haproxy/src/pool.c
Willy Tarreau 20dc3cd4a6 MINOR: pools: move the LRU cache heads to thread_info
The LRU cache head was an array of list, which causes false sharing
between 4 to 8 threads in the same cache line. Let's move it to the
thread_info structure instead. There's no need to do the same for the
pool_cache[] array since it's already quite large (32 pointers each).

By doing this the request rate increased by 1% on a 16-thread machine.
2020-06-29 10:36:37 +02:00

668 lines
18 KiB
C

/*
* Memory management functions.
*
* Copyright 2000-2007 Willy Tarreau <w@1wt.eu>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
*/
#include <errno.h>
#include <haproxy/activity-t.h>
#include <haproxy/api.h>
#include <haproxy/applet-t.h>
#include <haproxy/cfgparse.h>
#include <haproxy/channel.h>
#include <haproxy/cli.h>
#include <haproxy/errors.h>
#include <haproxy/global.h>
#include <haproxy/list.h>
#include <haproxy/pool.h>
#include <haproxy/stats-t.h>
#include <haproxy/stream_interface.h>
#include <haproxy/thread.h>
#include <haproxy/tools.h>
#ifdef CONFIG_HAP_LOCAL_POOLS
/* These are the most common pools, expected to be initialized first. These
* ones are allocated from an array, allowing to map them to an index.
*/
struct pool_head pool_base_start[MAX_BASE_POOLS] = { };
unsigned int pool_base_count = 0;
/* These ones are initialized per-thread on startup by init_pools() */
struct pool_cache_head pool_cache[MAX_THREADS][MAX_BASE_POOLS];
THREAD_LOCAL size_t pool_cache_bytes = 0; /* total cache size */
THREAD_LOCAL size_t pool_cache_count = 0; /* #cache objects */
#endif
static struct list pools = LIST_HEAD_INIT(pools);
int mem_poison_byte = -1;
#ifdef DEBUG_FAIL_ALLOC
static int mem_fail_rate = 0;
static int mem_should_fail(const struct pool_head *);
#endif
/* Try to find an existing shared pool with the same characteristics and
* returns it, otherwise creates this one. NULL is returned if no memory
* is available for a new creation. Two flags are supported :
* - MEM_F_SHARED to indicate that the pool may be shared with other users
* - MEM_F_EXACT to indicate that the size must not be rounded up
*/
struct pool_head *create_pool(char *name, unsigned int size, unsigned int flags)
{
struct pool_head *pool;
struct pool_head *entry;
struct list *start;
unsigned int align;
int idx __maybe_unused;
/* We need to store a (void *) at the end of the chunks. Since we know
* that the malloc() function will never return such a small size,
* let's round the size up to something slightly bigger, in order to
* ease merging of entries. Note that the rounding is a power of two.
* This extra (void *) is not accounted for in the size computation
* so that the visible parts outside are not affected.
*
* Note: for the LRU cache, we need to store 2 doubly-linked lists.
*/
if (!(flags & MEM_F_EXACT)) {
align = 4 * sizeof(void *); // 2 lists = 4 pointers min
size = ((size + POOL_EXTRA + align - 1) & -align) - POOL_EXTRA;
}
/* TODO: thread: we do not lock pool list for now because all pools are
* created during HAProxy startup (so before threads creation) */
start = &pools;
pool = NULL;
list_for_each_entry(entry, &pools, list) {
if (entry->size == size) {
/* either we can share this place and we take it, or
* we look for a shareable one or for the next position
* before which we will insert a new one.
*/
if (flags & entry->flags & MEM_F_SHARED) {
/* we can share this one */
pool = entry;
DPRINTF(stderr, "Sharing %s with %s\n", name, pool->name);
break;
}
}
else if (entry->size > size) {
/* insert before this one */
start = &entry->list;
break;
}
}
if (!pool) {
#ifdef CONFIG_HAP_LOCAL_POOLS
if (pool_base_count < MAX_BASE_POOLS)
pool = &pool_base_start[pool_base_count++];
if (!pool) {
/* look for a freed entry */
for (entry = pool_base_start; entry != pool_base_start + MAX_BASE_POOLS; entry++) {
if (!entry->size) {
pool = entry;
break;
}
}
}
#endif
if (!pool)
pool = calloc(1, sizeof(*pool));
if (!pool)
return NULL;
if (name)
strlcpy2(pool->name, name, sizeof(pool->name));
pool->size = size;
pool->flags = flags;
LIST_ADDQ(start, &pool->list);
#ifdef CONFIG_HAP_LOCAL_POOLS
/* update per-thread pool cache if necessary */
idx = pool_get_index(pool);
if (idx >= 0) {
int thr;
for (thr = 0; thr < MAX_THREADS; thr++)
pool_cache[thr][idx].size = size;
}
#endif
HA_SPIN_INIT(&pool->lock);
}
pool->users++;
return pool;
}
#ifdef CONFIG_HAP_LOCAL_POOLS
/* Evicts some of the oldest objects from the local cache, pushing them to the
* global pool.
*/
void pool_evict_from_cache()
{
struct pool_cache_item *item;
struct pool_cache_head *ph;
do {
item = LIST_PREV(&ti->pool_lru_head, struct pool_cache_item *, by_lru);
/* note: by definition we remove oldest objects so they also are the
* oldest in their own pools, thus their next is the pool's head.
*/
ph = LIST_NEXT(&item->by_pool, struct pool_cache_head *, list);
LIST_DEL(&item->by_pool);
LIST_DEL(&item->by_lru);
ph->count--;
pool_cache_count--;
pool_cache_bytes -= ph->size;
__pool_free(pool_base_start + (ph - pool_cache[tid]), item);
} while (pool_cache_bytes > CONFIG_HAP_POOL_CACHE_SIZE * 7 / 8);
}
#endif
#ifdef CONFIG_HAP_LOCKLESS_POOLS
/* Allocates new entries for pool <pool> until there are at least <avail> + 1
* available, then returns the last one for immediate use, so that at least
* <avail> are left available in the pool upon return. NULL is returned if the
* last entry could not be allocated. It's important to note that at least one
* allocation is always performed even if there are enough entries in the pool.
* A call to the garbage collector is performed at most once in case malloc()
* returns an error, before returning NULL.
*/
void *__pool_refill_alloc(struct pool_head *pool, unsigned int avail)
{
void *ptr = NULL, **free_list;
int failed = 0;
int size = pool->size;
int limit = pool->limit;
int allocated = pool->allocated, allocated_orig = allocated;
/* stop point */
avail += pool->used;
while (1) {
if (limit && allocated >= limit) {
_HA_ATOMIC_ADD(&pool->allocated, allocated - allocated_orig);
activity[tid].pool_fail++;
return NULL;
}
swrate_add_scaled(&pool->needed_avg, POOL_AVG_SAMPLES, pool->allocated, POOL_AVG_SAMPLES/4);
ptr = pool_alloc_area(size + POOL_EXTRA);
if (!ptr) {
_HA_ATOMIC_ADD(&pool->failed, 1);
if (failed) {
activity[tid].pool_fail++;
return NULL;
}
failed++;
pool_gc(pool);
continue;
}
if (++allocated > avail)
break;
free_list = pool->free_list;
do {
*POOL_LINK(pool, ptr) = free_list;
__ha_barrier_store();
} while (_HA_ATOMIC_CAS(&pool->free_list, &free_list, ptr) == 0);
}
__ha_barrier_atomic_store();
_HA_ATOMIC_ADD(&pool->allocated, allocated - allocated_orig);
_HA_ATOMIC_ADD(&pool->used, 1);
#ifdef DEBUG_MEMORY_POOLS
/* keep track of where the element was allocated from */
*POOL_LINK(pool, ptr) = (void *)pool;
#endif
return ptr;
}
void *pool_refill_alloc(struct pool_head *pool, unsigned int avail)
{
void *ptr;
ptr = __pool_refill_alloc(pool, avail);
return ptr;
}
/*
* This function frees whatever can be freed in pool <pool>.
*/
void pool_flush(struct pool_head *pool)
{
struct pool_free_list cmp, new;
void **next, *temp;
int removed = 0;
if (!pool)
return;
HA_SPIN_LOCK(POOL_LOCK, &pool->lock);
do {
cmp.free_list = pool->free_list;
cmp.seq = pool->seq;
new.free_list = NULL;
new.seq = cmp.seq + 1;
} while (!_HA_ATOMIC_DWCAS(&pool->free_list, &cmp, &new));
__ha_barrier_atomic_store();
HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock);
next = cmp.free_list;
while (next) {
temp = next;
next = *POOL_LINK(pool, temp);
removed++;
pool_free_area(temp, pool->size + POOL_EXTRA);
}
pool->free_list = next;
_HA_ATOMIC_SUB(&pool->allocated, removed);
/* here, we should have pool->allocate == pool->used */
}
/*
* This function frees whatever can be freed in all pools, but respecting
* the minimum thresholds imposed by owners. It makes sure to be alone to
* run by using thread_isolate(). <pool_ctx> is unused.
*/
void pool_gc(struct pool_head *pool_ctx)
{
struct pool_head *entry;
int isolated = thread_isolated();
if (!isolated)
thread_isolate();
list_for_each_entry(entry, &pools, list) {
while ((int)((volatile int)entry->allocated - (volatile int)entry->used) > (int)entry->minavail) {
struct pool_free_list cmp, new;
cmp.seq = entry->seq;
__ha_barrier_load();
cmp.free_list = entry->free_list;
__ha_barrier_load();
if (cmp.free_list == NULL)
break;
new.free_list = *POOL_LINK(entry, cmp.free_list);
new.seq = cmp.seq + 1;
if (HA_ATOMIC_DWCAS(&entry->free_list, &cmp, &new) == 0)
continue;
pool_free_area(cmp.free_list, entry->size + POOL_EXTRA);
_HA_ATOMIC_SUB(&entry->allocated, 1);
}
}
if (!isolated)
thread_release();
}
#else /* CONFIG_HAP_LOCKLESS_POOLS */
/* Allocates new entries for pool <pool> until there are at least <avail> + 1
* available, then returns the last one for immediate use, so that at least
* <avail> are left available in the pool upon return. NULL is returned if the
* last entry could not be allocated. It's important to note that at least one
* allocation is always performed even if there are enough entries in the pool.
* A call to the garbage collector is performed at most once in case malloc()
* returns an error, before returning NULL.
*/
void *__pool_refill_alloc(struct pool_head *pool, unsigned int avail)
{
void *ptr = NULL;
int failed = 0;
#ifdef DEBUG_FAIL_ALLOC
if (mem_should_fail(pool))
return NULL;
#endif
/* stop point */
avail += pool->used;
while (1) {
if (pool->limit && pool->allocated >= pool->limit) {
activity[tid].pool_fail++;
return NULL;
}
swrate_add_scaled(&pool->needed_avg, POOL_AVG_SAMPLES, pool->allocated, POOL_AVG_SAMPLES/4);
HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock);
ptr = pool_alloc_area(pool->size + POOL_EXTRA);
#ifdef DEBUG_MEMORY_POOLS
/* keep track of where the element was allocated from. This
* is done out of the lock so that the system really allocates
* the data without harming other threads waiting on the lock.
*/
if (ptr)
*POOL_LINK(pool, ptr) = (void *)pool;
#endif
HA_SPIN_LOCK(POOL_LOCK, &pool->lock);
if (!ptr) {
pool->failed++;
if (failed) {
activity[tid].pool_fail++;
return NULL;
}
failed++;
pool_gc(pool);
continue;
}
if (++pool->allocated > avail)
break;
*POOL_LINK(pool, ptr) = (void *)pool->free_list;
pool->free_list = ptr;
}
pool->used++;
return ptr;
}
void *pool_refill_alloc(struct pool_head *pool, unsigned int avail)
{
void *ptr;
HA_SPIN_LOCK(POOL_LOCK, &pool->lock);
ptr = __pool_refill_alloc(pool, avail);
HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock);
return ptr;
}
/*
* This function frees whatever can be freed in pool <pool>.
*/
void pool_flush(struct pool_head *pool)
{
void *temp;
if (!pool)
return;
while (1) {
HA_SPIN_LOCK(POOL_LOCK, &pool->lock);
temp = pool->free_list;
if (!temp) {
HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock);
break;
}
pool->free_list = *POOL_LINK(pool, temp);
pool->allocated--;
HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock);
pool_free_area(temp, pool->size + POOL_EXTRA);
}
/* here, we should have pool->allocated == pool->used */
}
/*
* This function frees whatever can be freed in all pools, but respecting
* the minimum thresholds imposed by owners. It makes sure to be alone to
* run by using thread_isolate(). <pool_ctx> is unused.
*/
void pool_gc(struct pool_head *pool_ctx)
{
struct pool_head *entry;
int isolated = thread_isolated();
if (!isolated)
thread_isolate();
list_for_each_entry(entry, &pools, list) {
void *temp;
//qfprintf(stderr, "Flushing pool %s\n", entry->name);
while (entry->free_list &&
(int)(entry->allocated - entry->used) > (int)entry->minavail) {
temp = entry->free_list;
entry->free_list = *POOL_LINK(entry, temp);
entry->allocated--;
pool_free_area(temp, entry->size + POOL_EXTRA);
}
}
if (!isolated)
thread_release();
}
#endif
/*
* This function destroys a pool by freeing it completely, unless it's still
* in use. This should be called only under extreme circumstances. It always
* returns NULL if the resulting pool is empty, easing the clearing of the old
* pointer, otherwise it returns the pool.
* .
*/
void *pool_destroy(struct pool_head *pool)
{
if (pool) {
pool_flush(pool);
if (pool->used)
return pool;
pool->users--;
if (!pool->users) {
LIST_DEL(&pool->list);
#ifndef CONFIG_HAP_LOCKLESS_POOLS
HA_SPIN_DESTROY(&pool->lock);
#endif
#ifdef CONFIG_HAP_LOCAL_POOLS
if ((pool - pool_base_start) < MAX_BASE_POOLS)
memset(pool, 0, sizeof(*pool));
else
#endif
free(pool);
}
}
return NULL;
}
/* This destroys all pools on exit. It is *not* thread safe. */
void pool_destroy_all()
{
struct pool_head *entry, *back;
list_for_each_entry_safe(entry, back, &pools, list)
pool_destroy(entry);
}
/* This function dumps memory usage information into the trash buffer. */
void dump_pools_to_trash()
{
struct pool_head *entry;
unsigned long allocated, used;
int nbpools;
allocated = used = nbpools = 0;
chunk_printf(&trash, "Dumping pools usage. Use SIGQUIT to flush them.\n");
list_for_each_entry(entry, &pools, list) {
#ifndef CONFIG_HAP_LOCKLESS_POOLS
HA_SPIN_LOCK(POOL_LOCK, &entry->lock);
#endif
chunk_appendf(&trash, " - Pool %s (%u bytes) : %u allocated (%u bytes), %u used, needed_avg %u, %u failures, %u users, @%p=%02d%s\n",
entry->name, entry->size, entry->allocated,
entry->size * entry->allocated, entry->used,
swrate_avg(entry->needed_avg, POOL_AVG_SAMPLES), entry->failed,
entry->users, entry, (int)pool_get_index(entry),
(entry->flags & MEM_F_SHARED) ? " [SHARED]" : "");
allocated += entry->allocated * entry->size;
used += entry->used * entry->size;
nbpools++;
#ifndef CONFIG_HAP_LOCKLESS_POOLS
HA_SPIN_UNLOCK(POOL_LOCK, &entry->lock);
#endif
}
chunk_appendf(&trash, "Total: %d pools, %lu bytes allocated, %lu used.\n",
nbpools, allocated, used);
}
/* Dump statistics on pools usage. */
void dump_pools(void)
{
dump_pools_to_trash();
qfprintf(stderr, "%s", trash.area);
}
/* This function returns the total number of failed pool allocations */
int pool_total_failures()
{
struct pool_head *entry;
int failed = 0;
list_for_each_entry(entry, &pools, list)
failed += entry->failed;
return failed;
}
/* This function returns the total amount of memory allocated in pools (in bytes) */
unsigned long pool_total_allocated()
{
struct pool_head *entry;
unsigned long allocated = 0;
list_for_each_entry(entry, &pools, list)
allocated += entry->allocated * entry->size;
return allocated;
}
/* This function returns the total amount of memory used in pools (in bytes) */
unsigned long pool_total_used()
{
struct pool_head *entry;
unsigned long used = 0;
list_for_each_entry(entry, &pools, list)
used += entry->used * entry->size;
return used;
}
/* This function dumps memory usage information onto the stream interface's
* read buffer. It returns 0 as long as it does not complete, non-zero upon
* completion. No state is used.
*/
static int cli_io_handler_dump_pools(struct appctx *appctx)
{
struct stream_interface *si = appctx->owner;
dump_pools_to_trash();
if (ci_putchk(si_ic(si), &trash) == -1) {
si_rx_room_blk(si);
return 0;
}
return 1;
}
/* callback used to create early pool <name> of size <size> and store the
* resulting pointer into <ptr>. If the allocation fails, it quits with after
* emitting an error message.
*/
void create_pool_callback(struct pool_head **ptr, char *name, unsigned int size)
{
*ptr = create_pool(name, size, MEM_F_SHARED);
if (!*ptr) {
ha_alert("Failed to allocate pool '%s' of size %u : %s. Aborting.\n",
name, size, strerror(errno));
exit(1);
}
}
/* Initializes all per-thread arrays on startup */
static void init_pools()
{
#ifdef CONFIG_HAP_LOCAL_POOLS
int thr, idx;
for (thr = 0; thr < MAX_THREADS; thr++) {
for (idx = 0; idx < MAX_BASE_POOLS; idx++) {
LIST_INIT(&pool_cache[thr][idx].list);
pool_cache[thr][idx].size = 0;
}
LIST_INIT(&ha_thread_info[thr].pool_lru_head);
}
#endif
}
INITCALL0(STG_PREPARE, init_pools);
/* register cli keywords */
static struct cli_kw_list cli_kws = {{ },{
{ { "show", "pools", NULL }, "show pools : report information about the memory pools usage", NULL, cli_io_handler_dump_pools },
{{},}
}};
INITCALL1(STG_REGISTER, cli_register_kw, &cli_kws);
#ifdef DEBUG_FAIL_ALLOC
#define MEM_FAIL_MAX_CHAR 32
#define MEM_FAIL_MAX_STR 128
static int mem_fail_cur_idx;
static char mem_fail_str[MEM_FAIL_MAX_CHAR * MEM_FAIL_MAX_STR];
__decl_thread(static HA_SPINLOCK_T mem_fail_lock);
int mem_should_fail(const struct pool_head *pool)
{
int ret = 0;
int n;
if (mem_fail_rate > 0 && !(global.mode & MODE_STARTING)) {
int randnb = ha_random() % 100;
if (mem_fail_rate > randnb)
ret = 1;
else
ret = 0;
}
HA_SPIN_LOCK(POOL_LOCK, &mem_fail_lock);
n = snprintf(&mem_fail_str[mem_fail_cur_idx * MEM_FAIL_MAX_CHAR],
MEM_FAIL_MAX_CHAR - 2,
"%d %.18s %d %d", mem_fail_cur_idx, pool->name, ret, tid);
while (n < MEM_FAIL_MAX_CHAR - 1)
mem_fail_str[mem_fail_cur_idx * MEM_FAIL_MAX_CHAR + n++] = ' ';
if (mem_fail_cur_idx < MEM_FAIL_MAX_STR - 1)
mem_fail_str[mem_fail_cur_idx * MEM_FAIL_MAX_CHAR + n] = '\n';
else
mem_fail_str[mem_fail_cur_idx * MEM_FAIL_MAX_CHAR + n] = 0;
mem_fail_cur_idx++;
if (mem_fail_cur_idx == MEM_FAIL_MAX_STR)
mem_fail_cur_idx = 0;
HA_SPIN_UNLOCK(POOL_LOCK, &mem_fail_lock);
return ret;
}
/* config parser for global "tune.fail-alloc" */
static int mem_parse_global_fail_alloc(char **args, int section_type, struct proxy *curpx,
struct proxy *defpx, const char *file, int line,
char **err)
{
if (too_many_args(1, args, err, NULL))
return -1;
mem_fail_rate = atoi(args[1]);
if (mem_fail_rate < 0 || mem_fail_rate > 100) {
memprintf(err, "'%s' expects a numeric value between 0 and 100.", args[0]);
return -1;
}
return 0;
}
#endif
/* register global config keywords */
static struct cfg_kw_list mem_cfg_kws = {ILH, {
#ifdef DEBUG_FAIL_ALLOC
{ CFG_GLOBAL, "tune.fail-alloc", mem_parse_global_fail_alloc },
#endif
{ 0, NULL, NULL }
}};
INITCALL1(STG_REGISTER, cfg_register_keywords, &mem_cfg_kws);
/*
* Local variables:
* c-indent-level: 8
* c-basic-offset: 8
* End:
*/