Index: sys/amd64/include/vmparam.h =================================================================== --- sys/amd64/include/vmparam.h +++ sys/amd64/include/vmparam.h @@ -227,4 +227,10 @@ #define ZERO_REGION_SIZE (2 * 1024 * 1024) /* 2MB */ +/* + * Use a fairly large batch size since we expect amd64 systems to have lots of + * memory. + */ +#define VM_BATCHQUEUE_SIZE 31 + #endif /* _MACHINE_VMPARAM_H_ */ Index: sys/kern/subr_witness.c =================================================================== --- sys/kern/subr_witness.c +++ sys/kern/subr_witness.c @@ -603,7 +603,6 @@ * CDEV */ { "vm map (system)", &lock_class_mtx_sleep }, - { "vm pagequeue", &lock_class_mtx_sleep }, { "vnode interlock", &lock_class_mtx_sleep }, { "cdev", &lock_class_mtx_sleep }, { NULL, NULL }, @@ -613,11 +612,11 @@ { "vm map (user)", &lock_class_sx }, { "vm object", &lock_class_rw }, { "vm page", &lock_class_mtx_sleep }, - { "vm pagequeue", &lock_class_mtx_sleep }, { "pmap pv global", &lock_class_rw }, { "pmap", &lock_class_mtx_sleep }, { "pmap pv list", &lock_class_rw }, { "vm page free queue", &lock_class_mtx_sleep }, + { "vm pagequeue", &lock_class_mtx_sleep }, { NULL, NULL }, /* * kqueue/VFS interaction Index: sys/vm/vm_object.c =================================================================== --- sys/vm/vm_object.c +++ sys/vm/vm_object.c @@ -720,14 +720,11 @@ vm_object_terminate_pages(vm_object_t object) { vm_page_t p, p_next; - struct mtx *mtx, *mtx1; - struct vm_pagequeue *pq, *pq1; - int dequeued; + struct mtx *mtx; VM_OBJECT_ASSERT_WLOCKED(object); mtx = NULL; - pq = NULL; /* * Free any remaining pageable pages. This also removes them from the @@ -737,60 +734,21 @@ */ TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { vm_page_assert_unbusied(p); - if ((object->flags & OBJ_UNMANAGED) == 0) { + if ((object->flags & OBJ_UNMANAGED) == 0) /* * vm_page_free_prep() only needs the page * lock for managed pages. */ - mtx1 = vm_page_lockptr(p); - if (mtx1 != mtx) { - if (mtx != NULL) - mtx_unlock(mtx); - if (pq != NULL) { - vm_pagequeue_cnt_add(pq, dequeued); - vm_pagequeue_unlock(pq); - pq = NULL; - } - mtx = mtx1; - mtx_lock(mtx); - } - } + vm_page_change_lock(p, &mtx); p->object = NULL; if (p->wire_count != 0) - goto unlist; - VM_CNT_INC(v_pfree); - p->flags &= ~PG_ZERO; - if (p->queue != PQ_NONE) { - KASSERT(p->queue < PQ_COUNT, ("vm_object_terminate: " - "page %p is not queued", p)); - pq1 = vm_page_pagequeue(p); - if (pq != pq1) { - if (pq != NULL) { - vm_pagequeue_cnt_add(pq, dequeued); - vm_pagequeue_unlock(pq); - } - pq = pq1; - vm_pagequeue_lock(pq); - dequeued = 0; - } - p->queue = PQ_NONE; - TAILQ_REMOVE(&pq->pq_pl, p, plinks.q); - dequeued--; - } - if (vm_page_free_prep(p, true)) continue; -unlist: - TAILQ_REMOVE(&object->memq, p, listq); - } - if (pq != NULL) { - vm_pagequeue_cnt_add(pq, dequeued); - vm_pagequeue_unlock(pq); + VM_CNT_INC(v_pfree); + vm_page_free(p); } if (mtx != NULL) mtx_unlock(mtx); - vm_page_free_phys_pglist(&object->memq); - /* * If the object contained any pages, then reset it to an empty state. * None of the object's fields, including "resident_page_count", were @@ -1973,7 +1931,6 @@ { vm_page_t p, next; struct mtx *mtx; - struct pglist pgl; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_UNMANAGED) == 0 || @@ -1982,7 +1939,6 @@ if (object->resident_page_count == 0) return; vm_object_pip_add(object, 1); - TAILQ_INIT(&pgl); again: p = vm_page_find_least(object, start); mtx = NULL; @@ -2036,13 +1992,10 @@ } if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_all(p); - p->flags &= ~PG_ZERO; - if (vm_page_free_prep(p, false)) - TAILQ_INSERT_TAIL(&pgl, p, listq); + vm_page_free(p); } if (mtx != NULL) mtx_unlock(mtx); - vm_page_free_phys_pglist(&pgl); vm_object_pip_wakeup(object); } Index: sys/vm/vm_page.h =================================================================== --- sys/vm/vm_page.h +++ sys/vm/vm_page.h @@ -93,8 +93,11 @@ * * In general, operations on this structure's mutable fields are * synchronized using either one of or a combination of the lock on the - * object that the page belongs to (O), the pool lock for the page (P), - * or the lock for either the free or paging queue (Q). If a field is + * object that the page belongs to (O), the page lock (P), + * the per-domain lock for the free queues (F), or the page's queue + * lock (Q). The physical address of a page is used to select its page + * lock from a pool. The queue lock for a page depends on the value of + * its queue field and described in detail below. If a field is * annotated below with two of these locks, then holding either lock is * sufficient for read access, but both locks are required for write * access. An annotation of (C) indicates that the field is immutable. @@ -143,6 +146,29 @@ * causing the thread to block. vm_page_sleep_if_busy() can be used to * sleep until the page's busy state changes, after which the caller * must re-lookup the page and re-evaluate its state. + * + * The queue field is the index of the page queue containing the + * page, or PQ_NONE if the page is not enqueued. The queue lock of a + * page is the page queue lock corresponding to the page queue index, + * or the page lock (P) for the page if it is not enqueued. To modify + * the queue field, the queue lock for the old value of the field must + * be held. It is invalid for a page's queue field to transition + * between two distinct page queue indices. That is, when updating + * the queue field, either the new value or the old value must be + * PQ_NONE. + * + * To avoid contention on page queue locks, page queue operations + * (enqueue, dequeue, requeue) are batched using per-CPU queues. + * A deferred operation is requested by inserting an entry into a + * batch queue; the entry is simply a pointer to the page, and the + * request type is encoded in the page's aflags field using the values + * in PGA_QUEUE_STATE_MASK. The type-stability of struct vm_pages is + * crucial to this scheme since the processing of entries in a given + * batch queue may be deferred indefinitely. In particular, a page + * may be freed before its pending batch queue entries have been + * processed. The page lock (P) must be held to schedule a batched + * queue operation, and the page queue lock must be held in order to + * process batch queue entries for the page queue. */ #if PAGE_SIZE == 4096 @@ -174,7 +200,7 @@ TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ vm_object_t object; /* which object am I in (O,P) */ vm_pindex_t pindex; /* offset into object (O,P) */ - vm_paddr_t phys_addr; /* physical address of page */ + vm_paddr_t phys_addr; /* physical address of page (C) */ struct md_page md; /* machine dependent stuff */ u_int wire_count; /* wired down maps refs (P) */ volatile u_int busy_lock; /* busy owners lock */ @@ -182,11 +208,11 @@ uint16_t flags; /* page PG_* flags (P) */ uint8_t aflags; /* access is atomic */ uint8_t oflags; /* page VPO_* flags (O) */ - uint8_t queue; /* page queue index (P,Q) */ + uint8_t queue; /* page queue index (Q) */ int8_t psind; /* pagesizes[] index (O) */ int8_t segind; /* vm_phys segment index (C) */ - uint8_t order; /* index of the buddy queue */ - uint8_t pool; /* vm_phys freepool index (Q) */ + uint8_t order; /* index of the buddy queue (F) */ + uint8_t pool; /* vm_phys freepool index (F) */ u_char act_count; /* page usage count (P) */ /* NOTE that these must support one bit per DEV_BSIZE in a page */ /* so, on normal X86 kernels, they must be at least 8 bits wide */ @@ -314,10 +340,38 @@ * * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has * at least one executable mapping. It is not consumed by the MI VM layer. + * + * PGA_ENQUEUED is set and cleared when a page is inserted into or removed + * from a page queue, respectively. It determines whether the plinks.q field + * of the page is valid. To set or clear this flag, the queue lock for the + * page must be held: the page queue lock corresponding to the page's "queue" + * field if its value is not PQ_NONE, and the page lock otherwise. + * + * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page + * queue, and cleared when the dequeue request is processed. A page may + * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue + * is requested after the page is scheduled to be enqueued but before it is + * actually inserted into the page queue. The page lock must be held to set + * this flag, and the queue lock for the page must be held to clear it. + * + * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued + * in its page queue. The page lock must be held to set this flag, and the + * queue lock for the page must be held to clear it. + * + * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of + * the inactive queue, thus bypassing LRU. The page lock must be held to + * set this flag, and the queue lock for the page must be held to clear it. */ #define PGA_WRITEABLE 0x01 /* page may be mapped writeable */ #define PGA_REFERENCED 0x02 /* page has been referenced */ #define PGA_EXECUTABLE 0x04 /* page may be mapped executable */ +#define PGA_ENQUEUED 0x08 /* page is enqueued in a page queue */ +#define PGA_DEQUEUE 0x10 /* page is due to be dequeued */ +#define PGA_REQUEUE 0x20 /* page is due to be requeued */ +#define PGA_REQUEUE_HEAD 0x40 /* page requeue should bypass LRU */ + +#define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \ + PGA_REQUEUE_HEAD) /* * Page flags. If changed at any other time than page allocation or @@ -484,13 +538,13 @@ void vm_page_deactivate(vm_page_t); void vm_page_deactivate_noreuse(vm_page_t); void vm_page_dequeue(vm_page_t m); +void vm_page_dequeue_deferred(vm_page_t m); void vm_page_dequeue_locked(vm_page_t m); +void vm_page_drain_pqbatch(void); vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t); -void vm_page_free_phys_pglist(struct pglist *tq); -bool vm_page_free_prep(vm_page_t m, bool pagequeue_locked); +bool vm_page_free_prep(vm_page_t m); vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr); void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); -void vm_page_init_marker(vm_page_t m, int queue); int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); void vm_page_launder(vm_page_t m); vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t); @@ -752,6 +806,24 @@ return (m->queue == PQ_LAUNDRY || m->queue == PQ_UNSWAPPABLE); } +/* + * vm_page_enqueued: + * + * Return true if the page is logically enqueued and no deferred + * dequeue is pending. + */ +static inline bool +vm_page_enqueued(vm_page_t m) +{ + + vm_page_assert_locked(m); + + if ((m->aflags & PGA_DEQUEUE) != 0) + return (false); + atomic_thread_fence_acq(); + return (m->queue != PQ_NONE); +} + /* * vm_page_held: * Index: sys/vm/vm_page.c =================================================================== --- sys/vm/vm_page.c +++ sys/vm/vm_page.c @@ -102,6 +102,7 @@ #include #include #include +#include #include #include #include @@ -131,13 +132,10 @@ extern void uma_startup(void *, int); extern int vmem_startup_count(void); -/* - * Associated with page of user-allocatable memory is a - * page structure. - */ - struct vm_domain vm_dom[MAXMEMDOM]; +static DPCPU_DEFINE(struct vm_batchqueue, pqbatch[MAXMEMDOM][PQ_COUNT]); + struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT]; struct mtx_padalign __exclusive_cache_line vm_domainset_lock; @@ -176,7 +174,8 @@ static void vm_page_alloc_check(vm_page_t m); static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits); -static void vm_page_enqueue(uint8_t queue, vm_page_t m); +static void vm_page_dequeue_complete(vm_page_t m); +static void vm_page_enqueue(vm_page_t m, uint8_t queue); static void vm_page_init(void *dummy); static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred); @@ -443,12 +442,13 @@ * Nonetheless, it write busies and initializes the hold count to one as * safety precautions. */ -void -vm_page_init_marker(vm_page_t marker, int queue) +static void +vm_page_init_marker(vm_page_t marker, int queue, uint8_t aflags) { bzero(marker, sizeof(*marker)); marker->flags = PG_MARKER; + marker->aflags = aflags; marker->busy_lock = VPB_SINGLE_EXCLUSIVER; marker->queue = queue; marker->hold_count = 1; @@ -481,14 +481,32 @@ TAILQ_INIT(&pq->pq_pl); mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue", MTX_DEF | MTX_DUPOK); - vm_page_init_marker(&vmd->vmd_markers[i], i); + vm_page_init_marker(&vmd->vmd_markers[i], i, 0); } mtx_init(&vmd->vmd_free_mtx, "vm page free queue", NULL, MTX_DEF); mtx_init(&vmd->vmd_pageout_mtx, "vm pageout lock", NULL, MTX_DEF); - vm_page_init_marker(&vmd->vmd_inacthead, PQ_INACTIVE); + snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain); + + /* + * inacthead is used to provide FIFO ordering for LRU-bypassing + * insertions. + */ + vm_page_init_marker(&vmd->vmd_inacthead, PQ_INACTIVE, PGA_ENQUEUED); TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_INACTIVE].pq_pl, &vmd->vmd_inacthead, plinks.q); - snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain); + + /* + * The clock pages are used to implement active queue scanning without + * requeues. Scans start at clock[0], which is advanced after the scan + * ends. When the two clock hands meet, they are reset and scanning + * resumes from the head of the queue. + */ + vm_page_init_marker(&vmd->vmd_clock[0], PQ_ACTIVE, PGA_ENQUEUED); + vm_page_init_marker(&vmd->vmd_clock[1], PQ_ACTIVE, PGA_ENQUEUED); + TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl, + &vmd->vmd_clock[0], plinks.q); + TAILQ_INSERT_TAIL(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl, + &vmd->vmd_clock[1], plinks.q); } /* @@ -1847,6 +1865,7 @@ KASSERT(m != NULL, ("missing page")); found: + vm_page_dequeue(m); vm_page_alloc_check(m); /* @@ -2043,8 +2062,10 @@ #if VM_NRESERVLEVEL > 0 found: #endif - for (m = m_ret; m < &m_ret[npages]; m++) + for (m = m_ret; m < &m_ret[npages]; m++) { + vm_page_dequeue(m); vm_page_alloc_check(m); + } /* * Initialize the pages. Only the PG_ZERO flag is inherited. @@ -2188,6 +2209,7 @@ goto again; return (NULL); } + vm_page_dequeue(m); vm_page_alloc_check(m); /* @@ -2381,7 +2403,7 @@ vm_reserv_size(level)) - pa); #endif } else if (object->memattr == VM_MEMATTR_DEFAULT && - m->queue != PQ_NONE && !vm_page_busied(m)) { + vm_page_enqueued(m) && !vm_page_busied(m)) { /* * The page is allocated but eligible for * relocation. Extend the current run by one @@ -2532,7 +2554,7 @@ error = EINVAL; else if (object->memattr != VM_MEMATTR_DEFAULT) error = EINVAL; - else if (m->queue != PQ_NONE && !vm_page_busied(m)) { + else if (vm_page_enqueued(m) && !vm_page_busied(m)) { KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has an unexpected memattr", m)); @@ -2592,7 +2614,8 @@ */ if (object->ref_count != 0) pmap_remove_all(m); - m_new->aflags = m->aflags; + m_new->aflags = m->aflags & + ~PGA_QUEUE_STATE_MASK; KASSERT(m_new->oflags == VPO_UNMANAGED, ("page %p is managed", m_new)); m_new->oflags = m->oflags & VPO_NOSYNC; @@ -2604,7 +2627,7 @@ vm_page_remque(m); vm_page_replace_checked(m_new, object, m->pindex, m); - if (vm_page_free_prep(m, false)) + if (vm_page_free_prep(m)) SLIST_INSERT_HEAD(&free, m, plinks.s.ss); @@ -2618,7 +2641,7 @@ m->flags &= ~PG_ZERO; vm_page_remque(m); vm_page_remove(m); - if (vm_page_free_prep(m, false)) + if (vm_page_free_prep(m)) SLIST_INSERT_HEAD(&free, m, plinks.s.ss); KASSERT(m->dirty == 0, @@ -3061,113 +3084,297 @@ return (&vm_pagequeue_domain(m)->vmd_pagequeues[m->queue]); } +static struct mtx * +vm_page_pagequeue_lockptr(vm_page_t m) +{ + + if (m->queue == PQ_NONE) + return (NULL); + return (&vm_page_pagequeue(m)->pq_mutex); +} + +static inline void +vm_pqbatch_process_page(struct vm_pagequeue *pq, vm_page_t m) +{ + struct vm_domain *vmd; + uint8_t aflags; + + vm_pagequeue_assert_locked(pq); + KASSERT(pq == vm_page_pagequeue(m), + ("page %p doesn't belong to %p", m, pq)); + + aflags = m->aflags; + if ((aflags & PGA_DEQUEUE) != 0) { + if (__predict_true((aflags & PGA_ENQUEUED) != 0)) { + TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); + vm_pagequeue_cnt_dec(pq); + } + vm_page_dequeue_complete(m); + } else if ((aflags & (PGA_REQUEUE | PGA_REQUEUE_HEAD)) != 0) { + if ((aflags & PGA_ENQUEUED) != 0) + TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); + else { + vm_pagequeue_cnt_inc(pq); + vm_page_aflag_set(m, PGA_ENQUEUED); + } + if ((aflags & PGA_REQUEUE_HEAD) != 0) { + KASSERT(m->queue == PQ_INACTIVE, + ("head enqueue not supported for page %p", m)); + vmd = vm_pagequeue_domain(m); + TAILQ_INSERT_BEFORE(&vmd->vmd_inacthead, m, plinks.q); + } else + TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); + + /* + * PGA_REQUEUE and PGA_REQUEUE_HEAD must be cleared after + * setting PGA_ENQUEUED in order to synchronize with the + * page daemon. + */ + vm_page_aflag_clear(m, PGA_REQUEUE | PGA_REQUEUE_HEAD); + } +} + +static void +vm_pqbatch_process(struct vm_pagequeue *pq, struct vm_batchqueue *bq, + uint8_t queue) +{ + vm_page_t m; + int i; + + for (i = 0; i < bq->bq_cnt; i++) { + m = bq->bq_pa[i]; + if (__predict_false(m->queue != queue)) + continue; + vm_pqbatch_process_page(pq, m); + } + vm_batchqueue_init(bq); +} + +static void +vm_pqbatch_submit_page(vm_page_t m, uint8_t queue) +{ + struct vm_batchqueue *bq; + struct vm_pagequeue *pq; + int domain; + + vm_page_assert_locked(m); + KASSERT(queue < PQ_COUNT, ("invalid queue %d", queue)); + + domain = vm_phys_domain(m); + pq = &vm_pagequeue_domain(m)->vmd_pagequeues[queue]; + + critical_enter(); + bq = DPCPU_PTR(pqbatch[domain][queue]); + if (vm_batchqueue_insert(bq, m)) { + critical_exit(); + return; + } + if (!vm_pagequeue_trylock(pq)) { + critical_exit(); + vm_pagequeue_lock(pq); + critical_enter(); + bq = DPCPU_PTR(pqbatch[domain][queue]); + } + vm_pqbatch_process(pq, bq, queue); + + /* + * The page may have been logically dequeued before we acquired the + * page queue lock. In this case, the page lock prevents the page + * from being logically enqueued elsewhere. + */ + if (__predict_true(m->queue == queue)) + vm_pqbatch_process_page(pq, m); + else { + KASSERT(m->queue == PQ_NONE, + ("invalid queue transition for page %p", m)); + KASSERT((m->aflags & PGA_ENQUEUED) == 0, + ("page %p is enqueued with invalid queue index", m)); + vm_page_aflag_clear(m, PGA_QUEUE_STATE_MASK); + } + vm_pagequeue_unlock(pq); + critical_exit(); +} + /* - * vm_page_dequeue: + * vm_page_drain_pqbatch: [ internal use only ] + * + * Force all per-CPU page queue batch queues to be drained. This is + * intended for use in severe memory shortages, to ensure that pages + * do not remain stuck in the batch queues. + */ +void +vm_page_drain_pqbatch(void) +{ + struct thread *td; + struct vm_domain *vmd; + struct vm_pagequeue *pq; + int cpu, domain, queue; + + td = curthread; + CPU_FOREACH(cpu) { + thread_lock(td); + sched_bind(td, cpu); + thread_unlock(td); + + for (domain = 0; domain < vm_ndomains; domain++) { + vmd = VM_DOMAIN(domain); + for (queue = 0; queue < PQ_COUNT; queue++) { + pq = &vmd->vmd_pagequeues[queue]; + vm_pagequeue_lock(pq); + critical_enter(); + vm_pqbatch_process(pq, + DPCPU_PTR(pqbatch[domain][queue]), queue); + critical_exit(); + vm_pagequeue_unlock(pq); + } + } + } + thread_lock(td); + sched_unbind(td); + thread_unlock(td); +} + +/* + * Complete the logical removal of a page from a page queue. We must be + * careful to synchronize with the page daemon, which may be concurrently + * examining the page with only the page lock held. The page must not be + * in a state where it appears to be logically enqueued. + */ +static void +vm_page_dequeue_complete(vm_page_t m) +{ + + m->queue = PQ_NONE; + atomic_thread_fence_rel(); + vm_page_aflag_clear(m, PGA_QUEUE_STATE_MASK); +} + +/* + * vm_page_dequeue_deferred: [ internal use only ] * - * Remove the given page from its current page queue. + * Request removal of the given page from its current page + * queue. Physical removal from the queue may be deferred + * indefinitely. * * The page must be locked. */ void -vm_page_dequeue(vm_page_t m) +vm_page_dequeue_deferred(vm_page_t m) { - struct vm_pagequeue *pq; + int queue; vm_page_assert_locked(m); - KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued", - m)); - pq = vm_page_pagequeue(m); - vm_pagequeue_lock(pq); - m->queue = PQ_NONE; - TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); - vm_pagequeue_cnt_dec(pq); - vm_pagequeue_unlock(pq); + + queue = m->queue; + if (queue == PQ_NONE) { + KASSERT((m->aflags & PGA_QUEUE_STATE_MASK) == 0, + ("page %p has queue state", m)); + return; + } + if ((m->aflags & PGA_DEQUEUE) == 0) + vm_page_aflag_set(m, PGA_DEQUEUE); + vm_pqbatch_submit_page(m, queue); } /* * vm_page_dequeue_locked: * - * Remove the given page from its current page queue. + * Remove the page from its page queue, which must be locked. + * If the page lock is not held, there is no guarantee that the + * page will not be enqueued by another thread before this function + * returns. In this case, it is up to the caller to ensure that + * no other threads hold a reference to the page. * - * The page and page queue must be locked. + * The page queue lock must be held. If the page is not already + * logically dequeued, the page lock must be held as well. */ void vm_page_dequeue_locked(vm_page_t m) { struct vm_pagequeue *pq; - vm_page_lock_assert(m, MA_OWNED); pq = vm_page_pagequeue(m); + + KASSERT(m->queue != PQ_NONE, + ("%s: page %p queue field is PQ_NONE", __func__, m)); vm_pagequeue_assert_locked(pq); - m->queue = PQ_NONE; - TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); - vm_pagequeue_cnt_dec(pq); + KASSERT((m->aflags & PGA_DEQUEUE) != 0 || + mtx_owned(vm_page_lockptr(m)), + ("%s: queued unlocked page %p", __func__, m)); + + if ((m->aflags & PGA_ENQUEUED) != 0) { + TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); + vm_pagequeue_cnt_dec(pq); + } + vm_page_dequeue_complete(m); } /* - * vm_page_enqueue: - * - * Add the given page to the specified page queue. + * vm_page_dequeue: * - * The page must be locked. + * Remove the page from whichever page queue it's in, if any. + * If the page lock is not held, there is no guarantee that the + * page will not be enqueued by another thread before this function + * returns. In this case, it is up to the caller to ensure that + * no other threads hold a reference to the page. */ -static void -vm_page_enqueue(uint8_t queue, vm_page_t m) +void +vm_page_dequeue(vm_page_t m) { - struct vm_pagequeue *pq; + struct mtx *lock, *lock1; - vm_page_lock_assert(m, MA_OWNED); - KASSERT(queue < PQ_COUNT, - ("vm_page_enqueue: invalid queue %u request for page %p", - queue, m)); - pq = &vm_pagequeue_domain(m)->vmd_pagequeues[queue]; - vm_pagequeue_lock(pq); - m->queue = queue; - TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); - vm_pagequeue_cnt_inc(pq); - vm_pagequeue_unlock(pq); + lock = vm_page_pagequeue_lockptr(m); + for (;;) { + if (lock == NULL) + return; + mtx_lock(lock); + if ((lock1 = vm_page_pagequeue_lockptr(m)) == lock) + break; + mtx_unlock(lock); + lock = lock1; + } + KASSERT(lock == vm_page_pagequeue_lockptr(m), + ("%s: page %p migrated directly between queues", __func__, m)); + vm_page_dequeue_locked(m); + mtx_unlock(lock); } /* - * vm_page_requeue: - * - * Move the given page to the tail of its current page queue. - * - * The page must be locked. + * Schedule the given page for insertion into the specified page queue. + * Physical insertion of the page may be deferred indefinitely. */ -void -vm_page_requeue(vm_page_t m) +static void +vm_page_enqueue(vm_page_t m, uint8_t queue) { - struct vm_pagequeue *pq; - vm_page_lock_assert(m, MA_OWNED); - KASSERT(m->queue != PQ_NONE, - ("vm_page_requeue: page %p is not queued", m)); - pq = vm_page_pagequeue(m); - vm_pagequeue_lock(pq); - TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); - TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); - vm_pagequeue_unlock(pq); + vm_page_assert_locked(m); + KASSERT(m->queue == PQ_NONE && (m->aflags & PGA_QUEUE_STATE_MASK) == 0, + ("%s: page %p is already enqueued", __func__, m)); + + m->queue = queue; + if ((m->aflags & PGA_REQUEUE) == 0) + vm_page_aflag_set(m, PGA_REQUEUE); + vm_pqbatch_submit_page(m, queue); } /* - * vm_page_requeue_locked: + * vm_page_requeue: [ internal use only ] * - * Move the given page to the tail of its current page queue. + * Schedule a requeue of the given page. * - * The page queue must be locked. + * The page must be locked. */ void -vm_page_requeue_locked(vm_page_t m) +vm_page_requeue(vm_page_t m) { - struct vm_pagequeue *pq; + vm_page_assert_locked(m); KASSERT(m->queue != PQ_NONE, - ("vm_page_requeue_locked: page %p is not queued", m)); - pq = vm_page_pagequeue(m); - vm_pagequeue_assert_locked(pq); - TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); - TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); + ("%s: page %p is not logically enqueued", __func__, m)); + + if ((m->aflags & PGA_REQUEUE) == 0) + vm_page_aflag_set(m, PGA_REQUEUE); + vm_pqbatch_submit_page(m, m->queue); } /* @@ -3185,18 +3392,18 @@ int queue; vm_page_lock_assert(m, MA_OWNED); - if ((queue = m->queue) != PQ_ACTIVE) { - if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { - if (m->act_count < ACT_INIT) - m->act_count = ACT_INIT; - if (queue != PQ_NONE) - vm_page_dequeue(m); - vm_page_enqueue(PQ_ACTIVE, m); - } - } else { - if (m->act_count < ACT_INIT) + + if ((queue = m->queue) == PQ_ACTIVE || m->wire_count > 0 || + (m->oflags & VPO_UNMANAGED) != 0) { + if (queue == PQ_ACTIVE && m->act_count < ACT_INIT) m->act_count = ACT_INIT; + return; } + + vm_page_remque(m); + if (m->act_count < ACT_INIT) + m->act_count = ACT_INIT; + vm_page_enqueue(m, PQ_ACTIVE); } /* @@ -3207,11 +3414,10 @@ * the page to the free list only if this function returns true. * * The object must be locked. The page must be locked if it is - * managed. For a queued managed page, the pagequeue_locked - * argument specifies whether the page queue is already locked. + * managed. */ bool -vm_page_free_prep(vm_page_t m, bool pagequeue_locked) +vm_page_free_prep(vm_page_t m) { #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP) @@ -3227,14 +3433,14 @@ if ((m->oflags & VPO_UNMANAGED) == 0) { vm_page_lock_assert(m, MA_OWNED); KASSERT(!pmap_page_is_mapped(m), - ("vm_page_free_toq: freeing mapped page %p", m)); + ("vm_page_free_prep: freeing mapped page %p", m)); } else KASSERT(m->queue == PQ_NONE, - ("vm_page_free_toq: unmanaged page %p is queued", m)); + ("vm_page_free_prep: unmanaged page %p is queued", m)); VM_CNT_INC(v_tfree); if (vm_page_sbusied(m)) - panic("vm_page_free: freeing busy page %p", m); + panic("vm_page_free_prep: freeing busy page %p", m); vm_page_remove(m); @@ -3250,21 +3456,23 @@ return (false); } - if (m->queue != PQ_NONE) { - if (pagequeue_locked) - vm_page_dequeue_locked(m); - else - vm_page_dequeue(m); - } + /* + * Pages need not be dequeued before they are returned to the physical + * memory allocator, but they must at least be marked for a deferred + * dequeue. + */ + if ((m->oflags & VPO_UNMANAGED) == 0) + vm_page_dequeue_deferred(m); + m->valid = 0; vm_page_undirty(m); if (m->wire_count != 0) - panic("vm_page_free: freeing wired page %p", m); + panic("vm_page_free_prep: freeing wired page %p", m); if (m->hold_count != 0) { m->flags &= ~PG_ZERO; KASSERT((m->flags & PG_UNHOLDFREE) == 0, - ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); + ("vm_page_free_prep: freeing PG_UNHOLDFREE page %p", m)); m->flags |= PG_UNHOLDFREE; return (false); } @@ -3283,36 +3491,6 @@ return (true); } -void -vm_page_free_phys_pglist(struct pglist *tq) -{ - struct vm_domain *vmd; - vm_page_t m; - int cnt; - - if (TAILQ_EMPTY(tq)) - return; - vmd = NULL; - cnt = 0; - TAILQ_FOREACH(m, tq, listq) { - if (vmd != vm_pagequeue_domain(m)) { - if (vmd != NULL) { - vm_domain_free_unlock(vmd); - vm_domain_freecnt_inc(vmd, cnt); - cnt = 0; - } - vmd = vm_pagequeue_domain(m); - vm_domain_free_lock(vmd); - } - vm_phys_free_pages(m, 0); - cnt++; - } - if (vmd != NULL) { - vm_domain_free_unlock(vmd); - vm_domain_freecnt_inc(vmd, cnt); - } -} - /* * vm_page_free_toq: * @@ -3327,7 +3505,7 @@ { struct vm_domain *vmd; - if (!vm_page_free_prep(m, false)) + if (!vm_page_free_prep(m)) return; vmd = vm_pagequeue_domain(m); @@ -3425,22 +3603,25 @@ KASSERT(queue < PQ_COUNT || queue == PQ_NONE, ("vm_page_unwire: invalid queue %u request for page %p", queue, m)); + if ((m->oflags & VPO_UNMANAGED) == 0) + vm_page_assert_locked(m); unwired = vm_page_unwire_noq(m); - if (unwired && (m->oflags & VPO_UNMANAGED) == 0 && m->object != NULL) { - if (m->queue == queue) { + if (!unwired || (m->oflags & VPO_UNMANAGED) != 0 || m->object == NULL) + return (unwired); + + if (m->queue == queue) { + if (queue == PQ_ACTIVE) + vm_page_reference(m); + else if (queue != PQ_NONE) + vm_page_requeue(m); + } else { + vm_page_dequeue(m); + if (queue != PQ_NONE) { + vm_page_enqueue(m, queue); if (queue == PQ_ACTIVE) - vm_page_reference(m); - else if (queue != PQ_NONE) - vm_page_requeue(m); - } else { - vm_page_remque(m); - if (queue != PQ_NONE) { - vm_page_enqueue(queue, m); - if (queue == PQ_ACTIVE) - /* Initialize act_count. */ - vm_page_activate(m); - } + /* Initialize act_count. */ + vm_page_activate(m); } } return (unwired); @@ -3476,65 +3657,32 @@ } /* - * Move the specified page to the inactive queue, or requeue the page if it is - * already in the inactive queue. - * - * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive - * queue. However, setting "noreuse" to TRUE will accelerate the specified - * page's reclamation, but it will not unmap the page from any address space. - * This is implemented by inserting the page near the head of the inactive - * queue, using a marker page to guide FIFO insertion ordering. + * Move the specified page to the tail of the inactive queue, or requeue + * the page if it is already in the inactive queue. * * The page must be locked. */ -static inline void -_vm_page_deactivate(vm_page_t m, boolean_t noreuse) +void +vm_page_deactivate(vm_page_t m) { - struct vm_pagequeue *pq; - int queue; vm_page_assert_locked(m); - if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { - pq = &vm_pagequeue_domain(m)->vmd_pagequeues[PQ_INACTIVE]; - /* Avoid multiple acquisitions of the inactive queue lock. */ - queue = m->queue; - if (queue == PQ_INACTIVE) { - vm_pagequeue_lock(pq); - vm_page_dequeue_locked(m); - } else { - if (queue != PQ_NONE) - vm_page_dequeue(m); - vm_pagequeue_lock(pq); - } - m->queue = PQ_INACTIVE; - if (noreuse) - TAILQ_INSERT_BEFORE( - &vm_pagequeue_domain(m)->vmd_inacthead, m, - plinks.q); - else - TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); - vm_pagequeue_cnt_inc(pq); - vm_pagequeue_unlock(pq); - } -} - -/* - * Move the specified page to the inactive queue, or requeue the page if it is - * already in the inactive queue. - * - * The page must be locked. - */ -void -vm_page_deactivate(vm_page_t m) -{ + if (m->wire_count > 0 || (m->oflags & VPO_UNMANAGED) != 0) + return; - _vm_page_deactivate(m, FALSE); + if (!vm_page_inactive(m)) { + vm_page_remque(m); + vm_page_enqueue(m, PQ_INACTIVE); + } else + vm_page_requeue(m); } /* - * Move the specified page to the inactive queue with the expectation - * that it is unlikely to be reused. + * Move the specified page close to the head of the inactive queue, + * bypassing LRU. A marker page is used to maintain FIFO ordering. + * As with regular enqueues, we use a per-CPU batch queue to reduce + * contention on the page queue lock. * * The page must be locked. */ @@ -3542,7 +3690,17 @@ vm_page_deactivate_noreuse(vm_page_t m) { - _vm_page_deactivate(m, TRUE); + vm_page_assert_locked(m); + + if (m->wire_count > 0 || (m->oflags & VPO_UNMANAGED) != 0) + return; + + if (!vm_page_inactive(m)) + vm_page_remque(m); + m->queue = PQ_INACTIVE; + if ((m->aflags & PGA_REQUEUE_HEAD) == 0) + vm_page_aflag_set(m, PGA_REQUEUE_HEAD); + vm_pqbatch_submit_page(m, PQ_INACTIVE); } /* @@ -3555,13 +3713,14 @@ { vm_page_assert_locked(m); - if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { - if (m->queue == PQ_LAUNDRY) - vm_page_requeue(m); - else { - vm_page_remque(m); - vm_page_enqueue(PQ_LAUNDRY, m); - } + if (m->wire_count > 0 || (m->oflags & VPO_UNMANAGED) != 0) + return; + + if (m->queue == PQ_LAUNDRY) + vm_page_requeue(m); + else { + vm_page_remque(m); + vm_page_enqueue(m, PQ_LAUNDRY); } } @@ -3577,9 +3736,9 @@ vm_page_assert_locked(m); KASSERT(m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0, ("page %p already unswappable", m)); - if (m->queue != PQ_NONE) - vm_page_dequeue(m); - vm_page_enqueue(PQ_UNSWAPPABLE, m); + + vm_page_remque(m); + vm_page_enqueue(m, PQ_UNSWAPPABLE); } /* Index: sys/vm/vm_pageout.c =================================================================== --- sys/vm/vm_pageout.c +++ sys/vm/vm_pageout.c @@ -201,103 +201,134 @@ CTLFLAG_RW, &vm_page_max_wired, 0, "System-wide limit to wired page count"); static u_int isqrt(u_int num); -static boolean_t vm_pageout_fallback_object_lock(vm_page_t, vm_page_t *); static int vm_pageout_launder(struct vm_domain *vmd, int launder, bool in_shortfall); static void vm_pageout_laundry_worker(void *arg); -static boolean_t vm_pageout_page_lock(vm_page_t, vm_page_t *); + +struct scan_state { + struct vm_batchqueue bq; + struct vm_pagequeue *pq; + vm_page_t marker; + int maxscan; + int scanned; +}; + +static void +vm_pageout_init_scan(struct scan_state *ss, struct vm_pagequeue *pq, + vm_page_t marker, vm_page_t after, int maxscan) +{ + + vm_pagequeue_assert_locked(pq); + KASSERT((marker->aflags & PGA_ENQUEUED) == 0, + ("marker %p already enqueued", marker)); + + if (after == NULL) + TAILQ_INSERT_HEAD(&pq->pq_pl, marker, plinks.q); + else + TAILQ_INSERT_AFTER(&pq->pq_pl, after, marker, plinks.q); + vm_page_aflag_set(marker, PGA_ENQUEUED); + + vm_batchqueue_init(&ss->bq); + ss->pq = pq; + ss->marker = marker; + ss->maxscan = maxscan; + ss->scanned = 0; + vm_pagequeue_unlock(pq); +} + +static void +vm_pageout_end_scan(struct scan_state *ss) +{ + struct vm_pagequeue *pq; + + pq = ss->pq; + vm_pagequeue_assert_locked(pq); + KASSERT((ss->marker->aflags & PGA_ENQUEUED) != 0, + ("marker %p not enqueued", ss->marker)); + + TAILQ_REMOVE(&pq->pq_pl, ss->marker, plinks.q); + vm_page_aflag_clear(ss->marker, PGA_ENQUEUED); + VM_CNT_ADD(v_pdpages, ss->scanned); +} /* - * vm_pageout_fallback_object_lock: - * - * Lock vm object currently associated with `m'. VM_OBJECT_TRYWLOCK is - * known to have failed and page queue must be either PQ_ACTIVE or - * PQ_INACTIVE. To avoid lock order violation, unlock the page queue - * while locking the vm object. Use marker page to detect page queue - * changes and maintain notion of next page on page queue. Return - * TRUE if no changes were detected, FALSE otherwise. vm object is - * locked on return. - * - * This function depends on both the lock portion of struct vm_object - * and normal struct vm_page being type stable. + * Ensure that the page has not been dequeued after a pageout batch was + * collected. See vm_page_dequeue_complete(). */ -static boolean_t -vm_pageout_fallback_object_lock(vm_page_t m, vm_page_t *next) +static inline bool +vm_pageout_page_queued(vm_page_t m, int queue) { - struct vm_page marker; - struct vm_pagequeue *pq; - boolean_t unchanged; - vm_object_t object; - int queue; - queue = m->queue; - vm_page_init_marker(&marker, queue); - pq = vm_page_pagequeue(m); - object = m->object; - - TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); - vm_pagequeue_unlock(pq); - vm_page_unlock(m); - VM_OBJECT_WLOCK(object); - vm_page_lock(m); - vm_pagequeue_lock(pq); + vm_page_assert_locked(m); - /* - * The page's object might have changed, and/or the page might - * have moved from its original position in the queue. If the - * page's object has changed, then the caller should abandon - * processing the page because the wrong object lock was - * acquired. Use the marker's plinks.q, not the page's, to - * determine if the page has been moved. The state of the - * page's plinks.q can be indeterminate; whereas, the marker's - * plinks.q must be valid. - */ - *next = TAILQ_NEXT(&marker, plinks.q); - unchanged = m->object == object && - m == TAILQ_PREV(&marker, pglist, plinks.q); - KASSERT(!unchanged || m->queue == queue, - ("page %p queue %d %d", m, queue, m->queue)); - TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); - return (unchanged); + if ((m->aflags & PGA_DEQUEUE) != 0) + return (false); + atomic_thread_fence_acq(); + return (m->queue == queue); } /* - * Lock the page while holding the page queue lock. Use marker page - * to detect page queue changes and maintain notion of next page on - * page queue. Return TRUE if no changes were detected, FALSE - * otherwise. The page is locked on return. The page queue lock might - * be dropped and reacquired. + * Add a small number of queued pages to a batch queue for later processing + * without the corresponding queue lock held. The caller must have enqueued a + * marker page at the desired start point for the scan. Pages will be + * physically dequeued if the caller so requests. Otherwise, the returned + * batch may contain marker pages, and it is up to the caller to handle them. * - * This function depends on normal struct vm_page being type stable. + * When processing the batch queue, vm_pageout_page_queued() must be used to + * determine whether the page was logically dequeued by another thread. Once + * this check is performed, the page lock guarantees that the page will not be + * disassociated from the queue. */ -static boolean_t -vm_pageout_page_lock(vm_page_t m, vm_page_t *next) +static __always_inline void +vm_pageout_collect_batch(struct scan_state *ss, const bool dequeue) { - struct vm_page marker; struct vm_pagequeue *pq; - boolean_t unchanged; - int queue; + vm_page_t m, marker; - vm_page_lock_assert(m, MA_NOTOWNED); - if (vm_page_trylock(m)) - return (TRUE); + marker = ss->marker; + pq = ss->pq; - queue = m->queue; - vm_page_init_marker(&marker, queue); - pq = vm_page_pagequeue(m); + KASSERT((marker->aflags & PGA_ENQUEUED) != 0, + ("marker %p not enqueued", ss->marker)); - TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); - vm_pagequeue_unlock(pq); - vm_page_lock(m); vm_pagequeue_lock(pq); + for (m = TAILQ_NEXT(marker, plinks.q); m != NULL && + ss->scanned < ss->maxscan && ss->bq.bq_cnt < VM_BATCHQUEUE_SIZE; + m = TAILQ_NEXT(m, plinks.q), ss->scanned++) { + if ((m->flags & PG_MARKER) == 0) { + KASSERT((m->aflags & PGA_ENQUEUED) != 0, + ("page %p not enqueued", m)); + KASSERT((m->flags & PG_FICTITIOUS) == 0, + ("Fictitious page %p cannot be in page queue", m)); + KASSERT((m->oflags & VPO_UNMANAGED) == 0, + ("Unmanaged page %p cannot be in page queue", m)); + } else if (dequeue) + continue; + + (void)vm_batchqueue_insert(&ss->bq, m); + if (dequeue) { + TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); + vm_page_aflag_clear(m, PGA_ENQUEUED); + } + } + TAILQ_REMOVE(&pq->pq_pl, marker, plinks.q); + if (__predict_true(m != NULL)) + TAILQ_INSERT_BEFORE(m, marker, plinks.q); + else + TAILQ_INSERT_TAIL(&pq->pq_pl, marker, plinks.q); + if (dequeue) + vm_pagequeue_cnt_add(pq, -ss->bq.bq_cnt); + vm_pagequeue_unlock(pq); +} + +/* Return the next page to be scanned, or NULL if the scan is complete. */ +static __always_inline vm_page_t +vm_pageout_next(struct scan_state *ss, const bool dequeue) +{ - /* Page queue might have changed. */ - *next = TAILQ_NEXT(&marker, plinks.q); - unchanged = m == TAILQ_PREV(&marker, pglist, plinks.q); - KASSERT(!unchanged || m->queue == queue, - ("page %p queue %d %d", m, queue, m->queue)); - TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); - return (unchanged); + if (ss->bq.bq_cnt == 0) + vm_pageout_collect_batch(ss, dequeue); + return (vm_batchqueue_pop(&ss->bq)); } /* @@ -353,12 +384,12 @@ break; } vm_page_test_dirty(p); - if (p->dirty == 0) { + if (p->dirty == 0 || !vm_page_in_laundry(p)) { ib = 0; break; } vm_page_lock(p); - if (!vm_page_in_laundry(p) || vm_page_held(p)) { + if (vm_page_held(p)) { vm_page_unlock(p); ib = 0; break; @@ -381,10 +412,10 @@ if ((p = vm_page_next(ps)) == NULL || vm_page_busied(p)) break; vm_page_test_dirty(p); - if (p->dirty == 0) + if (p->dirty == 0 || !vm_page_in_laundry(p)) break; vm_page_lock(p); - if (!vm_page_in_laundry(p) || vm_page_held(p)) { + if (vm_page_held(p)) { vm_page_unlock(p); break; } @@ -675,13 +706,18 @@ static int vm_pageout_launder(struct vm_domain *vmd, int launder, bool in_shortfall) { + struct scan_state ss; struct vm_pagequeue *pq; + struct mtx *mtx; vm_object_t object; - vm_page_t m, marker, next; - int act_delta, error, maxscan, numpagedout, queue, starting_target; + vm_page_t m, marker; + int act_delta, error, numpagedout, queue, starting_target; int vnodes_skipped; - bool pageout_ok, queue_locked; + bool obj_locked, pageout_ok; + mtx = NULL; + obj_locked = false; + object = NULL; starting_target = launder; vnodes_skipped = 0; @@ -691,10 +727,6 @@ * we've reached the end of the queue. A single iteration of this loop * may cause more than one page to be laundered because of clustering. * - * maxscan ensures that we don't re-examine requeued pages. Any - * additional pages written as part of a cluster are subtracted from - * maxscan since they must be taken from the laundry queue. - * * As an optimization, we avoid laundering from PQ_UNSWAPPABLE when no * swap devices are configured. */ @@ -704,53 +736,68 @@ queue = PQ_LAUNDRY; scan: - pq = &vmd->vmd_pagequeues[queue]; marker = &vmd->vmd_markers[queue]; - + pq = &vmd->vmd_pagequeues[queue]; vm_pagequeue_lock(pq); - maxscan = pq->pq_cnt; - queue_locked = true; - for (m = TAILQ_FIRST(&pq->pq_pl); - m != NULL && maxscan-- > 0 && launder > 0; - m = next) { - vm_pagequeue_assert_locked(pq); - KASSERT(queue_locked, ("unlocked laundry queue")); - KASSERT(vm_page_in_laundry(m), - ("page %p has an inconsistent queue", m)); - next = TAILQ_NEXT(m, plinks.q); - if ((m->flags & PG_MARKER) != 0) + vm_pageout_init_scan(&ss, pq, marker, NULL, pq->pq_cnt); + while (launder > 0 && (m = vm_pageout_next(&ss, false)) != NULL) { + if (__predict_false((m->flags & PG_MARKER) != 0)) continue; - KASSERT((m->flags & PG_FICTITIOUS) == 0, - ("PG_FICTITIOUS page %p cannot be in laundry queue", m)); - KASSERT((m->oflags & VPO_UNMANAGED) == 0, - ("VPO_UNMANAGED page %p cannot be in laundry queue", m)); - if (!vm_pageout_page_lock(m, &next) || m->hold_count != 0) { - vm_page_unlock(m); + + vm_page_change_lock(m, &mtx); + +recheck: + /* + * The page may have been disassociated from the queue + * while locks were dropped. + */ + if (!vm_pageout_page_queued(m, queue)) + continue; + + /* + * A requeue was requested, so this page gets a second + * chance. + */ + if ((m->aflags & PGA_REQUEUE) != 0) { + vm_page_requeue(m); continue; } + + /* + * Held pages are essentially stuck in the queue. + * + * Wired pages may not be freed. Complete their removal + * from the queue now to avoid needless revisits during + * future scans. + */ + if (m->hold_count != 0) + continue; if (m->wire_count != 0) { - vm_page_dequeue_locked(m); - vm_page_unlock(m); + vm_page_dequeue_deferred(m); continue; } - object = m->object; - if ((!VM_OBJECT_TRYWLOCK(object) && - (!vm_pageout_fallback_object_lock(m, &next) || - vm_page_held(m))) || vm_page_busied(m)) { - VM_OBJECT_WUNLOCK(object); - if (m->wire_count != 0 && vm_page_pagequeue(m) == pq) - vm_page_dequeue_locked(m); - vm_page_unlock(m); - continue; + + if (object != m->object) { + if (obj_locked) { + VM_OBJECT_WUNLOCK(object); + obj_locked = false; + } + object = m->object; + } + if (!obj_locked) { + if (!VM_OBJECT_TRYWLOCK(object)) { + mtx_unlock(mtx); + /* Depends on type-stability. */ + VM_OBJECT_WLOCK(object); + obj_locked = true; + mtx_lock(mtx); + goto recheck; + } else + obj_locked = true; } - /* - * Unlock the laundry queue, invalidating the 'next' pointer. - * Use a marker to remember our place in the laundry queue. - */ - TAILQ_INSERT_AFTER(&pq->pq_pl, m, marker, plinks.q); - vm_pagequeue_unlock(pq); - queue_locked = false; + if (vm_page_busied(m)) + continue; /* * Invalid pages can be easily freed. They cannot be @@ -799,9 +846,11 @@ */ if (!in_shortfall) launder--; - goto drop_page; - } else if ((object->flags & OBJ_DEAD) == 0) - goto requeue_page; + continue; + } else if ((object->flags & OBJ_DEAD) == 0) { + vm_page_requeue(m); + continue; + } } /* @@ -836,11 +885,8 @@ else pageout_ok = true; if (!pageout_ok) { -requeue_page: - vm_pagequeue_lock(pq); - queue_locked = true; - vm_page_requeue_locked(m); - goto drop_page; + vm_page_requeue(m); + continue; } /* @@ -859,24 +905,25 @@ error = vm_pageout_clean(m, &numpagedout); if (error == 0) { launder -= numpagedout; - maxscan -= numpagedout - 1; + ss.scanned += numpagedout; } else if (error == EDEADLK) { pageout_lock_miss++; vnodes_skipped++; } - goto relock_queue; + mtx = NULL; + obj_locked = false; } -drop_page: - vm_page_unlock(m); + } + if (mtx != NULL) { + mtx_unlock(mtx); + mtx = NULL; + } + if (obj_locked) { VM_OBJECT_WUNLOCK(object); -relock_queue: - if (!queue_locked) { - vm_pagequeue_lock(pq); - queue_locked = true; - } - next = TAILQ_NEXT(marker, plinks.q); - TAILQ_REMOVE(&pq->pq_pl, marker, plinks.q); + obj_locked = false; } + vm_pagequeue_lock(pq); + vm_pageout_end_scan(&ss); vm_pagequeue_unlock(pq); if (launder > 0 && queue == PQ_UNSWAPPABLE) { @@ -1077,6 +1124,56 @@ } } +static int +vm_pageout_reinsert_inactive_page(struct scan_state *ss, vm_page_t m) +{ + struct vm_domain *vmd; + + if (!vm_page_inactive(m) || (m->aflags & PGA_ENQUEUED) != 0) + return (0); + vm_page_aflag_set(m, PGA_ENQUEUED); + if ((m->aflags & PGA_REQUEUE_HEAD) != 0) { + vmd = vm_pagequeue_domain(m); + TAILQ_INSERT_BEFORE(&vmd->vmd_inacthead, m, plinks.q); + vm_page_aflag_clear(m, PGA_REQUEUE | PGA_REQUEUE_HEAD); + } else if ((m->aflags & PGA_REQUEUE) != 0) { + TAILQ_INSERT_TAIL(&ss->pq->pq_pl, m, plinks.q); + vm_page_aflag_clear(m, PGA_REQUEUE | PGA_REQUEUE_HEAD); + } else + TAILQ_INSERT_BEFORE(ss->marker, m, plinks.q); + return (1); +} + +/* + * Re-add stuck pages to the inactive queue. We will examine them again + * during the next scan. If the queue state of a page has changed since + * it was physically removed from the page queue in + * vm_pageout_collect_batch(), don't do anything with that page. + */ +static void +vm_pageout_reinsert_inactive(struct scan_state *ss, struct vm_batchqueue *bq, + vm_page_t m) +{ + struct vm_pagequeue *pq; + int delta; + + delta = 0; + pq = ss->pq; + + if (m != NULL) { + if (vm_batchqueue_insert(bq, m)) + return; + vm_pagequeue_lock(pq); + delta += vm_pageout_reinsert_inactive_page(ss, m); + } else + vm_pagequeue_lock(pq); + while ((m = vm_batchqueue_pop(bq)) != NULL) + delta += vm_pageout_reinsert_inactive_page(ss, m); + vm_pagequeue_cnt_add(pq, delta); + vm_pagequeue_unlock(pq); + vm_batchqueue_init(bq); +} + /* * vm_pageout_scan does the dirty work for the pageout daemon. * @@ -1089,13 +1186,16 @@ static bool vm_pageout_scan(struct vm_domain *vmd, int pass, int shortage) { - vm_page_t m, marker, next; + struct scan_state ss; + struct vm_batchqueue rq; + struct mtx *mtx; + vm_page_t m, marker; struct vm_pagequeue *pq; vm_object_t object; long min_scan; - int act_delta, addl_page_shortage, deficit, inactq_shortage, maxscan; - int page_shortage, scan_tick, scanned, starting_page_shortage; - boolean_t queue_locked; + int act_delta, addl_page_shortage, deficit, inactq_shortage, max_scan; + int page_shortage, scan_tick, starting_page_shortage; + bool obj_locked; /* * If we need to reclaim memory ask kernel caches to return @@ -1136,79 +1236,85 @@ page_shortage = deficit = 0; starting_page_shortage = page_shortage; + mtx = NULL; + obj_locked = false; + object = NULL; + vm_batchqueue_init(&rq); + /* * Start scanning the inactive queue for pages that we can free. The * scan will stop when we reach the target or we have scanned the * entire queue. (Note that m->act_count is not used to make * decisions for the inactive queue, only for the active queue.) */ - pq = &vmd->vmd_pagequeues[PQ_INACTIVE]; marker = &vmd->vmd_markers[PQ_INACTIVE]; - maxscan = pq->pq_cnt; + pq = &vmd->vmd_pagequeues[PQ_INACTIVE]; vm_pagequeue_lock(pq); - queue_locked = TRUE; - for (m = TAILQ_FIRST(&pq->pq_pl); - m != NULL && maxscan-- > 0 && page_shortage > 0; - m = next) { - vm_pagequeue_assert_locked(pq); - KASSERT(queue_locked, ("unlocked inactive queue")); - KASSERT(vm_page_inactive(m), ("Inactive queue %p", m)); + vm_pageout_init_scan(&ss, pq, marker, NULL, pq->pq_cnt); + while (page_shortage > 0 && (m = vm_pageout_next(&ss, true)) != NULL) { + KASSERT((m->flags & PG_MARKER) == 0, + ("marker page %p was dequeued", m)); - VM_CNT_INC(v_pdpages); - next = TAILQ_NEXT(m, plinks.q); + vm_page_change_lock(m, &mtx); +recheck: /* - * skip marker pages + * The page may have been disassociated from the queue + * while locks were dropped. */ - if (m->flags & PG_MARKER) + if (!vm_pageout_page_queued(m, PQ_INACTIVE)) { + addl_page_shortage++; continue; + } - KASSERT((m->flags & PG_FICTITIOUS) == 0, - ("Fictitious page %p cannot be in inactive queue", m)); - KASSERT((m->oflags & VPO_UNMANAGED) == 0, - ("Unmanaged page %p cannot be in inactive queue", m)); + /* + * The page was re-enqueued after the page queue lock was + * dropped, or a requeue was requested. This page gets a second + * chance. + */ + if ((m->aflags & (PGA_ENQUEUED | PGA_REQUEUE | + PGA_REQUEUE_HEAD)) != 0) + goto reinsert; /* - * The page or object lock acquisitions fail if the - * page was removed from the queue or moved to a - * different position within the queue. In either - * case, addl_page_shortage should not be incremented. + * Held pages are essentially stuck in the queue. So, + * they ought to be discounted from the inactive count. + * See the calculation of inactq_shortage before the + * loop over the active queue below. + * + * Wired pages may not be freed. Complete their removal + * from the queue now to avoid needless revisits during + * future scans. */ - if (!vm_pageout_page_lock(m, &next)) - goto unlock_page; - else if (m->wire_count != 0) { - /* - * Wired pages may not be freed, and unwiring a queued - * page will cause it to be requeued. Thus, remove them - * from the queue now to avoid unnecessary revisits. - */ - vm_page_dequeue_locked(m); + if (m->hold_count != 0) { addl_page_shortage++; - goto unlock_page; - } else if (m->hold_count != 0) { - /* - * Held pages are essentially stuck in the - * queue. So, they ought to be discounted - * from the inactive count. See the - * calculation of inactq_shortage before the - * loop over the active queue below. - */ + goto reinsert; + } + if (m->wire_count != 0) { addl_page_shortage++; - goto unlock_page; + vm_page_dequeue_deferred(m); + continue; } - object = m->object; - if (!VM_OBJECT_TRYWLOCK(object)) { - if (!vm_pageout_fallback_object_lock(m, &next)) - goto unlock_object; - else if (m->wire_count != 0) { - vm_page_dequeue_locked(m); - addl_page_shortage++; - goto unlock_object; - } else if (m->hold_count != 0) { - addl_page_shortage++; - goto unlock_object; + + if (object != m->object) { + if (obj_locked) { + VM_OBJECT_WUNLOCK(object); + obj_locked = false; } + object = m->object; } + if (!obj_locked) { + if (!VM_OBJECT_TRYWLOCK(object)) { + mtx_unlock(mtx); + /* Depends on type-stability. */ + VM_OBJECT_WLOCK(object); + obj_locked = true; + mtx_lock(mtx); + goto recheck; + } else + obj_locked = true; + } + if (vm_page_busied(m)) { /* * Don't mess with busy pages. Leave them at @@ -1219,26 +1325,8 @@ * inactive count. */ addl_page_shortage++; -unlock_object: - VM_OBJECT_WUNLOCK(object); -unlock_page: - vm_page_unlock(m); - continue; + goto reinsert; } - KASSERT(!vm_page_held(m), ("Held page %p", m)); - - /* - * Dequeue the inactive page and unlock the inactive page - * queue, invalidating the 'next' pointer. Dequeueing the - * page here avoids a later reacquisition (and release) of - * the inactive page queue lock when vm_page_activate(), - * vm_page_free(), or vm_page_launder() is called. Use a - * marker to remember our place in the inactive queue. - */ - TAILQ_INSERT_AFTER(&pq->pq_pl, m, marker, plinks.q); - vm_page_dequeue_locked(m); - vm_pagequeue_unlock(pq); - queue_locked = FALSE; /* * Invalid pages can be easily freed. They cannot be @@ -1276,14 +1364,10 @@ * queue. */ m->act_count += act_delta + ACT_ADVANCE; - goto drop_page; + continue; } else if ((object->flags & OBJ_DEAD) == 0) { - vm_pagequeue_lock(pq); - queue_locked = TRUE; - m->queue = PQ_INACTIVE; - TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); - vm_pagequeue_cnt_inc(pq); - goto drop_page; + vm_page_aflag_set(m, PGA_REQUEUE); + goto reinsert; } } @@ -1309,23 +1393,39 @@ */ if (m->dirty == 0) { free_page: + /* + * Because we dequeued the page and have already + * checked for concurrent dequeue and enqueue + * requests, we can safely disassociate the page + * from the inactive queue. + */ + KASSERT((m->aflags & PGA_QUEUE_STATE_MASK) == 0, + ("page %p has queue state", m)); + m->queue = PQ_NONE; vm_page_free(m); - VM_CNT_INC(v_dfree); - --page_shortage; + page_shortage--; } else if ((object->flags & OBJ_DEAD) == 0) vm_page_launder(m); -drop_page: - vm_page_unlock(m); + continue; +reinsert: + vm_pageout_reinsert_inactive(&ss, &rq, m); + } + if (mtx != NULL) { + mtx_unlock(mtx); + mtx = NULL; + } + if (obj_locked) { VM_OBJECT_WUNLOCK(object); - if (!queue_locked) { - vm_pagequeue_lock(pq); - queue_locked = TRUE; - } - next = TAILQ_NEXT(marker, plinks.q); - TAILQ_REMOVE(&pq->pq_pl, marker, plinks.q); + obj_locked = false; } + vm_pageout_reinsert_inactive(&ss, &rq, NULL); + vm_pageout_reinsert_inactive(&ss, &ss.bq, NULL); + vm_pagequeue_lock(pq); + vm_pageout_end_scan(&ss); vm_pagequeue_unlock(pq); + VM_CNT_ADD(v_dfree, starting_page_shortage - page_shortage); + /* * Wake up the laundry thread so that it can perform any needed * laundering. If we didn't meet our target, we're in shortfall and @@ -1386,9 +1486,9 @@ vm_paging_target(vmd) + deficit + addl_page_shortage; inactq_shortage *= act_scan_laundry_weight; + marker = &vmd->vmd_markers[PQ_ACTIVE]; pq = &vmd->vmd_pagequeues[PQ_ACTIVE]; vm_pagequeue_lock(pq); - maxscan = pq->pq_cnt; /* * If we're just idle polling attempt to visit every @@ -1401,43 +1501,55 @@ min_scan /= hz * vm_pageout_update_period; } else min_scan = 0; - if (min_scan > 0 || (inactq_shortage > 0 && maxscan > 0)) + if (min_scan > 0 || (inactq_shortage > 0 && pq->pq_cnt > 0)) vmd->vmd_last_active_scan = scan_tick; /* * Scan the active queue for pages that can be deactivated. Update * the per-page activity counter and use it to identify deactivation * candidates. Held pages may be deactivated. + * + * To avoid requeuing each page that remains in the active queue, we + * implement the CLOCK algorithm. To maintain consistency in the + * generic page queue code, pages are inserted at the tail of the + * active queue. We thus use two hands, represented by marker pages: + * scans begin at the first hand, which precedes the second hand in + * the queue. When the two hands meet, they are moved back to the + * head and tail of the queue, respectively, and scanning resumes. */ - for (m = TAILQ_FIRST(&pq->pq_pl), scanned = 0; m != NULL && (scanned < - min_scan || (inactq_shortage > 0 && scanned < maxscan)); m = next, - scanned++) { - KASSERT(m->queue == PQ_ACTIVE, - ("vm_pageout_scan: page %p isn't active", m)); - next = TAILQ_NEXT(m, plinks.q); - if ((m->flags & PG_MARKER) != 0) - continue; - KASSERT((m->flags & PG_FICTITIOUS) == 0, - ("Fictitious page %p cannot be in active queue", m)); - KASSERT((m->oflags & VPO_UNMANAGED) == 0, - ("Unmanaged page %p cannot be in active queue", m)); - if (!vm_pageout_page_lock(m, &next)) { - vm_page_unlock(m); - continue; + max_scan = inactq_shortage > 0 ? pq->pq_cnt : min_scan; +act_scan: + vm_pageout_init_scan(&ss, pq, marker, &vmd->vmd_clock[0], max_scan); + while ((m = vm_pageout_next(&ss, false)) != NULL) { + if (__predict_false(m == &vmd->vmd_clock[1])) { + vm_pagequeue_lock(pq); + TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_clock[0], plinks.q); + TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_clock[1], plinks.q); + TAILQ_INSERT_HEAD(&pq->pq_pl, &vmd->vmd_clock[0], + plinks.q); + TAILQ_INSERT_TAIL(&pq->pq_pl, &vmd->vmd_clock[1], + plinks.q); + max_scan -= ss.scanned; + vm_pageout_end_scan(&ss); + goto act_scan; } + if (__predict_false((m->flags & PG_MARKER) != 0)) + continue; + + vm_page_change_lock(m, &mtx); /* - * The count for page daemon pages is updated after checking - * the page for eligibility. + * The page may have been disassociated from the queue + * while locks were dropped. */ - VM_CNT_INC(v_pdpages); + if (!vm_pageout_page_queued(m, PQ_ACTIVE)) + continue; /* * Wired pages are dequeued lazily. */ if (m->wire_count != 0) { - vm_page_dequeue_locked(m); - vm_page_unlock(m); + vm_page_dequeue_deferred(m); continue; } @@ -1476,14 +1588,7 @@ } else m->act_count -= min(m->act_count, ACT_DECLINE); - /* - * Move this page to the tail of the active, inactive or laundry - * queue depending on usage. - */ if (m->act_count == 0) { - /* Dequeue to avoid later lock recursion. */ - vm_page_dequeue_locked(m); - /* * When not short for inactive pages, let dirty pages go * through the inactive queue before moving to the @@ -1515,11 +1620,18 @@ inactq_shortage--; } } - } else - vm_page_requeue_locked(m); - vm_page_unlock(m); + } + } + if (mtx != NULL) { + mtx_unlock(mtx); + mtx = NULL; } + vm_pagequeue_lock(pq); + TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_clock[0], plinks.q); + TAILQ_INSERT_AFTER(&pq->pq_pl, marker, &vmd->vmd_clock[0], plinks.q); + vm_pageout_end_scan(&ss); vm_pagequeue_unlock(pq); + if (pass > 0) vm_swapout_run_idle(); return (page_shortage <= 0); Index: sys/vm/vm_pagequeue.h =================================================================== --- sys/vm/vm_pagequeue.h +++ sys/vm/vm_pagequeue.h @@ -73,8 +73,17 @@ const char * const pq_name; } __aligned(CACHE_LINE_SIZE); -#include +#ifndef VM_BATCHQUEUE_SIZE +#define VM_BATCHQUEUE_SIZE 7 +#endif + +struct vm_batchqueue { + vm_page_t bq_pa[VM_BATCHQUEUE_SIZE]; + int bq_cnt; +} __aligned(CACHE_LINE_SIZE); + #include +#include struct sysctl_oid; /* @@ -82,12 +91,12 @@ * and accounting. * * Lock Key: - * f vmd_free_mtx - * p vmd_pageout_mtx - * d vm_domainset_lock - * a atomic - * c const after boot - * q page queue lock + * f vmd_free_mtx + * p vmd_pageout_mtx + * d vm_domainset_lock + * a atomic + * c const after boot + * q page queue lock */ struct vm_domain { struct vm_pagequeue vmd_pagequeues[PQ_COUNT]; @@ -107,8 +116,9 @@ boolean_t vmd_oom; int vmd_oom_seq; int vmd_last_active_scan; - struct vm_page vmd_markers[PQ_COUNT]; /* markers for queue scans */ + struct vm_page vmd_markers[PQ_COUNT]; /* (q) markers for queue scans */ struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */ + struct vm_page vmd_clock[2]; /* markers for active queue scan */ int vmd_pageout_wanted; /* (a, p) pageout daemon wait channel */ int vmd_pageout_pages_needed; /* (d) page daemon waiting for pages? */ @@ -144,6 +154,7 @@ #define vm_pagequeue_assert_locked(pq) mtx_assert(&(pq)->pq_mutex, MA_OWNED) #define vm_pagequeue_lock(pq) mtx_lock(&(pq)->pq_mutex) #define vm_pagequeue_lockptr(pq) (&(pq)->pq_mutex) +#define vm_pagequeue_trylock(pq) mtx_trylock(&(pq)->pq_mutex) #define vm_pagequeue_unlock(pq) mtx_unlock(&(pq)->pq_mutex) #define vm_domain_free_assert_locked(n) \ @@ -154,6 +165,8 @@ mtx_lock(vm_domain_free_lockptr((d))) #define vm_domain_free_lockptr(d) \ (&(d)->vmd_free_mtx) +#define vm_domain_free_trylock(d) \ + mtx_trylock(vm_domain_free_lockptr((d))) #define vm_domain_free_unlock(d) \ mtx_unlock(vm_domain_free_lockptr((d))) @@ -172,14 +185,39 @@ vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend) { -#ifdef notyet vm_pagequeue_assert_locked(pq); -#endif pq->pq_cnt += addend; } #define vm_pagequeue_cnt_inc(pq) vm_pagequeue_cnt_add((pq), 1) #define vm_pagequeue_cnt_dec(pq) vm_pagequeue_cnt_add((pq), -1) +static inline void +vm_batchqueue_init(struct vm_batchqueue *bq) +{ + + bq->bq_cnt = 0; +} + +static inline bool +vm_batchqueue_insert(struct vm_batchqueue *bq, vm_page_t m) +{ + + if (bq->bq_cnt < nitems(bq->bq_pa)) { + bq->bq_pa[bq->bq_cnt++] = m; + return (true); + } + return (false); +} + +static inline vm_page_t +vm_batchqueue_pop(struct vm_batchqueue *bq) +{ + + if (bq->bq_cnt == 0) + return (NULL); + return (bq->bq_pa[--bq->bq_cnt]); +} + void vm_domain_set(struct vm_domain *vmd); void vm_domain_clear(struct vm_domain *vmd); int vm_domain_allocate(struct vm_domain *vmd, int req, int npages); Index: sys/vm/vm_phys.c =================================================================== --- sys/vm/vm_phys.c +++ sys/vm/vm_phys.c @@ -354,9 +354,9 @@ m->order = order; if (tail) - TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q); + TAILQ_INSERT_TAIL(&fl[order].pl, m, listq); else - TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q); + TAILQ_INSERT_HEAD(&fl[order].pl, m, listq); fl[order].lcnt++; } @@ -364,7 +364,7 @@ vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order) { - TAILQ_REMOVE(&fl[order].pl, m, plinks.q); + TAILQ_REMOVE(&fl[order].pl, m, listq); fl[order].lcnt--; m->order = VM_NFREEORDER; } @@ -1196,7 +1196,7 @@ oind++) { for (pind = 0; pind < VM_NFREEPOOL; pind++) { fl = (*seg->free_queues)[pind]; - TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) { + TAILQ_FOREACH(m_ret, &fl[oind].pl, listq) { /* * Is the size of this allocation request * larger than the largest block size? Index: sys/vm/vm_swapout.c =================================================================== --- sys/vm/vm_swapout.c +++ sys/vm/vm_swapout.c @@ -399,8 +399,15 @@ swapout_flags = vm_pageout_req_swapout; vm_pageout_req_swapout = 0; mtx_unlock(&vm_daemon_mtx); - if (swapout_flags) + if (swapout_flags != 0) { + /* + * Drain the per-CPU page queue batches as a deadlock + * avoidance measure. + */ + if ((swapout_flags & VM_SWAP_NORMAL) != 0) + vm_page_drain_pqbatch(); swapout_procs(swapout_flags); + } /* * scan the processes for exceeding their rlimits or if