| version 1.87, 2003/04/09 18:22:13 |
version 1.87.2.1, 2004/08/03 10:52:55 |
| Line 59 __KERNEL_RCSID(0, "$NetBSD$"); |
|
| Line 59 __KERNEL_RCSID(0, "$NetBSD$"); |
|
| /* |
/* |
| * Pool resource management utility. |
* Pool resource management utility. |
| * |
* |
| * Memory is allocated in pages which are split into pieces according |
* Memory is allocated in pages which are split into pieces according to |
| * to the pool item size. Each page is kept on a list headed by `pr_pagelist' |
* the pool item size. Each page is kept on one of three lists in the |
| * in the pool structure and the individual pool items are on a linked list |
* pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages', |
| * headed by `ph_itemlist' in each page header. The memory for building |
* for empty, full and partially-full pages respectively. The individual |
| * the page list is either taken from the allocated pages themselves (for |
* pool items are on a linked list headed by `ph_itemlist' in each page |
| * small pool items) or taken from an internal pool of page headers (`phpool'). |
* header. The memory for building the page list is either taken from |
| |
* the allocated pages themselves (for small pool items) or taken from |
| |
* an internal pool of page headers (`phpool'). |
| */ |
*/ |
| |
|
| /* List of all pools */ |
/* List of all pools */ |
| Line 89 struct simplelock pool_head_slock = SIMP |
|
| Line 91 struct simplelock pool_head_slock = SIMP |
|
| |
|
| struct pool_item_header { |
struct pool_item_header { |
| /* Page headers */ |
/* Page headers */ |
| TAILQ_ENTRY(pool_item_header) |
LIST_ENTRY(pool_item_header) |
| ph_pagelist; /* pool page list */ |
ph_pagelist; /* pool page list */ |
| TAILQ_HEAD(,pool_item) ph_itemlist; /* chunk list for this page */ |
TAILQ_HEAD(,pool_item) ph_itemlist; /* chunk list for this page */ |
| LIST_ENTRY(pool_item_header) |
SPLAY_ENTRY(pool_item_header) |
| ph_hashlist; /* Off-page page headers */ |
ph_node; /* Off-page page headers */ |
| unsigned int ph_nmissing; /* # of chunks in use */ |
unsigned int ph_nmissing; /* # of chunks in use */ |
| caddr_t ph_page; /* this page's address */ |
caddr_t ph_page; /* this page's address */ |
| struct timeval ph_time; /* last referenced */ |
struct timeval ph_time; /* last referenced */ |
| }; |
}; |
| TAILQ_HEAD(pool_pagelist,pool_item_header); |
|
| |
|
| struct pool_item { |
struct pool_item { |
| #ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
| Line 109 struct pool_item { |
|
| Line 110 struct pool_item { |
|
| TAILQ_ENTRY(pool_item) pi_list; |
TAILQ_ENTRY(pool_item) pi_list; |
| }; |
}; |
| |
|
| #define PR_HASH_INDEX(pp,addr) \ |
|
| (((u_long)(addr) >> (pp)->pr_alloc->pa_pageshift) & \ |
|
| (PR_HASHTABSIZE - 1)) |
|
| |
|
| #define POOL_NEEDS_CATCHUP(pp) \ |
#define POOL_NEEDS_CATCHUP(pp) \ |
| ((pp)->pr_nitems < (pp)->pr_minitems) |
((pp)->pr_nitems < (pp)->pr_minitems) |
| |
|
| Line 150 static void pool_cache_reclaim(struct po |
|
| Line 147 static void pool_cache_reclaim(struct po |
|
| static int pool_catchup(struct pool *); |
static int pool_catchup(struct pool *); |
| static void pool_prime_page(struct pool *, caddr_t, |
static void pool_prime_page(struct pool *, caddr_t, |
| struct pool_item_header *); |
struct pool_item_header *); |
| |
static void pool_update_curpage(struct pool *); |
| |
|
| void *pool_allocator_alloc(struct pool *, int); |
void *pool_allocator_alloc(struct pool *, int); |
| void pool_allocator_free(struct pool *, void *); |
void pool_allocator_free(struct pool *, void *); |
| |
|
| |
static void pool_print_pagelist(struct pool_pagelist *, |
| |
void (*)(const char *, ...)); |
| static void pool_print1(struct pool *, const char *, |
static void pool_print1(struct pool *, const char *, |
| void (*)(const char *, ...)); |
void (*)(const char *, ...)); |
| |
|
| |
static int pool_chk_page(struct pool *, const char *, |
| |
struct pool_item_header *); |
| |
|
| /* |
/* |
| * Pool log entry. An array of these is allocated in pool_init(). |
* Pool log entry. An array of these is allocated in pool_init(). |
| */ |
*/ |
| Line 275 pr_enter_check(struct pool *pp, void (*p |
|
| Line 278 pr_enter_check(struct pool *pp, void (*p |
|
| #define pr_enter_check(pp, pr) |
#define pr_enter_check(pp, pr) |
| #endif /* POOL_DIAGNOSTIC */ |
#endif /* POOL_DIAGNOSTIC */ |
| |
|
| |
static __inline int |
| |
phtree_compare(struct pool_item_header *a, struct pool_item_header *b) |
| |
{ |
| |
if (a->ph_page < b->ph_page) |
| |
return (-1); |
| |
else if (a->ph_page > b->ph_page) |
| |
return (1); |
| |
else |
| |
return (0); |
| |
} |
| |
|
| |
SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare); |
| |
SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare); |
| |
|
| /* |
/* |
| * Return the pool page header based on page address. |
* Return the pool page header based on page address. |
| */ |
*/ |
| static __inline struct pool_item_header * |
static __inline struct pool_item_header * |
| pr_find_pagehead(struct pool *pp, caddr_t page) |
pr_find_pagehead(struct pool *pp, caddr_t page) |
| { |
{ |
| struct pool_item_header *ph; |
struct pool_item_header *ph, tmp; |
| |
|
| if ((pp->pr_roflags & PR_PHINPAGE) != 0) |
if ((pp->pr_roflags & PR_PHINPAGE) != 0) |
| return ((struct pool_item_header *)(page + pp->pr_phoffset)); |
return ((struct pool_item_header *)(page + pp->pr_phoffset)); |
| |
|
| for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]); |
tmp.ph_page = page; |
| ph != NULL; |
ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp); |
| ph = LIST_NEXT(ph, ph_hashlist)) { |
return ph; |
| if (ph->ph_page == page) |
|
| return (ph); |
|
| } |
|
| return (NULL); |
|
| } |
} |
| |
|
| /* |
/* |
| Line 304 pr_rmpage(struct pool *pp, struct pool_i |
|
| Line 317 pr_rmpage(struct pool *pp, struct pool_i |
|
| { |
{ |
| int s; |
int s; |
| |
|
| |
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock) || pq != NULL); |
| |
|
| /* |
/* |
| * If the page was idle, decrement the idle page count. |
* If the page was idle, decrement the idle page count. |
| */ |
*/ |
| Line 322 pr_rmpage(struct pool *pp, struct pool_i |
|
| Line 337 pr_rmpage(struct pool *pp, struct pool_i |
|
| /* |
/* |
| * Unlink a page from the pool and release it (or queue it for release). |
* Unlink a page from the pool and release it (or queue it for release). |
| */ |
*/ |
| TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| |
if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
| |
SPLAY_REMOVE(phtree, &pp->pr_phtree, ph); |
| if (pq) { |
if (pq) { |
| TAILQ_INSERT_HEAD(pq, ph, ph_pagelist); |
LIST_INSERT_HEAD(pq, ph, ph_pagelist); |
| } else { |
} else { |
| pool_allocator_free(pp, ph->ph_page); |
pool_allocator_free(pp, ph->ph_page); |
| if ((pp->pr_roflags & PR_PHINPAGE) == 0) { |
if ((pp->pr_roflags & PR_PHINPAGE) == 0) { |
| LIST_REMOVE(ph, ph_hashlist); |
|
| s = splvm(); |
s = splvm(); |
| pool_put(&phpool, ph); |
pool_put(&phpool, ph); |
| splx(s); |
splx(s); |
| Line 337 pr_rmpage(struct pool *pp, struct pool_i |
|
| Line 353 pr_rmpage(struct pool *pp, struct pool_i |
|
| pp->pr_npages--; |
pp->pr_npages--; |
| pp->pr_npagefree++; |
pp->pr_npagefree++; |
| |
|
| if (pp->pr_curpage == ph) { |
pool_update_curpage(pp); |
| /* |
} |
| * Find a new non-empty page header, if any. |
|
| * Start search from the page head, to increase the |
|
| * chance for "high water" pages to be freed. |
|
| */ |
|
| TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) |
|
| if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
| break; |
|
| |
|
| pp->pr_curpage = ph; |
/* |
| } |
* Initialize all the pools listed in the "pools" link set. |
| |
*/ |
| |
void |
| |
link_pool_init(void) |
| |
{ |
| |
__link_set_decl(pools, struct link_pool_init); |
| |
struct link_pool_init * const *pi; |
| |
|
| |
__link_set_foreach(pi, pools) |
| |
pool_init((*pi)->pp, (*pi)->size, (*pi)->align, |
| |
(*pi)->align_offset, (*pi)->flags, (*pi)->wchan, |
| |
(*pi)->palloc); |
| } |
} |
| |
|
| /* |
/* |
|
|
| pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags, |
pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags, |
| const char *wchan, struct pool_allocator *palloc) |
const char *wchan, struct pool_allocator *palloc) |
| { |
{ |
| int off, slack, i; |
int off, slack; |
| |
size_t trysize, phsize; |
| |
int s; |
| |
|
| #ifdef POOL_DIAGNOSTIC |
#ifdef POOL_DIAGNOSTIC |
| /* |
/* |
| Line 425 pool_init(struct pool *pp, size_t size, |
|
| Line 447 pool_init(struct pool *pp, size_t size, |
|
| /* |
/* |
| * Initialize the pool structure. |
* Initialize the pool structure. |
| */ |
*/ |
| TAILQ_INIT(&pp->pr_pagelist); |
LIST_INIT(&pp->pr_emptypages); |
| |
LIST_INIT(&pp->pr_fullpages); |
| |
LIST_INIT(&pp->pr_partpages); |
| TAILQ_INIT(&pp->pr_cachelist); |
TAILQ_INIT(&pp->pr_cachelist); |
| pp->pr_curpage = NULL; |
pp->pr_curpage = NULL; |
| pp->pr_npages = 0; |
pp->pr_npages = 0; |
| Line 451 pool_init(struct pool *pp, size_t size, |
|
| Line 475 pool_init(struct pool *pp, size_t size, |
|
| |
|
| /* |
/* |
| * Decide whether to put the page header off page to avoid |
* Decide whether to put the page header off page to avoid |
| * wasting too large a part of the page. Off-page page headers |
* wasting too large a part of the page or too big item. |
| * go on a hash table, so we can match a returned item |
* Off-page page headers go on a hash table, so we can match |
| * with its header based on the page address. |
* a returned item with its header based on the page address. |
| * We use 1/16 of the page size as the threshold (XXX: tune) |
* We use 1/16 of the page size and about 8 times of the item |
| |
* size as the threshold (XXX: tune) |
| |
* |
| |
* However, we'll put the header into the page if we can put |
| |
* it without wasting any items. |
| |
* |
| |
* Silently enforce `0 <= ioff < align'. |
| */ |
*/ |
| if (pp->pr_size < palloc->pa_pagesz/16) { |
pp->pr_itemoffset = ioff %= align; |
| |
/* See the comment below about reserved bytes. */ |
| |
trysize = palloc->pa_pagesz - ((align - ioff) % align); |
| |
phsize = ALIGN(sizeof(struct pool_item_header)); |
| |
if (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) || |
| |
trysize / pp->pr_size == (trysize - phsize) / pp->pr_size) { |
| /* Use the end of the page for the page header */ |
/* Use the end of the page for the page header */ |
| pp->pr_roflags |= PR_PHINPAGE; |
pp->pr_roflags |= PR_PHINPAGE; |
| pp->pr_phoffset = off = palloc->pa_pagesz - |
pp->pr_phoffset = off = palloc->pa_pagesz - phsize; |
| ALIGN(sizeof(struct pool_item_header)); |
|
| } else { |
} else { |
| /* The page header will be taken from our page header pool */ |
/* The page header will be taken from our page header pool */ |
| pp->pr_phoffset = 0; |
pp->pr_phoffset = 0; |
| off = palloc->pa_pagesz; |
off = palloc->pa_pagesz; |
| for (i = 0; i < PR_HASHTABSIZE; i++) { |
SPLAY_INIT(&pp->pr_phtree); |
| LIST_INIT(&pp->pr_hashtab[i]); |
|
| } |
|
| } |
} |
| |
|
| /* |
/* |
| * Alignment is to take place at `ioff' within the item. This means |
* Alignment is to take place at `ioff' within the item. This means |
| * we must reserve up to `align - 1' bytes on the page to allow |
* we must reserve up to `align - 1' bytes on the page to allow |
| * appropriate positioning of each item. |
* appropriate positioning of each item. |
| * |
|
| * Silently enforce `0 <= ioff < align'. |
|
| */ |
*/ |
| pp->pr_itemoffset = ioff = ioff % align; |
|
| pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size; |
pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size; |
| KASSERT(pp->pr_itemsperpage != 0); |
KASSERT(pp->pr_itemsperpage != 0); |
| |
|
| Line 538 pool_init(struct pool *pp, size_t size, |
|
| Line 567 pool_init(struct pool *pp, size_t size, |
|
| simple_unlock(&pool_head_slock); |
simple_unlock(&pool_head_slock); |
| |
|
| /* Insert this into the list of pools using this allocator. */ |
/* Insert this into the list of pools using this allocator. */ |
| |
s = splvm(); |
| simple_lock(&palloc->pa_slock); |
simple_lock(&palloc->pa_slock); |
| 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); |
| } |
} |
| |
|
| /* |
/* |
| Line 551 pool_destroy(struct pool *pp) |
|
| Line 582 pool_destroy(struct pool *pp) |
|
| { |
{ |
| struct pool_item_header *ph; |
struct pool_item_header *ph; |
| struct pool_cache *pc; |
struct pool_cache *pc; |
| |
int s; |
| |
|
| /* Locking order: pool_allocator -> pool */ |
/* Locking order: pool_allocator -> pool */ |
| |
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); |
| simple_unlock(&pp->pr_alloc->pa_slock); |
simple_unlock(&pp->pr_alloc->pa_slock); |
| |
splx(s); |
| |
|
| /* Destroy all caches for this pool. */ |
/* Destroy all caches for this pool. */ |
| while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL) |
while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL) |
| Line 570 pool_destroy(struct pool *pp) |
|
| Line 604 pool_destroy(struct pool *pp) |
|
| #endif |
#endif |
| |
|
| /* Remove all pages */ |
/* Remove all pages */ |
| while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL) |
while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) |
| pr_rmpage(pp, ph, NULL); |
pr_rmpage(pp, ph, NULL); |
| |
KASSERT(LIST_EMPTY(&pp->pr_fullpages)); |
| |
KASSERT(LIST_EMPTY(&pp->pr_partpages)); |
| |
|
| /* Remove from global pool list */ |
/* Remove from global pool list */ |
| simple_lock(&pool_head_slock); |
simple_lock(&pool_head_slock); |
| Line 600 pool_set_drain_hook(struct pool *pp, voi |
|
| Line 636 pool_set_drain_hook(struct pool *pp, voi |
|
| pp->pr_drain_hook_arg = arg; |
pp->pr_drain_hook_arg = arg; |
| } |
} |
| |
|
| static __inline struct pool_item_header * |
static struct pool_item_header * |
| pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags) |
pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags) |
| { |
{ |
| struct pool_item_header *ph; |
struct pool_item_header *ph; |
| Line 634 pool_get(struct pool *pp, int flags) |
|
| Line 670 pool_get(struct pool *pp, int flags) |
|
| void *v; |
void *v; |
| |
|
| #ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
| |
if (__predict_false(pp->pr_itemsperpage == 0)) |
| |
panic("pool_get: pool %p: pr_itemsperpage is zero, " |
| |
"pool not initialized?", pp); |
| if (__predict_false(curlwp == NULL && doing_shutdown == 0 && |
if (__predict_false(curlwp == NULL && doing_shutdown == 0 && |
| (flags & PR_WAITOK) != 0)) |
(flags & PR_WAITOK) != 0)) |
| panic("pool_get: %s: must have NOWAIT", pp->pr_wchan); |
panic("pool_get: %s: must have NOWAIT", pp->pr_wchan); |
| Line 729 pool_get(struct pool *pp, int flags) |
|
| Line 768 pool_get(struct pool *pp, int flags) |
|
| v = pool_allocator_alloc(pp, flags); |
v = pool_allocator_alloc(pp, flags); |
| if (__predict_true(v != NULL)) |
if (__predict_true(v != NULL)) |
| ph = pool_alloc_item_header(pp, v, flags); |
ph = pool_alloc_item_header(pp, v, flags); |
| simple_lock(&pp->pr_slock); |
|
| pr_enter(pp, file, line); |
|
| |
|
| if (__predict_false(v == NULL || ph == NULL)) { |
if (__predict_false(v == NULL || ph == NULL)) { |
| if (v != NULL) |
if (v != NULL) |
| pool_allocator_free(pp, v); |
pool_allocator_free(pp, v); |
| |
|
| |
simple_lock(&pp->pr_slock); |
| |
pr_enter(pp, file, line); |
| |
|
| /* |
/* |
| * We were unable to allocate a page or item |
* We were unable to allocate a page or item |
| * header, but we released the lock during |
* header, but we released the lock during |
| Line 767 pool_get(struct pool *pp, int flags) |
|
| Line 807 pool_get(struct pool *pp, int flags) |
|
| } |
} |
| |
|
| /* We have more memory; add it to the pool */ |
/* We have more memory; add it to the pool */ |
| |
simple_lock(&pp->pr_slock); |
| |
pr_enter(pp, file, line); |
| pool_prime_page(pp, v, ph); |
pool_prime_page(pp, v, ph); |
| pp->pr_npagealloc++; |
pp->pr_npagealloc++; |
| |
|
| /* Start the allocation process over. */ |
/* Start the allocation process over. */ |
| goto startover; |
goto startover; |
| } |
} |
| |
|
| if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) { |
if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) { |
| pr_leave(pp); |
pr_leave(pp); |
| simple_unlock(&pp->pr_slock); |
simple_unlock(&pp->pr_slock); |
| Line 814 pool_get(struct pool *pp, int flags) |
|
| Line 855 pool_get(struct pool *pp, int flags) |
|
| panic("pool_get: nidle inconsistent"); |
panic("pool_get: nidle inconsistent"); |
| #endif |
#endif |
| pp->pr_nidle--; |
pp->pr_nidle--; |
| |
|
| |
/* |
| |
* This page was previously empty. Move it to the list of |
| |
* partially-full pages. This page is already curpage. |
| |
*/ |
| |
LIST_REMOVE(ph, ph_pagelist); |
| |
LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist); |
| } |
} |
| ph->ph_nmissing++; |
ph->ph_nmissing++; |
| if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) { |
if (TAILQ_EMPTY(&ph->ph_itemlist)) { |
| #ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
| if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) { |
if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) { |
| pr_leave(pp); |
pr_leave(pp); |
| Line 826 pool_get(struct pool *pp, int flags) |
|
| Line 874 pool_get(struct pool *pp, int flags) |
|
| } |
} |
| #endif |
#endif |
| /* |
/* |
| * Find a new non-empty page header, if any. |
* This page is now full. Move it to the full list |
| * Start search from the page head, to increase |
* and select a new current page. |
| * the chance for "high water" pages to be freed. |
|
| * |
|
| * Migrate empty pages to the end of the list. This |
|
| * will speed the update of curpage as pages become |
|
| * idle. Empty pages intermingled with idle pages |
|
| * is no big deal. As soon as a page becomes un-empty, |
|
| * it will move back to the head of the list. |
|
| */ |
*/ |
| TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist); |
| TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) |
pool_update_curpage(pp); |
| if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
| break; |
|
| |
|
| pp->pr_curpage = ph; |
|
| } |
} |
| |
|
| pp->pr_nget++; |
pp->pr_nget++; |
| Line 935 pool_do_put(struct pool *pp, void *v) |
|
| Line 972 pool_do_put(struct pool *pp, void *v) |
|
| } |
} |
| |
|
| /* |
/* |
| * If this page is now complete, do one of two things: |
* If this page is now empty, do one of two things: |
| |
* |
| |
* (1) If we have more pages than the page high water mark, |
| |
* free the page back to the system. ONLY CONSIDER |
| |
* FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE |
| |
* CLAIM. |
| * |
* |
| * (1) If we have more pages than the page high water |
* (2) Otherwise, move the page to the empty page list. |
| * mark, free the page back to the system. |
|
| * |
* |
| * (2) Move it to the end of the page list, so that |
* Either way, select a new current page (so we use a partially-full |
| * we minimize our chances of fragmenting the |
* page if one is available). |
| * pool. Idle pages migrate to the end (along with |
|
| * completely empty pages, so that we find un-empty |
|
| * pages more quickly when we update curpage) of the |
|
| * list so they can be more easily swept up by |
|
| * the pagedaemon when pages are scarce. |
|
| */ |
*/ |
| if (ph->ph_nmissing == 0) { |
if (ph->ph_nmissing == 0) { |
| pp->pr_nidle++; |
pp->pr_nidle++; |
| if (pp->pr_npages > pp->pr_maxpages || |
if (pp->pr_npages > pp->pr_minpages && |
| (pp->pr_alloc->pa_flags & PA_WANT) != 0) { |
(pp->pr_npages > pp->pr_maxpages || |
| |
(pp->pr_alloc->pa_flags & PA_WANT) != 0)) { |
| |
simple_unlock(&pp->pr_slock); |
| pr_rmpage(pp, ph, NULL); |
pr_rmpage(pp, ph, NULL); |
| |
simple_lock(&pp->pr_slock); |
| } else { |
} else { |
| TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist); |
| |
|
| /* |
/* |
| * Update the timestamp on the page. A page must |
* Update the timestamp on the page. A page must |
| Line 966 pool_do_put(struct pool *pp, void *v) |
|
| Line 1005 pool_do_put(struct pool *pp, void *v) |
|
| s = splclock(); |
s = splclock(); |
| ph->ph_time = mono_time; |
ph->ph_time = mono_time; |
| splx(s); |
splx(s); |
| |
|
| /* |
|
| * Update the current page pointer. Just look for |
|
| * the first page with any free items. |
|
| * |
|
| * XXX: Maybe we want an option to look for the |
|
| * page with the fewest available items, to minimize |
|
| * fragmentation? |
|
| */ |
|
| TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) |
|
| if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
| break; |
|
| |
|
| pp->pr_curpage = ph; |
|
| } |
} |
| |
pool_update_curpage(pp); |
| } |
} |
| |
|
| /* |
/* |
| * If the page has just become un-empty, move it to the head of |
* If the page was previously completely full, move it to the |
| * the list, and make it the current page. The next allocation |
* partially-full list and make it the current page. The next |
| * will get the item from this page, instead of further fragmenting |
* allocation will get the item from this page, instead of |
| * the pool. |
* further fragmenting the pool. |
| */ |
*/ |
| else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) { |
else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) { |
| TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist); |
| pp->pr_curpage = ph; |
pp->pr_curpage = ph; |
| } |
} |
| } |
} |
| Line 1050 pool_prime(struct pool *pp, int n) |
|
| Line 1077 pool_prime(struct pool *pp, int n) |
|
| cp = pool_allocator_alloc(pp, PR_NOWAIT); |
cp = pool_allocator_alloc(pp, PR_NOWAIT); |
| if (__predict_true(cp != NULL)) |
if (__predict_true(cp != NULL)) |
| ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); |
ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); |
| simple_lock(&pp->pr_slock); |
|
| |
|
| if (__predict_false(cp == NULL || ph == NULL)) { |
if (__predict_false(cp == NULL || ph == NULL)) { |
| if (cp != NULL) |
if (cp != NULL) |
| pool_allocator_free(pp, cp); |
pool_allocator_free(pp, cp); |
| |
simple_lock(&pp->pr_slock); |
| break; |
break; |
| } |
} |
| |
|
| |
simple_lock(&pp->pr_slock); |
| pool_prime_page(pp, cp, ph); |
pool_prime_page(pp, cp, ph); |
| pp->pr_npagealloc++; |
pp->pr_npagealloc++; |
| pp->pr_minpages++; |
pp->pr_minpages++; |
| Line 1083 pool_prime_page(struct pool *pp, caddr_t |
|
| Line 1111 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)); |
| |
|
| #ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
| if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0) |
if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0) |
| panic("pool_prime_page: %s: unaligned page", pp->pr_wchan); |
panic("pool_prime_page: %s: unaligned page", pp->pr_wchan); |
| #endif |
#endif |
| |
|
| if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
|
| LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)], |
|
| ph, ph_hashlist); |
|
| |
|
| /* |
/* |
| * Insert page header. |
* Insert page header. |
| */ |
*/ |
| TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist); |
LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist); |
| TAILQ_INIT(&ph->ph_itemlist); |
TAILQ_INIT(&ph->ph_itemlist); |
| ph->ph_page = storage; |
ph->ph_page = storage; |
| ph->ph_nmissing = 0; |
ph->ph_nmissing = 0; |
| memset(&ph->ph_time, 0, sizeof(ph->ph_time)); |
s = splclock(); |
| |
ph->ph_time = mono_time; |
| |
splx(s); |
| |
if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
| |
SPLAY_INSERT(phtree, &pp->pr_phtree, ph); |
| |
|
| pp->pr_nidle++; |
pp->pr_nidle++; |
| |
|
| Line 1148 pool_prime_page(struct pool *pp, caddr_t |
|
| Line 1179 pool_prime_page(struct pool *pp, caddr_t |
|
| |
|
| /* |
/* |
| * Used by pool_get() when nitems drops below the low water mark. This |
* Used by pool_get() when nitems drops below the low water mark. This |
| * is used to catch up nitmes with the low water mark. |
* is used to catch up pr_nitems with the low water mark. |
| * |
* |
| * Note 1, we never wait for memory here, we let the caller decide what to do. |
* Note 1, we never wait for memory here, we let the caller decide what to do. |
| * |
* |
| Line 1173 pool_catchup(struct pool *pp) |
|
| Line 1204 pool_catchup(struct pool *pp) |
|
| cp = pool_allocator_alloc(pp, PR_NOWAIT); |
cp = pool_allocator_alloc(pp, PR_NOWAIT); |
| if (__predict_true(cp != NULL)) |
if (__predict_true(cp != NULL)) |
| ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); |
ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); |
| simple_lock(&pp->pr_slock); |
|
| if (__predict_false(cp == NULL || ph == NULL)) { |
if (__predict_false(cp == NULL || ph == NULL)) { |
| if (cp != NULL) |
if (cp != NULL) |
| pool_allocator_free(pp, cp); |
pool_allocator_free(pp, cp); |
| error = ENOMEM; |
error = ENOMEM; |
| |
simple_lock(&pp->pr_slock); |
| break; |
break; |
| } |
} |
| |
simple_lock(&pp->pr_slock); |
| pool_prime_page(pp, cp, ph); |
pool_prime_page(pp, cp, ph); |
| pp->pr_npagealloc++; |
pp->pr_npagealloc++; |
| } |
} |
| Line 1187 pool_catchup(struct pool *pp) |
|
| Line 1219 pool_catchup(struct pool *pp) |
|
| return (error); |
return (error); |
| } |
} |
| |
|
| |
static void |
| |
pool_update_curpage(struct pool *pp) |
| |
{ |
| |
|
| |
pp->pr_curpage = LIST_FIRST(&pp->pr_partpages); |
| |
if (pp->pr_curpage == NULL) { |
| |
pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages); |
| |
} |
| |
} |
| |
|
| void |
void |
| pool_setlowat(struct pool *pp, int n) |
pool_setlowat(struct pool *pp, int n) |
| { |
{ |
| Line 1260 pool_reclaim(struct pool *pp) |
|
| Line 1302 pool_reclaim(struct pool *pp) |
|
| struct pool_cache *pc; |
struct pool_cache *pc; |
| struct timeval curtime; |
struct timeval curtime; |
| struct pool_pagelist pq; |
struct pool_pagelist pq; |
| |
struct timeval diff; |
| int s; |
int s; |
| |
|
| if (pp->pr_drain_hook != NULL) { |
if (pp->pr_drain_hook != NULL) { |
| Line 1273 pool_reclaim(struct pool *pp) |
|
| Line 1316 pool_reclaim(struct pool *pp) |
|
| return (0); |
return (0); |
| pr_enter(pp, file, line); |
pr_enter(pp, file, line); |
| |
|
| TAILQ_INIT(&pq); |
LIST_INIT(&pq); |
| |
|
| /* |
/* |
| * Reclaim items from the pool's caches. |
* Reclaim items from the pool's caches. |
| Line 1285 pool_reclaim(struct pool *pp) |
|
| Line 1328 pool_reclaim(struct pool *pp) |
|
| curtime = mono_time; |
curtime = mono_time; |
| splx(s); |
splx(s); |
| |
|
| for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) { |
for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) { |
| phnext = TAILQ_NEXT(ph, ph_pagelist); |
phnext = LIST_NEXT(ph, ph_pagelist); |
| |
|
| /* Check our minimum page claim */ |
/* Check our minimum page claim */ |
| if (pp->pr_npages <= pp->pr_minpages) |
if (pp->pr_npages <= pp->pr_minpages) |
| break; |
break; |
| |
|
| if (ph->ph_nmissing == 0) { |
KASSERT(ph->ph_nmissing == 0); |
| struct timeval diff; |
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) |
continue; |
| continue; |
|
| |
|
| /* |
/* |
| * If freeing this page would put us below |
* If freeing this page would put us below |
| * the low water mark, stop now. |
* the low water mark, stop now. |
| */ |
*/ |
| if ((pp->pr_nitems - pp->pr_itemsperpage) < |
if ((pp->pr_nitems - pp->pr_itemsperpage) < |
| pp->pr_minitems) |
pp->pr_minitems) |
| break; |
break; |
| |
|
| pr_rmpage(pp, ph, &pq); |
pr_rmpage(pp, ph, &pq); |
| } |
|
| } |
} |
| |
|
| pr_leave(pp); |
pr_leave(pp); |
| simple_unlock(&pp->pr_slock); |
simple_unlock(&pp->pr_slock); |
| if (TAILQ_EMPTY(&pq)) |
if (LIST_EMPTY(&pq)) |
| return (0); |
return (0); |
| |
|
| while ((ph = TAILQ_FIRST(&pq)) != NULL) { |
while ((ph = LIST_FIRST(&pq)) != NULL) { |
| TAILQ_REMOVE(&pq, ph, ph_pagelist); |
LIST_REMOVE(ph, ph_pagelist); |
| pool_allocator_free(pp, ph->ph_page); |
pool_allocator_free(pp, ph->ph_page); |
| if (pp->pr_roflags & PR_PHINPAGE) { |
if (pp->pr_roflags & PR_PHINPAGE) { |
| continue; |
continue; |
| } |
} |
| LIST_REMOVE(ph, ph_hashlist); |
|
| s = splvm(); |
s = splvm(); |
| pool_put(&phpool, ph); |
pool_put(&phpool, ph); |
| splx(s); |
splx(s); |
| Line 1407 pool_printit(struct pool *pp, const char |
|
| Line 1447 pool_printit(struct pool *pp, const char |
|
| } |
} |
| |
|
| static void |
static void |
| pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...)) |
pool_print_pagelist(struct pool_pagelist *pl, void (*pr)(const char *, ...)) |
| { |
{ |
| struct pool_item_header *ph; |
struct pool_item_header *ph; |
| struct pool_cache *pc; |
|
| struct pool_cache_group *pcg; |
|
| #ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
| struct pool_item *pi; |
struct pool_item *pi; |
| #endif |
#endif |
| |
|
| |
LIST_FOREACH(ph, pl, ph_pagelist) { |
| |
(*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n", |
| |
ph->ph_page, ph->ph_nmissing, |
| |
(u_long)ph->ph_time.tv_sec, |
| |
(u_long)ph->ph_time.tv_usec); |
| |
#ifdef DIAGNOSTIC |
| |
TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) { |
| |
if (pi->pi_magic != PI_MAGIC) { |
| |
(*pr)("\t\t\titem %p, magic 0x%x\n", |
| |
pi, pi->pi_magic); |
| |
} |
| |
} |
| |
#endif |
| |
} |
| |
} |
| |
|
| |
static void |
| |
pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...)) |
| |
{ |
| |
struct pool_item_header *ph; |
| |
struct pool_cache *pc; |
| |
struct pool_cache_group *pcg; |
| int i, print_log = 0, print_pagelist = 0, print_cache = 0; |
int i, print_log = 0, print_pagelist = 0, print_cache = 0; |
| char c; |
char c; |
| |
|
| Line 1444 pool_print1(struct pool *pp, const char |
|
| Line 1505 pool_print1(struct pool *pp, const char |
|
| if (print_pagelist == 0) |
if (print_pagelist == 0) |
| goto skip_pagelist; |
goto skip_pagelist; |
| |
|
| if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL) |
if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) |
| (*pr)("\n\tpage list:\n"); |
(*pr)("\n\tempty page list:\n"); |
| for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) { |
pool_print_pagelist(&pp->pr_emptypages, pr); |
| (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n", |
if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL) |
| ph->ph_page, ph->ph_nmissing, |
(*pr)("\n\tfull page list:\n"); |
| (u_long)ph->ph_time.tv_sec, |
pool_print_pagelist(&pp->pr_fullpages, pr); |
| (u_long)ph->ph_time.tv_usec); |
if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL) |
| #ifdef DIAGNOSTIC |
(*pr)("\n\tpartial-page list:\n"); |
| TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) { |
pool_print_pagelist(&pp->pr_partpages, pr); |
| if (pi->pi_magic != PI_MAGIC) { |
|
| (*pr)("\t\t\titem %p, magic 0x%x\n", |
|
| pi, pi->pi_magic); |
|
| } |
|
| } |
|
| #endif |
|
| } |
|
| if (pp->pr_curpage == NULL) |
if (pp->pr_curpage == NULL) |
| (*pr)("\tno current page\n"); |
(*pr)("\tno current page\n"); |
| else |
else |
| (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page); |
(*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page); |
| |
|
| skip_pagelist: |
skip_pagelist: |
| |
|
| if (print_log == 0) |
if (print_log == 0) |
| goto skip_log; |
goto skip_log; |
| |
|
| Line 1477 pool_print1(struct pool *pp, const char |
|
| Line 1531 pool_print1(struct pool *pp, const char |
|
| pr_printlog(pp, NULL, pr); |
pr_printlog(pp, NULL, pr); |
| |
|
| skip_log: |
skip_log: |
| |
|
| if (print_cache == 0) |
if (print_cache == 0) |
| goto skip_cache; |
goto skip_cache; |
| |
|
| Line 1504 pool_print1(struct pool *pp, const char |
|
| Line 1557 pool_print1(struct pool *pp, const char |
|
| } |
} |
| |
|
| skip_cache: |
skip_cache: |
| |
|
| pr_enter_check(pp, pr); |
pr_enter_check(pp, pr); |
| } |
} |
| |
|
| int |
static int |
| pool_chk(struct pool *pp, const char *label) |
pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph) |
| { |
{ |
| struct pool_item_header *ph; |
struct pool_item *pi; |
| int r = 0; |
caddr_t page; |
| |
int n; |
| |
|
| simple_lock(&pp->pr_slock); |
page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask); |
| |
if (page != ph->ph_page && |
| |
(pp->pr_roflags & PR_PHINPAGE) != 0) { |
| |
if (label != NULL) |
| |
printf("%s: ", label); |
| |
printf("pool(%p:%s): page inconsistency: page %p;" |
| |
" at page head addr %p (p %p)\n", pp, |
| |
pp->pr_wchan, ph->ph_page, |
| |
ph, page); |
| |
return 1; |
| |
} |
| |
|
| |
for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0; |
| |
pi != NULL; |
| |
pi = TAILQ_NEXT(pi,pi_list), n++) { |
| |
|
| TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) { |
#ifdef DIAGNOSTIC |
| struct pool_item *pi; |
if (pi->pi_magic != PI_MAGIC) { |
| int n; |
|
| caddr_t page; |
|
| |
|
| page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask); |
|
| if (page != ph->ph_page && |
|
| (pp->pr_roflags & PR_PHINPAGE) != 0) { |
|
| if (label != NULL) |
if (label != NULL) |
| printf("%s: ", label); |
printf("%s: ", label); |
| printf("pool(%p:%s): page inconsistency: page %p;" |
printf("pool(%s): free list modified: magic=%x;" |
| " at page head addr %p (p %p)\n", pp, |
" page %p; item ordinal %d;" |
| pp->pr_wchan, ph->ph_page, |
" addr %p (p %p)\n", |
| ph, page); |
pp->pr_wchan, pi->pi_magic, ph->ph_page, |
| r++; |
n, pi, page); |
| goto out; |
panic("pool"); |
| } |
} |
| |
#endif |
| |
page = |
| |
(caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask); |
| |
if (page == ph->ph_page) |
| |
continue; |
| |
|
| for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0; |
if (label != NULL) |
| pi != NULL; |
printf("%s: ", label); |
| pi = TAILQ_NEXT(pi,pi_list), n++) { |
printf("pool(%p:%s): page inconsistency: page %p;" |
| |
" item ordinal %d; addr %p (p %p)\n", pp, |
| |
pp->pr_wchan, ph->ph_page, |
| |
n, pi, page); |
| |
return 1; |
| |
} |
| |
return 0; |
| |
} |
| |
|
| #ifdef DIAGNOSTIC |
|
| if (pi->pi_magic != PI_MAGIC) { |
|
| if (label != NULL) |
|
| printf("%s: ", label); |
|
| printf("pool(%s): free list modified: magic=%x;" |
|
| " page %p; item ordinal %d;" |
|
| " addr %p (p %p)\n", |
|
| pp->pr_wchan, pi->pi_magic, ph->ph_page, |
|
| n, pi, page); |
|
| panic("pool"); |
|
| } |
|
| #endif |
|
| page = |
|
| (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask); |
|
| if (page == ph->ph_page) |
|
| continue; |
|
| |
|
| if (label != NULL) |
int |
| printf("%s: ", label); |
pool_chk(struct pool *pp, const char *label) |
| printf("pool(%p:%s): page inconsistency: page %p;" |
{ |
| " item ordinal %d; addr %p (p %p)\n", pp, |
struct pool_item_header *ph; |
| pp->pr_wchan, ph->ph_page, |
int r = 0; |
| n, pi, page); |
|
| r++; |
simple_lock(&pp->pr_slock); |
| |
LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) { |
| |
r = pool_chk_page(pp, label, ph); |
| |
if (r) { |
| |
goto out; |
| |
} |
| |
} |
| |
LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) { |
| |
r = pool_chk_page(pp, label, ph); |
| |
if (r) { |
| |
goto out; |
| |
} |
| |
} |
| |
LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) { |
| |
r = pool_chk_page(pp, label, ph); |
| |
if (r) { |
| goto out; |
goto out; |
| } |
} |
| } |
} |
| |
|
| out: |
out: |
| simple_unlock(&pp->pr_slock); |
simple_unlock(&pp->pr_slock); |
| return (r); |
return (r); |
| Line 1923 pool_allocator_alloc(struct pool *org, i |
|
| Line 1996 pool_allocator_alloc(struct pool *org, i |
|
| int s, freed; |
int s, freed; |
| void *res; |
void *res; |
| |
|
| |
LOCK_ASSERT(!simple_lock_held(&org->pr_slock)); |
| |
|
| do { |
do { |
| if ((res = (*pa->pa_alloc)(org, flags)) != NULL) |
if ((res = (*pa->pa_alloc)(org, flags)) != NULL) |
| return (res); |
return (res); |
| Line 1993 pool_allocator_free(struct pool *pp, voi |
|
| Line 2068 pool_allocator_free(struct pool *pp, voi |
|
| struct pool_allocator *pa = pp->pr_alloc; |
struct pool_allocator *pa = pp->pr_alloc; |
| int s; |
int s; |
| |
|
| |
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); |
| |
|
| (*pa->pa_free)(pp, v); |
(*pa->pa_free)(pp, v); |
| |
|
| s = splvm(); |
s = splvm(); |
| Line 2036 pool_page_free(struct pool *pp, void *v) |
|
| Line 2113 pool_page_free(struct pool *pp, void *v) |
|
| void * |
void * |
| pool_subpage_alloc(struct pool *pp, int flags) |
pool_subpage_alloc(struct pool *pp, int flags) |
| { |
{ |
| |
void *v; |
| return (pool_get(&psppool, flags)); |
int s; |
| |
s = splvm(); |
| |
v = pool_get(&psppool, flags); |
| |
splx(s); |
| |
return v; |
| } |
} |
| |
|
| void |
void |
| pool_subpage_free(struct pool *pp, void *v) |
pool_subpage_free(struct pool *pp, void *v) |
| { |
{ |
| |
int s; |
| |
s = splvm(); |
| pool_put(&psppool, v); |
pool_put(&psppool, v); |
| |
splx(s); |
| } |
} |
| |
|
| /* We don't provide a real nointr allocator. Maybe later. */ |
/* We don't provide a real nointr allocator. Maybe later. */ |