From owner-svn-src-all@FreeBSD.ORG Sun Mar 15 06:41:47 2009 Return-Path: Delivered-To: svn-src-all@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id E5E621065673; Sun, 15 Mar 2009 06:41:47 +0000 (UTC) (envelope-from jeff@FreeBSD.org) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:4f8:fff6::2c]) by mx1.freebsd.org (Postfix) with ESMTP id D37188FC17; Sun, 15 Mar 2009 06:41:47 +0000 (UTC) (envelope-from jeff@FreeBSD.org) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.3/8.14.3) with ESMTP id n2F6fl6m019402; Sun, 15 Mar 2009 06:41:47 GMT (envelope-from jeff@svn.freebsd.org) Received: (from jeff@localhost) by svn.freebsd.org (8.14.3/8.14.3/Submit) id n2F6fluw019399; Sun, 15 Mar 2009 06:41:47 GMT (envelope-from jeff@svn.freebsd.org) Message-Id: <200903150641.n2F6fluw019399@svn.freebsd.org> From: Jeff Roberson Date: Sun, 15 Mar 2009 06:41:47 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cc: Subject: svn commit: r189845 - in head/sys: kern sys X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sun, 15 Mar 2009 06:41:48 -0000 Author: jeff Date: Sun Mar 15 06:41:47 2009 New Revision: 189845 URL: http://svn.freebsd.org/changeset/base/189845 Log: - Implement a new mechanism for resetting lock profiling. We now guarantee that all cpus have acknowledged the cleared enable int by scheduling the resetting thread on each cpu in succession. Since all lock profiling happens within a critical section this guarantees that all cpus have left lock profiling before we clear the datastructures. - Assert that the per-thread queue of locks lock profiling is aware of is clear on thread exit. There were several cases where this was not true that slows lock profiling and leaks information. - Remove all objects from all lists before clearing any per-cpu information in reset. Lock profiling objects can migrate between per-cpu caches and previously these migrated objects could be zero'd before they'd been removed Discussed with: attilio Sponsored by: Nokia Modified: head/sys/kern/kern_thread.c head/sys/kern/subr_lock.c head/sys/sys/lock_profile.h Modified: head/sys/kern/kern_thread.c ============================================================================== --- head/sys/kern/kern_thread.c Sun Mar 15 06:40:57 2009 (r189844) +++ head/sys/kern/kern_thread.c Sun Mar 15 06:41:47 2009 (r189845) @@ -306,6 +306,8 @@ thread_alloc(void) void thread_free(struct thread *td) { + + lock_profile_thread_exit(td); if (td->td_cpuset) cpuset_rel(td->td_cpuset); td->td_cpuset = NULL; @@ -439,6 +441,7 @@ thread_wait(struct proc *p) /* Wait for any remaining threads to exit cpu_throw(). */ while (p->p_exitthreads) sched_relinquish(curthread); + lock_profile_thread_exit(td); cpuset_rel(td->td_cpuset); td->td_cpuset = NULL; cpu_thread_clean(td); Modified: head/sys/kern/subr_lock.c ============================================================================== --- head/sys/kern/subr_lock.c Sun Mar 15 06:40:57 2009 (r189844) +++ head/sys/kern/subr_lock.c Sun Mar 15 06:41:47 2009 (r189845) @@ -46,9 +46,11 @@ __FBSDID("$FreeBSD$"); #include #include #include +#include #include #include #include +#include #include #include @@ -186,7 +188,8 @@ struct lock_prof_cpu { struct lock_prof_cpu *lp_cpu[MAXCPU]; -int lock_prof_enable = 0; +volatile int lock_prof_enable = 0; +static volatile int lock_prof_resetting; /* SWAG: sbuf size = avg stat. line size * number of locks */ #define LPROF_SBUF_SIZE 256 * 400 @@ -239,25 +242,77 @@ lock_prof_init(void *arg) } SYSINIT(lockprof, SI_SUB_SMP, SI_ORDER_ANY, lock_prof_init, NULL); +/* + * To be certain that lock profiling has idled on all cpus before we + * reset, we schedule the resetting thread on all active cpus. Since + * all operations happen within critical sections we can be sure that + * it is safe to zero the profiling structures. + */ +static void +lock_prof_idle(void) +{ + struct thread *td; + int cpu; + + td = curthread; + thread_lock(td); + for (cpu = 0; cpu <= mp_maxid; cpu++) { + if (CPU_ABSENT(cpu)) + continue; + sched_bind(td, cpu); + } + sched_unbind(td); + thread_unlock(td); +} + +static void +lock_prof_reset_wait(void) +{ + + /* + * Spin relinquishing our cpu so that lock_prof_idle may + * run on it. + */ + while (lock_prof_resetting) + sched_relinquish(curthread); +} + static void lock_prof_reset(void) { struct lock_prof_cpu *lpc; int enabled, i, cpu; + /* + * We not only race with acquiring and releasing locks but also + * thread exit. To be certain that threads exit without valid head + * pointers they must see resetting set before enabled is cleared. + * Otherwise a lock may not be removed from a per-thread list due + * to disabled being set but not wait for reset() to remove it below. + */ + atomic_store_rel_int(&lock_prof_resetting, 1); enabled = lock_prof_enable; lock_prof_enable = 0; - pause("lpreset", hz / 10); + lock_prof_idle(); + /* + * Some objects may have migrated between CPUs. Clear all links + * before we zero the structures. Some items may still be linked + * into per-thread lists as well. + */ for (cpu = 0; cpu <= mp_maxid; cpu++) { lpc = lp_cpu[cpu]; for (i = 0; i < LPROF_CACHE_SIZE; i++) { LIST_REMOVE(&lpc->lpc_types[0].lpt_objs[i], lpo_link); LIST_REMOVE(&lpc->lpc_types[1].lpt_objs[i], lpo_link); } + } + for (cpu = 0; cpu <= mp_maxid; cpu++) { + lpc = lp_cpu[cpu]; bzero(lpc, sizeof(*lpc)); lock_prof_init_type(&lpc->lpc_types[0]); lock_prof_init_type(&lpc->lpc_types[1]); } + atomic_store_rel_int(&lock_prof_resetting, 0); lock_prof_enable = enabled; } @@ -351,7 +406,7 @@ retry_sbufops: "max", "wait_max", "total", "wait_total", "count", "avg", "wait_avg", "cnt_hold", "cnt_lock", "name"); enabled = lock_prof_enable; lock_prof_enable = 0; - pause("lpreset", hz / 10); + lock_prof_idle(); t = ticks; for (cpu = 0; cpu <= mp_maxid; cpu++) { if (lp_cpu[cpu] == NULL) @@ -461,16 +516,13 @@ lock_profile_object_lookup(struct lock_o if (l->lpo_obj == lo && l->lpo_file == file && l->lpo_line == line) return (l); - critical_enter(); type = &lp_cpu[PCPU_GET(cpuid)]->lpc_types[spin]; l = LIST_FIRST(&type->lpt_lpoalloc); if (l == NULL) { lock_prof_rejected++; - critical_exit(); return (NULL); } LIST_REMOVE(l, lpo_link); - critical_exit(); l->lpo_obj = lo; l->lpo_file = file; l->lpo_line = line; @@ -497,18 +549,49 @@ lock_profile_obtain_lock_success(struct spin = (LOCK_CLASS(lo)->lc_flags & LC_SPINLOCK) ? 1 : 0; if (spin && lock_prof_skipspin == 1) return; + critical_enter(); + /* Recheck enabled now that we're in a critical section. */ + if (lock_prof_enable == 0) + goto out; l = lock_profile_object_lookup(lo, spin, file, line); if (l == NULL) - return; + goto out; l->lpo_cnt++; if (++l->lpo_ref > 1) - return; + goto out; l->lpo_contest_locking = contested; l->lpo_acqtime = nanoseconds(); if (waittime && (l->lpo_acqtime > waittime)) l->lpo_waittime = l->lpo_acqtime - waittime; else l->lpo_waittime = 0; +out: + critical_exit(); +} + +void +lock_profile_thread_exit(struct thread *td) +{ +#ifdef INVARIANTS + struct lock_profile_object *l; + + MPASS(curthread->td_critnest == 0); +#endif + /* + * If lock profiling was disabled we have to wait for reset to + * clear our pointers before we can exit safely. + */ + lock_prof_reset_wait(); +#ifdef INVARIANTS + LIST_FOREACH(l, &td->td_lprof[0], lpo_link) + printf("thread still holds lock acquired at %s:%d\n", + l->lpo_file, l->lpo_line); + LIST_FOREACH(l, &td->td_lprof[1], lpo_link) + printf("thread still holds lock acquired at %s:%d\n", + l->lpo_file, l->lpo_line); +#endif + MPASS(LIST_FIRST(&td->td_lprof[0]) == NULL); + MPASS(LIST_FIRST(&td->td_lprof[1]) == NULL); } void @@ -521,11 +604,20 @@ lock_profile_release_lock(struct lock_ob struct lpohead *head; int spin; - if (!lock_prof_enable || (lo->lo_flags & LO_NOPROFILE)) + if (lo->lo_flags & LO_NOPROFILE) return; spin = (LOCK_CLASS(lo)->lc_flags & LC_SPINLOCK) ? 1 : 0; head = &curthread->td_lprof[spin]; + if (LIST_FIRST(head) == NULL) + return; critical_enter(); + /* Recheck enabled now that we're in a critical section. */ + if (lock_prof_enable == 0 && lock_prof_resetting == 1) + goto out; + /* + * If lock profiling is not enabled we still want to remove the + * lpo from our queue. + */ LIST_FOREACH(l, head, lpo_link) if (l->lpo_obj == lo) break; Modified: head/sys/sys/lock_profile.h ============================================================================== --- head/sys/sys/lock_profile.h Sun Mar 15 06:40:57 2009 (r189844) +++ head/sys/sys/lock_profile.h Sun Mar 15 06:41:47 2009 (r189845) @@ -43,11 +43,13 @@ LIST_HEAD(lpohead, lock_profile_object); u_int64_t nanoseconds(void); #endif -extern int lock_prof_enable; +extern volatile int lock_prof_enable; void lock_profile_obtain_lock_success(struct lock_object *lo, int contested, uint64_t waittime, const char *file, int line); void lock_profile_release_lock(struct lock_object *lo); +void lock_profile_thread_exit(struct thread *td); + static inline void lock_profile_obtain_lock_failed(struct lock_object *lo, int *contested, @@ -61,21 +63,10 @@ lock_profile_obtain_lock_failed(struct l #else /* !LOCK_PROFILING */ -static inline void -lock_profile_release_lock(struct lock_object *lo) -{ -} - -static inline void -lock_profile_obtain_lock_failed(struct lock_object *lo, int *contested, uint64_t *waittime) -{ -} - -static inline void -lock_profile_obtain_lock_success(struct lock_object *lo, int contested, uint64_t waittime, - const char *file, int line) -{ -} +#define lock_profile_release_lock(lo) +#define lock_profile_obtain_lock_failed(lo, contested, waittime) +#define lock_profile_obtain_lock_success(lo, contested, waittime, file, line) +#define lock_profile_thread_exit(td) #endif /* !LOCK_PROFILING */