From b0655179358fbd6917b144860a6e1b180e5bc26b Mon Sep 17 00:00:00 2001 From: jeff Date: Wed, 2 Apr 2008 11:20:30 +0000 Subject: Implement per-cpu callout threads, wheels, and locks. - Move callout thread creation from kern_intr.c to kern_timeout.c - Call callout_tick() on every processor via hardclock_cpu() rather than inspecting callout internal details in kern_clock.c. - Remove callout implementation details from callout.h - Package up all of the global variables into a per-cpu callout structure. - Start one thread per-cpu. Threads are not strictly bound. They prefer to execute on the native cpu but may migrate temporarily if interrupts are starving callout processing. - Run all callouts by default in the thread for cpu0 to maintain current ordering and concurrency guarantees. Many consumers may not properly handle concurrent execution. - The new callout_reset_on() api allows specifying a particular cpu to execute the callout on. This may migrate a callout to a new cpu. callout_reset() schedules on the last assigned cpu while callout_reset_curcpu() schedules on the current cpu. Reviewed by: phk Sponsored by: Nokia --- sys/kern/kern_timeout.c | 370 ++++++++++++++++++++++++++++++++---------------- 1 file changed, 246 insertions(+), 124 deletions(-) (limited to 'sys/kern/kern_timeout.c') diff --git a/sys/kern/kern_timeout.c b/sys/kern/kern_timeout.c index 679430f..b187ab7 100644 --- a/sys/kern/kern_timeout.c +++ b/sys/kern/kern_timeout.c @@ -39,15 +39,19 @@ __FBSDID("$FreeBSD$"); #include #include +#include #include #include +#include #include #include #include +#include #include #include #include #include +#include static int avg_depth; SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0, @@ -65,36 +69,53 @@ SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0, * TODO: * allocate more timeout table slots when table overflows. */ - -/* Exported to machdep.c and/or kern_clock.c. */ -struct callout *callout; -struct callout_list callfree; int callwheelsize, callwheelbits, callwheelmask; -struct callout_tailq *callwheel; -int softticks; /* Like ticks, but for softclock(). */ -struct mtx callout_lock; -static struct callout *nextsoftcheck; /* Next callout to be checked. */ +struct callout_cpu { + struct mtx cc_lock; + struct callout *cc_callout; + struct callout_tailq *cc_callwheel; + struct callout_list cc_callfree; + struct callout *cc_next; + struct callout *cc_curr; + void *cc_cookie; + int cc_softticks; + int cc_cancel; + int cc_waiting; +}; + +#ifdef SMP +struct callout_cpu cc_cpu[MAXCPU]; +#define CC_CPU(cpu) (&cc_cpu[(cpu)]) +#define CC_SELF() CC_CPU(PCPU_GET(cpuid)) +#else +struct callout_cpu cc_cpu; +#define CC_CPU(cpu) &cc_cpu +#define CC_SELF() &cc_cpu +#endif +#define CC_LOCK(cc) mtx_lock_spin(&(cc)->cc_lock) +#define CC_UNLOCK(cc) mtx_unlock_spin(&(cc)->cc_lock) + +static int timeout_cpu; + +MALLOC_DEFINE(M_CALLOUT, "callout", "Callout datastructures"); /** - * Locked by callout_lock: - * curr_callout - If a callout is in progress, it is curr_callout. + * Locked by cc_lock: + * cc_curr - If a callout is in progress, it is curr_callout. * If curr_callout is non-NULL, threads waiting in - * callout_drain() will be woken up as soon as the + * callout_drain() will be woken up as soon as the * relevant callout completes. - * curr_cancelled - Changing to 1 with both callout_lock and c_lock held + * cc_cancel - Changing to 1 with both callout_lock and c_lock held * guarantees that the current callout will not run. * The softclock() function sets this to 0 before it * drops callout_lock to acquire c_lock, and it calls * the handler only if curr_cancelled is still 0 after * c_lock is successfully acquired. - * callout_wait - If a thread is waiting in callout_drain(), then + * cc_waiting - If a thread is waiting in callout_drain(), then * callout_wait is nonzero. Set only when * curr_callout is non-NULL. */ -static struct callout *curr_callout; -static int curr_cancelled; -static int callout_wait; /* * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization @@ -105,6 +126,10 @@ static int callout_wait; caddr_t kern_timeout_callwheel_alloc(caddr_t v) { + struct callout_cpu *cc; + + timeout_cpu = PCPU_GET(cpuid); + cc = CC_CPU(timeout_cpu); /* * Calculate callout wheel size */ @@ -114,13 +139,34 @@ kern_timeout_callwheel_alloc(caddr_t v) ; callwheelmask = callwheelsize - 1; - callout = (struct callout *)v; - v = (caddr_t)(callout + ncallout); - callwheel = (struct callout_tailq *)v; - v = (caddr_t)(callwheel + callwheelsize); + cc->cc_callout = (struct callout *)v; + v = (caddr_t)(cc->cc_callout + ncallout); + cc->cc_callwheel = (struct callout_tailq *)v; + v = (caddr_t)(cc->cc_callwheel + callwheelsize); return(v); } +static void +callout_cpu_init(struct callout_cpu *cc) +{ + struct callout *c; + int i; + + mtx_init(&cc->cc_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); + SLIST_INIT(&cc->cc_callfree); + for (i = 0; i < callwheelsize; i++) { + TAILQ_INIT(&cc->cc_callwheel[i]); + } + if (cc->cc_callout == NULL) + return; + for (i = 0; i < ncallout; i++) { + c = &cc->cc_callout[i]; + callout_init(c, 0); + c->c_flags = CALLOUT_LOCAL_ALLOC; + SLIST_INSERT_HEAD(&cc->cc_callfree, c, c_links.sle); + } +} + /* * kern_timeout_callwheel_init() - initialize previously reserved callwheel * space. @@ -131,18 +177,88 @@ kern_timeout_callwheel_alloc(caddr_t v) void kern_timeout_callwheel_init(void) { - int i; + callout_cpu_init(CC_CPU(timeout_cpu)); +} - SLIST_INIT(&callfree); - for (i = 0; i < ncallout; i++) { - callout_init(&callout[i], 0); - callout[i].c_flags = CALLOUT_LOCAL_ALLOC; - SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle); +/* + * Start standard softclock thread. + */ +void *softclock_ih; + +static void +start_softclock(void *dummy) +{ + struct callout_cpu *cc; +#ifdef SMP + int cpu; +#endif + + cc = CC_CPU(timeout_cpu); + if (swi_add(&clk_intr_event, "clock", softclock, cc, SWI_CLOCK, + INTR_MPSAFE, &softclock_ih)) + panic("died while creating standard software ithreads"); + cc->cc_cookie = softclock_ih; +#ifdef SMP + for (cpu = 0; cpu <= mp_maxid; cpu++) { + if (cpu == timeout_cpu) + continue; + if (CPU_ABSENT(cpu)) + continue; + cc = CC_CPU(cpu); + if (swi_add(NULL, "clock", softclock, cc, SWI_CLOCK, + INTR_MPSAFE, &cc->cc_cookie)) + panic("died while creating standard software ithreads"); + cc->cc_callout = NULL; /* Only cpu0 handles timeout(). */ + cc->cc_callwheel = malloc( + sizeof(struct callout_tailq) * callwheelsize, M_CALLOUT, + M_WAITOK); + callout_cpu_init(cc); } - for (i = 0; i < callwheelsize; i++) { - TAILQ_INIT(&callwheel[i]); +#endif +} + +SYSINIT(start_softclock, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softclock, NULL); + +void +callout_tick(void) +{ + int need_softclock = 0; + struct callout_cpu *cc; + + /* + * Process callouts at a very low cpu priority, so we don't keep the + * relatively high clock interrupt priority any longer than necessary. + */ + cc = CC_SELF(); + mtx_lock_spin_flags(&cc->cc_lock, MTX_QUIET); + if (!TAILQ_EMPTY(&cc->cc_callwheel[ticks & callwheelmask])) { + need_softclock = 1; + } else if (cc->cc_softticks + 1 == ticks) + ++cc->cc_softticks; + mtx_unlock_spin_flags(&cc->cc_lock, MTX_QUIET); + /* + * swi_sched acquires the thread lock, so we don't want to call it + * with cc_lock held; incorrect locking order. + */ + if (need_softclock) + swi_sched(cc->cc_cookie, 0); +} + +static struct callout_cpu * +callout_lock(struct callout *c) +{ + struct callout_cpu *cc; + int cpu; + + for (;;) { + cpu = c->c_cpu; + cc = CC_CPU(cpu); + CC_LOCK(cc); + if (cpu == c->c_cpu) + break; + CC_UNLOCK(cc); } - mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); + return (cc); } /* @@ -162,8 +278,9 @@ kern_timeout_callwheel_init(void) * Run periodic events from timeout queue. */ void -softclock(void *dummy) +softclock(void *arg) { + struct callout_cpu *cc; struct callout *c; struct callout_tailq *bucket; int curticks; @@ -188,15 +305,16 @@ softclock(void *dummy) gcalls = 0; depth = 0; steps = 0; - mtx_lock_spin(&callout_lock); - while (softticks != ticks) { - softticks++; + cc = (struct callout_cpu *)arg; + CC_LOCK(cc); + while (cc->cc_softticks != ticks) { + cc->cc_softticks++; /* - * softticks may be modified by hard clock, so cache + * cc_softticks may be modified by hard clock, so cache * it while we work on a given bucket. */ - curticks = softticks; - bucket = &callwheel[curticks & callwheelmask]; + curticks = cc->cc_softticks; + bucket = &cc->cc_callwheel[curticks & callwheelmask]; c = TAILQ_FIRST(bucket); while (c) { depth++; @@ -204,12 +322,12 @@ softclock(void *dummy) c = TAILQ_NEXT(c, c_links.tqe); ++steps; if (steps >= MAX_SOFTCLOCK_STEPS) { - nextsoftcheck = c; + cc->cc_next = c; /* Give interrupts a chance. */ - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); ; /* nothing */ - mtx_lock_spin(&callout_lock); - c = nextsoftcheck; + CC_LOCK(cc); + c = cc->cc_next; steps = 0; } } else { @@ -219,7 +337,7 @@ softclock(void *dummy) struct lock_object *c_lock; int c_flags, sharedlock; - nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); + cc->cc_next = TAILQ_NEXT(c, c_links.tqe); TAILQ_REMOVE(bucket, c, c_links.tqe); class = (c->c_lock != NULL) ? LOCK_CLASS(c->c_lock) : NULL; @@ -231,26 +349,25 @@ softclock(void *dummy) c_flags = c->c_flags; if (c->c_flags & CALLOUT_LOCAL_ALLOC) { c->c_flags = CALLOUT_LOCAL_ALLOC; - curr_callout = c; } else { c->c_flags = (c->c_flags & ~CALLOUT_PENDING); - curr_callout = c; } - curr_cancelled = 0; - mtx_unlock_spin(&callout_lock); + cc->cc_curr = c; + cc->cc_cancel = 0; + CC_UNLOCK(cc); if (c_lock != NULL) { class->lc_lock(c_lock, sharedlock); /* * The callout may have been cancelled * while we switched locks. */ - if (curr_cancelled) { + if (cc->cc_cancel) { class->lc_unlock(c_lock); goto skip; } /* The callout cannot be stopped now. */ - curr_cancelled = 1; + cc->cc_cancel = 1; if (c_lock == &Giant.lock_object) { gcalls++; @@ -295,7 +412,7 @@ softclock(void *dummy) if ((c_flags & CALLOUT_RETURNUNLOCKED) == 0) class->lc_unlock(c_lock); skip: - mtx_lock_spin(&callout_lock); + CC_LOCK(cc); /* * If the current callout is locally * allocated (from timeout(9)) @@ -311,22 +428,22 @@ softclock(void *dummy) CALLOUT_LOCAL_ALLOC, ("corrupted callout")); c->c_func = NULL; - SLIST_INSERT_HEAD(&callfree, c, + SLIST_INSERT_HEAD(&cc->cc_callfree, c, c_links.sle); } - curr_callout = NULL; - if (callout_wait) { + cc->cc_curr = NULL; + if (cc->cc_waiting) { /* * There is someone waiting * for the callout to complete. */ - callout_wait = 0; - mtx_unlock_spin(&callout_lock); - wakeup(&callout_wait); - mtx_lock_spin(&callout_lock); + cc->cc_waiting = 0; + CC_UNLOCK(cc); + wakeup(&cc->cc_waiting); + CC_LOCK(cc); } steps = 0; - c = nextsoftcheck; + c = cc->cc_next; } } } @@ -334,8 +451,8 @@ softclock(void *dummy) avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8; avg_lockcalls += (lockcalls * 1000 - avg_lockcalls) >> 8; avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8; - nextsoftcheck = NULL; - mtx_unlock_spin(&callout_lock); + cc->cc_next = NULL; + CC_UNLOCK(cc); } /* @@ -360,22 +477,22 @@ timeout(ftn, arg, to_ticks) void *arg; int to_ticks; { + struct callout_cpu *cc; struct callout *new; struct callout_handle handle; - mtx_lock_spin(&callout_lock); - + cc = CC_CPU(timeout_cpu); + CC_LOCK(cc); /* Fill in the next free callout structure. */ - new = SLIST_FIRST(&callfree); + new = SLIST_FIRST(&cc->cc_callfree); if (new == NULL) /* XXX Attempt to malloc first */ panic("timeout table full"); - SLIST_REMOVE_HEAD(&callfree, c_links.sle); - + SLIST_REMOVE_HEAD(&cc->cc_callfree, c_links.sle); callout_reset(new, to_ticks, ftn, arg); - handle.callout = new; - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); + return (handle); } @@ -385,6 +502,7 @@ untimeout(ftn, arg, handle) void *arg; struct callout_handle handle; { + struct callout_cpu *cc; /* * Check for a handle that was initialized @@ -394,10 +512,10 @@ untimeout(ftn, arg, handle) if (handle.callout == NULL) return; - mtx_lock_spin(&callout_lock); + cc = callout_lock(handle.callout); if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) callout_stop(handle.callout); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); } void @@ -423,24 +541,29 @@ callout_handle_init(struct callout_handle *handle) * callout_deactivate() - marks the callout as having been serviced */ int -callout_reset(c, to_ticks, ftn, arg) - struct callout *c; - int to_ticks; - void (*ftn)(void *); - void *arg; +callout_reset_on(struct callout *c, int to_ticks, void (*ftn)(void *), + void *arg, int cpu) { + struct callout_cpu *cc; int cancelled = 0; - mtx_lock_spin(&callout_lock); - if (c == curr_callout) { + /* + * Don't allow migration of pre-allocated callouts lest they + * become unbalanced. + */ + if (c->c_flags & CALLOUT_LOCAL_ALLOC) + cpu = c->c_cpu; +retry: + cc = callout_lock(c); + if (cc->cc_curr == c) { /* * We're being asked to reschedule a callout which is * currently in progress. If there is a lock then we * can cancel the callout if it has not really started. */ - if (c->c_lock != NULL && !curr_cancelled) - cancelled = curr_cancelled = 1; - if (callout_wait) { + if (c->c_lock != NULL && !cc->cc_cancel) + cancelled = cc->cc_cancel = 1; + if (cc->cc_waiting) { /* * Someone has called callout_drain to kill this * callout. Don't reschedule. @@ -448,34 +571,30 @@ callout_reset(c, to_ticks, ftn, arg) CTR4(KTR_CALLOUT, "%s %p func %p arg %p", cancelled ? "cancelled" : "failed to cancel", c, c->c_func, c->c_arg); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); return (cancelled); } } if (c->c_flags & CALLOUT_PENDING) { - if (nextsoftcheck == c) { - nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); + if (cc->cc_next == c) { + cc->cc_next = TAILQ_NEXT(c, c_links.tqe); } - TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, + TAILQ_REMOVE(&cc->cc_callwheel[c->c_time & callwheelmask], c, c_links.tqe); cancelled = 1; - - /* - * Part of the normal "stop a pending callout" process - * is to clear the CALLOUT_ACTIVE and CALLOUT_PENDING - * flags. We're not going to bother doing that here, - * because we're going to be setting those flags ten lines - * after this point, and we're holding callout_lock - * between now and then. - */ + c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); } - /* - * We could unlock callout_lock here and lock it again before the - * TAILQ_INSERT_TAIL, but there's no point since doing this setup - * doesn't take much time. + * If the lock must migrate we have to check the state again as + * we can't hold both the new and old locks simultaneously. */ + if (c->c_cpu != cpu) { + c->c_cpu = cpu; + CC_UNLOCK(cc); + goto retry; + } + if (to_ticks <= 0) to_ticks = 1; @@ -483,11 +602,11 @@ callout_reset(c, to_ticks, ftn, arg) c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); c->c_func = ftn; c->c_time = ticks + to_ticks; - TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask], + TAILQ_INSERT_TAIL(&cc->cc_callwheel[c->c_time & callwheelmask], c, c_links.tqe); CTR5(KTR_CALLOUT, "%sscheduled %p func %p arg %p in %d", cancelled ? "re" : "", c, c->c_func, c->c_arg, to_ticks); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); return (cancelled); } @@ -497,6 +616,7 @@ _callout_stop_safe(c, safe) struct callout *c; int safe; { + struct callout_cpu *cc; struct lock_class *class; int use_lock, sq_locked; @@ -517,7 +637,7 @@ _callout_stop_safe(c, safe) sq_locked = 0; again: - mtx_lock_spin(&callout_lock); + cc = callout_lock(c); /* * If the callout isn't pending, it's not on the queue, so * don't attempt to remove it from the queue. We can try to @@ -530,12 +650,12 @@ again: * If it wasn't on the queue and it isn't the current * callout, then we can't stop it, so just bail. */ - if (c != curr_callout) { + if (cc->cc_curr != c) { CTR3(KTR_CALLOUT, "failed to stop %p func %p arg %p", c, c->c_func, c->c_arg); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); if (sq_locked) - sleepq_release(&callout_wait); + sleepq_release(&cc->cc_curr); return (0); } @@ -546,19 +666,19 @@ again: * just wait for the current invocation to * finish. */ - while (c == curr_callout) { + while (cc->cc_curr == c) { /* * Use direct calls to sleepqueue interface * instead of cv/msleep in order to avoid - * a LOR between callout_lock and sleepqueue + * a LOR between cc_lock and sleepqueue * chain spinlocks. This piece of code * emulates a msleep_spin() call actually. * * If we already have the sleepqueue chain * locked, then we can safely block. If we * don't already have it locked, however, - * we have to drop the callout_lock to lock + * we have to drop the cc_lock to lock * it. This opens several races, so we * restart at the beginning once we have * both locks. If nothing has changed, then @@ -566,26 +686,25 @@ again: * set. */ if (!sq_locked) { - mtx_unlock_spin(&callout_lock); - sleepq_lock(&callout_wait); + CC_UNLOCK(cc); + sleepq_lock(&cc->cc_curr); sq_locked = 1; goto again; } - - callout_wait = 1; + cc->cc_waiting = 1; DROP_GIANT(); - mtx_unlock_spin(&callout_lock); - sleepq_add(&callout_wait, - &callout_lock.lock_object, "codrain", + CC_UNLOCK(cc); + sleepq_add(&cc->cc_curr, + &cc->cc_lock.lock_object, "codrain", SLEEPQ_SLEEP, 0); - sleepq_wait(&callout_wait, 0); + sleepq_wait(&cc->cc_curr, 0); sq_locked = 0; /* Reacquire locks previously released. */ PICKUP_GIANT(); - mtx_lock_spin(&callout_lock); + CC_LOCK(cc); } - } else if (use_lock && !curr_cancelled) { + } else if (use_lock && !cc->cc_cancel) { /* * The current callout is waiting for its * lock which we hold. Cancel the callout @@ -593,37 +712,38 @@ again: * lock, the callout will be skipped in * softclock(). */ - curr_cancelled = 1; + cc->cc_cancel = 1; CTR3(KTR_CALLOUT, "cancelled %p func %p arg %p", c, c->c_func, c->c_arg); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); KASSERT(!sq_locked, ("sleepqueue chain locked")); return (1); } CTR3(KTR_CALLOUT, "failed to stop %p func %p arg %p", c, c->c_func, c->c_arg); - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); KASSERT(!sq_locked, ("sleepqueue chain still locked")); return (0); } if (sq_locked) - sleepq_release(&callout_wait); + sleepq_release(&cc->cc_curr); c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); - if (nextsoftcheck == c) { - nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); + if (cc->cc_next == c) { + cc->cc_next = TAILQ_NEXT(c, c_links.tqe); } - TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe); + TAILQ_REMOVE(&cc->cc_callwheel[c->c_time & callwheelmask], c, + c_links.tqe); CTR3(KTR_CALLOUT, "cancelled %p func %p arg %p", c, c->c_func, c->c_arg); if (c->c_flags & CALLOUT_LOCAL_ALLOC) { c->c_func = NULL; - SLIST_INSERT_HEAD(&callfree, c, c_links.sle); + SLIST_INSERT_HEAD(&cc->cc_callfree, c, c_links.sle); } - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); return (1); } @@ -640,6 +760,7 @@ callout_init(c, mpsafe) c->c_lock = &Giant.lock_object; c->c_flags = 0; } + c->c_cpu = timeout_cpu; } void @@ -658,6 +779,7 @@ _callout_init_lock(c, lock, flags) (LC_SPINLOCK | LC_SLEEPABLE)), ("%s: invalid lock class", __func__)); c->c_flags = flags & (CALLOUT_RETURNUNLOCKED | CALLOUT_SHAREDLOCK); + c->c_cpu = timeout_cpu; } #ifdef APM_FIXUP_CALLTODO @@ -709,7 +831,7 @@ adjust_timeout_calltodo(time_change) */ /* don't collide with softclock() */ - mtx_lock_spin(&callout_lock); + CC_LOCK(cc); for (p = calltodo.c_next; p != NULL; p = p->c_next) { p->c_time -= delta_ticks; @@ -720,7 +842,7 @@ adjust_timeout_calltodo(time_change) /* take back the ticks the timer didn't use (p->c_time <= 0) */ delta_ticks = -p->c_time; } - mtx_unlock_spin(&callout_lock); + CC_UNLOCK(cc); return; } -- cgit v1.1