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 --- include/linux/sched.h | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index 55e30d114477..4c530278391b 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -71,6 +71,7 @@ struct sched_param { #include #include #include +#include #include #include #include @@ -1326,6 +1327,7 @@ 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_NUMA struct mempolicy *mempolicy; short il_next; @@ -2285,6 +2287,13 @@ static inline void inc_syscw(struct task_struct *tsk) #define TASK_SIZE_OF(tsk) TASK_SIZE #endif +/* + * Call the function if the target task is executing on a CPU right now: + */ +extern void task_oncpu_function_call(struct task_struct *p, + void (*func) (void *info), void *info); + + #ifdef CONFIG_MM_OWNER extern void mm_update_next_owner(struct mm_struct *mm); extern void mm_init_owner(struct mm_struct *mm, struct task_struct *p); -- cgit v1.2.3 From 6c594c21fcb02c662f11c97be4d7d2b73060a205 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 12:34:15 +0100 Subject: perfcounters: add task migrations counter Impact: add new feature, new sw counter Add a counter that counts the number of cross-CPU migrations a task is suffering. Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 8 +++++--- include/linux/sched.h | 3 ++- kernel/perf_counter.c | 49 ++++++++++++++++++++++++++++++++++++++++++++ kernel/sched.c | 7 +++++-- 4 files changed, 61 insertions(+), 6 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index d2a16563415f..f30486fc55d7 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -42,6 +42,8 @@ enum hw_event_types { PERF_COUNT_BRANCH_INSTRUCTIONS = 4, PERF_COUNT_BRANCH_MISSES = 5, + PERF_HW_EVENTS_MAX = 6, + /* * Special "software" counters provided by the kernel, even if * the hardware does not support performance counters. These @@ -50,11 +52,11 @@ enum hw_event_types { */ PERF_COUNT_CPU_CLOCK = -1, PERF_COUNT_TASK_CLOCK = -2, - /* - * Future software events: - */ PERF_COUNT_PAGE_FAULTS = -3, PERF_COUNT_CONTEXT_SWITCHES = -4, + PERF_COUNT_CPU_MIGRATIONS = -5, + + PERF_SW_EVENTS_MIN = -6, }; /* diff --git a/include/linux/sched.h b/include/linux/sched.h index 4c530278391b..2e15be8fc792 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1014,6 +1014,8 @@ struct sched_entity { u64 last_wakeup; u64 avg_overlap; + u64 nr_migrations; + #ifdef CONFIG_SCHEDSTATS u64 wait_start; u64 wait_max; @@ -1029,7 +1031,6 @@ struct sched_entity { u64 exec_max; u64 slice_max; - u64 nr_migrations; u64 nr_migrations_cold; u64 nr_failed_migrations_affine; u64 nr_failed_migrations_running; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 09287091c526..fb11e351e44e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -936,6 +936,52 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = { .hw_perf_counter_read = context_switches_perf_counter_read, }; +static inline u64 get_cpu_migrations(void) +{ + return current->se.nr_migrations; +} + +static void cpu_migrations_perf_counter_update(struct perf_counter *counter) +{ + u64 prev, now; + s64 delta; + + prev = atomic64_read(&counter->hw.prev_count); + now = get_cpu_migrations(); + + atomic64_set(&counter->hw.prev_count, now); + + delta = now - prev; + if (WARN_ON_ONCE(delta < 0)) + delta = 0; + + atomic64_add(delta, &counter->count); +} + +static void cpu_migrations_perf_counter_read(struct perf_counter *counter) +{ + cpu_migrations_perf_counter_update(counter); +} + +static void cpu_migrations_perf_counter_enable(struct perf_counter *counter) +{ + /* + * se.nr_migrations is a per-task value already, + * so we dont have to clear it on switch-in. + */ +} + +static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) +{ + cpu_migrations_perf_counter_update(counter); +} + +static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { + .hw_perf_counter_enable = cpu_migrations_perf_counter_enable, + .hw_perf_counter_disable = cpu_migrations_perf_counter_disable, + .hw_perf_counter_read = cpu_migrations_perf_counter_read, +}; + static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { @@ -951,6 +997,9 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_CONTEXT_SWITCHES: hw_ops = &perf_ops_context_switches; break; + case PERF_COUNT_CPU_MIGRATIONS: + hw_ops = &perf_ops_cpu_migrations; + break; default: break; } diff --git a/kernel/sched.c b/kernel/sched.c index 5c3f4106314e..382cfdb5e38d 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -1852,12 +1852,14 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) p->se.sleep_start -= clock_offset; if (p->se.block_start) p->se.block_start -= clock_offset; +#endif if (old_cpu != new_cpu) { - schedstat_inc(p, se.nr_migrations); + p->se.nr_migrations++; +#ifdef CONFIG_SCHEDSTATS if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); - } #endif + } p->se.vruntime -= old_cfsrq->min_vruntime - new_cfsrq->min_vruntime; @@ -2375,6 +2377,7 @@ static void __sched_fork(struct task_struct *p) p->se.exec_start = 0; p->se.sum_exec_runtime = 0; p->se.prev_sum_exec_runtime = 0; + p->se.nr_migrations = 0; p->se.last_wakeup = 0; p->se.avg_overlap = 0; -- cgit v1.2.3 From 23a185ca8abbeef64b6ffc33059b1d630e43ec10 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 9 Feb 2009 22:42:47 +1100 Subject: perf_counters: make software counters work as per-cpu counters Impact: kernel crash fix Yanmin Zhang reported that using a PERF_COUNT_TASK_CLOCK software counter as a per-cpu counter would reliably crash the system, because it calls __task_delta_exec with a null pointer. The page fault, context switch and cpu migration counters also won't function correctly as per-cpu counters since they reference the current task. This fixes the problem by redirecting the task_clock counter to the cpu_clock counter when used as a per-cpu counter, and by implementing per-cpu page fault, context switch and cpu migration counters. Along the way, this: - Initializes counter->ctx earlier, in perf_counter_alloc, so that sw_perf_counter_init can use it - Adds code to kernel/sched.c to count task migrations into each cpu, in rq->nr_migrations_in - Exports the per-cpu context switch and task migration counts via new functions added to kernel/sched.c - Makes sure that if sw_perf_counter_init fails, we don't try to initialize the counter as a hardware counter. Since the user has passed a negative, non-raw event type, they clearly don't intend for it to be interpreted as a hardware event. Reported-by: "Zhang Yanmin" Signed-off-by: Paul Mackerras Signed-off-by: Ingo Molnar --- include/linux/sched.h | 2 ++ kernel/perf_counter.c | 78 +++++++++++++++++++++++++++++---------------------- kernel/sched.c | 17 +++++++++++ 3 files changed, 64 insertions(+), 33 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index b85b10abf770..1e5f70062a9c 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -137,6 +137,8 @@ extern unsigned long nr_running(void); extern unsigned long nr_uninterruptible(void); extern unsigned long nr_active(void); extern unsigned long nr_iowait(void); +extern u64 cpu_nr_switches(int cpu); +extern u64 cpu_nr_migrations(int cpu); struct seq_file; struct cfs_rq; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f27a7e9f3c41..544193cbc478 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -20,6 +20,8 @@ #include #include #include +#include +#include /* * Each CPU has a list of per CPU counters: @@ -502,7 +504,6 @@ perf_install_in_context(struct perf_counter_context *ctx, { struct task_struct *task = ctx->task; - counter->ctx = ctx; if (!task) { /* * Per cpu counters are installed via an smp call and @@ -1417,11 +1418,19 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -static u64 get_page_faults(void) +#ifdef CONFIG_VM_EVENT_COUNTERS +#define cpu_page_faults() __get_cpu_var(vm_event_states).event[PGFAULT] +#else +#define cpu_page_faults() 0 +#endif + +static u64 get_page_faults(struct perf_counter *counter) { - struct task_struct *curr = current; + struct task_struct *curr = counter->ctx->task; - return curr->maj_flt + curr->min_flt; + if (curr) + return curr->maj_flt + curr->min_flt; + return cpu_page_faults(); } static void page_faults_perf_counter_update(struct perf_counter *counter) @@ -1430,7 +1439,7 @@ static void page_faults_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_page_faults(); + now = get_page_faults(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1446,11 +1455,7 @@ static void page_faults_perf_counter_read(struct perf_counter *counter) static int page_faults_perf_counter_enable(struct perf_counter *counter) { - /* - * page-faults is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_page_faults(counter)); return 0; } @@ -1465,11 +1470,13 @@ static const struct hw_perf_counter_ops perf_ops_page_faults = { .read = page_faults_perf_counter_read, }; -static u64 get_context_switches(void) +static u64 get_context_switches(struct perf_counter *counter) { - struct task_struct *curr = current; + struct task_struct *curr = counter->ctx->task; - return curr->nvcsw + curr->nivcsw; + if (curr) + return curr->nvcsw + curr->nivcsw; + return cpu_nr_switches(smp_processor_id()); } static void context_switches_perf_counter_update(struct perf_counter *counter) @@ -1478,7 +1485,7 @@ static void context_switches_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_context_switches(); + now = get_context_switches(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1494,11 +1501,7 @@ static void context_switches_perf_counter_read(struct perf_counter *counter) static int context_switches_perf_counter_enable(struct perf_counter *counter) { - /* - * ->nvcsw + curr->nivcsw is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_context_switches(counter)); return 0; } @@ -1513,9 +1516,13 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = { .read = context_switches_perf_counter_read, }; -static inline u64 get_cpu_migrations(void) +static inline u64 get_cpu_migrations(struct perf_counter *counter) { - return current->se.nr_migrations; + struct task_struct *curr = counter->ctx->task; + + if (curr) + return curr->se.nr_migrations; + return cpu_nr_migrations(smp_processor_id()); } static void cpu_migrations_perf_counter_update(struct perf_counter *counter) @@ -1524,7 +1531,7 @@ static void cpu_migrations_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_cpu_migrations(); + now = get_cpu_migrations(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1540,11 +1547,7 @@ static void cpu_migrations_perf_counter_read(struct perf_counter *counter) static int cpu_migrations_perf_counter_enable(struct perf_counter *counter) { - /* - * se.nr_migrations is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_cpu_migrations(counter)); return 0; } @@ -1569,7 +1572,14 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_TASK_CLOCK: - hw_ops = &perf_ops_task_clock; + /* + * If the user instantiates this as a per-cpu counter, + * use the cpu_clock counter instead. + */ + if (counter->ctx->task) + hw_ops = &perf_ops_task_clock; + else + hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_PAGE_FAULTS: hw_ops = &perf_ops_page_faults; @@ -1592,6 +1602,7 @@ sw_perf_counter_init(struct perf_counter *counter) static struct perf_counter * perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu, + struct perf_counter_context *ctx, struct perf_counter *group_leader, gfp_t gfpflags) { @@ -1623,6 +1634,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->wakeup_pending = 0; counter->group_leader = group_leader; counter->hw_ops = NULL; + counter->ctx = ctx; counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) @@ -1631,7 +1643,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hw_ops = NULL; if (!hw_event->raw && hw_event->type < 0) hw_ops = sw_perf_counter_init(counter); - if (!hw_ops) + else hw_ops = hw_perf_counter_init(counter); if (!hw_ops) { @@ -1707,7 +1719,8 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, } ret = -EINVAL; - counter = perf_counter_alloc(&hw_event, cpu, group_leader, GFP_KERNEL); + counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader, + GFP_KERNEL); if (!counter) goto err_put_context; @@ -1777,15 +1790,14 @@ inherit_counter(struct perf_counter *parent_counter, parent_counter = parent_counter->parent; child_counter = perf_counter_alloc(&parent_counter->hw_event, - parent_counter->cpu, group_leader, - GFP_KERNEL); + parent_counter->cpu, child_ctx, + group_leader, GFP_KERNEL); if (!child_counter) return NULL; /* * Link it up in the child's context: */ - child_counter->ctx = child_ctx; child_counter->task = child; list_add_counter(child_counter, child_ctx); child_ctx->nr_counters++; diff --git a/kernel/sched.c b/kernel/sched.c index 8db1a4cf2082..173768f142ad 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -558,6 +558,7 @@ struct rq { struct load_weight load; unsigned long nr_load_updates; u64 nr_switches; + u64 nr_migrations_in; struct cfs_rq cfs; struct rt_rq rt; @@ -1908,6 +1909,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) #endif if (old_cpu != new_cpu) { p->se.nr_migrations++; + new_rq->nr_migrations_in++; #ifdef CONFIG_SCHEDSTATS if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); @@ -2810,6 +2812,21 @@ unsigned long nr_active(void) return running + uninterruptible; } +/* + * Externally visible per-cpu scheduler statistics: + * cpu_nr_switches(cpu) - number of context switches on that cpu + * cpu_nr_migrations(cpu) - number of migrations into that cpu + */ +u64 cpu_nr_switches(int cpu) +{ + return cpu_rq(cpu)->nr_switches; +} + +u64 cpu_nr_migrations(int cpu) +{ + return cpu_rq(cpu)->nr_migrations_in; +} + /* * Update rq->cpu_load[] statistics. This function is usually called every * scheduler tick (TICK_NSEC). -- cgit v1.2.3 From 4a0deca657f3dbb8a707b5dc8f173beec01e7ed2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:12 +0100 Subject: perf_counter: generic context switch event Impact: cleanup Use the generic software events for context switches. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.283522645@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/sched.h | 1 - kernel/perf_counter.c | 60 ++++----------------------------------------------- kernel/sched.c | 6 ------ 3 files changed, 4 insertions(+), 63 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index 75b2fc5306d8..7ed41f7c5ace 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -138,7 +138,6 @@ extern unsigned long nr_running(void); extern unsigned long nr_uninterruptible(void); extern unsigned long nr_active(void); extern unsigned long nr_iowait(void); -extern u64 cpu_nr_switches(int cpu); extern u64 cpu_nr_migrations(int cpu); extern unsigned long get_parent_ip(unsigned long addr); diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4c4e9eb37ab0..99d5930f0a52 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -710,10 +710,13 @@ 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 pt_regs *regs; if (likely(!cpuctx->task_ctx)) return; + regs = task_pt_regs(task); + perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs); __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; @@ -1667,58 +1670,6 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -/* - * Software counter: context switches - */ - -static u64 get_context_switches(struct perf_counter *counter) -{ - struct task_struct *curr = counter->ctx->task; - - if (curr) - return curr->nvcsw + curr->nivcsw; - return cpu_nr_switches(smp_processor_id()); -} - -static void context_switches_perf_counter_update(struct perf_counter *counter) -{ - u64 prev, now; - s64 delta; - - prev = atomic64_read(&counter->hw.prev_count); - now = get_context_switches(counter); - - atomic64_set(&counter->hw.prev_count, now); - - delta = now - prev; - - atomic64_add(delta, &counter->count); -} - -static void context_switches_perf_counter_read(struct perf_counter *counter) -{ - context_switches_perf_counter_update(counter); -} - -static int context_switches_perf_counter_enable(struct perf_counter *counter) -{ - if (counter->prev_state <= PERF_COUNTER_STATE_OFF) - atomic64_set(&counter->hw.prev_count, - get_context_switches(counter)); - return 0; -} - -static void context_switches_perf_counter_disable(struct perf_counter *counter) -{ - context_switches_perf_counter_update(counter); -} - -static const struct hw_perf_counter_ops perf_ops_context_switches = { - .enable = context_switches_perf_counter_enable, - .disable = context_switches_perf_counter_disable, - .read = context_switches_perf_counter_read, -}; - /* * Software counter: cpu migrations */ @@ -1808,11 +1759,8 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_PAGE_FAULTS: case PERF_COUNT_PAGE_FAULTS_MIN: case PERF_COUNT_PAGE_FAULTS_MAJ: - hw_ops = &perf_ops_generic; - break; case PERF_COUNT_CONTEXT_SWITCHES: - if (!counter->hw_event.exclude_kernel) - hw_ops = &perf_ops_context_switches; + hw_ops = &perf_ops_generic; break; case PERF_COUNT_CPU_MIGRATIONS: if (!counter->hw_event.exclude_kernel) diff --git a/kernel/sched.c b/kernel/sched.c index 39e708602169..f76e3c0188a2 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2900,14 +2900,8 @@ unsigned long nr_active(void) /* * Externally visible per-cpu scheduler statistics: - * cpu_nr_switches(cpu) - number of context switches on that cpu * cpu_nr_migrations(cpu) - number of migrations into that cpu */ -u64 cpu_nr_switches(int cpu) -{ - return cpu_rq(cpu)->nr_switches; -} - u64 cpu_nr_migrations(int cpu) { return cpu_rq(cpu)->nr_migrations_in; -- cgit v1.2.3 From a26b89f05d194413c7238e0bea071054f6b5d3c8 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Fri, 3 Apr 2009 16:43:34 +0200 Subject: sched, hw-branch-tracer: add wait_task_context_switch() function to sched.h Add a function to wait until some other task has been switched out at least once. This differs from wait_task_inactive() subtly, in that the latter will wait until the task has left the CPU. Signed-off-by: Markus Metzger Cc: markus.t.metzger@gmail.com Cc: roland@redhat.com Cc: eranian@googlemail.com Cc: oleg@redhat.com Cc: juan.villacis@intel.com Cc: ak@linux.jf.intel.com LKML-Reference: <20090403144549.794157000@intel.com> Signed-off-by: Ingo Molnar --- include/linux/sched.h | 2 ++ kernel/sched.c | 43 +++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 45 insertions(+) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index b94f3541f67b..a5b9a83065fa 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1993,8 +1993,10 @@ extern void set_task_comm(struct task_struct *tsk, char *from); extern char *get_task_comm(char *to, struct task_struct *tsk); #ifdef CONFIG_SMP +extern void wait_task_context_switch(struct task_struct *p); extern unsigned long wait_task_inactive(struct task_struct *, long match_state); #else +static inline void wait_task_context_switch(struct task_struct *p) {} static inline unsigned long wait_task_inactive(struct task_struct *p, long match_state) { diff --git a/kernel/sched.c b/kernel/sched.c index 6cc1fd5d5072..f91bc8141dc3 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2002,6 +2002,49 @@ migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) return 1; } +/* + * wait_task_context_switch - wait for a thread to complete at least one + * context switch. + * + * @p must not be current. + */ +void wait_task_context_switch(struct task_struct *p) +{ + unsigned long nvcsw, nivcsw, flags; + int running; + struct rq *rq; + + nvcsw = p->nvcsw; + nivcsw = p->nivcsw; + for (;;) { + /* + * The runqueue is assigned before the actual context + * switch. We need to take the runqueue lock. + * + * We could check initially without the lock but it is + * very likely that we need to take the lock in every + * iteration. + */ + rq = task_rq_lock(p, &flags); + running = task_running(rq, p); + task_rq_unlock(rq, &flags); + + if (likely(!running)) + break; + /* + * The switch count is incremented before the actual + * context switch. We thus wait for two switches to be + * sure at least one completed. + */ + if ((p->nvcsw - nvcsw) > 1) + break; + if ((p->nivcsw - nivcsw) > 1) + break; + + cpu_relax(); + } +} + /* * wait_task_inactive - wait for a thread to unschedule. * -- cgit v1.2.3 From e2b371f00a6f529f6362654239bdec8dcd510760 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Fri, 3 Apr 2009 16:43:35 +0200 Subject: mm, x86, ptrace, bts: defer branch trace stopping When a ptraced task is unlinked, we need to stop branch tracing for that task. Since the unlink is called with interrupts disabled, and we need interrupts enabled to stop branch tracing, we defer the work. Collect all branch tracing related stuff in a branch tracing context. Reviewed-by: Oleg Nesterov Signed-off-by: Markus Metzger Cc: Andrew Morton Cc: Peter Zijlstra Cc: roland@redhat.com Cc: eranian@googlemail.com Cc: juan.villacis@intel.com Cc: ak@linux.jf.intel.com LKML-Reference: <20090403144550.712401000@intel.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/processor.h | 4 - arch/x86/kernel/ptrace.c | 254 ++++++++++++++++++++++++++------------- include/linux/mm.h | 3 +- include/linux/sched.h | 9 +- mm/mlock.c | 13 +- 5 files changed, 179 insertions(+), 104 deletions(-) (limited to 'include/linux/sched.h') diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h index 34c52370f2fe..2483807e06e7 100644 --- a/arch/x86/include/asm/processor.h +++ b/arch/x86/include/asm/processor.h @@ -458,10 +458,6 @@ struct thread_struct { /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */ struct ds_context *ds_ctx; #endif /* CONFIG_X86_DS */ -#ifdef CONFIG_X86_PTRACE_BTS -/* the signal to send on a bts buffer overflow */ - unsigned int bts_ovfl_signal; -#endif /* CONFIG_X86_PTRACE_BTS */ }; static inline unsigned long native_get_debugreg(int regno) diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index fe9345c967de..7c21d1e8cae7 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -22,6 +22,7 @@ #include #include #include +#include #include #include @@ -577,17 +578,119 @@ static int ioperm_get(struct task_struct *target, } #ifdef CONFIG_X86_PTRACE_BTS +/* + * A branch trace store context. + * + * Contexts may only be installed by ptrace_bts_config() and only for + * ptraced tasks. + * + * Contexts are destroyed when the tracee is detached from the tracer. + * The actual destruction work requires interrupts enabled, so the + * work is deferred and will be scheduled during __ptrace_unlink(). + * + * Contexts hold an additional task_struct reference on the traced + * task, as well as a reference on the tracer's mm. + * + * Ptrace already holds a task_struct for the duration of ptrace operations, + * but since destruction is deferred, it may be executed after both + * tracer and tracee exited. + */ +struct bts_context { + /* The branch trace handle. */ + struct bts_tracer *tracer; + + /* The buffer used to store the branch trace and its size. */ + void *buffer; + unsigned int size; + + /* The mm that paid for the above buffer. */ + struct mm_struct *mm; + + /* The task this context belongs to. */ + struct task_struct *task; + + /* The signal to send on a bts buffer overflow. */ + unsigned int bts_ovfl_signal; + + /* The work struct to destroy a context. */ + struct work_struct work; +}; + +static inline void alloc_bts_buffer(struct bts_context *context, + unsigned int size) +{ + void *buffer; + + buffer = alloc_locked_buffer(size); + if (buffer) { + context->buffer = buffer; + context->size = size; + context->mm = get_task_mm(current); + } +} + +static inline void free_bts_buffer(struct bts_context *context) +{ + if (!context->buffer) + return; + + kfree(context->buffer); + context->buffer = NULL; + + refund_locked_buffer_memory(context->mm, context->size); + context->size = 0; + + mmput(context->mm); + context->mm = NULL; +} + +static void free_bts_context_work(struct work_struct *w) +{ + struct bts_context *context; + + context = container_of(w, struct bts_context, work); + + ds_release_bts(context->tracer); + put_task_struct(context->task); + free_bts_buffer(context); + kfree(context); +} + +static inline void free_bts_context(struct bts_context *context) +{ + INIT_WORK(&context->work, free_bts_context_work); + schedule_work(&context->work); +} + +static inline struct bts_context *alloc_bts_context(struct task_struct *task) +{ + struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL); + if (context) { + context->task = task; + task->bts = context; + + get_task_struct(task); + } + + return context; +} + static int ptrace_bts_read_record(struct task_struct *child, size_t index, struct bts_struct __user *out) { + struct bts_context *context; const struct bts_trace *trace; struct bts_struct bts; const unsigned char *at; int error; - trace = ds_read_bts(child->bts); + context = child->bts; + if (!context) + return -ESRCH; + + trace = ds_read_bts(context->tracer); if (!trace) - return -EPERM; + return -ESRCH; at = trace->ds.top - ((index + 1) * trace->ds.size); if ((void *)at < trace->ds.begin) @@ -596,7 +699,7 @@ static int ptrace_bts_read_record(struct task_struct *child, size_t index, if (!trace->read) return -EOPNOTSUPP; - error = trace->read(child->bts, at, &bts); + error = trace->read(context->tracer, at, &bts); if (error < 0) return error; @@ -610,13 +713,18 @@ static int ptrace_bts_drain(struct task_struct *child, long size, struct bts_struct __user *out) { + struct bts_context *context; const struct bts_trace *trace; const unsigned char *at; int error, drained = 0; - trace = ds_read_bts(child->bts); + context = child->bts; + if (!context) + return -ESRCH; + + trace = ds_read_bts(context->tracer); if (!trace) - return -EPERM; + return -ESRCH; if (!trace->read) return -EOPNOTSUPP; @@ -627,9 +735,8 @@ static int ptrace_bts_drain(struct task_struct *child, for (at = trace->ds.begin; (void *)at < trace->ds.top; out++, drained++, at += trace->ds.size) { struct bts_struct bts; - int error; - error = trace->read(child->bts, at, &bts); + error = trace->read(context->tracer, at, &bts); if (error < 0) return error; @@ -639,35 +746,18 @@ static int ptrace_bts_drain(struct task_struct *child, memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); - error = ds_reset_bts(child->bts); + error = ds_reset_bts(context->tracer); if (error < 0) return error; return drained; } -static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size) -{ - child->bts_buffer = alloc_locked_buffer(size); - if (!child->bts_buffer) - return -ENOMEM; - - child->bts_size = size; - - return 0; -} - -static void ptrace_bts_free_buffer(struct task_struct *child) -{ - free_locked_buffer(child->bts_buffer, child->bts_size); - child->bts_buffer = NULL; - child->bts_size = 0; -} - static int ptrace_bts_config(struct task_struct *child, long cfg_size, const struct ptrace_bts_config __user *ucfg) { + struct bts_context *context; struct ptrace_bts_config cfg; unsigned int flags = 0; @@ -677,28 +767,31 @@ static int ptrace_bts_config(struct task_struct *child, if (copy_from_user(&cfg, ucfg, sizeof(cfg))) return -EFAULT; - if (child->bts) { - ds_release_bts(child->bts); - child->bts = NULL; - } + context = child->bts; + if (!context) + context = alloc_bts_context(child); + if (!context) + return -ENOMEM; if (cfg.flags & PTRACE_BTS_O_SIGNAL) { if (!cfg.signal) return -EINVAL; - child->thread.bts_ovfl_signal = cfg.signal; return -EOPNOTSUPP; + context->bts_ovfl_signal = cfg.signal; } - if ((cfg.flags & PTRACE_BTS_O_ALLOC) && - (cfg.size != child->bts_size)) { - int error; + ds_release_bts(context->tracer); + context->tracer = NULL; - ptrace_bts_free_buffer(child); + if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) { + free_bts_buffer(context); + if (!cfg.size) + return 0; - error = ptrace_bts_allocate_buffer(child, cfg.size); - if (error < 0) - return error; + alloc_bts_buffer(context, cfg.size); + if (!context->buffer) + return -ENOMEM; } if (cfg.flags & PTRACE_BTS_O_TRACE) @@ -707,15 +800,13 @@ static int ptrace_bts_config(struct task_struct *child, if (cfg.flags & PTRACE_BTS_O_SCHED) flags |= BTS_TIMESTAMPS; - child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size, - /* ovfl = */ NULL, /* th = */ (size_t)-1, - flags); - if (IS_ERR(child->bts)) { - int error = PTR_ERR(child->bts); - - ptrace_bts_free_buffer(child); - child->bts = NULL; + context->tracer = ds_request_bts(child, context->buffer, context->size, + NULL, (size_t)-1, flags); + if (unlikely(IS_ERR(context->tracer))) { + int error = PTR_ERR(context->tracer); + free_bts_buffer(context); + context->tracer = NULL; return error; } @@ -726,20 +817,25 @@ static int ptrace_bts_status(struct task_struct *child, long cfg_size, struct ptrace_bts_config __user *ucfg) { + struct bts_context *context; const struct bts_trace *trace; struct ptrace_bts_config cfg; + context = child->bts; + if (!context) + return -ESRCH; + if (cfg_size < sizeof(cfg)) return -EIO; - trace = ds_read_bts(child->bts); + trace = ds_read_bts(context->tracer); if (!trace) - return -EPERM; + return -ESRCH; memset(&cfg, 0, sizeof(cfg)); - cfg.size = trace->ds.end - trace->ds.begin; - cfg.signal = child->thread.bts_ovfl_signal; - cfg.bts_size = sizeof(struct bts_struct); + cfg.size = trace->ds.end - trace->ds.begin; + cfg.signal = context->bts_ovfl_signal; + cfg.bts_size = sizeof(struct bts_struct); if (cfg.signal) cfg.flags |= PTRACE_BTS_O_SIGNAL; @@ -758,67 +854,56 @@ static int ptrace_bts_status(struct task_struct *child, static int ptrace_bts_clear(struct task_struct *child) { + struct bts_context *context; const struct bts_trace *trace; - trace = ds_read_bts(child->bts); + context = child->bts; + if (!context) + return -ESRCH; + + trace = ds_read_bts(context->tracer); if (!trace) - return -EPERM; + return -ESRCH; memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); - return ds_reset_bts(child->bts); + return ds_reset_bts(context->tracer); } static int ptrace_bts_size(struct task_struct *child) { + struct bts_context *context; const struct bts_trace *trace; - trace = ds_read_bts(child->bts); + context = child->bts; + if (!context) + return -ESRCH; + + trace = ds_read_bts(context->tracer); if (!trace) - return -EPERM; + return -ESRCH; return (trace->ds.top - trace->ds.begin) / trace->ds.size; } -static void ptrace_bts_fork(struct task_struct *tsk) +static inline void ptrace_bts_fork(struct task_struct *tsk) { tsk->bts = NULL; - tsk->bts_buffer = NULL; - tsk->bts_size = 0; - tsk->thread.bts_ovfl_signal = 0; } -static void ptrace_bts_untrace(struct task_struct *child) +/* + * Called from __ptrace_unlink() after the child has been moved back + * to its original parent. + */ +static inline void ptrace_bts_untrace(struct task_struct *child) { if (unlikely(child->bts)) { - ds_release_bts(child->bts); + free_bts_context(child->bts); child->bts = NULL; - - /* We cannot update total_vm and locked_vm since - child's mm is already gone. But we can reclaim the - memory. */ - kfree(child->bts_buffer); - child->bts_buffer = NULL; - child->bts_size = 0; } } - -static void ptrace_bts_detach(struct task_struct *child) -{ - /* - * Ptrace_detach() races with ptrace_untrace() in case - * the child dies and is reaped by another thread. - * - * We only do the memory accounting at this point and - * leave the buffer deallocation and the bts tracer - * release to ptrace_bts_untrace() which will be called - * later on with tasklist_lock held. - */ - release_locked_buffer(child->bts_buffer, child->bts_size); -} #else static inline void ptrace_bts_fork(struct task_struct *tsk) {} -static inline void ptrace_bts_detach(struct task_struct *child) {} static inline void ptrace_bts_untrace(struct task_struct *child) {} #endif /* CONFIG_X86_PTRACE_BTS */ @@ -843,7 +928,6 @@ void ptrace_disable(struct task_struct *child) #ifdef TIF_SYSCALL_EMU clear_tsk_thread_flag(child, TIF_SYSCALL_EMU); #endif - ptrace_bts_detach(child); } #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION diff --git a/include/linux/mm.h b/include/linux/mm.h index bff1f0d475c7..64d8ed2538ae 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -13,6 +13,7 @@ #include #include #include +#include struct mempolicy; struct anon_vma; @@ -1321,6 +1322,6 @@ void vmemmap_populate_print_last(void); extern void *alloc_locked_buffer(size_t size); extern void free_locked_buffer(void *buffer, size_t size); -extern void release_locked_buffer(void *buffer, size_t size); +extern void refund_locked_buffer_memory(struct mm_struct *mm, size_t size); #endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */ diff --git a/include/linux/sched.h b/include/linux/sched.h index a5b9a83065fa..52b8cd049c2e 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -96,8 +96,8 @@ struct exec_domain; struct futex_pi_state; struct robust_list_head; struct bio; -struct bts_tracer; struct fs_struct; +struct bts_context; /* * List of flags we want to share for kernel threads, @@ -1210,12 +1210,7 @@ struct task_struct { * This is the tracer handle for the ptrace BTS extension. * This field actually belongs to the ptracer task. */ - struct bts_tracer *bts; - /* - * The buffer to hold the BTS data. - */ - void *bts_buffer; - size_t bts_size; + struct bts_context *bts; #endif /* CONFIG_X86_PTRACE_BTS */ /* PID/PID hash table linkage. */ diff --git a/mm/mlock.c b/mm/mlock.c index cbe9e0581b75..749383b442c7 100644 --- a/mm/mlock.c +++ b/mm/mlock.c @@ -660,21 +660,20 @@ void *alloc_locked_buffer(size_t size) return buffer; } -void release_locked_buffer(void *buffer, size_t size) +void refund_locked_buffer_memory(struct mm_struct *mm, size_t size) { unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT; - down_write(¤t->mm->mmap_sem); + down_write(&mm->mmap_sem); - current->mm->total_vm -= pgsz; - current->mm->locked_vm -= pgsz; + mm->total_vm -= pgsz; + mm->locked_vm -= pgsz; - up_write(¤t->mm->mmap_sem); + up_write(&mm->mmap_sem); } void free_locked_buffer(void *buffer, size_t size) { - release_locked_buffer(buffer, size); - + refund_locked_buffer_memory(current->mm, size); kfree(buffer); } -- cgit v1.2.3 From 0f4814065ff8c24ca8bfd75c9b73502be152c287 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Fri, 3 Apr 2009 16:43:48 +0200 Subject: x86, ptrace: add bts context unconditionally Add the ptrace bts context field to task_struct unconditionally. Initialize the field directly in copy_process(). Remove all the unneeded functionality used to initialize that field. Signed-off-by: Markus Metzger Cc: roland@redhat.com Cc: eranian@googlemail.com Cc: oleg@redhat.com Cc: juan.villacis@intel.com Cc: ak@linux.jf.intel.com LKML-Reference: <20090403144603.292754000@intel.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/ptrace.h | 9 ++++----- arch/x86/kernel/ptrace.c | 20 +------------------- include/linux/ptrace.h | 10 ---------- include/linux/sched.h | 2 -- kernel/fork.c | 4 ++-- kernel/ptrace.c | 10 ---------- 6 files changed, 7 insertions(+), 48 deletions(-) (limited to 'include/linux/sched.h') diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h index e304b66abeea..5cdd19f20b5b 100644 --- a/arch/x86/include/asm/ptrace.h +++ b/arch/x86/include/asm/ptrace.h @@ -235,12 +235,11 @@ extern int do_get_thread_area(struct task_struct *p, int idx, extern int do_set_thread_area(struct task_struct *p, int idx, struct user_desc __user *info, int can_allocate); -extern void x86_ptrace_untrace(struct task_struct *); -extern void x86_ptrace_fork(struct task_struct *child, - unsigned long clone_flags); +#ifdef CONFIG_X86_PTRACE_BTS +extern void ptrace_bts_untrace(struct task_struct *tsk); -#define arch_ptrace_untrace(tsk) x86_ptrace_untrace(tsk) -#define arch_ptrace_fork(child, flags) x86_ptrace_fork(child, flags) +#define arch_ptrace_untrace(tsk) ptrace_bts_untrace(tsk) +#endif /* CONFIG_X86_PTRACE_BTS */ #endif /* __KERNEL__ */ diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index adbb24322d8f..b32a8ee53381 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -887,37 +887,19 @@ static int ptrace_bts_size(struct task_struct *child) return (trace->ds.top - trace->ds.begin) / trace->ds.size; } -static inline void ptrace_bts_fork(struct task_struct *tsk) -{ - tsk->bts = NULL; -} - /* * Called from __ptrace_unlink() after the child has been moved back * to its original parent. */ -static inline void ptrace_bts_untrace(struct task_struct *child) +void ptrace_bts_untrace(struct task_struct *child) { if (unlikely(child->bts)) { free_bts_context(child->bts); child->bts = NULL; } } -#else -static inline void ptrace_bts_fork(struct task_struct *tsk) {} -static inline void ptrace_bts_untrace(struct task_struct *child) {} #endif /* CONFIG_X86_PTRACE_BTS */ -void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags) -{ - ptrace_bts_fork(child); -} - -void x86_ptrace_untrace(struct task_struct *child) -{ - ptrace_bts_untrace(child); -} - /* * Called by kernel/ptrace.c when detaching.. * diff --git a/include/linux/ptrace.h b/include/linux/ptrace.h index 67c15653fc23..59e133d39d50 100644 --- a/include/linux/ptrace.h +++ b/include/linux/ptrace.h @@ -95,7 +95,6 @@ extern void __ptrace_link(struct task_struct *child, struct task_struct *new_parent); extern void __ptrace_unlink(struct task_struct *child); extern void exit_ptrace(struct task_struct *tracer); -extern void ptrace_fork(struct task_struct *task, unsigned long clone_flags); #define PTRACE_MODE_READ 1 #define PTRACE_MODE_ATTACH 2 /* Returns 0 on success, -errno on denial. */ @@ -327,15 +326,6 @@ static inline void user_enable_block_step(struct task_struct *task) #define arch_ptrace_untrace(task) do { } while (0) #endif -#ifndef arch_ptrace_fork -/* - * Do machine-specific work to initialize a new task. - * - * This is called from copy_process(). - */ -#define arch_ptrace_fork(child, clone_flags) do { } while (0) -#endif - extern int task_current_syscall(struct task_struct *target, long *callno, unsigned long args[6], unsigned int maxargs, unsigned long *sp, unsigned long *pc); diff --git a/include/linux/sched.h b/include/linux/sched.h index 52b8cd049c2e..451186a22ef5 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1205,13 +1205,11 @@ struct task_struct { struct list_head ptraced; struct list_head ptrace_entry; -#ifdef CONFIG_X86_PTRACE_BTS /* * This is the tracer handle for the ptrace BTS extension. * This field actually belongs to the ptracer task. */ struct bts_context *bts; -#endif /* CONFIG_X86_PTRACE_BTS */ /* PID/PID hash table linkage. */ struct pid_link pids[PIDTYPE_MAX]; diff --git a/kernel/fork.c b/kernel/fork.c index 660c2b8765bc..69bde7a22e9b 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1086,8 +1086,8 @@ static struct task_struct *copy_process(unsigned long clone_flags, #ifdef CONFIG_DEBUG_MUTEXES p->blocked_on = NULL; /* not blocked yet */ #endif - if (unlikely(current->ptrace)) - ptrace_fork(p, clone_flags); + + p->bts = NULL; /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); diff --git a/kernel/ptrace.c b/kernel/ptrace.c index aaad0ec34194..321127d965c2 100644 --- a/kernel/ptrace.c +++ b/kernel/ptrace.c @@ -26,16 +26,6 @@ #include -/* - * Initialize a new task whose father had been ptraced. - * - * Called from copy_process(). - */ -void ptrace_fork(struct task_struct *child, unsigned long clone_flags) -{ - arch_ptrace_fork(child, clone_flags); -} - /* * ptrace a task: make the debugger its new parent and * move it to the ptrace list. -- cgit v1.2.3 From 05725f7eb4b8acb147c5fc7b91397b1f6bcab00d Mon Sep 17 00:00:00 2001 From: Jiri Pirko Date: Tue, 14 Apr 2009 20:17:16 +0200 Subject: rculist: use list_entry_rcu in places where it's appropriate Use previously introduced list_entry_rcu instead of an open-coded list_entry + rcu_dereference combination. Signed-off-by: Jiri Pirko Reviewed-by: Paul E. McKenney Cc: dipankar@in.ibm.com LKML-Reference: <20090414181715.GA3634@psychotron.englab.brq.redhat.com> Signed-off-by: Ingo Molnar --- include/linux/sched.h | 8 +++++--- ipc/sem.c | 4 ++-- security/integrity/ima/ima_fs.c | 4 ++-- security/smack/smackfs.c | 8 ++++---- 4 files changed, 13 insertions(+), 11 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index b4c38bc8049c..886df41e7452 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -77,6 +77,7 @@ struct sched_param { #include #include #include +#include #include #include @@ -2010,7 +2011,8 @@ static inline unsigned long wait_task_inactive(struct task_struct *p, } #endif -#define next_task(p) list_entry(rcu_dereference((p)->tasks.next), struct task_struct, tasks) +#define next_task(p) \ + list_entry_rcu((p)->tasks.next, struct task_struct, tasks) #define for_each_process(p) \ for (p = &init_task ; (p = next_task(p)) != &init_task ; ) @@ -2049,8 +2051,8 @@ int same_thread_group(struct task_struct *p1, struct task_struct *p2) static inline struct task_struct *next_thread(const struct task_struct *p) { - return list_entry(rcu_dereference(p->thread_group.next), - struct task_struct, thread_group); + return list_entry_rcu(p->thread_group.next, + struct task_struct, thread_group); } static inline int thread_group_empty(struct task_struct *p) diff --git a/ipc/sem.c b/ipc/sem.c index 16a2189e96f9..87c2b641fd7b 100644 --- a/ipc/sem.c +++ b/ipc/sem.c @@ -1290,8 +1290,8 @@ void exit_sem(struct task_struct *tsk) int i; rcu_read_lock(); - un = list_entry(rcu_dereference(ulp->list_proc.next), - struct sem_undo, list_proc); + un = list_entry_rcu(ulp->list_proc.next, + struct sem_undo, list_proc); if (&un->list_proc == &ulp->list_proc) semid = -1; else diff --git a/security/integrity/ima/ima_fs.c b/security/integrity/ima/ima_fs.c index ffbe259700b1..510186f0b72e 100644 --- a/security/integrity/ima/ima_fs.c +++ b/security/integrity/ima/ima_fs.c @@ -84,8 +84,8 @@ static void *ima_measurements_next(struct seq_file *m, void *v, loff_t *pos) * against concurrent list-extension */ rcu_read_lock(); - qe = list_entry(rcu_dereference(qe->later.next), - struct ima_queue_entry, later); + qe = list_entry_rcu(qe->later.next, + struct ima_queue_entry, later); rcu_read_unlock(); (*pos)++; diff --git a/security/smack/smackfs.c b/security/smack/smackfs.c index e03a7e19c73b..11d2cb19d7a6 100644 --- a/security/smack/smackfs.c +++ b/security/smack/smackfs.c @@ -734,8 +734,8 @@ static void smk_netlbladdr_insert(struct smk_netlbladdr *new) return; } - m = list_entry(rcu_dereference(smk_netlbladdr_list.next), - struct smk_netlbladdr, list); + m = list_entry_rcu(smk_netlbladdr_list.next, + struct smk_netlbladdr, list); /* the comparison '>' is a bit hacky, but works */ if (new->smk_mask.s_addr > m->smk_mask.s_addr) { @@ -748,8 +748,8 @@ static void smk_netlbladdr_insert(struct smk_netlbladdr *new) list_add_rcu(&new->list, &m->list); return; } - m_next = list_entry(rcu_dereference(m->list.next), - struct smk_netlbladdr, list); + m_next = list_entry_rcu(m->list.next, + struct smk_netlbladdr, list); if (new->smk_mask.s_addr > m_next->smk_mask.s_addr) { list_add_rcu(&new->list, &m->list); return; -- cgit v1.2.3 From 261842b7c9099f56de2eb969c8ad65402d68e00e Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 16 Apr 2009 21:41:52 -0400 Subject: tracing: add same level recursion detection The tracing infrastructure allows for recursion. That is, an interrupt may interrupt the act of tracing an event, and that interrupt may very well perform its own trace. This is a recursive trace, and is fine to do. The problem arises when there is a bug, and the utility doing the trace calls something that recurses back into the tracer. This recursion is not caused by an external event like an interrupt, but by code that is not expected to recurse. The result could be a lockup. This patch adds a bitmask to the task structure that keeps track of the trace recursion. To find the interrupt depth, the following algorithm is used: level = hardirq_count() + softirq_count() + in_nmi; Here, level will be the depth of interrutps and softirqs, and even handles the nmi. Then the corresponding bit is set in the recursion bitmask. If the bit was already set, we know we had a recursion at the same level and we warn about it and fail the writing to the buffer. After the data has been committed to the buffer, we clear the bit. No atomics are needed. The only races are with interrupts and they reset the bitmask before returning anywy. [ Impact: detect same irq level trace recursion ] Signed-off-by: Steven Rostedt --- include/linux/ftrace.h | 7 +++++++ include/linux/init_task.h | 1 + include/linux/sched.h | 4 +++- kernel/trace/ring_buffer.c | 42 ++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 53 insertions(+), 1 deletion(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h index 97c83e1bc589..39b95c56587e 100644 --- a/include/linux/ftrace.h +++ b/include/linux/ftrace.h @@ -488,8 +488,15 @@ static inline int test_tsk_trace_graph(struct task_struct *tsk) extern int ftrace_dump_on_oops; +#ifdef CONFIG_PREEMPT +#define INIT_TRACE_RECURSION .trace_recursion = 0, +#endif + #endif /* CONFIG_TRACING */ +#ifndef INIT_TRACE_RECURSION +#define INIT_TRACE_RECURSION +#endif #ifdef CONFIG_HW_BRANCH_TRACER diff --git a/include/linux/init_task.h b/include/linux/init_task.h index dcfb93337e9a..6fc218529863 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -187,6 +187,7 @@ extern struct cred init_cred; INIT_TRACE_IRQFLAGS \ INIT_LOCKDEP \ INIT_FTRACE_GRAPH \ + INIT_TRACE_RECURSION \ } diff --git a/include/linux/sched.h b/include/linux/sched.h index b4c38bc8049c..7ede5e490913 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1428,7 +1428,9 @@ struct task_struct { #ifdef CONFIG_TRACING /* state flags for use by tracers */ unsigned long trace; -#endif + /* bitmask of trace recursion */ + unsigned long trace_recursion; +#endif /* CONFIG_TRACING */ }; /* Future-safe accessor for struct task_struct's cpus_allowed. */ diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 84a6055f37c9..b421b0ea9112 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1481,6 +1481,40 @@ rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, return event; } +static int trace_irq_level(void) +{ + return hardirq_count() + softirq_count() + in_nmi(); +} + +static int trace_recursive_lock(void) +{ + int level; + + level = trace_irq_level(); + + if (unlikely(current->trace_recursion & (1 << level))) { + /* Disable all tracing before we do anything else */ + tracing_off_permanent(); + WARN_ON_ONCE(1); + return -1; + } + + current->trace_recursion |= 1 << level; + + return 0; +} + +static void trace_recursive_unlock(void) +{ + int level; + + level = trace_irq_level(); + + WARN_ON_ONCE(!current->trace_recursion & (1 << level)); + + current->trace_recursion &= ~(1 << level); +} + static DEFINE_PER_CPU(int, rb_need_resched); /** @@ -1514,6 +1548,9 @@ ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) /* If we are tracing schedule, we don't want to recurse */ resched = ftrace_preempt_disable(); + if (trace_recursive_lock()) + goto out_nocheck; + cpu = raw_smp_processor_id(); if (!cpumask_test_cpu(cpu, buffer->cpumask)) @@ -1543,6 +1580,9 @@ ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) return event; out: + trace_recursive_unlock(); + + out_nocheck: ftrace_preempt_enable(resched); return NULL; } @@ -1581,6 +1621,8 @@ int ring_buffer_unlock_commit(struct ring_buffer *buffer, rb_commit(cpu_buffer, event); + trace_recursive_unlock(); + /* * Only the last preempt count needs to restore preemption. */ -- cgit v1.2.3 From 3bcac0263f0b45e67a64034ebcb69eb9abb742f4 Mon Sep 17 00:00:00 2001 From: David Howells Date: Wed, 29 Apr 2009 13:45:05 +0100 Subject: SELinux: Don't flush inherited SIGKILL during execve() Don't flush inherited SIGKILL during execve() in SELinux's post cred commit hook. This isn't really a security problem: if the SIGKILL came before the credentials were changed, then we were right to receive it at the time, and should honour it; if it came after the creds were changed, then we definitely should honour it; and in any case, all that will happen is that the process will be scrapped before it ever returns to userspace. Signed-off-by: David Howells Signed-off-by: Oleg Nesterov Signed-off-by: James Morris --- include/linux/sched.h | 1 + kernel/signal.c | 11 ++++++++--- security/selinux/hooks.c | 9 +++++---- 3 files changed, 14 insertions(+), 7 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index 1d19c025f9d2..d3b787c7aef3 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1875,6 +1875,7 @@ extern void sched_dead(struct task_struct *p); extern void proc_caches_init(void); extern void flush_signals(struct task_struct *); +extern void __flush_signals(struct task_struct *); extern void ignore_signals(struct task_struct *); extern void flush_signal_handlers(struct task_struct *, int force_default); extern int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info); diff --git a/kernel/signal.c b/kernel/signal.c index 1c8814481a11..f93efec14ff5 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -238,14 +238,19 @@ void flush_sigqueue(struct sigpending *queue) /* * Flush all pending signals for a task. */ +void __flush_signals(struct task_struct *t) +{ + clear_tsk_thread_flag(t, TIF_SIGPENDING); + flush_sigqueue(&t->pending); + flush_sigqueue(&t->signal->shared_pending); +} + void flush_signals(struct task_struct *t) { unsigned long flags; spin_lock_irqsave(&t->sighand->siglock, flags); - clear_tsk_thread_flag(t, TIF_SIGPENDING); - flush_sigqueue(&t->pending); - flush_sigqueue(&t->signal->shared_pending); + __flush_signals(t); spin_unlock_irqrestore(&t->sighand->siglock, flags); } diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c index dd19ba81201f..5a345115036c 100644 --- a/security/selinux/hooks.c +++ b/security/selinux/hooks.c @@ -2394,11 +2394,12 @@ static void selinux_bprm_committed_creds(struct linux_binprm *bprm) memset(&itimer, 0, sizeof itimer); for (i = 0; i < 3; i++) do_setitimer(i, &itimer, NULL); - flush_signals(current); spin_lock_irq(¤t->sighand->siglock); - flush_signal_handlers(current, 1); - sigemptyset(¤t->blocked); - recalc_sigpending(); + if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) { + __flush_signals(current); + flush_signal_handlers(current, 1); + sigemptyset(¤t->blocked); + } spin_unlock_irq(¤t->sighand->siglock); } -- cgit v1.2.3 From 5e751e992f3fb08ba35e1ca8095ec8fbf9eda523 Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 8 May 2009 13:55:22 +0100 Subject: CRED: Rename cred_exec_mutex to reflect that it's a guard against ptrace Rename cred_exec_mutex to reflect that it's a guard against foreign intervention on a process's credential state, such as is made by ptrace(). The attachment of a debugger to a process affects execve()'s calculation of the new credential state - _and_ also setprocattr()'s calculation of that state. Signed-off-by: David Howells Signed-off-by: James Morris --- fs/compat.c | 6 +++--- fs/exec.c | 10 +++++----- include/linux/init_task.h | 4 ++-- include/linux/sched.h | 4 +++- kernel/cred.c | 4 ++-- kernel/ptrace.c | 9 +++++---- 6 files changed, 20 insertions(+), 17 deletions(-) (limited to 'include/linux/sched.h') diff --git a/fs/compat.c b/fs/compat.c index 681ed81e6be0..bb2a9b2e8173 100644 --- a/fs/compat.c +++ b/fs/compat.c @@ -1488,7 +1488,7 @@ int compat_do_execve(char * filename, if (!bprm) goto out_files; - retval = mutex_lock_interruptible(¤t->cred_exec_mutex); + retval = mutex_lock_interruptible(¤t->cred_guard_mutex); if (retval < 0) goto out_free; current->in_execve = 1; @@ -1550,7 +1550,7 @@ int compat_do_execve(char * filename, /* execve succeeded */ current->fs->in_exec = 0; current->in_execve = 0; - mutex_unlock(¤t->cred_exec_mutex); + mutex_unlock(¤t->cred_guard_mutex); acct_update_integrals(current); free_bprm(bprm); if (displaced) @@ -1573,7 +1573,7 @@ out_unmark: out_unlock: current->in_execve = 0; - mutex_unlock(¤t->cred_exec_mutex); + mutex_unlock(¤t->cred_guard_mutex); out_free: free_bprm(bprm); diff --git a/fs/exec.c b/fs/exec.c index 639177b0eeac..998e856c3079 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -1045,7 +1045,7 @@ void install_exec_creds(struct linux_binprm *bprm) commit_creds(bprm->cred); bprm->cred = NULL; - /* cred_exec_mutex must be held at least to this point to prevent + /* cred_guard_mutex must be held at least to this point to prevent * ptrace_attach() from altering our determination of the task's * credentials; any time after this it may be unlocked */ @@ -1055,7 +1055,7 @@ EXPORT_SYMBOL(install_exec_creds); /* * determine how safe it is to execute the proposed program - * - the caller must hold current->cred_exec_mutex to protect against + * - the caller must hold current->cred_guard_mutex to protect against * PTRACE_ATTACH */ int check_unsafe_exec(struct linux_binprm *bprm) @@ -1297,7 +1297,7 @@ int do_execve(char * filename, if (!bprm) goto out_files; - retval = mutex_lock_interruptible(¤t->cred_exec_mutex); + retval = mutex_lock_interruptible(¤t->cred_guard_mutex); if (retval < 0) goto out_free; current->in_execve = 1; @@ -1360,7 +1360,7 @@ int do_execve(char * filename, /* execve succeeded */ current->fs->in_exec = 0; current->in_execve = 0; - mutex_unlock(¤t->cred_exec_mutex); + mutex_unlock(¤t->cred_guard_mutex); acct_update_integrals(current); free_bprm(bprm); if (displaced) @@ -1383,7 +1383,7 @@ out_unmark: out_unlock: current->in_execve = 0; - mutex_unlock(¤t->cred_exec_mutex); + mutex_unlock(¤t->cred_guard_mutex); out_free: free_bprm(bprm); diff --git a/include/linux/init_task.h b/include/linux/init_task.h index d87247d2641f..7f54ba942429 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -145,8 +145,8 @@ extern struct cred init_cred; .group_leader = &tsk, \ .real_cred = &init_cred, \ .cred = &init_cred, \ - .cred_exec_mutex = \ - __MUTEX_INITIALIZER(tsk.cred_exec_mutex), \ + .cred_guard_mutex = \ + __MUTEX_INITIALIZER(tsk.cred_guard_mutex), \ .comm = "swapper", \ .thread = INIT_THREAD, \ .fs = &init_fs, \ diff --git a/include/linux/sched.h b/include/linux/sched.h index 3fa82b353c98..5932ace22400 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1247,7 +1247,9 @@ struct task_struct { * credentials (COW) */ const struct cred *cred; /* effective (overridable) subjective task * credentials (COW) */ - struct mutex cred_exec_mutex; /* execve vs ptrace cred calculation mutex */ + struct mutex cred_guard_mutex; /* guard against foreign influences on + * credential calculations + * (notably. ptrace) */ char comm[TASK_COMM_LEN]; /* executable name excluding path - access with [gs]et_task_comm (which lock diff --git a/kernel/cred.c b/kernel/cred.c index 3a039189d707..1bb4d7e5d616 100644 --- a/kernel/cred.c +++ b/kernel/cred.c @@ -167,7 +167,7 @@ EXPORT_SYMBOL(prepare_creds); /* * Prepare credentials for current to perform an execve() - * - The caller must hold current->cred_exec_mutex + * - The caller must hold current->cred_guard_mutex */ struct cred *prepare_exec_creds(void) { @@ -276,7 +276,7 @@ int copy_creds(struct task_struct *p, unsigned long clone_flags) struct cred *new; int ret; - mutex_init(&p->cred_exec_mutex); + mutex_init(&p->cred_guard_mutex); if ( #ifdef CONFIG_KEYS diff --git a/kernel/ptrace.c b/kernel/ptrace.c index 0692ab5a0d67..27ac80298bfa 100644 --- a/kernel/ptrace.c +++ b/kernel/ptrace.c @@ -185,10 +185,11 @@ int ptrace_attach(struct task_struct *task) if (same_thread_group(task, current)) goto out; - /* Protect exec's credential calculations against our interference; - * SUID, SGID and LSM creds get determined differently under ptrace. + /* Protect the target's credential calculations against our + * interference; SUID, SGID and LSM creds get determined differently + * under ptrace. */ - retval = mutex_lock_interruptible(&task->cred_exec_mutex); + retval = mutex_lock_interruptible(&task->cred_guard_mutex); if (retval < 0) goto out; @@ -232,7 +233,7 @@ repeat: bad: write_unlock_irqrestore(&tasklist_lock, flags); task_unlock(task); - mutex_unlock(&task->cred_exec_mutex); + mutex_unlock(&task->cred_guard_mutex); out: return retval; } -- cgit v1.2.3 From 789f90fcf6b0b54e655740e9396c954378542c79 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 15:19:27 +0200 Subject: perf_counter: per user mlock gift Instead of a per-process mlock gift for perf-counters, use a per-user gift so that there is less of a DoS potential. [ Impact: allow less worst-case unprivileged memory consumption ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo LKML-Reference: <20090515132018.496182835@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/sched.h | 4 ++++ kernel/perf_counter.c | 22 +++++++++++++++------- 2 files changed, 19 insertions(+), 7 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index d1857580a132..ff59d1231519 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -674,6 +674,10 @@ struct user_struct { struct work_struct work; #endif #endif + +#ifdef CONFIG_PERF_COUNTERS + atomic_long_t locked_vm; +#endif }; extern int uids_sysfs_init(void); diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0173738dd548..93f4a0e4b873 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -45,7 +45,7 @@ static atomic_t nr_munmap_tracking __read_mostly; static atomic_t nr_comm_tracking __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ -int sysctl_perf_counter_mlock __read_mostly = 128; /* 'free' kb per counter */ +int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ /* * Lock for (sysadmin-configurable) counter reservations: @@ -1522,6 +1522,9 @@ static void perf_mmap_close(struct vm_area_struct *vma) if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { + struct user_struct *user = current_user(); + + atomic_long_sub(counter->data->nr_pages + 1, &user->locked_vm); vma->vm_mm->locked_vm -= counter->data->nr_locked; perf_mmap_data_free(counter); mutex_unlock(&counter->mmap_mutex); @@ -1537,11 +1540,13 @@ static struct vm_operations_struct perf_mmap_vmops = { static int perf_mmap(struct file *file, struct vm_area_struct *vma) { struct perf_counter *counter = file->private_data; + struct user_struct *user = current_user(); unsigned long vma_size; unsigned long nr_pages; + unsigned long user_locked, user_lock_limit; unsigned long locked, lock_limit; + long user_extra, extra; int ret = 0; - long extra; if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) return -EINVAL; @@ -1569,15 +1574,17 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; } - extra = nr_pages /* + 1 only account the data pages */; - extra -= sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); - if (extra < 0) - extra = 0; + user_extra = nr_pages + 1; + user_lock_limit = sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); + user_locked = atomic_long_read(&user->locked_vm) + user_extra; - locked = vma->vm_mm->locked_vm + extra; + extra = 0; + if (user_locked > user_lock_limit) + extra = user_locked - user_lock_limit; lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; lock_limit >>= PAGE_SHIFT; + locked = vma->vm_mm->locked_vm + extra; if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { ret = -EPERM; @@ -1590,6 +1597,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; atomic_set(&counter->mmap_count, 1); + atomic_long_add(user_extra, &user->locked_vm); vma->vm_mm->locked_vm += extra; counter->data->nr_locked = extra; unlock: -- cgit v1.2.3 From dce48a84adf1806676319f6f480e30a6daa012f9 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 11 Apr 2009 10:43:41 +0200 Subject: sched, timers: move calc_load() to scheduler Dimitri Sivanich noticed that xtime_lock is held write locked across calc_load() which iterates over all online CPUs. That can cause long latencies for xtime_lock readers on large SMP systems. The load average calculation is an rough estimate anyway so there is no real need to protect the readers vs. the update. It's not a problem when the avenrun array is updated while a reader copies the values. Instead of iterating over all online CPUs let the scheduler_tick code update the number of active tasks shortly before the avenrun update happens. The avenrun update itself is handled by the CPU which calls do_timer(). [ Impact: reduce xtime_lock write locked section ] Signed-off-by: Thomas Gleixner Acked-by: Peter Zijlstra --- include/linux/sched.h | 2 +- kernel/sched.c | 84 +++++++++++++++++++++++++++++++++++++++++------ kernel/sched_idletask.c | 3 +- kernel/time/timekeeping.c | 2 +- kernel/timer.c | 54 ++---------------------------- 5 files changed, 80 insertions(+), 65 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index b4c38bc8049c..6eb4892efe45 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -135,8 +135,8 @@ DECLARE_PER_CPU(unsigned long, process_counts); extern int nr_processes(void); extern unsigned long nr_running(void); extern unsigned long nr_uninterruptible(void); -extern unsigned long nr_active(void); extern unsigned long nr_iowait(void); +extern void calc_global_load(void); extern unsigned long get_parent_ip(unsigned long addr); diff --git a/kernel/sched.c b/kernel/sched.c index 8908d190a348..f4eb88153bd1 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -630,6 +630,10 @@ struct rq { struct list_head migration_queue; #endif + /* calc_load related fields */ + unsigned long calc_load_update; + long calc_load_active; + #ifdef CONFIG_SCHED_HRTICK #ifdef CONFIG_SMP int hrtick_csd_pending; @@ -1728,6 +1732,8 @@ static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) } #endif +static void calc_load_account_active(struct rq *this_rq); + #include "sched_stats.h" #include "sched_idletask.c" #include "sched_fair.c" @@ -2856,19 +2862,57 @@ unsigned long nr_iowait(void) return sum; } -unsigned long nr_active(void) +/* Variables and functions for calc_load */ +static atomic_long_t calc_load_tasks; +static unsigned long calc_load_update; +unsigned long avenrun[3]; +EXPORT_SYMBOL(avenrun); + +static unsigned long +calc_load(unsigned long load, unsigned long exp, unsigned long active) { - unsigned long i, running = 0, uninterruptible = 0; + load *= exp; + load += active * (FIXED_1 - exp); + return load >> FSHIFT; +} - for_each_online_cpu(i) { - running += cpu_rq(i)->nr_running; - uninterruptible += cpu_rq(i)->nr_uninterruptible; - } +/* + * calc_load - update the avenrun load estimates 10 ticks after the + * CPUs have updated calc_load_tasks. + */ +void calc_global_load(void) +{ + unsigned long upd = calc_load_update + 10; + long active; + + if (time_before(jiffies, upd)) + return; - if (unlikely((long)uninterruptible < 0)) - uninterruptible = 0; + active = atomic_long_read(&calc_load_tasks); + active = active > 0 ? active * FIXED_1 : 0; - return running + uninterruptible; + avenrun[0] = calc_load(avenrun[0], EXP_1, active); + avenrun[1] = calc_load(avenrun[1], EXP_5, active); + avenrun[2] = calc_load(avenrun[2], EXP_15, active); + + calc_load_update += LOAD_FREQ; +} + +/* + * Either called from update_cpu_load() or from a cpu going idle + */ +static void calc_load_account_active(struct rq *this_rq) +{ + long nr_active, delta; + + nr_active = this_rq->nr_running; + nr_active += (long) this_rq->nr_uninterruptible; + + if (nr_active != this_rq->calc_load_active) { + delta = nr_active - this_rq->calc_load_active; + this_rq->calc_load_active = nr_active; + atomic_long_add(delta, &calc_load_tasks); + } } /* @@ -2899,6 +2943,11 @@ static void update_cpu_load(struct rq *this_rq) new_load += scale-1; this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; } + + if (time_after_eq(jiffies, this_rq->calc_load_update)) { + this_rq->calc_load_update += LOAD_FREQ; + calc_load_account_active(this_rq); + } } #ifdef CONFIG_SMP @@ -7091,6 +7140,14 @@ static void migrate_dead_tasks(unsigned int dead_cpu) } } + +/* + * remove the tasks which were accounted by rq from calc_load_tasks. + */ +static void calc_global_load_remove(struct rq *rq) +{ + atomic_long_sub(rq->calc_load_active, &calc_load_tasks); +} #endif /* CONFIG_HOTPLUG_CPU */ #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) @@ -7325,6 +7382,8 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) /* Update our root-domain */ rq = cpu_rq(cpu); spin_lock_irqsave(&rq->lock, flags); + rq->calc_load_update = calc_load_update; + rq->calc_load_active = 0; if (rq->rd) { BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); @@ -7364,7 +7423,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) cpuset_unlock(); migrate_nr_uninterruptible(rq); BUG_ON(rq->nr_running != 0); - + calc_global_load_remove(rq); /* * No need to migrate the tasks: it was best-effort if * they didn't take sched_hotcpu_mutex. Just wake up @@ -9059,6 +9118,8 @@ void __init sched_init(void) rq = cpu_rq(i); spin_lock_init(&rq->lock); rq->nr_running = 0; + rq->calc_load_active = 0; + rq->calc_load_update = jiffies + LOAD_FREQ; init_cfs_rq(&rq->cfs, rq); init_rt_rq(&rq->rt, rq); #ifdef CONFIG_FAIR_GROUP_SCHED @@ -9166,6 +9227,9 @@ void __init sched_init(void) * when this runqueue becomes "idle". */ init_idle(current, smp_processor_id()); + + calc_load_update = jiffies + LOAD_FREQ; + /* * During early bootup we pretend to be a normal task: */ diff --git a/kernel/sched_idletask.c b/kernel/sched_idletask.c index 8a21a2e28c13..499672c10cbd 100644 --- a/kernel/sched_idletask.c +++ b/kernel/sched_idletask.c @@ -22,7 +22,8 @@ static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int sy static struct task_struct *pick_next_task_idle(struct rq *rq) { schedstat_inc(rq, sched_goidle); - + /* adjust the active tasks as we might go into a long sleep */ + calc_load_account_active(rq); return rq->idle; } diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 687dff49f6e7..52a8bf8931f3 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -22,7 +22,7 @@ /* * This read-write spinlock protects us from races in SMP while - * playing with xtime and avenrun. + * playing with xtime. */ __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); diff --git a/kernel/timer.c b/kernel/timer.c index cffffad01c31..6a21d7af9620 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -1122,47 +1122,6 @@ void update_process_times(int user_tick) run_posix_cpu_timers(p); } -/* - * Nr of active tasks - counted in fixed-point numbers - */ -static unsigned long count_active_tasks(void) -{ - return nr_active() * FIXED_1; -} - -/* - * Hmm.. Changed this, as the GNU make sources (load.c) seems to - * imply that avenrun[] is the standard name for this kind of thing. - * Nothing else seems to be standardized: the fractional size etc - * all seem to differ on different machines. - * - * Requires xtime_lock to access. - */ -unsigned long avenrun[3]; - -EXPORT_SYMBOL(avenrun); - -/* - * calc_load - given tick count, update the avenrun load estimates. - * This is called while holding a write_lock on xtime_lock. - */ -static inline void calc_load(unsigned long ticks) -{ - unsigned long active_tasks; /* fixed-point */ - static int count = LOAD_FREQ; - - count -= ticks; - if (unlikely(count < 0)) { - active_tasks = count_active_tasks(); - do { - CALC_LOAD(avenrun[0], EXP_1, active_tasks); - CALC_LOAD(avenrun[1], EXP_5, active_tasks); - CALC_LOAD(avenrun[2], EXP_15, active_tasks); - count += LOAD_FREQ; - } while (count < 0); - } -} - /* * This function runs timers and the timer-tq in bottom half context. */ @@ -1186,16 +1145,6 @@ void run_local_timers(void) softlockup_tick(); } -/* - * Called by the timer interrupt. xtime_lock must already be taken - * by the timer IRQ! - */ -static inline void update_times(unsigned long ticks) -{ - update_wall_time(); - calc_load(ticks); -} - /* * The 64-bit jiffies value is not atomic - you MUST NOT read it * without sampling the sequence number in xtime_lock. @@ -1205,7 +1154,8 @@ static inline void update_times(unsigned long ticks) void do_timer(unsigned long ticks) { jiffies_64 += ticks; - update_times(ticks); + update_wall_time(); + calc_global_load(); } #ifdef __ARCH_WANT_SYS_ALARM -- cgit v1.2.3 From 2d02494f5a90f2e4b3c4c6acc85ec94674cdc431 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 2 May 2009 20:08:52 +0200 Subject: sched, timers: cleanup avenrun users avenrun is an rough estimate so we don't have to worry about consistency of the three avenrun values. Remove the xtime lock dependency and provide a function to scale the values. Cleanup the users. [ Impact: cleanup ] Signed-off-by: Thomas Gleixner Acked-by: Peter Zijlstra --- fs/proc/loadavg.c | 18 ++++++------------ include/linux/sched.h | 1 + kernel/sched.c | 15 +++++++++++++++ kernel/timer.c | 32 ++++++-------------------------- 4 files changed, 28 insertions(+), 38 deletions(-) (limited to 'include/linux/sched.h') diff --git a/fs/proc/loadavg.c b/fs/proc/loadavg.c index 9bca39cf99ee..1afa4dd4cae2 100644 --- a/fs/proc/loadavg.c +++ b/fs/proc/loadavg.c @@ -12,20 +12,14 @@ static int loadavg_proc_show(struct seq_file *m, void *v) { - int a, b, c; - unsigned long seq; + unsigned long avnrun[3]; - do { - seq = read_seqbegin(&xtime_lock); - a = avenrun[0] + (FIXED_1/200); - b = avenrun[1] + (FIXED_1/200); - c = avenrun[2] + (FIXED_1/200); - } while (read_seqretry(&xtime_lock, seq)); + get_avenrun(avnrun, FIXED_1/200, 0); - seq_printf(m, "%d.%02d %d.%02d %d.%02d %ld/%d %d\n", - LOAD_INT(a), LOAD_FRAC(a), - LOAD_INT(b), LOAD_FRAC(b), - LOAD_INT(c), LOAD_FRAC(c), + seq_printf(m, "%lu.%02lu %lu.%02lu %lu.%02lu %ld/%d %d\n", + LOAD_INT(avnrun[0]), LOAD_FRAC(avnrun[0]), + LOAD_INT(avnrun[1]), LOAD_FRAC(avnrun[1]), + LOAD_INT(avnrun[2]), LOAD_FRAC(avnrun[2]), nr_running(), nr_threads, task_active_pid_ns(current)->last_pid); return 0; diff --git a/include/linux/sched.h b/include/linux/sched.h index 6eb4892efe45..de7b3b217772 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -116,6 +116,7 @@ struct fs_struct; * 11 bit fractions. */ extern unsigned long avenrun[]; /* Load averages */ +extern void get_avenrun(unsigned long *loads, unsigned long offset, int shift); #define FSHIFT 11 /* nr of bits of precision */ #define FIXED_1 (1<= 0) { - tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC; - tp.tv_sec++; - } - info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); + ktime_get_ts(&tp); + monotonic_to_bootbased(&tp); + info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); - info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT); - info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT); - info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT); + get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); - info->procs = nr_threads; - } while (read_seqretry(&xtime_lock, seq)); + info->procs = nr_threads; si_meminfo(info); si_swapinfo(info); -- cgit v1.2.3 From 4200efd9acda4accf24640f1e77d24fdcdb524df Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 19 May 2009 09:22:19 +0200 Subject: sched: properly define the sched_group::cpumask and sched_domain::span fields Properly document the variable-size structure tricks we are doing wrt. struct sched_group and sched_domain, and use the field[0] GCC extension instead of defining a vla array. Dont use unions for this, as pointed out by Linus. [ Impact: cleanup, un-confuse Sparse and LLVM ] Reported-by: Jeff Garzik Acked-by: Linus Torvalds LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/sched.h | 25 ++++++++++++++++++++++--- kernel/sched.c | 5 +++-- 2 files changed, 25 insertions(+), 5 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/sched.h b/include/linux/sched.h index de7b3b217772..dbb1043e8656 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -839,7 +839,17 @@ struct sched_group { */ u32 reciprocal_cpu_power; - unsigned long cpumask[]; + /* + * The CPUs this group covers. + * + * NOTE: this field is variable length. (Allocated dynamically + * by attaching extra space to the end of the structure, + * depending on how many CPUs the kernel has booted up with) + * + * It is also be embedded into static data structures at build + * time. (See 'struct static_sched_group' in kernel/sched.c) + */ + unsigned long cpumask[0]; }; static inline struct cpumask *sched_group_cpus(struct sched_group *sg) @@ -925,8 +935,17 @@ struct sched_domain { char *name; #endif - /* span of all CPUs in this domain */ - unsigned long span[]; + /* + * Span of all CPUs in this domain. + * + * NOTE: this field is variable length. (Allocated dynamically + * by attaching extra space to the end of the structure, + * depending on how many CPUs the kernel has booted up with) + * + * It is also be embedded into static data structures at build + * time. (See 'struct static_sched_domain' in kernel/sched.c) + */ + unsigned long span[0]; }; static inline struct cpumask *sched_domain_span(struct sched_domain *sd) diff --git a/kernel/sched.c b/kernel/sched.c index 497c09ba61e7..228acae8821f 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -7948,8 +7948,9 @@ int sched_smt_power_savings = 0, sched_mc_power_savings = 0; /* * The cpus mask in sched_group and sched_domain hangs off the end. - * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space - * for nr_cpu_ids < CONFIG_NR_CPUS. + * + * ( See the the comments in include/linux/sched.h:struct sched_group + * and struct sched_domain. ) */ struct static_sched_group { struct sched_group sg; -- 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 'include/linux/sched.h') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index e9021a908020..b4f64402a82a 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 503afaa0afa7..d87247d2641f 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 f612941ef46e..071309005468 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 ff59d1231519..9714d450f417 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 f9dfedd94af0..99ad4063ee4a 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 d32fef4d38e5..e72a09f5355b 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 08584c16049f..06ea3eae886e 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 082ff5a2767a0679ee543f14883adbafb631ffbe Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:29:00 +0200 Subject: perf_counter: Change pctrl() behaviour Instead of en/dis-abling all counters acting on a particular task, en/dis- able all counters we created. [ v2: fix crash on first counter enable ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.916937244@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/init_task.h | 10 +++++ include/linux/perf_counter.h | 3 ++ include/linux/sched.h | 2 + kernel/perf_counter.c | 87 ++++++++++++-------------------------------- 4 files changed, 39 insertions(+), 63 deletions(-) (limited to 'include/linux/sched.h') diff --git a/include/linux/init_task.h b/include/linux/init_task.h index d87247d2641f..353c0ac7723a 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -108,6 +108,15 @@ extern struct group_info init_groups; extern struct cred init_cred; +#ifdef CONFIG_PERF_COUNTERS +# define INIT_PERF_COUNTERS(tsk) \ + .perf_counter_mutex = \ + __MUTEX_INITIALIZER(tsk.perf_counter_mutex), \ + .perf_counter_list = LIST_HEAD_INIT(tsk.perf_counter_list), +#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) @@ -171,6 +180,7 @@ 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 4ab8050eb9e8..4159ee5940f8 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -469,6 +469,9 @@ struct perf_counter { int oncpu; int cpu; + struct list_head owner_entry; + struct task_struct *owner; + /* mmap bits */ struct mutex mmap_mutex; atomic_t mmap_count; diff --git a/include/linux/sched.h b/include/linux/sched.h index 9714d450f417..bc9326dcdde1 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1389,6 +1389,8 @@ struct task_struct { #endif #ifdef CONFIG_PERF_COUNTERS struct perf_counter_context *perf_counter_ctxp; + struct mutex perf_counter_mutex; + struct list_head perf_counter_list; #endif #ifdef CONFIG_NUMA struct mempolicy *mempolicy; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0e97f8961333..4c86a6369764 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1076,79 +1076,26 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) __perf_counter_sched_in(ctx, cpuctx, cpu); } -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_ctxp; struct perf_counter *counter; - unsigned long flags; - - if (!ctx || !ctx->nr_counters) - return 0; - - local_irq_save(flags); - __perf_counter_task_sched_out(ctx); - - spin_lock(&ctx->lock); - - /* - * Disable all the counters: - */ - perf_disable(); - - list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state != PERF_COUNTER_STATE_ERROR) { - update_group_times(counter); - counter->state = PERF_COUNTER_STATE_OFF; - } - } - - perf_enable(); - - spin_unlock_irqrestore(&ctx->lock, flags); + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_enable(counter); + mutex_unlock(¤t->perf_counter_mutex); return 0; } -int perf_counter_task_enable(void) +int perf_counter_task_disable(void) { - struct task_struct *curr = current; - struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; - unsigned long flags; - int cpu; - - if (!ctx || !ctx->nr_counters) - return 0; - - local_irq_save(flags); - cpu = smp_processor_id(); - - __perf_counter_task_sched_out(ctx); - - spin_lock(&ctx->lock); - /* - * Disable all the counters: - */ - perf_disable(); - - list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state > PERF_COUNTER_STATE_OFF) - continue; - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = - ctx->time - counter->total_time_enabled; - counter->hw_event.disabled = 0; - } - perf_enable(); - - spin_unlock(&ctx->lock); - - perf_counter_task_sched_in(curr, cpu); - - local_irq_restore(flags); + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_disable(counter); + mutex_unlock(¤t->perf_counter_mutex); return 0; } @@ -1416,6 +1363,11 @@ static int perf_release(struct inode *inode, struct file *file) perf_counter_remove_from_context(counter); mutex_unlock(&ctx->mutex); + mutex_lock(&counter->owner->perf_counter_mutex); + list_del_init(&counter->owner_entry); + mutex_unlock(&counter->owner->perf_counter_mutex); + put_task_struct(counter->owner); + free_counter(counter); put_context(ctx); @@ -3272,6 +3224,12 @@ SYSCALL_DEFINE5(perf_counter_open, perf_install_in_context(ctx, counter, cpu); mutex_unlock(&ctx->mutex); + counter->owner = current; + get_task_struct(current); + mutex_lock(¤t->perf_counter_mutex); + list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); + mutex_unlock(¤t->perf_counter_mutex); + fput_light(counter_file, fput_needed2); out_fput: @@ -3488,6 +3446,9 @@ void perf_counter_init_task(struct task_struct *child) child->perf_counter_ctxp = NULL; + mutex_init(&child->perf_counter_mutex); + INIT_LIST_HEAD(&child->perf_counter_list); + /* * This is executed from the parent task context, so inherit * counters that have been marked for cloning. -- cgit v1.2.3