version 1.6, 1998/08/01 23:44:21 |
version 1.111.4.2, 2006/04/22 11:39:59 |
|
|
/* $NetBSD$ */ |
/* $NetBSD$ */ |
|
|
/*- |
/*- |
* Copyright (c) 1997 The NetBSD Foundation, Inc. |
* Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc. |
* All rights reserved. |
* All rights reserved. |
* |
* |
* This code is derived from software contributed to The NetBSD Foundation |
* This code is derived from software contributed to The NetBSD Foundation |
* by Paul Kranenburg. |
* by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace |
|
* Simulation Facility, NASA Ames Research Center. |
* |
* |
* Redistribution and use in source and binary forms, with or without |
* Redistribution and use in source and binary forms, with or without |
* modification, are permitted provided that the following conditions |
* modification, are permitted provided that the following conditions |
|
|
* documentation and/or other materials provided with the distribution. |
* documentation and/or other materials provided with the distribution. |
* 3. All advertising materials mentioning features or use of this software |
* 3. All advertising materials mentioning features or use of this software |
* must display the following acknowledgement: |
* must display the following acknowledgement: |
* This product includes software developed by the NetBSD |
* This product includes software developed by the NetBSD |
* Foundation, Inc. and its contributors. |
* Foundation, Inc. and its contributors. |
* 4. Neither the name of The NetBSD Foundation nor the names of its |
* 4. Neither the name of The NetBSD Foundation nor the names of its |
* contributors may be used to endorse or promote products derived |
* contributors may be used to endorse or promote products derived |
* from this software without specific prior written permission. |
* from this software without specific prior written permission. |
|
|
* POSSIBILITY OF SUCH DAMAGE. |
* POSSIBILITY OF SUCH DAMAGE. |
*/ |
*/ |
|
|
|
#include <sys/cdefs.h> |
|
__KERNEL_RCSID(0, "$NetBSD$"); |
|
|
|
#include "opt_pool.h" |
|
#include "opt_poollog.h" |
|
#include "opt_lockdebug.h" |
|
|
#include <sys/param.h> |
#include <sys/param.h> |
#include <sys/systm.h> |
#include <sys/systm.h> |
#include <sys/proc.h> |
#include <sys/proc.h> |
|
|
#include <sys/malloc.h> |
#include <sys/malloc.h> |
#include <sys/lock.h> |
#include <sys/lock.h> |
#include <sys/pool.h> |
#include <sys/pool.h> |
|
#include <sys/syslog.h> |
|
|
#include <vm/vm.h> |
|
#include <vm/vm_kern.h> |
|
|
|
#if defined(UVM) |
|
#include <uvm/uvm.h> |
#include <uvm/uvm.h> |
#endif |
|
|
|
/* |
/* |
* 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 */ |
TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head); |
LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head); |
|
|
/* Private pool for page header structures */ |
/* Private pool for page header structures */ |
static struct pool phpool; |
#define PHPOOL_MAX 8 |
|
static struct pool phpool[PHPOOL_MAX]; |
|
#define PHPOOL_FREELIST_NELEM(idx) (((idx) == 0) ? 0 : (1 << (idx))) |
|
|
|
#ifdef POOL_SUBPAGE |
|
/* Pool of subpages for use by normal pools. */ |
|
static struct pool psppool; |
|
#endif |
|
|
|
static void *pool_page_alloc_meta(struct pool *, int); |
|
static void pool_page_free_meta(struct pool *, void *); |
|
|
|
/* allocator for pool metadata */ |
|
static struct pool_allocator pool_allocator_meta = { |
|
pool_page_alloc_meta, pool_page_free_meta |
|
}; |
|
|
/* # of seconds to retain page after last use */ |
/* # of seconds to retain page after last use */ |
int pool_inactive_time = 10; |
int pool_inactive_time = 10; |
|
|
/* Next candidate for drainage (see pool_drain()) */ |
/* Next candidate for drainage (see pool_drain()) */ |
static struct pool *drainpp = NULL; |
static struct pool *drainpp; |
|
|
|
/* This spin lock protects both pool_head and drainpp. */ |
|
struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER; |
|
|
|
typedef uint8_t pool_item_freelist_t; |
|
|
struct pool_item_header { |
struct pool_item_header { |
/* Page headers */ |
/* Page headers */ |
TAILQ_ENTRY(pool_item_header) |
|
ph_pagelist; /* pool page list */ |
|
TAILQ_HEAD(,pool_item) ph_itemlist; /* chunk list for this page */ |
|
LIST_ENTRY(pool_item_header) |
LIST_ENTRY(pool_item_header) |
ph_hashlist; /* Off-page page headers */ |
ph_pagelist; /* pool page list */ |
int ph_nmissing; /* # of chunks in use */ |
SPLAY_ENTRY(pool_item_header) |
|
ph_node; /* Off-page page headers */ |
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 */ |
|
union { |
|
/* !PR_NOTOUCH */ |
|
struct { |
|
LIST_HEAD(, pool_item) |
|
phu_itemlist; /* chunk list for this page */ |
|
} phu_normal; |
|
/* PR_NOTOUCH */ |
|
struct { |
|
uint16_t |
|
phu_off; /* start offset in page */ |
|
pool_item_freelist_t |
|
phu_firstfree; /* first free item */ |
|
/* |
|
* XXX it might be better to use |
|
* a simple bitmap and ffs(3) |
|
*/ |
|
} phu_notouch; |
|
} ph_u; |
|
uint16_t ph_nmissing; /* # of chunks in use */ |
}; |
}; |
|
#define ph_itemlist ph_u.phu_normal.phu_itemlist |
|
#define ph_off ph_u.phu_notouch.phu_off |
|
#define ph_firstfree ph_u.phu_notouch.phu_firstfree |
|
|
struct pool_item { |
struct pool_item { |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
int pi_magic; |
u_int pi_magic; |
#define PI_MAGIC 0xdeadbeef |
|
#endif |
#endif |
|
#define PI_MAGIC 0xdeadbeefU |
/* Other entries use only this list entry */ |
/* Other entries use only this list entry */ |
TAILQ_ENTRY(pool_item) pi_list; |
LIST_ENTRY(pool_item) pi_list; |
}; |
}; |
|
|
|
#define POOL_NEEDS_CATCHUP(pp) \ |
|
((pp)->pr_nitems < (pp)->pr_minitems) |
|
|
#define PR_HASH_INDEX(pp,addr) \ |
/* |
(((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1)) |
* Pool cache management. |
|
* |
|
* Pool caches provide a way for constructed objects to be cached by the |
|
* pool subsystem. This can lead to performance improvements by avoiding |
|
* needless object construction/destruction; it is deferred until absolutely |
|
* necessary. |
|
* |
|
* Caches are grouped into cache groups. Each cache group references |
|
* up to 16 constructed objects. When a cache allocates an object |
|
* from the pool, it calls the object's constructor and places it into |
|
* a cache group. When a cache group frees an object back to the pool, |
|
* it first calls the object's destructor. This allows the object to |
|
* persist in constructed form while freed to the cache. |
|
* |
|
* Multiple caches may exist for each pool. This allows a single |
|
* object type to have multiple constructed forms. The pool references |
|
* each cache, so that when a pool is drained by the pagedaemon, it can |
|
* drain each individual cache as well. Each time a cache is drained, |
|
* the most idle cache group is freed to the pool in its entirety. |
|
* |
|
* Pool caches are layed on top of pools. By layering them, we can avoid |
|
* the complexity of cache management for pools which would not benefit |
|
* from it. |
|
*/ |
|
|
|
/* The cache group pool. */ |
|
static struct pool pcgpool; |
|
|
static struct pool_item_header |
static void pool_cache_reclaim(struct pool_cache *, struct pool_pagelist *, |
*pr_find_pagehead __P((struct pool *, caddr_t)); |
struct pool_cache_grouplist *); |
static void pr_rmpage __P((struct pool *, struct pool_item_header *)); |
static void pcg_grouplist_free(struct pool_cache_grouplist *); |
static int pool_prime_page __P((struct pool *, caddr_t)); |
|
static void *pool_page_alloc __P((unsigned long, int, int)); |
static int pool_catchup(struct pool *); |
static void pool_page_free __P((void *, unsigned long, int)); |
static void pool_prime_page(struct pool *, caddr_t, |
int pool_chk __P((struct pool *, char *)); |
struct pool_item_header *); |
|
static void pool_update_curpage(struct pool *); |
|
|
|
static int pool_grow(struct pool *, int); |
|
void *pool_allocator_alloc(struct pool *, int); |
|
void pool_allocator_free(struct pool *, void *); |
|
|
|
static void pool_print_pagelist(struct pool *, struct pool_pagelist *, |
|
void (*)(const char *, ...)); |
|
static void pool_print1(struct pool *, const char *, |
|
void (*)(const char *, ...)); |
|
|
|
static int pool_chk_page(struct pool *, const char *, |
|
struct pool_item_header *); |
|
|
#ifdef POOL_DIAGNOSTIC |
|
/* |
/* |
* Pool log entry. An array of these is allocated in pool_create(). |
* Pool log entry. An array of these is allocated in pool_init(). |
*/ |
*/ |
struct pool_log { |
struct pool_log { |
const char *pl_file; |
const char *pl_file; |
long pl_line; |
long pl_line; |
int pl_action; |
int pl_action; |
#define PRLOG_GET 1 |
#define PRLOG_GET 1 |
#define PRLOG_PUT 2 |
#define PRLOG_PUT 2 |
void *pl_addr; |
void *pl_addr; |
}; |
}; |
|
|
|
#ifdef POOL_DIAGNOSTIC |
/* Number of entries in pool log buffers */ |
/* Number of entries in pool log buffers */ |
int pool_logsize = 10; |
#ifndef POOL_LOGSIZE |
|
#define POOL_LOGSIZE 10 |
|
#endif |
|
|
static void pr_log __P((struct pool *, void *, int, const char *, long)); |
int pool_logsize = POOL_LOGSIZE; |
static void pr_printlog __P((struct pool *)); |
|
|
|
static __inline__ void |
static inline void |
pr_log(pp, v, action, file, line) |
pr_log(struct pool *pp, void *v, int action, const char *file, long line) |
struct pool *pp; |
|
void *v; |
|
int action; |
|
const char *file; |
|
long line; |
|
{ |
{ |
int n = pp->pr_curlogentry; |
int n = pp->pr_curlogentry; |
struct pool_log *pl; |
struct pool_log *pl; |
|
|
if ((pp->pr_flags & PR_LOGGING) == 0) |
if ((pp->pr_roflags & PR_LOGGING) == 0) |
return; |
return; |
|
|
/* |
/* |
Line 160 pr_log(pp, v, action, file, line) |
|
Line 239 pr_log(pp, v, action, file, line) |
|
} |
} |
|
|
static void |
static void |
pr_printlog(pp) |
pr_printlog(struct pool *pp, struct pool_item *pi, |
struct pool *pp; |
void (*pr)(const char *, ...)) |
{ |
{ |
int i = pp->pr_logsize; |
int i = pp->pr_logsize; |
int n = pp->pr_curlogentry; |
int n = pp->pr_curlogentry; |
|
|
if ((pp->pr_flags & PR_LOGGING) == 0) |
if ((pp->pr_roflags & PR_LOGGING) == 0) |
return; |
return; |
|
|
pool_print(pp, "printlog"); |
|
|
|
/* |
/* |
* Print all entries in this pool's log. |
* Print all entries in this pool's log. |
*/ |
*/ |
while (i-- > 0) { |
while (i-- > 0) { |
struct pool_log *pl = &pp->pr_log[n]; |
struct pool_log *pl = &pp->pr_log[n]; |
if (pl->pl_action != 0) { |
if (pl->pl_action != 0) { |
printf("log entry %d:\n", i); |
if (pi == NULL || pi == pl->pl_addr) { |
printf("\taction = %s, addr = %p\n", |
(*pr)("\tlog entry %d:\n", i); |
pl->pl_action == PRLOG_GET ? "get" : "put", |
(*pr)("\t\taction = %s, addr = %p\n", |
pl->pl_addr); |
pl->pl_action == PRLOG_GET ? "get" : "put", |
printf("\tfile: %s at line %lu\n", |
pl->pl_addr); |
pl->pl_file, pl->pl_line); |
(*pr)("\t\tfile: %s at line %lu\n", |
|
pl->pl_file, pl->pl_line); |
|
} |
} |
} |
if (++n >= pp->pr_logsize) |
if (++n >= pp->pr_logsize) |
n = 0; |
n = 0; |
} |
} |
} |
} |
|
|
|
static inline void |
|
pr_enter(struct pool *pp, const char *file, long line) |
|
{ |
|
|
|
if (__predict_false(pp->pr_entered_file != NULL)) { |
|
printf("pool %s: reentrancy at file %s line %ld\n", |
|
pp->pr_wchan, file, line); |
|
printf(" previous entry at file %s line %ld\n", |
|
pp->pr_entered_file, pp->pr_entered_line); |
|
panic("pr_enter"); |
|
} |
|
|
|
pp->pr_entered_file = file; |
|
pp->pr_entered_line = line; |
|
} |
|
|
|
static inline void |
|
pr_leave(struct pool *pp) |
|
{ |
|
|
|
if (__predict_false(pp->pr_entered_file == NULL)) { |
|
printf("pool %s not entered?\n", pp->pr_wchan); |
|
panic("pr_leave"); |
|
} |
|
|
|
pp->pr_entered_file = NULL; |
|
pp->pr_entered_line = 0; |
|
} |
|
|
|
static inline void |
|
pr_enter_check(struct pool *pp, void (*pr)(const char *, ...)) |
|
{ |
|
|
|
if (pp->pr_entered_file != NULL) |
|
(*pr)("\n\tcurrently entered from file %s line %ld\n", |
|
pp->pr_entered_file, pp->pr_entered_line); |
|
} |
#else |
#else |
#define pr_log(pp, v, action, file, line) |
#define pr_log(pp, v, action, file, line) |
#define pr_printlog(pp) |
#define pr_printlog(pp, pi, pr) |
#endif |
#define pr_enter(pp, file, line) |
|
#define pr_leave(pp) |
|
#define pr_enter_check(pp, pr) |
|
#endif /* POOL_DIAGNOSTIC */ |
|
|
|
static inline int |
|
pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph, |
|
const void *v) |
|
{ |
|
const char *cp = v; |
|
int idx; |
|
|
|
KASSERT(pp->pr_roflags & PR_NOTOUCH); |
|
idx = (cp - ph->ph_page - ph->ph_off) / pp->pr_size; |
|
KASSERT(idx < pp->pr_itemsperpage); |
|
return idx; |
|
} |
|
|
|
#define PR_FREELIST_ALIGN(p) \ |
|
roundup((uintptr_t)(p), sizeof(pool_item_freelist_t)) |
|
#define PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1)) |
|
#define PR_INDEX_USED ((pool_item_freelist_t)-1) |
|
#define PR_INDEX_EOL ((pool_item_freelist_t)-2) |
|
|
|
static inline void |
|
pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph, |
|
void *obj) |
|
{ |
|
int idx = pr_item_notouch_index(pp, ph, obj); |
|
pool_item_freelist_t *freelist = PR_FREELIST(ph); |
|
|
|
KASSERT(freelist[idx] == PR_INDEX_USED); |
|
freelist[idx] = ph->ph_firstfree; |
|
ph->ph_firstfree = idx; |
|
} |
|
|
|
static inline void * |
|
pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph) |
|
{ |
|
int idx = ph->ph_firstfree; |
|
pool_item_freelist_t *freelist = PR_FREELIST(ph); |
|
|
|
KASSERT(freelist[idx] != PR_INDEX_USED); |
|
ph->ph_firstfree = freelist[idx]; |
|
freelist[idx] = PR_INDEX_USED; |
|
|
|
return ph->ph_page + ph->ph_off + idx * pp->pr_size; |
|
} |
|
|
|
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(pp, page) |
pr_find_pagehead(struct pool *pp, caddr_t page) |
struct pool *pp; |
|
caddr_t page; |
|
{ |
{ |
struct pool_item_header *ph; |
struct pool_item_header *ph, tmp; |
|
|
if ((pp->pr_flags & 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); |
|
|
static void |
|
pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq) |
|
{ |
|
struct pool_item_header *ph; |
|
int s; |
|
|
|
while ((ph = LIST_FIRST(pq)) != NULL) { |
|
LIST_REMOVE(ph, ph_pagelist); |
|
pool_allocator_free(pp, ph->ph_page); |
|
if ((pp->pr_roflags & PR_PHINPAGE) == 0) { |
|
s = splvm(); |
|
pool_put(pp->pr_phpool, ph); |
|
splx(s); |
|
} |
} |
} |
return (NULL); |
|
} |
} |
|
|
/* |
/* |
* Remove a page from the pool. |
* Remove a page from the pool. |
*/ |
*/ |
static __inline__ void |
static inline void |
pr_rmpage(pp, ph) |
pr_rmpage(struct pool *pp, struct pool_item_header *ph, |
struct pool *pp; |
struct pool_pagelist *pq) |
struct pool_item_header *ph; |
|
{ |
{ |
|
|
|
LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
|
|
/* |
/* |
* Unlink a page from the pool and release it. |
* If the page was idle, decrement the idle page count. |
*/ |
*/ |
TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); |
|
(*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype); |
|
pp->pr_npages--; |
|
pp->pr_npagefree++; |
|
|
|
if (ph->ph_nmissing == 0) { |
if (ph->ph_nmissing == 0) { |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
if (pp->pr_nidle == 0) |
if (pp->pr_nidle == 0) |
panic("pr_rmpage: nidle inconsistent"); |
panic("pr_rmpage: nidle inconsistent"); |
|
if (pp->pr_nitems < pp->pr_itemsperpage) |
|
panic("pr_rmpage: nitems inconsistent"); |
#endif |
#endif |
pp->pr_nidle--; |
pp->pr_nidle--; |
} |
} |
|
|
if ((pp->pr_flags & PR_PHINPAGE) == 0) { |
pp->pr_nitems -= pp->pr_itemsperpage; |
LIST_REMOVE(ph, ph_hashlist); |
|
pool_put(&phpool, ph); |
|
} |
|
|
|
if (pp->pr_curpage == ph) { |
/* |
/* |
* Unlink the page from the pool and queue it for release. |
* Find a new non-empty page header, if any. |
*/ |
* Start search from the page head, to increase the |
LIST_REMOVE(ph, ph_pagelist); |
* chance for "high water" pages to be freed. |
if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
*/ |
SPLAY_REMOVE(phtree, &pp->pr_phtree, ph); |
for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; |
LIST_INSERT_HEAD(pq, ph, ph_pagelist); |
ph = TAILQ_NEXT(ph, ph_pagelist)) |
|
if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
break; |
|
|
|
pp->pr_curpage = ph; |
pp->pr_npages--; |
} |
pp->pr_npagefree++; |
|
|
|
pool_update_curpage(pp); |
} |
} |
|
|
/* |
/* |
* Allocate and initialize a pool. |
* Initialize all the pools listed in the "pools" link set. |
*/ |
*/ |
struct pool * |
void |
pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype) |
link_pool_init(void) |
size_t size; |
|
u_int align; |
|
u_int ioff; |
|
int nitems; |
|
char *wchan; |
|
size_t pagesz; |
|
void *(*alloc) __P((unsigned long, int, int)); |
|
void (*release) __P((void *, unsigned long, int)); |
|
int mtype; |
|
{ |
{ |
struct pool *pp; |
__link_set_decl(pools, struct link_pool_init); |
int flags; |
struct link_pool_init * const *pi; |
|
|
pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT); |
|
if (pp == NULL) |
|
return (NULL); |
|
|
|
flags = PR_FREEHEADER; |
|
#ifdef POOL_DIAGNOSTIC |
|
if (pool_logsize != 0) |
|
flags |= PR_LOGGING; |
|
#endif |
|
|
|
pool_init(pp, size, align, ioff, flags, wchan, pagesz, |
|
alloc, release, mtype); |
|
|
|
if (nitems != 0) { |
|
if (pool_prime(pp, nitems, NULL) != 0) { |
|
pool_destroy(pp); |
|
return (NULL); |
|
} |
|
} |
|
|
|
return (pp); |
__link_set_foreach(pi, pools) |
|
pool_init((*pi)->pp, (*pi)->size, (*pi)->align, |
|
(*pi)->align_offset, (*pi)->flags, (*pi)->wchan, |
|
(*pi)->palloc); |
} |
} |
|
|
/* |
/* |
Line 309 pool_create(size, align, ioff, nitems, w |
|
Line 465 pool_create(size, align, ioff, nitems, w |
|
* static pools that must be initialized before malloc() is available. |
* static pools that must be initialized before malloc() is available. |
*/ |
*/ |
void |
void |
pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype) |
pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags, |
struct pool *pp; |
const char *wchan, struct pool_allocator *palloc) |
size_t size; |
{ |
u_int align; |
#ifdef DEBUG |
u_int ioff; |
struct pool *pp1; |
int flags; |
#endif |
char *wchan; |
size_t trysize, phsize; |
size_t pagesz; |
int off, slack, s; |
void *(*alloc) __P((unsigned long, int, int)); |
|
void (*release) __P((void *, unsigned long, int)); |
KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >= |
int mtype; |
PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1)); |
{ |
|
int off, slack; |
#ifdef DEBUG |
|
/* |
/* |
* Check that the pool hasn't already been initialised and |
* Check arguments and construct default values. |
* added to the list of all pools. |
*/ |
*/ |
if (!powerof2(pagesz) || pagesz > PAGE_SIZE) |
LIST_FOREACH(pp1, &pool_head, pr_poollist) { |
panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz); |
if (pp == pp1) |
|
panic("pool_init: pool %s already initialised", |
if (alloc == NULL && release == NULL) { |
wchan); |
alloc = pool_page_alloc; |
} |
release = pool_page_free; |
#endif |
pagesz = PAGE_SIZE; /* Rounds to PAGE_SIZE anyhow. */ |
|
} else if ((alloc != NULL && release != NULL) == 0) { |
#ifdef POOL_DIAGNOSTIC |
/* If you specifiy one, must specify both. */ |
/* |
panic("pool_init: must specify alloc and release together"); |
* Always log if POOL_DIAGNOSTIC is defined. |
} |
*/ |
|
if (pool_logsize != 0) |
if (pagesz == 0) |
flags |= PR_LOGGING; |
pagesz = PAGE_SIZE; |
#endif |
|
|
|
if (palloc == NULL) |
|
palloc = &pool_allocator_kmem; |
|
#ifdef POOL_SUBPAGE |
|
if (size > palloc->pa_pagesz) { |
|
if (palloc == &pool_allocator_kmem) |
|
palloc = &pool_allocator_kmem_fullpage; |
|
else if (palloc == &pool_allocator_nointr) |
|
palloc = &pool_allocator_nointr_fullpage; |
|
} |
|
#endif /* POOL_SUBPAGE */ |
|
if ((palloc->pa_flags & PA_INITIALIZED) == 0) { |
|
if (palloc->pa_pagesz == 0) |
|
palloc->pa_pagesz = PAGE_SIZE; |
|
|
|
TAILQ_INIT(&palloc->pa_list); |
|
|
|
simple_lock_init(&palloc->pa_slock); |
|
palloc->pa_pagemask = ~(palloc->pa_pagesz - 1); |
|
palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1; |
|
palloc->pa_flags |= PA_INITIALIZED; |
|
} |
|
|
if (align == 0) |
if (align == 0) |
align = ALIGN(1); |
align = ALIGN(1); |
|
|
|
if (size < sizeof(struct pool_item)) |
|
size = sizeof(struct pool_item); |
|
|
|
size = roundup(size, align); |
|
#ifdef DIAGNOSTIC |
|
if (size > palloc->pa_pagesz) |
|
panic("pool_init: pool item size (%lu) too large", |
|
(u_long)size); |
|
#endif |
|
|
/* |
/* |
* Initialize the pool structure. |
* Initialize the pool structure. |
*/ |
*/ |
TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist); |
LIST_INIT(&pp->pr_emptypages); |
TAILQ_INIT(&pp->pr_pagelist); |
LIST_INIT(&pp->pr_fullpages); |
|
LIST_INIT(&pp->pr_partpages); |
|
LIST_INIT(&pp->pr_cachelist); |
pp->pr_curpage = NULL; |
pp->pr_curpage = NULL; |
pp->pr_npages = 0; |
pp->pr_npages = 0; |
pp->pr_minitems = 0; |
pp->pr_minitems = 0; |
pp->pr_minpages = 0; |
pp->pr_minpages = 0; |
pp->pr_maxpages = UINT_MAX; |
pp->pr_maxpages = UINT_MAX; |
pp->pr_flags = flags; |
pp->pr_roflags = flags; |
pp->pr_size = ALIGN(size); |
pp->pr_flags = 0; |
|
pp->pr_size = size; |
pp->pr_align = align; |
pp->pr_align = align; |
pp->pr_wchan = wchan; |
pp->pr_wchan = wchan; |
pp->pr_mtype = mtype; |
pp->pr_alloc = palloc; |
pp->pr_alloc = alloc; |
pp->pr_nitems = 0; |
pp->pr_free = release; |
pp->pr_nout = 0; |
pp->pr_pagesz = pagesz; |
pp->pr_hardlimit = UINT_MAX; |
pp->pr_pagemask = ~(pagesz - 1); |
pp->pr_hardlimit_warning = NULL; |
pp->pr_pageshift = ffs(pagesz) - 1; |
pp->pr_hardlimit_ratecap.tv_sec = 0; |
|
pp->pr_hardlimit_ratecap.tv_usec = 0; |
|
pp->pr_hardlimit_warning_last.tv_sec = 0; |
|
pp->pr_hardlimit_warning_last.tv_usec = 0; |
|
pp->pr_drain_hook = NULL; |
|
pp->pr_drain_hook_arg = NULL; |
|
|
/* |
/* |
* 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 < 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_roflags & PR_NOTOUCH) == 0 && |
|
(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_flags |= PR_PHINPAGE; |
pp->pr_roflags |= PR_PHINPAGE; |
pp->pr_phoffset = off = |
pp->pr_phoffset = off = palloc->pa_pagesz - phsize; |
pagesz - 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 = pagesz; |
off = palloc->pa_pagesz; |
bzero(pp->pr_hashtab, sizeof(pp->pr_hashtab)); |
SPLAY_INIT(&pp->pr_phtree); |
} |
} |
|
|
/* |
/* |
* 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); |
|
if ((pp->pr_roflags & PR_NOTOUCH)) { |
|
int idx; |
|
|
|
for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx); |
|
idx++) { |
|
/* nothing */ |
|
} |
|
if (idx >= PHPOOL_MAX) { |
|
/* |
|
* if you see this panic, consider to tweak |
|
* PHPOOL_MAX and PHPOOL_FREELIST_NELEM. |
|
*/ |
|
panic("%s: too large itemsperpage(%d) for PR_NOTOUCH", |
|
pp->pr_wchan, pp->pr_itemsperpage); |
|
} |
|
pp->pr_phpool = &phpool[idx]; |
|
} else if ((pp->pr_roflags & PR_PHINPAGE) == 0) { |
|
pp->pr_phpool = &phpool[0]; |
|
} |
|
#if defined(DIAGNOSTIC) |
|
else { |
|
pp->pr_phpool = NULL; |
|
} |
|
#endif |
|
|
/* |
/* |
* Use the slack between the chunks and the page header |
* Use the slack between the chunks and the page header |
Line 408 pool_init(pp, size, align, ioff, flags, |
|
Line 637 pool_init(pp, size, align, ioff, flags, |
|
pp->pr_npagealloc = 0; |
pp->pr_npagealloc = 0; |
pp->pr_npagefree = 0; |
pp->pr_npagefree = 0; |
pp->pr_hiwat = 0; |
pp->pr_hiwat = 0; |
|
pp->pr_nidle = 0; |
|
|
#ifdef POOL_DIAGNOSTIC |
#ifdef POOL_DIAGNOSTIC |
if ((flags & PR_LOGGING) != 0) { |
if (flags & PR_LOGGING) { |
pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log), |
if (kmem_map == NULL || |
M_TEMP, M_NOWAIT); |
(pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log), |
if (pp->pr_log == NULL) |
M_TEMP, M_NOWAIT)) == NULL) |
pp->pr_flags &= ~PR_LOGGING; |
pp->pr_roflags &= ~PR_LOGGING; |
pp->pr_curlogentry = 0; |
pp->pr_curlogentry = 0; |
pp->pr_logsize = pool_logsize; |
pp->pr_logsize = pool_logsize; |
} |
} |
#endif |
#endif |
|
|
simple_lock_init(&pp->pr_lock); |
pp->pr_entered_file = NULL; |
|
pp->pr_entered_line = 0; |
|
|
|
simple_lock_init(&pp->pr_slock); |
|
|
/* |
/* |
* Initialize private page header pool if we haven't done so yet. |
* Initialize private page header pool and cache magazine pool if we |
|
* haven't done so yet. |
|
* XXX LOCKING. |
*/ |
*/ |
if (phpool.pr_size == 0) { |
if (phpool[0].pr_size == 0) { |
pool_init(&phpool, sizeof(struct pool_item_header), 0, 0, |
int idx; |
0, "phpool", 0, 0, 0, 0); |
for (idx = 0; idx < PHPOOL_MAX; idx++) { |
|
static char phpool_names[PHPOOL_MAX][6+1+6+1]; |
|
int nelem; |
|
size_t sz; |
|
|
|
nelem = PHPOOL_FREELIST_NELEM(idx); |
|
snprintf(phpool_names[idx], sizeof(phpool_names[idx]), |
|
"phpool-%d", nelem); |
|
sz = sizeof(struct pool_item_header); |
|
if (nelem) { |
|
sz = PR_FREELIST_ALIGN(sz) |
|
+ nelem * sizeof(pool_item_freelist_t); |
|
} |
|
pool_init(&phpool[idx], sz, 0, 0, 0, |
|
phpool_names[idx], &pool_allocator_meta); |
|
} |
|
#ifdef POOL_SUBPAGE |
|
pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0, |
|
PR_RECURSIVE, "psppool", &pool_allocator_meta); |
|
#endif |
|
pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0, |
|
0, "pcgpool", &pool_allocator_meta); |
} |
} |
|
|
return; |
/* Insert into the list of all pools. */ |
|
simple_lock(&pool_head_slock); |
|
LIST_INSERT_HEAD(&pool_head, pp, pr_poollist); |
|
simple_unlock(&pool_head_slock); |
|
|
|
/* Insert this into the list of pools using this allocator. */ |
|
s = splvm(); |
|
simple_lock(&palloc->pa_slock); |
|
TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list); |
|
simple_unlock(&palloc->pa_slock); |
|
splx(s); |
} |
} |
|
|
/* |
/* |
* De-commision a pool resource. |
* De-commision a pool resource. |
*/ |
*/ |
void |
void |
pool_destroy(pp) |
pool_destroy(struct pool *pp) |
struct pool *pp; |
|
{ |
{ |
|
struct pool_pagelist pq; |
struct pool_item_header *ph; |
struct pool_item_header *ph; |
|
int s; |
|
|
|
/* Remove from global pool list */ |
|
simple_lock(&pool_head_slock); |
|
LIST_REMOVE(pp, pr_poollist); |
|
if (drainpp == pp) |
|
drainpp = NULL; |
|
simple_unlock(&pool_head_slock); |
|
|
|
/* Remove this pool from its allocator's list of pools. */ |
|
s = splvm(); |
|
simple_lock(&pp->pr_alloc->pa_slock); |
|
TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list); |
|
simple_unlock(&pp->pr_alloc->pa_slock); |
|
splx(s); |
|
|
|
s = splvm(); |
|
simple_lock(&pp->pr_slock); |
|
|
|
KASSERT(LIST_EMPTY(&pp->pr_cachelist)); |
|
|
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
if (pp->pr_nget - pp->pr_nput != 0) { |
if (pp->pr_nout != 0) { |
pr_printlog(pp); |
pr_printlog(pp, NULL, printf); |
panic("pool_destroy: pool busy: still out: %lu\n", |
panic("pool_destroy: pool busy: still out: %u", |
pp->pr_nget - pp->pr_nput); |
pp->pr_nout); |
} |
} |
#endif |
#endif |
|
|
|
KASSERT(LIST_EMPTY(&pp->pr_fullpages)); |
|
KASSERT(LIST_EMPTY(&pp->pr_partpages)); |
|
|
/* Remove all pages */ |
/* Remove all pages */ |
if ((pp->pr_flags & PR_STATIC) == 0) |
LIST_INIT(&pq); |
while ((ph = pp->pr_pagelist.tqh_first) != NULL) |
while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) |
pr_rmpage(pp, ph); |
pr_rmpage(pp, ph, &pq); |
|
|
/* Remove from global pool list */ |
simple_unlock(&pp->pr_slock); |
TAILQ_REMOVE(&pool_head, pp, pr_poollist); |
splx(s); |
drainpp = NULL; |
|
|
pr_pagelist_free(pp, &pq); |
|
|
#ifdef POOL_DIAGNOSTIC |
#ifdef POOL_DIAGNOSTIC |
if ((pp->pr_flags & PR_LOGGING) != 0) |
if ((pp->pr_roflags & PR_LOGGING) != 0) |
free(pp->pr_log, M_TEMP); |
free(pp->pr_log, M_TEMP); |
#endif |
#endif |
|
} |
|
|
if (pp->pr_flags & PR_FREEHEADER) |
void |
free(pp, M_POOL); |
pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg) |
|
{ |
|
|
|
/* XXX no locking -- must be used just after pool_init() */ |
|
#ifdef DIAGNOSTIC |
|
if (pp->pr_drain_hook != NULL) |
|
panic("pool_set_drain_hook(%s): already set", pp->pr_wchan); |
|
#endif |
|
pp->pr_drain_hook = fn; |
|
pp->pr_drain_hook_arg = arg; |
} |
} |
|
|
|
static struct pool_item_header * |
|
pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags) |
|
{ |
|
struct pool_item_header *ph; |
|
int s; |
|
|
|
LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0); |
|
|
|
if ((pp->pr_roflags & PR_PHINPAGE) != 0) |
|
ph = (struct pool_item_header *) (storage + pp->pr_phoffset); |
|
else { |
|
s = splvm(); |
|
ph = pool_get(pp->pr_phpool, flags); |
|
splx(s); |
|
} |
|
|
|
return (ph); |
|
} |
|
|
/* |
/* |
* Grab an item from the pool; must be called at appropriate spl level |
* Grab an item from the pool; must be called at appropriate spl level |
*/ |
*/ |
#ifdef POOL_DIAGNOSTIC |
|
void * |
void * |
_pool_get(pp, flags, file, line) |
#ifdef POOL_DIAGNOSTIC |
struct pool *pp; |
_pool_get(struct pool *pp, int flags, const char *file, long line) |
int flags; |
|
const char *file; |
|
long line; |
|
#else |
#else |
void * |
pool_get(struct pool *pp, int flags) |
pool_get(pp, flags) |
|
struct pool *pp; |
|
int flags; |
|
#endif |
#endif |
{ |
{ |
void *v; |
|
struct pool_item *pi; |
struct pool_item *pi; |
struct pool_item_header *ph; |
struct pool_item_header *ph; |
|
void *v; |
|
|
|
#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 && |
|
(flags & PR_WAITOK) != 0)) |
|
panic("pool_get: %s: must have NOWAIT", pp->pr_wchan); |
|
|
|
#endif /* DIAGNOSTIC */ |
|
#ifdef LOCKDEBUG |
|
if (flags & PR_WAITOK) |
|
simple_lock_only_held(NULL, "pool_get(PR_WAITOK)"); |
|
SCHED_ASSERT_UNLOCKED(); |
|
#endif |
|
|
|
simple_lock(&pp->pr_slock); |
|
pr_enter(pp, file, line); |
|
|
|
startover: |
|
/* |
|
* Check to see if we've reached the hard limit. If we have, |
|
* and we can wait, then wait until an item has been returned to |
|
* the pool. |
|
*/ |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK)) { |
if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) { |
pr_printlog(pp); |
pr_leave(pp); |
panic("pool_get: static"); |
simple_unlock(&pp->pr_slock); |
|
panic("pool_get: %s: crossed hard limit", pp->pr_wchan); |
} |
} |
#endif |
#endif |
|
if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) { |
|
if (pp->pr_drain_hook != NULL) { |
|
/* |
|
* Since the drain hook is going to free things |
|
* back to the pool, unlock, call the hook, re-lock, |
|
* and check the hardlimit condition again. |
|
*/ |
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
(*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags); |
|
simple_lock(&pp->pr_slock); |
|
pr_enter(pp, file, line); |
|
if (pp->pr_nout < pp->pr_hardlimit) |
|
goto startover; |
|
} |
|
|
|
if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) { |
|
/* |
|
* XXX: A warning isn't logged in this case. Should |
|
* it be? |
|
*/ |
|
pp->pr_flags |= PR_WANTED; |
|
pr_leave(pp); |
|
ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock); |
|
pr_enter(pp, file, line); |
|
goto startover; |
|
} |
|
|
|
/* |
|
* Log a message that the hard limit has been hit. |
|
*/ |
|
if (pp->pr_hardlimit_warning != NULL && |
|
ratecheck(&pp->pr_hardlimit_warning_last, |
|
&pp->pr_hardlimit_ratecap)) |
|
log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning); |
|
|
|
pp->pr_nfail++; |
|
|
simple_lock(&pp->pr_lock); |
pr_leave(pp); |
if (curproc == NULL && (flags & PR_WAITOK) != 0) |
simple_unlock(&pp->pr_slock); |
panic("pool_get: must have NOWAIT"); |
return (NULL); |
|
} |
|
|
/* |
/* |
* The convention we use is that if `curpage' is not NULL, then |
* The convention we use is that if `curpage' is not NULL, then |
Line 507 pool_get(pp, flags) |
|
Line 881 pool_get(pp, flags) |
|
* never points at a page header which has PR_PHINPAGE set and |
* never points at a page header which has PR_PHINPAGE set and |
* has no items in its bucket. |
* has no items in its bucket. |
*/ |
*/ |
again: |
|
if ((ph = pp->pr_curpage) == NULL) { |
if ((ph = pp->pr_curpage) == NULL) { |
void *v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype); |
int error; |
if (v == NULL) { |
|
if (flags & PR_URGENT) |
#ifdef DIAGNOSTIC |
panic("pool_get: urgent"); |
if (pp->pr_nitems != 0) { |
|
simple_unlock(&pp->pr_slock); |
|
printf("pool_get: %s: curpage NULL, nitems %u\n", |
|
pp->pr_wchan, pp->pr_nitems); |
|
panic("pool_get: nitems inconsistent"); |
|
} |
|
#endif |
|
|
|
/* |
|
* Call the back-end page allocator for more memory. |
|
* Release the pool lock, as the back-end page allocator |
|
* may block. |
|
*/ |
|
pr_leave(pp); |
|
error = pool_grow(pp, flags); |
|
pr_enter(pp, file, line); |
|
if (error != 0) { |
|
/* |
|
* We were unable to allocate a page or item |
|
* header, but we released the lock during |
|
* allocation, so perhaps items were freed |
|
* back to the pool. Check for this case. |
|
*/ |
|
if (pp->pr_curpage != NULL) |
|
goto startover; |
|
|
if ((flags & PR_WAITOK) == 0) { |
if ((flags & PR_WAITOK) == 0) { |
pp->pr_nfail++; |
pp->pr_nfail++; |
simple_unlock(&pp->pr_lock); |
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
return (NULL); |
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; |
pp->pr_flags |= PR_WANTED; |
simple_unlock(&pp->pr_lock); |
/* PA_WANTED is already set on the allocator. */ |
tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0); |
pr_leave(pp); |
simple_lock(&pp->pr_lock); |
ltsleep(pp, PSWP, pp->pr_wchan, hz, &pp->pr_slock); |
} else { |
pr_enter(pp, file, line); |
pp->pr_npagealloc++; |
|
pool_prime_page(pp, v); |
|
} |
} |
|
|
goto again; |
/* Start the allocation process over. */ |
|
goto startover; |
} |
} |
|
if (pp->pr_roflags & PR_NOTOUCH) { |
|
#ifdef DIAGNOSTIC |
|
if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) { |
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
panic("pool_get: %s: page empty", pp->pr_wchan); |
|
} |
|
#endif |
|
v = pr_item_notouch_get(pp, ph); |
|
#ifdef POOL_DIAGNOSTIC |
|
pr_log(pp, v, PRLOG_GET, file, line); |
|
#endif |
|
} else { |
|
v = pi = LIST_FIRST(&ph->ph_itemlist); |
|
if (__predict_false(v == NULL)) { |
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
panic("pool_get: %s: page empty", pp->pr_wchan); |
|
} |
|
#ifdef DIAGNOSTIC |
|
if (__predict_false(pp->pr_nitems == 0)) { |
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
printf("pool_get: %s: items on itemlist, nitems %u\n", |
|
pp->pr_wchan, pp->pr_nitems); |
|
panic("pool_get: nitems inconsistent"); |
|
} |
|
#endif |
|
|
if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL) |
#ifdef POOL_DIAGNOSTIC |
panic("pool_get: %s: page empty", pp->pr_wchan); |
pr_log(pp, v, PRLOG_GET, file, line); |
|
#endif |
pr_log(pp, v, PRLOG_GET, file, line); |
|
|
|
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
if (pi->pi_magic != PI_MAGIC) { |
if (__predict_false(pi->pi_magic != PI_MAGIC)) { |
pr_printlog(pp); |
pr_printlog(pp, pi, printf); |
panic("pool_get(%s): free list modified: magic=%x; page %p;" |
panic("pool_get(%s): free list modified: " |
" item addr %p\n", |
"magic=%x; page %p; item addr %p\n", |
pp->pr_wchan, pi->pi_magic, ph->ph_page, pi); |
pp->pr_wchan, pi->pi_magic, ph->ph_page, pi); |
} |
} |
#endif |
#endif |
|
|
/* |
/* |
* Remove from item list. |
* Remove from item list. |
*/ |
*/ |
TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list); |
LIST_REMOVE(pi, pi_list); |
|
} |
|
pp->pr_nitems--; |
|
pp->pr_nout++; |
if (ph->ph_nmissing == 0) { |
if (ph->ph_nmissing == 0) { |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
if (pp->pr_nidle == 0) |
if (__predict_false(pp->pr_nidle == 0)) |
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 (ph->ph_nmissing == pp->pr_itemsperpage) { |
|
#ifdef DIAGNOSTIC |
|
if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 && |
|
!LIST_EMPTY(&ph->ph_itemlist))) { |
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
panic("pool_get: %s: nmissing inconsistent", |
|
pp->pr_wchan); |
|
} |
|
#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. |
|
* |
|
* First, move the now empty page to the head of |
|
* the page list. |
|
*/ |
*/ |
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_fullpages, ph, ph_pagelist); |
while ((ph = TAILQ_NEXT(ph, ph_pagelist)) != NULL) |
pool_update_curpage(pp); |
if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
break; |
|
|
|
pp->pr_curpage = ph; |
|
} |
} |
|
|
pp->pr_nget++; |
pp->pr_nget++; |
simple_unlock(&pp->pr_lock); |
pr_leave(pp); |
|
|
|
/* |
|
* If we have a low water mark and we are now below that low |
|
* water mark, add more items to the pool. |
|
*/ |
|
if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) { |
|
/* |
|
* XXX: Should we log a warning? Should we set up a timeout |
|
* to try again in a second or so? The latter could break |
|
* a caller's assumptions about interrupt protection, etc. |
|
*/ |
|
} |
|
|
|
simple_unlock(&pp->pr_slock); |
return (v); |
return (v); |
} |
} |
|
|
/* |
/* |
* Return resource to the pool; must be called at appropriate spl level |
* Internal version of pool_put(). Pool is already locked/entered. |
*/ |
*/ |
#ifdef POOL_DIAGNOSTIC |
static void |
void |
pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq) |
_pool_put(pp, v, file, line) |
|
struct pool *pp; |
|
void *v; |
|
const char *file; |
|
long line; |
|
#else |
|
void |
|
pool_put(pp, v) |
|
struct pool *pp; |
|
void *v; |
|
#endif |
|
{ |
{ |
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; |
|
|
page = (caddr_t)((u_long)v & pp->pr_pagemask); |
LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
|
SCHED_ASSERT_UNLOCKED(); |
|
|
simple_lock(&pp->pr_lock); |
page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask); |
|
|
pr_log(pp, v, PRLOG_PUT, file, line); |
#ifdef DIAGNOSTIC |
|
if (__predict_false(pp->pr_nout == 0)) { |
|
printf("pool %s: putting with none out\n", |
|
pp->pr_wchan); |
|
panic("pool_put"); |
|
} |
|
#endif |
|
|
if ((ph = pr_find_pagehead(pp, page)) == NULL) { |
if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) { |
pr_printlog(pp); |
pr_printlog(pp, NULL, printf); |
panic("pool_put: %s: page header missing", pp->pr_wchan); |
panic("pool_put: %s: page header missing", pp->pr_wchan); |
} |
} |
|
|
|
#ifdef LOCKDEBUG |
|
/* |
|
* Check if we're freeing a locked simple lock. |
|
*/ |
|
simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size); |
|
#endif |
|
|
/* |
/* |
* Return to item list. |
* Return to item list. |
*/ |
*/ |
|
if (pp->pr_roflags & PR_NOTOUCH) { |
|
pr_item_notouch_put(pp, ph, v); |
|
} else { |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
pi->pi_magic = PI_MAGIC; |
pi->pi_magic = PI_MAGIC; |
|
#endif |
|
#ifdef DEBUG |
|
{ |
|
int i, *ip = v; |
|
|
|
for (i = 0; i < pp->pr_size / sizeof(int); i++) { |
|
*ip++ = PI_MAGIC; |
|
} |
|
} |
#endif |
#endif |
TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list); |
|
|
LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list); |
|
} |
|
KDASSERT(ph->ph_nmissing != 0); |
ph->ph_nmissing--; |
ph->ph_nmissing--; |
pp->pr_nput++; |
pp->pr_nput++; |
|
pp->pr_nitems++; |
|
pp->pr_nout--; |
|
|
/* Cancel "pool empty" condition if it exists */ |
/* Cancel "pool empty" condition if it exists */ |
if (pp->pr_curpage == NULL) |
if (pp->pr_curpage == NULL) |
|
Line 1104 pool_put(pp, v) |
|
|
|
if (pp->pr_flags & PR_WANTED) { |
if (pp->pr_flags & PR_WANTED) { |
pp->pr_flags &= ~PR_WANTED; |
pp->pr_flags &= ~PR_WANTED; |
|
if (ph->ph_nmissing == 0) |
|
pp->pr_nidle++; |
wakeup((caddr_t)pp); |
wakeup((caddr_t)pp); |
simple_unlock(&pp->pr_lock); |
|
return; |
return; |
} |
} |
|
|
/* |
/* |
* If this page is now complete, move it to the end of the pagelist. |
* If this page is now empty, do one of two things: |
* If this page has just become un-empty, move it the head. |
* |
*/ |
* (1) If we have more pages than the page high water mark, |
if (ph->ph_nmissing == 0) { |
* free the page back to the system. ONLY CONSIDER |
|
* FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE |
|
* CLAIM. |
|
* |
|
* (2) Otherwise, move the page to the empty page list. |
|
* |
|
* Either way, select a new current page (so we use a partially-full |
|
* page if one is available). |
|
*/ |
|
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 && |
#if 0 |
(pp->pr_npages > pp->pr_maxpages || |
timeout(pool_drain, 0, pool_inactive_time*hz); |
(pp->pr_alloc->pa_flags & PA_WANT) != 0)) { |
#else |
pr_rmpage(pp, ph, pq); |
pr_rmpage(pp, ph); |
|
#endif |
|
} 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); |
ph->ph_time = time; |
|
|
|
/* XXX - update curpage */ |
|
for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; |
|
ph = TAILQ_NEXT(ph, ph_pagelist)) |
|
if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) |
|
break; |
|
|
|
pp->pr_curpage = ph; |
/* |
|
* Update the timestamp on the page. A page must |
|
* be idle for some period of time before it can |
|
* be reclaimed by the pagedaemon. This minimizes |
|
* ping-pong'ing for memory. |
|
*/ |
|
getmicrotime(&ph->ph_time); |
} |
} |
|
pool_update_curpage(pp); |
} |
} |
|
|
simple_unlock(&pp->pr_lock); |
/* |
|
* If the page was previously completely full, move it to the |
|
* partially-full list and make it the current page. The next |
|
* allocation will get the item from this page, instead of |
|
* further fragmenting the pool. |
|
*/ |
|
else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) { |
|
LIST_REMOVE(ph, ph_pagelist); |
|
LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist); |
|
pp->pr_curpage = ph; |
|
} |
} |
} |
|
|
/* |
/* |
* Add N items to the pool. |
* Return resource to the pool; must be called at appropriate spl level |
*/ |
*/ |
int |
#ifdef POOL_DIAGNOSTIC |
pool_prime(pp, n, storage) |
void |
struct pool *pp; |
_pool_put(struct pool *pp, void *v, const char *file, long line) |
int n; |
|
caddr_t storage; |
|
{ |
{ |
caddr_t cp; |
struct pool_pagelist pq; |
int newnitems, newpages; |
|
|
|
#ifdef DIAGNOSTIC |
LIST_INIT(&pq); |
if (storage && !(pp->pr_flags & PR_STATIC)) |
|
panic("pool_prime: static"); |
simple_lock(&pp->pr_slock); |
/* !storage && static caught below */ |
pr_enter(pp, file, line); |
|
|
|
pr_log(pp, v, PRLOG_PUT, file, line); |
|
|
|
pool_do_put(pp, v, &pq); |
|
|
|
pr_leave(pp); |
|
simple_unlock(&pp->pr_slock); |
|
|
|
pr_pagelist_free(pp, &pq); |
|
} |
|
#undef pool_put |
|
#endif /* POOL_DIAGNOSTIC */ |
|
|
|
void |
|
pool_put(struct pool *pp, void *v) |
|
{ |
|
struct pool_pagelist pq; |
|
|
|
LIST_INIT(&pq); |
|
|
|
simple_lock(&pp->pr_slock); |
|
pool_do_put(pp, v, &pq); |
|
simple_unlock(&pp->pr_slock); |
|
|
|
pr_pagelist_free(pp, &pq); |
|
} |
|
|
|
#ifdef POOL_DIAGNOSTIC |
|
#define pool_put(h, v) _pool_put((h), (v), __FILE__, __LINE__) |
#endif |
#endif |
|
|
newnitems = pp->pr_minitems + n; |
/* |
newpages = |
* pool_grow: grow a pool by a page. |
roundup(pp->pr_itemsperpage,newnitems) / pp->pr_itemsperpage |
* |
- pp->pr_minpages; |
* => called with pool locked. |
|
* => unlock and relock the pool. |
|
* => return with pool locked. |
|
*/ |
|
|
simple_lock(&pp->pr_lock); |
static int |
while (newpages-- > 0) { |
pool_grow(struct pool *pp, int flags) |
|
{ |
|
struct pool_item_header *ph = NULL; |
|
char *cp; |
|
|
if (pp->pr_flags & PR_STATIC) { |
simple_unlock(&pp->pr_slock); |
cp = storage; |
cp = pool_allocator_alloc(pp, flags); |
storage += pp->pr_pagesz; |
if (__predict_true(cp != NULL)) { |
} else { |
ph = pool_alloc_item_header(pp, cp, flags); |
cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype); |
} |
|
if (__predict_false(cp == NULL || ph == NULL)) { |
|
if (cp != NULL) { |
|
pool_allocator_free(pp, cp); |
} |
} |
|
simple_lock(&pp->pr_slock); |
|
return ENOMEM; |
|
} |
|
|
if (cp == NULL) { |
simple_lock(&pp->pr_slock); |
simple_unlock(&pp->pr_lock); |
pool_prime_page(pp, cp, ph); |
return (ENOMEM); |
pp->pr_npagealloc++; |
} |
return 0; |
|
} |
|
|
pool_prime_page(pp, cp); |
/* |
|
* Add N items to the pool. |
|
*/ |
|
int |
|
pool_prime(struct pool *pp, int n) |
|
{ |
|
int newpages; |
|
int error = 0; |
|
|
|
simple_lock(&pp->pr_slock); |
|
|
|
newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage; |
|
|
|
while (newpages-- > 0) { |
|
error = pool_grow(pp, PR_NOWAIT); |
|
if (error) { |
|
break; |
|
} |
pp->pr_minpages++; |
pp->pr_minpages++; |
} |
} |
|
|
pp->pr_minitems = newnitems; |
|
|
|
if (pp->pr_minpages >= pp->pr_maxpages) |
if (pp->pr_minpages >= pp->pr_maxpages) |
pp->pr_maxpages = pp->pr_minpages + 1; /* XXX */ |
pp->pr_maxpages = pp->pr_minpages + 1; /* XXX */ |
|
|
simple_unlock(&pp->pr_lock); |
simple_unlock(&pp->pr_slock); |
return (0); |
return error; |
} |
} |
|
|
/* |
/* |
* Add a page worth of items to the pool. |
* Add a page worth of items to the pool. |
|
* |
|
* Note, we must be called with the pool descriptor LOCKED. |
*/ |
*/ |
int |
static void |
pool_prime_page(pp, storage) |
pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph) |
struct pool *pp; |
|
caddr_t storage; |
|
{ |
{ |
struct pool_item *pi; |
struct pool_item *pi; |
struct pool_item_header *ph; |
|
caddr_t cp = storage; |
caddr_t cp = storage; |
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; |
|
|
if ((pp->pr_flags & PR_PHINPAGE) != 0) { |
LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); |
ph = (struct pool_item_header *)(cp + pp->pr_phoffset); |
|
} else { |
#ifdef DIAGNOSTIC |
ph = pool_get(&phpool, PR_URGENT); |
if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0) |
LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)], |
panic("pool_prime_page: %s: unaligned page", pp->pr_wchan); |
ph, ph_hashlist); |
#endif |
} |
|
|
|
/* |
/* |
* 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); |
LIST_INIT(&ph->ph_itemlist); |
ph->ph_page = storage; |
ph->ph_page = storage; |
ph->ph_nmissing = 0; |
ph->ph_nmissing = 0; |
ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0; |
getmicrotime(&ph->ph_time); |
|
if ((pp->pr_roflags & PR_PHINPAGE) == 0) |
|
SPLAY_INSERT(phtree, &pp->pr_phtree, ph); |
|
|
pp->pr_nidle++; |
pp->pr_nidle++; |
|
|
Line 765 pool_prime_page(pp, storage) |
|
Line 1313 pool_prime_page(pp, storage) |
|
* Insert remaining chunks on the bucket list. |
* Insert remaining chunks on the bucket list. |
*/ |
*/ |
n = pp->pr_itemsperpage; |
n = pp->pr_itemsperpage; |
|
pp->pr_nitems += n; |
|
|
while (n--) { |
if (pp->pr_roflags & PR_NOTOUCH) { |
pi = (struct pool_item *)cp; |
pool_item_freelist_t *freelist = PR_FREELIST(ph); |
|
int i; |
|
|
|
ph->ph_off = cp - storage; |
|
ph->ph_firstfree = 0; |
|
for (i = 0; i < n - 1; i++) |
|
freelist[i] = i + 1; |
|
freelist[n - 1] = PR_INDEX_EOL; |
|
} else { |
|
while (n--) { |
|
pi = (struct pool_item *)cp; |
|
|
|
KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0); |
|
|
/* Insert on page list */ |
/* Insert on page list */ |
TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list); |
LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list); |
#ifdef DIAGNOSTIC |
#ifdef DIAGNOSTIC |
pi->pi_magic = PI_MAGIC; |
pi->pi_magic = PI_MAGIC; |
#endif |
#endif |
cp = (caddr_t)(cp + pp->pr_size); |
cp = (caddr_t)(cp + pp->pr_size); |
|
} |
} |
} |
|
|
/* |
/* |
Line 785 pool_prime_page(pp, storage) |
|
Line 1347 pool_prime_page(pp, storage) |
|
|
|
if (++pp->pr_npages > pp->pr_hiwat) |
if (++pp->pr_npages > pp->pr_hiwat) |
pp->pr_hiwat = pp->pr_npages; |
pp->pr_hiwat = pp->pr_npages; |
|
} |
|
|
return (0); |
/* |
|
* Used by pool_get() when nitems drops below the low water mark. This |
|
* 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 2, we must be called with the pool already locked, and we return |
|
* with it locked. |
|
*/ |
|
static int |
|
pool_catchup(struct pool *pp) |
|
{ |
|
int error = 0; |
|
|
|
while (POOL_NEEDS_CATCHUP(pp)) { |
|
error = pool_grow(pp, PR_NOWAIT); |
|
if (error) { |
|
break; |
|
} |
|
} |
|
return error; |
} |
} |
|
|
void |
static void |
pool_setlowat(pp, n) |
pool_update_curpage(struct pool *pp) |
pool_handle_t pp; |
|
int n; |
|
{ |
{ |
pp->pr_minitems = n; |
|
if (n == 0) { |
pp->pr_curpage = LIST_FIRST(&pp->pr_partpages); |
pp->pr_minpages = 0; |
if (pp->pr_curpage == NULL) { |
return; |
pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages); |
} |
} |
pp->pr_minpages = |
|
roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage; |
|
} |
} |
|
|
void |
void |
pool_sethiwat(pp, n) |
pool_setlowat(struct pool *pp, int n) |
pool_handle_t pp; |
|
int n; |
|
{ |
{ |
if (n == 0) { |
|
pp->pr_maxpages = 0; |
simple_lock(&pp->pr_slock); |
return; |
|
|
pp->pr_minitems = n; |
|
pp->pr_minpages = (n == 0) |
|
? 0 |
|
: roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage; |
|
|
|
/* Make sure we're caught up with the newly-set low water mark. */ |
|
if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) { |
|
/* |
|
* XXX: Should we log a warning? Should we set up a timeout |
|
* to try again in a second or so? The latter could break |
|
* a caller's assumptions about interrupt protection, etc. |
|
*/ |
} |
} |
pp->pr_maxpages = |
|
roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage; |
|
} |
|
|
|
|
simple_unlock(&pp->pr_slock); |
|
} |
|
|
/* |
void |
* Default page allocator. |
pool_sethiwat(struct pool *pp, int n) |
*/ |
|
static void * |
|
pool_page_alloc(sz, flags, mtype) |
|
unsigned long sz; |
|
int flags; |
|
int mtype; |
|
{ |
{ |
|
|
#if defined(UVM) |
simple_lock(&pp->pr_slock); |
return ((void *)uvm_km_alloc_poolpage()); |
|
#else |
pp->pr_maxpages = (n == 0) |
return ((void *)kmem_alloc_poolpage()); |
? 0 |
#endif |
: roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage; |
|
|
|
simple_unlock(&pp->pr_slock); |
} |
} |
|
|
static void |
void |
pool_page_free(v, sz, mtype) |
pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap) |
void *v; |
|
unsigned long sz; |
|
int mtype; |
|
{ |
{ |
|
|
#if defined(UVM) |
simple_lock(&pp->pr_slock); |
uvm_km_free_poolpage((vm_offset_t)v); |
|
#else |
pp->pr_hardlimit = n; |
kmem_free_poolpage((vm_offset_t)v); |
pp->pr_hardlimit_warning = warnmess; |
#endif |
pp->pr_hardlimit_ratecap.tv_sec = ratecap; |
|
pp->pr_hardlimit_warning_last.tv_sec = 0; |
|
pp->pr_hardlimit_warning_last.tv_usec = 0; |
|
|
|
/* |
|
* In-line version of pool_sethiwat(), because we don't want to |
|
* release the lock. |
|
*/ |
|
pp->pr_maxpages = (n == 0) |
|
? 0 |
|
: roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage; |
|
|
|
simple_unlock(&pp->pr_slock); |
} |
} |
|
|
/* |
/* |
* Release all complete pages that have not been used recently. |
* Release all complete pages that have not been used recently. |
*/ |
*/ |
void |
int |
pool_reclaim (pp) |
#ifdef POOL_DIAGNOSTIC |
pool_handle_t pp; |
_pool_reclaim(struct pool *pp, const char *file, long line) |
|
#else |
|
pool_reclaim(struct pool *pp) |
|
#endif |
{ |
{ |
struct pool_item_header *ph, *phnext; |
struct pool_item_header *ph, *phnext; |
struct timeval curtime = time; |
struct pool_cache *pc; |
|
struct pool_pagelist pq; |
|
struct pool_cache_grouplist pcgl; |
|
struct timeval curtime, diff; |
|
|
if (pp->pr_flags & PR_STATIC) |
if (pp->pr_drain_hook != NULL) { |
return; |
/* |
|
* The drain hook must be called with the pool unlocked. |
|
*/ |
|
(*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT); |
|
} |
|
|
if (simple_lock_try(&pp->pr_lock) == 0) |
if (simple_lock_try(&pp->pr_slock) == 0) |
return; |
return (0); |
|
pr_enter(pp, file, line); |
|
|
for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) { |
LIST_INIT(&pq); |
phnext = TAILQ_NEXT(ph, ph_pagelist); |
LIST_INIT(&pcgl); |
|
|
|
/* |
|
* Reclaim items from the pool's caches. |
|
*/ |
|
LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) |
|
pool_cache_reclaim(pc, &pq, &pcgl); |
|
|
|
getmicrotime(&curtime); |
|
|
|
for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) { |
|
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; |
|
pr_rmpage(pp, ph); |
/* |
} |
* If freeing this page would put us below |
|
* the low water mark, stop now. |
|
*/ |
|
if ((pp->pr_nitems - pp->pr_itemsperpage) < |
|
pp->pr_minitems) |
|
break; |
|
|
|
pr_rmpage(pp, ph, &pq); |
} |
} |
|
|
simple_unlock(&pp->pr_lock); |
pr_leave(pp); |
} |
simple_unlock(&pp->pr_slock); |
|
if (LIST_EMPTY(&pq) && LIST_EMPTY(&pcgl)) |
|
return 0; |
|
|
|
pr_pagelist_free(pp, &pq); |
|
pcg_grouplist_free(&pcgl); |
|
return (1); |
|
} |
|
|
/* |
/* |
* Drain pools, one at a time. |
* Drain pools, one at a time. |
|
* |
|
* Note, we must never be called from an interrupt context. |
*/ |
*/ |
void |
void |
pool_drain(arg) |
pool_drain(void *arg) |
void *arg; |
|
{ |
{ |
struct pool *pp; |
struct pool *pp; |
int s = splimp(); |
int s; |
|
|
/* XXX:lock pool head */ |
|
if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL) { |
|
splx(s); |
|
return; |
|
} |
|
|
|
pp = drainpp; |
|
drainpp = TAILQ_NEXT(pp, pr_poollist); |
|
/* XXX:unlock pool head */ |
|
|
|
pool_reclaim(pp); |
pp = NULL; |
|
s = splvm(); |
|
simple_lock(&pool_head_slock); |
|
if (drainpp == NULL) { |
|
drainpp = LIST_FIRST(&pool_head); |
|
} |
|
if (drainpp) { |
|
pp = drainpp; |
|
drainpp = LIST_NEXT(pp, pr_poollist); |
|
} |
|
simple_unlock(&pool_head_slock); |
|
if (pp) |
|
pool_reclaim(pp); |
splx(s); |
splx(s); |
} |
} |
|
|
|
|
#ifdef DEBUG |
|
/* |
/* |
* Diagnostic helpers. |
* Diagnostic helpers. |
*/ |
*/ |
void |
void |
pool_print(pp, label) |
pool_print(struct pool *pp, const char *modif) |
struct pool *pp; |
|
char *label; |
|
{ |
{ |
|
int s; |
|
|
if (label != NULL) |
s = splvm(); |
printf("%s: ", label); |
if (simple_lock_try(&pp->pr_slock) == 0) { |
|
printf("pool %s is locked; try again later\n", |
printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n" |
pp->pr_wchan); |
" npages %u minitems %u itemsperpage %u itemoffset %u\n" |
splx(s); |
" nidle %lu\n", |
return; |
pp->pr_wchan, |
} |
pp->pr_nget, |
pool_print1(pp, modif, printf); |
pp->pr_nput, |
simple_unlock(&pp->pr_slock); |
pp->pr_npagealloc, |
splx(s); |
pp->pr_npagefree, |
|
pp->pr_npages, |
|
pp->pr_minitems, |
|
pp->pr_itemsperpage, |
|
pp->pr_itemoffset, |
|
pp->pr_nidle); |
|
} |
} |
|
|
int |
void |
pool_chk(pp, label) |
pool_printall(const char *modif, void (*pr)(const char *, ...)) |
|
{ |
struct pool *pp; |
struct pool *pp; |
char *label; |
|
|
if (simple_lock_try(&pool_head_slock) == 0) { |
|
(*pr)("WARNING: pool_head_slock is locked\n"); |
|
} else { |
|
simple_unlock(&pool_head_slock); |
|
} |
|
|
|
LIST_FOREACH(pp, &pool_head, pr_poollist) { |
|
pool_printit(pp, modif, pr); |
|
} |
|
} |
|
|
|
void |
|
pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...)) |
{ |
{ |
struct pool_item_header *ph; |
|
int r = 0; |
|
|
|
simple_lock(&pp->pr_lock); |
if (pp == NULL) { |
|
(*pr)("Must specify a pool to print.\n"); |
|
return; |
|
} |
|
|
for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; |
/* |
ph = TAILQ_NEXT(ph, ph_pagelist)) { |
* Called from DDB; interrupts should be blocked, and all |
|
* other processors should be paused. We can skip locking |
|
* the pool in this case. |
|
* |
|
* We do a simple_lock_try() just to print the lock |
|
* status, however. |
|
*/ |
|
|
struct pool_item *pi; |
if (simple_lock_try(&pp->pr_slock) == 0) |
int n; |
(*pr)("WARNING: pool %s is locked\n", pp->pr_wchan); |
caddr_t page; |
else |
|
simple_unlock(&pp->pr_slock); |
|
|
page = (caddr_t)((u_long)ph & pp->pr_pagemask); |
pool_print1(pp, modif, pr); |
if (page != ph->ph_page) { |
} |
if (label != NULL) |
|
printf("%s: ", label); |
|
printf("pool(%s): page inconsistency: page %p;" |
|
" at page head addr %p (p %p)\n", |
|
pp->pr_wchan, ph->ph_page, |
|
ph, page); |
|
r++; |
|
goto out; |
|
} |
|
|
|
for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0; |
static void |
pi != NULL; |
pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl, |
pi = TAILQ_NEXT(pi,pi_list), n++) { |
void (*pr)(const char *, ...)) |
|
{ |
|
struct pool_item_header *ph; |
|
#ifdef DIAGNOSTIC |
|
struct pool_item *pi; |
|
#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 |
#ifdef DIAGNOSTIC |
if (pi->pi_magic != PI_MAGIC) { |
if (!(pp->pr_roflags & PR_NOTOUCH)) { |
if (label != NULL) |
LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) { |
printf("%s: ", label); |
if (pi->pi_magic != PI_MAGIC) { |
printf("pool(%s): free list modified: magic=%x;" |
(*pr)("\t\t\titem %p, magic 0x%x\n", |
" page %p; item ordinal %d;" |
pi, pi->pi_magic); |
" addr %p (p %p)\n", |
} |
pp->pr_wchan, pi->pi_magic, ph->ph_page, |
|
n, pi, page); |
|
panic("pool"); |
|
} |
} |
|
} |
#endif |
#endif |
page = (caddr_t)((u_long)pi & pp->pr_pagemask); |
} |
if (page == ph->ph_page) |
} |
continue; |
|
|
|
|
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; |
|
char c; |
|
|
|
while ((c = *modif++) != '\0') { |
|
if (c == 'l') |
|
print_log = 1; |
|
if (c == 'p') |
|
print_pagelist = 1; |
|
if (c == 'c') |
|
print_cache = 1; |
|
} |
|
|
|
(*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n", |
|
pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset, |
|
pp->pr_roflags); |
|
(*pr)("\talloc %p\n", pp->pr_alloc); |
|
(*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n", |
|
pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages); |
|
(*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n", |
|
pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit); |
|
|
|
(*pr)("\n\tnget %lu, nfail %lu, nput %lu\n", |
|
pp->pr_nget, pp->pr_nfail, pp->pr_nput); |
|
(*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n", |
|
pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle); |
|
|
|
if (print_pagelist == 0) |
|
goto skip_pagelist; |
|
|
|
if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) |
|
(*pr)("\n\tempty page list:\n"); |
|
pool_print_pagelist(pp, &pp->pr_emptypages, pr); |
|
if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL) |
|
(*pr)("\n\tfull page list:\n"); |
|
pool_print_pagelist(pp, &pp->pr_fullpages, pr); |
|
if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL) |
|
(*pr)("\n\tpartial-page list:\n"); |
|
pool_print_pagelist(pp, &pp->pr_partpages, pr); |
|
|
|
if (pp->pr_curpage == NULL) |
|
(*pr)("\tno current page\n"); |
|
else |
|
(*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page); |
|
|
|
skip_pagelist: |
|
if (print_log == 0) |
|
goto skip_log; |
|
|
|
(*pr)("\n"); |
|
if ((pp->pr_roflags & PR_LOGGING) == 0) |
|
(*pr)("\tno log\n"); |
|
else |
|
pr_printlog(pp, NULL, pr); |
|
|
|
skip_log: |
|
if (print_cache == 0) |
|
goto skip_cache; |
|
|
|
#define PR_GROUPLIST(pcg) \ |
|
(*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail); \ |
|
for (i = 0; i < PCG_NOBJECTS; i++) { \ |
|
if (pcg->pcg_objects[i].pcgo_pa != \ |
|
POOL_PADDR_INVALID) { \ |
|
(*pr)("\t\t\t%p, 0x%llx\n", \ |
|
pcg->pcg_objects[i].pcgo_va, \ |
|
(unsigned long long) \ |
|
pcg->pcg_objects[i].pcgo_pa); \ |
|
} else { \ |
|
(*pr)("\t\t\t%p\n", \ |
|
pcg->pcg_objects[i].pcgo_va); \ |
|
} \ |
|
} |
|
|
|
LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) { |
|
(*pr)("\tcache %p\n", pc); |
|
(*pr)("\t hits %lu misses %lu ngroups %lu nitems %lu\n", |
|
pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems); |
|
(*pr)("\t full groups:\n"); |
|
LIST_FOREACH(pcg, &pc->pc_fullgroups, pcg_list) { |
|
PR_GROUPLIST(pcg); |
|
} |
|
(*pr)("\t partial groups:\n"); |
|
LIST_FOREACH(pcg, &pc->pc_partgroups, pcg_list) { |
|
PR_GROUPLIST(pcg); |
|
} |
|
(*pr)("\t empty groups:\n"); |
|
LIST_FOREACH(pcg, &pc->pc_emptygroups, pcg_list) { |
|
PR_GROUPLIST(pcg); |
|
} |
|
} |
|
#undef PR_GROUPLIST |
|
|
|
skip_cache: |
|
pr_enter_check(pp, pr); |
|
} |
|
|
|
static int |
|
pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph) |
|
{ |
|
struct pool_item *pi; |
|
caddr_t page; |
|
int n; |
|
|
|
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; |
|
} |
|
|
|
if ((pp->pr_roflags & PR_NOTOUCH) != 0) |
|
return 0; |
|
|
|
for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0; |
|
pi != NULL; |
|
pi = LIST_NEXT(pi,pi_list), n++) { |
|
|
|
#ifdef DIAGNOSTIC |
|
if (pi->pi_magic != PI_MAGIC) { |
if (label != NULL) |
if (label != NULL) |
printf("%s: ", label); |
printf("%s: ", label); |
printf("pool(%s): page inconsistency: page %p;" |
printf("pool(%s): free list modified: magic=%x;" |
" item ordinal %d; addr %p (p %p)\n", |
" page %p; item ordinal %d;" |
pp->pr_wchan, ph->ph_page, |
" addr %p (p %p)\n", |
|
pp->pr_wchan, pi->pi_magic, ph->ph_page, |
n, pi, page); |
n, pi, page); |
r++; |
panic("pool"); |
|
} |
|
#endif |
|
page = |
|
(caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask); |
|
if (page == ph->ph_page) |
|
continue; |
|
|
|
if (label != NULL) |
|
printf("%s: ", label); |
|
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; |
|
} |
|
|
|
|
|
int |
|
pool_chk(struct pool *pp, const char *label) |
|
{ |
|
struct pool_item_header *ph; |
|
int r = 0; |
|
|
|
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_lock); |
simple_unlock(&pp->pr_slock); |
return (r); |
return (r); |
} |
} |
|
|
|
/* |
|
* pool_cache_init: |
|
* |
|
* Initialize a pool cache. |
|
* |
|
* NOTE: If the pool must be protected from interrupts, we expect |
|
* to be called at the appropriate interrupt priority level. |
|
*/ |
|
void |
|
pool_cache_init(struct pool_cache *pc, struct pool *pp, |
|
int (*ctor)(void *, void *, int), |
|
void (*dtor)(void *, void *), |
|
void *arg) |
|
{ |
|
|
|
LIST_INIT(&pc->pc_emptygroups); |
|
LIST_INIT(&pc->pc_fullgroups); |
|
LIST_INIT(&pc->pc_partgroups); |
|
simple_lock_init(&pc->pc_slock); |
|
|
|
pc->pc_pool = pp; |
|
|
|
pc->pc_ctor = ctor; |
|
pc->pc_dtor = dtor; |
|
pc->pc_arg = arg; |
|
|
|
pc->pc_hits = 0; |
|
pc->pc_misses = 0; |
|
|
|
pc->pc_ngroups = 0; |
|
|
|
pc->pc_nitems = 0; |
|
|
|
simple_lock(&pp->pr_slock); |
|
LIST_INSERT_HEAD(&pp->pr_cachelist, pc, pc_poollist); |
|
simple_unlock(&pp->pr_slock); |
|
} |
|
|
|
/* |
|
* pool_cache_destroy: |
|
* |
|
* Destroy a pool cache. |
|
*/ |
|
void |
|
pool_cache_destroy(struct pool_cache *pc) |
|
{ |
|
struct pool *pp = pc->pc_pool; |
|
|
|
/* First, invalidate the entire cache. */ |
|
pool_cache_invalidate(pc); |
|
|
|
/* ...and remove it from the pool's cache list. */ |
|
simple_lock(&pp->pr_slock); |
|
LIST_REMOVE(pc, pc_poollist); |
|
simple_unlock(&pp->pr_slock); |
|
} |
|
|
|
static inline void * |
|
pcg_get(struct pool_cache_group *pcg, paddr_t *pap) |
|
{ |
|
void *object; |
|
u_int idx; |
|
|
|
KASSERT(pcg->pcg_avail <= PCG_NOBJECTS); |
|
KASSERT(pcg->pcg_avail != 0); |
|
idx = --pcg->pcg_avail; |
|
|
|
KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL); |
|
object = pcg->pcg_objects[idx].pcgo_va; |
|
if (pap != NULL) |
|
*pap = pcg->pcg_objects[idx].pcgo_pa; |
|
pcg->pcg_objects[idx].pcgo_va = NULL; |
|
|
|
return (object); |
|
} |
|
|
|
static inline void |
|
pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa) |
|
{ |
|
u_int idx; |
|
|
|
KASSERT(pcg->pcg_avail < PCG_NOBJECTS); |
|
idx = pcg->pcg_avail++; |
|
|
|
KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL); |
|
pcg->pcg_objects[idx].pcgo_va = object; |
|
pcg->pcg_objects[idx].pcgo_pa = pa; |
|
} |
|
|
|
static void |
|
pcg_grouplist_free(struct pool_cache_grouplist *pcgl) |
|
{ |
|
struct pool_cache_group *pcg; |
|
int s; |
|
|
|
s = splvm(); |
|
while ((pcg = LIST_FIRST(pcgl)) != NULL) { |
|
LIST_REMOVE(pcg, pcg_list); |
|
pool_put(&pcgpool, pcg); |
|
} |
|
splx(s); |
|
} |
|
|
|
/* |
|
* pool_cache_get{,_paddr}: |
|
* |
|
* Get an object from a pool cache (optionally returning |
|
* the physical address of the object). |
|
*/ |
|
void * |
|
pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap) |
|
{ |
|
struct pool_cache_group *pcg; |
|
void *object; |
|
|
|
#ifdef LOCKDEBUG |
|
if (flags & PR_WAITOK) |
|
simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)"); |
|
#endif |
|
|
|
simple_lock(&pc->pc_slock); |
|
|
|
pcg = LIST_FIRST(&pc->pc_partgroups); |
|
if (pcg == NULL) { |
|
pcg = LIST_FIRST(&pc->pc_fullgroups); |
|
if (pcg != NULL) { |
|
LIST_REMOVE(pcg, pcg_list); |
|
LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list); |
|
} |
|
} |
|
if (pcg == NULL) { |
|
|
|
/* |
|
* No groups with any available objects. Allocate |
|
* a new object, construct it, and return it to |
|
* the caller. We will allocate a group, if necessary, |
|
* when the object is freed back to the cache. |
|
*/ |
|
pc->pc_misses++; |
|
simple_unlock(&pc->pc_slock); |
|
object = pool_get(pc->pc_pool, flags); |
|
if (object != NULL && pc->pc_ctor != NULL) { |
|
if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) { |
|
pool_put(pc->pc_pool, object); |
|
return (NULL); |
|
} |
|
} |
|
if (object != NULL && pap != NULL) { |
|
#ifdef POOL_VTOPHYS |
|
*pap = POOL_VTOPHYS(object); |
|
#else |
|
*pap = POOL_PADDR_INVALID; |
|
#endif |
|
} |
|
return (object); |
|
} |
|
|
|
pc->pc_hits++; |
|
pc->pc_nitems--; |
|
object = pcg_get(pcg, pap); |
|
|
|
if (pcg->pcg_avail == 0) { |
|
LIST_REMOVE(pcg, pcg_list); |
|
LIST_INSERT_HEAD(&pc->pc_emptygroups, pcg, pcg_list); |
|
} |
|
simple_unlock(&pc->pc_slock); |
|
|
|
return (object); |
|
} |
|
|
|
/* |
|
* pool_cache_put{,_paddr}: |
|
* |
|
* Put an object back to the pool cache (optionally caching the |
|
* physical address of the object). |
|
*/ |
|
void |
|
pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa) |
|
{ |
|
struct pool_cache_group *pcg; |
|
int s; |
|
|
|
if (__predict_false((pc->pc_pool->pr_flags & PR_WANTED) != 0)) { |
|
goto destruct; |
|
} |
|
|
|
simple_lock(&pc->pc_slock); |
|
|
|
pcg = LIST_FIRST(&pc->pc_partgroups); |
|
if (pcg == NULL) { |
|
pcg = LIST_FIRST(&pc->pc_emptygroups); |
|
if (pcg != NULL) { |
|
LIST_REMOVE(pcg, pcg_list); |
|
LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list); |
|
} |
|
} |
|
if (pcg == NULL) { |
|
|
|
/* |
|
* No empty groups to free the object to. Attempt to |
|
* allocate one. |
|
*/ |
|
simple_unlock(&pc->pc_slock); |
|
s = splvm(); |
|
pcg = pool_get(&pcgpool, PR_NOWAIT); |
|
splx(s); |
|
if (pcg == NULL) { |
|
destruct: |
|
|
|
/* |
|
* Unable to allocate a cache group; destruct the object |
|
* and free it back to the pool. |
|
*/ |
|
pool_cache_destruct_object(pc, object); |
|
return; |
|
} |
|
memset(pcg, 0, sizeof(*pcg)); |
|
simple_lock(&pc->pc_slock); |
|
pc->pc_ngroups++; |
|
LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list); |
|
} |
|
|
|
pc->pc_nitems++; |
|
pcg_put(pcg, object, pa); |
|
|
|
if (pcg->pcg_avail == PCG_NOBJECTS) { |
|
LIST_REMOVE(pcg, pcg_list); |
|
LIST_INSERT_HEAD(&pc->pc_fullgroups, pcg, pcg_list); |
|
} |
|
simple_unlock(&pc->pc_slock); |
|
} |
|
|
|
/* |
|
* pool_cache_destruct_object: |
|
* |
|
* Force destruction of an object and its release back into |
|
* the pool. |
|
*/ |
|
void |
|
pool_cache_destruct_object(struct pool_cache *pc, void *object) |
|
{ |
|
|
|
if (pc->pc_dtor != NULL) |
|
(*pc->pc_dtor)(pc->pc_arg, object); |
|
pool_put(pc->pc_pool, object); |
|
} |
|
|
|
static void |
|
pool_do_cache_invalidate_grouplist(struct pool_cache_grouplist *pcgsl, |
|
struct pool_cache *pc, struct pool_pagelist *pq, |
|
struct pool_cache_grouplist *pcgdl) |
|
{ |
|
struct pool_cache_group *pcg, *npcg; |
|
void *object; |
|
|
|
for (pcg = LIST_FIRST(pcgsl); pcg != NULL; pcg = npcg) { |
|
npcg = LIST_NEXT(pcg, pcg_list); |
|
while (pcg->pcg_avail != 0) { |
|
pc->pc_nitems--; |
|
object = pcg_get(pcg, NULL); |
|
if (pc->pc_dtor != NULL) |
|
(*pc->pc_dtor)(pc->pc_arg, object); |
|
pool_do_put(pc->pc_pool, object, pq); |
|
} |
|
pc->pc_ngroups--; |
|
LIST_REMOVE(pcg, pcg_list); |
|
LIST_INSERT_HEAD(pcgdl, pcg, pcg_list); |
|
} |
|
} |
|
|
|
static void |
|
pool_do_cache_invalidate(struct pool_cache *pc, struct pool_pagelist *pq, |
|
struct pool_cache_grouplist *pcgl) |
|
{ |
|
|
|
LOCK_ASSERT(simple_lock_held(&pc->pc_slock)); |
|
LOCK_ASSERT(simple_lock_held(&pc->pc_pool->pr_slock)); |
|
|
|
pool_do_cache_invalidate_grouplist(&pc->pc_fullgroups, pc, pq, pcgl); |
|
pool_do_cache_invalidate_grouplist(&pc->pc_partgroups, pc, pq, pcgl); |
|
|
|
KASSERT(LIST_EMPTY(&pc->pc_partgroups)); |
|
KASSERT(LIST_EMPTY(&pc->pc_fullgroups)); |
|
KASSERT(pc->pc_nitems == 0); |
|
} |
|
|
|
/* |
|
* pool_cache_invalidate: |
|
* |
|
* Invalidate a pool cache (destruct and release all of the |
|
* cached objects). |
|
*/ |
|
void |
|
pool_cache_invalidate(struct pool_cache *pc) |
|
{ |
|
struct pool_pagelist pq; |
|
struct pool_cache_grouplist pcgl; |
|
|
|
LIST_INIT(&pq); |
|
LIST_INIT(&pcgl); |
|
|
|
simple_lock(&pc->pc_slock); |
|
simple_lock(&pc->pc_pool->pr_slock); |
|
|
|
pool_do_cache_invalidate(pc, &pq, &pcgl); |
|
|
|
simple_unlock(&pc->pc_pool->pr_slock); |
|
simple_unlock(&pc->pc_slock); |
|
|
|
pr_pagelist_free(pc->pc_pool, &pq); |
|
pcg_grouplist_free(&pcgl); |
|
} |
|
|
|
/* |
|
* pool_cache_reclaim: |
|
* |
|
* Reclaim a pool cache for pool_reclaim(). |
|
*/ |
|
static void |
|
pool_cache_reclaim(struct pool_cache *pc, struct pool_pagelist *pq, |
|
struct pool_cache_grouplist *pcgl) |
|
{ |
|
|
|
/* |
|
* We're locking in the wrong order (normally pool_cache -> pool, |
|
* but the pool is already locked when we get here), so we have |
|
* to use trylock. If we can't lock the pool_cache, it's not really |
|
* a big deal here. |
|
*/ |
|
if (simple_lock_try(&pc->pc_slock) == 0) |
|
return; |
|
|
|
pool_do_cache_invalidate(pc, pq, pcgl); |
|
|
|
simple_unlock(&pc->pc_slock); |
|
} |
|
|
|
/* |
|
* Pool backend allocators. |
|
* |
|
* Each pool has a backend allocator that handles allocation, deallocation, |
|
* and any additional draining that might be needed. |
|
* |
|
* We provide two standard allocators: |
|
* |
|
* pool_allocator_kmem - the default when no allocator is specified |
|
* |
|
* pool_allocator_nointr - used for pools that will not be accessed |
|
* in interrupt context. |
|
*/ |
|
void *pool_page_alloc(struct pool *, int); |
|
void pool_page_free(struct pool *, void *); |
|
|
|
#ifdef POOL_SUBPAGE |
|
struct pool_allocator pool_allocator_kmem_fullpage = { |
|
pool_page_alloc, pool_page_free, 0, |
|
}; |
|
#else |
|
struct pool_allocator pool_allocator_kmem = { |
|
pool_page_alloc, pool_page_free, 0, |
|
}; |
|
#endif |
|
|
|
void *pool_page_alloc_nointr(struct pool *, int); |
|
void pool_page_free_nointr(struct pool *, void *); |
|
|
|
#ifdef POOL_SUBPAGE |
|
struct pool_allocator pool_allocator_nointr_fullpage = { |
|
pool_page_alloc_nointr, pool_page_free_nointr, 0, |
|
}; |
|
#else |
|
struct pool_allocator pool_allocator_nointr = { |
|
pool_page_alloc_nointr, pool_page_free_nointr, 0, |
|
}; |
#endif |
#endif |
|
|
|
#ifdef POOL_SUBPAGE |
|
void *pool_subpage_alloc(struct pool *, int); |
|
void pool_subpage_free(struct pool *, void *); |
|
|
|
struct pool_allocator pool_allocator_kmem = { |
|
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
|
}; |
|
|
|
void *pool_subpage_alloc_nointr(struct pool *, int); |
|
void pool_subpage_free_nointr(struct pool *, void *); |
|
|
|
struct pool_allocator pool_allocator_nointr = { |
|
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, |
|
}; |
|
#endif /* POOL_SUBPAGE */ |
|
|
|
/* |
|
* We have at least three different resources for the same allocation and |
|
* 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 *pp, *start; |
|
int s, freed; |
|
void *res; |
|
|
|
LOCK_ASSERT(!simple_lock_held(&org->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; |
|
} |
|
|
|
/* |
|
* Drain all pools, that use this allocator. |
|
* We do this to reclaim VA space. |
|
* pa_alloc is responsible for waiting for |
|
* 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; |
|
|
|
s = splvm(); |
|
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); |
|
} while (freed); |
|
return (NULL); |
|
} |
|
|
|
void |
|
pool_allocator_free(struct pool *pp, void *v) |
|
{ |
|
struct pool_allocator *pa = pp->pr_alloc; |
|
int s; |
|
|
|
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); |
|
|
|
(*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 * |
|
pool_page_alloc(struct pool *pp, int flags) |
|
{ |
|
boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE; |
|
|
|
return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok)); |
|
} |
|
|
|
void |
|
pool_page_free(struct pool *pp, void *v) |
|
{ |
|
|
|
uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v); |
|
} |
|
|
|
static void * |
|
pool_page_alloc_meta(struct pool *pp, int flags) |
|
{ |
|
boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE; |
|
|
|
return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok)); |
|
} |
|
|
|
static void |
|
pool_page_free_meta(struct pool *pp, void *v) |
|
{ |
|
|
|
uvm_km_free_poolpage(kmem_map, (vaddr_t) v); |
|
} |
|
|
|
#ifdef POOL_SUBPAGE |
|
/* Sub-page allocator, for machines with large hardware pages. */ |
|
void * |
|
pool_subpage_alloc(struct pool *pp, int flags) |
|
{ |
|
void *v; |
|
int s; |
|
s = splvm(); |
|
v = pool_get(&psppool, flags); |
|
splx(s); |
|
return v; |
|
} |
|
|
|
void |
|
pool_subpage_free(struct pool *pp, void *v) |
|
{ |
|
int s; |
|
s = splvm(); |
|
pool_put(&psppool, v); |
|
splx(s); |
|
} |
|
|
|
/* We don't provide a real nointr allocator. Maybe later. */ |
|
void * |
|
pool_subpage_alloc_nointr(struct pool *pp, int flags) |
|
{ |
|
|
|
return (pool_subpage_alloc(pp, flags)); |
|
} |
|
|
|
void |
|
pool_subpage_free_nointr(struct pool *pp, void *v) |
|
{ |
|
|
|
pool_subpage_free(pp, v); |
|
} |
|
#endif /* POOL_SUBPAGE */ |
|
void * |
|
pool_page_alloc_nointr(struct pool *pp, int flags) |
|
{ |
|
boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE; |
|
|
|
return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok)); |
|
} |
|
|
|
void |
|
pool_page_free_nointr(struct pool *pp, void *v) |
|
{ |
|
|
|
uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v); |
|
} |