From b326e9560a28fc3e950637ef51847ed8f05c1335 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Sat, 5 Dec 2009 09:44:31 +0100 Subject: hw-breakpoints: Use overflow handler instead of the event callback struct perf_event::event callback was called when a breakpoint triggers. But this is a rather opaque callback, pretty tied-only to the breakpoint API and not really integrated into perf as it triggers even when we don't overflow. We prefer to use overflow_handler() as it fits into the perf events rules, being called only when we overflow. Reported-by: Peter Zijlstra Signed-off-by: Frederic Weisbecker Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo Cc: "K. Prasad" --- kernel/perf_event.c | 24 +++++++++--------------- 1 file changed, 9 insertions(+), 15 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 6b7ddba1dd64..fd43ff4ac860 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4286,15 +4286,8 @@ static void bp_perf_event_destroy(struct perf_event *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); + + err = register_perf_hw_breakpoint(bp); if (err) return ERR_PTR(err); @@ -4390,7 +4383,7 @@ 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, + perf_overflow_handler_t overflow_handler, gfp_t gfpflags) { const struct pmu *pmu; @@ -4433,10 +4426,10 @@ perf_event_alloc(struct perf_event_attr *attr, event->state = PERF_EVENT_STATE_INACTIVE; - if (!callback && parent_event) - callback = parent_event->callback; + if (!overflow_handler && parent_event) + overflow_handler = parent_event->overflow_handler; - event->callback = callback; + event->overflow_handler = overflow_handler; if (attr->disabled) event->state = PERF_EVENT_STATE_OFF; @@ -4776,7 +4769,8 @@ err_put_context: */ struct perf_event * perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, - pid_t pid, perf_callback_t callback) + pid_t pid, + perf_overflow_handler_t overflow_handler) { struct perf_event *event; struct perf_event_context *ctx; @@ -4793,7 +4787,7 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, } event = perf_event_alloc(attr, cpu, ctx, NULL, - NULL, callback, GFP_KERNEL); + NULL, overflow_handler, GFP_KERNEL); if (IS_ERR(event)) { err = PTR_ERR(event); goto err_put_context; -- cgit v1.2.3 From 44234adcdce38f83c56e05f808ce656175b4beeb Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 9 Dec 2009 09:25:48 +0100 Subject: hw-breakpoints: Modify breakpoints without unregistering them Currently, when ptrace needs to modify a breakpoint, like disabling it, changing its address, type or len, it calls modify_user_hw_breakpoint(). This latter will perform the heavy and racy task of unregistering the old breakpoint and registering a new one. This is racy as someone else might steal the reserved breakpoint slot under us, which is undesired as the breakpoint is only supposed to be modified, sometimes in the middle of a debugging workflow. We don't want our slot to be stolen in the middle. So instead of unregistering/registering the breakpoint, just disable it while we modify its breakpoint fields and re-enable it after if necessary. Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Prasad LKML-Reference: <1260347148-5519-1-git-send-regression-fweisbec@gmail.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/ptrace.c | 57 ++++++++++++++++++++----------------------- include/linux/hw_breakpoint.h | 4 +-- include/linux/perf_event.h | 5 +++- kernel/hw_breakpoint.c | 42 +++++++++++++++++++++++-------- kernel/perf_event.c | 4 +-- 5 files changed, 66 insertions(+), 46 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index b361d28061d0..7079ddaf0731 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -595,7 +595,7 @@ static unsigned long ptrace_get_dr7(struct perf_event *bp[]) return dr7; } -static struct perf_event * +static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type, struct task_struct *tsk, int disabled) { @@ -609,11 +609,11 @@ ptrace_modify_breakpoint(struct perf_event *bp, int len, int type, * written the address register first */ if (!bp) - return ERR_PTR(-EINVAL); + return -EINVAL; err = arch_bp_generic_fields(len, type, &gen_len, &gen_type); if (err) - return ERR_PTR(err); + return err; attr = bp->attr; attr.bp_len = gen_len; @@ -658,28 +658,17 @@ restore: if (!second_pass) continue; - thread->ptrace_bps[i] = NULL; - bp = ptrace_modify_breakpoint(bp, len, type, + rc = ptrace_modify_breakpoint(bp, len, type, tsk, 1); - if (IS_ERR(bp)) { - rc = PTR_ERR(bp); - thread->ptrace_bps[i] = NULL; + if (rc) break; - } - thread->ptrace_bps[i] = bp; } continue; } - bp = ptrace_modify_breakpoint(bp, len, type, tsk, 0); - - /* Incorrect bp, or we have a bug in bp API */ - if (IS_ERR(bp)) { - rc = PTR_ERR(bp); - thread->ptrace_bps[i] = NULL; + rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0); + if (rc) break; - } - thread->ptrace_bps[i] = bp; } /* * Make a second pass to free the remaining unused breakpoints @@ -737,26 +726,32 @@ static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr, attr.disabled = 1; bp = register_user_hw_breakpoint(&attr, ptrace_triggered, tsk); + + /* + * 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; } else { + int err; + bp = t->ptrace_bps[nr]; - t->ptrace_bps[nr] = NULL; attr = bp->attr; attr.bp_addr = addr; - bp = modify_user_hw_breakpoint(bp, &attr); + err = modify_user_hw_breakpoint(bp, &attr); + if (err) + return err; } - /* - * 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; } diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h index 42da1ce19ec0..69f07a9f1277 100644 --- a/include/linux/hw_breakpoint.h +++ b/include/linux/hw_breakpoint.h @@ -55,7 +55,7 @@ register_user_hw_breakpoint(struct perf_event_attr *attr, struct task_struct *tsk); /* FIXME: only change from the attr, and don't unregister */ -extern struct perf_event * +extern int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr); /* @@ -91,7 +91,7 @@ static inline struct perf_event * register_user_hw_breakpoint(struct perf_event_attr *attr, perf_overflow_handler_t triggered, struct task_struct *tsk) { return NULL; } -static inline struct perf_event * +static inline int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) { return NULL; } static inline struct perf_event * diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index bf3329413e18..64a53f74c9a9 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -872,6 +872,8 @@ extern void perf_output_copy(struct perf_output_handle *handle, const void *buf, unsigned int len); extern int perf_swevent_get_recursion_context(void); extern void perf_swevent_put_recursion_context(int rctx); +extern void perf_event_enable(struct perf_event *event); +extern void perf_event_disable(struct perf_event *event); #else static inline void perf_event_task_sched_in(struct task_struct *task, int cpu) { } @@ -902,7 +904,8 @@ static inline void perf_event_fork(struct task_struct *tsk) { } static inline void perf_event_init(void) { } static inline int perf_swevent_get_recursion_context(void) { return -1; } static inline void perf_swevent_put_recursion_context(int rctx) { } - +static inline void perf_event_enable(struct perf_event *event) { } +static inline void perf_event_disable(struct perf_event *event) { } #endif #define perf_output_put(handle, x) \ diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c index 03a0773ac2b2..366eedf949c0 100644 --- a/kernel/hw_breakpoint.c +++ b/kernel/hw_breakpoint.c @@ -320,18 +320,40 @@ EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); * @triggered: callback to trigger when we hit the breakpoint * @tsk: pointer to 'task_struct' of the process to which the address belongs */ -struct perf_event * -modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) +int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) { - /* - * 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); + u64 old_addr = bp->attr.bp_addr; + int old_type = bp->attr.bp_type; + int old_len = bp->attr.bp_len; + int err = 0; + + perf_event_disable(bp); + + bp->attr.bp_addr = attr->bp_addr; + bp->attr.bp_type = attr->bp_type; + bp->attr.bp_len = attr->bp_len; + + if (attr->disabled) + goto end; - return perf_event_create_kernel_counter(attr, -1, bp->ctx->task->pid, - bp->overflow_handler); + err = arch_validate_hwbkpt_settings(bp, bp->ctx->task); + if (!err) + perf_event_enable(bp); + + if (err) { + bp->attr.bp_addr = old_addr; + bp->attr.bp_type = old_type; + bp->attr.bp_len = old_len; + if (!bp->attr.disabled) + perf_event_enable(bp); + + return err; + } + +end: + bp->attr.disabled = attr->disabled; + + return 0; } EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); diff --git a/kernel/perf_event.c b/kernel/perf_event.c index fd43ff4ac860..3b0cf86eee84 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -567,7 +567,7 @@ static void __perf_event_disable(void *info) * is the current context on this CPU and preemption is disabled, * hence we can't get into perf_event_task_sched_out for this context. */ -static void perf_event_disable(struct perf_event *event) +void perf_event_disable(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; struct task_struct *task = ctx->task; @@ -971,7 +971,7 @@ static void __perf_event_enable(void *info) * perf_event_for_each_child or perf_event_for_each as described * for perf_event_disable. */ -static void perf_event_enable(struct perf_event *event) +void perf_event_enable(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; struct task_struct *task = ctx->task; -- cgit v1.2.3 From aa5452d70c0d559310598b243b8b1033c10056e7 Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Wed, 9 Dec 2009 11:28:13 +0800 Subject: perf_event: Clean up __perf_event_init_context() Clean up the code a bit: - define 'perf_cpu_context' variable with 'static' - use kzalloc() instead of kmalloc() and memset() Signed-off-by: Xiao Guangrong Reviewed-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <4B1F194D.7080306@cn.fujitsu.com> 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 3b0cf86eee84..2b06c45bfba9 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -36,7 +36,7 @@ /* * Each CPU has a list of per CPU events: */ -DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); +static DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); int perf_max_events __read_mostly = 1; static int perf_reserved_percpu __read_mostly; @@ -1579,7 +1579,6 @@ static void __perf_event_init_context(struct perf_event_context *ctx, struct task_struct *task) { - memset(ctx, 0, sizeof(*ctx)); spin_lock_init(&ctx->lock); mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->group_list); @@ -1654,7 +1653,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) } if (!ctx) { - ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); + ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL); err = -ENOMEM; if (!ctx) goto errout; @@ -5105,7 +5104,7 @@ int perf_event_init_task(struct task_struct *child) * First allocate and initialize a context for the child. */ - child_ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); + child_ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL); if (!child_ctx) return -ENOMEM; -- cgit v1.2.3 From b93f7978ad6b46133e9453b90ccc057dc2429e75 Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Wed, 9 Dec 2009 11:29:44 +0800 Subject: perf_event: Allocate children's perf_event_ctxp at the right time In current code, children task will allocate memory for 'child->perf_event_ctxp' if the parent is counted, we can do it only if the parent allowed children inherit it. It can save memory and reduce overhead. Signed-off-by: Xiao Guangrong Reviewed-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <4B1F19A8.5040805@cn.fujitsu.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 37 ++++++++++++++++++++++--------------- 1 file changed, 22 insertions(+), 15 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 2b06c45bfba9..77641ae6b23f 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -5083,7 +5083,7 @@ again: */ int perf_event_init_task(struct task_struct *child) { - struct perf_event_context *child_ctx, *parent_ctx; + struct perf_event_context *child_ctx = NULL, *parent_ctx; struct perf_event_context *cloned_ctx; struct perf_event *event; struct task_struct *parent = current; @@ -5098,20 +5098,6 @@ int perf_event_init_task(struct task_struct *child) if (likely(!parent->perf_event_ctxp)) return 0; - /* - * This is executed from the parent task context, so inherit - * events that have been marked for cloning. - * First allocate and initialize a context for the child. - */ - - child_ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL); - if (!child_ctx) - return -ENOMEM; - - __perf_event_init_context(child_ctx, child); - child->perf_event_ctxp = child_ctx; - get_task_struct(child); - /* * If the parent's context is a clone, pin it so it won't get * swapped under us. @@ -5142,6 +5128,26 @@ int perf_event_init_task(struct task_struct *child) continue; } + if (!child->perf_event_ctxp) { + /* + * This is executed from the parent task context, so + * inherit events that have been marked for cloning. + * First allocate and initialize a context for the + * child. + */ + + child_ctx = kzalloc(sizeof(struct perf_event_context), + GFP_KERNEL); + if (!child_ctx) { + ret = -ENOMEM; + goto exit; + } + + __perf_event_init_context(child_ctx, child); + child->perf_event_ctxp = child_ctx; + get_task_struct(child); + } + ret = inherit_group(event, parent, parent_ctx, child, child_ctx); if (ret) { @@ -5170,6 +5176,7 @@ int perf_event_init_task(struct task_struct *child) get_ctx(child_ctx->parent_ctx); } +exit: mutex_unlock(&parent_ctx->mutex); perf_unpin_context(parent_ctx); -- cgit v1.2.3 From ec89a06fd4e12301f11ab039ee07d2353a18addc Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Wed, 9 Dec 2009 11:30:36 +0800 Subject: perf_event: Cleanup for cpu_clock_perf_event_update() Using atomic64_xchg() instead of atomic64_read() and atomic64_set(). Signed-off-by: Xiao Guangrong Reviewed-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <4B1F19DC.90204@cn.fujitsu.com> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 77641ae6b23f..94e1b28333ae 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4079,8 +4079,7 @@ static void cpu_clock_perf_event_update(struct perf_event *event) u64 now; now = cpu_clock(cpu); - prev = atomic64_read(&event->hw.prev_count); - atomic64_set(&event->hw.prev_count, now); + prev = atomic64_xchg(&event->hw.prev_count, now); atomic64_add(now - prev, &event->count); } -- cgit v1.2.3 From 21140f4d3387aa2213f1deea0128df1dbf924379 Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Thu, 10 Dec 2009 14:00:51 +0800 Subject: perf_event: Fix perf_swevent_hrtimer() variable initialization fix: [] ? printk+0x1d/0x24 [] ? perf_prepare_sample+0x269/0x280 [] warn_slowpath_common+0x71/0xd0 [] ? perf_prepare_sample+0x269/0x280 [] warn_slowpath_null+0x1a/0x20 [] perf_prepare_sample+0x269/0x280 [] ? cpu_clock+0x53/0x90 [] __perf_event_overflow+0x2a8/0x300 [] perf_event_overflow+0x1b/0x30 [] perf_swevent_hrtimer+0x7f/0x120 This is because 'data.raw' variable not initialize. Signed-off-by: Xiao Guangrong Acked-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: Paul Mackerras LKML-Reference: <4B208E93.1010801@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 94e1b28333ae..3a5d6c4786bb 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.raw = NULL; data.period = event->hw.last_period; regs = get_irq_regs(); /* -- cgit v1.2.3 From 5e855db5d8fec44e6604eb245aa9077bbd3f0d05 Mon Sep 17 00:00:00 2001 From: Xiao Guangrong Date: Thu, 10 Dec 2009 17:08:54 +0800 Subject: perf_event: Fix variable initialization in other codepaths Signed-off-by: Xiao Guangrong Cc: Peter Zijlstra Cc: Frederic Weisbecker Cc: Paul Mackerras LKML-Reference: <4B20BAA6.7010609@cn.fujitsu.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.c | 4 ++++ kernel/perf_event.c | 1 + 2 files changed, 5 insertions(+) (limited to 'kernel/perf_event.c') diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index d35f26076ae5..1342f236e32a 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -1632,6 +1632,7 @@ static void intel_pmu_drain_bts_buffer(struct cpu_hw_events *cpuc) data.period = event->hw.last_period; data.addr = 0; + data.raw = NULL; regs.ip = 0; /* @@ -1749,6 +1750,7 @@ static int p6_pmu_handle_irq(struct pt_regs *regs) u64 val; data.addr = 0; + data.raw = NULL; cpuc = &__get_cpu_var(cpu_hw_events); @@ -1794,6 +1796,7 @@ static int intel_pmu_handle_irq(struct pt_regs *regs) u64 ack, status; data.addr = 0; + data.raw = NULL; cpuc = &__get_cpu_var(cpu_hw_events); @@ -1857,6 +1860,7 @@ static int amd_pmu_handle_irq(struct pt_regs *regs) u64 val; data.addr = 0; + data.raw = NULL; cpuc = &__get_cpu_var(cpu_hw_events); diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 3a5d6c4786bb..d891ec4a8100 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4300,6 +4300,7 @@ void perf_bp_event(struct perf_event *bp, void *data) struct perf_sample_data sample; struct pt_regs *regs = data; + sample.raw = NULL; sample.addr = bp->attr.bp_addr; if (!perf_exclude_event(bp, regs)) -- cgit v1.2.3 From e625cce1b73fb38b74e5387226534f7bcbfc36fe Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 17 Nov 2009 18:02:06 +0100 Subject: perf_event: Convert to raw_spinlock Convert locks which cannot be sleeping locks in preempt-rt to raw_spinlocks. Signed-off-by: Thomas Gleixner Acked-by: Peter Zijlstra Acked-by: Ingo Molnar --- include/linux/perf_event.h | 2 +- kernel/hw_breakpoint.c | 4 +- kernel/perf_event.c | 106 ++++++++++++++++++++++----------------------- 3 files changed, 56 insertions(+), 56 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 64a53f74c9a9..da7bdc23f279 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -681,7 +681,7 @@ struct perf_event_context { * Protect the states of the events in the list, * nr_active, and the list: */ - spinlock_t lock; + raw_spinlock_t lock; /* * Protect the list of events. Locking either mutex or lock * is sufficient to ensure the list doesn't change; to change diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c index 366eedf949c0..dbcbf6a33a08 100644 --- a/kernel/hw_breakpoint.c +++ b/kernel/hw_breakpoint.c @@ -96,7 +96,7 @@ static int task_bp_pinned(struct task_struct *tsk) list = &ctx->event_list; - spin_lock_irqsave(&ctx->lock, flags); + raw_spin_lock_irqsave(&ctx->lock, flags); /* * The current breakpoint counter is not included in the list @@ -107,7 +107,7 @@ static int task_bp_pinned(struct task_struct *tsk) count++; } - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); return count; } diff --git a/kernel/perf_event.c b/kernel/perf_event.c index e73e53c7582f..9052d6c8c9fd 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -203,14 +203,14 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags) * if so. If we locked the right context, then it * can't get swapped on us any more. */ - spin_lock_irqsave(&ctx->lock, *flags); + raw_spin_lock_irqsave(&ctx->lock, *flags); if (ctx != rcu_dereference(task->perf_event_ctxp)) { - spin_unlock_irqrestore(&ctx->lock, *flags); + raw_spin_unlock_irqrestore(&ctx->lock, *flags); goto retry; } if (!atomic_inc_not_zero(&ctx->refcount)) { - spin_unlock_irqrestore(&ctx->lock, *flags); + raw_spin_unlock_irqrestore(&ctx->lock, *flags); ctx = NULL; } } @@ -231,7 +231,7 @@ static struct perf_event_context *perf_pin_task_context(struct task_struct *task ctx = perf_lock_task_context(task, &flags); if (ctx) { ++ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } return ctx; } @@ -240,9 +240,9 @@ static void perf_unpin_context(struct perf_event_context *ctx) { unsigned long flags; - spin_lock_irqsave(&ctx->lock, flags); + raw_spin_lock_irqsave(&ctx->lock, flags); --ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); put_ctx(ctx); } @@ -427,7 +427,7 @@ static void __perf_event_remove_from_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * Protect the list operation against NMI by disabling the * events on a global level. @@ -449,7 +449,7 @@ static void __perf_event_remove_from_context(void *info) } perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } @@ -488,12 +488,12 @@ retry: task_oncpu_function_call(task, __perf_event_remove_from_context, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the context is active we need to retry the smp call. */ if (ctx->nr_active && !list_empty(&event->group_entry)) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -504,7 +504,7 @@ retry: */ if (!list_empty(&event->group_entry)) list_del_event(event, ctx); - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } /* @@ -535,7 +535,7 @@ static void __perf_event_disable(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * If the event is on, turn it off. @@ -551,7 +551,7 @@ static void __perf_event_disable(void *info) event->state = PERF_EVENT_STATE_OFF; } - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -584,12 +584,12 @@ void perf_event_disable(struct perf_event *event) retry: task_oncpu_function_call(task, __perf_event_disable, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the event is still active, we need to retry the cross-call. */ if (event->state == PERF_EVENT_STATE_ACTIVE) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -602,7 +602,7 @@ void perf_event_disable(struct perf_event *event) event->state = PERF_EVENT_STATE_OFF; } - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } static int @@ -770,7 +770,7 @@ static void __perf_install_in_context(void *info) cpuctx->task_ctx = ctx; } - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); @@ -820,7 +820,7 @@ static void __perf_install_in_context(void *info) unlock: perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -856,12 +856,12 @@ retry: task_oncpu_function_call(task, __perf_install_in_context, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * we need to retry the smp call. */ if (ctx->is_active && list_empty(&event->group_entry)) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -872,7 +872,7 @@ retry: */ if (list_empty(&event->group_entry)) add_event_to_ctx(event, ctx); - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } /* @@ -917,7 +917,7 @@ static void __perf_event_enable(void *info) cpuctx->task_ctx = ctx; } - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); @@ -959,7 +959,7 @@ static void __perf_event_enable(void *info) } unlock: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -985,7 +985,7 @@ void perf_event_enable(struct perf_event *event) return; } - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); if (event->state >= PERF_EVENT_STATE_INACTIVE) goto out; @@ -1000,10 +1000,10 @@ void perf_event_enable(struct perf_event *event) event->state = PERF_EVENT_STATE_OFF; retry: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); task_oncpu_function_call(task, __perf_event_enable, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the context is active and the event is still off, @@ -1020,7 +1020,7 @@ void perf_event_enable(struct perf_event *event) __perf_event_mark_enabled(event, ctx); out: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } static int perf_event_refresh(struct perf_event *event, int refresh) @@ -1042,7 +1042,7 @@ void __perf_event_sched_out(struct perf_event_context *ctx, { struct perf_event *event; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 0; if (likely(!ctx->nr_events)) goto out; @@ -1055,7 +1055,7 @@ void __perf_event_sched_out(struct perf_event_context *ctx, } perf_enable(); out: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1193,8 +1193,8 @@ void perf_event_task_sched_out(struct task_struct *task, * order we take the locks because no other cpu could * be trying to lock both of these tasks. */ - spin_lock(&ctx->lock); - spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); + raw_spin_lock(&ctx->lock); + raw_spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); if (context_equiv(ctx, next_ctx)) { /* * XXX do we need a memory barrier of sorts @@ -1208,8 +1208,8 @@ void perf_event_task_sched_out(struct task_struct *task, perf_event_sync_stat(ctx, next_ctx); } - spin_unlock(&next_ctx->lock); - spin_unlock(&ctx->lock); + raw_spin_unlock(&next_ctx->lock); + raw_spin_unlock(&ctx->lock); } rcu_read_unlock(); @@ -1251,7 +1251,7 @@ __perf_event_sched_in(struct perf_event_context *ctx, struct perf_event *event; int can_add_hw = 1; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; if (likely(!ctx->nr_events)) goto out; @@ -1306,7 +1306,7 @@ __perf_event_sched_in(struct perf_event_context *ctx, } perf_enable(); out: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1370,7 +1370,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) struct hw_perf_event *hwc; u64 interrupts, freq; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (event->state != PERF_EVENT_STATE_ACTIVE) continue; @@ -1425,7 +1425,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) perf_enable(); } } - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1438,7 +1438,7 @@ static void rotate_ctx(struct perf_event_context *ctx) if (!ctx->nr_events) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * Rotate the first entry last (works just fine for group events too): */ @@ -1449,7 +1449,7 @@ static void rotate_ctx(struct perf_event_context *ctx) } perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } void perf_event_task_tick(struct task_struct *curr, int cpu) @@ -1498,7 +1498,7 @@ static void perf_event_enable_on_exec(struct task_struct *task) __perf_event_task_sched_out(ctx); - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); list_for_each_entry(event, &ctx->group_list, group_entry) { if (!event->attr.enable_on_exec) @@ -1516,7 +1516,7 @@ static void perf_event_enable_on_exec(struct task_struct *task) if (enabled) unclone_ctx(ctx); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); perf_event_task_sched_in(task, smp_processor_id()); out: @@ -1542,10 +1542,10 @@ static void __perf_event_read(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); update_context_time(ctx); update_event_times(event); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); event->pmu->read(event); } @@ -1563,10 +1563,10 @@ static u64 perf_event_read(struct perf_event *event) struct perf_event_context *ctx = event->ctx; unsigned long flags; - spin_lock_irqsave(&ctx->lock, flags); + raw_spin_lock_irqsave(&ctx->lock, flags); update_context_time(ctx); update_event_times(event); - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } return atomic64_read(&event->count); @@ -1579,7 +1579,7 @@ static void __perf_event_init_context(struct perf_event_context *ctx, struct task_struct *task) { - spin_lock_init(&ctx->lock); + raw_spin_lock_init(&ctx->lock); mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->group_list); INIT_LIST_HEAD(&ctx->event_list); @@ -1649,7 +1649,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) ctx = perf_lock_task_context(task, &flags); if (ctx) { unclone_ctx(ctx); - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } if (!ctx) { @@ -1987,7 +1987,7 @@ static int perf_event_period(struct perf_event *event, u64 __user *arg) if (!value) return -EINVAL; - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); if (event->attr.freq) { if (value > sysctl_perf_event_sample_rate) { ret = -EINVAL; @@ -2000,7 +2000,7 @@ static int perf_event_period(struct perf_event *event, u64 __user *arg) event->hw.sample_period = value; } unlock: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); return ret; } @@ -4992,7 +4992,7 @@ void perf_event_exit_task(struct task_struct *child) * reading child->perf_event_ctxp, we wait until it has * incremented the context's refcount before we do put_ctx below. */ - spin_lock(&child_ctx->lock); + raw_spin_lock(&child_ctx->lock); child->perf_event_ctxp = NULL; /* * If this context is a clone; unclone it so it can't get @@ -5001,7 +5001,7 @@ void perf_event_exit_task(struct task_struct *child) */ unclone_ctx(child_ctx); update_context_time(child_ctx); - spin_unlock_irqrestore(&child_ctx->lock, flags); + raw_spin_unlock_irqrestore(&child_ctx->lock, flags); /* * Report the task dead after unscheduling the events so that we @@ -5292,11 +5292,11 @@ perf_set_reserve_percpu(struct sysdev_class *class, perf_reserved_percpu = val; for_each_online_cpu(cpu) { cpuctx = &per_cpu(perf_cpu_context, cpu); - spin_lock_irq(&cpuctx->ctx.lock); + raw_spin_lock_irq(&cpuctx->ctx.lock); mpt = min(perf_max_events - cpuctx->ctx.nr_events, perf_max_events - perf_reserved_percpu); cpuctx->max_pertask = mpt; - spin_unlock_irq(&cpuctx->ctx.lock); + raw_spin_unlock_irq(&cpuctx->ctx.lock); } spin_unlock(&perf_resource_lock); -- cgit v1.2.3 From 0f624e7e5625f4c30c836b7a5decfe2553582391 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Tue, 15 Dec 2009 19:40:32 +1100 Subject: perf_event: Fix incorrect range check on cpu number It is quite legitimate for CPUs to be numbered sparsely, meaning that it possible for an online CPU to have a number which is greater than the total count of possible CPUs. Currently find_get_context() has a sanity check on the cpu number where it checks it against num_possible_cpus(). This test can fail for a legitimate cpu number if the cpu_possible_mask is sparsely populated. This fixes the problem by checking the CPU number against nr_cpumask_bits instead, since that is the appropriate check to ensure that the cpu number is same to pass to cpu_isset() subsequently. Reported-by: Michael Neuling Signed-off-by: Paul Mackerras Tested-by: Michael Neuling Acked-by: Peter Zijlstra Cc: LKML-Reference: <20091215084032.GA18661@brick.ozlabs.ibm.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 d891ec4a8100..8823b0885183 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1603,7 +1603,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) return ERR_PTR(-EACCES); - if (cpu < 0 || cpu > num_possible_cpus()) + if (cpu < 0 || cpu >= nr_cpumask_bits) return ERR_PTR(-EINVAL); /* -- cgit v1.2.3 From f13c12c634e124d5d31f912b969d542a016d6105 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 15 Dec 2009 19:43:11 +0100 Subject: perf_events: Fix perf_event_attr layout The miss-alignment of bp_addr created a 32bit hole, causing different structure packings on 32 and 64 bit machines. Fix that by moving __reserve_2 into that hole. Further, remove the useless struct and redundant __bp_reserve muck. Signed-off-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: Arnaldo Carvalho de Melo LKML-Reference: <1260902591.8023.781.camel@laptop> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 12 +++--------- kernel/perf_event.c | 2 +- 2 files changed, 4 insertions(+), 10 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 64a53f74c9a9..5fcbf7d2712a 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -211,17 +211,11 @@ struct perf_event_attr { __u32 wakeup_watermark; /* bytes before wakeup */ }; - struct { /* Hardware breakpoint info */ - __u64 bp_addr; - __u32 bp_type; - __u32 bp_len; - __u64 __bp_reserved_1; - __u64 __bp_reserved_2; - }; - __u32 __reserved_2; - __u64 __reserved_3; + __u64 bp_addr; + __u32 bp_type; + __u32 bp_len; }; /* diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 8823b0885183..0dd8e5d02c66 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4564,7 +4564,7 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr, if (attr->type >= PERF_TYPE_MAX) return -EINVAL; - if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) + if (attr->__reserved_1 || attr->__reserved_2) return -EINVAL; if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) -- cgit v1.2.3 From f4c4176f21533e22bcc292030da72bcfa105f5b8 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 16 Dec 2009 17:55:54 +0100 Subject: perf events: Allow per-task-per-cpu counters In order to allow for per-task-per-cpu counters, useful for scalability when profiling task hierarchies, we allow installing events with event->cpu != -1 in task contexts. __perf_event_sched_in() already skips events where ->cpu mis-matches the current cpu, fix up __perf_install_in_context() and __perf_event_enable() to also respect this filter. This does lead to vary hard to interpret enabled/running times for such counters, but I don't see a simple solution for that. Signed-off-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: fweisbec@gmail.com Cc: Paul Mackerras LKML-Reference: <20091216165904.831451147@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 0dd8e5d02c66..2e0aaa34fc7e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -782,6 +782,9 @@ static void __perf_install_in_context(void *info) add_event_to_ctx(event, ctx); + if (event->cpu != -1 && event->cpu != smp_processor_id()) + goto unlock; + /* * Don't put the event on if it is disabled or if * it is in a group and the group isn't on. @@ -925,6 +928,9 @@ static void __perf_event_enable(void *info) goto unlock; __perf_event_mark_enabled(event, ctx); + if (event->cpu != -1 && event->cpu != smp_processor_id()) + goto unlock; + /* * If the event is in a group and isn't the group leader, * then don't put it on unless the group is on. @@ -1595,10 +1601,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) unsigned long flags; int err; - /* - * If cpu is not a wildcard then this is a percpu event: - */ - if (cpu != -1) { + if (pid == -1 && cpu != -1) { /* Must be root to operate on a CPU event: */ if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) return ERR_PTR(-EACCES); -- cgit v1.2.3 From f6325e30ebd6fc870315b017a5d4a6ab15bf790b Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Thu, 17 Dec 2009 11:43:08 -0600 Subject: cpumask: use cpu_online in kernel/perf_event.c Also, we want to check against nr_cpu_ids, not num_possible_cpus(). The latter works, but the correct bounds check is < nr_cpu_ids. Signed-off-by: Rusty Russell To: Thomas Gleixner --- 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 8ab86988bd24..97d1a3dd7a59 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1614,7 +1614,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) * offline CPU and activate it when the CPU comes up, but * that's for later. */ - if (!cpu_isset(cpu, cpu_online_map)) + if (!cpu_online(cpu)) return ERR_PTR(-ENODEV); cpuctx = &per_cpu(perf_cpu_context, cpu); -- cgit v1.2.3 From 5d27c23df09b702868d9a3bff86ec6abd22963ac Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 17 Dec 2009 13:16:32 +0100 Subject: perf events: Dont report side-band events on each cpu for per-task-per-cpu events Acme noticed that his FORK/MMAP numbers were inflated by about the same factor as his cpu-count. This led to the discovery of a few more sites that need to respect the event->cpu filter. Reported-by: Arnaldo Carvalho de Melo Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20091217121830.215333434@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 32 ++++++++++++++++++-------------- 1 file changed, 18 insertions(+), 14 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 8ab86988bd24..03cc061398d1 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1381,6 +1381,9 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) if (event->state != PERF_EVENT_STATE_ACTIVE) continue; + if (event->cpu != -1 && event->cpu != smp_processor_id()) + continue; + hwc = &event->hw; interrupts = hwc->interrupts; @@ -3265,6 +3268,9 @@ static void perf_event_task_output(struct perf_event *event, static int perf_event_task_match(struct perf_event *event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.comm || event->attr.mmap || event->attr.task) return 1; @@ -3290,12 +3296,11 @@ static void perf_event_task_event(struct perf_task_event *task_event) rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_task_ctx(&cpuctx->ctx, task_event); - put_cpu_var(perf_cpu_context); - if (!ctx) ctx = rcu_dereference(task_event->task->perf_event_ctxp); if (ctx) perf_event_task_ctx(ctx, task_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); } @@ -3372,6 +3377,9 @@ static void perf_event_comm_output(struct perf_event *event, static int perf_event_comm_match(struct perf_event *event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.comm) return 1; @@ -3408,15 +3416,10 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_comm_ctx(&cpuctx->ctx, comm_event); - put_cpu_var(perf_cpu_context); - - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ ctx = rcu_dereference(current->perf_event_ctxp); if (ctx) perf_event_comm_ctx(ctx, comm_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); } @@ -3491,6 +3494,9 @@ static void perf_event_mmap_output(struct perf_event *event, static int perf_event_mmap_match(struct perf_event *event, struct perf_mmap_event *mmap_event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.mmap) return 1; @@ -3564,15 +3570,10 @@ got_name: rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_mmap_ctx(&cpuctx->ctx, mmap_event); - put_cpu_var(perf_cpu_context); - - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ ctx = rcu_dereference(current->perf_event_ctxp); if (ctx) perf_event_mmap_ctx(ctx, mmap_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); kfree(buf); @@ -3863,6 +3864,9 @@ static int perf_swevent_match(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (!perf_swevent_is_counting(event)) return 0; -- cgit v1.2.3 From 628ff7c1d8d8466a5ad8078bd0206a130f8b8a51 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Fri, 18 Dec 2009 09:41:24 -0800 Subject: anonfd: Allow making anon files read-only It seems a couple places such as arch/ia64/kernel/perfmon.c and drivers/infiniband/core/uverbs_main.c could use anon_inode_getfile() instead of a private pseudo-fs + alloc_file(), if only there were a way to get a read-only file. So provide this by having anon_inode_getfile() create a read-only file if we pass O_RDONLY in flags. Signed-off-by: Roland Dreier Signed-off-by: Al Viro --- fs/anon_inodes.c | 12 ++++++++++-- fs/eventfd.c | 2 +- fs/eventpoll.c | 2 +- fs/signalfd.c | 2 +- fs/timerfd.c | 2 +- kernel/perf_event.c | 2 +- virt/kvm/kvm_main.c | 4 ++-- 7 files changed, 17 insertions(+), 9 deletions(-) (limited to 'kernel/perf_event.c') diff --git a/fs/anon_inodes.c b/fs/anon_inodes.c index 2c994591f4d7..598237e97221 100644 --- a/fs/anon_inodes.c +++ b/fs/anon_inodes.c @@ -89,11 +89,19 @@ struct file *anon_inode_getfile(const char *name, struct qstr this; struct path path; struct file *file; + fmode_t mode; int error; if (IS_ERR(anon_inode_inode)) return ERR_PTR(-ENODEV); + switch (flags & O_ACCMODE) { + case O_RDONLY: mode = FMODE_READ; break; + case O_WRONLY: mode = FMODE_WRITE; break; + case O_RDWR: mode = FMODE_READ | FMODE_WRITE; break; + default: return ERR_PTR(-EINVAL); + } + if (fops->owner && !try_module_get(fops->owner)) return ERR_PTR(-ENOENT); @@ -121,13 +129,13 @@ struct file *anon_inode_getfile(const char *name, d_instantiate(path.dentry, anon_inode_inode); error = -ENFILE; - file = alloc_file(&path, FMODE_READ | FMODE_WRITE, fops); + file = alloc_file(&path, mode, fops); if (!file) goto err_dput; file->f_mapping = anon_inode_inode->i_mapping; file->f_pos = 0; - file->f_flags = O_RDWR | (flags & O_NONBLOCK); + file->f_flags = flags & (O_ACCMODE | O_NONBLOCK); file->f_version = 0; file->private_data = priv; diff --git a/fs/eventfd.c b/fs/eventfd.c index 8b47e4200e65..d26402ff06ea 100644 --- a/fs/eventfd.c +++ b/fs/eventfd.c @@ -339,7 +339,7 @@ struct file *eventfd_file_create(unsigned int count, int flags) ctx->flags = flags; file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, - flags & EFD_SHARED_FCNTL_FLAGS); + O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS)); if (IS_ERR(file)) eventfd_free_ctx(ctx); diff --git a/fs/eventpoll.c b/fs/eventpoll.c index 366c503f9657..bd056a5b4efc 100644 --- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -1206,7 +1206,7 @@ SYSCALL_DEFINE1(epoll_create1, int, flags) * a file structure and a free file descriptor. */ error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep, - flags & O_CLOEXEC); + O_RDWR | (flags & O_CLOEXEC)); if (error < 0) ep_free(ep); diff --git a/fs/signalfd.c b/fs/signalfd.c index b07565c94386..1dabe4ee02fe 100644 --- a/fs/signalfd.c +++ b/fs/signalfd.c @@ -236,7 +236,7 @@ SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask, * anon_inode_getfd() will install the fd. */ ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx, - flags & (O_CLOEXEC | O_NONBLOCK)); + O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK))); if (ufd < 0) kfree(ctx); } else { diff --git a/fs/timerfd.c b/fs/timerfd.c index b042bd7034b1..1bfc95ad5f71 100644 --- a/fs/timerfd.c +++ b/fs/timerfd.c @@ -200,7 +200,7 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS); ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx, - flags & TFD_SHARED_FCNTL_FLAGS); + O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS)); if (ufd < 0) kfree(ctx); diff --git a/kernel/perf_event.c b/kernel/perf_event.c index e0eb4a2fe183..1f38270f08c7 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4724,7 +4724,7 @@ SYSCALL_DEFINE5(perf_event_open, if (IS_ERR(event)) goto err_put_context; - err = anon_inode_getfd("[perf_event]", &perf_fops, event, 0); + err = anon_inode_getfd("[perf_event]", &perf_fops, event, O_RDWR); if (err < 0) goto err_free_put_context; diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c index e1f2bf8d7b1e..b5af88167613 100644 --- a/virt/kvm/kvm_main.c +++ b/virt/kvm/kvm_main.c @@ -1177,7 +1177,7 @@ static struct file_operations kvm_vcpu_fops = { */ static int create_vcpu_fd(struct kvm_vcpu *vcpu) { - return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0); + return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR); } /* @@ -1638,7 +1638,7 @@ static int kvm_dev_ioctl_create_vm(void) kvm = kvm_create_vm(); if (IS_ERR(kvm)) return PTR_ERR(kvm); - fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0); + fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); if (fd < 0) kvm_put_kvm(kvm); -- cgit v1.2.3 From 05cbaa2853cdfc255fdd04e65a82bfe9208c4e52 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 30 Dec 2009 16:00:35 +0100 Subject: perf: Fix NULL deref in inheritance code Liming found a NULL deref when a task has a perf context but no counters when it forks. This can occur in two cases, a race during construction where the fork hits after installing the context but before the first counter gets inserted, or more reproducably, a fork after the last counter is closed (which leaves the context around). Reported-by: Wang Liming Signed-off-by: Peter Zijlstra Cc: Frederic Weisbecker Cc: Paul Mackerras CC: LKML-Reference: <1262185684.7135.222.camel@laptop> 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 03cc061398d1..58ed1dae5875 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -5148,7 +5148,7 @@ int perf_event_init_task(struct task_struct *child) GFP_KERNEL); if (!child_ctx) { ret = -ENOMEM; - goto exit; + break; } __perf_event_init_context(child_ctx, child); @@ -5164,7 +5164,7 @@ int perf_event_init_task(struct task_struct *child) } } - if (inherited_all) { + if (child_ctx && inherited_all) { /* * Mark the child context as a clone of the parent * context, or of whatever the parent is a clone of. @@ -5184,7 +5184,6 @@ int perf_event_init_task(struct task_struct *child) get_ctx(child_ctx->parent_ctx); } -exit: mutex_unlock(&parent_ctx->mutex); perf_unpin_context(parent_ctx); -- cgit v1.2.3 From 22e190851f8709c48baf00ed9ce6144cdc54d025 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 18 Jan 2010 09:12:32 +0100 Subject: perf: Honour event state for aux stream data Anton reported that perf record kept receiving events even after calling ioctl(PERF_EVENT_IOC_DISABLE). It turns out that FORK,COMM and MMAP events didn't respect the disabled state and kept flowing in. Reported-by: Anton Blanchard Signed-off-by: Peter Zijlstra Tested-by: Anton Blanchard LKML-Reference: <1263459187.4244.265.camel@laptop> CC: stable@kernel.org 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 603c0d8b5df1..d27746bd3a06 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -3268,6 +3268,9 @@ static void perf_event_task_output(struct perf_event *event, static int perf_event_task_match(struct perf_event *event) { + if (event->state != PERF_EVENT_STATE_ACTIVE) + return 0; + if (event->cpu != -1 && event->cpu != smp_processor_id()) return 0; @@ -3377,6 +3380,9 @@ static void perf_event_comm_output(struct perf_event *event, static int perf_event_comm_match(struct perf_event *event) { + if (event->state != PERF_EVENT_STATE_ACTIVE) + return 0; + if (event->cpu != -1 && event->cpu != smp_processor_id()) return 0; @@ -3494,6 +3500,9 @@ static void perf_event_mmap_output(struct perf_event *event, static int perf_event_mmap_match(struct perf_event *event, struct perf_mmap_event *mmap_event) { + if (event->state != PERF_EVENT_STATE_ACTIVE) + return 0; + if (event->cpu != -1 && event->cpu != smp_processor_id()) return 0; -- cgit v1.2.3