Enable per-tcache tcache_max

1. add tcache_max and nhbins into tcache_t so that they are per-tcache,
   with one auto tcache per thread, it's also per-thread;
2. add mallctl for each thread to set its own tcache_max (of its auto tcache);
3. store the maximum number of items in each bin instead of using a global storage;
4. add tests for the modifications above.
5. Rename `nhbins` and `tcache_maxclass` to `global_do_not_change_nhbins` and `global_do_not_change_tcache_maxclass`.
This commit is contained in:
guangli-dai
2023-08-06 11:38:30 -07:00
committed by Qi Wang
parent fbca96c433
commit a442d9b895
15 changed files with 528 additions and 222 deletions

View File

@@ -12,7 +12,7 @@
bool opt_tcache = true;
/* tcache_maxclass is set to 32KB by default. */
/* global_do_not_change_tcache_maxclass is set to 32KB by default. */
size_t opt_tcache_max = ((size_t)1) << 15;
/* Reasonable defaults for min and max values. */
@@ -57,16 +57,18 @@ size_t opt_tcache_gc_delay_bytes = 0;
unsigned opt_lg_tcache_flush_small_div = 1;
unsigned opt_lg_tcache_flush_large_div = 1;
cache_bin_info_t *tcache_bin_info;
/* Total stack size required (per tcache). Include the padding above. */
static size_t tcache_bin_alloc_size;
static size_t tcache_bin_alloc_alignment;
/* Number of cache bins enabled, including both large and small. */
unsigned nhbins;
/* Max size class to be cached (can be small or large). */
size_t tcache_maxclass;
/*
* Number of cache bins enabled, including both large and small. This value
* is only used to initialize tcache_nhbins in the per-thread tcache.
* Directly modifying it will not affect threads already launched.
*/
unsigned global_do_not_change_nhbins;
/*
* Max size class to be cached (can be small or large). This value is only used
* to initialize tcache_max in the per-thread tcache. Directly modifying it
* will not affect threads already launched.
*/
size_t global_do_not_change_tcache_maxclass;
tcaches_t *tcaches;
@@ -127,9 +129,9 @@ tcache_gc_small(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
cache_bin_t *cache_bin = &tcache->bins[szind];
cache_bin_sz_t ncached = cache_bin_ncached_get_local(cache_bin,
&tcache_bin_info[szind]);
&cache_bin->bin_info);
cache_bin_sz_t low_water = cache_bin_low_water_get(cache_bin,
&tcache_bin_info[szind]);
&cache_bin->bin_info);
assert(!tcache_slow->bin_refilled[szind]);
size_t nflush = low_water - (low_water >> 2);
@@ -152,7 +154,7 @@ tcache_gc_small(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
* Reduce fill count by 2X. Limit lg_fill_div such that
* the fill count is always at least 1.
*/
if ((cache_bin_info_ncached_max(&tcache_bin_info[szind])
if ((cache_bin_info_ncached_max(&cache_bin->bin_info)
>> (tcache_slow->lg_fill_div[szind] + 1)) >= 1) {
tcache_slow->lg_fill_div[szind]++;
}
@@ -165,9 +167,9 @@ tcache_gc_large(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
assert(szind >= SC_NBINS);
cache_bin_t *cache_bin = &tcache->bins[szind];
cache_bin_sz_t ncached = cache_bin_ncached_get_local(cache_bin,
&tcache_bin_info[szind]);
&cache_bin->bin_info);
cache_bin_sz_t low_water = cache_bin_low_water_get(cache_bin,
&tcache_bin_info[szind]);
&cache_bin->bin_info);
tcache_bin_flush_large(tsd, tcache, cache_bin, szind,
(unsigned)(ncached - low_water + (low_water >> 2)));
}
@@ -187,7 +189,7 @@ tcache_event(tsd_t *tsd) {
tcache_bin_flush_stashed(tsd, tcache, cache_bin, szind, is_small);
cache_bin_sz_t low_water = cache_bin_low_water_get(cache_bin,
&tcache_bin_info[szind]);
&cache_bin->bin_info);
if (low_water > 0) {
if (is_small) {
tcache_gc_small(tsd, tcache_slow, tcache, szind);
@@ -208,7 +210,7 @@ tcache_event(tsd_t *tsd) {
cache_bin_low_water_set(cache_bin);
tcache_slow->next_gc_bin++;
if (tcache_slow->next_gc_bin == nhbins) {
if (tcache_slow->next_gc_bin == tcache_nhbins_get(tcache)) {
tcache_slow->next_gc_bin = 0;
}
}
@@ -233,10 +235,10 @@ tcache_alloc_small_hard(tsdn_t *tsdn, arena_t *arena,
void *ret;
assert(tcache_slow->arena != NULL);
unsigned nfill = cache_bin_info_ncached_max(&tcache_bin_info[binind])
unsigned nfill = cache_bin_info_ncached_max(&cache_bin->bin_info)
>> tcache_slow->lg_fill_div[binind];
arena_cache_bin_fill_small(tsdn, arena, cache_bin,
&tcache_bin_info[binind], binind, nfill);
&cache_bin->bin_info, binind, nfill);
tcache_slow->bin_refilled[binind] = true;
ret = cache_bin_alloc(cache_bin, tcache_success);
@@ -318,7 +320,7 @@ tcache_bin_flush_impl(tsd_t *tsd, tcache_t *tcache, cache_bin_t *cache_bin,
if (small) {
assert(binind < SC_NBINS);
} else {
assert(binind < nhbins);
assert(binind < tcache_nhbins_get(tcache));
}
arena_t *tcache_arena = tcache_slow->arena;
assert(tcache_arena != NULL);
@@ -508,18 +510,18 @@ tcache_bin_flush_bottom(tsd_t *tsd, tcache_t *tcache, cache_bin_t *cache_bin,
tcache_bin_flush_stashed(tsd, tcache, cache_bin, binind, small);
cache_bin_sz_t ncached = cache_bin_ncached_get_local(cache_bin,
&tcache_bin_info[binind]);
&cache_bin->bin_info);
assert((cache_bin_sz_t)rem <= ncached);
unsigned nflush = ncached - rem;
CACHE_BIN_PTR_ARRAY_DECLARE(ptrs, nflush);
cache_bin_init_ptr_array_for_flush(cache_bin, &tcache_bin_info[binind],
cache_bin_init_ptr_array_for_flush(cache_bin, &cache_bin->bin_info,
&ptrs, nflush);
tcache_bin_flush_impl(tsd, tcache, cache_bin, binind, &ptrs, nflush,
small);
cache_bin_finish_flush(cache_bin, &tcache_bin_info[binind], &ptrs,
cache_bin_finish_flush(cache_bin, &cache_bin->bin_info, &ptrs,
ncached - rem);
}
@@ -548,7 +550,7 @@ tcache_bin_flush_large(tsd_t *tsd, tcache_t *tcache, cache_bin_t *cache_bin,
void
tcache_bin_flush_stashed(tsd_t *tsd, tcache_t *tcache, cache_bin_t *cache_bin,
szind_t binind, bool is_small) {
cache_bin_info_t *info = &tcache_bin_info[binind];
cache_bin_info_t *info = &cache_bin->bin_info;
/*
* The two below are for assertion only. The content of original cached
* items remain unchanged -- the stashed items reside on the other end
@@ -633,15 +635,31 @@ tcache_arena_reassociate(tsdn_t *tsdn, tcache_slow_t *tcache_slow,
tcache_arena_associate(tsdn, tcache_slow, tcache, arena);
}
static void
tcache_max_and_nhbins_init(tcache_t *tcache) {
assert(tcache != NULL);
assert(global_do_not_change_tcache_maxclass != 0);
assert(global_do_not_change_nhbins != 0);
tcache->tcache_max = global_do_not_change_tcache_maxclass;
tcache->tcache_nhbins = global_do_not_change_nhbins;
assert(tcache->tcache_nhbins == sz_size2index(tcache->tcache_max) + 1);
}
bool
tsd_tcache_enabled_data_init(tsd_t *tsd) {
/* Called upon tsd initialization. */
tsd_tcache_enabled_set(tsd, opt_tcache);
/*
* tcache is not available yet, but we need to set up its tcache_max
* and tcache_nhbins in advance.
*/
tcache_t *tcache = tsd_tcachep_get(tsd);
tcache_max_and_nhbins_init(tcache);
tsd_slow_update(tsd);
if (opt_tcache) {
/* Trigger tcache init. */
tsd_tcache_data_init(tsd);
tsd_tcache_data_init(tsd, NULL);
}
return false;
@@ -649,7 +667,7 @@ tsd_tcache_enabled_data_init(tsd_t *tsd) {
static void
tcache_init(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
void *mem) {
void *mem, cache_bin_info_t *tcache_bin_info) {
tcache->tcache_slow = tcache_slow;
tcache_slow->tcache = tcache;
@@ -660,17 +678,19 @@ tcache_init(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
/*
* We reserve cache bins for all small size classes, even if some may
* not get used (i.e. bins higher than nhbins). This allows the fast
* and common paths to access cache bin metadata safely w/o worrying
* about which ones are disabled.
* not get used (i.e. bins higher than tcache_nhbins). This allows
* the fast and common paths to access cache bin metadata safely w/o
* worrying about which ones are disabled.
*/
unsigned n_reserved_bins = nhbins < SC_NBINS ? SC_NBINS : nhbins;
unsigned tcache_nhbins = tcache_nhbins_get(tcache);
unsigned n_reserved_bins = tcache_nhbins < SC_NBINS ? SC_NBINS
: tcache_nhbins;
memset(tcache->bins, 0, sizeof(cache_bin_t) * n_reserved_bins);
size_t cur_offset = 0;
cache_bin_preincrement(tcache_bin_info, nhbins, mem,
cache_bin_preincrement(tcache_bin_info, tcache_nhbins, mem,
&cur_offset);
for (unsigned i = 0; i < nhbins; i++) {
for (unsigned i = 0; i < tcache_nhbins; i++) {
if (i < SC_NBINS) {
tcache_slow->lg_fill_div[i] = 1;
tcache_slow->bin_refilled[i] = false;
@@ -682,12 +702,12 @@ tcache_init(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
&cur_offset);
}
/*
* For small size classes beyond tcache_maxclass (i.e. nhbins < NBINS),
* their cache bins are initialized to a state to safely and efficiently
* fail all fastpath alloc / free, so that no additional check around
* nhbins is needed on fastpath.
* For small size classes beyond tcache_max(i.e.
* tcache_nhbins< NBINS), their cache bins are initialized to a state
* to safely and efficiently fail all fastpath alloc / free, so that
* no additional check around tcache_nhbins is needed on fastpath.
*/
for (unsigned i = nhbins; i < SC_NBINS; i++) {
for (unsigned i = tcache_nhbins; i < SC_NBINS; i++) {
/* Disabled small bins. */
cache_bin_t *cache_bin = &tcache->bins[i];
void *fake_stack = mem;
@@ -699,19 +719,102 @@ tcache_init(tsd_t *tsd, tcache_slow_t *tcache_slow, tcache_t *tcache,
}
cache_bin_postincrement(mem, &cur_offset);
/* Sanity check that the whole stack is used. */
assert(cur_offset == tcache_bin_alloc_size);
if (config_debug) {
/* Sanity check that the whole stack is used. */
size_t size, alignment;
cache_bin_info_compute_alloc(tcache_bin_info, tcache_nhbins,
&size, &alignment);
assert(cur_offset == size);
}
}
static inline unsigned
tcache_ncached_max_compute(szind_t szind, unsigned current_nhbins) {
if (szind >= SC_NBINS) {
assert(szind < current_nhbins);
return opt_tcache_nslots_large;
}
unsigned slab_nregs = bin_infos[szind].nregs;
/* We may modify these values; start with the opt versions. */
unsigned nslots_small_min = opt_tcache_nslots_small_min;
unsigned nslots_small_max = opt_tcache_nslots_small_max;
/*
* Clamp values to meet our constraints -- even, nonzero, min < max, and
* suitable for a cache bin size.
*/
if (opt_tcache_nslots_small_max > CACHE_BIN_NCACHED_MAX) {
nslots_small_max = CACHE_BIN_NCACHED_MAX;
}
if (nslots_small_min % 2 != 0) {
nslots_small_min++;
}
if (nslots_small_max % 2 != 0) {
nslots_small_max--;
}
if (nslots_small_min < 2) {
nslots_small_min = 2;
}
if (nslots_small_max < 2) {
nslots_small_max = 2;
}
if (nslots_small_min > nslots_small_max) {
nslots_small_min = nslots_small_max;
}
unsigned candidate;
if (opt_lg_tcache_nslots_mul < 0) {
candidate = slab_nregs >> (-opt_lg_tcache_nslots_mul);
} else {
candidate = slab_nregs << opt_lg_tcache_nslots_mul;
}
if (candidate % 2 != 0) {
/*
* We need the candidate size to be even -- we assume that we
* can divide by two and get a positive number (e.g. when
* flushing).
*/
++candidate;
}
if (candidate <= nslots_small_min) {
return nslots_small_min;
} else if (candidate <= nslots_small_max) {
return candidate;
} else {
return nslots_small_max;
}
}
static void
tcache_bin_info_compute(cache_bin_info_t *tcache_bin_info,
unsigned tcache_nhbins) {
for (szind_t i = 0; i < tcache_nhbins; i++) {
unsigned ncached_max = tcache_ncached_max_compute(i,
tcache_nhbins);
cache_bin_info_init(&tcache_bin_info[i], ncached_max);
}
for (szind_t i = tcache_nhbins; i < SC_NBINS; i++) {
/* Disabled small bins. */
cache_bin_info_init(&tcache_bin_info[i], 0);
}
}
/* Initialize auto tcache (embedded in TSD). */
bool
tsd_tcache_data_init(tsd_t *tsd) {
tsd_tcache_data_init(tsd_t *tsd, arena_t *arena) {
tcache_slow_t *tcache_slow = tsd_tcache_slowp_get_unsafe(tsd);
tcache_t *tcache = tsd_tcachep_get_unsafe(tsd);
assert(cache_bin_still_zero_initialized(&tcache->bins[0]));
size_t alignment = tcache_bin_alloc_alignment;
size_t size = sz_sa2u(tcache_bin_alloc_size, alignment);
unsigned tcache_nhbins = tcache_nhbins_get(tcache);
size_t size, alignment;
/* Takes 146B stack space. */
cache_bin_info_t tcache_bin_info[TCACHE_NBINS_MAX] = {0};
tcache_bin_info_compute(tcache_bin_info, tcache_nhbins);
cache_bin_info_compute_alloc(tcache_bin_info, tcache_nhbins,
&size, &alignment);
size = sz_sa2u(size, alignment);
void *mem = ipallocztm(tsd_tsdn(tsd), size, alignment, true, NULL,
true, arena_get(TSDN_NULL, 0, true));
@@ -719,7 +822,7 @@ tsd_tcache_data_init(tsd_t *tsd) {
return true;
}
tcache_init(tsd, tcache_slow, tcache, mem);
tcache_init(tsd, tcache_slow, tcache, mem, tcache_bin_info);
/*
* Initialization is a bit tricky here. After malloc init is done, all
* threads can rely on arena_choose and associate tcache accordingly.
@@ -729,14 +832,15 @@ tsd_tcache_data_init(tsd_t *tsd) {
* arena_choose_hard() will re-associate properly.
*/
tcache_slow->arena = NULL;
arena_t *arena;
if (!malloc_initialized()) {
/* If in initialization, assign to a0. */
arena = arena_get(tsd_tsdn(tsd), 0, false);
tcache_arena_associate(tsd_tsdn(tsd), tcache_slow, tcache,
arena);
} else {
arena = arena_choose(tsd, NULL);
if (arena == NULL) {
arena = arena_choose(tsd, NULL);
}
/* This may happen if thread.tcache.enabled is used. */
if (tcache_slow->arena == NULL) {
tcache_arena_associate(tsd_tsdn(tsd), tcache_slow,
@@ -756,21 +860,29 @@ tcache_create_explicit(tsd_t *tsd) {
* the beginning of the whole allocation (for freeing). The makes sure
* the cache bins have the requested alignment.
*/
size_t size = tcache_bin_alloc_size + sizeof(tcache_t)
unsigned tcache_nhbins = global_do_not_change_nhbins;
size_t tcache_size, alignment;
cache_bin_info_t tcache_bin_info[TCACHE_NBINS_MAX] = {0};
tcache_bin_info_compute(tcache_bin_info, tcache_nhbins);
cache_bin_info_compute_alloc(tcache_bin_info, tcache_nhbins,
&tcache_size, &alignment);
size_t size = tcache_size + sizeof(tcache_t)
+ sizeof(tcache_slow_t);
/* Naturally align the pointer stacks. */
size = PTR_CEILING(size);
size = sz_sa2u(size, tcache_bin_alloc_alignment);
size = sz_sa2u(size, alignment);
void *mem = ipallocztm(tsd_tsdn(tsd), size, tcache_bin_alloc_alignment,
void *mem = ipallocztm(tsd_tsdn(tsd), size, alignment,
true, NULL, true, arena_get(TSDN_NULL, 0, true));
if (mem == NULL) {
return NULL;
}
tcache_t *tcache = (void *)((byte_t *)mem + tcache_bin_alloc_size);
tcache_t *tcache = (void *)((byte_t *)mem + tcache_size);
tcache_slow_t *tcache_slow =
(void *)((byte_t *)mem + tcache_bin_alloc_size + sizeof(tcache_t));
tcache_init(tsd, tcache_slow, tcache, mem);
(void *)((byte_t *)mem + tcache_size + sizeof(tcache_t));
tcache_max_and_nhbins_init(tcache);
tcache_init(tsd, tcache_slow, tcache, mem, tcache_bin_info);
tcache_arena_associate(tsd_tsdn(tsd), tcache_slow, tcache,
arena_ichoose(tsd, NULL));
@@ -783,7 +895,7 @@ tcache_flush_cache(tsd_t *tsd, tcache_t *tcache) {
tcache_slow_t *tcache_slow = tcache->tcache_slow;
assert(tcache_slow->arena != NULL);
for (unsigned i = 0; i < nhbins; i++) {
for (unsigned i = 0; i < tcache_nhbins_get(tcache); i++) {
cache_bin_t *cache_bin = &tcache->bins[i];
if (i < SC_NBINS) {
tcache_bin_flush_small(tsd, tcache, cache_bin, i, 0);
@@ -811,7 +923,7 @@ tcache_destroy(tsd_t *tsd, tcache_t *tcache, bool tsd_tcache) {
if (tsd_tcache) {
cache_bin_t *cache_bin = &tcache->bins[0];
cache_bin_assert_empty(cache_bin, &tcache_bin_info[0]);
cache_bin_assert_empty(cache_bin, &cache_bin->bin_info);
}
idalloctm(tsd_tsdn(tsd), tcache_slow->dyn_alloc, NULL, NULL, true,
true);
@@ -849,13 +961,9 @@ tcache_cleanup(tsd_t *tsd) {
assert(!cache_bin_still_zero_initialized(&tcache->bins[0]));
tcache_destroy(tsd, tcache, true);
if (config_debug) {
/*
* For debug testing only, we want to pretend we're still in the
* zero-initialized state.
*/
memset(tcache->bins, 0, sizeof(cache_bin_t) * nhbins);
}
/* Make sure all bins used are reinitialized to the clean state. */
memset(tcache->bins, 0, sizeof(cache_bin_t) *
tcache_nhbins_get(tcache));
}
void
@@ -863,7 +971,7 @@ tcache_stats_merge(tsdn_t *tsdn, tcache_t *tcache, arena_t *arena) {
cassert(config_stats);
/* Merge and reset tcache stats. */
for (unsigned i = 0; i < nhbins; i++) {
for (unsigned i = 0; i < tcache_nhbins_get(tcache); i++) {
cache_bin_t *cache_bin = &tcache->bins[i];
if (i < SC_NBINS) {
bin_t *bin = arena_bin_choose(tsdn, arena, i, NULL);
@@ -986,97 +1094,18 @@ tcaches_destroy(tsd_t *tsd, unsigned ind) {
}
}
static unsigned
tcache_ncached_max_compute(szind_t szind) {
if (szind >= SC_NBINS) {
assert(szind < nhbins);
return opt_tcache_nslots_large;
}
unsigned slab_nregs = bin_infos[szind].nregs;
/* We may modify these values; start with the opt versions. */
unsigned nslots_small_min = opt_tcache_nslots_small_min;
unsigned nslots_small_max = opt_tcache_nslots_small_max;
/*
* Clamp values to meet our constraints -- even, nonzero, min < max, and
* suitable for a cache bin size.
*/
if (opt_tcache_nslots_small_max > CACHE_BIN_NCACHED_MAX) {
nslots_small_max = CACHE_BIN_NCACHED_MAX;
}
if (nslots_small_min % 2 != 0) {
nslots_small_min++;
}
if (nslots_small_max % 2 != 0) {
nslots_small_max--;
}
if (nslots_small_min < 2) {
nslots_small_min = 2;
}
if (nslots_small_max < 2) {
nslots_small_max = 2;
}
if (nslots_small_min > nslots_small_max) {
nslots_small_min = nslots_small_max;
}
unsigned candidate;
if (opt_lg_tcache_nslots_mul < 0) {
candidate = slab_nregs >> (-opt_lg_tcache_nslots_mul);
} else {
candidate = slab_nregs << opt_lg_tcache_nslots_mul;
}
if (candidate % 2 != 0) {
/*
* We need the candidate size to be even -- we assume that we
* can divide by two and get a positive number (e.g. when
* flushing).
*/
++candidate;
}
if (candidate <= nslots_small_min) {
return nslots_small_min;
} else if (candidate <= nslots_small_max) {
return candidate;
} else {
return nslots_small_max;
}
}
bool
tcache_boot(tsdn_t *tsdn, base_t *base) {
tcache_maxclass = sz_s2u(opt_tcache_max);
assert(tcache_maxclass <= TCACHE_MAXCLASS_LIMIT);
nhbins = sz_size2index(tcache_maxclass) + 1;
global_do_not_change_tcache_maxclass = sz_s2u(opt_tcache_max);
assert(global_do_not_change_tcache_maxclass <= TCACHE_MAXCLASS_LIMIT);
global_do_not_change_nhbins =
sz_size2index(global_do_not_change_tcache_maxclass) + 1;
if (malloc_mutex_init(&tcaches_mtx, "tcaches", WITNESS_RANK_TCACHES,
malloc_mutex_rank_exclusive)) {
return true;
}
/* Initialize tcache_bin_info. See comments in tcache_init(). */
unsigned n_reserved_bins = nhbins < SC_NBINS ? SC_NBINS : nhbins;
size_t size = n_reserved_bins * sizeof(cache_bin_info_t);
tcache_bin_info = (cache_bin_info_t *)base_alloc(tsdn, base, size,
CACHELINE);
if (tcache_bin_info == NULL) {
return true;
}
for (szind_t i = 0; i < nhbins; i++) {
unsigned ncached_max = tcache_ncached_max_compute(i);
cache_bin_info_init(&tcache_bin_info[i], ncached_max);
}
for (szind_t i = nhbins; i < SC_NBINS; i++) {
/* Disabled small bins. */
cache_bin_info_init(&tcache_bin_info[i], 0);
assert(tcache_small_bin_disabled(i, NULL));
}
cache_bin_info_compute_alloc(tcache_bin_info, nhbins,
&tcache_bin_alloc_size, &tcache_bin_alloc_alignment);
return false;
}