| version 1.116, 2006/04/15 14:23:11 |
version 1.116.2.1, 2006/06/19 04:07:16 |
| Line 82 static struct pool phpool[PHPOOL_MAX]; |
|
| Line 82 static struct pool phpool[PHPOOL_MAX]; |
|
| static struct pool psppool; |
static struct pool psppool; |
| #endif |
#endif |
| |
|
| |
static SLIST_HEAD(, pool_allocator) pa_deferinitq = |
| |
SLIST_HEAD_INITIALIZER(pa_deferinitq); |
| |
|
| static void *pool_page_alloc_meta(struct pool *, int); |
static void *pool_page_alloc_meta(struct pool *, int); |
| static void pool_page_free_meta(struct pool *, void *); |
static void pool_page_free_meta(struct pool *, void *); |
| |
|
| /* allocator for pool metadata */ |
/* allocator for pool metadata */ |
| static struct pool_allocator pool_allocator_meta = { |
static struct pool_allocator pool_allocator_meta = { |
| pool_page_alloc_meta, pool_page_free_meta |
pool_page_alloc_meta, pool_page_free_meta, |
| |
.pa_backingmapptr = &kmem_map, |
| }; |
}; |
| |
|
| /* # of seconds to retain page after last use */ |
/* # of seconds to retain page after last use */ |
| Line 184 static void pool_prime_page(struct pool |
|
| Line 188 static void pool_prime_page(struct pool |
|
| static void pool_update_curpage(struct pool *); |
static void pool_update_curpage(struct pool *); |
| |
|
| static int pool_grow(struct pool *, int); |
static int pool_grow(struct pool *, int); |
| void *pool_allocator_alloc(struct pool *, int); |
static void *pool_allocator_alloc(struct pool *, int); |
| void pool_allocator_free(struct pool *, void *); |
static void pool_allocator_free(struct pool *, void *); |
| |
|
| static void pool_print_pagelist(struct pool *, struct pool_pagelist *, |
static void pool_print_pagelist(struct pool *, struct pool_pagelist *, |
| void (*)(const char *, ...)); |
void (*)(const char *, ...)); |
| Line 443 pr_rmpage(struct pool *pp, struct pool_i |
|
| Line 447 pr_rmpage(struct pool *pp, struct pool_i |
|
| pool_update_curpage(pp); |
pool_update_curpage(pp); |
| } |
} |
| |
|
| |
static boolean_t |
| |
pa_starved_p(struct pool_allocator *pa) |
| |
{ |
| |
|
| |
if (pa->pa_backingmap != NULL) { |
| |
return vm_map_starved_p(pa->pa_backingmap); |
| |
} |
| |
return FALSE; |
| |
} |
| |
|
| |
static int |
| |
pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) |
| |
{ |
| |
struct pool *pp = obj; |
| |
struct pool_allocator *pa = pp->pr_alloc; |
| |
|
| |
KASSERT(&pp->pr_reclaimerentry == ce); |
| |
pool_reclaim(pp); |
| |
if (!pa_starved_p(pa)) { |
| |
return CALLBACK_CHAIN_ABORT; |
| |
} |
| |
return CALLBACK_CHAIN_CONTINUE; |
| |
} |
| |
|
| |
static void |
| |
pool_reclaim_register(struct pool *pp) |
| |
{ |
| |
struct vm_map *map = pp->pr_alloc->pa_backingmap; |
| |
int s; |
| |
|
| |
if (map == NULL) { |
| |
return; |
| |
} |
| |
|
| |
s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */ |
| |
callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback, |
| |
&pp->pr_reclaimerentry, pp, pool_reclaim_callback); |
| |
splx(s); |
| |
} |
| |
|
| |
static void |
| |
pool_reclaim_unregister(struct pool *pp) |
| |
{ |
| |
struct vm_map *map = pp->pr_alloc->pa_backingmap; |
| |
int s; |
| |
|
| |
if (map == NULL) { |
| |
return; |
| |
} |
| |
|
| |
s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */ |
| |
callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback, |
| |
&pp->pr_reclaimerentry); |
| |
splx(s); |
| |
} |
| |
|
| |
static void |
| |
pa_reclaim_register(struct pool_allocator *pa) |
| |
{ |
| |
struct vm_map *map = *pa->pa_backingmapptr; |
| |
struct pool *pp; |
| |
|
| |
KASSERT(pa->pa_backingmap == NULL); |
| |
if (map == NULL) { |
| |
SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q); |
| |
return; |
| |
} |
| |
pa->pa_backingmap = map; |
| |
TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) { |
| |
pool_reclaim_register(pp); |
| |
} |
| |
} |
| |
|
| /* |
/* |
| * Initialize all the pools listed in the "pools" link set. |
* Initialize all the pools listed in the "pools" link set. |
| */ |
*/ |
| void |
void |
| link_pool_init(void) |
pool_subsystem_init(void) |
| { |
{ |
| |
struct pool_allocator *pa; |
| __link_set_decl(pools, struct link_pool_init); |
__link_set_decl(pools, struct link_pool_init); |
| struct link_pool_init * const *pi; |
struct link_pool_init * const *pi; |
| |
|
| Line 456 link_pool_init(void) |
|
| Line 534 link_pool_init(void) |
|
| pool_init((*pi)->pp, (*pi)->size, (*pi)->align, |
pool_init((*pi)->pp, (*pi)->size, (*pi)->align, |
| (*pi)->align_offset, (*pi)->flags, (*pi)->wchan, |
(*pi)->align_offset, (*pi)->flags, (*pi)->wchan, |
| (*pi)->palloc); |
(*pi)->palloc); |
| |
|
| |
while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) { |
| |
KASSERT(pa->pa_backingmapptr != NULL); |
| |
KASSERT(*pa->pa_backingmapptr != NULL); |
| |
SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q); |
| |
pa_reclaim_register(pa); |
| |
} |
| } |
} |
| |
|
| /* |
/* |
| Line 516 pool_init(struct pool *pp, size_t size, |
|
| Line 601 pool_init(struct pool *pp, size_t size, |
|
| simple_lock_init(&palloc->pa_slock); |
simple_lock_init(&palloc->pa_slock); |
| palloc->pa_pagemask = ~(palloc->pa_pagesz - 1); |
palloc->pa_pagemask = ~(palloc->pa_pagesz - 1); |
| palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1; |
palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1; |
| |
|
| |
if (palloc->pa_backingmapptr != NULL) { |
| |
pa_reclaim_register(palloc); |
| |
} |
| palloc->pa_flags |= PA_INITIALIZED; |
palloc->pa_flags |= PA_INITIALIZED; |
| } |
} |
| |
|
| Line 697 pool_init(struct pool *pp, size_t size, |
|
| Line 786 pool_init(struct pool *pp, size_t size, |
|
| TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list); |
TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list); |
| simple_unlock(&palloc->pa_slock); |
simple_unlock(&palloc->pa_slock); |
| splx(s); |
splx(s); |
| |
pool_reclaim_register(pp); |
| } |
} |
| |
|
| /* |
/* |
| Line 717 pool_destroy(struct pool *pp) |
|
| Line 807 pool_destroy(struct pool *pp) |
|
| simple_unlock(&pool_head_slock); |
simple_unlock(&pool_head_slock); |
| |
|
| /* Remove this pool from its allocator's list of pools. */ |
/* Remove this pool from its allocator's list of pools. */ |
| |
pool_reclaim_unregister(pp); |
| s = splvm(); |
s = splvm(); |
| simple_lock(&pp->pr_alloc->pa_slock); |
simple_lock(&pp->pr_alloc->pa_slock); |
| TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list); |
TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list); |
| Line 911 pool_get(struct pool *pp, int flags) |
|
| Line 1002 pool_get(struct pool *pp, int flags) |
|
| if (pp->pr_curpage != NULL) |
if (pp->pr_curpage != NULL) |
| goto startover; |
goto startover; |
| |
|
| if ((flags & PR_WAITOK) == 0) { |
pp->pr_nfail++; |
| pp->pr_nfail++; |
|
| pr_leave(pp); |
|
| simple_unlock(&pp->pr_slock); |
|
| return (NULL); |
|
| } |
|
| |
|
| /* |
|
| * Wait for items to be returned to this pool. |
|
| * |
|
| * wake up once a second and try again, |
|
| * as the check in pool_cache_put_paddr() is racy. |
|
| */ |
|
| pp->pr_flags |= PR_WANTED; |
|
| /* PA_WANTED is already set on the allocator. */ |
|
| pr_leave(pp); |
pr_leave(pp); |
| ltsleep(pp, PSWP, pp->pr_wchan, hz, &pp->pr_slock); |
simple_unlock(&pp->pr_slock); |
| pr_enter(pp, file, line); |
return (NULL); |
| } |
} |
| |
|
| /* Start the allocation process over. */ |
/* Start the allocation process over. */ |
| Line 1045 pool_do_put(struct pool *pp, void *v, st |
|
| Line 1122 pool_do_put(struct pool *pp, void *v, st |
|
| struct pool_item *pi = v; |
struct pool_item *pi = v; |
| struct pool_item_header *ph; |
struct pool_item_header *ph; |
| caddr_t page; |
caddr_t page; |
| int s; |
|
| |
|
| LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
| SCHED_ASSERT_UNLOCKED(); |
SCHED_ASSERT_UNLOCKED(); |
| Line 1128 pool_do_put(struct pool *pp, void *v, st |
|
| Line 1204 pool_do_put(struct pool *pp, void *v, st |
|
| pp->pr_nidle++; |
pp->pr_nidle++; |
| if (pp->pr_npages > pp->pr_minpages && |
if (pp->pr_npages > pp->pr_minpages && |
| (pp->pr_npages > pp->pr_maxpages || |
(pp->pr_npages > pp->pr_maxpages || |
| (pp->pr_alloc->pa_flags & PA_WANT) != 0)) { |
pa_starved_p(pp->pr_alloc))) { |
| pr_rmpage(pp, ph, pq); |
pr_rmpage(pp, ph, pq); |
| } else { |
} else { |
| LIST_REMOVE(ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| Line 1140 pool_do_put(struct pool *pp, void *v, st |
|
| Line 1216 pool_do_put(struct pool *pp, void *v, st |
|
| * be reclaimed by the pagedaemon. This minimizes |
* be reclaimed by the pagedaemon. This minimizes |
| * ping-pong'ing for memory. |
* ping-pong'ing for memory. |
| */ |
*/ |
| s = splclock(); |
getmicrotime(&ph->ph_time); |
| ph->ph_time = mono_time; |
|
| splx(s); |
|
| } |
} |
| pool_update_curpage(pp); |
pool_update_curpage(pp); |
| } |
} |
| Line 1278 pool_prime_page(struct pool *pp, caddr_t |
|
| Line 1352 pool_prime_page(struct pool *pp, caddr_t |
|
| unsigned int align = pp->pr_align; |
unsigned int align = pp->pr_align; |
| unsigned int ioff = pp->pr_itemoffset; |
unsigned int ioff = pp->pr_itemoffset; |
| int n; |
int n; |
| int s; |
|
| |
|
| LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
| |
|
| Line 1294 pool_prime_page(struct pool *pp, caddr_t |
|
| Line 1367 pool_prime_page(struct pool *pp, caddr_t |
|
| LIST_INIT(&ph->ph_itemlist); |
LIST_INIT(&ph->ph_itemlist); |
| ph->ph_page = storage; |
ph->ph_page = storage; |
| ph->ph_nmissing = 0; |
ph->ph_nmissing = 0; |
| s = splclock(); |
getmicrotime(&ph->ph_time); |
| ph->ph_time = mono_time; |
|
| splx(s); |
|
| if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
| SPLAY_INSERT(phtree, &pp->pr_phtree, ph); |
SPLAY_INSERT(phtree, &pp->pr_phtree, ph); |
| |
|
| Line 1462 pool_reclaim(struct pool *pp) |
|
| Line 1533 pool_reclaim(struct pool *pp) |
|
| struct pool_pagelist pq; |
struct pool_pagelist pq; |
| struct pool_cache_grouplist pcgl; |
struct pool_cache_grouplist pcgl; |
| struct timeval curtime, diff; |
struct timeval curtime, diff; |
| int s; |
|
| |
|
| if (pp->pr_drain_hook != NULL) { |
if (pp->pr_drain_hook != NULL) { |
| /* |
/* |
| Line 1484 pool_reclaim(struct pool *pp) |
|
| Line 1554 pool_reclaim(struct pool *pp) |
|
| LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) |
LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) |
| pool_cache_reclaim(pc, &pq, &pcgl); |
pool_cache_reclaim(pc, &pq, &pcgl); |
| |
|
| s = splclock(); |
getmicrotime(&curtime); |
| curtime = mono_time; |
|
| splx(s); |
|
| |
|
| for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) { |
for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) { |
| phnext = LIST_NEXT(ph, ph_pagelist); |
phnext = LIST_NEXT(ph, ph_pagelist); |
| Line 1497 pool_reclaim(struct pool *pp) |
|
| Line 1565 pool_reclaim(struct pool *pp) |
|
| |
|
| KASSERT(ph->ph_nmissing == 0); |
KASSERT(ph->ph_nmissing == 0); |
| timersub(&curtime, &ph->ph_time, &diff); |
timersub(&curtime, &ph->ph_time, &diff); |
| if (diff.tv_sec < pool_inactive_time) |
if (diff.tv_sec < pool_inactive_time |
| |
&& !pa_starved_p(pp->pr_alloc)) |
| continue; |
continue; |
| |
|
| /* |
/* |
| Line 2181 void pool_page_free(struct pool *, void |
|
| Line 2250 void pool_page_free(struct pool *, void |
|
| #ifdef POOL_SUBPAGE |
#ifdef POOL_SUBPAGE |
| struct pool_allocator pool_allocator_kmem_fullpage = { |
struct pool_allocator pool_allocator_kmem_fullpage = { |
| pool_page_alloc, pool_page_free, 0, |
pool_page_alloc, pool_page_free, 0, |
| |
.pa_backingmapptr = &kmem_map, |
| }; |
}; |
| #else |
#else |
| struct pool_allocator pool_allocator_kmem = { |
struct pool_allocator pool_allocator_kmem = { |
| pool_page_alloc, pool_page_free, 0, |
pool_page_alloc, pool_page_free, 0, |
| |
.pa_backingmapptr = &kmem_map, |
| }; |
}; |
| #endif |
#endif |
| |
|
| Line 2194 void pool_page_free_nointr(struct pool * |
|
| Line 2265 void pool_page_free_nointr(struct pool * |
|
| #ifdef POOL_SUBPAGE |
#ifdef POOL_SUBPAGE |
| struct pool_allocator pool_allocator_nointr_fullpage = { |
struct pool_allocator pool_allocator_nointr_fullpage = { |
| pool_page_alloc_nointr, pool_page_free_nointr, 0, |
pool_page_alloc_nointr, pool_page_free_nointr, 0, |
| |
.pa_backingmapptr = &kernel_map, |
| }; |
}; |
| #else |
#else |
| struct pool_allocator pool_allocator_nointr = { |
struct pool_allocator pool_allocator_nointr = { |
| pool_page_alloc_nointr, pool_page_free_nointr, 0, |
pool_page_alloc_nointr, pool_page_free_nointr, 0, |
| |
.pa_backingmapptr = &kernel_map, |
| }; |
}; |
| #endif |
#endif |
| |
|
| Line 2207 void pool_subpage_free(struct pool *, vo |
|
| Line 2280 void pool_subpage_free(struct pool *, vo |
|
| |
|
| struct pool_allocator pool_allocator_kmem = { |
struct pool_allocator pool_allocator_kmem = { |
| pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
| |
.pa_backingmapptr = &kmem_map, |
| }; |
}; |
| |
|
| void *pool_subpage_alloc_nointr(struct pool *, int); |
void *pool_subpage_alloc_nointr(struct pool *, int); |
| Line 2214 void pool_subpage_free_nointr(struct poo |
|
| Line 2288 void pool_subpage_free_nointr(struct poo |
|
| |
|
| struct pool_allocator pool_allocator_nointr = { |
struct pool_allocator pool_allocator_nointr = { |
| pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
| |
.pa_backingmapptr = &kmem_map, |
| }; |
}; |
| #endif /* POOL_SUBPAGE */ |
#endif /* POOL_SUBPAGE */ |
| |
|
| /* |
static void * |
| * We have at least three different resources for the same allocation and |
pool_allocator_alloc(struct pool *pp, int flags) |
| * each resource can be depleted. First, we have the ready elements in the |
|
| * pool. Then we have the resource (typically a vm_map) for this allocator. |
|
| * Finally, we have physical memory. Waiting for any of these can be |
|
| * unnecessary when any other is freed, but the kernel doesn't support |
|
| * sleeping on multiple wait channels, so we have to employ another strategy. |
|
| * |
|
| * The caller sleeps on the pool (so that it can be awakened when an item |
|
| * is returned to the pool), but we set PA_WANT on the allocator. When a |
|
| * page is returned to the allocator and PA_WANT is set, pool_allocator_free |
|
| * will wake up all sleeping pools belonging to this allocator. |
|
| * |
|
| * XXX Thundering herd. |
|
| */ |
|
| void * |
|
| pool_allocator_alloc(struct pool *org, int flags) |
|
| { |
{ |
| struct pool_allocator *pa = org->pr_alloc; |
struct pool_allocator *pa = pp->pr_alloc; |
| struct pool *pp, *start; |
|
| int s, freed; |
|
| void *res; |
void *res; |
| |
|
| LOCK_ASSERT(!simple_lock_held(&org->pr_slock)); |
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); |
| |
|
| do { |
|
| if ((res = (*pa->pa_alloc)(org, flags)) != NULL) |
|
| return (res); |
|
| if ((flags & PR_WAITOK) == 0) { |
|
| /* |
|
| * We only run the drain hookhere if PR_NOWAIT. |
|
| * In other cases, the hook will be run in |
|
| * pool_reclaim(). |
|
| */ |
|
| if (org->pr_drain_hook != NULL) { |
|
| (*org->pr_drain_hook)(org->pr_drain_hook_arg, |
|
| flags); |
|
| if ((res = (*pa->pa_alloc)(org, flags)) != NULL) |
|
| return (res); |
|
| } |
|
| break; |
|
| } |
|
| |
|
| |
res = (*pa->pa_alloc)(pp, flags); |
| |
if (res == NULL && (flags & PR_WAITOK) == 0) { |
| /* |
/* |
| * Drain all pools, that use this allocator. |
* We only run the drain hook here if PR_NOWAIT. |
| * We do this to reclaim VA space. |
* In other cases, the hook will be run in |
| * pa_alloc is responsible for waiting for |
* pool_reclaim(). |
| * physical memory. |
|
| * |
|
| * XXX We risk looping forever if start if someone |
|
| * calls pool_destroy on "start". But there is no |
|
| * other way to have potentially sleeping pool_reclaim, |
|
| * non-sleeping locks on pool_allocator, and some |
|
| * stirring of drained pools in the allocator. |
|
| * |
|
| * XXX Maybe we should use pool_head_slock for locking |
|
| * the allocators? |
|
| */ |
*/ |
| freed = 0; |
if (pp->pr_drain_hook != NULL) { |
| |
(*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags); |
| s = splvm(); |
res = (*pa->pa_alloc)(pp, flags); |
| simple_lock(&pa->pa_slock); |
|
| pp = start = TAILQ_FIRST(&pa->pa_list); |
|
| do { |
|
| TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list); |
|
| TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list); |
|
| simple_unlock(&pa->pa_slock); |
|
| freed = pool_reclaim(pp); |
|
| simple_lock(&pa->pa_slock); |
|
| } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start && |
|
| freed == 0); |
|
| |
|
| if (freed == 0) { |
|
| /* |
|
| * We set PA_WANT here, the caller will most likely |
|
| * sleep waiting for pages (if not, this won't hurt |
|
| * that much), and there is no way to set this in |
|
| * the caller without violating locking order. |
|
| */ |
|
| pa->pa_flags |= PA_WANT; |
|
| } |
} |
| simple_unlock(&pa->pa_slock); |
} |
| splx(s); |
return res; |
| } while (freed); |
|
| return (NULL); |
|
| } |
} |
| |
|
| void |
static void |
| pool_allocator_free(struct pool *pp, void *v) |
pool_allocator_free(struct pool *pp, void *v) |
| { |
{ |
| struct pool_allocator *pa = pp->pr_alloc; |
struct pool_allocator *pa = pp->pr_alloc; |
| int s; |
|
| |
|
| LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); |
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); |
| |
|
| (*pa->pa_free)(pp, v); |
(*pa->pa_free)(pp, v); |
| |
|
| s = splvm(); |
|
| simple_lock(&pa->pa_slock); |
|
| if ((pa->pa_flags & PA_WANT) == 0) { |
|
| simple_unlock(&pa->pa_slock); |
|
| splx(s); |
|
| return; |
|
| } |
|
| |
|
| TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) { |
|
| simple_lock(&pp->pr_slock); |
|
| if ((pp->pr_flags & PR_WANTED) != 0) { |
|
| pp->pr_flags &= ~PR_WANTED; |
|
| wakeup(pp); |
|
| } |
|
| simple_unlock(&pp->pr_slock); |
|
| } |
|
| pa->pa_flags &= ~PA_WANT; |
|
| simple_unlock(&pa->pa_slock); |
|
| splx(s); |
|
| } |
} |
| |
|
| void * |
void * |