struct perf_event *bp); +void hw_breakpoint_pmu_unthrottle(struct perf_event *bp); + +extern void +arch_fill_perf_breakpoint(struct perf_event *bp); + +unsigned long encode_dr7(int drnum, unsigned int len, unsigned int type); +int decode_dr7(unsigned long dr7, int bpnum, unsigned *len, unsigned *type); + +extern int arch_bp_generic_fields(int x86_len, int x86_type, + int *gen_len, int *gen_type); + +extern struct pmu perf_ops_bp; + #endif /* __KERNEL__ */ #endif /* _I386_HW_BREAKPOINT_H */ diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h index 61aafb71c7ef..820f3000f736 100644 --- a/arch/x86/include/asm/processor.h +++ b/arch/x86/include/asm/processor.h @@ -423,6 +423,8 @@ extern unsigned int xstate_size; extern void free_thread_xstate(struct task_struct *); extern struct kmem_cache *task_xstate_cachep; +struct perf_event; + struct thread_struct { /* Cached TLS descriptors: */ struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; @@ -444,12 +446,10 @@ struct thread_struct { unsigned long fs; #endif unsigned long gs; - /* Hardware debugging registers: */ - unsigned long debugreg[HBP_NUM]; - unsigned long debugreg6; - unsigned long debugreg7; - /* Hardware breakpoint info */ - struct hw_breakpoint *hbp[HBP_NUM]; + /* Save middle states of ptrace breakpoints */ + struct perf_event *ptrace_bps[HBP_NUM]; + /* Debug status used for traps, single steps, etc... */ + unsigned long debugreg6; /* Fault info: */ unsigned long cr2; unsigned long trap_no; diff --git a/arch/x86/kernel/hw_breakpoint.c b/arch/x86/kernel/hw_breakpoint.c index 9316a9de4de3..e622620790bd 100644 --- a/arch/x86/kernel/hw_breakpoint.c +++ b/arch/x86/kernel/hw_breakpoint.c @@ -15,6 +15,7 @@ * * Copyright (C) 2007 Alan Stern * Copyright (C) 2009 IBM Corporation + * Copyright (C) 2009 Frederic Weisbecker */ /* @@ -22,6 +23,8 @@ * using the CPU's debug registers. */ +#include +#include #include #include #include @@ -38,26 +41,24 @@ #include #include -/* Unmasked kernel DR7 value */ -static unsigned long kdr7; +/* Per cpu debug control register value */ +DEFINE_PER_CPU(unsigned long, dr7); + +/* Per cpu debug address registers values */ +static DEFINE_PER_CPU(unsigned long, cpu_debugreg[HBP_NUM]); /* - * Masks for the bits corresponding to registers DR0 - DR3 in DR7 register. - * Used to clear and verify the status of bits corresponding to DR0 - DR3 + * Stores the breakpoints currently in use on each breakpoint address + * register for each cpus */ -static const unsigned long dr7_masks[HBP_NUM] = { - 0x000f0003, /* LEN0, R/W0, G0, L0 */ - 0x00f0000c, /* LEN1, R/W1, G1, L1 */ - 0x0f000030, /* LEN2, R/W2, G2, L2 */ - 0xf00000c0 /* LEN3, R/W3, G3, L3 */ -}; +static DEFINE_PER_CPU(struct perf_event *, bp_per_reg[HBP_NUM]); /* * Encode the length, type, Exact, and Enable bits for a particular breakpoint * as stored in debug register 7. */ -static unsigned long encode_dr7(int drnum, unsigned int len, unsigned int type) +unsigned long encode_dr7(int drnum, unsigned int len, unsigned int type) { unsigned long bp_info; @@ -68,64 +69,89 @@ static unsigned long encode_dr7(int drnum, unsigned int len, unsigned int type) return bp_info; } -void arch_update_kernel_hw_breakpoint(void *unused) +/* + * Decode the length and type bits for a particular breakpoint as + * stored in debug register 7. Return the "enabled" status. + */ +int decode_dr7(unsigned long dr7, int bpnum, unsigned *len, unsigned *type) { - struct hw_breakpoint *bp; - int i, cpu = get_cpu(); - unsigned long temp_kdr7 = 0; - - /* Don't allow debug exceptions while we update the registers */ - set_debugreg(0UL, 7); + int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE); - for (i = hbp_kernel_pos; i < HBP_NUM; i++) { - per_cpu(this_hbp_kernel[i], cpu) = bp = hbp_kernel[i]; - if (bp) { - temp_kdr7 |= encode_dr7(i, bp->info.len, bp->info.type); - set_debugreg(bp->info.address, i); - } - } + *len = (bp_info & 0xc) | 0x40; + *type = (bp_info & 0x3) | 0x80; - /* No need to set DR6. Update the debug registers with kernel-space - * breakpoint values from kdr7 and user-space requests from the - * current process - */ - kdr7 = temp_kdr7; - set_debugreg(kdr7 | current->thread.debugreg7, 7); - put_cpu(); + return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3; } /* - * Install the thread breakpoints in their debug registers. + * Install a perf counter breakpoint. + * + * We seek a free debug address register and use it for this + * breakpoint. Eventually we enable it in the debug control register. + * + * Atomic: we hold the counter->ctx->lock and we only handle variables + * and registers local to this cpu. */ -void arch_install_thread_hw_breakpoint(struct task_struct *tsk) +int arch_install_hw_breakpoint(struct perf_event *bp) { - struct thread_struct *thread = &(tsk->thread); - - switch (hbp_kernel_pos) { - case 4: - set_debugreg(thread->debugreg[3], 3); - case 3: - set_debugreg(thread->debugreg[2], 2); - case 2: - set_debugreg(thread->debugreg[1], 1); - case 1: - set_debugreg(thread->debugreg[0], 0); - default: - break; + struct arch_hw_breakpoint *info = counter_arch_bp(bp); + unsigned long *dr7; + int i; + + for (i = 0; i < HBP_NUM; i++) { + struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]); + + if (!*slot) { + *slot = bp; + break; + } } - /* No need to set DR6 */ - set_debugreg((kdr7 | thread->debugreg7), 7); + if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot")) + return -EBUSY; + + set_debugreg(info->address, i); + __get_cpu_var(cpu_debugreg[i]) = info->address; + + dr7 = &__get_cpu_var(dr7); + *dr7 |= encode_dr7(i, info->len, info->type); + + set_debugreg(*dr7, 7); + + return 0; } /* - * Install the debug register values for just the kernel, no thread. + * Uninstall the breakpoint contained in the given counter. + * + * First we search the debug address register it uses and then we disable + * it. + * + * Atomic: we hold the counter->ctx->lock and we only handle variables + * and registers local to this cpu. */ -void arch_uninstall_thread_hw_breakpoint(void) +void arch_uninstall_hw_breakpoint(struct perf_event *bp) { - /* Clear the user-space portion of debugreg7 by setting only kdr7 */ - set_debugreg(kdr7, 7); + struct arch_hw_breakpoint *info = counter_arch_bp(bp); + unsigned long *dr7; + int i; + + for (i = 0; i < HBP_NUM; i++) { + struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]); + + if (*slot == bp) { + *slot = NULL; + break; + } + } + + if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot")) + return; + dr7 = &__get_cpu_var(dr7); + *dr7 &= ~encode_dr7(i, info->len, info->type); + + set_debugreg(*dr7, 7); } static int get_hbp_len(u8 hbp_len) @@ -133,17 +159,17 @@ static int get_hbp_len(u8 hbp_len) unsigned int len_in_bytes = 0; switch (hbp_len) { - case HW_BREAKPOINT_LEN_1: + case X86_BREAKPOINT_LEN_1: len_in_bytes = 1; break; - case HW_BREAKPOINT_LEN_2: + case X86_BREAKPOINT_LEN_2: len_in_bytes = 2; break; - case HW_BREAKPOINT_LEN_4: + case X86_BREAKPOINT_LEN_4: len_in_bytes = 4; break; #ifdef CONFIG_X86_64 - case HW_BREAKPOINT_LEN_8: + case X86_BREAKPOINT_LEN_8: len_in_bytes = 8; break; #endif @@ -178,67 +204,146 @@ static int arch_check_va_in_kernelspace(unsigned long va, u8 hbp_len) /* * Store a breakpoint's encoded address, length, and type. */ -static int arch_store_info(struct hw_breakpoint *bp, struct task_struct *tsk) +static int arch_store_info(struct perf_event *bp) { - /* - * User-space requests will always have the address field populated - * Symbol names from user-space are rejected - */ - if (tsk && bp->info.name) - return -EINVAL; + struct arch_hw_breakpoint *info = counter_arch_bp(bp); /* * For kernel-addresses, either the address or symbol name can be * specified. */ - if (bp->info.name) - bp->info.address = (unsigned long) - kallsyms_lookup_name(bp->info.name); - if (bp->info.address) + if (info->name) + info->address = (unsigned long) + kallsyms_lookup_name(info->name); + if (info->address) return 0; + return -EINVAL; } -/* - * Validate the arch-specific HW Breakpoint register settings - */ -int arch_validate_hwbkpt_settings(struct hw_breakpoint *bp, - struct task_struct *tsk) +int arch_bp_generic_fields(int x86_len, int x86_type, + int *gen_len, int *gen_type) { - unsigned int align; - int ret = -EINVAL; + /* Len */ + switch (x86_len) { + case X86_BREAKPOINT_LEN_1: + *gen_len = HW_BREAKPOINT_LEN_1; + break; + case X86_BREAKPOINT_LEN_2: + *gen_len = HW_BREAKPOINT_LEN_2; + break; + case X86_BREAKPOINT_LEN_4: + *gen_len = HW_BREAKPOINT_LEN_4; + break; +#ifdef CONFIG_X86_64 + case X86_BREAKPOINT_LEN_8: + *gen_len = HW_BREAKPOINT_LEN_8; + break; +#endif + default: + return -EINVAL; + } - switch (bp->info.type) { - /* - * Ptrace-refactoring code - * For now, we'll allow instruction breakpoint only for user-space - * addresses - */ - case HW_BREAKPOINT_EXECUTE: - if ((!arch_check_va_in_userspace(bp->info.address, - bp->info.len)) && - bp->info.len != HW_BREAKPOINT_LEN_EXECUTE) - return ret; + /* Type */ + switch (x86_type) { + case X86_BREAKPOINT_EXECUTE: + *gen_type = HW_BREAKPOINT_X; break; - case HW_BREAKPOINT_WRITE: + case X86_BREAKPOINT_WRITE: + *gen_type = HW_BREAKPOINT_W; break; - case HW_BREAKPOINT_RW: + case X86_BREAKPOINT_RW: + *gen_type = HW_BREAKPOINT_W | HW_BREAKPOINT_R; break; default: - return ret; + return -EINVAL; } - switch (bp->info.len) { + return 0; +} + + +static int arch_build_bp_info(struct perf_event *bp) +{ + struct arch_hw_breakpoint *info = counter_arch_bp(bp); + + info->address = bp->attr.bp_addr; + + /* Len */ + switch (bp->attr.bp_len) { case HW_BREAKPOINT_LEN_1: - align = 0; + info->len = X86_BREAKPOINT_LEN_1; break; case HW_BREAKPOINT_LEN_2: - align = 1; + info->len = X86_BREAKPOINT_LEN_2; break; case HW_BREAKPOINT_LEN_4: - align = 3; + info->len = X86_BREAKPOINT_LEN_4; break; #ifdef CONFIG_X86_64 case HW_BREAKPOINT_LEN_8: + info->len = X86_BREAKPOINT_LEN_8; + break; +#endif + default: + return -EINVAL; + } + + /* Type */ + switch (bp->attr.bp_type) { + case HW_BREAKPOINT_W: + info->type = X86_BREAKPOINT_WRITE; + break; + case HW_BREAKPOINT_W | HW_BREAKPOINT_R: + info->type = X86_BREAKPOINT_RW; + break; + case HW_BREAKPOINT_X: + info->type = X86_BREAKPOINT_EXECUTE; + break; + default: + return -EINVAL; + } + + return 0; +} +/* + * Validate the arch-specific HW Breakpoint register settings + */ +int arch_validate_hwbkpt_settings(struct perf_event *bp, + struct task_struct *tsk) +{ + struct arch_hw_breakpoint *info = counter_arch_bp(bp); + unsigned int align; + int ret; + + + ret = arch_build_bp_info(bp); + if (ret) + return ret; + + ret = -EINVAL; + + if (info->type == X86_BREAKPOINT_EXECUTE) + /* + * Ptrace-refactoring code + * For now, we'll allow instruction breakpoint only for user-space + * addresses + */ + if ((!arch_check_va_in_userspace(info->address, info->len)) && + info->len != X86_BREAKPOINT_EXECUTE) + return ret; + + switch (info->len) { + case X86_BREAKPOINT_LEN_1: + align = 0; + break; + case X86_BREAKPOINT_LEN_2: + align = 1; + break; + case X86_BREAKPOINT_LEN_4: + align = 3; + break; +#ifdef CONFIG_X86_64 + case X86_BREAKPOINT_LEN_8: align = 7; break; #endif @@ -246,8 +351,8 @@ int arch_validate_hwbkpt_settings(struct hw_breakpoint *bp, return ret; } - if (bp->triggered) - ret = arch_store_info(bp, tsk); + if (bp->callback) + ret = arch_store_info(bp); if (ret < 0) return ret; @@ -255,44 +360,47 @@ int arch_validate_hwbkpt_settings(struct hw_breakpoint *bp, * Check that the low-order bits of the address are appropriate * for the alignment implied by len. */ - if (bp->info.address & align) + if (info->address & align) return -EINVAL; /* Check that the virtual address is in the proper range */ if (tsk) { - if (!arch_check_va_in_userspace(bp->info.address, bp->info.len)) + if (!arch_check_va_in_userspace(info->address, info->len)) return -EFAULT; } else { - if (!arch_check_va_in_kernelspace(bp->info.address, - bp->info.len)) + if (!arch_check_va_in_kernelspace(info->address, info->len)) return -EFAULT; } + return 0; } -void arch_update_user_hw_breakpoint(int pos, struct task_struct *tsk) +/* + * Release the user breakpoints used by ptrace + */ +void flush_ptrace_hw_breakpoint(struct task_struct *tsk) { - struct thread_struct *thread = &(tsk->thread); - struct hw_breakpoint *bp = thread->hbp[pos]; - - thread->debugreg7 &= ~dr7_masks[pos]; - if (bp) { - thread->debugreg[pos] = bp->info.address; - thread->debugreg7 |= encode_dr7(pos, bp->info.len, - bp->info.type); - } else - thread->debugreg[pos] = 0; + int i; + struct thread_struct *t = &tsk->thread; + + for (i = 0; i < HBP_NUM; i++) { + unregister_hw_breakpoint(t->ptrace_bps[i]); + t->ptrace_bps[i] = NULL; + } } -void arch_flush_thread_hw_breakpoint(struct task_struct *tsk) +#ifdef CONFIG_KVM +void hw_breakpoint_restore(void) { - int i; - struct thread_struct *thread = &(tsk->thread); - - thread->debugreg7 = 0; - for (i = 0; i < HBP_NUM; i++) - thread->debugreg[i] = 0; + set_debugreg(__get_cpu_var(cpu_debugreg[0]), 0); + set_debugreg(__get_cpu_var(cpu_debugreg[1]), 1); + set_debugreg(__get_cpu_var(cpu_debugreg[2]), 2); + set_debugreg(__get_cpu_var(cpu_debugreg[3]), 3); + set_debugreg(current->thread.debugreg6, 6); + set_debugreg(__get_cpu_var(dr7), 7); } +EXPORT_SYMBOL_GPL(hw_breakpoint_restore); +#endif /* * Handle debug exception notifications. @@ -313,7 +421,7 @@ void arch_flush_thread_hw_breakpoint(struct task_struct *tsk) static int __kprobes hw_breakpoint_handler(struct die_args *args) { int i, cpu, rc = NOTIFY_STOP; - struct hw_breakpoint *bp; + struct perf_event *bp; unsigned long dr7, dr6; unsigned long *dr6_p; @@ -325,10 +433,6 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args) if ((dr6 & DR_TRAP_BITS) == 0) return NOTIFY_DONE; - /* Lazy debug register switching */ - if (!test_tsk_thread_flag(current, TIF_DEBUG)) - arch_uninstall_thread_hw_breakpoint(); - get_debugreg(dr7, 7); /* Disable breakpoints during exception handling */ set_debugreg(0UL, 7); @@ -344,17 +448,18 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args) for (i = 0; i < HBP_NUM; ++i) { if (likely(!(dr6 & (DR_TRAP0 << i)))) continue; + /* - * Find the corresponding hw_breakpoint structure and - * invoke its triggered callback. + * The counter may be concurrently released but that can only + * occur from a call_rcu() path. We can then safely fetch + * the breakpoint, use its callback, touch its counter + * while we are in an rcu_read_lock() path. */ - if (i >= hbp_kernel_pos) - bp = per_cpu(this_hbp_kernel[i], cpu); - else { - bp = current->thread.hbp[i]; - if (bp) - rc = NOTIFY_DONE; - } + rcu_read_lock(); + + bp = per_cpu(bp_per_reg[i], cpu); + if (bp) + rc = NOTIFY_DONE; /* * Reset the 'i'th TRAP bit in dr6 to denote completion of * exception handling @@ -362,19 +467,23 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args) (*dr6_p) &= ~(DR_TRAP0 << i); /* * bp can be NULL due to lazy debug register switching - * or due to the delay between updates of hbp_kernel_pos - * and this_hbp_kernel. + * or due to concurrent perf counter removing. */ - if (!bp) - continue; + if (!bp) { + rcu_read_unlock(); + break; + } + + (bp->callback)(bp, args->regs); - (bp->triggered)(bp, args->regs); + rcu_read_unlock(); } if (dr6 & (~DR_TRAP_BITS)) rc = NOTIFY_DONE; set_debugreg(dr7, 7); put_cpu(); + return rc; } @@ -389,3 +498,13 @@ int __kprobes hw_breakpoint_exceptions_notify( return hw_breakpoint_handler(data); } + +void hw_breakpoint_pmu_read(struct perf_event *bp) +{ + /* TODO */ +} + +void hw_breakpoint_pmu_unthrottle(struct perf_event *bp) +{ + /* TODO */ +} diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c index cf8ee0016307..744508e7cfdd 100644 --- a/arch/x86/kernel/process.c +++ b/arch/x86/kernel/process.c @@ -10,6 +10,7 @@ #include #include #include +#include #include #include #include @@ -18,7 +19,6 @@ #include #include #include -#include unsigned long idle_halt; EXPORT_SYMBOL(idle_halt); @@ -47,8 +47,6 @@ void free_thread_xstate(struct task_struct *tsk) kmem_cache_free(task_xstate_cachep, tsk->thread.xstate); tsk->thread.xstate = NULL; } - if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG))) - flush_thread_hw_breakpoint(tsk); WARN(tsk->thread.ds_ctx, "leaking DS context\n"); } @@ -107,8 +105,7 @@ void flush_thread(void) } #endif - if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG))) - flush_thread_hw_breakpoint(tsk); + flush_ptrace_hw_breakpoint(tsk); memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); /* * Forget coprocessor state.. diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c index 209e74801763..d5bd3132ee70 100644 --- a/arch/x86/kernel/process_32.c +++ b/arch/x86/kernel/process_32.c @@ -59,7 +59,6 @@ #include #include #include -#include asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); @@ -264,9 +263,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp, p->thread.io_bitmap_ptr = NULL; tsk = current; err = -ENOMEM; - if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG))) - if (copy_thread_hw_breakpoint(tsk, p, clone_flags)) - goto out; + + memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, @@ -287,13 +285,10 @@ int copy_thread(unsigned long clone_flags, unsigned long sp, err = do_set_thread_area(p, -1, (struct user_desc __user *)childregs->si, 0); -out: if (err && p->thread.io_bitmap_ptr) { kfree(p->thread.io_bitmap_ptr); p->thread.io_bitmap_max = 0; } - if (err) - flush_thread_hw_breakpoint(p); clear_tsk_thread_flag(p, TIF_DS_AREA_MSR); p->thread.ds_ctx = NULL; @@ -437,23 +432,6 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p) lazy_load_gs(next->gs); percpu_write(current_task, next_p); - /* - * There's a problem with moving the arch_install_thread_hw_breakpoint() - * call before current is updated. Suppose a kernel breakpoint is - * triggered in between the two, the hw-breakpoint handler will see that - * the 'current' task does not have TIF_DEBUG flag set and will think it - * is leftover from an old task (lazy switching) and will erase it. Then - * until the next context switch, no user-breakpoints will be installed. - * - * The real problem is that it's impossible to update both current and - * physical debug registers at the same instant, so there will always be - * a window in which they disagree and a breakpoint might get triggered. - * Since we use lazy switching, we are forced to assume that a - * disagreement means that current is correct and the exception is due - * to lazy debug register switching. - */ - if (unlikely(test_tsk_thread_flag(next_p, TIF_DEBUG))) - arch_install_thread_hw_breakpoint(next_p); return prev_p; } diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c index 72edac026a78..5bafdec34441 100644 --- a/arch/x86/kernel/process_64.c +++ b/arch/x86/kernel/process_64.c @@ -53,7 +53,6 @@ #include #include #include -#include asmlinkage extern void ret_from_fork(void); @@ -244,8 +243,6 @@ void release_thread(struct task_struct *dead_task) BUG(); } } - if (unlikely(dead_task->thread.debugreg7)) - flush_thread_hw_breakpoint(dead_task); } static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr) @@ -309,9 +306,7 @@ int copy_thread(unsigned long clone_flags, unsigned long sp, savesegment(ds, p->thread.ds); err = -ENOMEM; - if (unlikely(test_tsk_thread_flag(me, TIF_DEBUG))) - if (copy_thread_hw_breakpoint(me, p, clone_flags)) - goto out; + memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); @@ -351,8 +346,6 @@ out: kfree(p->thread.io_bitmap_ptr); p->thread.io_bitmap_max = 0; } - if (err) - flush_thread_hw_breakpoint(p); return err; } @@ -508,23 +501,6 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p) */ if (preload_fpu) __math_state_restore(); - /* - * There's a problem with moving the arch_install_thread_hw_breakpoint() - * call before current is updated. Suppose a kernel breakpoint is - * triggered in between the two, the hw-breakpoint handler will see that - * the 'current' task does not have TIF_DEBUG flag set and will think it - * is leftover from an old task (lazy switching) and will erase it. Then - * until the next context switch, no user-breakpoints will be installed. - * - * The real problem is that it's impossible to update both current and - * physical debug registers at the same instant, so there will always be - * a window in which they disagree and a breakpoint might get triggered. - * Since we use lazy switching, we are forced to assume that a - * disagreement means that current is correct and the exception is due - * to lazy debug register switching. - */ - if (unlikely(test_tsk_thread_flag(next_p, TIF_DEBUG))) - arch_install_thread_hw_breakpoint(next_p); return prev_p; } diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index 267cb85b479c..e79610d95971 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -22,6 +22,8 @@ #include #include #include +#include +#include #include #include @@ -441,54 +443,59 @@ static int genregs_set(struct task_struct *target, return ret; } -/* - * Decode the length and type bits for a particular breakpoint as - * stored in debug register 7. Return the "enabled" status. - */ -static int decode_dr7(unsigned long dr7, int bpnum, unsigned *len, - unsigned *type) -{ - int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE); - - *len = (bp_info & 0xc) | 0x40; - *type = (bp_info & 0x3) | 0x80; - return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3; -} - -static void ptrace_triggered(struct hw_breakpoint *bp, struct pt_regs *regs) +static void ptrace_triggered(struct perf_event *bp, void *data) { - struct thread_struct *thread = &(current->thread); int i; + struct thread_struct *thread = &(current->thread); /* * Store in the virtual DR6 register the fact that the breakpoint * was hit so the thread's debugger will see it. */ - for (i = 0; i < hbp_kernel_pos; i++) - /* - * We will check bp->info.address against the address stored in - * thread's hbp structure and not debugreg[i]. This is to ensure - * that the corresponding bit for 'i' in DR7 register is enabled - */ - if (bp->info.address == thread->hbp[i]->info.address) + for (i = 0; i < HBP_NUM; i++) { + if (thread->ptrace_bps[i] == bp) break; + } thread->debugreg6 |= (DR_TRAP0 << i); } +/* + * Walk through every ptrace breakpoints for this thread and + * build the dr7 value on top of their attributes. + * + */ +static unsigned long ptrace_get_dr7(struct perf_event *bp[]) +{ + int i; + int dr7 = 0; + struct arch_hw_breakpoint *info; + + for (i = 0; i < HBP_NUM; i++) { + if (bp[i] && !bp[i]->attr.disabled) { + info = counter_arch_bp(bp[i]); + dr7 |= encode_dr7(i, info->len, info->type); + } + } + + return dr7; +} + /* * Handle ptrace writes to debug register 7. */ static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data) { struct thread_struct *thread = &(tsk->thread); - unsigned long old_dr7 = thread->debugreg7; + unsigned long old_dr7; int i, orig_ret = 0, rc = 0; int enabled, second_pass = 0; unsigned len, type; - struct hw_breakpoint *bp; + int gen_len, gen_type; + struct perf_event *bp; data &= ~DR_CONTROL_RESERVED; + old_dr7 = ptrace_get_dr7(thread->ptrace_bps); restore: /* * Loop through all the hardware breakpoints, making the @@ -496,11 +503,12 @@ restore: */ for (i = 0; i < HBP_NUM; i++) { enabled = decode_dr7(data, i, &len, &type); - bp = thread->hbp[i]; + bp = thread->ptrace_bps[i]; if (!enabled) { if (bp) { - /* Don't unregister the breakpoints right-away, + /* + * Don't unregister the breakpoints right-away, * unless all register_user_hw_breakpoint() * requests have succeeded. This prevents * any window of opportunity for debug @@ -508,27 +516,45 @@ restore: */ if (!second_pass) continue; - unregister_user_hw_breakpoint(tsk, bp); - kfree(bp); + thread->ptrace_bps[i] = NULL; + unregister_hw_breakpoint(bp); } continue; } + + /* + * We shoud have at least an inactive breakpoint at this + * slot. It means the user is writing dr7 without having + * written the address register first + */ if (!bp) { - rc = -ENOMEM; - bp = kzalloc(sizeof(struct hw_breakpoint), GFP_KERNEL); - if (bp) { - bp->info.address = thread->debugreg[i]; - bp->triggered = ptrace_triggered; - bp->info.len = len; - bp->info.type = type; - rc = register_user_hw_breakpoint(tsk, bp); - if (rc) - kfree(bp); - } - } else - rc = modify_user_hw_breakpoint(tsk, bp); + rc = -EINVAL; + break; + } + + rc = arch_bp_generic_fields(len, type, &gen_len, &gen_type); if (rc) break; + + /* + * This is a temporary thing as bp is unregistered/registered + * to simulate modification + */ + bp = modify_user_hw_breakpoint(bp, bp->attr.bp_addr, gen_len, + gen_type, bp->callback, + tsk, true); + thread->ptrace_bps[i] = NULL; + + if (!bp) { /* incorrect bp, or we have a bug in bp API */ + rc = -EINVAL; + break; + } + if (IS_ERR(bp)) { + rc = PTR_ERR(bp); + bp = NULL; + break; + } + thread->ptrace_bps[i] = bp; } /* * Make a second pass to free the remaining unused breakpoints @@ -553,15 +579,63 @@ static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n) struct thread_struct *thread = &(tsk->thread); unsigned long val = 0; - if (n < HBP_NUM) - val = thread->debugreg[n]; - else if (n == 6) + if (n < HBP_NUM) { + struct perf_event *bp; + bp = thread->ptrace_bps[n]; + if (!bp) + return 0; + val = bp->hw.info.address; + } else if (n == 6) { val = thread->debugreg6; - else if (n == 7) - val = thread->debugreg7; + } else if (n == 7) { + val = ptrace_get_dr7(thread->ptrace_bps); + } return val; } +static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr, + unsigned long addr) +{ + struct perf_event *bp; + struct thread_struct *t = &tsk->thread; + + if (!t->ptrace_bps[nr]) { + /* + * Put stub len and type to register (reserve) an inactive but + * correct bp + */ + bp = register_user_hw_breakpoint(addr, HW_BREAKPOINT_LEN_1, + HW_BREAKPOINT_W, + ptrace_triggered, tsk, + false); + } else { + bp = t->ptrace_bps[nr]; + t->ptrace_bps[nr] = NULL; + bp = modify_user_hw_breakpoint(bp, addr, bp->attr.bp_len, + bp->attr.bp_type, + bp->callback, + tsk, + bp->attr.disabled); + } + + if (!bp) + return -EIO; + /* + * CHECKME: the previous code returned -EIO if the addr wasn't a + * valid task virtual addr. The new one will return -EINVAL in this + * case. + * -EINVAL may be what we want for in-kernel breakpoints users, but + * -EIO looks better for ptrace, since we refuse a register writing + * for the user. And anyway this is the previous behaviour. + */ + if (IS_ERR(bp)) + return PTR_ERR(bp); + + t->ptrace_bps[nr] = bp; + + return 0; +} + /* * Handle PTRACE_POKEUSR calls for the debug register area. */ @@ -575,19 +649,13 @@ int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val) return -EIO; if (n == 6) { - tsk->thread.debugreg6 = val; + thread->debugreg6 = val; goto ret_path; } if (n < HBP_NUM) { - if (thread->hbp[n]) { - if (arch_check_va_in_userspace(val, - thread->hbp[n]->info.len) == 0) { - rc = -EIO; - goto ret_path; - } - thread->hbp[n]->info.address = val; - } - thread->debugreg[n] = val; + rc = ptrace_set_breakpoint_addr(tsk, n, val); + if (rc) + return rc; } /* All that's left is DR7 */ if (n == 7) diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index 213a7a3e4562..565ebc65920e 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -64,7 +64,6 @@ #include #include #include -#include #include #include @@ -328,7 +327,6 @@ notrace static void __cpuinit start_secondary(void *unused) x86_cpuinit.setup_percpu_clockev(); wmb(); - load_debug_registers(); cpu_idle(); } @@ -1269,7 +1267,6 @@ void cpu_disable_common(void) remove_cpu_from_maps(cpu); unlock_vector_lock(); fixup_irqs(); - hw_breakpoint_disable(); } int native_cpu_disable(void) diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index fc2974adf9b6..22dee7aa7813 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -42,6 +42,7 @@ #define CREATE_TRACE_POINTS #include "trace.h" +#include #include #include #include @@ -3643,14 +3644,15 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) trace_kvm_entry(vcpu->vcpu_id); kvm_x86_ops->run(vcpu, kvm_run); - if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) { - set_debugreg(current->thread.debugreg[0], 0); - set_debugreg(current->thread.debugreg[1], 1); - set_debugreg(current->thread.debugreg[2], 2); - set_debugreg(current->thread.debugreg[3], 3); - set_debugreg(current->thread.debugreg6, 6); - set_debugreg(current->thread.debugreg7, 7); - } + /* + * If the guest has used debug registers, at least dr7 + * will be disabled while returning to the host. + * If we don't have active breakpoints in the host, we don't + * care about the messed up debug address registers. But if + * we have some of them active, restore the old state. + */ + if (__get_cpu_var(dr7) & DR_GLOBAL_ENABLE_MASK) + hw_breakpoint_restore(); set_bit(KVM_REQ_KICK, &vcpu->requests); local_irq_enable(); diff --git a/arch/x86/power/cpu.c b/arch/x86/power/cpu.c index e09a44fc4664..0a979f3e5b8a 100644 --- a/arch/x86/power/cpu.c +++ b/arch/x86/power/cpu.c @@ -105,7 +105,6 @@ static void __save_processor_state(struct saved_context *ctxt) ctxt->cr4 = read_cr4(); ctxt->cr8 = read_cr8(); #endif - hw_breakpoint_disable(); } /* Needed by apm.c */ @@ -144,11 +143,6 @@ static void fix_processor_context(void) #endif load_TR_desc(); /* This does ltr */ load_LDT(¤t->active_mm->context); /* This does lldt */ - - /* - * Now maybe reload the debug registers - */ - load_debug_registers(); } /** diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h index 61ccc8f17eac..7eba9b92e5f3 100644 --- a/include/linux/hw_breakpoint.h +++ b/include/linux/hw_breakpoint.h @@ -1,136 +1,131 @@ #ifndef _LINUX_HW_BREAKPOINT_H #define _LINUX_HW_BREAKPOINT_H +#include -#ifdef __KERNEL__ -#include -#include -#include - -/** - * struct hw_breakpoint - unified kernel/user-space hardware breakpoint - * @triggered: callback invoked after target address access - * @info: arch-specific breakpoint info (address, length, and type) - * - * %hw_breakpoint structures are the kernel's way of representing - * hardware breakpoints. These are data breakpoints - * (also known as "watchpoints", triggered on data access), and the breakpoint's - * target address can be located in either kernel space or user space. - * - * The breakpoint's address, length, and type are highly - * architecture-specific. The values are encoded in the @info field; you - * specify them when registering the breakpoint. To examine the encoded - * values use hw_breakpoint_get_{kaddress,uaddress,len,type}(), declared - * below. - * - * The address is specified as a regular kernel pointer (for kernel-space - * breakponts) or as an %__user pointer (for user-space breakpoints). - * With register_user_hw_breakpoint(), the address must refer to a - * location in user space. The breakpoint will be active only while the - * requested task is running. Conversely with - * register_kernel_hw_breakpoint(), the address must refer to a location - * in kernel space, and the breakpoint will be active on all CPUs - * regardless of the current task. - * - * The length is the breakpoint's extent in bytes, which is subject to - * certain limitations. include/asm/hw_breakpoint.h contains macros - * defining the available lengths for a specific architecture. Note that - * the address's alignment must match the length. The breakpoint will - * catch accesses to any byte in the range from address to address + - * (length - 1). - * - * The breakpoint's type indicates the sort of access that will cause it - * to trigger. Possible values may include: - * - * %HW_BREAKPOINT_RW (triggered on read or write access), - * %HW_BREAKPOINT_WRITE (triggered on write access), and - * %HW_BREAKPOINT_READ (triggered on read access). - * - * Appropriate macros are defined in include/asm/hw_breakpoint.h; not all - * possibilities are available on all architectures. Execute breakpoints - * must have length equal to the special value %HW_BREAKPOINT_LEN_EXECUTE. - * - * When a breakpoint gets hit, the @triggered callback is - * invoked in_interrupt with a pointer to the %hw_breakpoint structure and the - * processor registers. - * Data breakpoints occur after the memory access has taken place. - * Breakpoints are disabled during execution @triggered, to avoid - * recursive traps and allow unhindered access to breakpointed memory. - * - * This sample code sets a breakpoint on pid_max and registers a callback - * function for writes to that variable. Note that it is not portable - * as written, because not all architectures support HW_BREAKPOINT_LEN_4. - * - * ---------------------------------------------------------------------- - * - * #include - * - * struct hw_breakpoint my_bp; - * - * static void my_triggered(struct hw_breakpoint *bp, struct pt_regs *regs) - * { - * printk(KERN_DEBUG "Inside triggered routine of breakpoint exception\n"); - * dump_stack(); - * ............... - * } - * - * static struct hw_breakpoint my_bp; - * - * static int init_module(void) - * { - * ...................... - * my_bp.info.type = HW_BREAKPOINT_WRITE; - * my_bp.info.len = HW_BREAKPOINT_LEN_4; - * - * my_bp.installed = (void *)my_bp_installed; - * - * rc = register_kernel_hw_breakpoint(&my_bp); - * ...................... - * } - * - * static void cleanup_module(void) - * { - * ...................... - * unregister_kernel_hw_breakpoint(&my_bp); - * ...................... - * } - * - * ---------------------------------------------------------------------- - */ -struct hw_breakpoint { - void (*triggered)(struct hw_breakpoint *, struct pt_regs *); - struct arch_hw_breakpoint info; +enum { + HW_BREAKPOINT_LEN_1 = 1, + HW_BREAKPOINT_LEN_2 = 2, + HW_BREAKPOINT_LEN_4 = 4, + HW_BREAKPOINT_LEN_8 = 8, }; -/* - * len and type values are defined in include/asm/hw_breakpoint.h. - * Available values vary according to the architecture. On i386 the - * possibilities are: - * - * HW_BREAKPOINT_LEN_1 - * HW_BREAKPOINT_LEN_2 - * HW_BREAKPOINT_LEN_4 - * HW_BREAKPOINT_RW - * HW_BREAKPOINT_READ - * - * On other architectures HW_BREAKPOINT_LEN_8 may be available, and the - * 1-, 2-, and 4-byte lengths may be unavailable. There also may be - * HW_BREAKPOINT_WRITE. You can use #ifdef to check at compile time. - */ +enum { + HW_BREAKPOINT_R = 1, + HW_BREAKPOINT_W = 2, + HW_BREAKPOINT_X = 4, +}; + +static inline struct arch_hw_breakpoint *counter_arch_bp(struct perf_event *bp) +{ + return &bp->hw.info; +} + +static inline unsigned long hw_breakpoint_addr(struct perf_event *bp) +{ + return bp->attr.bp_addr; +} + +static inline int hw_breakpoint_type(struct perf_event *bp) +{ + return bp->attr.bp_type; +} + +static inline int hw_breakpoint_len(struct perf_event *bp) +{ + return bp->attr.bp_len; +} + +#ifdef CONFIG_HAVE_HW_BREAKPOINT +extern struct perf_event * +register_user_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active); + +/* FIXME: only change from the attr, and don't unregister */ +extern struct perf_event * +modify_user_hw_breakpoint(struct perf_event *bp, + unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active); -extern int register_user_hw_breakpoint(struct task_struct *tsk, - struct hw_breakpoint *bp); -extern int modify_user_hw_breakpoint(struct task_struct *tsk, - struct hw_breakpoint *bp); -extern void unregister_user_hw_breakpoint(struct task_struct *tsk, - struct hw_breakpoint *bp); /* * Kernel breakpoints are not associated with any particular thread. */ -extern int register_kernel_hw_breakpoint(struct hw_breakpoint *bp); -extern void unregister_kernel_hw_breakpoint(struct hw_breakpoint *bp); +extern struct perf_event * +register_wide_hw_breakpoint_cpu(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + int cpu, + bool active); + +extern struct perf_event ** +register_wide_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + bool active); + +extern int register_perf_hw_breakpoint(struct perf_event *bp); +extern int __register_perf_hw_breakpoint(struct perf_event *bp); +extern void unregister_hw_breakpoint(struct perf_event *bp); +extern void unregister_wide_hw_breakpoint(struct perf_event **cpu_events); + +extern int reserve_bp_slot(struct perf_event *bp); +extern void release_bp_slot(struct perf_event *bp); + +extern void flush_ptrace_hw_breakpoint(struct task_struct *tsk); + +#else /* !CONFIG_HAVE_HW_BREAKPOINT */ + +static inline struct perf_event * +register_user_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active) { return NULL; } +static inline struct perf_event * +modify_user_hw_breakpoint(struct perf_event *bp, + unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active) { return NULL; } +static inline struct perf_event * +register_wide_hw_breakpoint_cpu(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + int cpu, + bool active) { return NULL; } +static inline struct perf_event ** +register_wide_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + bool active) { return NULL; } +static inline int +register_perf_hw_breakpoint(struct perf_event *bp) { return -ENOSYS; } +static inline int +__register_perf_hw_breakpoint(struct perf_event *bp) { return -ENOSYS; } +static inline void unregister_hw_breakpoint(struct perf_event *bp) { } +static inline void +unregister_wide_hw_breakpoint(struct perf_event **cpu_events) { } +static inline int +reserve_bp_slot(struct perf_event *bp) {return -ENOSYS; } +static inline void release_bp_slot(struct perf_event *bp) { } + +static inline void flush_ptrace_hw_breakpoint(struct task_struct *tsk) { } -extern unsigned int hbp_kernel_pos; +#endif /* CONFIG_HAVE_HW_BREAKPOINT */ -#endif /* __KERNEL__ */ -#endif /* _LINUX_HW_BREAKPOINT_H */ +#endif /* _LINUX_HW_BREAKPOINT_H */ diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 8d54e6d25eeb..cead64ea6c15 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -18,6 +18,10 @@ #include #include +#ifdef CONFIG_HAVE_HW_BREAKPOINT +#include +#endif + /* * User-space ABI bits: */ @@ -31,6 +35,7 @@ enum perf_type_id { PERF_TYPE_TRACEPOINT = 2, PERF_TYPE_HW_CACHE = 3, PERF_TYPE_RAW = 4, + PERF_TYPE_BREAKPOINT = 5, PERF_TYPE_MAX, /* non-ABI */ }; @@ -207,6 +212,15 @@ struct perf_event_attr { __u32 wakeup_events; /* wakeup every n events */ __u32 wakeup_watermark; /* bytes before wakeup */ }; + + union { + struct { /* Hardware breakpoint info */ + __u64 bp_addr; + __u32 bp_type; + __u32 bp_len; + }; + }; + __u32 __reserved_2; __u64 __reserved_3; @@ -476,6 +490,11 @@ struct hw_perf_event { atomic64_t count; struct hrtimer hrtimer; }; +#ifdef CONFIG_HAVE_HW_BREAKPOINT + union { /* breakpoint */ + struct arch_hw_breakpoint info; + }; +#endif }; atomic64_t prev_count; u64 sample_period; @@ -588,7 +607,7 @@ struct perf_event { u64 tstamp_running; u64 tstamp_stopped; - struct perf_event_attr attr; + struct perf_event_attr attr; struct hw_perf_event hw; struct perf_event_context *ctx; @@ -643,6 +662,8 @@ struct perf_event { perf_callback_t callback; + perf_callback_t event_callback; + #endif /* CONFIG_PERF_EVENTS */ }; @@ -831,6 +852,7 @@ extern int sysctl_perf_event_sample_rate; extern void perf_event_init(void); extern void perf_tp_event(int event_id, u64 addr, u64 count, void *record, int entry_size); +extern void perf_bp_event(struct perf_event *event, void *data); #ifndef perf_misc_flags #define perf_misc_flags(regs) (user_mode(regs) ? PERF_RECORD_MISC_USER : \ @@ -865,6 +887,8 @@ static inline int perf_event_task_enable(void) { return -EINVAL; } static inline void perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { } +static inline void +perf_bp_event(struct perf_event *event, void *data) { } static inline void perf_event_mmap(struct vm_area_struct *vma) { } static inline void perf_event_comm(struct task_struct *tsk) { } diff --git a/kernel/exit.c b/kernel/exit.c index e61891f80123..266f8920628a 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -49,6 +49,7 @@ #include #include #include +#include #include #include @@ -979,6 +980,10 @@ NORET_TYPE void do_exit(long code) proc_exit_connector(tsk); + /* + * 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. diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c index c1f64e65a9f3..08f6d0163201 100644 --- a/kernel/hw_breakpoint.c +++ b/kernel/hw_breakpoint.c @@ -15,6 +15,7 @@ * * Copyright (C) 2007 Alan Stern * Copyright (C) IBM Corporation, 2009 + * Copyright (C) 2009, Frederic Weisbecker */ /* @@ -35,334 +36,242 @@ #include #include -#include +#include + #include #ifdef CONFIG_X86 #include #endif -/* - * Spinlock that protects all (un)register operations over kernel/user-space - * breakpoint requests - */ -static DEFINE_SPINLOCK(hw_breakpoint_lock); - -/* Array of kernel-space breakpoint structures */ -struct hw_breakpoint *hbp_kernel[HBP_NUM]; - -/* - * Per-processor copy of hbp_kernel[]. Used only when hbp_kernel is being - * modified but we need the older copy to handle any hbp exceptions. It will - * sync with hbp_kernel[] value after updation is done through IPIs. - */ -DEFINE_PER_CPU(struct hw_breakpoint*, this_hbp_kernel[HBP_NUM]); - -/* - * Kernel breakpoints grow downwards, starting from HBP_NUM - * 'hbp_kernel_pos' denotes lowest numbered breakpoint register occupied for - * kernel-space request. We will initialise it here and not in an __init - * routine because load_debug_registers(), which uses this variable can be - * called very early during CPU initialisation. - */ -unsigned int hbp_kernel_pos = HBP_NUM; -/* - * An array containing refcount of threads using a given bkpt register - * Accesses are synchronised by acquiring hw_breakpoint_lock - */ -unsigned int hbp_user_refcount[HBP_NUM]; +static atomic_t bp_slot; -/* - * Load the debug registers during startup of a CPU. - */ -void load_debug_registers(void) +int reserve_bp_slot(struct perf_event *bp) { - unsigned long flags; - struct task_struct *tsk = current; - - spin_lock_bh(&hw_breakpoint_lock); - - /* Prevent IPIs for new kernel breakpoint updates */ - local_irq_save(flags); - arch_update_kernel_hw_breakpoint(NULL); - local_irq_restore(flags); - - if (test_tsk_thread_flag(tsk, TIF_DEBUG)) - arch_install_thread_hw_breakpoint(tsk); - - spin_unlock_bh(&hw_breakpoint_lock); -} + if (atomic_inc_return(&bp_slot) == HBP_NUM) { + atomic_dec(&bp_slot); -/* - * Erase all the hardware breakpoint info associated with a thread. - * - * If tsk != current then tsk must not be usable (for example, a - * child being cleaned up from a failed fork). - */ -void flush_thread_hw_breakpoint(struct task_struct *tsk) -{ - int i; - struct thread_struct *thread = &(tsk->thread); - - spin_lock_bh(&hw_breakpoint_lock); - - /* The thread no longer has any breakpoints associated with it */ - clear_tsk_thread_flag(tsk, TIF_DEBUG); - for (i = 0; i < HBP_NUM; i++) { - if (thread->hbp[i]) { - hbp_user_refcount[i]--; - kfree(thread->hbp[i]); - thread->hbp[i] = NULL; - } + return -ENOSPC; } - arch_flush_thread_hw_breakpoint(tsk); - - /* Actually uninstall the breakpoints if necessary */ - if (tsk == current) - arch_uninstall_thread_hw_breakpoint(); - spin_unlock_bh(&hw_breakpoint_lock); + return 0; } -/* - * Copy the hardware breakpoint info from a thread to its cloned child. - */ -int copy_thread_hw_breakpoint(struct task_struct *tsk, - struct task_struct *child, unsigned long clone_flags) +void release_bp_slot(struct perf_event *bp) { - /* - * We will assume that breakpoint settings are not inherited - * and the child starts out with no debug registers set. - * But what about CLONE_PTRACE? - */ - clear_tsk_thread_flag(child, TIF_DEBUG); - - /* We will call flush routine since the debugregs are not inherited */ - arch_flush_thread_hw_breakpoint(child); - - return 0; + atomic_dec(&bp_slot); } -static int __register_user_hw_breakpoint(int pos, struct task_struct *tsk, - struct hw_breakpoint *bp) +int __register_perf_hw_breakpoint(struct perf_event *bp) { - struct thread_struct *thread = &(tsk->thread); - int rc; + int ret; - /* Do not overcommit. Fail if kernel has used the hbp registers */ - if (pos >= hbp_kernel_pos) - return -ENOSPC; + ret = reserve_bp_slot(bp); + if (ret) + return ret; - rc = arch_validate_hwbkpt_settings(bp, tsk); - if (rc) - return rc; + if (!bp->attr.disabled) + ret = arch_validate_hwbkpt_settings(bp, bp->ctx->task); - thread->hbp[pos] = bp; - hbp_user_refcount[pos]++; + return ret; +} - arch_update_user_hw_breakpoint(pos, tsk); - /* - * Does it need to be installed right now? - * Otherwise it will get installed the next time tsk runs - */ - if (tsk == current) - arch_install_thread_hw_breakpoint(tsk); +int register_perf_hw_breakpoint(struct perf_event *bp) +{ + bp->callback = perf_bp_event; - return rc; + return __register_perf_hw_breakpoint(bp); } /* - * Modify the address of a hbp register already in use by the task - * Do not invoke this in-lieu of a __unregister_user_hw_breakpoint() + * Register a breakpoint bound to a task and a given cpu. + * If cpu is -1, the breakpoint is active for the task in every cpu + * If the task is -1, the breakpoint is active for every tasks in the given + * cpu. */ -static int __modify_user_hw_breakpoint(int pos, struct task_struct *tsk, - struct hw_breakpoint *bp) +static struct perf_event * +register_user_hw_breakpoint_cpu(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + pid_t pid, + int cpu, + bool active) { - struct thread_struct *thread = &(tsk->thread); - - if ((pos >= hbp_kernel_pos) || (arch_validate_hwbkpt_settings(bp, tsk))) - return -EINVAL; - - if (thread->hbp[pos] == NULL) - return -EINVAL; - - thread->hbp[pos] = bp; + struct perf_event_attr *attr; + struct perf_event *bp; + + attr = kzalloc(sizeof(*attr), GFP_KERNEL); + if (!attr) + return ERR_PTR(-ENOMEM); + + attr->type = PERF_TYPE_BREAKPOINT; + attr->size = sizeof(*attr); + attr->bp_addr = addr; + attr->bp_len = len; + attr->bp_type = type; /* - * 'pos' must be that of a hbp register already used by 'tsk' - * Otherwise arch_modify_user_hw_breakpoint() will fail + * Such breakpoints are used by debuggers to trigger signals when + * we hit the excepted memory op. We can't miss such events, they + * must be pinned. */ - arch_update_user_hw_breakpoint(pos, tsk); + attr->pinned = 1; - if (tsk == current) - arch_install_thread_hw_breakpoint(tsk); + if (!active) + attr->disabled = 1; - return 0; -} - -static void __unregister_user_hw_breakpoint(int pos, struct task_struct *tsk) -{ - hbp_user_refcount[pos]--; - tsk->thread.hbp[pos] = NULL; + bp = perf_event_create_kernel_counter(attr, cpu, pid, triggered); + kfree(attr); - arch_update_user_hw_breakpoint(pos, tsk); - - if (tsk == current) - arch_install_thread_hw_breakpoint(tsk); + return bp; } /** * register_user_hw_breakpoint - register a hardware breakpoint for user space + * @addr: is the memory address that triggers the breakpoint + * @len: the length of the access to the memory (1 byte, 2 bytes etc...) + * @type: the type of the access to the memory (read/write/exec) + * @triggered: callback to trigger when we hit the breakpoint * @tsk: pointer to 'task_struct' of the process to which the address belongs - * @bp: the breakpoint structure to register - * - * @bp.info->name or @bp.info->address, @bp.info->len, @bp.info->type and - * @bp->triggered must be set properly before invocation + * @active: should we activate it while registering it * */ -int register_user_hw_breakpoint(struct task_struct *tsk, - struct hw_breakpoint *bp) +struct perf_event * +register_user_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active) { - struct thread_struct *thread = &(tsk->thread); - int i, rc = -ENOSPC; - - spin_lock_bh(&hw_breakpoint_lock); - - for (i = 0; i < hbp_kernel_pos; i++) { - if (!thread->hbp[i]) { - rc = __register_user_hw_breakpoint(i, tsk, bp); - break; - } - } - if (!rc) - set_tsk_thread_flag(tsk, TIF_DEBUG); - - spin_unlock_bh(&hw_breakpoint_lock); - return rc; + return register_user_hw_breakpoint_cpu(addr, len, type, triggered, + tsk->pid, -1, active); } EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); /** * modify_user_hw_breakpoint - modify a user-space hardware breakpoint + * @bp: the breakpoint structure to modify + * @addr: is the memory address that triggers the breakpoint + * @len: the length of the access to the memory (1 byte, 2 bytes etc...) + * @type: the type of the access to the memory (read/write/exec) + * @triggered: callback to trigger when we hit the breakpoint * @tsk: pointer to 'task_struct' of the process to which the address belongs - * @bp: the breakpoint structure to unregister - * + * @active: should we activate it while registering it */ -int modify_user_hw_breakpoint(struct task_struct *tsk, struct hw_breakpoint *bp) +struct perf_event * +modify_user_hw_breakpoint(struct perf_event *bp, + unsigned long addr, + int len, + int type, + perf_callback_t triggered, + struct task_struct *tsk, + bool active) { - struct thread_struct *thread = &(tsk->thread); - int i, ret = -ENOENT; + /* + * FIXME: do it without unregistering + * - We don't want to lose our slot + * - If the new bp is incorrect, don't lose the older one + */ + unregister_hw_breakpoint(bp); - spin_lock_bh(&hw_breakpoint_lock); - for (i = 0; i < hbp_kernel_pos; i++) { - if (bp == thread->hbp[i]) { - ret = __modify_user_hw_breakpoint(i, tsk, bp); - break; - } - } - spin_unlock_bh(&hw_breakpoint_lock); - return ret; + return register_user_hw_breakpoint(addr, len, type, triggered, + tsk, active); } EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); /** - * unregister_user_hw_breakpoint - unregister a user-space hardware breakpoint - * @tsk: pointer to 'task_struct' of the process to which the address belongs + * unregister_hw_breakpoint - unregister a user-space hardware breakpoint * @bp: the breakpoint structure to unregister - * */ -void unregister_user_hw_breakpoint(struct task_struct *tsk, - struct hw_breakpoint *bp) +void unregister_hw_breakpoint(struct perf_event *bp) { - struct thread_struct *thread = &(tsk->thread); - int i, pos = -1, hbp_counter = 0; - - spin_lock_bh(&hw_breakpoint_lock); - for (i = 0; i < hbp_kernel_pos; i++) { - if (thread->hbp[i]) - hbp_counter++; - if (bp == thread->hbp[i]) - pos = i; - } - if (pos >= 0) { - __unregister_user_hw_breakpoint(pos, tsk); - hbp_counter--; - } - if (!hbp_counter) - clear_tsk_thread_flag(tsk, TIF_DEBUG); - - spin_unlock_bh(&hw_breakpoint_lock); + if (!bp) + return; + perf_event_release_kernel(bp); +} +EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); + +static struct perf_event * +register_kernel_hw_breakpoint_cpu(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + int cpu, + bool active) +{ + return register_user_hw_breakpoint_cpu(addr, len, type, triggered, + -1, cpu, active); } -EXPORT_SYMBOL_GPL(unregister_user_hw_breakpoint); /** - * register_kernel_hw_breakpoint - register a hardware breakpoint for kernel space - * @bp: the breakpoint structure to register - * - * @bp.info->name or @bp.info->address, @bp.info->len, @bp.info->type and - * @bp->triggered must be set properly before invocation + * register_wide_hw_breakpoint - register a wide breakpoint in the kernel + * @addr: is the memory address that triggers the breakpoint + * @len: the length of the access to the memory (1 byte, 2 bytes etc...) + * @type: the type of the access to the memory (read/write/exec) + * @triggered: callback to trigger when we hit the breakpoint + * @active: should we activate it while registering it * + * @return a set of per_cpu pointers to perf events */ -int register_kernel_hw_breakpoint(struct hw_breakpoint *bp) +struct perf_event ** +register_wide_hw_breakpoint(unsigned long addr, + int len, + int type, + perf_callback_t triggered, + bool active) { - int rc; + struct perf_event **cpu_events, **pevent, *bp; + long err; + int cpu; + + cpu_events = alloc_percpu(typeof(*cpu_events)); + if (!cpu_events) + return ERR_PTR(-ENOMEM); - rc = arch_validate_hwbkpt_settings(bp, NULL); - if (rc) - return rc; + for_each_possible_cpu(cpu) { + pevent = per_cpu_ptr(cpu_events, cpu); + bp = register_kernel_hw_breakpoint_cpu(addr, len, type, + triggered, cpu, active); - spin_lock_bh(&hw_breakpoint_lock); + *pevent = bp; - rc = -ENOSPC; - /* Check if we are over-committing */ - if ((hbp_kernel_pos > 0) && (!hbp_user_refcount[hbp_kernel_pos-1])) { - hbp_kernel_pos--; - hbp_kernel[hbp_kernel_pos] = bp; - on_each_cpu(arch_update_kernel_hw_breakpoint, NULL, 1); - rc = 0; + if (IS_ERR(bp) || !bp) { + err = PTR_ERR(bp); + goto fail; + } } - spin_unlock_bh(&hw_breakpoint_lock); - return rc; + return cpu_events; + +fail: + for_each_possible_cpu(cpu) { + pevent = per_cpu_ptr(cpu_events, cpu); + if (IS_ERR(*pevent) || !*pevent) + break; + unregister_hw_breakpoint(*pevent); + } + free_percpu(cpu_events); + /* return the error if any */ + return ERR_PTR(err); } -EXPORT_SYMBOL_GPL(register_kernel_hw_breakpoint); /** - * unregister_kernel_hw_breakpoint - unregister a HW breakpoint for kernel space - * @bp: the breakpoint structure to unregister - * - * Uninstalls and unregisters @bp. + * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel + * @cpu_events: the per cpu set of events to unregister */ -void unregister_kernel_hw_breakpoint(struct hw_breakpoint *bp) +void unregister_wide_hw_breakpoint(struct perf_event **cpu_events) { - int i, j; - - spin_lock_bh(&hw_breakpoint_lock); - - /* Find the 'bp' in our list of breakpoints for kernel */ - for (i = hbp_kernel_pos; i < HBP_NUM; i++) - if (bp == hbp_kernel[i]) - break; + int cpu; + struct perf_event **pevent; - /* Check if we did not find a match for 'bp'. If so return early */ - if (i == HBP_NUM) { - spin_unlock_bh(&hw_breakpoint_lock); - return; + for_each_possible_cpu(cpu) { + pevent = per_cpu_ptr(cpu_events, cpu); + unregister_hw_breakpoint(*pevent); } - - /* - * We'll shift the breakpoints one-level above to compact if - * unregistration creates a hole - */ - for (j = i; j > hbp_kernel_pos; j--) - hbp_kernel[j] = hbp_kernel[j-1]; - - hbp_kernel[hbp_kernel_pos] = NULL; - on_each_cpu(arch_update_kernel_hw_breakpoint, NULL, 1); - hbp_kernel_pos++; - - spin_unlock_bh(&hw_breakpoint_lock); + free_percpu(cpu_events); } -EXPORT_SYMBOL_GPL(unregister_kernel_hw_breakpoint); + static struct notifier_block hw_breakpoint_exceptions_nb = { .notifier_call = hw_breakpoint_exceptions_notify, @@ -374,5 +283,12 @@ static int __init init_hw_breakpoint(void) { return register_die_notifier(&hw_breakpoint_exceptions_nb); } - core_initcall(init_hw_breakpoint); + + +struct pmu perf_ops_bp = { + .enable = arch_install_hw_breakpoint, + .disable = arch_uninstall_hw_breakpoint, + .read = hw_breakpoint_pmu_read, + .unthrottle = hw_breakpoint_pmu_unthrottle +}; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 5087125e2a00..98dc56b2ebe4 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -29,6 +29,7 @@ #include #include #include +#include #include @@ -4229,6 +4230,51 @@ static void perf_event_free_filter(struct perf_event *event) #endif /* CONFIG_EVENT_PROFILE */ +#ifdef CONFIG_HAVE_HW_BREAKPOINT +static void bp_perf_event_destroy(struct perf_event *event) +{ + release_bp_slot(event); +} + +static const struct pmu *bp_perf_event_init(struct perf_event *bp) +{ + int err; + /* + * The breakpoint is already filled if we haven't created the counter + * through perf syscall + * FIXME: manage to get trigerred to NULL if it comes from syscalls + */ + if (!bp->callback) + err = register_perf_hw_breakpoint(bp); + else + err = __register_perf_hw_breakpoint(bp); + if (err) + return ERR_PTR(err); + + bp->destroy = bp_perf_event_destroy; + + return &perf_ops_bp; +} + +void perf_bp_event(struct perf_event *bp, void *regs) +{ + /* TODO */ +} +#else +static void bp_perf_event_destroy(struct perf_event *event) +{ +} + +static const struct pmu *bp_perf_event_init(struct perf_event *bp) +{ + return NULL; +} + +void perf_bp_event(struct perf_event *bp, void *regs) +{ +} +#endif + atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; static void sw_perf_event_destroy(struct perf_event *event) @@ -4375,6 +4421,11 @@ perf_event_alloc(struct perf_event_attr *attr, pmu = tp_perf_event_init(event); break; + case PERF_TYPE_BREAKPOINT: + pmu = bp_perf_event_init(event); + break; + + default: break; } @@ -4686,7 +4737,7 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, ctx = find_get_context(pid, cpu); if (IS_ERR(ctx)) - return NULL ; + return NULL; event = perf_event_alloc(attr, cpu, ctx, NULL, NULL, callback, GFP_KERNEL); diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index 91c3d0e9a5a1..d72f06ff263f 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -11,14 +11,11 @@ #include #include #include +#include #include #include -#ifdef CONFIG_KSYM_TRACER -#include -#endif - enum trace_type { __TRACE_FIRST_TYPE = 0, diff --git a/kernel/trace/trace_entries.h b/kernel/trace/trace_entries.h index e19747d4f860..c16a08f399df 100644 --- a/kernel/trace/trace_entries.h +++ b/kernel/trace/trace_entries.h @@ -372,11 +372,11 @@ FTRACE_ENTRY(ksym_trace, ksym_trace_entry, F_STRUCT( __field( unsigned long, ip ) __field( unsigned char, type ) - __array( char , ksym_name, KSYM_NAME_LEN ) __array( char , cmd, TASK_COMM_LEN ) + __field( unsigned long, addr ) ), - F_printk("ip: %pF type: %d ksym_name: %s cmd: %s", + F_printk("ip: %pF type: %d ksym_name: %pS cmd: %s", (void *)__entry->ip, (unsigned int)__entry->type, - __entry->ksym_name, __entry->cmd) + (void *)__entry->addr, __entry->cmd) ); diff --git a/kernel/trace/trace_ksym.c b/kernel/trace/trace_ksym.c index 6d5609c67378..fea83eeeef09 100644 --- a/kernel/trace/trace_ksym.c +++ b/kernel/trace/trace_ksym.c @@ -29,7 +29,11 @@ #include "trace_stat.h" #include "trace.h" -/* For now, let us restrict the no. of symbols traced simultaneously to number +#include +#include + +/* + * For now, let us restrict the no. of symbols traced simultaneously to number * of available hardware breakpoint registers. */ #define KSYM_TRACER_MAX HBP_NUM @@ -37,8 +41,10 @@ #define KSYM_TRACER_OP_LEN 3 /* rw- */ struct trace_ksym { - struct hw_breakpoint *ksym_hbp; + struct perf_event **ksym_hbp; unsigned long ksym_addr; + int type; + int len; #ifdef CONFIG_PROFILE_KSYM_TRACER unsigned long counter; #endif @@ -75,10 +81,11 @@ void ksym_collect_stats(unsigned long hbp_hit_addr) } #endif /* CONFIG_PROFILE_KSYM_TRACER */ -void ksym_hbp_handler(struct hw_breakpoint *hbp, struct pt_regs *regs) +void ksym_hbp_handler(struct perf_event *hbp, void *data) { struct ring_buffer_event *event; struct ksym_trace_entry *entry; + struct pt_regs *regs = data; struct ring_buffer *buffer; int pc; @@ -96,12 +103,12 @@ void ksym_hbp_handler(struct hw_breakpoint *hbp, struct pt_regs *regs) entry = ring_buffer_event_data(event); entry->ip = instruction_pointer(regs); - entry->type = hbp->info.type; - strlcpy(entry->ksym_name, hbp->info.name, KSYM_SYMBOL_LEN); + entry->type = hw_breakpoint_type(hbp); + entry->addr = hw_breakpoint_addr(hbp); strlcpy(entry->cmd, current->comm, TASK_COMM_LEN); #ifdef CONFIG_PROFILE_KSYM_TRACER - ksym_collect_stats(hbp->info.address); + ksym_collect_stats(hw_breakpoint_addr(hbp)); #endif /* CONFIG_PROFILE_KSYM_TRACER */ trace_buffer_unlock_commit(buffer, event, 0, pc); @@ -120,31 +127,21 @@ static int ksym_trace_get_access_type(char *str) int access = 0; if (str[0] == 'r') - access += 4; - else if (str[0] != '-') - return -EINVAL; + access |= HW_BREAKPOINT_R; if (str[1] == 'w') - access += 2; - else if (str[1] != '-') - return -EINVAL; + access |= HW_BREAKPOINT_W; - if (str[2] != '-') - return -EINVAL; + if (str[2] == 'x') + access |= HW_BREAKPOINT_X; switch (access) { - case 6: - access = HW_BREAKPOINT_RW; - break; - case 4: - access = -EINVAL; - break; - case 2: - access = HW_BREAKPOINT_WRITE; - break; + case HW_BREAKPOINT_W: + case HW_BREAKPOINT_W | HW_BREAKPOINT_R: + return access; + default: + return -EINVAL; } - - return access; } /* @@ -194,36 +191,33 @@ int process_new_ksym_entry(char *ksymname, int op, unsigned long addr) if (!entry) return -ENOMEM; - entry->ksym_hbp = kzalloc(sizeof(struct hw_breakpoint), GFP_KERNEL); - if (!entry->ksym_hbp) - goto err; - - entry->ksym_hbp->info.name = kstrdup(ksymname, GFP_KERNEL); - if (!entry->ksym_hbp->info.name) - goto err; - - entry->ksym_hbp->info.type = op; - entry->ksym_addr = entry->ksym_hbp->info.address = addr; -#ifdef CONFIG_X86 - entry->ksym_hbp->info.len = HW_BREAKPOINT_LEN_4; -#endif - entry->ksym_hbp->triggered = (void *)ksym_hbp_handler; + entry->type = op; + entry->ksym_addr = addr; + entry->len = HW_BREAKPOINT_LEN_4; + + ret = -EAGAIN; + entry->ksym_hbp = register_wide_hw_breakpoint(entry->ksym_addr, + entry->len, entry->type, + ksym_hbp_handler, true); + if (IS_ERR(entry->ksym_hbp)) { + entry->ksym_hbp = NULL; + ret = PTR_ERR(entry->ksym_hbp); + } - ret = register_kernel_hw_breakpoint(entry->ksym_hbp); - if (ret < 0) { + if (!entry->ksym_hbp) { printk(KERN_INFO "ksym_tracer request failed. Try again" " later!!\n"); - ret = -EAGAIN; goto err; } + hlist_add_head_rcu(&(entry->ksym_hlist), &ksym_filter_head); ksym_filter_entry_count++; + return 0; + err: - if (entry->ksym_hbp) - kfree(entry->ksym_hbp->info.name); - kfree(entry->ksym_hbp); kfree(entry); + return ret; } @@ -244,10 +238,10 @@ static ssize_t ksym_trace_filter_read(struct file *filp, char __user *ubuf, mutex_lock(&ksym_tracer_mutex); hlist_for_each_entry(entry, node, &ksym_filter_head, ksym_hlist) { - ret = trace_seq_printf(s, "%s:", entry->ksym_hbp->info.name); - if (entry->ksym_hbp->info.type == HW_BREAKPOINT_WRITE) + ret = trace_seq_printf(s, "%pS:", (void *)entry->ksym_addr); + if (entry->type == HW_BREAKPOINT_W) ret = trace_seq_puts(s, "-w-\n"); - else if (entry->ksym_hbp->info.type == HW_BREAKPOINT_RW) + else if (entry->type == (HW_BREAKPOINT_W | HW_BREAKPOINT_R)) ret = trace_seq_puts(s, "rw-\n"); WARN_ON_ONCE(!ret); } @@ -269,12 +263,10 @@ static void __ksym_trace_reset(void) mutex_lock(&ksym_tracer_mutex); hlist_for_each_entry_safe(entry, node, node1, &ksym_filter_head, ksym_hlist) { - unregister_kernel_hw_breakpoint(entry->ksym_hbp); + unregister_wide_hw_breakpoint(entry->ksym_hbp); ksym_filter_entry_count--; hlist_del_rcu(&(entry->ksym_hlist)); synchronize_rcu(); - kfree(entry->ksym_hbp->info.name); - kfree(entry->ksym_hbp); kfree(entry); } mutex_unlock(&ksym_tracer_mutex); @@ -327,7 +319,7 @@ static ssize_t ksym_trace_filter_write(struct file *file, hlist_for_each_entry(entry, node, &ksym_filter_head, ksym_hlist) { if (entry->ksym_addr == ksym_addr) { /* Check for malformed request: (6) */ - if (entry->ksym_hbp->info.type != op) + if (entry->type != op) changed = 1; else goto out; @@ -335,18 +327,21 @@ static ssize_t ksym_trace_filter_write(struct file *file, } } if (changed) { - unregister_kernel_hw_breakpoint(entry->ksym_hbp); - entry->ksym_hbp->info.type = op; + unregister_wide_hw_breakpoint(entry->ksym_hbp); + entry->type = op; if (op > 0) { - ret = register_kernel_hw_breakpoint(entry->ksym_hbp); - if (ret == 0) + entry->ksym_hbp = + register_wide_hw_breakpoint(entry->ksym_addr, + entry->len, entry->type, + ksym_hbp_handler, true); + if (IS_ERR(entry->ksym_hbp)) + entry->ksym_hbp = NULL; + if (!entry->ksym_hbp) goto out; } ksym_filter_entry_count--; hlist_del_rcu(&(entry->ksym_hlist)); synchronize_rcu(); - kfree(entry->ksym_hbp->info.name); - kfree(entry->ksym_hbp); kfree(entry); ret = 0; goto out; @@ -413,16 +408,16 @@ static enum print_line_t ksym_trace_output(struct trace_iterator *iter) trace_assign_type(field, entry); - ret = trace_seq_printf(s, "%11s-%-5d [%03d] %-30s ", field->cmd, - entry->pid, iter->cpu, field->ksym_name); + ret = trace_seq_printf(s, "%11s-%-5d [%03d] %pS", field->cmd, + entry->pid, iter->cpu, (char *)field->addr); if (!ret) return TRACE_TYPE_PARTIAL_LINE; switch (field->type) { - case HW_BREAKPOINT_WRITE: + case HW_BREAKPOINT_W: ret = trace_seq_printf(s, " W "); break; - case HW_BREAKPOINT_RW: + case HW_BREAKPOINT_R | HW_BREAKPOINT_W: ret = trace_seq_printf(s, " RW "); break; default: @@ -490,14 +485,13 @@ static int ksym_tracer_stat_show(struct seq_file *m, void *v) entry = hlist_entry(stat, struct trace_ksym, ksym_hlist); - if (entry->ksym_hbp) - access_type = entry->ksym_hbp->info.type; + access_type = entry->type; switch (access_type) { - case HW_BREAKPOINT_WRITE: + case HW_BREAKPOINT_W: seq_puts(m, " W "); break; - case HW_BREAKPOINT_RW: + case HW_BREAKPOINT_R | HW_BREAKPOINT_W: seq_puts(m, " RW "); break; default: diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c index 7179c12e4f0f..27c5072c2e6b 100644 --- a/kernel/trace/trace_selftest.c +++ b/kernel/trace/trace_selftest.c @@ -828,7 +828,8 @@ trace_selftest_startup_ksym(struct tracer *trace, struct trace_array *tr) ksym_selftest_dummy = 0; /* Register the read-write tracing request */ - ret = process_new_ksym_entry(KSYM_SELFTEST_ENTRY, HW_BREAKPOINT_RW, + ret = process_new_ksym_entry(KSYM_SELFTEST_ENTRY, + HW_BREAKPOINT_R | HW_BREAKPOINT_W, (unsigned long)(&ksym_selftest_dummy)); if (ret < 0) { -- cgit v1.2.3 From 559fdc3c1b624edb1933a875022fe7e27934d11c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 16 Nov 2009 12:45:14 +0100 Subject: perf_event: Optimize perf_output_lock() The purpose of perf_output_{un,}lock() is to: 1) avoid publishing incomplete data [ possible when publishing a head that is ahead of an entry that is still being written ] 2) guarantee fwd progress [ a simple refcount on pending writers doesn't need to drop to 0, making it so would end up implementing something like forced quiecent states of RCU ] To satisfy the above without undue complexity it serializes between CPUs, this means that a pending writer can only be the same cpu in a nested context, and since (under normal operation) a cpu always makes progress we're good -- if the head is only published when the bottom most writer completes. Now we don't need to disable IRQs in order to serialize between CPUs, disabling preemption ought to be sufficient, esp since we already deal with nesting due to NMIs. This avoids potentially expensive (and needless) local IRQ disable/enable ops. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Cc: Frederic Weisbecker LKML-Reference: <1258373161.26714.254.camel@laptop> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 1 - kernel/perf_event.c | 21 +++++++++++---------- 2 files changed, 11 insertions(+), 11 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index df4e73e33774..7f87563c8485 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -714,7 +714,6 @@ struct perf_output_handle { int nmi; int sample; int locked; - unsigned long flags; }; #ifdef CONFIG_PERF_EVENTS diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 6f4ed3b4cd73..3256e36ad251 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -2674,20 +2674,21 @@ static void perf_output_wakeup(struct perf_output_handle *handle) static void perf_output_lock(struct perf_output_handle *handle) { struct perf_mmap_data *data = handle->data; - int cpu; + int cur, cpu = get_cpu(); handle->locked = 0; - local_irq_save(handle->flags); - cpu = smp_processor_id(); - - if (in_nmi() && atomic_read(&data->lock) == cpu) - return; + for (;;) { + cur = atomic_cmpxchg(&data->lock, -1, cpu); + if (cur == -1) { + handle->locked = 1; + break; + } + if (cur == cpu) + break; - while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) cpu_relax(); - - handle->locked = 1; + } } static void perf_output_unlock(struct perf_output_handle *handle) @@ -2733,7 +2734,7 @@ again: if (atomic_xchg(&data->wakeup, 0)) perf_output_wakeup(handle); out: - local_irq_restore(handle->flags); + put_cpu(); } void perf_output_copy(struct perf_output_handle *handle, -- cgit v1.2.3 From 453f19eea7dbad837425e9b07d84568d14898794 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:43 +0100 Subject: perf: Allow for custom overflow handlers in-kernel perf users might wish to have custom actions on the sample interrupt. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.222339539@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 6 ++++++ kernel/perf_event.c | 8 +++++++- 2 files changed, 13 insertions(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index b5cdac0de370..a430ac3074af 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -567,6 +567,8 @@ struct perf_pending_entry { typedef void (*perf_callback_t)(struct perf_event *, void *); +struct perf_sample_data; + /** * struct perf_event - performance event kernel representation: */ @@ -658,6 +660,10 @@ struct perf_event { struct pid_namespace *ns; u64 id; + void (*overflow_handler)(struct perf_event *event, + int nmi, struct perf_sample_data *data, + struct pt_regs *regs); + #ifdef CONFIG_EVENT_PROFILE struct event_filter *filter; #endif diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 3852e2656bb0..1dfb6cc4fdea 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3710,7 +3710,11 @@ static int __perf_event_overflow(struct perf_event *event, int nmi, perf_event_disable(event); } - perf_event_output(event, nmi, data, regs); + if (event->overflow_handler) + event->overflow_handler(event, nmi, data, regs); + else + perf_event_output(event, nmi, data, regs); + return ret; } @@ -4836,6 +4840,8 @@ inherit_event(struct perf_event *parent_event, if (parent_event->attr.freq) child_event->hw.sample_period = parent_event->hw.sample_period; + child_event->overflow_handler = parent_event->overflow_handler; + /* * Link it up in the child's context: */ -- cgit v1.2.3 From 0cff784ae41cc125368ae77f1c01328ae2fdc6b3 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:44 +0100 Subject: perf: Optimize some swcounter attr.sample_period==1 paths Avoid the rather expensive perf_swevent_set_period() if we know we have to sample every single event anyway. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.299508332@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 1dfb6cc4fdea..8e55b440e28a 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3759,16 +3759,16 @@ again: return nr; } -static void perf_swevent_overflow(struct perf_event *event, +static void perf_swevent_overflow(struct perf_event *event, u64 overflow, int nmi, struct perf_sample_data *data, struct pt_regs *regs) { struct hw_perf_event *hwc = &event->hw; int throttle = 0; - u64 overflow; data->period = event->hw.last_period; - overflow = perf_swevent_set_period(event); + if (!overflow) + overflow = perf_swevent_set_period(event); if (hwc->interrupts == MAX_INTERRUPTS) return; @@ -3801,14 +3801,19 @@ static void perf_swevent_add(struct perf_event *event, u64 nr, atomic64_add(nr, &event->count); + if (!regs) + return; + if (!hwc->sample_period) return; - if (!regs) + if (nr == 1 && hwc->sample_period == 1 && !event->attr.freq) + return perf_swevent_overflow(event, 1, nmi, data, regs); + + if (atomic64_add_negative(nr, &hwc->period_left)) return; - if (!atomic64_add_negative(nr, &hwc->period_left)) - perf_swevent_overflow(event, nmi, data, regs); + perf_swevent_overflow(event, 0, nmi, data, regs); } static int perf_swevent_is_counting(struct perf_event *event) -- cgit v1.2.3 From 81520183878a8813c71c9372de28bb70913ba549 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:45 +0100 Subject: perf: Optimize perf_swevent_ctx_event() Remove a rcu_read_{,un}lock() pair and a few conditionals. We can remove the rcu_read_lock() by increasing the scope of one in the calling function. We can do away with the system_state check if the machine still boots after this patch (seems to be the case). We can do away with the list_empty() check because the bare list_for_each_entry_rcu() reduces to that now that we've removed everything else. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.378188589@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 8e55b440e28a..cda17acfcaf8 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3886,15 +3886,10 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_swevent_match(event, type, event_id, data, regs)) perf_swevent_add(event, nr, nmi, data, regs); } - rcu_read_unlock(); } static int *perf_swevent_recursion_context(struct perf_cpu_context *cpuctx) @@ -3926,9 +3921,9 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id, (*recursion)++; barrier(); + rcu_read_lock(); perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, nr, nmi, data, regs); - rcu_read_lock(); /* * doesn't really matter which of the child contexts the * events ends up in. -- cgit v1.2.3 From d6ff86cfb50a72df820e7e839836d55d245306fb Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:46 +0100 Subject: perf: Optimize perf_event_task_ctx() Remove a rcu_read_{,un}lock() pair and a few conditionals. We can remove the rcu_read_lock() by increasing the scope of one in the calling function. We can do away with the system_state check if the machine still boots after this patch (seems to be the case). We can do away with the list_empty() check because the bare list_for_each_entry_rcu() reduces to that now that we've removed everything else. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.452227115@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index cda17acfcaf8..2afb305944af 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3267,15 +3267,10 @@ static void perf_event_task_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_task_match(event)) perf_event_task_output(event, task_event); } - rcu_read_unlock(); } static void perf_event_task_event(struct perf_task_event *task_event) @@ -3283,11 +3278,11 @@ static void perf_event_task_event(struct perf_task_event *task_event) struct perf_cpu_context *cpuctx; struct perf_event_context *ctx = task_event->task_ctx; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_task_ctx(&cpuctx->ctx, task_event); put_cpu_var(perf_cpu_context); - rcu_read_lock(); if (!ctx) ctx = rcu_dereference(task_event->task->perf_event_ctxp); if (ctx) -- cgit v1.2.3 From f6595f3a9680c86b6332f881a7ae2cbbcfdc8619 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:47 +0100 Subject: perf: Optimize perf_event_comm_ctx() Remove a rcu_read_{,un}lock() pair and a few conditionals. We can remove the rcu_read_lock() by increasing the scope of one in the calling function. We can do away with the system_state check if the machine still boots after this patch (seems to be the case). We can do away with the list_empty() check because the bare list_for_each_entry_rcu() reduces to that now that we've removed everything else. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.527608793@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 2afb305944af..4deefaace90e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3374,15 +3374,10 @@ static void perf_event_comm_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_comm_match(event)) perf_event_comm_output(event, comm_event); } - rcu_read_unlock(); } static void perf_event_comm_event(struct perf_comm_event *comm_event) @@ -3401,11 +3396,11 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) comm_event->event_id.header.size = sizeof(comm_event->event_id) + size; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_comm_ctx(&cpuctx->ctx, comm_event); put_cpu_var(perf_cpu_context); - rcu_read_lock(); /* * doesn't really matter which of the child contexts the * events ends up in. -- cgit v1.2.3 From f6d9dd237da400effb265f3554c64413f8a3e7b4 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:48 +0100 Subject: perf: Optimize perf_event_mmap_ctx() Remove a rcu_read_{,un}lock() pair and a few conditionals. We can remove the rcu_read_lock() by increasing the scope of one in the calling function. We can do away with the system_state check if the machine still boots after this patch (seems to be the case). We can do away with the list_empty() check because the bare list_for_each_entry_rcu() reduces to that now that we've removed everything else. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.606459548@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 4deefaace90e..68fbf4ff6888 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3493,15 +3493,10 @@ static void perf_event_mmap_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_mmap_match(event, mmap_event)) perf_event_mmap_output(event, mmap_event); } - rcu_read_unlock(); } static void perf_event_mmap_event(struct perf_mmap_event *mmap_event) @@ -3557,11 +3552,11 @@ got_name: mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_mmap_ctx(&cpuctx->ctx, mmap_event); put_cpu_var(perf_cpu_context); - rcu_read_lock(); /* * doesn't really matter which of the child contexts the * events ends up in. -- cgit v1.2.3 From abf4868b8548cae18d4fe8bbfb4e207443be01be Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:49 +0100 Subject: perf: Fix PERF_FORMAT_GROUP scale info As Corey reported, the total_enabled and total_running times could occasionally be 0, even though there were events counted. It turns out this is because we record the times before reading the counter while the latter updates the times. This patch corrects that. While looking at this code I found that there is a lot of locking iffyness around, the following patches correct most of that. Reported-by: Corey Ashford Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.685559857@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 52 +++++++++++++++++++++------------------------------- 1 file changed, 21 insertions(+), 31 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 68fbf4ff6888..9a18ff28ea5b 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1784,30 +1784,15 @@ u64 perf_event_read_value(struct perf_event *event) } EXPORT_SYMBOL_GPL(perf_event_read_value); -static int perf_event_read_entry(struct perf_event *event, - u64 read_format, char __user *buf) -{ - int n = 0, count = 0; - u64 values[2]; - - values[n++] = perf_event_read_value(event); - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_event_id(event); - - count = n * sizeof(u64); - - if (copy_to_user(buf, values, count)) - return -EFAULT; - - return count; -} - static int perf_event_read_group(struct perf_event *event, u64 read_format, char __user *buf) { struct perf_event *leader = event->group_leader, *sub; - int n = 0, size = 0, err = -EFAULT; - u64 values[3]; + int n = 0, size = 0, ret = 0; + u64 values[5]; + u64 count; + + count = perf_event_read_value(leader); values[n++] = 1 + leader->nr_siblings; if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { @@ -1818,28 +1803,33 @@ static int perf_event_read_group(struct perf_event *event, values[n++] = leader->total_time_running + atomic64_read(&leader->child_total_time_running); } + values[n++] = count; + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(leader); size = n * sizeof(u64); if (copy_to_user(buf, values, size)) return -EFAULT; - err = perf_event_read_entry(leader, read_format, buf + size); - if (err < 0) - return err; - - size += err; + ret += size; list_for_each_entry(sub, &leader->sibling_list, group_entry) { - err = perf_event_read_entry(sub, read_format, - buf + size); - if (err < 0) - return err; + n = 0; - size += err; + values[n++] = perf_event_read_value(sub); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(sub); + + size = n * sizeof(u64); + + if (copy_to_user(buf + size, values, size)) + return -EFAULT; + + ret += size; } - return size; + return ret; } static int perf_event_read_one(struct perf_event *event, -- cgit v1.2.3 From 02ffdbc866c8b1c8644601e9aa6155700eed4c91 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:50 +0100 Subject: perf: Optimize perf_event_task_sched_out Remove an update_context_time() call from the perf_event_task_sched_out() path and into the branch its needed. The call was both superfluous, because __perf_event_sched_out() already does it, and wrong, because it was done without holding ctx->lock. Place it in perf_event_sync_stat(), which is the only place it is needed and which does already hold ctx->lock. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.779516394@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 9a18ff28ea5b..65f4dab0ce60 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1120,6 +1120,8 @@ static void perf_event_sync_stat(struct perf_event_context *ctx, if (!ctx->nr_stat) return; + update_context_time(ctx); + event = list_first_entry(&ctx->event_list, struct perf_event, event_entry); @@ -1163,8 +1165,6 @@ void perf_event_task_sched_out(struct task_struct *task, if (likely(!ctx || !cpuctx->task_ctx)) return; - update_context_time(ctx); - rcu_read_lock(); parent = rcu_dereference(ctx->parent_ctx); next_ctx = next->perf_event_ctxp; -- cgit v1.2.3 From f6f83785222b0ee037f7be90731f62a649292b5e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:51 +0100 Subject: perf: Optimize __perf_event_read() Both callers actually have IRQs disabled, no need doing so again. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.863685796@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 3 --- 1 file changed, 3 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 65f4dab0ce60..e66f6c400d13 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1517,7 +1517,6 @@ static void __perf_event_read(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_event *event = info; struct perf_event_context *ctx = event->ctx; - unsigned long flags; /* * If this is a task context, we need to check whether it is @@ -1529,12 +1528,10 @@ static void __perf_event_read(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - local_irq_save(flags); if (ctx->is_active) update_context_time(ctx); event->pmu->read(event); update_event_times(event); - local_irq_restore(flags); } static u64 perf_event_read(struct perf_event *event) -- cgit v1.2.3 From 3dbebf15c5d3e265f751eec72c1538a00da4be27 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:52 +0100 Subject: perf: Simplify __perf_event_sync_stat Removes constraints from __perf_event_read() by leaving it with a single callsite; this callsite had ctx->lock held, the other one does not. Removes some superfluous code from __perf_event_sync_stat(). Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212508.918544317@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index e66f6c400d13..af150bbcfc5b 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1061,8 +1061,6 @@ static int context_equiv(struct perf_event_context *ctx1, && !ctx1->pin_count && !ctx2->pin_count; } -static void __perf_event_read(void *event); - static void __perf_event_sync_stat(struct perf_event *event, struct perf_event *next_event) { @@ -1080,8 +1078,8 @@ static void __perf_event_sync_stat(struct perf_event *event, */ switch (event->state) { case PERF_EVENT_STATE_ACTIVE: - __perf_event_read(event); - break; + event->pmu->read(event); + /* fall-through */ case PERF_EVENT_STATE_INACTIVE: update_event_times(event); -- cgit v1.2.3 From 58e5ad1de3d6ad931c84f0cc8ef0655c922f30ad Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:53 +0100 Subject: perf: Simplify __perf_event_read cpuctx is always active, task context is always active for current the previous condition verifies that if its a task context its for current, hence we can assume ctx->is_active. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212509.000272254@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index af150bbcfc5b..028619dd6d0e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1526,10 +1526,9 @@ static void __perf_event_read(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - if (ctx->is_active) - update_context_time(ctx); - event->pmu->read(event); + update_context_time(ctx); update_event_times(event); + event->pmu->read(event); } static u64 perf_event_read(struct perf_event *event) -- cgit v1.2.3 From 2b8988c9f7defe319cffe0cd362a7cd356c86f62 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:54 +0100 Subject: perf: Fix time locking Most sites updating ctx->time and event times do so under ctx->lock, make sure they all do. This was made possible by removing the __perf_event_read() call from __perf_event_sync_stat(), which already had this lock taken. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212509.102316434@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 028619dd6d0e..fdfae888a67c 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1526,8 +1526,11 @@ static void __perf_event_read(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; + spin_lock(&ctx->lock); update_context_time(ctx); update_event_times(event); + spin_unlock(&ctx->lock); + event->pmu->read(event); } @@ -1541,7 +1544,13 @@ static u64 perf_event_read(struct perf_event *event) smp_call_function_single(event->oncpu, __perf_event_read, event, 1); } else if (event->state == PERF_EVENT_STATE_INACTIVE) { + struct perf_event_context *ctx = event->ctx; + unsigned long flags; + + spin_lock_irqsave(&ctx->lock, flags); + update_context_time(ctx); update_event_times(event); + spin_unlock_irqrestore(&ctx->lock, flags); } return atomic64_read(&event->count); -- cgit v1.2.3 From 59ed446f792cc07d37b1536b9c4664d14e25e425 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:55 +0100 Subject: perf: Fix event scaling for inherited counters Properly account the full hierarchy of counters for both the count (we already did so) and the scale times (new). Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212509.153379276@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 3 ++- kernel/perf_event.c | 48 +++++++++++++++++++++++++--------------------- 2 files changed, 28 insertions(+), 23 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index a430ac3074af..36fe89f72641 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -782,7 +782,8 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, pid_t pid, perf_callback_t callback); -extern u64 perf_event_read_value(struct perf_event *event); +extern u64 perf_event_read_value(struct perf_event *event, + u64 *enabled, u64 *running); struct perf_sample_data { u64 type; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index fdfae888a67c..80f40da9a01e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1774,14 +1774,25 @@ static int perf_event_read_size(struct perf_event *event) return size; } -u64 perf_event_read_value(struct perf_event *event) +u64 perf_event_read_value(struct perf_event *event, u64 *enabled, u64 *running) { struct perf_event *child; u64 total = 0; + *enabled = 0; + *running = 0; + total += perf_event_read(event); - list_for_each_entry(child, &event->child_list, child_list) + *enabled += event->total_time_enabled + + atomic64_read(&event->child_total_time_enabled); + *running += event->total_time_running + + atomic64_read(&event->child_total_time_running); + + list_for_each_entry(child, &event->child_list, child_list) { total += perf_event_read(child); + *enabled += child->total_time_enabled; + *running += child->total_time_running; + } return total; } @@ -1793,19 +1804,15 @@ static int perf_event_read_group(struct perf_event *event, struct perf_event *leader = event->group_leader, *sub; int n = 0, size = 0, ret = 0; u64 values[5]; - u64 count; + u64 count, enabled, running; - count = perf_event_read_value(leader); + count = perf_event_read_value(leader, &enabled, &running); values[n++] = 1 + leader->nr_siblings; - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = leader->total_time_enabled + - atomic64_read(&leader->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = leader->total_time_running + - atomic64_read(&leader->child_total_time_running); - } + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = enabled; + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = running; values[n++] = count; if (read_format & PERF_FORMAT_ID) values[n++] = primary_event_id(leader); @@ -1820,7 +1827,7 @@ static int perf_event_read_group(struct perf_event *event, list_for_each_entry(sub, &leader->sibling_list, group_entry) { n = 0; - values[n++] = perf_event_read_value(sub); + values[n++] = perf_event_read_value(sub, &enabled, &running); if (read_format & PERF_FORMAT_ID) values[n++] = primary_event_id(sub); @@ -1838,18 +1845,15 @@ static int perf_event_read_group(struct perf_event *event, static int perf_event_read_one(struct perf_event *event, u64 read_format, char __user *buf) { + u64 enabled, running; u64 values[4]; int n = 0; - values[n++] = perf_event_read_value(event); - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = event->total_time_enabled + - atomic64_read(&event->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = event->total_time_running + - atomic64_read(&event->child_total_time_running); - } + values[n++] = perf_event_read_value(event, &enabled, &running); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = enabled; + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = running; if (read_format & PERF_FORMAT_ID) values[n++] = primary_event_id(event); -- cgit v1.2.3 From 6f10581aeaa5543a3b7a8c7a87a064375ec357f8 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 20 Nov 2009 22:19:56 +0100 Subject: perf: Fix locking for PERF_FORMAT_GROUP We should hold event->child_mutex when iterating the inherited counters, we should hold ctx->mutex when iterating siblings. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212509.251030114@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 20 +++++++++++++------- 1 file changed, 13 insertions(+), 7 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 80f40da9a01e..3ede0981f4ac 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1782,6 +1782,7 @@ u64 perf_event_read_value(struct perf_event *event, u64 *enabled, u64 *running) *enabled = 0; *running = 0; + mutex_lock(&event->child_mutex); total += perf_event_read(event); *enabled += event->total_time_enabled + atomic64_read(&event->child_total_time_enabled); @@ -1793,6 +1794,7 @@ u64 perf_event_read_value(struct perf_event *event, u64 *enabled, u64 *running) *enabled += child->total_time_enabled; *running += child->total_time_running; } + mutex_unlock(&event->child_mutex); return total; } @@ -1802,10 +1804,12 @@ static int perf_event_read_group(struct perf_event *event, u64 read_format, char __user *buf) { struct perf_event *leader = event->group_leader, *sub; - int n = 0, size = 0, ret = 0; + int n = 0, size = 0, ret = -EFAULT; + struct perf_event_context *ctx = leader->ctx; u64 values[5]; u64 count, enabled, running; + mutex_lock(&ctx->mutex); count = perf_event_read_value(leader, &enabled, &running); values[n++] = 1 + leader->nr_siblings; @@ -1820,9 +1824,9 @@ static int perf_event_read_group(struct perf_event *event, size = n * sizeof(u64); if (copy_to_user(buf, values, size)) - return -EFAULT; + goto unlock; - ret += size; + ret = size; list_for_each_entry(sub, &leader->sibling_list, group_entry) { n = 0; @@ -1833,11 +1837,15 @@ static int perf_event_read_group(struct perf_event *event, size = n * sizeof(u64); - if (copy_to_user(buf + size, values, size)) - return -EFAULT; + if (copy_to_user(buf + size, values, size)) { + ret = -EFAULT; + goto unlock; + } ret += size; } +unlock: + mutex_unlock(&ctx->mutex); return ret; } @@ -1884,12 +1892,10 @@ perf_read_hw(struct perf_event *event, char __user *buf, size_t count) return -ENOSPC; WARN_ON_ONCE(event->ctx->parent_ctx); - mutex_lock(&event->child_mutex); if (read_format & PERF_FORMAT_GROUP) ret = perf_event_read_group(event, read_format, buf); else ret = perf_event_read_one(event, read_format, buf); - mutex_unlock(&event->child_mutex); return ret; } -- cgit v1.2.3 From 8904b18046c2f050107f6449e887e7c1142b9ab9 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Fri, 20 Nov 2009 22:19:57 +0100 Subject: perf_events: Fix default watermark calculation This patch fixes the default watermark value for the sampling buffer. With the existing calculation (watermark = max(PAGE_SIZE, max_size / 2)), no notification was ever received when the buffer was exactly 1 page. This was because you would never cross the threshold (there is no partial samples). In certain configuration, there was no possibilty detecting the problem because there was not enough space left to store the LOST record.In fact, there may be a more generic problem here. The kernel should ensure that there is alaways enough space to store one LOST record. This patch sets the default watermark to half the buffer size. With such limit, we are guaranteed to get a notification even with a single page buffer assuming no sample is bigger than a page. Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091120212509.344964101@chello.nl> Signed-off-by: Ingo Molnar LKML-Reference: <1256302576-6169-1-git-send-email-eranian@gmail.com> --- kernel/perf_event.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 3ede0981f4ac..718fa939b1a7 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -2340,7 +2340,7 @@ perf_mmap_data_init(struct perf_event *event, struct perf_mmap_data *data) } if (!data->watermark) - data->watermark = max_t(long, PAGE_SIZE, max_size / 2); + data->watermark = max_size / 2; rcu_assign_pointer(event->data, data); -- cgit v1.2.3 From ce71b9df8893ec954e56c5979df6da274f20f65e Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Sun, 22 Nov 2009 05:26:55 +0100 Subject: tracing: Use the perf recursion protection from trace event When we commit a trace to perf, we first check if we are recursing in the same buffer so that we don't mess-up the buffer with a recursing trace. But later on, we do the same check from perf to avoid commit recursion. The recursion check is desired early before we touch the buffer but we want to do this check only once. Then export the recursion protection from perf and use it from the trace events before submitting a trace. v2: Put appropriate Reported-by tag Reported-by: Peter Zijlstra Signed-off-by: Frederic Weisbecker Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Steven Rostedt Cc: Masami Hiramatsu Cc: Jason Baron LKML-Reference: <1258864015-10579-1-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- include/linux/ftrace_event.h | 9 ++--- include/linux/perf_event.h | 4 +++ include/trace/ftrace.h | 23 +++++++------ kernel/perf_event.c | 68 +++++++++++++++++++++++++------------- kernel/trace/trace_event_profile.c | 14 ++++---- kernel/trace/trace_kprobe.c | 48 ++++++++++----------------- kernel/trace/trace_syscalls.c | 47 ++++++++++---------------- 7 files changed, 106 insertions(+), 107 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 43360c1d8f70..47bbdf9c38d0 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -137,13 +137,8 @@ struct ftrace_event_call { #define FTRACE_MAX_PROFILE_SIZE 2048 -struct perf_trace_buf { - char buf[FTRACE_MAX_PROFILE_SIZE]; - int recursion; -}; - -extern struct perf_trace_buf *perf_trace_buf; -extern struct perf_trace_buf *perf_trace_buf_nmi; +extern char *perf_trace_buf; +extern char *perf_trace_buf_nmi; #define MAX_FILTER_PRED 32 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */ diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 36fe89f72641..74e98b1d3391 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -874,6 +874,8 @@ extern int perf_output_begin(struct perf_output_handle *handle, extern void perf_output_end(struct perf_output_handle *handle); extern void perf_output_copy(struct perf_output_handle *handle, const void *buf, unsigned int len); +extern int perf_swevent_get_recursion_context(int **recursion); +extern void perf_swevent_put_recursion_context(int *recursion); #else static inline void perf_event_task_sched_in(struct task_struct *task, int cpu) { } @@ -902,6 +904,8 @@ static inline void perf_event_mmap(struct vm_area_struct *vma) { } static inline void perf_event_comm(struct task_struct *tsk) { } static inline void perf_event_fork(struct task_struct *tsk) { } static inline void perf_event_init(void) { } +static int perf_swevent_get_recursion_context(int **recursion) { return -1; } +static void perf_swevent_put_recursion_context(int *recursion) { } #endif diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 4945d1c99864..c222ef5238bf 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -724,16 +724,19 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ static void ftrace_profile_##call(proto) \ { \ struct ftrace_data_offsets_##call __maybe_unused __data_offsets;\ + extern int perf_swevent_get_recursion_context(int **recursion); \ + extern void perf_swevent_put_recursion_context(int *recursion); \ struct ftrace_event_call *event_call = &event_##call; \ extern void perf_tp_event(int, u64, u64, void *, int); \ struct ftrace_raw_##call *entry; \ - struct perf_trace_buf *trace_buf; \ u64 __addr = 0, __count = 1; \ unsigned long irq_flags; \ struct trace_entry *ent; \ int __entry_size; \ int __data_size; \ + char *trace_buf; \ char *raw_data; \ + int *recursion; \ int __cpu; \ int pc; \ \ @@ -749,6 +752,10 @@ static void ftrace_profile_##call(proto) \ return; \ \ local_irq_save(irq_flags); \ + \ + if (perf_swevent_get_recursion_context(&recursion)) \ + goto end_recursion; \ + \ __cpu = smp_processor_id(); \ \ if (in_nmi()) \ @@ -759,13 +766,7 @@ static void ftrace_profile_##call(proto) \ if (!trace_buf) \ goto end; \ \ - trace_buf = per_cpu_ptr(trace_buf, __cpu); \ - if (trace_buf->recursion++) \ - goto end_recursion; \ - \ - barrier(); \ - \ - raw_data = trace_buf->buf; \ + raw_data = per_cpu_ptr(trace_buf, __cpu); \ \ *(u64 *)(&raw_data[__entry_size - sizeof(u64)]) = 0ULL; \ entry = (struct ftrace_raw_##call *)raw_data; \ @@ -780,9 +781,9 @@ static void ftrace_profile_##call(proto) \ perf_tp_event(event_call->id, __addr, __count, entry, \ __entry_size); \ \ -end_recursion: \ - trace_buf->recursion--; \ -end: \ +end: \ + perf_swevent_put_recursion_context(recursion); \ +end_recursion: \ local_irq_restore(irq_flags); \ \ } diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 718fa939b1a7..aba822722300 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3880,34 +3880,42 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx, } } -static int *perf_swevent_recursion_context(struct perf_cpu_context *cpuctx) +/* + * Must be called with preemption disabled + */ +int perf_swevent_get_recursion_context(int **recursion) { + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + if (in_nmi()) - return &cpuctx->recursion[3]; + *recursion = &cpuctx->recursion[3]; + else if (in_irq()) + *recursion = &cpuctx->recursion[2]; + else if (in_softirq()) + *recursion = &cpuctx->recursion[1]; + else + *recursion = &cpuctx->recursion[0]; - if (in_irq()) - return &cpuctx->recursion[2]; + if (**recursion) + return -1; - if (in_softirq()) - return &cpuctx->recursion[1]; + (**recursion)++; - return &cpuctx->recursion[0]; + return 0; } -static void do_perf_sw_event(enum perf_type_id type, u32 event_id, - u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) +void perf_swevent_put_recursion_context(int *recursion) { - struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); - int *recursion = perf_swevent_recursion_context(cpuctx); - struct perf_event_context *ctx; - - if (*recursion) - goto out; + (*recursion)--; +} - (*recursion)++; - barrier(); +static void __do_perf_sw_event(enum perf_type_id type, u32 event_id, + u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct perf_event_context *ctx; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); rcu_read_lock(); perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, @@ -3920,12 +3928,25 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id, if (ctx) perf_swevent_ctx_event(ctx, type, event_id, nr, nmi, data, regs); rcu_read_unlock(); +} - barrier(); - (*recursion)--; +static void do_perf_sw_event(enum perf_type_id type, u32 event_id, + u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + int *recursion; + + preempt_disable(); + + if (perf_swevent_get_recursion_context(&recursion)) + goto out; + + __do_perf_sw_event(type, event_id, nr, nmi, data, regs); + perf_swevent_put_recursion_context(recursion); out: - put_cpu_var(perf_cpu_context); + preempt_enable(); } void __perf_sw_event(u32 event_id, u64 nr, int nmi, @@ -4159,7 +4180,8 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record, if (!regs) regs = task_pt_regs(current); - do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, + /* Trace events already protected against recursion */ + __do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, &data, regs); } EXPORT_SYMBOL_GPL(perf_tp_event); diff --git a/kernel/trace/trace_event_profile.c b/kernel/trace/trace_event_profile.c index e0d351b01f5a..d9c60f80aa0d 100644 --- a/kernel/trace/trace_event_profile.c +++ b/kernel/trace/trace_event_profile.c @@ -9,31 +9,33 @@ #include "trace.h" -struct perf_trace_buf *perf_trace_buf; +char *perf_trace_buf; EXPORT_SYMBOL_GPL(perf_trace_buf); -struct perf_trace_buf *perf_trace_buf_nmi; +char *perf_trace_buf_nmi; EXPORT_SYMBOL_GPL(perf_trace_buf_nmi); +typedef typeof(char [FTRACE_MAX_PROFILE_SIZE]) perf_trace_t ; + /* Count the events in use (per event id, not per instance) */ static int total_profile_count; static int ftrace_profile_enable_event(struct ftrace_event_call *event) { - struct perf_trace_buf *buf; + char *buf; int ret = -ENOMEM; if (atomic_inc_return(&event->profile_count)) return 0; if (!total_profile_count) { - buf = alloc_percpu(struct perf_trace_buf); + buf = (char *)alloc_percpu(perf_trace_t); if (!buf) goto fail_buf; rcu_assign_pointer(perf_trace_buf, buf); - buf = alloc_percpu(struct perf_trace_buf); + buf = (char *)alloc_percpu(perf_trace_t); if (!buf) goto fail_buf_nmi; @@ -79,7 +81,7 @@ int ftrace_profile_enable(int event_id) static void ftrace_profile_disable_event(struct ftrace_event_call *event) { - struct perf_trace_buf *buf, *nmi_buf; + char *buf, *nmi_buf; if (!atomic_add_negative(-1, &event->profile_count)) return; diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c index 3696476f307d..22e6f68b05b3 100644 --- a/kernel/trace/trace_kprobe.c +++ b/kernel/trace/trace_kprobe.c @@ -1208,11 +1208,12 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp); struct ftrace_event_call *call = &tp->call; struct kprobe_trace_entry *entry; - struct perf_trace_buf *trace_buf; struct trace_entry *ent; int size, __size, i, pc, __cpu; unsigned long irq_flags; + char *trace_buf; char *raw_data; + int *recursion; pc = preempt_count(); __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args); @@ -1227,6 +1228,10 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, * This also protects the rcu read side */ local_irq_save(irq_flags); + + if (perf_swevent_get_recursion_context(&recursion)) + goto end_recursion; + __cpu = smp_processor_id(); if (in_nmi()) @@ -1237,18 +1242,7 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, if (!trace_buf) goto end; - trace_buf = per_cpu_ptr(trace_buf, __cpu); - - if (trace_buf->recursion++) - goto end_recursion; - - /* - * Make recursion update visible before entering perf_tp_event - * so that we protect from perf recursions. - */ - barrier(); - - raw_data = trace_buf->buf; + raw_data = per_cpu_ptr(trace_buf, __cpu); /* Zero dead bytes from alignment to avoid buffer leak to userspace */ *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL; @@ -1263,9 +1257,9 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, entry->args[i] = call_fetch(&tp->args[i].fetch, regs); perf_tp_event(call->id, entry->ip, 1, entry, size); -end_recursion: - trace_buf->recursion--; end: + perf_swevent_put_recursion_context(recursion); +end_recursion: local_irq_restore(irq_flags); return 0; @@ -1278,10 +1272,11 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp); struct ftrace_event_call *call = &tp->call; struct kretprobe_trace_entry *entry; - struct perf_trace_buf *trace_buf; struct trace_entry *ent; int size, __size, i, pc, __cpu; unsigned long irq_flags; + char *trace_buf; + int *recursion; char *raw_data; pc = preempt_count(); @@ -1297,6 +1292,10 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, * This also protects the rcu read side */ local_irq_save(irq_flags); + + if (perf_swevent_get_recursion_context(&recursion)) + goto end_recursion; + __cpu = smp_processor_id(); if (in_nmi()) @@ -1307,18 +1306,7 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, if (!trace_buf) goto end; - trace_buf = per_cpu_ptr(trace_buf, __cpu); - - if (trace_buf->recursion++) - goto end_recursion; - - /* - * Make recursion update visible before entering perf_tp_event - * so that we protect from perf recursions. - */ - barrier(); - - raw_data = trace_buf->buf; + raw_data = per_cpu_ptr(trace_buf, __cpu); /* Zero dead bytes from alignment to avoid buffer leak to userspace */ *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL; @@ -1334,9 +1322,9 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, entry->args[i] = call_fetch(&tp->args[i].fetch, regs); perf_tp_event(call->id, entry->ret_ip, 1, entry, size); -end_recursion: - trace_buf->recursion--; end: + perf_swevent_put_recursion_context(recursion); +end_recursion: local_irq_restore(irq_flags); return 0; diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c index 51213b0aa81b..0bb934875263 100644 --- a/kernel/trace/trace_syscalls.c +++ b/kernel/trace/trace_syscalls.c @@ -477,10 +477,11 @@ static int sys_prof_refcount_exit; static void prof_syscall_enter(struct pt_regs *regs, long id) { struct syscall_metadata *sys_data; - struct perf_trace_buf *trace_buf; struct syscall_trace_enter *rec; unsigned long flags; + char *trace_buf; char *raw_data; + int *recursion; int syscall_nr; int size; int cpu; @@ -505,6 +506,9 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) /* Protect the per cpu buffer, begin the rcu read side */ local_irq_save(flags); + if (perf_swevent_get_recursion_context(&recursion)) + goto end_recursion; + cpu = smp_processor_id(); if (in_nmi()) @@ -515,18 +519,7 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) if (!trace_buf) goto end; - trace_buf = per_cpu_ptr(trace_buf, cpu); - - if (trace_buf->recursion++) - goto end_recursion; - - /* - * Make recursion update visible before entering perf_tp_event - * so that we protect from perf recursions. - */ - barrier(); - - raw_data = trace_buf->buf; + raw_data = per_cpu_ptr(trace_buf, cpu); /* zero the dead bytes from align to not leak stack to user */ *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL; @@ -539,9 +532,9 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) (unsigned long *)&rec->args); perf_tp_event(sys_data->enter_id, 0, 1, rec, size); -end_recursion: - trace_buf->recursion--; end: + perf_swevent_put_recursion_context(recursion); +end_recursion: local_irq_restore(flags); } @@ -588,10 +581,11 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) { struct syscall_metadata *sys_data; struct syscall_trace_exit *rec; - struct perf_trace_buf *trace_buf; unsigned long flags; int syscall_nr; + char *trace_buf; char *raw_data; + int *recursion; int size; int cpu; @@ -617,6 +611,10 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) /* Protect the per cpu buffer, begin the rcu read side */ local_irq_save(flags); + + if (perf_swevent_get_recursion_context(&recursion)) + goto end_recursion; + cpu = smp_processor_id(); if (in_nmi()) @@ -627,18 +625,7 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) if (!trace_buf) goto end; - trace_buf = per_cpu_ptr(trace_buf, cpu); - - if (trace_buf->recursion++) - goto end_recursion; - - /* - * Make recursion update visible before entering perf_tp_event - * so that we protect from perf recursions. - */ - barrier(); - - raw_data = trace_buf->buf; + raw_data = per_cpu_ptr(trace_buf, cpu); /* zero the dead bytes from align to not leak stack to user */ *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL; @@ -652,9 +639,9 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) perf_tp_event(sys_data->exit_id, 0, 1, rec, size); -end_recursion: - trace_buf->recursion--; end: + perf_swevent_put_recursion_context(recursion); +end_recursion: local_irq_restore(flags); } -- cgit v1.2.3 From 96b02d78a7e47cd189f6b307c5513fec6b2155dc Mon Sep 17 00:00:00 2001 From: Márton Németh Date: Sat, 21 Nov 2009 23:10:15 +0100 Subject: perf_event: Remove redundant zero fill MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The buffer is first zeroed out by memset(). Then strncpy() is used to fill the content. The strncpy() function also pads the string till the end of the specified length, which is redundant. The strncpy() does not ensures that the string will be properly closed with 0. Use strlcpy() instead. The semantic match that finds this kind of pattern is as follows: (http://coccinelle.lip6.fr/) // @@ expression buffer; expression size; expression str; @@ memset(buffer, 0, size); ... - strncpy( + strlcpy( buffer, str, sizeof(buffer) ); @@ expression buffer; expression size; expression str; @@ memset(&buffer, 0, size); ... - strncpy( + strlcpy( &buffer, str, sizeof(buffer)); @@ expression buffer; identifier field; expression size; expression str; @@ memset(buffer, 0, size); ... - strncpy( + strlcpy( buffer->field, str, sizeof(buffer->field) ); @@ expression buffer; identifier field; expression size; expression str; @@ memset(&buffer, 0, size); ... - strncpy( + strlcpy( buffer.field, str, sizeof(buffer.field)); // On strncpy() vs strlcpy() see http://www.gratisoft.us/todd/papers/strlcpy.html . Signed-off-by: Márton Németh Cc: Julia Lawall Cc: cocci@diku.dk Cc: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <4B086547.5040100@freemail.hu> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index aba822722300..b26cb03c1914 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3391,7 +3391,7 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) char comm[TASK_COMM_LEN]; memset(comm, 0, sizeof(comm)); - strncpy(comm, comm_event->task->comm, sizeof(comm)); + strlcpy(comm, comm_event->task->comm, sizeof(comm)); size = ALIGN(strlen(comm)+1, sizeof(u64)); comm_event->comm = comm; -- cgit v1.2.3 From 645e8cc0c9f01f07f384fd522b782e5e6ae9de18 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 22 Nov 2009 12:20:19 +0100 Subject: perf_events: Fix modular build Fix: ERROR: "perf_swevent_put_recursion_context" [fs/ext4/ext4.ko] undefined! ERROR: "perf_swevent_get_recursion_context" [fs/ext4/ext4.ko] undefined! Cc: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Steven Rostedt Cc: Masami Hiramatsu Cc: Jason Baron LKML-Reference: <1258864015-10579-1-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index b26cb03c1914..abe1ef47496c 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3903,11 +3903,13 @@ int perf_swevent_get_recursion_context(int **recursion) return 0; } +EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context); void perf_swevent_put_recursion_context(int *recursion) { (*recursion)--; } +EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); static void __do_perf_sw_event(enum perf_type_id type, u32 event_id, u64 nr, int nmi, -- cgit v1.2.3 From a4234bfcf4d72a10a99176cdef007345e9c3b4aa Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 23 Nov 2009 10:57:59 +0100 Subject: perf_events: Optimize the swcounter hotpath The structure init creates a bit memcpy, which shows up big time in perf annotate output: : ffffffff810a859d <__perf_sw_event>: 1.68 : ffffffff810a859d: 55 push %rbp 1.69 : ffffffff810a859e: 41 89 fa mov %edi,%r10d 0.01 : ffffffff810a85a1: 49 89 c9 mov %rcx,%r9 0.00 : ffffffff810a85a4: 31 c0 xor %eax,%eax 1.71 : ffffffff810a85a6: b9 16 00 00 00 mov $0x16,%ecx 0.00 : ffffffff810a85ab: 48 89 e5 mov %rsp,%rbp 0.00 : ffffffff810a85ae: 48 83 ec 60 sub $0x60,%rsp 1.52 : ffffffff810a85b2: 48 8d 7d a0 lea -0x60(%rbp),%rdi 85.20 : ffffffff810a85b6: f3 ab rep stos %eax,%es:(%rdi) None of the callees depends on the structure being pre-initialized, so only initialize ->addr. This gets rid of the memcpy overhead. Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index abe1ef47496c..20df8aba8da5 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3954,12 +3954,12 @@ out: void __perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { - struct perf_sample_data data = { - .addr = addr, - }; + struct perf_sample_data data; - do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, - &data, regs); + data.addr = addr; + data.raw = NULL; + + do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, &data, regs); } static void perf_swevent_read(struct perf_event *event) -- cgit v1.2.3 From a66a3052e2d4c5815d7ad26887b1d4193206e691 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:23 +0100 Subject: perf_events: Undo copy/paste damage We had two almost identical functions, avoid the duplication. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker Cc: Mike Galbraith Cc: Arnaldo Carvalho de Melo LKML-Reference: <20091123103819.537537928@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 31 +++++++++---------------------- 1 file changed, 9 insertions(+), 22 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 20df8aba8da5..e2daa10bb5ce 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1704,16 +1704,10 @@ static void free_event(struct perf_event *event) call_rcu(&event->rcu_head, free_event_rcu); } -/* - * Called when the last reference to the file is gone. - */ -static int perf_release(struct inode *inode, struct file *file) +int perf_event_release_kernel(struct perf_event *event) { - struct perf_event *event = file->private_data; struct perf_event_context *ctx = event->ctx; - file->private_data = NULL; - WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); perf_event_remove_from_context(event); @@ -1728,26 +1722,19 @@ static int perf_release(struct inode *inode, struct file *file) return 0; } +EXPORT_SYMBOL_GPL(perf_event_release_kernel); -int perf_event_release_kernel(struct perf_event *event) +/* + * Called when the last reference to the file is gone. + */ +static int perf_release(struct inode *inode, struct file *file) { - struct perf_event_context *ctx = event->ctx; - - WARN_ON_ONCE(ctx->parent_ctx); - mutex_lock(&ctx->mutex); - perf_event_remove_from_context(event); - mutex_unlock(&ctx->mutex); - - mutex_lock(&event->owner->perf_event_mutex); - list_del_init(&event->owner_entry); - mutex_unlock(&event->owner->perf_event_mutex); - put_task_struct(event->owner); + struct perf_event *event = file->private_data; - free_event(event); + file->private_data = NULL; - return 0; + return perf_event_release_kernel(event); } -EXPORT_SYMBOL_GPL(perf_event_release_kernel); static int perf_event_read_size(struct perf_event *event) { -- cgit v1.2.3 From 6c2bfcbe58e0dd39554be88940149f5aa11e17d1 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:24 +0100 Subject: perf_events: Fix style nits Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker LKML-Reference: <20091123103819.613427378@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index e2daa10bb5ce..1f14481c2337 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -447,9 +447,8 @@ retry: * can remove the event safely, if the call above did not * succeed. */ - if (!list_empty(&event->group_entry)) { + if (!list_empty(&event->group_entry)) list_del_event(event, ctx); - } spin_unlock_irq(&ctx->lock); } @@ -1033,10 +1032,10 @@ void __perf_event_sched_out(struct perf_event_context *ctx, update_context_time(ctx); perf_disable(); - if (ctx->nr_active) + if (ctx->nr_active) { list_for_each_entry(event, &ctx->group_list, group_entry) group_sched_out(event, cpuctx, ctx); - + } perf_enable(); out: spin_unlock(&ctx->lock); -- cgit v1.2.3 From 2e2af50b1fab3c40636839a7f439c167ae559533 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:25 +0100 Subject: perf_events: Disable events when we detach them If we leave the event in STATE_INACTIVE, any read of the event after the detach will increase the running count but not the enabled count and cause funny scaling artefacts. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker LKML-Reference: <20091123103819.689055515@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 1f14481c2337..fb851ec34461 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -294,6 +294,8 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) if (event->group_leader != event) event->group_leader->nr_siblings--; + event->state = PERF_EVENT_STATE_OFF; + /* * If this was a group event with sibling events then * upgrade the siblings to singleton events by adding them -- cgit v1.2.3 From 5e942bb33371254a474653123cd9e13a4c89ee44 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:26 +0100 Subject: perf_events: Update the context time on exit It appeared we did call update_event_times() on exit, but we failed to update the context time, which renders the former moot. Locking is a bit iffy, we call update_event_times under ctx->mutex instead of ctx->lock - the next patch fixes this. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker LKML-Reference: <20091123103819.764207355@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index fb851ec34461..8be2574b89b6 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4983,6 +4983,7 @@ void perf_event_exit_task(struct task_struct *child) * the events from it. */ unclone_ctx(child_ctx); + update_context_time(child_ctx); spin_unlock_irqrestore(&child_ctx->lock, flags); /* -- cgit v1.2.3 From f67218c3e93abaf0f480bb94b53d234853ffe4de Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:27 +0100 Subject: perf_events: Fix __perf_event_exit_task() vs. update_event_times() locking Move the update_event_times() call in __perf_event_exit_task() into list_del_event() because that holds the proper lock (ctx->lock) and seems a more natural place to do the last time update. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker LKML-Reference: <20091123103819.842455480@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 78 ++++++++++++++++++++++++++--------------------------- 1 file changed, 39 insertions(+), 39 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 8be2574b89b6..50f11b5f8c3d 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -246,6 +246,44 @@ static void perf_unpin_context(struct perf_event_context *ctx) put_ctx(ctx); } +static inline u64 perf_clock(void) +{ + return cpu_clock(smp_processor_id()); +} + +/* + * Update the record of the current time in a context. + */ +static void update_context_time(struct perf_event_context *ctx) +{ + u64 now = perf_clock(); + + ctx->time += now - ctx->timestamp; + ctx->timestamp = now; +} + +/* + * Update the total_time_enabled and total_time_running fields for a event. + */ +static void update_event_times(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + u64 run_end; + + if (event->state < PERF_EVENT_STATE_INACTIVE || + event->group_leader->state < PERF_EVENT_STATE_INACTIVE) + return; + + event->total_time_enabled = ctx->time - event->tstamp_enabled; + + if (event->state == PERF_EVENT_STATE_INACTIVE) + run_end = event->tstamp_stopped; + else + run_end = ctx->time; + + event->total_time_running = run_end - event->tstamp_running; +} + /* * Add a event from the lists for its context. * Must be called with ctx->mutex and ctx->lock held. @@ -294,6 +332,7 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) if (event->group_leader != event) event->group_leader->nr_siblings--; + update_event_times(event); event->state = PERF_EVENT_STATE_OFF; /* @@ -454,44 +493,6 @@ retry: spin_unlock_irq(&ctx->lock); } -static inline u64 perf_clock(void) -{ - return cpu_clock(smp_processor_id()); -} - -/* - * Update the record of the current time in a context. - */ -static void update_context_time(struct perf_event_context *ctx) -{ - u64 now = perf_clock(); - - ctx->time += now - ctx->timestamp; - ctx->timestamp = now; -} - -/* - * Update the total_time_enabled and total_time_running fields for a event. - */ -static void update_event_times(struct perf_event *event) -{ - struct perf_event_context *ctx = event->ctx; - u64 run_end; - - if (event->state < PERF_EVENT_STATE_INACTIVE || - event->group_leader->state < PERF_EVENT_STATE_INACTIVE) - return; - - event->total_time_enabled = ctx->time - event->tstamp_enabled; - - if (event->state == PERF_EVENT_STATE_INACTIVE) - run_end = event->tstamp_stopped; - else - run_end = ctx->time; - - event->total_time_running = run_end - event->tstamp_running; -} - /* * Update total_time_enabled and total_time_running for all events in a group. */ @@ -4931,7 +4932,6 @@ __perf_event_exit_task(struct perf_event *child_event, { struct perf_event *parent_event; - update_event_times(child_event); perf_event_remove_from_context(child_event); parent_event = child_event->parent; -- cgit v1.2.3 From 4ed7c92d68a5387ba5f7030dc76eab03558e27f5 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 11:37:29 +0100 Subject: perf_events: Undo some recursion damage Make perf_swevent_get_recursion_context return a context number and disable preemption. This could be used to remove the IRQ disable from the trace bit and index the per-cpu buffer with. Signed-off-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: Paul Mackerras LKML-Reference: <20091123103819.993226816@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 8 ++--- include/trace/ftrace.h | 17 ++++++----- kernel/perf_event.c | 71 +++++++++++++++++++------------------------ kernel/trace/trace_kprobe.c | 14 +++++---- kernel/trace/trace_syscalls.c | 14 +++++---- 5 files changed, 61 insertions(+), 63 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 74e98b1d3391..43adbd7f0010 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -874,8 +874,8 @@ extern int perf_output_begin(struct perf_output_handle *handle, extern void perf_output_end(struct perf_output_handle *handle); extern void perf_output_copy(struct perf_output_handle *handle, const void *buf, unsigned int len); -extern int perf_swevent_get_recursion_context(int **recursion); -extern void perf_swevent_put_recursion_context(int *recursion); +extern int perf_swevent_get_recursion_context(void); +extern void perf_swevent_put_recursion_context(int rctx); #else static inline void perf_event_task_sched_in(struct task_struct *task, int cpu) { } @@ -904,8 +904,8 @@ static inline void perf_event_mmap(struct vm_area_struct *vma) { } static inline void perf_event_comm(struct task_struct *tsk) { } static inline void perf_event_fork(struct task_struct *tsk) { } static inline void perf_event_init(void) { } -static int perf_swevent_get_recursion_context(int **recursion) { return -1; } -static void perf_swevent_put_recursion_context(int *recursion) { } +static inline int perf_swevent_get_recursion_context(void) { return -1; } +static inline void perf_swevent_put_recursion_context(int rctx) { } #endif diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index c222ef5238bf..c3417c13e3ed 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -724,8 +724,8 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ static void ftrace_profile_##call(proto) \ { \ struct ftrace_data_offsets_##call __maybe_unused __data_offsets;\ - extern int perf_swevent_get_recursion_context(int **recursion); \ - extern void perf_swevent_put_recursion_context(int *recursion); \ + extern int perf_swevent_get_recursion_context(void); \ + extern void perf_swevent_put_recursion_context(int rctx); \ struct ftrace_event_call *event_call = &event_##call; \ extern void perf_tp_event(int, u64, u64, void *, int); \ struct ftrace_raw_##call *entry; \ @@ -736,8 +736,8 @@ static void ftrace_profile_##call(proto) \ int __data_size; \ char *trace_buf; \ char *raw_data; \ - int *recursion; \ int __cpu; \ + int rctx; \ int pc; \ \ pc = preempt_count(); \ @@ -753,8 +753,9 @@ static void ftrace_profile_##call(proto) \ \ local_irq_save(irq_flags); \ \ - if (perf_swevent_get_recursion_context(&recursion)) \ - goto end_recursion; \ + rctx = perf_swevent_get_recursion_context(); \ + if (rctx < 0) \ + goto end_recursion; \ \ __cpu = smp_processor_id(); \ \ @@ -781,9 +782,9 @@ static void ftrace_profile_##call(proto) \ perf_tp_event(event_call->id, __addr, __count, entry, \ __entry_size); \ \ -end: \ - perf_swevent_put_recursion_context(recursion); \ -end_recursion: \ +end: \ + perf_swevent_put_recursion_context(rctx); \ +end_recursion: \ local_irq_restore(irq_flags); \ \ } diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 50f11b5f8c3d..0b0d5f72fe7d 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3869,45 +3869,50 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx, } } -/* - * Must be called with preemption disabled - */ -int perf_swevent_get_recursion_context(int **recursion) +int perf_swevent_get_recursion_context(void) { - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); + int rctx; if (in_nmi()) - *recursion = &cpuctx->recursion[3]; + rctx = 3; else if (in_irq()) - *recursion = &cpuctx->recursion[2]; + rctx = 2; else if (in_softirq()) - *recursion = &cpuctx->recursion[1]; + rctx = 1; else - *recursion = &cpuctx->recursion[0]; + rctx = 0; - if (**recursion) + if (cpuctx->recursion[rctx]) { + put_cpu_var(perf_cpu_context); return -1; + } - (**recursion)++; + cpuctx->recursion[rctx]++; + barrier(); - return 0; + return rctx; } EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context); -void perf_swevent_put_recursion_context(int *recursion) +void perf_swevent_put_recursion_context(int rctx) { - (*recursion)--; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + barrier(); + cpuctx->recursion[rctx]++; + put_cpu_var(perf_cpu_context); } EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); -static void __do_perf_sw_event(enum perf_type_id type, u32 event_id, - u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) +static void do_perf_sw_event(enum perf_type_id type, u32 event_id, + u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) { + struct perf_cpu_context *cpuctx; struct perf_event_context *ctx; - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + cpuctx = &__get_cpu_var(perf_cpu_context); rcu_read_lock(); perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, nr, nmi, data, regs); @@ -3921,34 +3926,22 @@ static void __do_perf_sw_event(enum perf_type_id type, u32 event_id, rcu_read_unlock(); } -static void do_perf_sw_event(enum perf_type_id type, u32 event_id, - u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - int *recursion; - - preempt_disable(); - - if (perf_swevent_get_recursion_context(&recursion)) - goto out; - - __do_perf_sw_event(type, event_id, nr, nmi, data, regs); - - perf_swevent_put_recursion_context(recursion); -out: - preempt_enable(); -} - void __perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { struct perf_sample_data data; + int rctx; + + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) + return; data.addr = addr; data.raw = NULL; do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, &data, regs); + + perf_swevent_put_recursion_context(rctx); } static void perf_swevent_read(struct perf_event *event) @@ -4172,7 +4165,7 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record, regs = task_pt_regs(current); /* Trace events already protected against recursion */ - __do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, + do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, &data, regs); } EXPORT_SYMBOL_GPL(perf_tp_event); diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c index 22e6f68b05b3..79ce6a2bd74f 100644 --- a/kernel/trace/trace_kprobe.c +++ b/kernel/trace/trace_kprobe.c @@ -1213,7 +1213,7 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, unsigned long irq_flags; char *trace_buf; char *raw_data; - int *recursion; + int rctx; pc = preempt_count(); __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args); @@ -1229,7 +1229,8 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, */ local_irq_save(irq_flags); - if (perf_swevent_get_recursion_context(&recursion)) + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) goto end_recursion; __cpu = smp_processor_id(); @@ -1258,7 +1259,7 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, perf_tp_event(call->id, entry->ip, 1, entry, size); end: - perf_swevent_put_recursion_context(recursion); + perf_swevent_put_recursion_context(rctx); end_recursion: local_irq_restore(irq_flags); @@ -1276,8 +1277,8 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, int size, __size, i, pc, __cpu; unsigned long irq_flags; char *trace_buf; - int *recursion; char *raw_data; + int rctx; pc = preempt_count(); __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args); @@ -1293,7 +1294,8 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, */ local_irq_save(irq_flags); - if (perf_swevent_get_recursion_context(&recursion)) + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) goto end_recursion; __cpu = smp_processor_id(); @@ -1323,7 +1325,7 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, perf_tp_event(call->id, entry->ret_ip, 1, entry, size); end: - perf_swevent_put_recursion_context(recursion); + perf_swevent_put_recursion_context(rctx); end_recursion: local_irq_restore(irq_flags); diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c index 41b6dd963daa..9189cbe86079 100644 --- a/kernel/trace/trace_syscalls.c +++ b/kernel/trace/trace_syscalls.c @@ -481,8 +481,8 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) unsigned long flags; char *trace_buf; char *raw_data; - int *recursion; int syscall_nr; + int rctx; int size; int cpu; @@ -506,7 +506,8 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) /* Protect the per cpu buffer, begin the rcu read side */ local_irq_save(flags); - if (perf_swevent_get_recursion_context(&recursion)) + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) goto end_recursion; cpu = smp_processor_id(); @@ -530,7 +531,7 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) perf_tp_event(sys_data->enter_id, 0, 1, rec, size); end: - perf_swevent_put_recursion_context(recursion); + perf_swevent_put_recursion_context(rctx); end_recursion: local_irq_restore(flags); } @@ -582,7 +583,7 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) int syscall_nr; char *trace_buf; char *raw_data; - int *recursion; + int rctx; int size; int cpu; @@ -609,7 +610,8 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) /* Protect the per cpu buffer, begin the rcu read side */ local_irq_save(flags); - if (perf_swevent_get_recursion_context(&recursion)) + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) goto end_recursion; cpu = smp_processor_id(); @@ -634,7 +636,7 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) perf_tp_event(sys_data->exit_id, 0, 1, rec, size); end: - perf_swevent_put_recursion_context(recursion); + perf_swevent_put_recursion_context(rctx); end_recursion: local_irq_restore(flags); } -- cgit v1.2.3 From acd1d7c1f8f3d848a3c5327dc09f8c1efb971678 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Nov 2009 15:00:36 +0100 Subject: perf_events: Restore sanity to scaling land It is quite possible to call update_event_times() on a context that isn't actually running and thereby confuse the thing. perf stat was reporting !100% scale values for software counters (2e2af50b perf_events: Disable events when we detach them, solved the worst of that, but there was still some left). The thing that happens is that because we are not self-reaping (we have a caring parent) there is a time between the last schedule (out) and having do_exit() called which will detach the events. This period would be accounted as enabled,!running because the event->state==INACTIVE, even though !event->ctx->is_active. Similar issues could have been observed by calling read() on a event while the attached task was not scheduled in. Solve this by teaching update_event_times() about ctx->is_active. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Mike Galbraith Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker LKML-Reference: <1258984836.4531.480.camel@laptop> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 0b0d5f72fe7d..0aafe85362fd 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -274,7 +274,12 @@ static void update_event_times(struct perf_event *event) event->group_leader->state < PERF_EVENT_STATE_INACTIVE) return; - event->total_time_enabled = ctx->time - event->tstamp_enabled; + if (ctx->is_active) + run_end = ctx->time; + else + run_end = event->tstamp_stopped; + + event->total_time_enabled = run_end - event->tstamp_enabled; if (event->state == PERF_EVENT_STATE_INACTIVE) run_end = event->tstamp_stopped; -- cgit v1.2.3 From f5ffe02e5046003ae7e2ce70d3d1c2a73331268b Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Mon, 23 Nov 2009 15:42:34 +0100 Subject: perf: Add kernel side syscall events support for breakpoints Add the remaining necessary bits to support breakpoints created through perf syscall. We don't use the software counter interface as: - We don't need to check against recursion, this is already done in hardware breakpoints arch level. - We already know the perf event we are dealing with when the event is to be committed. Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Prasad LKML-Reference: <1258987355-8751-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 34 +++++++++++++++++++++++++--------- 1 file changed, 25 insertions(+), 9 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 0aafe85362fd..9425c9600c89 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3831,6 +3831,20 @@ static int perf_swevent_is_counting(struct perf_event *event) static int perf_tp_event_match(struct perf_event *event, struct perf_sample_data *data); +static int perf_exclude_event(struct perf_event *event, + struct pt_regs *regs) +{ + if (regs) { + if (event->attr.exclude_user && user_mode(regs)) + return 1; + + if (event->attr.exclude_kernel && !user_mode(regs)) + return 1; + } + + return 0; +} + static int perf_swevent_match(struct perf_event *event, enum perf_type_id type, u32 event_id, @@ -3842,16 +3856,12 @@ static int perf_swevent_match(struct perf_event *event, if (event->attr.type != type) return 0; + if (event->attr.config != event_id) return 0; - if (regs) { - if (event->attr.exclude_user && user_mode(regs)) - return 0; - - if (event->attr.exclude_kernel && !user_mode(regs)) - return 0; - } + if (perf_exclude_event(event, regs)) + return 0; if (event->attr.type == PERF_TYPE_TRACEPOINT && !perf_tp_event_match(event, data)) @@ -4282,9 +4292,15 @@ static const struct pmu *bp_perf_event_init(struct perf_event *bp) return &perf_ops_bp; } -void perf_bp_event(struct perf_event *bp, void *regs) +void perf_bp_event(struct perf_event *bp, void *data) { - /* TODO */ + struct perf_sample_data sample; + struct pt_regs *regs = data; + + sample.addr = bp->attr.bp_addr; + + if (!perf_exclude_event(bp, regs)) + perf_swevent_add(bp, 1, 1, &sample, regs); } #else static void bp_perf_event_destroy(struct perf_event *event) -- cgit v1.2.3 From 184d3da8ef0ca552dffa0fdd35c046e058a2cf9a Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Mon, 23 Nov 2009 21:40:49 -0800 Subject: perf_events: Fix bogus copy_to_user() in perf_event_read_group() When using an event group, the value and id for non leaders events were wrong due to invalid offset into the outgoing buffer. Signed-off-by: Stephane Eranian Acked-by: Peter Zijlstra Cc: paulus@samba.org Cc: perfmon2-devel@lists.sourceforge.net LKML-Reference: <4b0b71e1.0508d00a.075e.ffff84a3@mx.google.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 9425c9600c89..accfd7bfe387 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1831,7 +1831,7 @@ static int perf_event_read_group(struct perf_event *event, size = n * sizeof(u64); - if (copy_to_user(buf + size, values, size)) { + if (copy_to_user(buf + ret, values, size)) { ret = -EFAULT; goto unlock; } -- cgit v1.2.3 From fe6126722718e51fba4879517c11ac12d9775bcc Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 24 Nov 2009 20:38:22 +0100 Subject: perf_events: Fix bad software/trace event recursion counting Commit 4ed7c92d68a5387ba5f7030dc76eab03558e27f5 (perf_events: Undo some recursion damage) has introduced a bad reference counting of the recursion context. putting the context behaves like getting it, dropping every software/trace events after the first one in a context. Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Arjan van de Ven Cc: Li Zefan Cc: Steven Rostedt LKML-Reference: <1259091502-5171-1-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index accfd7bfe387..35df94e344f2 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3914,7 +3914,7 @@ void perf_swevent_put_recursion_context(int rctx) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); barrier(); - cpuctx->recursion[rctx]++; + cpuctx->recursion[rctx]--; put_cpu_var(perf_cpu_context); } EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); -- cgit v1.2.3 From c6567f642e20bcc79abed030f44be5b0d6da2ded Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Thu, 26 Nov 2009 05:35:41 +0100 Subject: hw-breakpoints: Improve in-kernel event creation error granularity In fail case, perf_event_create_kernel_counter() returns NULL instead of an error, which doesn't help us to inform the user about the origin of the problem from the outer most callers. Often we can just return -EINVAL, which doesn't help anyone when it's eventually about a memory allocation failure. Then, this patch makes perf_event_create_kernel_counter() always return a detailed error code. Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Paul Mackerras Cc: Prasad LKML-Reference: <1259210142-5714-2-git-send-regression-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 35df94e344f2..34a1b9d7633e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4780,14 +4780,17 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, */ ctx = find_get_context(pid, cpu); - if (IS_ERR(ctx)) - return NULL; + if (IS_ERR(ctx)) { + err = PTR_ERR(ctx); + goto err_exit; + } event = perf_event_alloc(attr, cpu, ctx, NULL, NULL, callback, GFP_KERNEL); - err = PTR_ERR(event); - if (IS_ERR(event)) + if (IS_ERR(event)) { + err = PTR_ERR(event); goto err_put_context; + } event->filp = NULL; WARN_ON_ONCE(ctx->parent_ctx); @@ -4804,11 +4807,10 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, return event; -err_put_context: - if (err < 0) - put_ctx(ctx); - - return NULL; + err_put_context: + put_ctx(ctx); + err_exit: + return ERR_PTR(err); } EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter); -- cgit v1.2.3 From 80bbf6b641c8843b9d751a1f299aa7ee073ab9d4 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 25 Nov 2009 21:20:53 +0100 Subject: hw-breakpoints: Fix unused function in off-case bp_perf_event_destroy() is unused in its off-case version, let's remove it to fix the following warning reported by Stephen Rothwell in linux-next: kernel/perf_event.c:4306: warning: 'bp_perf_event_destroy' defined but not used Reported-by: Stephen Rothwell Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra LKML-Reference: <1259180453-5813-1-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 4 ---- 1 file changed, 4 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 34a1b9d7633e..f8c7939c57cf 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4303,10 +4303,6 @@ void perf_bp_event(struct perf_event *bp, void *data) perf_swevent_add(bp, 1, 1, &sample, regs); } #else -static void bp_perf_event_destroy(struct perf_event *event) -{ -} - static const struct pmu *bp_perf_event_init(struct perf_event *bp) { return NULL; -- cgit v1.2.3 From b2e74a265ded1a185f762ebaab967e9e0d008dd8 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Thu, 26 Nov 2009 09:24:30 -0800 Subject: perf_events: Fix read() bogus counts when in error state When a pinned group cannot be scheduled it goes into error state. Normally a group cannot go out of error state without being explicitly re-enabled or disabled. There was a bug in per-thread mode, whereby upon termination of the thread, the group would transition from error to off leading to bogus counts and timing information returned by read(). Fix it by clearing the error state. Signed-off-by: Stephane Eranian Acked-by: Peter Zijlstra Cc: Paul Mackerras Cc: perfmon2-devel@lists.sourceforge.net LKML-Reference: <4b0eb9ce.0508d00a.573b.ffffeab6@mx.google.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index f8c7939c57cf..0b9ca2d834dd 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -338,7 +338,16 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) event->group_leader->nr_siblings--; update_event_times(event); - event->state = PERF_EVENT_STATE_OFF; + + /* + * If event was in error state, then keep it + * that way, otherwise bogus counts will be + * returned on read(). The only way to get out + * of error state is by explicit re-enabling + * of the event + */ + if (event->state > PERF_EVENT_STATE_OFF) + event->state = PERF_EVENT_STATE_OFF; /* * If this was a group event with sibling events then -- cgit v1.2.3 From 59d069eb5ae9b033ed1c124c92e1532c4a958991 Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Tue, 1 Dec 2009 17:30:08 +0800 Subject: perf_event: Initialize data.period in perf_swevent_hrtimer() In current code in perf_swevent_hrtimer(), data.period is not initialized, The result is obvious wrong: # ./perf record -f -e cpu-clock make # ./perf report # Samples: 1740 # # Overhead Command ...... # ........ ........ .......................................... # 1025422183050275328.00% sh libc-2.9.90.so ... 1025422183050275328.00% perl libperl.so ... 1025422168240043264.00% perl [kernel] ... 1025422030011210752.00% perl [kernel] ... Signed-off-by: Xiao Guangrong Acked-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: LKML-Reference: <4B14E220.2050107@cn.fujitsu.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 0b9ca2d834dd..040ee517c808 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4010,6 +4010,7 @@ static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer) event->pmu->read(event); data.addr = 0; + data.period = event->hw.last_period; regs = get_irq_regs(); /* * In case we exclude kernel IPs or are somehow not in interrupt -- cgit v1.2.3 From ec70ccd806111ba3caf596def91a8580138b12db Mon Sep 17 00:00:00 2001 From: Kristian Høgsberg Date: Tue, 1 Dec 2009 15:05:01 -0500 Subject: perf: Don't free perf_mmap_data until work has been done MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit In the CONFIG_PERF_USE_VMALLOC case, perf_mmap_data_free() only schedules the cleanup of the perf_mmap_data struct. In that case we have to wait until the work has been done before we free data. Signed-off-by: Kristian Høgsberg Cc: David S. Miller Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker Cc: LKML-Reference: <1259697901-1747-1-git-send-email-krh@bitplanet.net> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 040ee517c808..6b7ddba1dd64 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -2210,6 +2210,7 @@ static void perf_mmap_data_free(struct perf_mmap_data *data) perf_mmap_free_page((unsigned long)data->user_page); for (i = 0; i < data->nr_pages; i++) perf_mmap_free_page((unsigned long)data->data_pages[i]); + kfree(data); } #else @@ -2250,6 +2251,7 @@ static void perf_mmap_data_free_work(struct work_struct *work) perf_mmap_unmark_page(base + (i * PAGE_SIZE)); vfree(base); + kfree(data); } static void perf_mmap_data_free(struct perf_mmap_data *data) @@ -2355,7 +2357,6 @@ static void perf_mmap_data_free_rcu(struct rcu_head *rcu_head) data = container_of(rcu_head, struct perf_mmap_data, rcu_head); perf_mmap_data_free(data); - kfree(data); } static void perf_mmap_data_release(struct perf_event *event) -- cgit v1.2.3