diff options
Diffstat (limited to 'sys/vm')
-rw-r--r-- | sys/vm/memguard.c | 7 | ||||
-rw-r--r-- | sys/vm/pmap.h | 16 | ||||
-rw-r--r-- | sys/vm/swap_pager.c | 659 | ||||
-rw-r--r-- | sys/vm/swap_pager.h | 1 | ||||
-rw-r--r-- | sys/vm/uma.h | 1 | ||||
-rw-r--r-- | sys/vm/uma_core.c | 17 | ||||
-rw-r--r-- | sys/vm/uma_int.h | 3 | ||||
-rw-r--r-- | sys/vm/vm_extern.h | 1 | ||||
-rw-r--r-- | sys/vm/vm_fault.c | 53 | ||||
-rw-r--r-- | sys/vm/vm_glue.c | 517 | ||||
-rw-r--r-- | sys/vm/vm_kern.c | 19 | ||||
-rw-r--r-- | sys/vm/vm_map.c | 25 | ||||
-rw-r--r-- | sys/vm/vm_mmap.c | 2 | ||||
-rw-r--r-- | sys/vm/vm_object.c | 180 | ||||
-rw-r--r-- | sys/vm/vm_object.h | 22 | ||||
-rw-r--r-- | sys/vm/vm_page.c | 458 | ||||
-rw-r--r-- | sys/vm/vm_page.h | 40 | ||||
-rw-r--r-- | sys/vm/vm_pageout.c | 435 | ||||
-rw-r--r-- | sys/vm/vm_pageout.h | 9 | ||||
-rw-r--r-- | sys/vm/vm_param.h | 3 | ||||
-rw-r--r-- | sys/vm/vm_phys.c | 51 | ||||
-rw-r--r-- | sys/vm/vm_phys.h | 1 | ||||
-rw-r--r-- | sys/vm/vm_reserv.c | 14 | ||||
-rw-r--r-- | sys/vm/vm_reserv.h | 1 | ||||
-rw-r--r-- | sys/vm/vm_swapout.c | 975 | ||||
-rw-r--r-- | sys/vm/vm_swapout_dummy.c | 122 | ||||
-rw-r--r-- | sys/vm/vnode_pager.c | 143 |
27 files changed, 2238 insertions, 1537 deletions
diff --git a/sys/vm/memguard.c b/sys/vm/memguard.c index d4efc2b..9e1d7a4 100644 --- a/sys/vm/memguard.c +++ b/sys/vm/memguard.c @@ -283,7 +283,7 @@ v2sizev(vm_offset_t va) void * memguard_alloc(unsigned long req_size, int flags) { - vm_offset_t addr; + vm_offset_t addr, origaddr; u_long size_p, size_v; int do_guard, rv; @@ -327,7 +327,7 @@ memguard_alloc(unsigned long req_size, int flags) for (;;) { if (vmem_xalloc(memguard_arena, size_v, 0, 0, 0, memguard_cursor, VMEM_ADDR_MAX, - M_BESTFIT | M_NOWAIT, &addr) == 0) + M_BESTFIT | M_NOWAIT, &origaddr) == 0) break; /* * The map has no space. This may be due to @@ -342,11 +342,12 @@ memguard_alloc(unsigned long req_size, int flags) memguard_wrap++; memguard_cursor = memguard_base; } + addr = origaddr; if (do_guard) addr += PAGE_SIZE; rv = kmem_back(kmem_object, addr, size_p, flags); if (rv != KERN_SUCCESS) { - vmem_xfree(memguard_arena, addr, size_v); + vmem_xfree(memguard_arena, origaddr, size_v); memguard_fail_pgs++; addr = (vm_offset_t)NULL; goto out; diff --git a/sys/vm/pmap.h b/sys/vm/pmap.h index 1d18823..6033f37 100644 --- a/sys/vm/pmap.h +++ b/sys/vm/pmap.h @@ -100,9 +100,21 @@ extern vm_offset_t kernel_vm_end; /* * Flags for pmap_enter(). The bits in the low-order byte are reserved * for the protection code (vm_prot_t) that describes the fault type. + * Bits 24 through 31 are reserved for the pmap's internal use. */ -#define PMAP_ENTER_NOSLEEP 0x0100 -#define PMAP_ENTER_WIRED 0x0200 +#define PMAP_ENTER_NOSLEEP 0x00000100 +#define PMAP_ENTER_WIRED 0x00000200 +#define PMAP_ENTER_RESERVED 0xFF000000 + +/* + * Define the maximum number of machine-dependent reference bits that are + * cleared by a call to pmap_ts_referenced(). This limit serves two purposes. + * First, it bounds the cost of reference bit maintenance on widely shared + * pages. Second, it prevents numeric overflow during maintenance of a + * widely shared page's "act_count" field. An overflow could result in the + * premature deactivation of the page. + */ +#define PMAP_TS_REFERENCED_MAX 5 void pmap_activate(struct thread *td); void pmap_advise(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, diff --git a/sys/vm/swap_pager.c b/sys/vm/swap_pager.c index f5a766b..0703312 100644 --- a/sys/vm/swap_pager.c +++ b/sys/vm/swap_pager.c @@ -86,10 +86,12 @@ __FBSDID("$FreeBSD$"); #include <sys/namei.h> #include <sys/vnode.h> #include <sys/malloc.h> +#include <sys/pctrie.h> #include <sys/racct.h> #include <sys/resource.h> #include <sys/resourcevar.h> #include <sys/rwlock.h> +#include <sys/sbuf.h> #include <sys/sysctl.h> #include <sys/sysproto.h> #include <sys/blist.h> @@ -126,22 +128,17 @@ __FBSDID("$FreeBSD$"); #define SWB_NPAGES MAX_PAGEOUT_CLUSTER #endif +#define SWAP_META_PAGES PCTRIE_COUNT + /* - * The swblock structure maps an object and a small, fixed-size range - * of page indices to disk addresses within a swap area. - * The collection of these mappings is implemented as a hash table. - * Unused disk addresses within a swap area are allocated and managed - * using a blist. + * A swblk structure maps each page index within a + * SWAP_META_PAGES-aligned and sized range to the address of an + * on-disk swap block (or SWAPBLK_NONE). The collection of these + * mappings for an entire vm object is implemented as a pc-trie. */ -#define SWAP_META_PAGES 32 -#define SWAP_META_MASK (SWAP_META_PAGES - 1) - -struct swblock { - struct swblock *swb_hnext; - vm_object_t swb_object; - vm_pindex_t swb_index; - int swb_count; - daddr_t swb_pages[SWAP_META_PAGES]; +struct swblk { + vm_pindex_t p; + daddr_t d[SWAP_META_PAGES]; }; static MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "swap pager private data"); @@ -159,7 +156,7 @@ static vm_ooffset_t swap_reserved; SYSCTL_QUAD(_vm, OID_AUTO, swap_reserved, CTLFLAG_RD, &swap_reserved, 0, "Amount of swap storage needed to back all allocated anonymous memory."); static int overcommit = 0; -SYSCTL_INT(_vm, OID_AUTO, overcommit, CTLFLAG_RW, &overcommit, 0, +SYSCTL_INT(_vm, VM_OVERCOMMIT, overcommit, CTLFLAG_RW, &overcommit, 0, "Configure virtual memory overcommit behavior. See tuning(7) " "for details."); static unsigned long swzone; @@ -314,7 +311,7 @@ swap_release_by_cred(vm_ooffset_t decr, struct ucred *cred) #define SWM_FREE 0x02 /* free, period */ #define SWM_POP 0x04 /* pop out */ -int swap_pager_full = 2; /* swap space exhaustion (task killing) */ +static int swap_pager_full = 2; /* swap space exhaustion (task killing) */ static int swap_pager_almost_full = 1; /* swap space exhaustion (w/hysteresis)*/ static int nsw_rcount; /* free read buffers */ static int nsw_wcount_sync; /* limit write buffers / synchronous */ @@ -326,10 +323,10 @@ static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_async_max, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_async_max, "I", "Maximum running async swap ops"); - -static struct swblock **swhash; -static int swhash_mask; -static struct mtx swhash_mtx; +static int sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS); +SYSCTL_PROC(_vm, OID_AUTO, swap_fragmentation, CTLTYPE_STRING | CTLFLAG_RD | + CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_fragmentation, "A", + "Swap Fragmentation Info"); static struct sx sw_alloc_sx; @@ -344,7 +341,8 @@ static struct sx sw_alloc_sx; (&swap_pager_object_list[((int)(intptr_t)handle >> 4) & (NOBJLISTS-1)]) static struct pagerlst swap_pager_object_list[NOBJLISTS]; -static uma_zone_t swap_zone; +static uma_zone_t swblk_zone; +static uma_zone_t swpctrie_zone; /* * pagerops for OBJT_SWAP - "swap pager". Some ops are also global procedure @@ -402,12 +400,28 @@ static daddr_t swp_pager_getswapspace(int npages); /* * Metadata functions */ -static struct swblock **swp_pager_hash(vm_object_t object, vm_pindex_t index); static void swp_pager_meta_build(vm_object_t, vm_pindex_t, daddr_t); static void swp_pager_meta_free(vm_object_t, vm_pindex_t, vm_pindex_t); static void swp_pager_meta_free_all(vm_object_t); static daddr_t swp_pager_meta_ctl(vm_object_t, vm_pindex_t, int); +static void * +swblk_trie_alloc(struct pctrie *ptree) +{ + + return (uma_zalloc(swpctrie_zone, M_NOWAIT | (curproc == pageproc ? + M_USE_RESERVE : 0))); +} + +static void +swblk_trie_free(struct pctrie *ptree, void *node) +{ + + uma_zfree(swpctrie_zone, node); +} + +PCTRIE_DEFINE(SWAP, swblk, p, swblk_trie_alloc, swblk_trie_free); + /* * SWP_SIZECHECK() - update swap_pager_full indication * @@ -436,33 +450,6 @@ swp_sizecheck(void) } /* - * SWP_PAGER_HASH() - hash swap meta data - * - * This is an helper function which hashes the swapblk given - * the object and page index. It returns a pointer to a pointer - * to the object, or a pointer to a NULL pointer if it could not - * find a swapblk. - */ -static struct swblock ** -swp_pager_hash(vm_object_t object, vm_pindex_t index) -{ - struct swblock **pswap; - struct swblock *swap; - - index &= ~(vm_pindex_t)SWAP_META_MASK; - pswap = &swhash[(index ^ (int)(intptr_t)object) & swhash_mask]; - while ((swap = *pswap) != NULL) { - if (swap->swb_object == object && - swap->swb_index == index - ) { - break; - } - pswap = &swap->swb_hnext; - } - return (pswap); -} - -/* * SWAP_PAGER_INIT() - initialize the swap pager! * * Expected to be started from system init. NOTE: This code is run @@ -527,21 +514,25 @@ swap_pager_swap_init(void) mtx_unlock(&pbuf_mtx); /* - * Initialize our zone. Right now I'm just guessing on the number - * we need based on the number of pages in the system. Each swblock - * can hold 32 pages, so this is probably overkill. This reservation - * is typically limited to around 32MB by default. + * Initialize our zone, guessing on the number we need based + * on the number of pages in the system. */ n = vm_cnt.v_page_count / 2; - if (maxswzone && n > maxswzone / sizeof(struct swblock)) - n = maxswzone / sizeof(struct swblock); + if (maxswzone && n > maxswzone / sizeof(struct swblk)) + n = maxswzone / sizeof(struct swblk); + swpctrie_zone = uma_zcreate("swpctrie", pctrie_node_size(), NULL, NULL, + pctrie_zone_init, NULL, UMA_ALIGN_PTR, + UMA_ZONE_NOFREE | UMA_ZONE_VM); + if (swpctrie_zone == NULL) + panic("failed to create swap pctrie zone."); + swblk_zone = uma_zcreate("swblk", sizeof(struct swblk), NULL, NULL, + NULL, NULL, _Alignof(struct swblk) - 1, + UMA_ZONE_NOFREE | UMA_ZONE_VM); + if (swblk_zone == NULL) + panic("failed to create swap blk zone."); n2 = n; - swap_zone = uma_zcreate("SWAPMETA", sizeof(struct swblock), NULL, NULL, - NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); - if (swap_zone == NULL) - panic("failed to create swap_zone."); do { - if (uma_zone_reserve_kva(swap_zone, n)) + if (uma_zone_reserve_kva(swblk_zone, n)) break; /* * if the allocation failed, try a zone two thirds the @@ -549,25 +540,22 @@ swap_pager_swap_init(void) */ n -= ((n + 2) / 3); } while (n > 0); - if (n2 != n) - printf("Swap zone entries reduced from %lu to %lu.\n", n2, n); - swap_maxpages = n * SWAP_META_PAGES; - swzone = n * sizeof(struct swblock); - n2 = n; /* - * Initialize our meta-data hash table. The swapper does not need to - * be quite as efficient as the VM system, so we do not use an - * oversized hash table. - * - * n: size of hash table, must be power of 2 - * swhash_mask: hash table index mask + * Often uma_zone_reserve_kva() cannot reserve exactly the + * requested size. Account for the difference when + * calculating swap_maxpages. */ - for (n = 1; n < n2 / 8; n *= 2) - ; - swhash = malloc(sizeof(struct swblock *) * n, M_VMPGDATA, M_WAITOK | M_ZERO); - swhash_mask = n - 1; - mtx_init(&swhash_mtx, "swap_pager swhash", NULL, MTX_DEF); + n = uma_zone_get_max(swblk_zone); + + if (n < n2) + printf("Swap blk zone entries reduced from %lu to %lu.\n", + n2, n); + swap_maxpages = n * SWAP_META_PAGES; + swzone = n * sizeof(struct swblk); + if (!uma_zone_reserve_kva(swpctrie_zone, n)) + printf("Cannot reserve swap pctrie zone, " + "reduce kern.maxswzone.\n"); } static vm_object_t @@ -581,14 +569,20 @@ swap_pager_alloc_init(void *handle, struct ucred *cred, vm_ooffset_t size, return (NULL); crhold(cred); } + + /* + * The un_pager.swp.swp_blks trie is initialized by + * vm_object_allocate() to ensure the correct order of + * visibility to other threads. + */ object = vm_object_allocate(OBJT_SWAP, OFF_TO_IDX(offset + PAGE_MASK + size)); + object->handle = handle; if (cred != NULL) { object->cred = cred; object->charge = size; } - object->un_pager.swp.swp_bcount = 0; return (object); } @@ -798,6 +792,36 @@ swp_pager_freeswapspace(daddr_t blk, int npages) } /* + * SYSCTL_SWAP_FRAGMENTATION() - produce raw swap space stats + */ +static int +sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS) +{ + struct sbuf sbuf; + struct swdevt *sp; + const char *devname; + int error; + + error = sysctl_wire_old_buffer(req, 0); + if (error != 0) + return (error); + sbuf_new_for_sysctl(&sbuf, NULL, 128, req); + mtx_lock(&sw_dev_mtx); + TAILQ_FOREACH(sp, &swtailq, sw_list) { + if (vn_isdisk(sp->sw_vp, NULL)) + devname = devtoname(sp->sw_vp->v_rdev); + else + devname = "[file]"; + sbuf_printf(&sbuf, "\nFree space on device %s:\n", devname); + blist_stats(sp->sw_blist, &sbuf); + } + mtx_unlock(&sw_dev_mtx); + error = sbuf_finish(&sbuf); + sbuf_delete(&sbuf); + return (error); +} + +/* * SWAP_PAGER_FREESPACE() - frees swap blocks associated with a page * range within an object. * @@ -1507,7 +1531,7 @@ swp_pager_async_iodone(struct buf *bp) * so it doesn't clog the inactive list, * then finish the I/O. */ - vm_page_dirty(m); + MPASS(m->dirty == VM_PAGE_BITS_ALL); vm_page_lock(m); vm_page_activate(m); vm_page_unlock(m); @@ -1643,50 +1667,56 @@ swp_pager_force_pagein(vm_object_t object, vm_pindex_t pindex) static void swap_pager_swapoff(struct swdevt *sp) { - struct swblock *swap; - vm_object_t locked_obj, object; - vm_pindex_t pindex; - int i, j, retries; + struct swblk *sb; + vm_object_t object; + vm_pindex_t pi; + int i, retries; sx_assert(&swdev_syscall_lock, SA_XLOCKED); retries = 0; - locked_obj = NULL; full_rescan: - mtx_lock(&swhash_mtx); - for (i = 0; i <= swhash_mask; i++) { /* '<=' is correct here */ -restart: - for (swap = swhash[i]; swap != NULL; swap = swap->swb_hnext) { - object = swap->swb_object; - pindex = swap->swb_index; - for (j = 0; j < SWAP_META_PAGES; ++j) { - if (!swp_pager_isondev(swap->swb_pages[j], sp)) + mtx_lock(&vm_object_list_mtx); + TAILQ_FOREACH(object, &vm_object_list, object_list) { + if (object->type != OBJT_SWAP) + continue; + mtx_unlock(&vm_object_list_mtx); + /* Depends on type-stability. */ + VM_OBJECT_WLOCK(object); + + /* + * Dead objects are eventually terminated on their own. + */ + if ((object->flags & OBJ_DEAD) != 0) + goto next_obj; + + /* + * Sync with fences placed after pctrie + * initialization. We must not access pctrie below + * unless we checked that our object is swap and not + * dead. + */ + atomic_thread_fence_acq(); + if (object->type != OBJT_SWAP) + goto next_obj; + + for (pi = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( + &object->un_pager.swp.swp_blks, pi)) != NULL; ) { + pi = sb->p + SWAP_META_PAGES; + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] == SWAPBLK_NONE) continue; - if (locked_obj != object) { - if (locked_obj != NULL) - VM_OBJECT_WUNLOCK(locked_obj); - locked_obj = object; - if (!VM_OBJECT_TRYWLOCK(object)) { - mtx_unlock(&swhash_mtx); - /* Depends on type-stability. */ - VM_OBJECT_WLOCK(object); - mtx_lock(&swhash_mtx); - goto restart; - } - } - MPASS(locked_obj == object); - mtx_unlock(&swhash_mtx); - swp_pager_force_pagein(object, pindex + j); - mtx_lock(&swhash_mtx); - goto restart; + if (swp_pager_isondev(sb->d[i], sp)) + swp_pager_force_pagein(object, + sb->p + i); } } +next_obj: + VM_OBJECT_WUNLOCK(object); + mtx_lock(&vm_object_list_mtx); } - mtx_unlock(&swhash_mtx); - if (locked_obj != NULL) { - VM_OBJECT_WUNLOCK(locked_obj); - locked_obj = NULL; - } + mtx_unlock(&vm_object_list_mtx); + if (sp->sw_used) { /* * Objects may be locked or paging to the device being @@ -1729,85 +1759,120 @@ restart: static void swp_pager_meta_build(vm_object_t object, vm_pindex_t pindex, daddr_t swapblk) { - static volatile int exhausted; - struct swblock *swap; - struct swblock **pswap; - int idx; + static volatile int swblk_zone_exhausted, swpctrie_zone_exhausted; + struct swblk *sb, *sb1; + vm_pindex_t modpi, rdpi; + int error, i; VM_OBJECT_ASSERT_WLOCKED(object); + /* * Convert default object to swap object if necessary */ if (object->type != OBJT_SWAP) { + pctrie_init(&object->un_pager.swp.swp_blks); + + /* + * Ensure that swap_pager_swapoff()'s iteration over + * object_list does not see a garbage pctrie. + */ + atomic_thread_fence_rel(); + object->type = OBJT_SWAP; - object->un_pager.swp.swp_bcount = 0; KASSERT(object->handle == NULL, ("default pager with handle")); } - /* - * Locate hash entry. If not found create, but if we aren't adding - * anything just return. If we run out of space in the map we wait - * and, since the hash table may have changed, retry. - */ -retry: - mtx_lock(&swhash_mtx); - pswap = swp_pager_hash(object, pindex); - - if ((swap = *pswap) == NULL) { - int i; - + rdpi = rounddown(pindex, SWAP_META_PAGES); + sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); + if (sb == NULL) { if (swapblk == SWAPBLK_NONE) - goto done; - - swap = *pswap = uma_zalloc(swap_zone, M_NOWAIT | - (curproc == pageproc ? M_USE_RESERVE : 0)); - if (swap == NULL) { - mtx_unlock(&swhash_mtx); + return; + for (;;) { + sb = uma_zalloc(swblk_zone, M_NOWAIT | (curproc == + pageproc ? M_USE_RESERVE : 0)); + if (sb != NULL) { + sb->p = rdpi; + for (i = 0; i < SWAP_META_PAGES; i++) + sb->d[i] = SWAPBLK_NONE; + if (atomic_cmpset_int(&swblk_zone_exhausted, + 1, 0)) + printf("swblk zone ok\n"); + break; + } VM_OBJECT_WUNLOCK(object); - if (uma_zone_exhausted(swap_zone)) { - if (atomic_cmpset_int(&exhausted, 0, 1)) - printf("swap zone exhausted, " + if (uma_zone_exhausted(swblk_zone)) { + if (atomic_cmpset_int(&swblk_zone_exhausted, + 0, 1)) + printf("swap blk zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); - pause("swzonex", 10); + pause("swzonxb", 10); } else VM_WAIT; VM_OBJECT_WLOCK(object); - goto retry; + sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, + rdpi); + if (sb != NULL) + /* + * Somebody swapped out a nearby page, + * allocating swblk at the rdpi index, + * while we dropped the object lock. + */ + goto allocated; + } + for (;;) { + error = SWAP_PCTRIE_INSERT( + &object->un_pager.swp.swp_blks, sb); + if (error == 0) { + if (atomic_cmpset_int(&swpctrie_zone_exhausted, + 1, 0)) + printf("swpctrie zone ok\n"); + break; + } + VM_OBJECT_WUNLOCK(object); + if (uma_zone_exhausted(swpctrie_zone)) { + if (atomic_cmpset_int(&swpctrie_zone_exhausted, + 0, 1)) + printf("swap pctrie zone exhausted, " + "increase kern.maxswzone\n"); + vm_pageout_oom(VM_OOM_SWAPZ); + pause("swzonxp", 10); + } else + VM_WAIT; + VM_OBJECT_WLOCK(object); + sb1 = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, + rdpi); + if (sb1 != NULL) { + uma_zfree(swblk_zone, sb); + sb = sb1; + goto allocated; + } } - - if (atomic_cmpset_int(&exhausted, 1, 0)) - printf("swap zone ok\n"); - - swap->swb_hnext = NULL; - swap->swb_object = object; - swap->swb_index = pindex & ~(vm_pindex_t)SWAP_META_MASK; - swap->swb_count = 0; - - ++object->un_pager.swp.swp_bcount; - - for (i = 0; i < SWAP_META_PAGES; ++i) - swap->swb_pages[i] = SWAPBLK_NONE; } +allocated: + MPASS(sb->p == rdpi); - /* - * Delete prior contents of metadata - */ - idx = pindex & SWAP_META_MASK; - - if (swap->swb_pages[idx] != SWAPBLK_NONE) { - swp_pager_freeswapspace(swap->swb_pages[idx], 1); - --swap->swb_count; - } + modpi = pindex % SWAP_META_PAGES; + /* Delete prior contents of metadata. */ + if (sb->d[modpi] != SWAPBLK_NONE) + swp_pager_freeswapspace(sb->d[modpi], 1); + /* Enter block into metadata. */ + sb->d[modpi] = swapblk; /* - * Enter block into metadata + * Free the swblk if we end up with the empty page run. */ - swap->swb_pages[idx] = swapblk; - if (swapblk != SWAPBLK_NONE) - ++swap->swb_count; -done: - mtx_unlock(&swhash_mtx); + if (swapblk == SWAPBLK_NONE) { + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] != SWAPBLK_NONE) + break; + } + if (i == SWAP_META_PAGES) { + SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, + rdpi); + uma_zfree(swblk_zone, sb); + } + } } /* @@ -1821,42 +1886,40 @@ done: * with resident pages. */ static void -swp_pager_meta_free(vm_object_t object, vm_pindex_t index, vm_pindex_t count) +swp_pager_meta_free(vm_object_t object, vm_pindex_t pindex, vm_pindex_t count) { - struct swblock **pswap, *swap; - vm_pindex_t c; - daddr_t v; - int n, sidx; + struct swblk *sb; + vm_pindex_t last; + int i; + bool empty; - VM_OBJECT_ASSERT_LOCKED(object); + VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP || count == 0) return; - mtx_lock(&swhash_mtx); - for (c = 0; c < count;) { - pswap = swp_pager_hash(object, index); - sidx = index & SWAP_META_MASK; - n = SWAP_META_PAGES - sidx; - index += n; - if ((swap = *pswap) == NULL) { - c += n; - continue; - } - for (; c < count && sidx < SWAP_META_PAGES; ++c, ++sidx) { - if ((v = swap->swb_pages[sidx]) == SWAPBLK_NONE) + last = pindex + count - 1; + for (;;) { + sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, + rounddown(pindex, SWAP_META_PAGES)); + if (sb == NULL || sb->p > last) + break; + empty = true; + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] == SWAPBLK_NONE) continue; - swp_pager_freeswapspace(v, 1); - swap->swb_pages[sidx] = SWAPBLK_NONE; - if (--swap->swb_count == 0) { - *pswap = swap->swb_hnext; - uma_zfree(swap_zone, swap); - --object->un_pager.swp.swp_bcount; - c += SWAP_META_PAGES - sidx; - break; - } + if (pindex <= sb->p + i && sb->p + i <= last) { + swp_pager_freeswapspace(sb->d[i], 1); + sb->d[i] = SWAPBLK_NONE; + } else + empty = false; + } + pindex = sb->p + SWAP_META_PAGES; + if (empty) { + SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, + sb->p); + uma_zfree(swblk_zone, sb); } } - mtx_unlock(&swhash_mtx); } /* @@ -1868,36 +1931,23 @@ swp_pager_meta_free(vm_object_t object, vm_pindex_t index, vm_pindex_t count) static void swp_pager_meta_free_all(vm_object_t object) { - struct swblock **pswap, *swap; - vm_pindex_t index; - daddr_t v; + struct swblk *sb; + vm_pindex_t pindex; int i; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP) return; - index = 0; - while (object->un_pager.swp.swp_bcount != 0) { - mtx_lock(&swhash_mtx); - pswap = swp_pager_hash(object, index); - if ((swap = *pswap) != NULL) { - for (i = 0; i < SWAP_META_PAGES; ++i) { - v = swap->swb_pages[i]; - if (v != SWAPBLK_NONE) { - --swap->swb_count; - swp_pager_freeswapspace(v, 1); - } - } - if (swap->swb_count != 0) - panic( - "swap_pager_meta_free_all: swb_count != 0"); - *pswap = swap->swb_hnext; - uma_zfree(swap_zone, swap); - --object->un_pager.swp.swp_bcount; + for (pindex = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( + &object->un_pager.swp.swp_blks, pindex)) != NULL;) { + pindex = sb->p + SWAP_META_PAGES; + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] != SWAPBLK_NONE) + swp_pager_freeswapspace(sb->d[i], 1); } - mtx_unlock(&swhash_mtx); - index += SWAP_META_PAGES; + SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, sb->p); + uma_zfree(swblk_zone, sb); } } @@ -1911,9 +1961,6 @@ swp_pager_meta_free_all(vm_object_t object) * was invalid. This routine will automatically free any invalid * meta-data swapblks. * - * It is not possible to store invalid swapblks in the swap meta data - * (other then a literal 'SWAPBLK_NONE'), so we don't bother checking. - * * When acting on a busy resident page and paging is in progress, we * have to wait until paging is complete but otherwise can act on the * busy page. @@ -1924,43 +1971,45 @@ swp_pager_meta_free_all(vm_object_t object) static daddr_t swp_pager_meta_ctl(vm_object_t object, vm_pindex_t pindex, int flags) { - struct swblock **pswap; - struct swblock *swap; + struct swblk *sb; daddr_t r1; - int idx; + int i; + + if ((flags & (SWM_FREE | SWM_POP)) != 0) + VM_OBJECT_ASSERT_WLOCKED(object); + else + VM_OBJECT_ASSERT_LOCKED(object); - VM_OBJECT_ASSERT_LOCKED(object); /* - * The meta data only exists of the object is OBJT_SWAP + * The meta data only exists if the object is OBJT_SWAP * and even then might not be allocated yet. */ if (object->type != OBJT_SWAP) return (SWAPBLK_NONE); - r1 = SWAPBLK_NONE; - mtx_lock(&swhash_mtx); - pswap = swp_pager_hash(object, pindex); - - if ((swap = *pswap) != NULL) { - idx = pindex & SWAP_META_MASK; - r1 = swap->swb_pages[idx]; - - if (r1 != SWAPBLK_NONE) { - if (flags & SWM_FREE) { - swp_pager_freeswapspace(r1, 1); - r1 = SWAPBLK_NONE; - } - if (flags & (SWM_FREE|SWM_POP)) { - swap->swb_pages[idx] = SWAPBLK_NONE; - if (--swap->swb_count == 0) { - *pswap = swap->swb_hnext; - uma_zfree(swap_zone, swap); - --object->un_pager.swp.swp_bcount; - } - } + sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, + rounddown(pindex, SWAP_META_PAGES)); + if (sb == NULL) + return (SWAPBLK_NONE); + r1 = sb->d[pindex % SWAP_META_PAGES]; + if (r1 == SWAPBLK_NONE) + return (SWAPBLK_NONE); + if ((flags & (SWM_FREE | SWM_POP)) != 0) { + sb->d[pindex % SWAP_META_PAGES] = SWAPBLK_NONE; + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] != SWAPBLK_NONE) + break; + } + if (i == SWAP_META_PAGES) { + SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, + rounddown(pindex, SWAP_META_PAGES)); + uma_zfree(swblk_zone, sb); } } - mtx_unlock(&swhash_mtx); + if ((flags & SWM_FREE) != 0) { + swp_pager_freeswapspace(r1, 1); + r1 = SWAPBLK_NONE; + } return (r1); } @@ -1974,32 +2023,38 @@ swp_pager_meta_ctl(vm_object_t object, vm_pindex_t pindex, int flags) vm_pindex_t swap_pager_find_least(vm_object_t object, vm_pindex_t pindex) { - struct swblock **pswap, *swap; - vm_pindex_t i, j, lim; - int idx; + struct swblk *sb; + int i; VM_OBJECT_ASSERT_LOCKED(object); - if (object->type != OBJT_SWAP || object->un_pager.swp.swp_bcount == 0) + if (object->type != OBJT_SWAP) return (object->size); - mtx_lock(&swhash_mtx); - for (j = pindex; j < object->size; j = lim) { - pswap = swp_pager_hash(object, j); - lim = rounddown2(j + SWAP_META_PAGES, SWAP_META_PAGES); - if (lim > object->size) - lim = object->size; - if ((swap = *pswap) != NULL) { - for (idx = j & SWAP_META_MASK, i = j; i < lim; - i++, idx++) { - if (swap->swb_pages[idx] != SWAPBLK_NONE) - goto found; - } + sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, + rounddown(pindex, SWAP_META_PAGES)); + if (sb == NULL) + return (object->size); + if (sb->p < pindex) { + for (i = pindex % SWAP_META_PAGES; i < SWAP_META_PAGES; i++) { + if (sb->d[i] != SWAPBLK_NONE) + return (sb->p + i); } + sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, + roundup(pindex, SWAP_META_PAGES)); + if (sb == NULL) + return (object->size); } - i = object->size; -found: - mtx_unlock(&swhash_mtx); - return (i); + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->d[i] != SWAPBLK_NONE) + return (sb->p + i); + } + + /* + * We get here if a swblk is present in the trie but it + * doesn't map any blocks. + */ + MPASS(0); + return (object->size); } /* @@ -2035,7 +2090,7 @@ sys_swapon(struct thread *td, struct swapon_args *uap) * Swap metadata may not fit in the KVM if we have physical * memory of >1GB. */ - if (swap_zone == NULL) { + if (swblk_zone == NULL) { error = ENOMEM; goto done; } @@ -2071,15 +2126,16 @@ done: /* * Check that the total amount of swap currently configured does not * exceed half the theoretical maximum. If it does, print a warning - * message and return -1; otherwise, return 0. + * message. */ -static int -swapon_check_swzone(unsigned long npages) +static void +swapon_check_swzone(void) { - unsigned long maxpages; + unsigned long maxpages, npages; + npages = swap_total / PAGE_SIZE; /* absolute maximum we can handle assuming 100% efficiency */ - maxpages = uma_zone_get_max(swap_zone) * SWAP_META_PAGES; + maxpages = uma_zone_get_max(swblk_zone) * SWAP_META_PAGES; /* recommend using no more than half that amount */ if (npages > maxpages / 2) { @@ -2088,9 +2144,7 @@ swapon_check_swzone(unsigned long npages) npages, maxpages / 2); printf("warning: increase kern.maxswzone " "or reduce amount of swap.\n"); - return (-1); } - return (0); } static void @@ -2158,7 +2212,7 @@ swaponsomething(struct vnode *vp, void *id, u_long nblks, nswapdev++; swap_pager_avail += nblks - 2; swap_total += (vm_ooffset_t)nblks * PAGE_SIZE; - swapon_check_swzone(swap_total / PAGE_SIZE); + swapon_check_swzone(); swp_sizecheck(); mtx_unlock(&sw_dev_mtx); } @@ -2379,15 +2433,9 @@ SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD | CTLFLAG_MPSAFE, "Swap statistics by device"); /* - * vmspace_swap_count() - count the approximate swap usage in pages for a - * vmspace. - * - * The map must be locked. - * - * Swap usage is determined by taking the proportional swap used by - * VM objects backing the VM map. To make up for fractional losses, - * if the VM object has any swap use at all the associated map entries - * count for at least 1 swap page. + * Count the approximate swap usage in pages for a vmspace. The + * shadowed or not yet copied on write swap blocks are not accounted. + * The map must be locked. */ long vmspace_swap_count(struct vmspace *vmspace) @@ -2395,23 +2443,38 @@ vmspace_swap_count(struct vmspace *vmspace) vm_map_t map; vm_map_entry_t cur; vm_object_t object; - long count, n; + struct swblk *sb; + vm_pindex_t e, pi; + long count; + int i; map = &vmspace->vm_map; count = 0; for (cur = map->header.next; cur != &map->header; cur = cur->next) { - if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && - (object = cur->object.vm_object) != NULL) { - VM_OBJECT_WLOCK(object); - if (object->type == OBJT_SWAP && - object->un_pager.swp.swp_bcount != 0) { - n = (cur->end - cur->start) / PAGE_SIZE; - count += object->un_pager.swp.swp_bcount * - SWAP_META_PAGES * n / object->size + 1; + if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) + continue; + object = cur->object.vm_object; + if (object == NULL || object->type != OBJT_SWAP) + continue; + VM_OBJECT_RLOCK(object); + if (object->type != OBJT_SWAP) + goto unlock; + pi = OFF_TO_IDX(cur->offset); + e = pi + OFF_TO_IDX(cur->end - cur->start); + for (;; pi = sb->p + SWAP_META_PAGES) { + sb = SWAP_PCTRIE_LOOKUP_GE( + &object->un_pager.swp.swp_blks, pi); + if (sb == NULL || sb->p >= e) + break; + for (i = 0; i < SWAP_META_PAGES; i++) { + if (sb->p + i < e && + sb->d[i] != SWAPBLK_NONE) + count++; } - VM_OBJECT_WUNLOCK(object); } +unlock: + VM_OBJECT_RUNLOCK(object); } return (count); } diff --git a/sys/vm/swap_pager.h b/sys/vm/swap_pager.h index 0131297..1abded5 100644 --- a/sys/vm/swap_pager.h +++ b/sys/vm/swap_pager.h @@ -73,7 +73,6 @@ struct swdevt { #ifdef _KERNEL -extern int swap_pager_full; extern int swap_pager_avail; struct xswdev; diff --git a/sys/vm/uma.h b/sys/vm/uma.h index f4c2de8..55b9e61 100644 --- a/sys/vm/uma.h +++ b/sys/vm/uma.h @@ -296,6 +296,7 @@ uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor, #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */ #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */ #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */ +#define UMA_ALIGNOF(type) (_Alignof(type) - 1) /* Alignment fit for 'type' */ /* * Destroys an empty uma zone. If the zone is not empty uma complains loudly. diff --git a/sys/vm/uma_core.c b/sys/vm/uma_core.c index 26439dc..8504a72 100644 --- a/sys/vm/uma_core.c +++ b/sys/vm/uma_core.c @@ -1326,10 +1326,6 @@ keg_large_init(uma_keg_t keg) keg->uk_ipers = 1; keg->uk_rsize = keg->uk_size; - /* We can't do OFFPAGE if we're internal, bail out here. */ - if (keg->uk_flags & UMA_ZFLAG_INTERNAL) - return; - /* Check whether we have enough space to not do OFFPAGE. */ if ((keg->uk_flags & UMA_ZONE_OFFPAGE) == 0) { shsize = sizeof(struct uma_slab); @@ -1337,8 +1333,17 @@ keg_large_init(uma_keg_t keg) shsize = (shsize & ~UMA_ALIGN_PTR) + (UMA_ALIGN_PTR + 1); - if ((PAGE_SIZE * keg->uk_ppera) - keg->uk_rsize < shsize) - keg->uk_flags |= UMA_ZONE_OFFPAGE; + if (PAGE_SIZE * keg->uk_ppera - keg->uk_rsize < shsize) { + /* + * We can't do OFFPAGE if we're internal, in which case + * we need an extra page per allocation to contain the + * slab header. + */ + if ((keg->uk_flags & UMA_ZFLAG_INTERNAL) == 0) + keg->uk_flags |= UMA_ZONE_OFFPAGE; + else + keg->uk_ppera++; + } } if ((keg->uk_flags & UMA_ZONE_OFFPAGE) && diff --git a/sys/vm/uma_int.h b/sys/vm/uma_int.h index 8423d1c..ee315ab 100644 --- a/sys/vm/uma_int.h +++ b/sys/vm/uma_int.h @@ -28,6 +28,7 @@ * */ +#include <sys/_bitset.h> #include <sys/_task.h> /* @@ -210,7 +211,7 @@ struct uma_keg { vm_offset_t uk_kva; /* Zone base KVA */ uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */ - uint16_t uk_pgoff; /* Offset to uma_slab struct */ + uint32_t uk_pgoff; /* Offset to uma_slab struct */ uint16_t uk_ppera; /* pages per allocation from backend */ uint16_t uk_ipers; /* Items per slab */ uint32_t uk_flags; /* Internal flags */ diff --git a/sys/vm/vm_extern.h b/sys/vm/vm_extern.h index 0489b6e..8e55273 100644 --- a/sys/vm/vm_extern.h +++ b/sys/vm/vm_extern.h @@ -71,7 +71,6 @@ void kmem_init(vm_offset_t, vm_offset_t); void kmem_init_zero_region(void); void kmeminit(void); -void swapout_procs(int); int kernacc(void *, int, int); int useracc(void *, int, int); int vm_fault(vm_map_t, vm_offset_t, vm_prot_t, int); diff --git a/sys/vm/vm_fault.c b/sys/vm/vm_fault.c index d2147f6..92c761b 100644 --- a/sys/vm/vm_fault.c +++ b/sys/vm/vm_fault.c @@ -236,14 +236,15 @@ vm_fault_dirty(vm_map_entry_t entry, vm_page_t m, vm_prot_t prot, * written NOW so dirty it explicitly to save on * pmap_is_modified() calls later. * - * Also tell the backing pager, if any, that it should remove - * any swap backing since the page is now dirty. + * Also, since the page is now dirty, we can possibly tell + * the pager to release any swap backing the page. Calling + * the pager requires a write lock on the object. */ if (need_dirty) vm_page_dirty(m); if (!set_wd) vm_page_unlock(m); - if (need_dirty) + else if (need_dirty) vm_pager_page_unswapped(m); } @@ -266,8 +267,12 @@ static int vm_fault_soft_fast(struct faultstate *fs, vm_offset_t vaddr, vm_prot_t prot, int fault_type, int fault_flags, boolean_t wired, vm_page_t *m_hold) { - vm_page_t m; - int rv; + vm_page_t m, m_map; +#if defined(__amd64__) && VM_NRESERVLEVEL > 0 + vm_page_t m_super; + int flags; +#endif + int psind, rv; MPASS(fs->vp == NULL); m = vm_page_lookup(fs->first_object, fs->first_pindex); @@ -275,14 +280,46 @@ vm_fault_soft_fast(struct faultstate *fs, vm_offset_t vaddr, vm_prot_t prot, if (m == NULL || ((prot & VM_PROT_WRITE) != 0 && vm_page_busied(m)) || m->valid != VM_PAGE_BITS_ALL) return (KERN_FAILURE); - rv = pmap_enter(fs->map->pmap, vaddr, m, prot, fault_type | - PMAP_ENTER_NOSLEEP | (wired ? PMAP_ENTER_WIRED : 0), 0); + m_map = m; + psind = 0; +#if defined(__amd64__) && VM_NRESERVLEVEL > 0 + if ((m->flags & PG_FICTITIOUS) == 0 && + (m_super = vm_reserv_to_superpage(m)) != NULL && + rounddown2(vaddr, pagesizes[m_super->psind]) >= fs->entry->start && + roundup2(vaddr + 1, pagesizes[m_super->psind]) <= fs->entry->end && + (vaddr & (pagesizes[m_super->psind] - 1)) == (VM_PAGE_TO_PHYS(m) & + (pagesizes[m_super->psind] - 1)) && + pmap_ps_enabled(fs->map->pmap)) { + flags = PS_ALL_VALID; + if ((prot & VM_PROT_WRITE) != 0) { + /* + * Create a superpage mapping allowing write access + * only if none of the constituent pages are busy and + * all of them are already dirty (except possibly for + * the page that was faulted on). + */ + flags |= PS_NONE_BUSY; + if ((fs->first_object->flags & OBJ_UNMANAGED) == 0) + flags |= PS_ALL_DIRTY; + } + if (vm_page_ps_test(m_super, flags, m)) { + m_map = m_super; + psind = m_super->psind; + vaddr = rounddown2(vaddr, pagesizes[psind]); + /* Preset the modified bit for dirty superpages. */ + if ((flags & PS_ALL_DIRTY) != 0) + fault_type |= VM_PROT_WRITE; + } + } +#endif + rv = pmap_enter(fs->map->pmap, vaddr, m_map, prot, fault_type | + PMAP_ENTER_NOSLEEP | (wired ? PMAP_ENTER_WIRED : 0), psind); if (rv != KERN_SUCCESS) return (rv); vm_fault_fill_hold(m_hold, m); vm_fault_dirty(fs->entry, m, prot, fault_type, fault_flags, false); VM_OBJECT_RUNLOCK(fs->first_object); - if (!wired) + if (psind == 0 && !wired) vm_fault_prefault(fs, vaddr, PFBAK, PFFOR); vm_map_lookup_done(fs->map, fs->entry); curthread->td_ru.ru_minflt++; diff --git a/sys/vm/vm_glue.c b/sys/vm/vm_glue.c index 60b822e..14ec78a 100644 --- a/sys/vm/vm_glue.c +++ b/sys/vm/vm_glue.c @@ -101,13 +101,6 @@ __FBSDID("$FreeBSD$"); #include <machine/cpu.h> -#ifndef NO_SWAPPING -static int swapout(struct proc *); -static void swapclear(struct proc *); -static void vm_thread_swapin(struct thread *td); -static void vm_thread_swapout(struct thread *td); -#endif - /* * MPSAFE * @@ -308,10 +301,6 @@ SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0, SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0, ""); -#ifndef KSTACK_MAX_PAGES -#define KSTACK_MAX_PAGES 32 -#endif - /* * Create the kernel stack (including pcb for i386) for a new thread. * This routine directly affects the fork perf for a process and @@ -322,7 +311,7 @@ vm_thread_new(struct thread *td, int pages) { vm_object_t ksobj; vm_offset_t ks; - vm_page_t m, ma[KSTACK_MAX_PAGES]; + vm_page_t ma[KSTACK_MAX_PAGES]; struct kstack_cache_entry *ks_ce; int i; @@ -391,15 +380,10 @@ vm_thread_new(struct thread *td, int pages) * page of stack. */ VM_OBJECT_WLOCK(ksobj); - for (i = 0; i < pages; i++) { - /* - * Get a kernel stack page. - */ - m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY | - VM_ALLOC_NORMAL | VM_ALLOC_WIRED); - ma[i] = m; - m->valid = VM_PAGE_BITS_ALL; - } + (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | + VM_ALLOC_WIRED, ma, pages); + for (i = 0; i < pages; i++) + ma[i]->valid = VM_PAGE_BITS_ALL; VM_OBJECT_WUNLOCK(ksobj); pmap_qenter(ks, ma, pages); return (1); @@ -532,80 +516,6 @@ intr_prof_stack_use(struct thread *td, struct trapframe *frame) } #endif /* KSTACK_USAGE_PROF */ -#ifndef NO_SWAPPING -/* - * Allow a thread's kernel stack to be paged out. - */ -static void -vm_thread_swapout(struct thread *td) -{ - vm_object_t ksobj; - vm_page_t m; - int i, pages; - - cpu_thread_swapout(td); - pages = td->td_kstack_pages; - ksobj = td->td_kstack_obj; - pmap_qremove(td->td_kstack, pages); - VM_OBJECT_WLOCK(ksobj); - for (i = 0; i < pages; i++) { - m = vm_page_lookup(ksobj, i); - if (m == NULL) - panic("vm_thread_swapout: kstack already missing?"); - vm_page_dirty(m); - vm_page_lock(m); - vm_page_unwire(m, PQ_INACTIVE); - vm_page_unlock(m); - } - VM_OBJECT_WUNLOCK(ksobj); -} - -/* - * Bring the kernel stack for a specified thread back in. - */ -static void -vm_thread_swapin(struct thread *td) -{ - vm_object_t ksobj; - vm_page_t ma[KSTACK_MAX_PAGES]; - int pages; - - pages = td->td_kstack_pages; - ksobj = td->td_kstack_obj; - VM_OBJECT_WLOCK(ksobj); - for (int i = 0; i < pages; i++) - ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | - VM_ALLOC_WIRED); - for (int i = 0; i < pages;) { - int j, a, count, rv; - - vm_page_assert_xbusied(ma[i]); - if (ma[i]->valid == VM_PAGE_BITS_ALL) { - vm_page_xunbusy(ma[i]); - i++; - continue; - } - vm_object_pip_add(ksobj, 1); - for (j = i + 1; j < pages; j++) - if (ma[j]->valid == VM_PAGE_BITS_ALL) - break; - rv = vm_pager_has_page(ksobj, ma[i]->pindex, NULL, &a); - KASSERT(rv == 1, ("%s: missing page %p", __func__, ma[i])); - count = min(a + 1, j - i); - rv = vm_pager_get_pages(ksobj, ma + i, count, NULL, NULL); - KASSERT(rv == VM_PAGER_OK, ("%s: cannot get kstack for proc %d", - __func__, td->td_proc->p_pid)); - vm_object_pip_wakeup(ksobj); - for (j = i; j < i + count; j++) - vm_page_xunbusy(ma[j]); - i += count; - } - VM_OBJECT_WUNLOCK(ksobj); - pmap_qenter(td->td_kstack, ma, pages); - cpu_thread_swapin(td); -} -#endif /* !NO_SWAPPING */ - /* * Implement fork's actions on an address space. * Here we arrange for the address space to be copied or referenced, @@ -679,425 +589,8 @@ vm_waitproc(p) } void -faultin(p) - struct proc *p; -{ -#ifdef NO_SWAPPING - - PROC_LOCK_ASSERT(p, MA_OWNED); - if ((p->p_flag & P_INMEM) == 0) - panic("faultin: proc swapped out with NO_SWAPPING!"); -#else /* !NO_SWAPPING */ - struct thread *td; - - PROC_LOCK_ASSERT(p, MA_OWNED); - /* - * If another process is swapping in this process, - * just wait until it finishes. - */ - if (p->p_flag & P_SWAPPINGIN) { - while (p->p_flag & P_SWAPPINGIN) - msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0); - return; - } - if ((p->p_flag & P_INMEM) == 0) { - /* - * Don't let another thread swap process p out while we are - * busy swapping it in. - */ - ++p->p_lock; - p->p_flag |= P_SWAPPINGIN; - PROC_UNLOCK(p); - - /* - * We hold no lock here because the list of threads - * can not change while all threads in the process are - * swapped out. - */ - FOREACH_THREAD_IN_PROC(p, td) - vm_thread_swapin(td); - PROC_LOCK(p); - swapclear(p); - p->p_swtick = ticks; - - wakeup(&p->p_flag); - - /* Allow other threads to swap p out now. */ - --p->p_lock; - } -#endif /* NO_SWAPPING */ -} - -/* - * This swapin algorithm attempts to swap-in processes only if there - * is enough space for them. Of course, if a process waits for a long - * time, it will be swapped in anyway. - */ -void -swapper(void) -{ - struct proc *p; - struct thread *td; - struct proc *pp; - int slptime; - int swtime; - int ppri; - int pri; - -loop: - if (vm_page_count_min()) { - VM_WAIT; - goto loop; - } - - pp = NULL; - ppri = INT_MIN; - sx_slock(&allproc_lock); - FOREACH_PROC_IN_SYSTEM(p) { - PROC_LOCK(p); - if (p->p_state == PRS_NEW || - p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) { - PROC_UNLOCK(p); - continue; - } - swtime = (ticks - p->p_swtick) / hz; - FOREACH_THREAD_IN_PROC(p, td) { - /* - * An otherwise runnable thread of a process - * swapped out has only the TDI_SWAPPED bit set. - * - */ - thread_lock(td); - if (td->td_inhibitors == TDI_SWAPPED) { - slptime = (ticks - td->td_slptick) / hz; - pri = swtime + slptime; - if ((td->td_flags & TDF_SWAPINREQ) == 0) - pri -= p->p_nice * 8; - /* - * if this thread is higher priority - * and there is enough space, then select - * this process instead of the previous - * selection. - */ - if (pri > ppri) { - pp = p; - ppri = pri; - } - } - thread_unlock(td); - } - PROC_UNLOCK(p); - } - sx_sunlock(&allproc_lock); - - /* - * Nothing to do, back to sleep. - */ - if ((p = pp) == NULL) { - tsleep(&proc0, PVM, "swapin", MAXSLP * hz / 2); - goto loop; - } - PROC_LOCK(p); - - /* - * Another process may be bringing or may have already - * brought this process in while we traverse all threads. - * Or, this process may even be being swapped out again. - */ - if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) { - PROC_UNLOCK(p); - goto loop; - } - - /* - * We would like to bring someone in. (only if there is space). - * [What checks the space? ] - */ - faultin(p); - PROC_UNLOCK(p); - goto loop; -} - -void kick_proc0(void) { wakeup(&proc0); } - -#ifndef NO_SWAPPING - -/* - * Swap_idle_threshold1 is the guaranteed swapped in time for a process - */ -static int swap_idle_threshold1 = 2; -SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, - &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process"); - -/* - * Swap_idle_threshold2 is the time that a process can be idle before - * it will be swapped out, if idle swapping is enabled. - */ -static int swap_idle_threshold2 = 10; -SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, - &swap_idle_threshold2, 0, "Time before a process will be swapped out"); - -/* - * First, if any processes have been sleeping or stopped for at least - * "swap_idle_threshold1" seconds, they are swapped out. If, however, - * no such processes exist, then the longest-sleeping or stopped - * process is swapped out. Finally, and only as a last resort, if - * there are no sleeping or stopped processes, the longest-resident - * process is swapped out. - */ -void -swapout_procs(action) -int action; -{ - struct proc *p; - struct thread *td; - int didswap = 0; - -retry: - sx_slock(&allproc_lock); - FOREACH_PROC_IN_SYSTEM(p) { - struct vmspace *vm; - int minslptime = 100000; - int slptime; - - PROC_LOCK(p); - /* - * Watch out for a process in - * creation. It may have no - * address space or lock yet. - */ - if (p->p_state == PRS_NEW) { - PROC_UNLOCK(p); - continue; - } - /* - * An aio daemon switches its - * address space while running. - * Perform a quick check whether - * a process has P_SYSTEM. - * Filter out exiting processes. - */ - if ((p->p_flag & (P_SYSTEM | P_WEXIT)) != 0) { - PROC_UNLOCK(p); - continue; - } - _PHOLD_LITE(p); - PROC_UNLOCK(p); - sx_sunlock(&allproc_lock); - - /* - * Do not swapout a process that - * is waiting for VM data - * structures as there is a possible - * deadlock. Test this first as - * this may block. - * - * Lock the map until swapout - * finishes, or a thread of this - * process may attempt to alter - * the map. - */ - vm = vmspace_acquire_ref(p); - if (vm == NULL) - goto nextproc2; - if (!vm_map_trylock(&vm->vm_map)) - goto nextproc1; - - PROC_LOCK(p); - if (p->p_lock != 1 || (p->p_flag & (P_STOPPED_SINGLE | - P_TRACED | P_SYSTEM)) != 0) - goto nextproc; - - /* - * only aiod changes vmspace, however it will be - * skipped because of the if statement above checking - * for P_SYSTEM - */ - if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM) - goto nextproc; - - switch (p->p_state) { - default: - /* Don't swap out processes in any sort - * of 'special' state. */ - break; - - case PRS_NORMAL: - /* - * do not swapout a realtime process - * Check all the thread groups.. - */ - FOREACH_THREAD_IN_PROC(p, td) { - thread_lock(td); - if (PRI_IS_REALTIME(td->td_pri_class)) { - thread_unlock(td); - goto nextproc; - } - slptime = (ticks - td->td_slptick) / hz; - /* - * Guarantee swap_idle_threshold1 - * time in memory. - */ - if (slptime < swap_idle_threshold1) { - thread_unlock(td); - goto nextproc; - } - - /* - * Do not swapout a process if it is - * waiting on a critical event of some - * kind or there is a thread whose - * pageable memory may be accessed. - * - * This could be refined to support - * swapping out a thread. - */ - if (!thread_safetoswapout(td)) { - thread_unlock(td); - goto nextproc; - } - /* - * If the system is under memory stress, - * or if we are swapping - * idle processes >= swap_idle_threshold2, - * then swap the process out. - */ - if (((action & VM_SWAP_NORMAL) == 0) && - (((action & VM_SWAP_IDLE) == 0) || - (slptime < swap_idle_threshold2))) { - thread_unlock(td); - goto nextproc; - } - - if (minslptime > slptime) - minslptime = slptime; - thread_unlock(td); - } - - /* - * If the pageout daemon didn't free enough pages, - * or if this process is idle and the system is - * configured to swap proactively, swap it out. - */ - if ((action & VM_SWAP_NORMAL) || - ((action & VM_SWAP_IDLE) && - (minslptime > swap_idle_threshold2))) { - _PRELE(p); - if (swapout(p) == 0) - didswap++; - PROC_UNLOCK(p); - vm_map_unlock(&vm->vm_map); - vmspace_free(vm); - goto retry; - } - } -nextproc: - PROC_UNLOCK(p); - vm_map_unlock(&vm->vm_map); -nextproc1: - vmspace_free(vm); -nextproc2: - sx_slock(&allproc_lock); - PRELE(p); - } - sx_sunlock(&allproc_lock); - /* - * If we swapped something out, and another process needed memory, - * then wakeup the sched process. - */ - if (didswap) - wakeup(&proc0); -} - -static void -swapclear(p) - struct proc *p; -{ - struct thread *td; - - PROC_LOCK_ASSERT(p, MA_OWNED); - - FOREACH_THREAD_IN_PROC(p, td) { - thread_lock(td); - td->td_flags |= TDF_INMEM; - td->td_flags &= ~TDF_SWAPINREQ; - TD_CLR_SWAPPED(td); - if (TD_CAN_RUN(td)) - if (setrunnable(td)) { -#ifdef INVARIANTS - /* - * XXX: We just cleared TDI_SWAPPED - * above and set TDF_INMEM, so this - * should never happen. - */ - panic("not waking up swapper"); -#endif - } - thread_unlock(td); - } - p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT); - p->p_flag |= P_INMEM; -} - -static int -swapout(p) - struct proc *p; -{ - struct thread *td; - - PROC_LOCK_ASSERT(p, MA_OWNED); -#if defined(SWAP_DEBUG) - printf("swapping out %d\n", p->p_pid); -#endif - - /* - * The states of this process and its threads may have changed - * by now. Assuming that there is only one pageout daemon thread, - * this process should still be in memory. - */ - KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM, - ("swapout: lost a swapout race?")); - - /* - * remember the process resident count - */ - p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); - /* - * Check and mark all threads before we proceed. - */ - p->p_flag &= ~P_INMEM; - p->p_flag |= P_SWAPPINGOUT; - FOREACH_THREAD_IN_PROC(p, td) { - thread_lock(td); - if (!thread_safetoswapout(td)) { - thread_unlock(td); - swapclear(p); - return (EBUSY); - } - td->td_flags &= ~TDF_INMEM; - TD_SET_SWAPPED(td); - thread_unlock(td); - } - td = FIRST_THREAD_IN_PROC(p); - ++td->td_ru.ru_nswap; - PROC_UNLOCK(p); - - /* - * This list is stable because all threads are now prevented from - * running. The list is only modified in the context of a running - * thread in this process. - */ - FOREACH_THREAD_IN_PROC(p, td) - vm_thread_swapout(td); - - PROC_LOCK(p); - p->p_flag &= ~P_SWAPPINGOUT; - p->p_swtick = ticks; - return (0); -} -#endif /* !NO_SWAPPING */ diff --git a/sys/vm/vm_kern.c b/sys/vm/vm_kern.c index e1c3db0..3778f76 100644 --- a/sys/vm/vm_kern.c +++ b/sys/vm/vm_kern.c @@ -121,8 +121,7 @@ SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD, * a mapping on demand through vm_fault() will result in a panic. */ vm_offset_t -kva_alloc(size) - vm_size_t size; +kva_alloc(vm_size_t size) { vm_offset_t addr; @@ -143,9 +142,7 @@ kva_alloc(size) * This routine may not block on kernel maps. */ void -kva_free(addr, size) - vm_offset_t addr; - vm_size_t size; +kva_free(vm_offset_t addr, vm_size_t size) { size = round_page(size); @@ -430,9 +427,7 @@ kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size) * This routine may block. */ vm_offset_t -kmap_alloc_wait(map, size) - vm_map_t map; - vm_size_t size; +kmap_alloc_wait(vm_map_t map, vm_size_t size) { vm_offset_t addr; @@ -470,10 +465,7 @@ kmap_alloc_wait(map, size) * waiting for memory in that map. */ void -kmap_free_wakeup(map, addr, size) - vm_map_t map; - vm_offset_t addr; - vm_size_t size; +kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size) { vm_map_lock(map); @@ -517,8 +509,7 @@ kmem_init_zero_region(void) * `start' as allocated, and the range between `start' and `end' as free. */ void -kmem_init(start, end) - vm_offset_t start, end; +kmem_init(vm_offset_t start, vm_offset_t end) { vm_map_t m; diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c index ce1696f..4a5eadf 100644 --- a/sys/vm/vm_map.c +++ b/sys/vm/vm_map.c @@ -1556,6 +1556,18 @@ again: return (result); } +/* + * vm_map_find_min() is a variant of vm_map_find() that takes an + * additional parameter (min_addr) and treats the given address + * (*addr) differently. Specifically, it treats *addr as a hint + * and not as the minimum address where the mapping is created. + * + * This function works in two phases. First, it tries to + * allocate above the hint. If that fails and the hint is + * greater than min_addr, it performs a second pass, replacing + * the hint with min_addr as the minimum address for the + * allocation. + */ int vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr, @@ -1962,7 +1974,7 @@ vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, (pagesizes[p->psind] - 1)) == 0) { mask = atop(pagesizes[p->psind]) - 1; if (tmpidx + mask < psize && - vm_page_ps_is_valid(p)) { + vm_page_ps_test(p, PS_ALL_VALID, NULL)) { p += mask; threshold += mask; } @@ -3610,12 +3622,13 @@ vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP), ("bi-dir stack")); - sgp = (vm_size_t)stack_guard_page * PAGE_SIZE; if (addrbos < vm_map_min(map) || - addrbos > vm_map_max(map) || - addrbos + max_ssize < addrbos || - sgp >= max_ssize) - return (KERN_NO_SPACE); + addrbos + max_ssize > vm_map_max(map) || + addrbos + max_ssize <= addrbos) + return (KERN_INVALID_ADDRESS); + sgp = (vm_size_t)stack_guard_page * PAGE_SIZE; + if (sgp >= max_ssize) + return (KERN_INVALID_ARGUMENT); init_ssize = growsize; if (max_ssize < init_ssize + sgp) diff --git a/sys/vm/vm_mmap.c b/sys/vm/vm_mmap.c index 17631f9..258a96e 100644 --- a/sys/vm/vm_mmap.c +++ b/sys/vm/vm_mmap.c @@ -1196,7 +1196,7 @@ vm_mmap_vnode(struct thread *td, vm_size_t objsize, { struct vattr va; vm_object_t obj; - vm_offset_t foff; + vm_ooffset_t foff; struct ucred *cred; int error, flags, locktype; diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c index 5bf9e26..11e1f84 100644 --- a/sys/vm/vm_object.c +++ b/sys/vm/vm_object.c @@ -73,6 +73,7 @@ __FBSDID("$FreeBSD$"); #include <sys/mman.h> #include <sys/mount.h> #include <sys/kernel.h> +#include <sys/pctrie.h> #include <sys/sysctl.h> #include <sys/mutex.h> #include <sys/proc.h> /* for curproc, pageproc */ @@ -208,6 +209,7 @@ vm_object_zinit(void *mem, int size, int flags) object->paging_in_progress = 0; object->resident_page_count = 0; object->shadow_count = 0; + object->flags = OBJ_DEAD; mtx_lock(&vm_object_list_mtx); TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); @@ -223,6 +225,16 @@ _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object) LIST_INIT(&object->shadow_head); object->type = type; + if (type == OBJT_SWAP) + pctrie_init(&object->un_pager.swp.swp_blks); + + /* + * Ensure that swap_pager_swapoff() iteration over object_list + * sees up to date type and pctrie head if it observed + * non-dead object. + */ + atomic_thread_fence_rel(); + switch (type) { case OBJT_DEAD: panic("_vm_object_allocate: can't create OBJT_DEAD"); @@ -694,6 +706,89 @@ vm_object_destroy(vm_object_t object) } /* + * vm_object_terminate_pages removes any remaining pageable pages + * from the object and resets the object to an empty state. + */ +static void +vm_object_terminate_pages(vm_object_t object) +{ + vm_page_t p, p_next; + struct mtx *mtx, *mtx1; + struct vm_pagequeue *pq, *pq1; + + VM_OBJECT_ASSERT_WLOCKED(object); + + mtx = NULL; + pq = NULL; + + /* + * Free any remaining pageable pages. This also removes them from the + * paging queues. However, don't free wired pages, just remove them + * from the object. Rather than incrementally removing each page from + * the object, the page and object are reset to any empty state. + */ + TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { + vm_page_assert_unbusied(p); + 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_unlock(pq); + pq = NULL; + } + mtx = mtx1; + mtx_lock(mtx); + } + } + p->object = NULL; + if (p->wire_count != 0) + goto unlist; + PCPU_INC(cnt.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_unlock(pq); + pq = pq1; + vm_pagequeue_lock(pq); + } + } + if (vm_page_free_prep(p, true)) + continue; +unlist: + TAILQ_REMOVE(&object->memq, p, listq); + } + if (pq != NULL) + vm_pagequeue_unlock(pq); + 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 + * modified by the preceding loop. + */ + if (object->resident_page_count != 0) { + vm_radix_reclaim_allnodes(&object->rtree); + TAILQ_INIT(&object->memq); + object->resident_page_count = 0; + if (object->type == OBJT_VNODE) + vdrop(object->handle); + } +} + +/* * vm_object_terminate actually destroys the specified object, freeing * up all previously used resources. * @@ -703,7 +798,6 @@ vm_object_destroy(vm_object_t object) void vm_object_terminate(vm_object_t object) { - vm_page_t p, p_next; VM_OBJECT_ASSERT_WLOCKED(object); @@ -746,41 +840,8 @@ vm_object_terminate(vm_object_t object) ("vm_object_terminate: object with references, ref_count=%d", object->ref_count)); - /* - * Free any remaining pageable pages. This also removes them from the - * paging queues. However, don't free wired pages, just remove them - * from the object. Rather than incrementally removing each page from - * the object, the page and object are reset to any empty state. - */ - TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { - vm_page_assert_unbusied(p); - vm_page_lock(p); - /* - * Optimize the page's removal from the object by resetting - * its "object" field. Specifically, if the page is not - * wired, then the effect of this assignment is that - * vm_page_free()'s call to vm_page_remove() will return - * immediately without modifying the page or the object. - */ - p->object = NULL; - if (p->wire_count == 0) { - vm_page_free(p); - PCPU_INC(cnt.v_pfree); - } - vm_page_unlock(p); - } - /* - * If the object contained any pages, then reset it to an empty state. - * None of the object's fields, including "resident_page_count", were - * modified by the preceding loop. - */ - if (object->resident_page_count != 0) { - vm_radix_reclaim_allnodes(&object->rtree); - TAILQ_INIT(&object->memq); - object->resident_page_count = 0; - if (object->type == OBJT_VNODE) - vdrop(object->handle); - } + if ((object->flags & OBJ_PG_DTOR) == 0) + vm_object_terminate_pages(object); #if VM_NRESERVLEVEL > 0 if (__predict_false(!LIST_EMPTY(&object->rvq))) @@ -1022,8 +1083,8 @@ vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, * I/O. */ if (object->type == OBJT_VNODE && - (object->flags & OBJ_MIGHTBEDIRTY) != 0) { - vp = object->handle; + (object->flags & OBJ_MIGHTBEDIRTY) != 0 && + ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WUNLOCK(object); (void) vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); @@ -1898,6 +1959,8 @@ vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options) { vm_page_t p, next; + struct mtx *mtx; + struct pglist pgl; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_UNMANAGED) == 0 || @@ -1906,8 +1969,10 @@ vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 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; /* * Here, the variable "p" is either (1) the page with the least pindex @@ -1924,7 +1989,7 @@ again: * however, be invalidated if the option OBJPR_CLEANONLY is * not specified. */ - vm_page_lock(p); + vm_page_change_lock(p, &mtx); if (vm_page_xbusied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmopax", true); @@ -1932,13 +1997,14 @@ again: goto again; } if (p->wire_count != 0) { - if ((options & OBJPR_NOTMAPPED) == 0) + if ((options & OBJPR_NOTMAPPED) == 0 && + object->ref_count != 0) pmap_remove_all(p); if ((options & OBJPR_CLEANONLY) == 0) { p->valid = 0; vm_page_undirty(p); } - goto next; + continue; } if (vm_page_busied(p)) { VM_OBJECT_WUNLOCK(object); @@ -1949,17 +2015,21 @@ again: KASSERT((p->flags & PG_FICTITIOUS) == 0, ("vm_object_page_remove: page %p is fictitious", p)); if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) { - if ((options & OBJPR_NOTMAPPED) == 0) + if ((options & OBJPR_NOTMAPPED) == 0 && + object->ref_count != 0) pmap_remove_write(p); - if (p->dirty) - goto next; + if (p->dirty != 0) + continue; } - if ((options & OBJPR_NOTMAPPED) == 0) + if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_all(p); - vm_page_free(p); -next: - vm_page_unlock(p); + p->flags &= ~PG_ZERO; + if (vm_page_free_prep(p, false)) + TAILQ_INSERT_TAIL(&pgl, p, listq); } + if (mtx != NULL) + mtx_unlock(mtx); + vm_page_free_phys_pglist(&pgl); vm_object_pip_wakeup(object); } @@ -1982,7 +2052,7 @@ next: void vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { - struct mtx *mtx, *new_mtx; + struct mtx *mtx; vm_page_t p, next; VM_OBJECT_ASSERT_LOCKED(object); @@ -1999,17 +2069,7 @@ vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) mtx = NULL; for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); - - /* - * Avoid releasing and reacquiring the same page lock. - */ - new_mtx = vm_page_lockptr(p); - if (mtx != new_mtx) { - if (mtx != NULL) - mtx_unlock(mtx); - mtx = new_mtx; - mtx_lock(mtx); - } + vm_page_change_lock(p, &mtx); vm_page_deactivate_noreuse(p); } if (mtx != NULL) diff --git a/sys/vm/vm_object.h b/sys/vm/vm_object.h index 9b2192e..17a885e 100644 --- a/sys/vm/vm_object.h +++ b/sys/vm/vm_object.h @@ -70,6 +70,7 @@ #include <sys/queue.h> #include <sys/_lock.h> #include <sys/_mutex.h> +#include <sys/_pctrie.h> #include <sys/_rwlock.h> #include <vm/_vm_radix.h> @@ -86,12 +87,17 @@ * */ +#ifndef VM_PAGE_HAVE_PGLIST +TAILQ_HEAD(pglist, vm_page); +#define VM_PAGE_HAVE_PGLIST +#endif + struct vm_object { struct rwlock lock; TAILQ_ENTRY(vm_object) object_list; /* list of all objects */ LIST_HEAD(, vm_object) shadow_head; /* objects that this is a shadow for */ LIST_ENTRY(vm_object) shadow_list; /* chain of shadow objects */ - TAILQ_HEAD(respgs, vm_page) memq; /* list of resident pages */ + struct pglist memq; /* list of resident pages */ struct vm_radix rtree; /* root of the resident page radix trie*/ vm_pindex_t size; /* Object size */ int generation; /* generation ID */ @@ -151,13 +157,12 @@ struct vm_object { * the handle changed and hash-chain * invalid. * - * swp_bcount - number of swap 'swblock' metablocks, each - * contains up to 16 swapblk assignments. - * see vm/swap_pager.h + * swp_blks - pc-trie of the allocated swap blocks. + * */ struct { void *swp_tmpfs; - int swp_bcount; + struct pctrie swp_blks; } swp; } un_pager; struct ucred *cred; @@ -171,11 +176,12 @@ struct vm_object { #define OBJ_FICTITIOUS 0x0001 /* (c) contains fictitious pages */ #define OBJ_UNMANAGED 0x0002 /* (c) contains unmanaged pages */ #define OBJ_POPULATE 0x0004 /* pager implements populate() */ -#define OBJ_DEAD 0x0008 /* dead objects (during rundown) */ +#define OBJ_DEAD 0x0008 /* dead objects (during rundown) */ #define OBJ_NOSPLIT 0x0010 /* dont split this object */ #define OBJ_UMTXDEAD 0x0020 /* umtx pshared was terminated */ -#define OBJ_PIPWNT 0x0040 /* paging in progress wanted */ -#define OBJ_MIGHTBEDIRTY 0x0100 /* object might be dirty, only for vnode */ +#define OBJ_PIPWNT 0x0040 /* paging in progress wanted */ +#define OBJ_PG_DTOR 0x0080 /* dont reset object, leave that for dtor */ +#define OBJ_MIGHTBEDIRTY 0x0100 /* object might be dirty, only for vnode */ #define OBJ_TMPFS_NODE 0x0200 /* object belongs to tmpfs VREG node */ #define OBJ_TMPFS_DIRTY 0x0400 /* dirty tmpfs obj */ #define OBJ_COLORED 0x1000 /* pg_color is defined */ diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c index d8a9c21..16dc868 100644 --- a/sys/vm/vm_page.c +++ b/sys/vm/vm_page.c @@ -158,6 +158,7 @@ static uma_zone_t fakepg_zone; 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_free_phys(vm_page_t m); static void vm_page_free_wakeup(void); static void vm_page_init_fakepg(void *dummy); static int vm_page_insert_after(vm_page_t m, vm_object_t object, @@ -407,6 +408,29 @@ vm_page_domain_init(struct vm_domain *vmd) } /* + * Initialize a physical page in preparation for adding it to the free + * lists. + */ +static void +vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind) +{ + + m->object = NULL; + m->wire_count = 0; + m->busy_lock = VPB_UNBUSIED; + m->hold_count = 0; + m->flags = 0; + m->phys_addr = pa; + m->queue = PQ_NONE; + m->psind = 0; + m->segind = segind; + m->order = VM_NFREEORDER; + m->pool = VM_FREEPOOL_DEFAULT; + m->valid = m->dirty = 0; + pmap_page_init(m); +} + +/* * vm_page_startup: * * Initializes the resident memory module. Allocates physical memory for @@ -417,17 +441,15 @@ vm_page_domain_init(struct vm_domain *vmd) vm_offset_t vm_page_startup(vm_offset_t vaddr) { - vm_offset_t mapped; - vm_paddr_t high_avail, low_avail, page_range, size; - vm_paddr_t new_end; - int i; - vm_paddr_t pa; - vm_paddr_t last_pa; + struct vm_domain *vmd; + struct vm_phys_seg *seg; + vm_page_t m; char *list, *listend; - vm_paddr_t end; - vm_paddr_t biggestsize; - int biggestone; - int pages_per_zone; + vm_offset_t mapped; + vm_paddr_t end, high_avail, low_avail, new_end, page_range, size; + vm_paddr_t biggestsize, last_pa, pa; + u_long pagecount; + int biggestone, i, pages_per_zone, segind; biggestsize = 0; biggestone = 0; @@ -463,7 +485,8 @@ vm_page_startup(vm_offset_t vaddr) * in proportion to the zone structure size. */ pages_per_zone = howmany(sizeof(struct uma_zone) + - sizeof(struct uma_cache) * (mp_maxid + 1), UMA_SLAB_SIZE); + sizeof(struct uma_cache) * (mp_maxid + 1) + + roundup2(sizeof(struct uma_slab), sizeof(void *)), UMA_SLAB_SIZE); if (pages_per_zone > 1) { /* Reserve more pages so that we don't run out. */ boot_pages = UMA_BOOT_PAGES_ZONES * pages_per_zone; @@ -507,6 +530,8 @@ vm_page_startup(vm_offset_t vaddr) vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)vm_page_dump, vm_page_dump_size); +#else + (void)last_pa; #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) /* @@ -611,7 +636,9 @@ vm_page_startup(vm_offset_t vaddr) new_end = trunc_page(end - page_range * sizeof(struct vm_page)); mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); - vm_page_array = (vm_page_t) mapped; + vm_page_array = (vm_page_t)mapped; + vm_page_array_size = page_range; + #if VM_NRESERVLEVEL > 0 /* * Allocate physical memory for the reservation management system's @@ -638,33 +665,53 @@ vm_page_startup(vm_offset_t vaddr) vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]); /* - * Clear all of the page structures - */ - bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); - for (i = 0; i < page_range; i++) - vm_page_array[i].order = VM_NFREEORDER; - vm_page_array_size = page_range; - - /* * Initialize the physical memory allocator. */ vm_phys_init(); /* - * Add every available physical page that is not blacklisted to - * the free lists. + * Initialize the page structures and add every available page to the + * physical memory allocator's free lists. */ vm_cnt.v_page_count = 0; vm_cnt.v_free_count = 0; - for (i = 0; phys_avail[i + 1] != 0; i += 2) { - pa = phys_avail[i]; - last_pa = phys_avail[i + 1]; - while (pa < last_pa) { - vm_phys_add_page(pa); - pa += PAGE_SIZE; + for (segind = 0; segind < vm_phys_nsegs; segind++) { + seg = &vm_phys_segs[segind]; + for (m = seg->first_page, pa = seg->start; pa < seg->end; + m++, pa += PAGE_SIZE) + vm_page_init_page(m, pa, segind); + + /* + * Add the segment to the free lists only if it is covered by + * one of the ranges in phys_avail. Because we've added the + * ranges to the vm_phys_segs array, we can assume that each + * segment is either entirely contained in one of the ranges, + * or doesn't overlap any of them. + */ + for (i = 0; phys_avail[i + 1] != 0; i += 2) { + if (seg->start < phys_avail[i] || + seg->end > phys_avail[i + 1]) + continue; + + m = seg->first_page; + pagecount = (u_long)atop(seg->end - seg->start); + + mtx_lock(&vm_page_queue_free_mtx); + vm_phys_free_contig(m, pagecount); + vm_phys_freecnt_adj(m, (int)pagecount); + mtx_unlock(&vm_page_queue_free_mtx); + vm_cnt.v_page_count += (u_int)pagecount; + + vmd = &vm_dom[seg->domain]; + vmd->vmd_page_count += (u_int)pagecount; + vmd->vmd_segs |= 1UL << m->segind; + break; } } + /* + * Remove blacklisted pages from the physical memory allocator. + */ TAILQ_INIT(&blacklist_head); vm_page_blacklist_load(&list, &listend); vm_page_blacklist_check(list, listend); @@ -905,6 +952,23 @@ vm_page_flash(vm_page_t m) } /* + * Avoid releasing and reacquiring the same page lock. + */ +void +vm_page_change_lock(vm_page_t m, struct mtx **mtx) +{ + struct mtx *mtx1; + + mtx1 = vm_page_lockptr(m); + if (*mtx == mtx1) + return; + if (*mtx != NULL) + mtx_unlock(*mtx); + *mtx = mtx1; + mtx_lock(mtx1); +} + +/* * Keep page from being freed by the page daemon * much of the same effect as wiring, except much lower * overhead and should be used only for *very* temporary @@ -937,20 +1001,11 @@ vm_page_unhold(vm_page_t mem) void vm_page_unhold_pages(vm_page_t *ma, int count) { - struct mtx *mtx, *new_mtx; + struct mtx *mtx; mtx = NULL; for (; count != 0; count--) { - /* - * Avoid releasing and reacquiring the same page lock. - */ - new_mtx = vm_page_lockptr(*ma); - if (mtx != new_mtx) { - if (mtx != NULL) - mtx_unlock(mtx); - mtx = new_mtx; - mtx_lock(mtx); - } + vm_page_change_lock(*ma, &mtx); vm_page_unhold(*ma); ma++; } @@ -1989,7 +2044,7 @@ vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options) { - struct mtx *m_mtx, *new_mtx; + struct mtx *m_mtx; vm_object_t object; vm_paddr_t pa; vm_page_t m, m_run; @@ -2005,8 +2060,10 @@ vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, run_len = 0; m_mtx = NULL; for (m = m_start; m < m_end && run_len < npages; m += m_inc) { - KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0, - ("page %p is PG_FICTITIOUS or PG_MARKER", m)); + KASSERT((m->flags & PG_MARKER) == 0, + ("page %p is PG_MARKER", m)); + KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->wire_count == 1, + ("fictitious page %p has invalid wire count", m)); /* * If the current page would be the start of a run, check its @@ -2032,16 +2089,7 @@ vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, } else KASSERT(m_run != NULL, ("m_run == NULL")); - /* - * Avoid releasing and reacquiring the same page lock. - */ - new_mtx = vm_page_lockptr(m); - if (m_mtx != new_mtx) { - if (m_mtx != NULL) - mtx_unlock(m_mtx); - m_mtx = new_mtx; - mtx_lock(m_mtx); - } + vm_page_change_lock(m, &m_mtx); m_inc = 1; retry: if (m->wire_count != 0 || m->hold_count != 0) @@ -2191,7 +2239,7 @@ static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, vm_paddr_t high) { - struct mtx *m_mtx, *new_mtx; + struct mtx *m_mtx; struct spglist free; vm_object_t object; vm_paddr_t pa; @@ -2212,13 +2260,7 @@ vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, /* * Avoid releasing and reacquiring the same page lock. */ - new_mtx = vm_page_lockptr(m); - if (m_mtx != new_mtx) { - if (m_mtx != NULL) - mtx_unlock(m_mtx); - m_mtx = new_mtx; - mtx_lock(m_mtx); - } + vm_page_change_lock(m, &m_mtx); retry: if (m->wire_count != 0 || m->hold_count != 0) error = EBUSY; @@ -2331,12 +2373,7 @@ retry: * The new page must be deactivated * before the object is unlocked. */ - new_mtx = vm_page_lockptr(m_new); - if (m_mtx != new_mtx) { - mtx_unlock(m_mtx); - m_mtx = new_mtx; - mtx_lock(m_mtx); - } + vm_page_change_lock(m_new, &m_mtx); vm_page_deactivate(m_new); } else { m->flags &= ~PG_ZERO; @@ -2379,13 +2416,7 @@ unlock: mtx_lock(&vm_page_queue_free_mtx); do { SLIST_REMOVE_HEAD(&free, plinks.s.ss); - vm_phys_freecnt_adj(m, 1); -#if VM_NRESERVLEVEL > 0 - if (!vm_reserv_free_page(m)) -#else - if (true) -#endif - vm_phys_free_pages(m, 0); + vm_page_free_phys(m); } while ((m = SLIST_FIRST(&free)) != NULL); vm_page_zero_idle_wakeup(); vm_page_free_wakeup(); @@ -2722,7 +2753,7 @@ vm_page_activate(vm_page_t m) * * The page queues must be locked. */ -static inline void +static void vm_page_free_wakeup(void) { @@ -2748,17 +2779,30 @@ vm_page_free_wakeup(void) } /* - * vm_page_free_toq: + * vm_page_free_prep: * - * Returns the given page to the free list, - * disassociating it with any VM object. + * Prepares the given page to be put on the free list, + * disassociating it from any VM object. The caller may return + * 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. + * 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. */ -void -vm_page_free_toq(vm_page_t m) +bool +vm_page_free_prep(vm_page_t m, bool pagequeue_locked) { +#if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP) + if ((m->flags & PG_ZERO) != 0) { + uint64_t *p; + int i; + p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); + for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++) + KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx", + m, i, (uintmax_t)*p)); + } +#endif if ((m->oflags & VPO_UNMANAGED) == 0) { vm_page_lock_assert(m, MA_OWNED); KASSERT(!pmap_page_is_mapped(m), @@ -2777,16 +2821,20 @@ vm_page_free_toq(vm_page_t m) * callback routine until after we've put the page on the * appropriate free queue. */ - vm_page_remque(m); + if (m->queue != PQ_NONE) { + if (pagequeue_locked) + vm_page_dequeue_locked(m); + else + vm_page_dequeue(m); + } vm_page_remove(m); /* * If fictitious remove object association and * return, otherwise delay object association removal. */ - if ((m->flags & PG_FICTITIOUS) != 0) { - return; - } + if ((m->flags & PG_FICTITIOUS) != 0) + return (false); m->valid = 0; vm_page_undirty(m); @@ -2798,32 +2846,72 @@ vm_page_free_toq(vm_page_t m) KASSERT((m->flags & PG_UNHOLDFREE) == 0, ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); m->flags |= PG_UNHOLDFREE; - } else { - /* - * Restore the default memory attribute to the page. - */ - if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) - pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); + return (false); + } - /* - * Insert the page into the physical memory allocator's free - * page queues. - */ - mtx_lock(&vm_page_queue_free_mtx); - vm_phys_freecnt_adj(m, 1); + /* + * Restore the default memory attribute to the page. + */ + if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) + pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); + + return (true); +} + +/* + * Insert the page into the physical memory allocator's free page + * queues. This is the last step to free a page. + */ +static void +vm_page_free_phys(vm_page_t m) +{ + + mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); + + vm_phys_freecnt_adj(m, 1); #if VM_NRESERVLEVEL > 0 - if (!vm_reserv_free_page(m)) -#else - if (TRUE) + if (!vm_reserv_free_page(m)) #endif vm_phys_free_pages(m, 0); - if ((m->flags & PG_ZERO) != 0) - ++vm_page_zero_count; - else - vm_page_zero_idle_wakeup(); - vm_page_free_wakeup(); - mtx_unlock(&vm_page_queue_free_mtx); - } + if ((m->flags & PG_ZERO) != 0) + ++vm_page_zero_count; + else + vm_page_zero_idle_wakeup(); +} + +void +vm_page_free_phys_pglist(struct pglist *tq) +{ + vm_page_t m; + + if (TAILQ_EMPTY(tq)) + return; + mtx_lock(&vm_page_queue_free_mtx); + TAILQ_FOREACH(m, tq, listq) + vm_page_free_phys(m); + vm_page_free_wakeup(); + mtx_unlock(&vm_page_queue_free_mtx); +} + +/* + * vm_page_free_toq: + * + * Returns the given page to the free list, disassociating it + * from any VM object. + * + * The object must be locked. The page must be locked if it is + * managed. + */ +void +vm_page_free_toq(vm_page_t m) +{ + + if (!vm_page_free_prep(m, false)) + return; + mtx_lock(&vm_page_queue_free_mtx); + vm_page_free_phys(m); + vm_page_free_wakeup(); + mtx_unlock(&vm_page_queue_free_mtx); } /* @@ -3006,23 +3094,25 @@ vm_page_launder(vm_page_t m) * vm_page_try_to_free() * * Attempt to free the page. If we cannot free it, we do nothing. - * 1 is returned on success, 0 on failure. + * true is returned on success, false on failure. */ -int +bool vm_page_try_to_free(vm_page_t m) { - vm_page_lock_assert(m, MA_OWNED); + vm_page_assert_locked(m); if (m->object != NULL) VM_OBJECT_ASSERT_WLOCKED(m->object); - if (m->dirty || m->hold_count || m->wire_count || + if (m->dirty != 0 || m->hold_count != 0 || m->wire_count != 0 || (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) - return (0); - pmap_remove_all(m); - if (m->dirty) - return (0); + return (false); + if (m->object != NULL && m->object->ref_count != 0) { + pmap_remove_all(m); + if (m->dirty != 0) + return (false); + } vm_page_free(m); - return (1); + return (true); } /* @@ -3142,6 +3232,107 @@ retrylookup: } /* + * Return the specified range of pages from the given object. For each + * page offset within the range, if a page already exists within the object + * at that offset and it is busy, then wait for it to change state. If, + * instead, the page doesn't exist, then allocate it. + * + * The caller must always specify an allocation class. + * + * allocation classes: + * VM_ALLOC_NORMAL normal process request + * VM_ALLOC_SYSTEM system *really* needs the pages + * + * The caller must always specify that the pages are to be busied and/or + * wired. + * + * optional allocation flags: + * VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages + * VM_ALLOC_NOBUSY do not exclusive busy the page + * VM_ALLOC_NOWAIT do not sleep + * VM_ALLOC_SBUSY set page to sbusy state + * VM_ALLOC_WIRED wire the pages + * VM_ALLOC_ZERO zero and validate any invalid pages + * + * If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it + * may return a partial prefix of the requested range. + */ +int +vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, + vm_page_t *ma, int count) +{ + vm_page_t m; + int i; + bool sleep; + + VM_OBJECT_ASSERT_WLOCKED(object); + KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0, + ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed")); + KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 || + (allocflags & VM_ALLOC_WIRED) != 0, + ("vm_page_grab_pages: the pages must be busied or wired")); + KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || + (allocflags & VM_ALLOC_IGN_SBUSY) != 0, + ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch")); + if (count == 0) + return (0); + i = 0; +retrylookup: + m = vm_page_lookup(object, pindex + i); + for (; i < count; i++) { + if (m != NULL) { + sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? + vm_page_xbusied(m) : vm_page_busied(m); + if (sleep) { + if ((allocflags & VM_ALLOC_NOWAIT) != 0) + break; + /* + * Reference the page before unlocking and + * sleeping so that the page daemon is less + * likely to reclaim it. + */ + vm_page_aflag_set(m, PGA_REFERENCED); + vm_page_lock(m); + VM_OBJECT_WUNLOCK(object); + vm_page_busy_sleep(m, "grbmaw", (allocflags & + VM_ALLOC_IGN_SBUSY) != 0); + VM_OBJECT_WLOCK(object); + goto retrylookup; + } + if ((allocflags & VM_ALLOC_WIRED) != 0) { + vm_page_lock(m); + vm_page_wire(m); + vm_page_unlock(m); + } + if ((allocflags & (VM_ALLOC_NOBUSY | + VM_ALLOC_SBUSY)) == 0) + vm_page_xbusy(m); + if ((allocflags & VM_ALLOC_SBUSY) != 0) + vm_page_sbusy(m); + } else { + m = vm_page_alloc(object, pindex + i, (allocflags & + ~VM_ALLOC_IGN_SBUSY) | VM_ALLOC_COUNT(count - i)); + if (m == NULL) { + if ((allocflags & VM_ALLOC_NOWAIT) != 0) + break; + VM_OBJECT_WUNLOCK(object); + VM_WAIT; + VM_OBJECT_WLOCK(object); + goto retrylookup; + } + } + if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) { + if ((m->flags & PG_ZERO) == 0) + pmap_zero_page(m); + m->valid = VM_PAGE_BITS_ALL; + } + ma[i] = m; + m = vm_page_next(m); + } + return (i); +} + +/* * Mapping function for valid or dirty bits in a page. * * Inputs are required to range within a page. @@ -3459,16 +3650,17 @@ vm_page_is_valid(vm_page_t m, int base, int size) } /* - * vm_page_ps_is_valid: - * - * Returns TRUE if the entire (super)page is valid and FALSE otherwise. + * Returns true if all of the specified predicates are true for the entire + * (super)page and false otherwise. */ -boolean_t -vm_page_ps_is_valid(vm_page_t m) +bool +vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m) { + vm_object_t object; int i, npages; - VM_OBJECT_ASSERT_LOCKED(m->object); + object = m->object; + VM_OBJECT_ASSERT_LOCKED(object); npages = atop(pagesizes[m->psind]); /* @@ -3477,10 +3669,28 @@ vm_page_ps_is_valid(vm_page_t m) * occupy adjacent entries in vm_page_array[]. */ for (i = 0; i < npages; i++) { - if (m[i].valid != VM_PAGE_BITS_ALL) - return (FALSE); + /* Always test object consistency, including "skip_m". */ + if (m[i].object != object) + return (false); + if (&m[i] == skip_m) + continue; + if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i])) + return (false); + if ((flags & PS_ALL_DIRTY) != 0) { + /* + * Calling vm_page_test_dirty() or pmap_is_modified() + * might stop this case from spuriously returning + * "false". However, that would require a write lock + * on the object containing "m[i]". + */ + if (m[i].dirty != VM_PAGE_BITS_ALL) + return (false); + } + if ((flags & PS_ALL_VALID) != 0 && + m[i].valid != VM_PAGE_BITS_ALL) + return (false); } - return (TRUE); + return (true); } /* diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h index 1ee8dde..d27fe10 100644 --- a/sys/vm/vm_page.h +++ b/sys/vm/vm_page.h @@ -209,7 +209,10 @@ struct vm_page { #define PQ_LAUNDRY 2 #define PQ_COUNT 3 +#ifndef VM_PAGE_HAVE_PGLIST TAILQ_HEAD(pglist, vm_page); +#define VM_PAGE_HAVE_PGLIST +#endif SLIST_HEAD(spglist, vm_page); struct vm_pagequeue { @@ -387,6 +390,9 @@ vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); * vm_page_alloc_freelist(). Some functions support only a subset * of the flags, and ignore others, see the flags legend. * + * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*() + * and the vm_page_grab*() functions. See these functions for details. + * * Bits 0 - 1 define class. * Bits 2 - 15 dedicated for flags. * Legend: @@ -394,6 +400,7 @@ vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); * (c) - vm_page_alloc_contig() supports the flag. * (f) - vm_page_alloc_freelist() supports the flag. * (g) - vm_page_grab() supports the flag. + * (p) - vm_page_grab_pages() supports the flag. * Bits above 15 define the count of additional pages that the caller * intends to allocate. */ @@ -401,16 +408,16 @@ vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); #define VM_ALLOC_INTERRUPT 1 #define VM_ALLOC_SYSTEM 2 #define VM_ALLOC_CLASS_MASK 3 -#define VM_ALLOC_WIRED 0x0020 /* (acfg) Allocate non pageable page */ -#define VM_ALLOC_ZERO 0x0040 /* (acfg) Try to obtain a zeroed page */ +#define VM_ALLOC_WIRED 0x0020 /* (acfgp) Allocate a wired page */ +#define VM_ALLOC_ZERO 0x0040 /* (acfgp) Allocate a prezeroed page */ #define VM_ALLOC_NOOBJ 0x0100 /* (acg) No associated object */ -#define VM_ALLOC_NOBUSY 0x0200 /* (acg) Do not busy the page */ +#define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */ #define VM_ALLOC_IFCACHED 0x0400 #define VM_ALLOC_IFNOTCACHED 0x0800 -#define VM_ALLOC_IGN_SBUSY 0x1000 /* (g) Ignore shared busy flag */ +#define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy flag */ #define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */ -#define VM_ALLOC_SBUSY 0x4000 /* (acg) Shared busy the page */ -#define VM_ALLOC_NOWAIT 0x8000 /* (g) Do not sleep, return NULL */ +#define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */ +#define VM_ALLOC_NOWAIT 0x8000 /* (gp) Do not sleep */ #define VM_ALLOC_COUNT_SHIFT 16 #define VM_ALLOC_COUNT(count) ((count) << VM_ALLOC_COUNT_SHIFT) @@ -433,6 +440,18 @@ malloc2vm_flags(int malloc_flags) } #endif +/* + * Predicates supported by vm_page_ps_test(): + * + * PS_ALL_DIRTY is true only if the entire (super)page is dirty. + * However, it can be spuriously false when the (super)page has become + * dirty in the pmap but that information has not been propagated to the + * machine-independent layer. + */ +#define PS_ALL_DIRTY 0x1 +#define PS_ALL_VALID 0x2 +#define PS_NONE_BUSY 0x4 + void vm_page_busy_downgrade(vm_page_t m); void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared); void vm_page_flash(vm_page_t m); @@ -448,13 +467,17 @@ vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr); vm_page_t vm_page_alloc_freelist(int, int); +void vm_page_change_lock(vm_page_t m, struct mtx **mtx); vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int); -int vm_page_try_to_free (vm_page_t); +int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, + vm_page_t *ma, int count); 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_locked(vm_page_t m); 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); 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); int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); @@ -464,7 +487,7 @@ vm_page_t vm_page_next(vm_page_t m); int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *); struct vm_pagequeue *vm_page_pagequeue(vm_page_t m); vm_page_t vm_page_prev(vm_page_t m); -boolean_t vm_page_ps_is_valid(vm_page_t m); +bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m); void vm_page_putfake(vm_page_t m); void vm_page_readahead_finish(vm_page_t m); bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, @@ -483,6 +506,7 @@ void vm_page_set_valid_range(vm_page_t m, int base, int size); int vm_page_sleep_if_busy(vm_page_t m, const char *msg); vm_offset_t vm_page_startup(vm_offset_t vaddr); void vm_page_sunbusy(vm_page_t m); +bool vm_page_try_to_free(vm_page_t m); int vm_page_trysbusy(vm_page_t m); void vm_page_unhold_pages(vm_page_t *ma, int count); boolean_t vm_page_unwire(vm_page_t m, uint8_t queue); diff --git a/sys/vm/vm_pageout.c b/sys/vm/vm_pageout.c index c9646cf..793e2e9 100644 --- a/sys/vm/vm_pageout.c +++ b/sys/vm/vm_pageout.c @@ -141,19 +141,6 @@ SYSINIT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_SECOND, kproc_start, SDT_PROVIDER_DEFINE(vm); SDT_PROBE_DEFINE(vm, , , vm__lowmem_scan); -#if !defined(NO_SWAPPING) -/* the kernel process "vm_daemon"*/ -static void vm_daemon(void); -static struct proc *vmproc; - -static struct kproc_desc vm_kp = { - "vmdaemon", - vm_daemon, - &vmproc -}; -SYSINIT(vmdaemon, SI_SUB_KTHREAD_VM, SI_ORDER_FIRST, kproc_start, &vm_kp); -#endif - /* Pagedaemon activity rates, in subdivisions of one second. */ #define VM_LAUNDER_RATE 10 #define VM_INACT_SCAN_RATE 2 @@ -171,26 +158,11 @@ static enum { VM_LAUNDRY_SHORTFALL } vm_laundry_request = VM_LAUNDRY_IDLE; -#if !defined(NO_SWAPPING) -static int vm_pageout_req_swapout; /* XXX */ -static int vm_daemon_needed; -static struct mtx vm_daemon_mtx; -/* Allow for use by vm_pageout before vm_daemon is initialized. */ -MTX_SYSINIT(vm_daemon, &vm_daemon_mtx, "vm daemon", MTX_DEF); -#endif static int vm_pageout_update_period; static int disable_swap_pageouts; static int lowmem_period = 10; static time_t lowmem_uptime; -#if defined(NO_SWAPPING) -static int vm_swap_enabled = 0; -static int vm_swap_idle_enabled = 0; -#else -static int vm_swap_enabled = 1; -static int vm_swap_idle_enabled = 0; -#endif - static int vm_panic_on_oom = 0; SYSCTL_INT(_vm, OID_AUTO, panic_on_oom, @@ -198,56 +170,44 @@ SYSCTL_INT(_vm, OID_AUTO, panic_on_oom, "panic on out of memory instead of killing the largest process"); SYSCTL_INT(_vm, OID_AUTO, pageout_wakeup_thresh, - CTLFLAG_RW, &vm_pageout_wakeup_thresh, 0, + CTLFLAG_RWTUN, &vm_pageout_wakeup_thresh, 0, "free page threshold for waking up the pageout daemon"); SYSCTL_INT(_vm, OID_AUTO, pageout_update_period, - CTLFLAG_RW, &vm_pageout_update_period, 0, + CTLFLAG_RWTUN, &vm_pageout_update_period, 0, "Maximum active LRU update period"); -SYSCTL_INT(_vm, OID_AUTO, lowmem_period, CTLFLAG_RW, &lowmem_period, 0, +SYSCTL_INT(_vm, OID_AUTO, lowmem_period, CTLFLAG_RWTUN, &lowmem_period, 0, "Low memory callback period"); -#if defined(NO_SWAPPING) -SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, - CTLFLAG_RD, &vm_swap_enabled, 0, "Enable entire process swapout"); -SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, - CTLFLAG_RD, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); -#else -SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, - CTLFLAG_RW, &vm_swap_enabled, 0, "Enable entire process swapout"); -SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, - CTLFLAG_RW, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); -#endif - SYSCTL_INT(_vm, OID_AUTO, disable_swapspace_pageouts, - CTLFLAG_RW, &disable_swap_pageouts, 0, "Disallow swapout of dirty pages"); + CTLFLAG_RWTUN, &disable_swap_pageouts, 0, "Disallow swapout of dirty pages"); static int pageout_lock_miss; SYSCTL_INT(_vm, OID_AUTO, pageout_lock_miss, CTLFLAG_RD, &pageout_lock_miss, 0, "vget() lock misses during pageout"); SYSCTL_INT(_vm, OID_AUTO, pageout_oom_seq, - CTLFLAG_RW, &vm_pageout_oom_seq, 0, + CTLFLAG_RWTUN, &vm_pageout_oom_seq, 0, "back-to-back calls to oom detector to start OOM"); static int act_scan_laundry_weight = 3; -SYSCTL_INT(_vm, OID_AUTO, act_scan_laundry_weight, CTLFLAG_RW, +SYSCTL_INT(_vm, OID_AUTO, act_scan_laundry_weight, CTLFLAG_RWTUN, &act_scan_laundry_weight, 0, "weight given to clean vs. dirty pages in active queue scans"); static u_int vm_background_launder_target; -SYSCTL_UINT(_vm, OID_AUTO, background_launder_target, CTLFLAG_RW, +SYSCTL_UINT(_vm, OID_AUTO, background_launder_target, CTLFLAG_RWTUN, &vm_background_launder_target, 0, "background laundering target, in pages"); static u_int vm_background_launder_rate = 4096; -SYSCTL_UINT(_vm, OID_AUTO, background_launder_rate, CTLFLAG_RW, +SYSCTL_UINT(_vm, OID_AUTO, background_launder_rate, CTLFLAG_RWTUN, &vm_background_launder_rate, 0, "background laundering rate, in kilobytes per second"); static u_int vm_background_launder_max = 20 * 1024; -SYSCTL_UINT(_vm, OID_AUTO, background_launder_max, CTLFLAG_RW, +SYSCTL_UINT(_vm, OID_AUTO, background_launder_max, CTLFLAG_RWTUN, &vm_background_launder_max, 0, "background laundering cap, in kilobytes"); int vm_pageout_page_count = 32; @@ -261,11 +221,6 @@ 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); -#if !defined(NO_SWAPPING) -static void vm_pageout_map_deactivate_pages(vm_map_t, long); -static void vm_pageout_object_deactivate_pages(pmap_t, vm_object_t, long); -static void vm_req_vmdaemon(int req); -#endif static boolean_t vm_pageout_page_lock(vm_page_t, vm_page_t *); /* @@ -401,6 +356,8 @@ vm_pageout_cluster(vm_page_t m) */ vm_page_assert_unbusied(m); KASSERT(m->hold_count == 0, ("page %p is held", m)); + + pmap_remove_write(m); vm_page_unlock(m); mc[vm_pageout_page_count] = pb = ps = m; @@ -443,6 +400,7 @@ more: ib = 0; break; } + pmap_remove_write(p); vm_page_unlock(p); mc[--page_base] = pb = p; ++pageout_count; @@ -468,6 +426,7 @@ more: vm_page_unlock(p); break; } + pmap_remove_write(p); vm_page_unlock(p); mc[page_base + pageout_count] = ps = p; ++pageout_count; @@ -512,8 +471,8 @@ vm_pageout_flush(vm_page_t *mc, int count, int flags, int mreq, int *prunlen, VM_OBJECT_ASSERT_WLOCKED(object); /* - * Initiate I/O. Bump the vm_page_t->busy counter and - * mark the pages read-only. + * Initiate I/O. Mark the pages busy and verify that they're valid + * and read-only. * * We do not have to fixup the clean/dirty bits here... we can * allow the pager to do it after the I/O completes. @@ -525,8 +484,9 @@ vm_pageout_flush(vm_page_t *mc, int count, int flags, int mreq, int *prunlen, KASSERT(mc[i]->valid == VM_PAGE_BITS_ALL, ("vm_pageout_flush: partially invalid page %p index %d/%d", mc[i], i, count)); + KASSERT((mc[i]->aflags & PGA_WRITEABLE) == 0, + ("vm_pageout_flush: writeable page %p", mc[i])); vm_page_sbusy(mc[i]); - pmap_remove_write(mc[i]); } vm_object_pip_add(object, count); @@ -599,171 +559,6 @@ vm_pageout_flush(vm_page_t *mc, int count, int flags, int mreq, int *prunlen, return (numpagedout); } -#if !defined(NO_SWAPPING) -/* - * vm_pageout_object_deactivate_pages - * - * Deactivate enough pages to satisfy the inactive target - * requirements. - * - * The object and map must be locked. - */ -static void -vm_pageout_object_deactivate_pages(pmap_t pmap, vm_object_t first_object, - long desired) -{ - vm_object_t backing_object, object; - vm_page_t p; - int act_delta, remove_mode; - - VM_OBJECT_ASSERT_LOCKED(first_object); - if ((first_object->flags & OBJ_FICTITIOUS) != 0) - return; - for (object = first_object;; object = backing_object) { - if (pmap_resident_count(pmap) <= desired) - goto unlock_return; - VM_OBJECT_ASSERT_LOCKED(object); - if ((object->flags & OBJ_UNMANAGED) != 0 || - object->paging_in_progress != 0) - goto unlock_return; - - remove_mode = 0; - if (object->shadow_count > 1) - remove_mode = 1; - /* - * Scan the object's entire memory queue. - */ - TAILQ_FOREACH(p, &object->memq, listq) { - if (pmap_resident_count(pmap) <= desired) - goto unlock_return; - if (vm_page_busied(p)) - continue; - PCPU_INC(cnt.v_pdpages); - vm_page_lock(p); - if (p->wire_count != 0 || p->hold_count != 0 || - !pmap_page_exists_quick(pmap, p)) { - vm_page_unlock(p); - continue; - } - act_delta = pmap_ts_referenced(p); - if ((p->aflags & PGA_REFERENCED) != 0) { - if (act_delta == 0) - act_delta = 1; - vm_page_aflag_clear(p, PGA_REFERENCED); - } - if (!vm_page_active(p) && act_delta != 0) { - vm_page_activate(p); - p->act_count += act_delta; - } else if (vm_page_active(p)) { - if (act_delta == 0) { - p->act_count -= min(p->act_count, - ACT_DECLINE); - if (!remove_mode && p->act_count == 0) { - pmap_remove_all(p); - vm_page_deactivate(p); - } else - vm_page_requeue(p); - } else { - vm_page_activate(p); - if (p->act_count < ACT_MAX - - ACT_ADVANCE) - p->act_count += ACT_ADVANCE; - vm_page_requeue(p); - } - } else if (vm_page_inactive(p)) - pmap_remove_all(p); - vm_page_unlock(p); - } - if ((backing_object = object->backing_object) == NULL) - goto unlock_return; - VM_OBJECT_RLOCK(backing_object); - if (object != first_object) - VM_OBJECT_RUNLOCK(object); - } -unlock_return: - if (object != first_object) - VM_OBJECT_RUNLOCK(object); -} - -/* - * deactivate some number of pages in a map, try to do it fairly, but - * that is really hard to do. - */ -static void -vm_pageout_map_deactivate_pages(map, desired) - vm_map_t map; - long desired; -{ - vm_map_entry_t tmpe; - vm_object_t obj, bigobj; - int nothingwired; - - if (!vm_map_trylock(map)) - return; - - bigobj = NULL; - nothingwired = TRUE; - - /* - * first, search out the biggest object, and try to free pages from - * that. - */ - tmpe = map->header.next; - while (tmpe != &map->header) { - if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { - obj = tmpe->object.vm_object; - if (obj != NULL && VM_OBJECT_TRYRLOCK(obj)) { - if (obj->shadow_count <= 1 && - (bigobj == NULL || - bigobj->resident_page_count < obj->resident_page_count)) { - if (bigobj != NULL) - VM_OBJECT_RUNLOCK(bigobj); - bigobj = obj; - } else - VM_OBJECT_RUNLOCK(obj); - } - } - if (tmpe->wired_count > 0) - nothingwired = FALSE; - tmpe = tmpe->next; - } - - if (bigobj != NULL) { - vm_pageout_object_deactivate_pages(map->pmap, bigobj, desired); - VM_OBJECT_RUNLOCK(bigobj); - } - /* - * Next, hunt around for other pages to deactivate. We actually - * do this search sort of wrong -- .text first is not the best idea. - */ - tmpe = map->header.next; - while (tmpe != &map->header) { - if (pmap_resident_count(vm_map_pmap(map)) <= desired) - break; - if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { - obj = tmpe->object.vm_object; - if (obj != NULL) { - VM_OBJECT_RLOCK(obj); - vm_pageout_object_deactivate_pages(map->pmap, obj, desired); - VM_OBJECT_RUNLOCK(obj); - } - } - tmpe = tmpe->next; - } - - /* - * Remove all mappings if a process is swapped out, this will free page - * table pages. - */ - if (desired == 0 && nothingwired) { - pmap_remove(vm_map_pmap(map), vm_map_min(map), - vm_map_max(map)); - } - - vm_map_unlock(map); -} -#endif /* !defined(NO_SWAPPING) */ - /* * Attempt to acquire all of the necessary locks to launder a page and * then call through the clustering layer to PUTPAGES. Wait a short @@ -822,7 +617,17 @@ vm_pageout_clean(vm_page_t m, int *numpagedout) goto unlock_mp; } VM_OBJECT_WLOCK(object); + + /* + * Ensure that the object and vnode were not disassociated + * while locks were dropped. + */ + if (vp->v_object != object) { + error = ENOENT; + goto unlock_all; + } vm_page_lock(m); + /* * While the object and page were unlocked, the page * may have been: @@ -1506,14 +1311,12 @@ drop_page: vm_pagequeue_unlock(pq); } -#if !defined(NO_SWAPPING) /* * Wakeup the swapout daemon if we didn't free the targeted number of * pages. */ - if (vm_swap_enabled && page_shortage > 0) - vm_req_vmdaemon(VM_SWAP_NORMAL); -#endif + if (page_shortage > 0) + vm_swapout_run(); /* * If the inactive queue scan fails repeatedly to meet its @@ -1663,18 +1466,8 @@ drop_page: vm_page_unlock(m); } vm_pagequeue_unlock(pq); -#if !defined(NO_SWAPPING) - /* - * Idle process swapout -- run once per second. - */ - if (vm_swap_idle_enabled) { - static long lsec; - if (time_second != lsec) { - vm_req_vmdaemon(VM_SWAP_IDLE); - lsec = time_second; - } - } -#endif + if (pass > 0) + vm_swapout_run_idle(); return (page_shortage <= 0); } @@ -1971,7 +1764,7 @@ vm_pageout_worker(void *arg) */ mtx_unlock(&vm_page_queue_free_mtx); if (pass >= 1) - pause("psleep", hz / VM_INACT_SCAN_RATE); + pause("pwait", hz / VM_INACT_SCAN_RATE); pass++; } else { /* @@ -2100,167 +1893,3 @@ pagedaemon_wakeup(void) wakeup(&vm_pageout_wanted); } } - -#if !defined(NO_SWAPPING) -static void -vm_req_vmdaemon(int req) -{ - static int lastrun = 0; - - mtx_lock(&vm_daemon_mtx); - vm_pageout_req_swapout |= req; - if ((ticks > (lastrun + hz)) || (ticks < lastrun)) { - wakeup(&vm_daemon_needed); - lastrun = ticks; - } - mtx_unlock(&vm_daemon_mtx); -} - -static void -vm_daemon(void) -{ - struct rlimit rsslim; - struct proc *p; - struct thread *td; - struct vmspace *vm; - int breakout, swapout_flags, tryagain, attempts; -#ifdef RACCT - uint64_t rsize, ravailable; -#endif - - while (TRUE) { - mtx_lock(&vm_daemon_mtx); - msleep(&vm_daemon_needed, &vm_daemon_mtx, PPAUSE, "psleep", -#ifdef RACCT - racct_enable ? hz : 0 -#else - 0 -#endif - ); - swapout_flags = vm_pageout_req_swapout; - vm_pageout_req_swapout = 0; - mtx_unlock(&vm_daemon_mtx); - if (swapout_flags) - swapout_procs(swapout_flags); - - /* - * scan the processes for exceeding their rlimits or if - * process is swapped out -- deactivate pages - */ - tryagain = 0; - attempts = 0; -again: - attempts++; - sx_slock(&allproc_lock); - FOREACH_PROC_IN_SYSTEM(p) { - vm_pindex_t limit, size; - - /* - * if this is a system process or if we have already - * looked at this process, skip it. - */ - PROC_LOCK(p); - if (p->p_state != PRS_NORMAL || - p->p_flag & (P_INEXEC | P_SYSTEM | P_WEXIT)) { - PROC_UNLOCK(p); - continue; - } - /* - * if the process is in a non-running type state, - * don't touch it. - */ - breakout = 0; - FOREACH_THREAD_IN_PROC(p, td) { - thread_lock(td); - if (!TD_ON_RUNQ(td) && - !TD_IS_RUNNING(td) && - !TD_IS_SLEEPING(td) && - !TD_IS_SUSPENDED(td)) { - thread_unlock(td); - breakout = 1; - break; - } - thread_unlock(td); - } - if (breakout) { - PROC_UNLOCK(p); - continue; - } - /* - * get a limit - */ - lim_rlimit_proc(p, RLIMIT_RSS, &rsslim); - limit = OFF_TO_IDX( - qmin(rsslim.rlim_cur, rsslim.rlim_max)); - - /* - * let processes that are swapped out really be - * swapped out set the limit to nothing (will force a - * swap-out.) - */ - if ((p->p_flag & P_INMEM) == 0) - limit = 0; /* XXX */ - vm = vmspace_acquire_ref(p); - _PHOLD_LITE(p); - PROC_UNLOCK(p); - if (vm == NULL) { - PRELE(p); - continue; - } - sx_sunlock(&allproc_lock); - - size = vmspace_resident_count(vm); - if (size >= limit) { - vm_pageout_map_deactivate_pages( - &vm->vm_map, limit); - size = vmspace_resident_count(vm); - } -#ifdef RACCT - if (racct_enable) { - rsize = IDX_TO_OFF(size); - PROC_LOCK(p); - if (p->p_state == PRS_NORMAL) - racct_set(p, RACCT_RSS, rsize); - ravailable = racct_get_available(p, RACCT_RSS); - PROC_UNLOCK(p); - if (rsize > ravailable) { - /* - * Don't be overly aggressive; this - * might be an innocent process, - * and the limit could've been exceeded - * by some memory hog. Don't try - * to deactivate more than 1/4th - * of process' resident set size. - */ - if (attempts <= 8) { - if (ravailable < rsize - - (rsize / 4)) { - ravailable = rsize - - (rsize / 4); - } - } - vm_pageout_map_deactivate_pages( - &vm->vm_map, - OFF_TO_IDX(ravailable)); - /* Update RSS usage after paging out. */ - size = vmspace_resident_count(vm); - rsize = IDX_TO_OFF(size); - PROC_LOCK(p); - if (p->p_state == PRS_NORMAL) - racct_set(p, RACCT_RSS, rsize); - PROC_UNLOCK(p); - if (rsize > ravailable) - tryagain = 1; - } - } -#endif - vmspace_free(vm); - sx_slock(&allproc_lock); - PRELE(p); - } - sx_sunlock(&allproc_lock); - if (tryagain != 0 && attempts <= 10) - goto again; - } -} -#endif /* !defined(NO_SWAPPING) */ diff --git a/sys/vm/vm_pageout.h b/sys/vm/vm_pageout.h index b44ca2f..6f8e0e9 100644 --- a/sys/vm/vm_pageout.h +++ b/sys/vm/vm_pageout.h @@ -77,12 +77,6 @@ extern int vm_pageout_page_count; extern bool vm_pageout_wanted; extern bool vm_pages_needed; -/* - * Swap out requests - */ -#define VM_SWAP_NORMAL 1 -#define VM_SWAP_IDLE 2 - #define VM_OOM_MEM 1 #define VM_OOM_SWAPZ 2 @@ -109,5 +103,8 @@ extern void vm_waitpfault(void); #ifdef _KERNEL int vm_pageout_flush(vm_page_t *, int, int, int, int *, boolean_t *); void vm_pageout_oom(int shortage); + +void vm_swapout_run(void); +void vm_swapout_run_idle(void); #endif #endif /* _VM_VM_PAGEOUT_H_ */ diff --git a/sys/vm/vm_param.h b/sys/vm/vm_param.h index 68e5cd1..20dc143 100644 --- a/sys/vm/vm_param.h +++ b/sys/vm/vm_param.h @@ -84,7 +84,8 @@ #define VM_V_PAGEOUT_FREE_MIN 9 /* vm_cnt.v_pageout_free_min */ #define VM_OBSOLETE_10 10 /* pageout algorithm */ #define VM_SWAPPING_ENABLED 11 /* swapping enabled */ -#define VM_MAXID 12 /* number of valid vm ids */ +#define VM_OVERCOMMIT 12 /* vm.overcommit */ +#define VM_MAXID 13 /* number of valid vm ids */ /* * Structure for swap device statistics diff --git a/sys/vm/vm_phys.c b/sys/vm/vm_phys.c index 484417b..60b452f 100644 --- a/sys/vm/vm_phys.c +++ b/sys/vm/vm_phys.c @@ -175,7 +175,6 @@ static vm_page_t vm_phys_alloc_seg_contig(struct vm_phys_seg *seg, vm_paddr_t boundary); static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain); static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end); -static int vm_phys_paddr_to_segind(vm_paddr_t pa); static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order); @@ -731,35 +730,6 @@ vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order) } /* - * Initialize a physical page and add it to the free lists. - */ -void -vm_phys_add_page(vm_paddr_t pa) -{ - vm_page_t m; - struct vm_domain *vmd; - - vm_cnt.v_page_count++; - m = vm_phys_paddr_to_vm_page(pa); - m->busy_lock = VPB_UNBUSIED; - m->phys_addr = pa; - m->queue = PQ_NONE; - m->segind = vm_phys_paddr_to_segind(pa); - vmd = vm_phys_domain(m); - vmd->vmd_page_count++; - vmd->vmd_segs |= 1UL << m->segind; - KASSERT(m->order == VM_NFREEORDER, - ("vm_phys_add_page: page %p has unexpected order %d", - m, m->order)); - m->pool = VM_FREEPOOL_DEFAULT; - pmap_page_init(m); - mtx_lock(&vm_page_queue_free_mtx); - vm_phys_freecnt_adj(m, 1); - vm_phys_free_pages(m, 0); - mtx_unlock(&vm_page_queue_free_mtx); -} - -/* * Allocate a contiguous, power of two-sized set of physical pages * from the free lists. * @@ -912,6 +882,7 @@ vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start, { long i; + bzero(range, page_count * sizeof(*range)); for (i = 0; i < page_count; i++) { vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr); range[i].oflags &= ~VPO_UNMANAGED; @@ -986,7 +957,7 @@ vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end, alloc: #endif fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES, - M_WAITOK | M_ZERO); + M_WAITOK); #ifdef VM_PHYSSEG_DENSE } #endif @@ -1067,24 +1038,6 @@ vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end) } /* - * Find the segment containing the given physical address. - */ -static int -vm_phys_paddr_to_segind(vm_paddr_t pa) -{ - struct vm_phys_seg *seg; - int segind; - - for (segind = 0; segind < vm_phys_nsegs; segind++) { - seg = &vm_phys_segs[segind]; - if (pa >= seg->start && pa < seg->end) - return (segind); - } - panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" , - (uintmax_t)pa); -} - -/* * Free a contiguous, power of two-sized set of physical pages. * * The free page queues must be locked. diff --git a/sys/vm/vm_phys.h b/sys/vm/vm_phys.h index ee4aa2d..c5dd58d 100644 --- a/sys/vm/vm_phys.h +++ b/sys/vm/vm_phys.h @@ -69,7 +69,6 @@ extern int vm_phys_nsegs; /* * The following functions are only to be used by the virtual memory system. */ -void vm_phys_add_page(vm_paddr_t pa); void vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end); vm_page_t vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary); diff --git a/sys/vm/vm_reserv.c b/sys/vm/vm_reserv.c index 7e2bfb6..ce3289e 100644 --- a/sys/vm/vm_reserv.c +++ b/sys/vm/vm_reserv.c @@ -1120,4 +1120,18 @@ vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water) return (new_end); } +/* + * Returns the superpage containing the given page. + */ +vm_page_t +vm_reserv_to_superpage(vm_page_t m) +{ + vm_reserv_t rv; + + VM_OBJECT_ASSERT_LOCKED(m->object); + rv = vm_reserv_from_page(m); + return (rv->object == m->object && rv->popcnt == VM_LEVEL_0_NPAGES ? + rv->pages : NULL); +} + #endif /* VM_NRESERVLEVEL > 0 */ diff --git a/sys/vm/vm_reserv.h b/sys/vm/vm_reserv.h index 8b33b48..3d9472d 100644 --- a/sys/vm/vm_reserv.h +++ b/sys/vm/vm_reserv.h @@ -64,6 +64,7 @@ void vm_reserv_rename(vm_page_t m, vm_object_t new_object, int vm_reserv_size(int level); vm_paddr_t vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water); +vm_page_t vm_reserv_to_superpage(vm_page_t m); #endif /* VM_NRESERVLEVEL > 0 */ #endif /* _KERNEL */ diff --git a/sys/vm/vm_swapout.c b/sys/vm/vm_swapout.c new file mode 100644 index 0000000..91b6422 --- /dev/null +++ b/sys/vm/vm_swapout.c @@ -0,0 +1,975 @@ +/*- + * Copyright (c) 1991 Regents of the University of California. + * All rights reserved. + * Copyright (c) 1994 John S. Dyson + * All rights reserved. + * Copyright (c) 1994 David Greenman + * All rights reserved. + * Copyright (c) 2005 Yahoo! Technologies Norway AS + * All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * The Mach Operating System project at Carnegie-Mellon University. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * from: @(#)vm_pageout.c 7.4 (Berkeley) 5/7/91 + * + * + * Copyright (c) 1987, 1990 Carnegie-Mellon University. + * All rights reserved. + * + * Authors: Avadis Tevanian, Jr., Michael Wayne Young + * + * Permission to use, copy, modify and distribute this software and + * its documentation is hereby granted, provided that both the copyright + * notice and this permission notice appear in all copies of the + * software, derivative works or modified versions, and any portions + * thereof, and that both notices appear in supporting documentation. + * + * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" + * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND + * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. + * + * Carnegie Mellon requests users of this software to return to + * + * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU + * School of Computer Science + * Carnegie Mellon University + * Pittsburgh PA 15213-3890 + * + * any improvements or extensions that they make and grant Carnegie the + * rights to redistribute these changes. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include "opt_kstack_pages.h" +#include "opt_kstack_max_pages.h" +#include "opt_vm.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/limits.h> +#include <sys/kernel.h> +#include <sys/eventhandler.h> +#include <sys/lock.h> +#include <sys/mutex.h> +#include <sys/proc.h> +#include <sys/_kstack_cache.h> +#include <sys/kthread.h> +#include <sys/ktr.h> +#include <sys/mount.h> +#include <sys/racct.h> +#include <sys/resourcevar.h> +#include <sys/sched.h> +#include <sys/sdt.h> +#include <sys/signalvar.h> +#include <sys/smp.h> +#include <sys/time.h> +#include <sys/vnode.h> +#include <sys/vmmeter.h> +#include <sys/rwlock.h> +#include <sys/sx.h> +#include <sys/sysctl.h> + +#include <vm/vm.h> +#include <vm/vm_param.h> +#include <vm/vm_object.h> +#include <vm/vm_page.h> +#include <vm/vm_map.h> +#include <vm/vm_pageout.h> +#include <vm/vm_pager.h> +#include <vm/vm_phys.h> +#include <vm/swap_pager.h> +#include <vm/vm_extern.h> +#include <vm/uma.h> + +/* the kernel process "vm_daemon" */ +static void vm_daemon(void); +static struct proc *vmproc; + +static struct kproc_desc vm_kp = { + "vmdaemon", + vm_daemon, + &vmproc +}; +SYSINIT(vmdaemon, SI_SUB_KTHREAD_VM, SI_ORDER_FIRST, kproc_start, &vm_kp); + +static int vm_swap_enabled = 1; +static int vm_swap_idle_enabled = 0; + +SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RW, + &vm_swap_enabled, 0, + "Enable entire process swapout"); +SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RW, + &vm_swap_idle_enabled, 0, + "Allow swapout on idle criteria"); + +/* + * Swap_idle_threshold1 is the guaranteed swapped in time for a process + */ +static int swap_idle_threshold1 = 2; +SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, + &swap_idle_threshold1, 0, + "Guaranteed swapped in time for a process"); + +/* + * Swap_idle_threshold2 is the time that a process can be idle before + * it will be swapped out, if idle swapping is enabled. + */ +static int swap_idle_threshold2 = 10; +SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, + &swap_idle_threshold2, 0, + "Time before a process will be swapped out"); + +static int vm_pageout_req_swapout; /* XXX */ +static int vm_daemon_needed; +static struct mtx vm_daemon_mtx; +/* Allow for use by vm_pageout before vm_daemon is initialized. */ +MTX_SYSINIT(vm_daemon, &vm_daemon_mtx, "vm daemon", MTX_DEF); + +static void swapclear(struct proc *); +static int swapout(struct proc *); +static void vm_swapout_map_deactivate_pages(vm_map_t, long); +static void vm_swapout_object_deactivate_pages(pmap_t, vm_object_t, long); +static void swapout_procs(int action); +static void vm_req_vmdaemon(int req); +static void vm_thread_swapin(struct thread *td); +static void vm_thread_swapout(struct thread *td); + +/* + * vm_swapout_object_deactivate_pages + * + * Deactivate enough pages to satisfy the inactive target + * requirements. + * + * The object and map must be locked. + */ +static void +vm_swapout_object_deactivate_pages(pmap_t pmap, vm_object_t first_object, + long desired) +{ + vm_object_t backing_object, object; + vm_page_t p; + int act_delta, remove_mode; + + VM_OBJECT_ASSERT_LOCKED(first_object); + if ((first_object->flags & OBJ_FICTITIOUS) != 0) + return; + for (object = first_object;; object = backing_object) { + if (pmap_resident_count(pmap) <= desired) + goto unlock_return; + VM_OBJECT_ASSERT_LOCKED(object); + if ((object->flags & OBJ_UNMANAGED) != 0 || + object->paging_in_progress != 0) + goto unlock_return; + + remove_mode = 0; + if (object->shadow_count > 1) + remove_mode = 1; + /* + * Scan the object's entire memory queue. + */ + TAILQ_FOREACH(p, &object->memq, listq) { + if (pmap_resident_count(pmap) <= desired) + goto unlock_return; + if (vm_page_busied(p)) + continue; + PCPU_INC(cnt.v_pdpages); + vm_page_lock(p); + if (p->wire_count != 0 || p->hold_count != 0 || + !pmap_page_exists_quick(pmap, p)) { + vm_page_unlock(p); + continue; + } + act_delta = pmap_ts_referenced(p); + if ((p->aflags & PGA_REFERENCED) != 0) { + if (act_delta == 0) + act_delta = 1; + vm_page_aflag_clear(p, PGA_REFERENCED); + } + if (!vm_page_active(p) && act_delta != 0) { + vm_page_activate(p); + p->act_count += act_delta; + } else if (vm_page_active(p)) { + if (act_delta == 0) { + p->act_count -= min(p->act_count, + ACT_DECLINE); + if (!remove_mode && p->act_count == 0) { + pmap_remove_all(p); + vm_page_deactivate(p); + } else + vm_page_requeue(p); + } else { + vm_page_activate(p); + if (p->act_count < ACT_MAX - + ACT_ADVANCE) + p->act_count += ACT_ADVANCE; + vm_page_requeue(p); + } + } else if (vm_page_inactive(p)) + pmap_remove_all(p); + vm_page_unlock(p); + } + if ((backing_object = object->backing_object) == NULL) + goto unlock_return; + VM_OBJECT_RLOCK(backing_object); + if (object != first_object) + VM_OBJECT_RUNLOCK(object); + } +unlock_return: + if (object != first_object) + VM_OBJECT_RUNLOCK(object); +} + +/* + * deactivate some number of pages in a map, try to do it fairly, but + * that is really hard to do. + */ +static void +vm_swapout_map_deactivate_pages(vm_map_t map, long desired) +{ + vm_map_entry_t tmpe; + vm_object_t obj, bigobj; + int nothingwired; + + if (!vm_map_trylock(map)) + return; + + bigobj = NULL; + nothingwired = TRUE; + + /* + * first, search out the biggest object, and try to free pages from + * that. + */ + tmpe = map->header.next; + while (tmpe != &map->header) { + if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { + obj = tmpe->object.vm_object; + if (obj != NULL && VM_OBJECT_TRYRLOCK(obj)) { + if (obj->shadow_count <= 1 && + (bigobj == NULL || + bigobj->resident_page_count < + obj->resident_page_count)) { + if (bigobj != NULL) + VM_OBJECT_RUNLOCK(bigobj); + bigobj = obj; + } else + VM_OBJECT_RUNLOCK(obj); + } + } + if (tmpe->wired_count > 0) + nothingwired = FALSE; + tmpe = tmpe->next; + } + + if (bigobj != NULL) { + vm_swapout_object_deactivate_pages(map->pmap, bigobj, desired); + VM_OBJECT_RUNLOCK(bigobj); + } + /* + * Next, hunt around for other pages to deactivate. We actually + * do this search sort of wrong -- .text first is not the best idea. + */ + tmpe = map->header.next; + while (tmpe != &map->header) { + if (pmap_resident_count(vm_map_pmap(map)) <= desired) + break; + if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { + obj = tmpe->object.vm_object; + if (obj != NULL) { + VM_OBJECT_RLOCK(obj); + vm_swapout_object_deactivate_pages(map->pmap, + obj, desired); + VM_OBJECT_RUNLOCK(obj); + } + } + tmpe = tmpe->next; + } + + /* + * Remove all mappings if a process is swapped out, this will free page + * table pages. + */ + if (desired == 0 && nothingwired) { + pmap_remove(vm_map_pmap(map), vm_map_min(map), + vm_map_max(map)); + } + + vm_map_unlock(map); +} + +/* + * Swap out requests + */ +#define VM_SWAP_NORMAL 1 +#define VM_SWAP_IDLE 2 + +void +vm_swapout_run(void) +{ + + if (vm_swap_enabled) + vm_req_vmdaemon(VM_SWAP_NORMAL); +} + +/* + * Idle process swapout -- run once per second when pagedaemons are + * reclaiming pages. + */ +void +vm_swapout_run_idle(void) +{ + static long lsec; + + if (!vm_swap_idle_enabled || time_second == lsec) + return; + vm_req_vmdaemon(VM_SWAP_IDLE); + lsec = time_second; +} + +static void +vm_req_vmdaemon(int req) +{ + static int lastrun = 0; + + mtx_lock(&vm_daemon_mtx); + vm_pageout_req_swapout |= req; + if ((ticks > (lastrun + hz)) || (ticks < lastrun)) { + wakeup(&vm_daemon_needed); + lastrun = ticks; + } + mtx_unlock(&vm_daemon_mtx); +} + +static void +vm_daemon(void) +{ + struct rlimit rsslim; + struct proc *p; + struct thread *td; + struct vmspace *vm; + int breakout, swapout_flags, tryagain, attempts; +#ifdef RACCT + uint64_t rsize, ravailable; +#endif + + while (TRUE) { + mtx_lock(&vm_daemon_mtx); + msleep(&vm_daemon_needed, &vm_daemon_mtx, PPAUSE, "psleep", +#ifdef RACCT + racct_enable ? hz : 0 +#else + 0 +#endif + ); + swapout_flags = vm_pageout_req_swapout; + vm_pageout_req_swapout = 0; + mtx_unlock(&vm_daemon_mtx); + if (swapout_flags) + swapout_procs(swapout_flags); + + /* + * scan the processes for exceeding their rlimits or if + * process is swapped out -- deactivate pages + */ + tryagain = 0; + attempts = 0; +again: + attempts++; + sx_slock(&allproc_lock); + FOREACH_PROC_IN_SYSTEM(p) { + vm_pindex_t limit, size; + + /* + * if this is a system process or if we have already + * looked at this process, skip it. + */ + PROC_LOCK(p); + if (p->p_state != PRS_NORMAL || + p->p_flag & (P_INEXEC | P_SYSTEM | P_WEXIT)) { + PROC_UNLOCK(p); + continue; + } + /* + * if the process is in a non-running type state, + * don't touch it. + */ + breakout = 0; + FOREACH_THREAD_IN_PROC(p, td) { + thread_lock(td); + if (!TD_ON_RUNQ(td) && + !TD_IS_RUNNING(td) && + !TD_IS_SLEEPING(td) && + !TD_IS_SUSPENDED(td)) { + thread_unlock(td); + breakout = 1; + break; + } + thread_unlock(td); + } + if (breakout) { + PROC_UNLOCK(p); + continue; + } + /* + * get a limit + */ + lim_rlimit_proc(p, RLIMIT_RSS, &rsslim); + limit = OFF_TO_IDX( + qmin(rsslim.rlim_cur, rsslim.rlim_max)); + + /* + * let processes that are swapped out really be + * swapped out set the limit to nothing (will force a + * swap-out.) + */ + if ((p->p_flag & P_INMEM) == 0) + limit = 0; /* XXX */ + vm = vmspace_acquire_ref(p); + _PHOLD_LITE(p); + PROC_UNLOCK(p); + if (vm == NULL) { + PRELE(p); + continue; + } + sx_sunlock(&allproc_lock); + + size = vmspace_resident_count(vm); + if (size >= limit) { + vm_swapout_map_deactivate_pages( + &vm->vm_map, limit); + size = vmspace_resident_count(vm); + } +#ifdef RACCT + if (racct_enable) { + rsize = IDX_TO_OFF(size); + PROC_LOCK(p); + if (p->p_state == PRS_NORMAL) + racct_set(p, RACCT_RSS, rsize); + ravailable = racct_get_available(p, RACCT_RSS); + PROC_UNLOCK(p); + if (rsize > ravailable) { + /* + * Don't be overly aggressive; this + * might be an innocent process, + * and the limit could've been exceeded + * by some memory hog. Don't try + * to deactivate more than 1/4th + * of process' resident set size. + */ + if (attempts <= 8) { + if (ravailable < rsize - + (rsize / 4)) { + ravailable = rsize - + (rsize / 4); + } + } + vm_swapout_map_deactivate_pages( + &vm->vm_map, + OFF_TO_IDX(ravailable)); + /* Update RSS usage after paging out. */ + size = vmspace_resident_count(vm); + rsize = IDX_TO_OFF(size); + PROC_LOCK(p); + if (p->p_state == PRS_NORMAL) + racct_set(p, RACCT_RSS, rsize); + PROC_UNLOCK(p); + if (rsize > ravailable) + tryagain = 1; + } + } +#endif + vmspace_free(vm); + sx_slock(&allproc_lock); + PRELE(p); + } + sx_sunlock(&allproc_lock); + if (tryagain != 0 && attempts <= 10) + goto again; + } +} + +/* + * Allow a thread's kernel stack to be paged out. + */ +static void +vm_thread_swapout(struct thread *td) +{ + vm_object_t ksobj; + vm_page_t m; + int i, pages; + + cpu_thread_swapout(td); + pages = td->td_kstack_pages; + ksobj = td->td_kstack_obj; + pmap_qremove(td->td_kstack, pages); + VM_OBJECT_WLOCK(ksobj); + for (i = 0; i < pages; i++) { + m = vm_page_lookup(ksobj, i); + if (m == NULL) + panic("vm_thread_swapout: kstack already missing?"); + vm_page_dirty(m); + vm_page_lock(m); + vm_page_unwire(m, PQ_INACTIVE); + vm_page_unlock(m); + } + VM_OBJECT_WUNLOCK(ksobj); +} + +/* + * Bring the kernel stack for a specified thread back in. + */ +static void +vm_thread_swapin(struct thread *td) +{ + vm_object_t ksobj; + vm_page_t ma[KSTACK_MAX_PAGES]; + int pages; + + pages = td->td_kstack_pages; + ksobj = td->td_kstack_obj; + VM_OBJECT_WLOCK(ksobj); + (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_WIRED, ma, + pages); + for (int i = 0; i < pages;) { + int j, a, count, rv; + + vm_page_assert_xbusied(ma[i]); + if (ma[i]->valid == VM_PAGE_BITS_ALL) { + vm_page_xunbusy(ma[i]); + i++; + continue; + } + vm_object_pip_add(ksobj, 1); + for (j = i + 1; j < pages; j++) + if (ma[j]->valid == VM_PAGE_BITS_ALL) + break; + rv = vm_pager_has_page(ksobj, ma[i]->pindex, NULL, &a); + KASSERT(rv == 1, ("%s: missing page %p", __func__, ma[i])); + count = min(a + 1, j - i); + rv = vm_pager_get_pages(ksobj, ma + i, count, NULL, NULL); + KASSERT(rv == VM_PAGER_OK, ("%s: cannot get kstack for proc %d", + __func__, td->td_proc->p_pid)); + vm_object_pip_wakeup(ksobj); + for (j = i; j < i + count; j++) + vm_page_xunbusy(ma[j]); + i += count; + } + VM_OBJECT_WUNLOCK(ksobj); + pmap_qenter(td->td_kstack, ma, pages); + cpu_thread_swapin(td); +} + +void +faultin(struct proc *p) +{ + struct thread *td; + + PROC_LOCK_ASSERT(p, MA_OWNED); + /* + * If another process is swapping in this process, + * just wait until it finishes. + */ + if (p->p_flag & P_SWAPPINGIN) { + while (p->p_flag & P_SWAPPINGIN) + msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0); + return; + } + if ((p->p_flag & P_INMEM) == 0) { + /* + * Don't let another thread swap process p out while we are + * busy swapping it in. + */ + ++p->p_lock; + p->p_flag |= P_SWAPPINGIN; + PROC_UNLOCK(p); + + /* + * We hold no lock here because the list of threads + * can not change while all threads in the process are + * swapped out. + */ + FOREACH_THREAD_IN_PROC(p, td) + vm_thread_swapin(td); + PROC_LOCK(p); + swapclear(p); + p->p_swtick = ticks; + + wakeup(&p->p_flag); + + /* Allow other threads to swap p out now. */ + --p->p_lock; + } +} + +/* + * This swapin algorithm attempts to swap-in processes only if there + * is enough space for them. Of course, if a process waits for a long + * time, it will be swapped in anyway. + */ +void +swapper(void) +{ + struct proc *p; + struct thread *td; + struct proc *pp; + int slptime; + int swtime; + int ppri; + int pri; + +loop: + if (vm_page_count_min()) { + VM_WAIT; + goto loop; + } + + pp = NULL; + ppri = INT_MIN; + sx_slock(&allproc_lock); + FOREACH_PROC_IN_SYSTEM(p) { + PROC_LOCK(p); + if (p->p_state == PRS_NEW || + p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) { + PROC_UNLOCK(p); + continue; + } + swtime = (ticks - p->p_swtick) / hz; + FOREACH_THREAD_IN_PROC(p, td) { + /* + * An otherwise runnable thread of a process + * swapped out has only the TDI_SWAPPED bit set. + */ + thread_lock(td); + if (td->td_inhibitors == TDI_SWAPPED) { + slptime = (ticks - td->td_slptick) / hz; + pri = swtime + slptime; + if ((td->td_flags & TDF_SWAPINREQ) == 0) + pri -= p->p_nice * 8; + /* + * if this thread is higher priority + * and there is enough space, then select + * this process instead of the previous + * selection. + */ + if (pri > ppri) { + pp = p; + ppri = pri; + } + } + thread_unlock(td); + } + PROC_UNLOCK(p); + } + sx_sunlock(&allproc_lock); + + /* + * Nothing to do, back to sleep. + */ + if ((p = pp) == NULL) { + tsleep(&proc0, PVM, "swapin", MAXSLP * hz / 2); + goto loop; + } + PROC_LOCK(p); + + /* + * Another process may be bringing or may have already + * brought this process in while we traverse all threads. + * Or, this process may even be being swapped out again. + */ + if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) { + PROC_UNLOCK(p); + goto loop; + } + + /* + * We would like to bring someone in. (only if there is space). + * [What checks the space? ] + */ + faultin(p); + PROC_UNLOCK(p); + goto loop; +} + +/* + * First, if any processes have been sleeping or stopped for at least + * "swap_idle_threshold1" seconds, they are swapped out. If, however, + * no such processes exist, then the longest-sleeping or stopped + * process is swapped out. Finally, and only as a last resort, if + * there are no sleeping or stopped processes, the longest-resident + * process is swapped out. + */ +static void +swapout_procs(int action) +{ + struct proc *p; + struct thread *td; + int didswap = 0; + +retry: + sx_slock(&allproc_lock); + FOREACH_PROC_IN_SYSTEM(p) { + struct vmspace *vm; + int minslptime = 100000; + int slptime; + + PROC_LOCK(p); + /* + * Watch out for a process in + * creation. It may have no + * address space or lock yet. + */ + if (p->p_state == PRS_NEW) { + PROC_UNLOCK(p); + continue; + } + /* + * An aio daemon switches its + * address space while running. + * Perform a quick check whether + * a process has P_SYSTEM. + * Filter out exiting processes. + */ + if ((p->p_flag & (P_SYSTEM | P_WEXIT)) != 0) { + PROC_UNLOCK(p); + continue; + } + _PHOLD_LITE(p); + PROC_UNLOCK(p); + sx_sunlock(&allproc_lock); + + /* + * Do not swapout a process that + * is waiting for VM data + * structures as there is a possible + * deadlock. Test this first as + * this may block. + * + * Lock the map until swapout + * finishes, or a thread of this + * process may attempt to alter + * the map. + */ + vm = vmspace_acquire_ref(p); + if (vm == NULL) + goto nextproc2; + if (!vm_map_trylock(&vm->vm_map)) + goto nextproc1; + + PROC_LOCK(p); + if (p->p_lock != 1 || (p->p_flag & (P_STOPPED_SINGLE | + P_TRACED | P_SYSTEM)) != 0) + goto nextproc; + + /* + * only aiod changes vmspace, however it will be + * skipped because of the if statement above checking + * for P_SYSTEM + */ + if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM) + goto nextproc; + + switch (p->p_state) { + default: + /* Don't swap out processes in any sort + * of 'special' state. */ + break; + + case PRS_NORMAL: + /* + * do not swapout a realtime process + * Check all the thread groups.. + */ + FOREACH_THREAD_IN_PROC(p, td) { + thread_lock(td); + if (PRI_IS_REALTIME(td->td_pri_class)) { + thread_unlock(td); + goto nextproc; + } + slptime = (ticks - td->td_slptick) / hz; + /* + * Guarantee swap_idle_threshold1 + * time in memory. + */ + if (slptime < swap_idle_threshold1) { + thread_unlock(td); + goto nextproc; + } + + /* + * Do not swapout a process if it is + * waiting on a critical event of some + * kind or there is a thread whose + * pageable memory may be accessed. + * + * This could be refined to support + * swapping out a thread. + */ + if (!thread_safetoswapout(td)) { + thread_unlock(td); + goto nextproc; + } + /* + * If the system is under memory stress, + * or if we are swapping + * idle processes >= swap_idle_threshold2, + * then swap the process out. + */ + if (((action & VM_SWAP_NORMAL) == 0) && + (((action & VM_SWAP_IDLE) == 0) || + (slptime < swap_idle_threshold2))) { + thread_unlock(td); + goto nextproc; + } + + if (minslptime > slptime) + minslptime = slptime; + thread_unlock(td); + } + + /* + * If the pageout daemon didn't free enough pages, + * or if this process is idle and the system is + * configured to swap proactively, swap it out. + */ + if ((action & VM_SWAP_NORMAL) || + ((action & VM_SWAP_IDLE) && + (minslptime > swap_idle_threshold2))) { + _PRELE(p); + if (swapout(p) == 0) + didswap++; + PROC_UNLOCK(p); + vm_map_unlock(&vm->vm_map); + vmspace_free(vm); + goto retry; + } + } +nextproc: + PROC_UNLOCK(p); + vm_map_unlock(&vm->vm_map); +nextproc1: + vmspace_free(vm); +nextproc2: + sx_slock(&allproc_lock); + PRELE(p); + } + sx_sunlock(&allproc_lock); + /* + * If we swapped something out, and another process needed memory, + * then wakeup the sched process. + */ + if (didswap) + wakeup(&proc0); +} + +static void +swapclear(struct proc *p) +{ + struct thread *td; + + PROC_LOCK_ASSERT(p, MA_OWNED); + + FOREACH_THREAD_IN_PROC(p, td) { + thread_lock(td); + td->td_flags |= TDF_INMEM; + td->td_flags &= ~TDF_SWAPINREQ; + TD_CLR_SWAPPED(td); + if (TD_CAN_RUN(td)) + if (setrunnable(td)) { +#ifdef INVARIANTS + /* + * XXX: We just cleared TDI_SWAPPED + * above and set TDF_INMEM, so this + * should never happen. + */ + panic("not waking up swapper"); +#endif + } + thread_unlock(td); + } + p->p_flag &= ~(P_SWAPPINGIN | P_SWAPPINGOUT); + p->p_flag |= P_INMEM; +} + +static int +swapout(struct proc *p) +{ + struct thread *td; + + PROC_LOCK_ASSERT(p, MA_OWNED); + + /* + * The states of this process and its threads may have changed + * by now. Assuming that there is only one pageout daemon thread, + * this process should still be in memory. + */ + KASSERT((p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) == + P_INMEM, ("swapout: lost a swapout race?")); + + /* + * remember the process resident count + */ + p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); + /* + * Check and mark all threads before we proceed. + */ + p->p_flag &= ~P_INMEM; + p->p_flag |= P_SWAPPINGOUT; + FOREACH_THREAD_IN_PROC(p, td) { + thread_lock(td); + if (!thread_safetoswapout(td)) { + thread_unlock(td); + swapclear(p); + return (EBUSY); + } + td->td_flags &= ~TDF_INMEM; + TD_SET_SWAPPED(td); + thread_unlock(td); + } + td = FIRST_THREAD_IN_PROC(p); + ++td->td_ru.ru_nswap; + PROC_UNLOCK(p); + + /* + * This list is stable because all threads are now prevented from + * running. The list is only modified in the context of a running + * thread in this process. + */ + FOREACH_THREAD_IN_PROC(p, td) + vm_thread_swapout(td); + + PROC_LOCK(p); + p->p_flag &= ~P_SWAPPINGOUT; + p->p_swtick = ticks; + return (0); +} diff --git a/sys/vm/vm_swapout_dummy.c b/sys/vm/vm_swapout_dummy.c new file mode 100644 index 0000000..9cdfcb8 --- /dev/null +++ b/sys/vm/vm_swapout_dummy.c @@ -0,0 +1,122 @@ +/*- + * Copyright (c) 1991 Regents of the University of California. + * All rights reserved. + * Copyright (c) 1994 John S. Dyson + * All rights reserved. + * Copyright (c) 1994 David Greenman + * All rights reserved. + * Copyright (c) 2005 Yahoo! Technologies Norway AS + * All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * The Mach Operating System project at Carnegie-Mellon University. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * from: @(#)vm_pageout.c 7.4 (Berkeley) 5/7/91 + * + * + * Copyright (c) 1987, 1990 Carnegie-Mellon University. + * All rights reserved. + * + * Authors: Avadis Tevanian, Jr., Michael Wayne Young + * + * Permission to use, copy, modify and distribute this software and + * its documentation is hereby granted, provided that both the copyright + * notice and this permission notice appear in all copies of the + * software, derivative works or modified versions, and any portions + * thereof, and that both notices appear in supporting documentation. + * + * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" + * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND + * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. + * + * Carnegie Mellon requests users of this software to return to + * + * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU + * School of Computer Science + * Carnegie Mellon University + * Pittsburgh PA 15213-3890 + * + * any improvements or extensions that they make and grant Carnegie the + * rights to redistribute these changes. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/lock.h> +#include <sys/mutex.h> +#include <sys/proc.h> +#include <sys/sysctl.h> +#include <sys/vmmeter.h> + +#include <vm/vm.h> +#include <vm/vm_param.h> +#include <vm/vm_pageout.h> + +static int vm_swap_enabled = 0; +SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RD, + &vm_swap_enabled, 0, + "Enable entire process swapout"); + +static int vm_swap_idle_enabled = 0; +SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RD, + &vm_swap_idle_enabled, 0, + "Allow swapout on idle criteria"); + +void +vm_swapout_run(void) +{ +} + +void +vm_swapout_run_idle(void) +{ +} + +void +faultin(struct proc *p) +{ + + PROC_LOCK_ASSERT(p, MA_OWNED); + if ((p->p_flag & P_INMEM) == 0) + panic("faultin: proc %p swapped out with NO_SWAPPING", p); +} + +void +swapper(void) +{ + + for (;;) + tsleep(&proc0, PVM, "swapin", MAXSLP * hz); +} diff --git a/sys/vm/vnode_pager.c b/sys/vm/vnode_pager.c index 4c0dae9..30b02c4 100644 --- a/sys/vm/vnode_pager.c +++ b/sys/vm/vnode_pager.c @@ -1138,6 +1138,23 @@ vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count, VM_OBJECT_WLOCK(object); } +static int +vn_off2bidx(vm_ooffset_t offset) +{ + + return ((offset & PAGE_MASK) / DEV_BSIZE); +} + +static bool +vn_dirty_blk(vm_page_t m, vm_ooffset_t offset) +{ + + KASSERT(IDX_TO_OFF(m->pindex) <= offset && + offset < IDX_TO_OFF(m->pindex + 1), + ("page %p pidx %ju offset %ju", m, (uintmax_t)m->pindex, + (uintmax_t)offset)); + return ((m->dirty & ((vm_page_bits_t)1 << vn_off2bidx(offset))) != 0); +} /* * This is now called from local media FS's to operate against their @@ -1154,10 +1171,12 @@ vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, { vm_object_t object; vm_page_t m; - vm_ooffset_t poffset; + vm_ooffset_t maxblksz, next_offset, poffset, prev_offset; struct uio auio; struct iovec aiov; + off_t prev_resid, wrsz; int count, error, i, maxsize, ncount, pgoff, ppscheck; + bool in_hole; static struct timeval lastfail; static int curfail; @@ -1192,8 +1211,14 @@ vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, * We do not under any circumstances truncate the valid bits, as * this will screw up bogus page replacement. */ - VM_OBJECT_WLOCK(object); + VM_OBJECT_RLOCK(object); if (maxsize + poffset > object->un_pager.vnp.vnp_size) { + if (!VM_OBJECT_TRYUPGRADE(object)) { + VM_OBJECT_RUNLOCK(object); + VM_OBJECT_WLOCK(object); + if (maxsize + poffset <= object->un_pager.vnp.vnp_size) + goto downgrade; + } if (object->un_pager.vnp.vnp_size > poffset) { maxsize = object->un_pager.vnp.vnp_size - poffset; ncount = btoc(maxsize); @@ -1218,35 +1243,105 @@ vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, } for (i = ncount; i < count; i++) rtvals[i] = VM_PAGER_BAD; +downgrade: + VM_OBJECT_LOCK_DOWNGRADE(object); } - for (i = 0; i < ncount - ((btoc(maxsize) & PAGE_MASK) != 0); i++) - MPASS(ma[i]->dirty == VM_PAGE_BITS_ALL); - VM_OBJECT_WUNLOCK(object); - aiov.iov_base = NULL; - aiov.iov_len = maxsize; auio.uio_iov = &aiov; - auio.uio_iovcnt = 1; - auio.uio_offset = poffset; auio.uio_segflg = UIO_NOCOPY; auio.uio_rw = UIO_WRITE; - auio.uio_resid = maxsize; auio.uio_td = NULL; - error = VOP_WRITE(vp, &auio, vnode_pager_putpages_ioflags(flags), - curthread->td_ucred); - PCPU_INC(cnt.v_vnodeout); - PCPU_ADD(cnt.v_vnodepgsout, ncount); - - ppscheck = 0; - if (error != 0 && (ppscheck = ppsratecheck(&lastfail, &curfail, 1)) - != 0) - printf("vnode_pager_putpages: I/O error %d\n", error); - if (auio.uio_resid != 0 && (ppscheck != 0 || - ppsratecheck(&lastfail, &curfail, 1) != 0)) - printf("vnode_pager_putpages: residual I/O %zd at %ju\n", - auio.uio_resid, (uintmax_t)ma[0]->pindex); - for (i = 0; i < ncount; i++) + maxblksz = roundup2(poffset + maxsize, DEV_BSIZE); + + for (prev_offset = poffset; prev_offset < maxblksz;) { + /* Skip clean blocks. */ + for (in_hole = true; in_hole && prev_offset < maxblksz;) { + m = ma[OFF_TO_IDX(prev_offset - poffset)]; + for (i = vn_off2bidx(prev_offset); + i < sizeof(vm_page_bits_t) * NBBY && + prev_offset < maxblksz; i++) { + if (vn_dirty_blk(m, prev_offset)) { + in_hole = false; + break; + } + prev_offset += DEV_BSIZE; + } + } + if (in_hole) + goto write_done; + + /* Find longest run of dirty blocks. */ + for (next_offset = prev_offset; next_offset < maxblksz;) { + m = ma[OFF_TO_IDX(next_offset - poffset)]; + for (i = vn_off2bidx(next_offset); + i < sizeof(vm_page_bits_t) * NBBY && + next_offset < maxblksz; i++) { + if (!vn_dirty_blk(m, next_offset)) + goto start_write; + next_offset += DEV_BSIZE; + } + } +start_write: + if (next_offset > poffset + maxsize) + next_offset = poffset + maxsize; + + /* + * Getting here requires finding a dirty block in the + * 'skip clean blocks' loop. + */ + MPASS(prev_offset < next_offset); + + VM_OBJECT_RUNLOCK(object); + aiov.iov_base = NULL; + auio.uio_iovcnt = 1; + auio.uio_offset = prev_offset; + prev_resid = auio.uio_resid = aiov.iov_len = next_offset - + prev_offset; + error = VOP_WRITE(vp, &auio, + vnode_pager_putpages_ioflags(flags), curthread->td_ucred); + + wrsz = prev_resid - auio.uio_resid; + if (wrsz == 0) { + if (ppsratecheck(&lastfail, &curfail, 1) != 0) { + vn_printf(vp, "vnode_pager_putpages: " + "zero-length write at %ju resid %zd\n", + auio.uio_offset, auio.uio_resid); + } + VM_OBJECT_RLOCK(object); + break; + } + + /* Adjust the starting offset for next iteration. */ + prev_offset += wrsz; + MPASS(auio.uio_offset == prev_offset); + + ppscheck = 0; + if (error != 0 && (ppscheck = ppsratecheck(&lastfail, + &curfail, 1)) != 0) + vn_printf(vp, "vnode_pager_putpages: I/O error %d\n", + error); + if (auio.uio_resid != 0 && (ppscheck != 0 || + ppsratecheck(&lastfail, &curfail, 1) != 0)) + vn_printf(vp, "vnode_pager_putpages: residual I/O %zd " + "at %ju\n", auio.uio_resid, + (uintmax_t)ma[0]->pindex); + VM_OBJECT_RLOCK(object); + if (error != 0 || auio.uio_resid != 0) + break; + } +write_done: + /* Mark completely processed pages. */ + for (i = 0; i < OFF_TO_IDX(prev_offset - poffset); i++) rtvals[i] = VM_PAGER_OK; + /* Mark partial EOF page. */ + if (prev_offset == poffset + maxsize && (prev_offset & PAGE_MASK) != 0) + rtvals[i++] = VM_PAGER_OK; + /* Unwritten pages in range, free bonus if the page is clean. */ + for (; i < ncount; i++) + rtvals[i] = ma[i]->dirty == 0 ? VM_PAGER_OK : VM_PAGER_ERROR; + VM_OBJECT_RUNLOCK(object); + PCPU_ADD(cnt.v_vnodepgsout, i); + PCPU_INC(cnt.v_vnodeout); return (rtvals[0]); } |