From 0793a61d4df8daeac6492dbf8d2f3e5713caae5e Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Thu, 4 Dec 2008 20:12:29 +0100 Subject: performance counters: core code Implement the core kernel bits of Performance Counters subsystem. The Linux Performance Counter subsystem provides an abstraction of performance counter hardware capabilities. It provides per task and per CPU counters, and it provides event capabilities on top of those. Performance counters are accessed via special file descriptors. There's one file descriptor per virtual counter used. The special file descriptor is opened via the perf_counter_open() system call: int perf_counter_open(u32 hw_event_type, u32 hw_event_period, u32 record_type, pid_t pid, int cpu); The syscall returns the new fd. The fd can be used via the normal VFS system calls: read() can be used to read the counter, fcntl() can be used to set the blocking mode, etc. Multiple counters can be kept open at a time, and the counters can be poll()ed. See more details in Documentation/perf-counters.txt. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- kernel/fork.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel/fork.c') diff --git a/kernel/fork.c b/kernel/fork.c index 2a372a0e206..441fadff1fa 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -975,6 +975,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, goto fork_out; rt_mutex_init_task(p); + perf_counter_init_task(p); #ifdef CONFIG_PROVE_LOCKING DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); -- cgit v1.2.3 From a63eaf34ae60bdb067a354cc8def2e8f4a01f5f4 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 22 May 2009 14:17:31 +1000 Subject: perf_counter: Dynamically allocate tasks' perf_counter_context struct This replaces the struct perf_counter_context in the task_struct with a pointer to a dynamically allocated perf_counter_context struct. The main reason for doing is this is to allow us to transfer a perf_counter_context from one task to another when we do lazy PMU switching in a later patch. This has a few side-benefits: the task_struct becomes a little smaller, we save some memory because only tasks that have perf_counters attached get a perf_counter_context allocated for them, and we can remove the inclusion of in sched.h, meaning that we don't end up recompiling nearly everything whenever perf_counter.h changes. The perf_counter_context structures are reference-counted and freed when the last reference is dropped. A context can have references from its task and the counters on its task. Counters can outlive the task so it is possible that a context will be freed well after its task has exited. Contexts are allocated on fork if the parent had a context, or otherwise the first time that a per-task counter is created on a task. In the latter case, we set the context pointer in the task struct locklessly using an atomic compare-and-exchange operation in case we raced with some other task in creating a context for the subject task. This also removes the task pointer from the perf_counter struct. The task pointer was not used anywhere and would make it harder to move a context from one task to another. Anything that needed to know which task a counter was attached to was already using counter->ctx->task. The __perf_counter_init_context function moves up in perf_counter.c so that it can be called from find_get_context, and now initializes the refcount, but is otherwise unchanged. We were potentially calling list_del_counter twice: once from __perf_counter_exit_task when the task exits and once from __perf_counter_remove_from_context when the counter's fd gets closed. This adds a check in list_del_counter so it doesn't do anything if the counter has already been removed from the lists. Since perf_counter_task_sched_in doesn't do anything if the task doesn't have a context, and leaves cpuctx->task_ctx = NULL, this adds code to __perf_install_in_context to set cpuctx->task_ctx if necessary, i.e. in the case where the current task adds the first counter to itself and thus creates a context for itself. This also adds similar code to __perf_counter_enable to handle a similar situation which can arise when the counters have been disabled using prctl; that also leaves cpuctx->task_ctx = NULL. [ Impact: refactor counter context management to prepare for new feature ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: <18966.10075.781053.231153@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic/apic.c | 1 + include/linux/init_task.h | 13 --- include/linux/perf_counter.h | 4 +- include/linux/sched.h | 6 +- kernel/exit.c | 3 +- kernel/fork.c | 1 + kernel/perf_counter.c | 218 +++++++++++++++++++++++++++---------------- 7 files changed, 145 insertions(+), 101 deletions(-) (limited to 'kernel/fork.c') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index e9021a90802..b4f64402a82 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -14,6 +14,7 @@ * Mikael Pettersson : PM converted to driver model. */ +#include #include #include #include diff --git a/include/linux/init_task.h b/include/linux/init_task.h index 503afaa0afa..d87247d2641 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -108,18 +108,6 @@ extern struct group_info init_groups; extern struct cred init_cred; -#ifdef CONFIG_PERF_COUNTERS -# define INIT_PERF_COUNTERS(tsk) \ - .perf_counter_ctx.counter_list = \ - LIST_HEAD_INIT(tsk.perf_counter_ctx.counter_list), \ - .perf_counter_ctx.event_list = \ - LIST_HEAD_INIT(tsk.perf_counter_ctx.event_list), \ - .perf_counter_ctx.lock = \ - __SPIN_LOCK_UNLOCKED(tsk.perf_counter_ctx.lock), -#else -# define INIT_PERF_COUNTERS(tsk) -#endif - /* * INIT_TASK is used to set up the first task table, touch at * your own risk!. Base=0, limit=0x1fffff (=2MB) @@ -183,7 +171,6 @@ extern struct cred init_cred; }, \ .dirties = INIT_PROP_LOCAL_SINGLE(dirties), \ INIT_IDS \ - INIT_PERF_COUNTERS(tsk) \ INIT_TRACE_IRQFLAGS \ INIT_LOCKDEP \ INIT_FTRACE_GRAPH \ diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index f612941ef46..07130900546 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -449,7 +449,6 @@ struct perf_counter { struct hw_perf_counter hw; struct perf_counter_context *ctx; - struct task_struct *task; struct file *filp; struct perf_counter *parent; @@ -498,7 +497,6 @@ struct perf_counter { * Used as a container for task counters and CPU counters as well: */ struct perf_counter_context { -#ifdef CONFIG_PERF_COUNTERS /* * Protect the states of the counters in the list, * nr_active, and the list: @@ -516,6 +514,7 @@ struct perf_counter_context { int nr_counters; int nr_active; int is_active; + atomic_t refcount; struct task_struct *task; /* @@ -523,7 +522,6 @@ struct perf_counter_context { */ u64 time; u64 timestamp; -#endif }; /** diff --git a/include/linux/sched.h b/include/linux/sched.h index ff59d123151..9714d450f41 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -71,7 +71,6 @@ struct sched_param { #include #include #include -#include #include #include #include @@ -99,6 +98,7 @@ struct robust_list_head; struct bio; struct bts_tracer; struct fs_struct; +struct perf_counter_context; /* * List of flags we want to share for kernel threads, @@ -1387,7 +1387,9 @@ struct task_struct { struct list_head pi_state_list; struct futex_pi_state *pi_state_cache; #endif - struct perf_counter_context perf_counter_ctx; +#ifdef CONFIG_PERF_COUNTERS + struct perf_counter_context *perf_counter_ctxp; +#endif #ifdef CONFIG_NUMA struct mempolicy *mempolicy; short il_next; diff --git a/kernel/exit.c b/kernel/exit.c index f9dfedd94af..99ad4063ee4 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -48,6 +48,7 @@ #include #include #include +#include #include #include @@ -159,7 +160,7 @@ static void delayed_put_task_struct(struct rcu_head *rhp) struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); #ifdef CONFIG_PERF_COUNTERS - WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list)); + WARN_ON_ONCE(tsk->perf_counter_ctxp); #endif trace_sched_process_free(tsk); put_task_struct(tsk); diff --git a/kernel/fork.c b/kernel/fork.c index d32fef4d38e..e72a09f5355 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -63,6 +63,7 @@ #include #include #include +#include #include #include diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 08584c16049..06ea3eae886 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -97,6 +97,17 @@ void perf_enable(void) hw_perf_enable(); } +static void get_ctx(struct perf_counter_context *ctx) +{ + atomic_inc(&ctx->refcount); +} + +static void put_ctx(struct perf_counter_context *ctx) +{ + if (atomic_dec_and_test(&ctx->refcount)) + kfree(ctx); +} + static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { @@ -118,11 +129,17 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) ctx->nr_counters++; } +/* + * Remove a counter from the lists for its context. + * Must be called with counter->mutex and ctx->mutex held. + */ static void list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { struct perf_counter *sibling, *tmp; + if (list_empty(&counter->list_entry)) + return; ctx->nr_counters--; list_del_init(&counter->list_entry); @@ -216,8 +233,6 @@ static void __perf_counter_remove_from_context(void *info) counter_sched_out(counter, cpuctx, ctx); - counter->task = NULL; - list_del_counter(counter, ctx); if (!ctx->task) { @@ -279,7 +294,6 @@ retry: */ if (!list_empty(&counter->list_entry)) { list_del_counter(counter, ctx); - counter->task = NULL; } spin_unlock_irq(&ctx->lock); } @@ -568,11 +582,17 @@ static void __perf_install_in_context(void *info) * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been * scheduled out before the smp call arrived. + * Or possibly this is the right context but it isn't + * on this cpu because it had no counters. */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } spin_lock_irqsave(&ctx->lock, flags); + ctx->is_active = 1; update_context_time(ctx); /* @@ -653,7 +673,6 @@ perf_install_in_context(struct perf_counter_context *ctx, return; } - counter->task = task; retry: task_oncpu_function_call(task, __perf_install_in_context, counter); @@ -693,10 +712,14 @@ static void __perf_counter_enable(void *info) * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } spin_lock_irqsave(&ctx->lock, flags); + ctx->is_active = 1; update_context_time(ctx); counter->prev_state = counter->state; @@ -852,10 +875,10 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, void perf_counter_task_sched_out(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter_context *ctx = task->perf_counter_ctxp; struct pt_regs *regs; - if (likely(!cpuctx->task_ctx)) + if (likely(!ctx || !cpuctx->task_ctx)) return; update_context_time(ctx); @@ -871,6 +894,8 @@ static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + if (!cpuctx->task_ctx) + return; __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; } @@ -969,8 +994,10 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, void perf_counter_task_sched_in(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter_context *ctx = task->perf_counter_ctxp; + if (likely(!ctx)) + return; __perf_counter_sched_in(ctx, cpuctx, cpu); cpuctx->task_ctx = ctx; } @@ -985,11 +1012,11 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) int perf_counter_task_disable(void) { struct task_struct *curr = current; - struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; unsigned long flags; - if (likely(!ctx->nr_counters)) + if (!ctx || !ctx->nr_counters) return 0; local_irq_save(flags); @@ -1020,12 +1047,12 @@ int perf_counter_task_disable(void) int perf_counter_task_enable(void) { struct task_struct *curr = current; - struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; unsigned long flags; int cpu; - if (likely(!ctx->nr_counters)) + if (!ctx || !ctx->nr_counters) return 0; local_irq_save(flags); @@ -1128,19 +1155,23 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) return; cpuctx = &per_cpu(perf_cpu_context, cpu); - ctx = &curr->perf_counter_ctx; + ctx = curr->perf_counter_ctxp; perf_adjust_freq(&cpuctx->ctx); - perf_adjust_freq(ctx); + if (ctx) + perf_adjust_freq(ctx); perf_counter_cpu_sched_out(cpuctx); - __perf_counter_task_sched_out(ctx); + if (ctx) + __perf_counter_task_sched_out(ctx); rotate_ctx(&cpuctx->ctx); - rotate_ctx(ctx); + if (ctx) + rotate_ctx(ctx); perf_counter_cpu_sched_in(cpuctx, cpu); - perf_counter_task_sched_in(curr, cpu); + if (ctx) + perf_counter_task_sched_in(curr, cpu); } /* @@ -1176,6 +1207,22 @@ static u64 perf_counter_read(struct perf_counter *counter) return atomic64_read(&counter->count); } +/* + * Initialize the perf_counter context in a task_struct: + */ +static void +__perf_counter_init_context(struct perf_counter_context *ctx, + struct task_struct *task) +{ + memset(ctx, 0, sizeof(*ctx)); + spin_lock_init(&ctx->lock); + mutex_init(&ctx->mutex); + INIT_LIST_HEAD(&ctx->counter_list); + INIT_LIST_HEAD(&ctx->event_list); + atomic_set(&ctx->refcount, 1); + ctx->task = task; +} + static void put_context(struct perf_counter_context *ctx) { if (ctx->task) @@ -1186,6 +1233,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) { struct perf_cpu_context *cpuctx; struct perf_counter_context *ctx; + struct perf_counter_context *tctx; struct task_struct *task; /* @@ -1225,15 +1273,36 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) if (!task) return ERR_PTR(-ESRCH); - ctx = &task->perf_counter_ctx; - ctx->task = task; - /* Reuse ptrace permission checks for now. */ if (!ptrace_may_access(task, PTRACE_MODE_READ)) { - put_context(ctx); + put_task_struct(task); return ERR_PTR(-EACCES); } + ctx = task->perf_counter_ctxp; + if (!ctx) { + ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); + if (!ctx) { + put_task_struct(task); + return ERR_PTR(-ENOMEM); + } + __perf_counter_init_context(ctx, task); + /* + * Make sure other cpus see correct values for *ctx + * once task->perf_counter_ctxp is visible to them. + */ + smp_wmb(); + tctx = cmpxchg(&task->perf_counter_ctxp, NULL, ctx); + if (tctx) { + /* + * We raced with some other task; use + * the context they set. + */ + kfree(ctx); + ctx = tctx; + } + } + return ctx; } @@ -1242,6 +1311,7 @@ static void free_counter_rcu(struct rcu_head *head) struct perf_counter *counter; counter = container_of(head, struct perf_counter, rcu_head); + put_ctx(counter->ctx); kfree(counter); } @@ -2247,7 +2317,7 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) perf_counter_comm_ctx(&cpuctx->ctx, comm_event); put_cpu_var(perf_cpu_context); - perf_counter_comm_ctx(¤t->perf_counter_ctx, comm_event); + perf_counter_comm_ctx(current->perf_counter_ctxp, comm_event); } void perf_counter_comm(struct task_struct *task) @@ -2256,7 +2326,9 @@ void perf_counter_comm(struct task_struct *task) if (!atomic_read(&nr_comm_tracking)) return; - + if (!current->perf_counter_ctxp) + return; + comm_event = (struct perf_comm_event){ .task = task, .event = { @@ -2372,7 +2444,7 @@ got_name: perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); put_cpu_var(perf_cpu_context); - perf_counter_mmap_ctx(¤t->perf_counter_ctx, mmap_event); + perf_counter_mmap_ctx(current->perf_counter_ctxp, mmap_event); kfree(buf); } @@ -2384,6 +2456,8 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, if (!atomic_read(&nr_mmap_tracking)) return; + if (!current->perf_counter_ctxp) + return; mmap_event = (struct perf_mmap_event){ .file = file, @@ -2985,6 +3059,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->pmu = NULL; counter->ctx = ctx; + get_ctx(ctx); counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) @@ -3149,21 +3224,6 @@ err_put_context: goto out_fput; } -/* - * Initialize the perf_counter context in a task_struct: - */ -static void -__perf_counter_init_context(struct perf_counter_context *ctx, - struct task_struct *task) -{ - memset(ctx, 0, sizeof(*ctx)); - spin_lock_init(&ctx->lock); - mutex_init(&ctx->mutex); - INIT_LIST_HEAD(&ctx->counter_list); - INIT_LIST_HEAD(&ctx->event_list); - ctx->task = task; -} - /* * inherit a counter from parent task to child task: */ @@ -3195,7 +3255,6 @@ inherit_counter(struct perf_counter *parent_counter, /* * Link it up in the child's context: */ - child_counter->task = child; add_counter_to_ctx(child_counter, child_ctx); child_counter->parent = parent_counter; @@ -3294,40 +3353,15 @@ __perf_counter_exit_task(struct task_struct *child, struct perf_counter *parent_counter; /* - * If we do not self-reap then we have to wait for the - * child task to unschedule (it will happen for sure), - * so that its counter is at its final count. (This - * condition triggers rarely - child tasks usually get - * off their CPU before the parent has a chance to - * get this far into the reaping action) + * Protect against concurrent operations on child_counter + * due its fd getting closed, etc. */ - if (child != current) { - wait_task_inactive(child, 0); - update_counter_times(child_counter); - list_del_counter(child_counter, child_ctx); - } else { - struct perf_cpu_context *cpuctx; - unsigned long flags; - - /* - * Disable and unlink this counter. - * - * Be careful about zapping the list - IRQ/NMI context - * could still be processing it: - */ - local_irq_save(flags); - perf_disable(); - - cpuctx = &__get_cpu_var(perf_cpu_context); + mutex_lock(&child_counter->mutex); - group_sched_out(child_counter, cpuctx, child_ctx); - update_counter_times(child_counter); + update_counter_times(child_counter); + list_del_counter(child_counter, child_ctx); - list_del_counter(child_counter, child_ctx); - - perf_enable(); - local_irq_restore(flags); - } + mutex_unlock(&child_counter->mutex); parent_counter = child_counter->parent; /* @@ -3346,19 +3380,29 @@ __perf_counter_exit_task(struct task_struct *child, * * Note: we may be running in child context, but the PID is not hashed * anymore so new counters will not be added. + * (XXX not sure that is true when we get called from flush_old_exec. + * -- paulus) */ void perf_counter_exit_task(struct task_struct *child) { struct perf_counter *child_counter, *tmp; struct perf_counter_context *child_ctx; + unsigned long flags; WARN_ON_ONCE(child != current); - child_ctx = &child->perf_counter_ctx; + child_ctx = child->perf_counter_ctxp; - if (likely(!child_ctx->nr_counters)) + if (likely(!child_ctx)) return; + local_irq_save(flags); + __perf_counter_task_sched_out(child_ctx); + child->perf_counter_ctxp = NULL; + local_irq_restore(flags); + + mutex_lock(&child_ctx->mutex); + again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) @@ -3371,6 +3415,10 @@ again: */ if (!list_empty(&child_ctx->counter_list)) goto again; + + mutex_unlock(&child_ctx->mutex); + + put_ctx(child_ctx); } /* @@ -3382,19 +3430,25 @@ void perf_counter_init_task(struct task_struct *child) struct perf_counter *counter; struct task_struct *parent = current; - child_ctx = &child->perf_counter_ctx; - parent_ctx = &parent->perf_counter_ctx; - - __perf_counter_init_context(child_ctx, child); + child->perf_counter_ctxp = NULL; /* * This is executed from the parent task context, so inherit - * counters that have been marked for cloning: + * counters that have been marked for cloning. + * First allocate and initialize a context for the child. */ - if (likely(!parent_ctx->nr_counters)) + child_ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); + if (!child_ctx) + return; + + parent_ctx = parent->perf_counter_ctxp; + if (likely(!parent_ctx || !parent_ctx->nr_counters)) return; + __perf_counter_init_context(child_ctx, child); + child->perf_counter_ctxp = child_ctx; + /* * Lock the parent list. No need to lock the child - not PID * hashed yet and not running, so nobody can access it. -- cgit v1.2.3 From e4cbb4e3ac8b09fdb11e39e5a5611bfab0a7cd1a Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 19 May 2009 15:50:30 +0200 Subject: perf_counter: Move child perfcounter init to after scheduler init Initialize a task's perfcounters (inherit from parent, etc.) after the child task's scheduler fields have been initialized already. [ Impact: cleanup ] Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur Cc: Mike Galbraith LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/fork.c') diff --git a/kernel/fork.c b/kernel/fork.c index e72a09f5355..675e01e9072 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -984,7 +984,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, goto fork_out; rt_mutex_init_task(p); - perf_counter_init_task(p); #ifdef CONFIG_PROVE_LOCKING DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); @@ -1096,6 +1095,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); + perf_counter_init_task(p); if ((retval = audit_alloc(p))) goto bad_fork_cleanup_policy; -- cgit v1.2.3 From 6ab423e0eaca827fbd201ca4ae7d4f8573a366b2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 14:45:27 +0200 Subject: perf_counter: Propagate inheritance failures down the fork() path Fail fork() when we fail inheritance for some reason (-ENOMEM most likely). Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525124600.324656474@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 4 ++-- kernel/fork.c | 6 +++++- kernel/perf_counter.c | 20 ++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-) (limited to 'kernel/fork.c') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index b1f2bac09f9..d3e85de9bf1 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -566,7 +566,7 @@ extern void perf_counter_task_sched_in(struct task_struct *task, int cpu); extern void perf_counter_task_sched_out(struct task_struct *task, struct task_struct *next, int cpu); extern void perf_counter_task_tick(struct task_struct *task, int cpu); -extern void perf_counter_init_task(struct task_struct *child); +extern int perf_counter_init_task(struct task_struct *child); extern void perf_counter_exit_task(struct task_struct *child); extern void perf_counter_do_pending(void); extern void perf_counter_print_debug(void); @@ -631,7 +631,7 @@ perf_counter_task_sched_out(struct task_struct *task, struct task_struct *next, int cpu) { } static inline void perf_counter_task_tick(struct task_struct *task, int cpu) { } -static inline void perf_counter_init_task(struct task_struct *child) { } +static inline int perf_counter_init_task(struct task_struct *child) { } static inline void perf_counter_exit_task(struct task_struct *child) { } static inline void perf_counter_do_pending(void) { } static inline void perf_counter_print_debug(void) { } diff --git a/kernel/fork.c b/kernel/fork.c index 675e01e9072..c07c3335cea 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1095,7 +1095,10 @@ static struct task_struct *copy_process(unsigned long clone_flags, /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); - perf_counter_init_task(p); + + retval = perf_counter_init_task(p); + if (retval) + goto bad_fork_cleanup_policy; if ((retval = audit_alloc(p))) goto bad_fork_cleanup_policy; @@ -1295,6 +1298,7 @@ bad_fork_cleanup_semundo: bad_fork_cleanup_audit: audit_free(p); bad_fork_cleanup_policy: + perf_counter_exit_task(p); #ifdef CONFIG_NUMA mpol_put(p->mempolicy); bad_fork_cleanup_cgroup: diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 217dbcce2eb..7a7a144870e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3434,18 +3434,23 @@ again: /* * Initialize the perf_counter context in task_struct */ -void perf_counter_init_task(struct task_struct *child) +int perf_counter_init_task(struct task_struct *child) { struct perf_counter_context *child_ctx, *parent_ctx; struct perf_counter *counter; struct task_struct *parent = current; int inherited_all = 1; + int ret = 0; child->perf_counter_ctxp = NULL; mutex_init(&child->perf_counter_mutex); INIT_LIST_HEAD(&child->perf_counter_list); + parent_ctx = parent->perf_counter_ctxp; + if (likely(!parent_ctx || !parent_ctx->nr_counters)) + return 0; + /* * This is executed from the parent task context, so inherit * counters that have been marked for cloning. @@ -3454,11 +3459,7 @@ void perf_counter_init_task(struct task_struct *child) child_ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); if (!child_ctx) - return; - - parent_ctx = parent->perf_counter_ctxp; - if (likely(!parent_ctx || !parent_ctx->nr_counters)) - return; + return -ENOMEM; __perf_counter_init_context(child_ctx, child); child->perf_counter_ctxp = child_ctx; @@ -3482,8 +3483,9 @@ void perf_counter_init_task(struct task_struct *child) continue; } - if (inherit_group(counter, parent, - parent_ctx, child, child_ctx)) { + ret = inherit_group(counter, parent, parent_ctx, + child, child_ctx); + if (ret) { inherited_all = 0; break; } @@ -3505,6 +3507,8 @@ void perf_counter_init_task(struct task_struct *child) } mutex_unlock(&parent_ctx->mutex); + + return ret; } static void __cpuinit perf_counter_init_cpu(int cpu) -- cgit v1.2.3 From bbbee90829304d156c12b171c0ac7e6e1aba8b90 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 29 May 2009 14:25:58 +0200 Subject: perf_counter: Ammend cleanup in fork() fail When fork() fails we cannot use perf_counter_exit_task() since that assumes to operate on current. Write a new helper that cleans up unused/clean contexts. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 2 ++ kernel/fork.c | 2 +- kernel/perf_counter.c | 43 ++++++++++++++++++++++++++++++++++++++++--- 3 files changed, 43 insertions(+), 4 deletions(-) (limited to 'kernel/fork.c') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 717bf3b59ba..519a41bba24 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -579,6 +579,7 @@ extern void perf_counter_task_sched_out(struct task_struct *task, extern void perf_counter_task_tick(struct task_struct *task, int cpu); extern int perf_counter_init_task(struct task_struct *child); extern void perf_counter_exit_task(struct task_struct *child); +extern void perf_counter_free_task(struct task_struct *task); extern void perf_counter_do_pending(void); extern void perf_counter_print_debug(void); extern void __perf_disable(void); @@ -644,6 +645,7 @@ static inline void perf_counter_task_tick(struct task_struct *task, int cpu) { } static inline int perf_counter_init_task(struct task_struct *child) { return 0; } static inline void perf_counter_exit_task(struct task_struct *child) { } +static inline void perf_counter_free_task(struct task_struct *task) { } static inline void perf_counter_do_pending(void) { } static inline void perf_counter_print_debug(void) { } static inline void perf_disable(void) { } diff --git a/kernel/fork.c b/kernel/fork.c index c07c3335cea..23bf757ed32 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1298,7 +1298,7 @@ bad_fork_cleanup_semundo: bad_fork_cleanup_audit: audit_free(p); bad_fork_cleanup_policy: - perf_counter_exit_task(p); + perf_counter_free_task(p); #ifdef CONFIG_NUMA mpol_put(p->mempolicy); bad_fork_cleanup_cgroup: diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0c000d305e0..79c3f26541d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3538,8 +3538,7 @@ static void sync_child_counter(struct perf_counter *child_counter, } static void -__perf_counter_exit_task(struct task_struct *child, - struct perf_counter *child_counter, +__perf_counter_exit_task(struct perf_counter *child_counter, struct perf_counter_context *child_ctx) { struct perf_counter *parent_counter; @@ -3605,7 +3604,7 @@ void perf_counter_exit_task(struct task_struct *child) again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) - __perf_counter_exit_task(child, child_counter, child_ctx); + __perf_counter_exit_task(child_counter, child_ctx); /* * If the last counter was a group counter, it will have appended all @@ -3620,6 +3619,44 @@ again: put_ctx(child_ctx); } +/* + * free an unexposed, unused context as created by inheritance by + * init_task below, used by fork() in case of fail. + */ +void perf_counter_free_task(struct task_struct *task) +{ + struct perf_counter_context *ctx = task->perf_counter_ctxp; + struct perf_counter *counter, *tmp; + + if (!ctx) + return; + + mutex_lock(&ctx->mutex); +again: + list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry) { + struct perf_counter *parent = counter->parent; + + if (WARN_ON_ONCE(!parent)) + continue; + + mutex_lock(&parent->child_mutex); + list_del_init(&counter->child_list); + mutex_unlock(&parent->child_mutex); + + fput(parent->filp); + + list_del_counter(counter, ctx); + free_counter(counter); + } + + if (!list_empty(&ctx->counter_list)) + goto again; + + mutex_unlock(&ctx->mutex); + + put_ctx(ctx); +} + /* * Initialize the perf_counter context in task_struct */ -- cgit v1.2.3 From 226f62fdd53d5b2c74e242aa11f6ad43d0285d3f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 3 Jun 2009 11:23:56 +0200 Subject: perf_counter: Add a comm hook for pure fork()s I noticed missing COMM events and found that we missed reporting them for pure forks. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'kernel/fork.c') diff --git a/kernel/fork.c b/kernel/fork.c index 23bf757ed32..b7d7a9f0bd7 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1412,6 +1412,12 @@ long do_fork(unsigned long clone_flags, if (clone_flags & CLONE_VFORK) { p->vfork_done = &vfork; init_completion(&vfork); + } else { + /* + * vfork will do an exec which will call + * set_task_comm() + */ + perf_counter_comm(p); } audit_finish_fork(p); -- cgit v1.2.3 From 60313ebed739b331e8e61079da27a11ee3b73a30 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 4 Jun 2009 16:53:44 +0200 Subject: perf_counter: Add fork event Create a fork event so that we can easily clone the comm and dso maps without having to generate all those events. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 10 ++++ kernel/fork.c | 4 +- kernel/perf_counter.c | 131 +++++++++++++++++++++++++++++++++++++------ 3 files changed, 126 insertions(+), 19 deletions(-) (limited to 'kernel/fork.c') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 37d5541d74c..380247bdb91 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -276,6 +276,14 @@ enum perf_event_type { PERF_EVENT_THROTTLE = 5, PERF_EVENT_UNTHROTTLE = 6, + /* + * struct { + * struct perf_event_header header; + * u32 pid, ppid; + * }; + */ + PERF_EVENT_FORK = 7, + /* * When header.misc & PERF_EVENT_MISC_OVERFLOW the event_type field * will be PERF_RECORD_* @@ -618,6 +626,7 @@ extern void perf_counter_munmap(unsigned long addr, unsigned long len, unsigned long pgoff, struct file *file); extern void perf_counter_comm(struct task_struct *tsk); +extern void perf_counter_fork(struct task_struct *tsk); extern void perf_counter_task_migration(struct task_struct *task, int cpu); @@ -673,6 +682,7 @@ perf_counter_munmap(unsigned long addr, unsigned long len, unsigned long pgoff, struct file *file) { } static inline void perf_counter_comm(struct task_struct *tsk) { } +static inline void perf_counter_fork(struct task_struct *tsk) { } static inline void perf_counter_init(void) { } static inline void perf_counter_task_migration(struct task_struct *task, int cpu) { } diff --git a/kernel/fork.c b/kernel/fork.c index b7d7a9f0bd7..f4466ca37ec 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1412,12 +1412,12 @@ long do_fork(unsigned long clone_flags, if (clone_flags & CLONE_VFORK) { p->vfork_done = &vfork; init_completion(&vfork); - } else { + } else if (!(clone_flags & CLONE_VM)) { /* * vfork will do an exec which will call * set_task_comm() */ - perf_counter_comm(p); + perf_counter_fork(p); } audit_finish_fork(p); diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0bb03f15a5b..78c58623a0d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -40,9 +40,9 @@ static int perf_reserved_percpu __read_mostly; static int perf_overcommit __read_mostly = 1; static atomic_t nr_counters __read_mostly; -static atomic_t nr_mmap_tracking __read_mostly; -static atomic_t nr_munmap_tracking __read_mostly; -static atomic_t nr_comm_tracking __read_mostly; +static atomic_t nr_mmap_counters __read_mostly; +static atomic_t nr_munmap_counters __read_mostly; +static atomic_t nr_comm_counters __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ @@ -1447,11 +1447,11 @@ static void free_counter(struct perf_counter *counter) atomic_dec(&nr_counters); if (counter->attr.mmap) - atomic_dec(&nr_mmap_tracking); + atomic_dec(&nr_mmap_counters); if (counter->attr.munmap) - atomic_dec(&nr_munmap_tracking); + atomic_dec(&nr_munmap_counters); if (counter->attr.comm) - atomic_dec(&nr_comm_tracking); + atomic_dec(&nr_comm_counters); if (counter->destroy) counter->destroy(counter); @@ -2475,6 +2475,105 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_end(&handle); } +/* + * fork tracking + */ + +struct perf_fork_event { + struct task_struct *task; + + struct { + struct perf_event_header header; + + u32 pid; + u32 ppid; + } event; +}; + +static void perf_counter_fork_output(struct perf_counter *counter, + struct perf_fork_event *fork_event) +{ + struct perf_output_handle handle; + int size = fork_event->event.header.size; + struct task_struct *task = fork_event->task; + int ret = perf_output_begin(&handle, counter, size, 0, 0); + + if (ret) + return; + + fork_event->event.pid = perf_counter_pid(counter, task); + fork_event->event.ppid = perf_counter_pid(counter, task->real_parent); + + perf_output_put(&handle, fork_event->event); + perf_output_end(&handle); +} + +static int perf_counter_fork_match(struct perf_counter *counter) +{ + if (counter->attr.comm || counter->attr.mmap || counter->attr.munmap) + return 1; + + return 0; +} + +static void perf_counter_fork_ctx(struct perf_counter_context *ctx, + struct perf_fork_event *fork_event) +{ + struct perf_counter *counter; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { + if (perf_counter_fork_match(counter)) + perf_counter_fork_output(counter, fork_event); + } + rcu_read_unlock(); +} + +static void perf_counter_fork_event(struct perf_fork_event *fork_event) +{ + struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_counter_fork_ctx(&cpuctx->ctx, fork_event); + put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_counter_ctxp); + if (ctx) + perf_counter_fork_ctx(ctx, fork_event); + rcu_read_unlock(); +} + +void perf_counter_fork(struct task_struct *task) +{ + struct perf_fork_event fork_event; + + if (!atomic_read(&nr_comm_counters) && + !atomic_read(&nr_mmap_counters) && + !atomic_read(&nr_munmap_counters)) + return; + + fork_event = (struct perf_fork_event){ + .task = task, + .event = { + .header = { + .type = PERF_EVENT_FORK, + .size = sizeof(fork_event.event), + }, + }, + }; + + perf_counter_fork_event(&fork_event); +} + /* * comm tracking */ @@ -2511,11 +2610,9 @@ static void perf_counter_comm_output(struct perf_counter *counter, perf_output_end(&handle); } -static int perf_counter_comm_match(struct perf_counter *counter, - struct perf_comm_event *comm_event) +static int perf_counter_comm_match(struct perf_counter *counter) { - if (counter->attr.comm && - comm_event->event.header.type == PERF_EVENT_COMM) + if (counter->attr.comm) return 1; return 0; @@ -2531,7 +2628,7 @@ static void perf_counter_comm_ctx(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_comm_match(counter, comm_event)) + if (perf_counter_comm_match(counter)) perf_counter_comm_output(counter, comm_event); } rcu_read_unlock(); @@ -2570,7 +2667,7 @@ void perf_counter_comm(struct task_struct *task) { struct perf_comm_event comm_event; - if (!atomic_read(&nr_comm_tracking)) + if (!atomic_read(&nr_comm_counters)) return; comm_event = (struct perf_comm_event){ @@ -2708,7 +2805,7 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, { struct perf_mmap_event mmap_event; - if (!atomic_read(&nr_mmap_tracking)) + if (!atomic_read(&nr_mmap_counters)) return; mmap_event = (struct perf_mmap_event){ @@ -2729,7 +2826,7 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, { struct perf_mmap_event mmap_event; - if (!atomic_read(&nr_munmap_tracking)) + if (!atomic_read(&nr_munmap_counters)) return; mmap_event = (struct perf_mmap_event){ @@ -3427,11 +3524,11 @@ done: atomic_inc(&nr_counters); if (counter->attr.mmap) - atomic_inc(&nr_mmap_tracking); + atomic_inc(&nr_mmap_counters); if (counter->attr.munmap) - atomic_inc(&nr_munmap_tracking); + atomic_inc(&nr_munmap_counters); if (counter->attr.comm) - atomic_inc(&nr_comm_tracking); + atomic_inc(&nr_comm_counters); return counter; } -- cgit v1.2.3