diff options
Diffstat (limited to 'deps/jemalloc/include/jemalloc/internal')
64 files changed, 2112 insertions, 1057 deletions
diff --git a/deps/jemalloc/include/jemalloc/internal/arena_externs.h b/deps/jemalloc/include/jemalloc/internal/arena_externs.h index 4b3732b41..a4523ae0c 100644 --- a/deps/jemalloc/include/jemalloc/internal/arena_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/arena_externs.h @@ -3,8 +3,8 @@ #include "jemalloc/internal/bin.h" #include "jemalloc/internal/extent_dss.h" +#include "jemalloc/internal/hook.h" #include "jemalloc/internal/pages.h" -#include "jemalloc/internal/size_classes.h" #include "jemalloc/internal/stats.h" extern ssize_t opt_dirty_decay_ms; @@ -16,13 +16,17 @@ extern const char *percpu_arena_mode_names[]; extern const uint64_t h_steps[SMOOTHSTEP_NSTEPS]; extern malloc_mutex_t arenas_lock; +extern size_t opt_oversize_threshold; +extern size_t oversize_threshold; + void arena_basic_stats_merge(tsdn_t *tsdn, arena_t *arena, unsigned *nthreads, const char **dss, ssize_t *dirty_decay_ms, ssize_t *muzzy_decay_ms, size_t *nactive, size_t *ndirty, size_t *nmuzzy); void arena_stats_merge(tsdn_t *tsdn, arena_t *arena, unsigned *nthreads, const char **dss, ssize_t *dirty_decay_ms, ssize_t *muzzy_decay_ms, size_t *nactive, size_t *ndirty, size_t *nmuzzy, arena_stats_t *astats, - bin_stats_t *bstats, arena_stats_large_t *lstats); + bin_stats_t *bstats, arena_stats_large_t *lstats, + arena_stats_extents_t *estats); void arena_extents_dirty_dalloc(tsdn_t *tsdn, arena_t *arena, extent_hooks_t **r_extent_hooks, extent_t *extent); #ifdef JEMALLOC_JET @@ -56,16 +60,17 @@ void *arena_malloc_hard(tsdn_t *tsdn, arena_t *arena, size_t size, szind_t ind, bool zero); void *arena_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, size_t alignment, bool zero, tcache_t *tcache); -void arena_prof_promote(tsdn_t *tsdn, const void *ptr, size_t usize); +void arena_prof_promote(tsdn_t *tsdn, void *ptr, size_t usize); void arena_dalloc_promoted(tsdn_t *tsdn, void *ptr, tcache_t *tcache, bool slow_path); -void arena_dalloc_bin_junked_locked(tsdn_t *tsdn, arena_t *arena, - extent_t *extent, void *ptr); +void arena_dalloc_bin_junked_locked(tsdn_t *tsdn, arena_t *arena, bin_t *bin, + szind_t binind, extent_t *extent, void *ptr); void arena_dalloc_small(tsdn_t *tsdn, void *ptr); bool arena_ralloc_no_move(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, - size_t extra, bool zero); + size_t extra, bool zero, size_t *newsize); void *arena_ralloc(tsdn_t *tsdn, arena_t *arena, void *ptr, size_t oldsize, - size_t size, size_t alignment, bool zero, tcache_t *tcache); + size_t size, size_t alignment, bool zero, tcache_t *tcache, + hook_ralloc_args_t *hook_args); dss_prec_t arena_dss_prec_get(arena_t *arena); bool arena_dss_prec_set(arena_t *arena, dss_prec_t dss_prec); ssize_t arena_dirty_decay_ms_default_get(void); @@ -79,7 +84,12 @@ void arena_nthreads_inc(arena_t *arena, bool internal); void arena_nthreads_dec(arena_t *arena, bool internal); size_t arena_extent_sn_next(arena_t *arena); arena_t *arena_new(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks); -void arena_boot(void); +bool arena_init_huge(void); +bool arena_is_huge(unsigned arena_ind); +arena_t *arena_choose_huge(tsd_t *tsd); +bin_t *arena_bin_choose_lock(tsdn_t *tsdn, arena_t *arena, szind_t binind, + unsigned *binshard); +void arena_boot(sc_data_t *sc_data); void arena_prefork0(tsdn_t *tsdn, arena_t *arena); void arena_prefork1(tsdn_t *tsdn, arena_t *arena); void arena_prefork2(tsdn_t *tsdn, arena_t *arena); diff --git a/deps/jemalloc/include/jemalloc/internal/arena_inlines_b.h b/deps/jemalloc/include/jemalloc/internal/arena_inlines_b.h index 2b7e77e72..dd926575f 100644 --- a/deps/jemalloc/include/jemalloc/internal/arena_inlines_b.h +++ b/deps/jemalloc/include/jemalloc/internal/arena_inlines_b.h @@ -4,10 +4,36 @@ #include "jemalloc/internal/jemalloc_internal_types.h" #include "jemalloc/internal/mutex.h" #include "jemalloc/internal/rtree.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/sz.h" #include "jemalloc/internal/ticker.h" +JEMALLOC_ALWAYS_INLINE bool +arena_has_default_hooks(arena_t *arena) { + return (extent_hooks_get(arena) == &extent_hooks_default); +} + +JEMALLOC_ALWAYS_INLINE arena_t * +arena_choose_maybe_huge(tsd_t *tsd, arena_t *arena, size_t size) { + if (arena != NULL) { + return arena; + } + + /* + * For huge allocations, use the dedicated huge arena if both are true: + * 1) is using auto arena selection (i.e. arena == NULL), and 2) the + * thread is not assigned to a manual arena. + */ + if (unlikely(size >= oversize_threshold)) { + arena_t *tsd_arena = tsd_arena_get(tsd); + if (tsd_arena == NULL || arena_is_auto(tsd_arena)) { + return arena_choose_huge(tsd); + } + } + + return arena_choose(tsd, NULL); +} + JEMALLOC_ALWAYS_INLINE prof_tctx_t * arena_prof_tctx_get(tsdn_t *tsdn, const void *ptr, alloc_ctx_t *alloc_ctx) { cassert(config_prof); @@ -28,7 +54,7 @@ arena_prof_tctx_get(tsdn_t *tsdn, const void *ptr, alloc_ctx_t *alloc_ctx) { } JEMALLOC_ALWAYS_INLINE void -arena_prof_tctx_set(tsdn_t *tsdn, const void *ptr, UNUSED size_t usize, +arena_prof_tctx_set(tsdn_t *tsdn, const void *ptr, size_t usize, alloc_ctx_t *alloc_ctx, prof_tctx_t *tctx) { cassert(config_prof); assert(ptr != NULL); @@ -47,7 +73,7 @@ arena_prof_tctx_set(tsdn_t *tsdn, const void *ptr, UNUSED size_t usize, } static inline void -arena_prof_tctx_reset(tsdn_t *tsdn, const void *ptr, UNUSED prof_tctx_t *tctx) { +arena_prof_tctx_reset(tsdn_t *tsdn, const void *ptr, prof_tctx_t *tctx) { cassert(config_prof); assert(ptr != NULL); @@ -57,6 +83,32 @@ arena_prof_tctx_reset(tsdn_t *tsdn, const void *ptr, UNUSED prof_tctx_t *tctx) { large_prof_tctx_reset(tsdn, extent); } +JEMALLOC_ALWAYS_INLINE nstime_t +arena_prof_alloc_time_get(tsdn_t *tsdn, const void *ptr, + alloc_ctx_t *alloc_ctx) { + cassert(config_prof); + assert(ptr != NULL); + + extent_t *extent = iealloc(tsdn, ptr); + /* + * Unlike arena_prof_prof_tctx_{get, set}, we only call this once we're + * sure we have a sampled allocation. + */ + assert(!extent_slab_get(extent)); + return large_prof_alloc_time_get(extent); +} + +JEMALLOC_ALWAYS_INLINE void +arena_prof_alloc_time_set(tsdn_t *tsdn, const void *ptr, alloc_ctx_t *alloc_ctx, + nstime_t t) { + cassert(config_prof); + assert(ptr != NULL); + + extent_t *extent = iealloc(tsdn, ptr); + assert(!extent_slab_get(extent)); + large_prof_alloc_time_set(extent, t); +} + JEMALLOC_ALWAYS_INLINE void arena_decay_ticks(tsdn_t *tsdn, arena_t *arena, unsigned nticks) { tsd_t *tsd; @@ -83,14 +135,33 @@ arena_decay_tick(tsdn_t *tsdn, arena_t *arena) { arena_decay_ticks(tsdn, arena, 1); } +/* Purge a single extent to retained / unmapped directly. */ +JEMALLOC_ALWAYS_INLINE void +arena_decay_extent(tsdn_t *tsdn,arena_t *arena, extent_hooks_t **r_extent_hooks, + extent_t *extent) { + size_t extent_size = extent_size_get(extent); + extent_dalloc_wrapper(tsdn, arena, + r_extent_hooks, extent); + if (config_stats) { + /* Update stats accordingly. */ + arena_stats_lock(tsdn, &arena->stats); + arena_stats_add_u64(tsdn, &arena->stats, + &arena->decay_dirty.stats->nmadvise, 1); + arena_stats_add_u64(tsdn, &arena->stats, + &arena->decay_dirty.stats->purged, extent_size >> LG_PAGE); + arena_stats_sub_zu(tsdn, &arena->stats, &arena->stats.mapped, + extent_size); + arena_stats_unlock(tsdn, &arena->stats); + } +} + JEMALLOC_ALWAYS_INLINE void * arena_malloc(tsdn_t *tsdn, arena_t *arena, size_t size, szind_t ind, bool zero, tcache_t *tcache, bool slow_path) { assert(!tsdn_null(tsdn) || tcache == NULL); - assert(size != 0); if (likely(tcache != NULL)) { - if (likely(size <= SMALL_MAXCLASS)) { + if (likely(size <= SC_SMALL_MAXCLASS)) { return tcache_alloc_small(tsdn_tsd(tsdn), arena, tcache, size, ind, zero, slow_path); } @@ -119,7 +190,7 @@ arena_salloc(tsdn_t *tsdn, const void *ptr) { szind_t szind = rtree_szind_read(tsdn, &extents_rtree, rtree_ctx, (uintptr_t)ptr, true); - assert(szind != NSIZES); + assert(szind != SC_NSIZES); return sz_index2size(szind); } @@ -152,12 +223,22 @@ arena_vsalloc(tsdn_t *tsdn, const void *ptr) { /* Only slab members should be looked up via interior pointers. */ assert(extent_addr_get(extent) == ptr || extent_slab_get(extent)); - assert(szind != NSIZES); + assert(szind != SC_NSIZES); return sz_index2size(szind); } static inline void +arena_dalloc_large_no_tcache(tsdn_t *tsdn, void *ptr, szind_t szind) { + if (config_prof && unlikely(szind < SC_NBINS)) { + arena_dalloc_promoted(tsdn, ptr, NULL, true); + } else { + extent_t *extent = iealloc(tsdn, ptr); + large_dalloc(tsdn, extent); + } +} + +static inline void arena_dalloc_no_tcache(tsdn_t *tsdn, void *ptr) { assert(ptr != NULL); @@ -173,7 +254,7 @@ arena_dalloc_no_tcache(tsdn_t *tsdn, void *ptr) { extent_t *extent = rtree_extent_read(tsdn, &extents_rtree, rtree_ctx, (uintptr_t)ptr, true); assert(szind == extent_szind_get(extent)); - assert(szind < NSIZES); + assert(szind < SC_NSIZES); assert(slab == extent_slab_get(extent)); } @@ -181,6 +262,21 @@ arena_dalloc_no_tcache(tsdn_t *tsdn, void *ptr) { /* Small allocation. */ arena_dalloc_small(tsdn, ptr); } else { + arena_dalloc_large_no_tcache(tsdn, ptr, szind); + } +} + +JEMALLOC_ALWAYS_INLINE void +arena_dalloc_large(tsdn_t *tsdn, void *ptr, tcache_t *tcache, szind_t szind, + bool slow_path) { + if (szind < nhbins) { + if (config_prof && unlikely(szind < SC_NBINS)) { + arena_dalloc_promoted(tsdn, ptr, tcache, slow_path); + } else { + tcache_dalloc_large(tsdn_tsd(tsdn), tcache, ptr, szind, + slow_path); + } + } else { extent_t *extent = iealloc(tsdn, ptr); large_dalloc(tsdn, extent); } @@ -203,7 +299,7 @@ arena_dalloc(tsdn_t *tsdn, void *ptr, tcache_t *tcache, if (alloc_ctx != NULL) { szind = alloc_ctx->szind; slab = alloc_ctx->slab; - assert(szind != NSIZES); + assert(szind != SC_NSIZES); } else { rtree_ctx = tsd_rtree_ctx(tsdn_tsd(tsdn)); rtree_szind_slab_read(tsdn, &extents_rtree, rtree_ctx, @@ -215,7 +311,7 @@ arena_dalloc(tsdn_t *tsdn, void *ptr, tcache_t *tcache, extent_t *extent = rtree_extent_read(tsdn, &extents_rtree, rtree_ctx, (uintptr_t)ptr, true); assert(szind == extent_szind_get(extent)); - assert(szind < NSIZES); + assert(szind < SC_NSIZES); assert(slab == extent_slab_get(extent)); } @@ -224,25 +320,14 @@ arena_dalloc(tsdn_t *tsdn, void *ptr, tcache_t *tcache, tcache_dalloc_small(tsdn_tsd(tsdn), tcache, ptr, szind, slow_path); } else { - if (szind < nhbins) { - if (config_prof && unlikely(szind < NBINS)) { - arena_dalloc_promoted(tsdn, ptr, tcache, - slow_path); - } else { - tcache_dalloc_large(tsdn_tsd(tsdn), tcache, ptr, - szind, slow_path); - } - } else { - extent_t *extent = iealloc(tsdn, ptr); - large_dalloc(tsdn, extent); - } + arena_dalloc_large(tsdn, ptr, tcache, szind, slow_path); } } static inline void arena_sdalloc_no_tcache(tsdn_t *tsdn, void *ptr, size_t size) { assert(ptr != NULL); - assert(size <= LARGE_MAXCLASS); + assert(size <= SC_LARGE_MAXCLASS); szind_t szind; bool slab; @@ -252,7 +337,7 @@ arena_sdalloc_no_tcache(tsdn_t *tsdn, void *ptr, size_t size) { * object, so base szind and slab on the given size. */ szind = sz_size2index(size); - slab = (szind < NBINS); + slab = (szind < SC_NBINS); } if ((config_prof && opt_prof) || config_debug) { @@ -264,7 +349,7 @@ arena_sdalloc_no_tcache(tsdn_t *tsdn, void *ptr, size_t size) { (uintptr_t)ptr, true, &szind, &slab); assert(szind == sz_size2index(size)); - assert((config_prof && opt_prof) || slab == (szind < NBINS)); + assert((config_prof && opt_prof) || slab == (szind < SC_NBINS)); if (config_debug) { extent_t *extent = rtree_extent_read(tsdn, @@ -278,8 +363,7 @@ arena_sdalloc_no_tcache(tsdn_t *tsdn, void *ptr, size_t size) { /* Small allocation. */ arena_dalloc_small(tsdn, ptr); } else { - extent_t *extent = iealloc(tsdn, ptr); - large_dalloc(tsdn, extent); + arena_dalloc_large_no_tcache(tsdn, ptr, szind); } } @@ -288,7 +372,7 @@ arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, alloc_ctx_t *alloc_ctx, bool slow_path) { assert(!tsdn_null(tsdn) || tcache == NULL); assert(ptr != NULL); - assert(size <= LARGE_MAXCLASS); + assert(size <= SC_LARGE_MAXCLASS); if (unlikely(tcache == NULL)) { arena_sdalloc_no_tcache(tsdn, ptr, size); @@ -297,7 +381,7 @@ arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, szind_t szind; bool slab; - UNUSED alloc_ctx_t local_ctx; + alloc_ctx_t local_ctx; if (config_prof && opt_prof) { if (alloc_ctx == NULL) { /* Uncommon case and should be a static check. */ @@ -318,7 +402,7 @@ arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, * object, so base szind and slab on the given size. */ szind = sz_size2index(size); - slab = (szind < NBINS); + slab = (szind < SC_NBINS); } if (config_debug) { @@ -336,18 +420,7 @@ arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, tcache_dalloc_small(tsdn_tsd(tsdn), tcache, ptr, szind, slow_path); } else { - if (szind < nhbins) { - if (config_prof && unlikely(szind < NBINS)) { - arena_dalloc_promoted(tsdn, ptr, tcache, - slow_path); - } else { - tcache_dalloc_large(tsdn_tsd(tsdn), - tcache, ptr, szind, slow_path); - } - } else { - extent_t *extent = iealloc(tsdn, ptr); - large_dalloc(tsdn, extent); - } + arena_dalloc_large(tsdn, ptr, tcache, szind, slow_path); } } diff --git a/deps/jemalloc/include/jemalloc/internal/arena_stats.h b/deps/jemalloc/include/jemalloc/internal/arena_stats.h index 5f3dca8b1..23949ed92 100644 --- a/deps/jemalloc/include/jemalloc/internal/arena_stats.h +++ b/deps/jemalloc/include/jemalloc/internal/arena_stats.h @@ -4,7 +4,9 @@ #include "jemalloc/internal/atomic.h" #include "jemalloc/internal/mutex.h" #include "jemalloc/internal/mutex_prof.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" + +JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS /* * In those architectures that support 64-bit atomics, we use atomic updates for @@ -33,6 +35,13 @@ struct arena_stats_large_s { * periodically merges into this counter. */ arena_stats_u64_t nrequests; /* Partially derived. */ + /* + * Number of tcache fills / flushes for large (similarly, periodically + * merged). Note that there is no large tcache batch-fill currently + * (i.e. only fill 1 at a time); however flush may be batched. + */ + arena_stats_u64_t nfills; /* Partially derived. */ + arena_stats_u64_t nflushes; /* Partially derived. */ /* Current number of allocations of this size class. */ size_t curlextents; /* Derived. */ @@ -48,6 +57,22 @@ struct arena_stats_decay_s { arena_stats_u64_t purged; }; +typedef struct arena_stats_extents_s arena_stats_extents_t; +struct arena_stats_extents_s { + /* + * Stats for a given index in the range [0, SC_NPSIZES] in an extents_t. + * We track both bytes and # of extents: two extents in the same bucket + * may have different sizes if adjacent size classes differ by more than + * a page, so bytes cannot always be derived from # of extents. + */ + atomic_zu_t ndirty; + atomic_zu_t dirty_bytes; + atomic_zu_t nmuzzy; + atomic_zu_t muzzy_bytes; + atomic_zu_t nretained; + atomic_zu_t retained_bytes; +}; + /* * Arena stats. Note that fields marked "derived" are not directly maintained * within the arena code; rather their values are derived during stats merge @@ -69,6 +94,9 @@ struct arena_stats_s { */ atomic_zu_t retained; /* Derived. */ + /* Number of extent_t structs allocated by base, but not being used. */ + atomic_zu_t extent_avail; + arena_stats_decay_t decay_dirty; arena_stats_decay_t decay_muzzy; @@ -80,22 +108,27 @@ struct arena_stats_s { atomic_zu_t allocated_large; /* Derived. */ arena_stats_u64_t nmalloc_large; /* Derived. */ arena_stats_u64_t ndalloc_large; /* Derived. */ + arena_stats_u64_t nfills_large; /* Derived. */ + arena_stats_u64_t nflushes_large; /* Derived. */ arena_stats_u64_t nrequests_large; /* Derived. */ + /* VM space had to be leaked (undocumented). Normally 0. */ + atomic_zu_t abandoned_vm; + /* Number of bytes cached in tcache associated with this arena. */ atomic_zu_t tcache_bytes; /* Derived. */ mutex_prof_data_t mutex_prof_data[mutex_prof_num_arena_mutexes]; /* One element for each large size class. */ - arena_stats_large_t lstats[NSIZES - NBINS]; + arena_stats_large_t lstats[SC_NSIZES - SC_NBINS]; /* Arena uptime. */ nstime_t uptime; }; static inline bool -arena_stats_init(UNUSED tsdn_t *tsdn, arena_stats_t *arena_stats) { +arena_stats_init(tsdn_t *tsdn, arena_stats_t *arena_stats) { if (config_debug) { for (size_t i = 0; i < sizeof(arena_stats_t); i++) { assert(((char *)arena_stats)[i] == 0); @@ -147,11 +180,11 @@ arena_stats_add_u64(tsdn_t *tsdn, arena_stats_t *arena_stats, #endif } -UNUSED static inline void +static inline void arena_stats_sub_u64(tsdn_t *tsdn, arena_stats_t *arena_stats, arena_stats_u64_t *p, uint64_t x) { #ifdef JEMALLOC_ATOMIC_U64 - UNUSED uint64_t r = atomic_fetch_sub_u64(p, x, ATOMIC_RELAXED); + uint64_t r = atomic_fetch_sub_u64(p, x, ATOMIC_RELAXED); assert(r - x <= r); #else malloc_mutex_assert_owner(tsdn, &arena_stats->mtx); @@ -176,7 +209,8 @@ arena_stats_accum_u64(arena_stats_u64_t *dst, uint64_t src) { } static inline size_t -arena_stats_read_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, atomic_zu_t *p) { +arena_stats_read_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, + atomic_zu_t *p) { #ifdef JEMALLOC_ATOMIC_U64 return atomic_load_zu(p, ATOMIC_RELAXED); #else @@ -186,8 +220,8 @@ arena_stats_read_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, atomic_zu_t *p) { } static inline void -arena_stats_add_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, atomic_zu_t *p, - size_t x) { +arena_stats_add_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, + atomic_zu_t *p, size_t x) { #ifdef JEMALLOC_ATOMIC_U64 atomic_fetch_add_zu(p, x, ATOMIC_RELAXED); #else @@ -198,10 +232,10 @@ arena_stats_add_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, atomic_zu_t *p, } static inline void -arena_stats_sub_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, atomic_zu_t *p, - size_t x) { +arena_stats_sub_zu(tsdn_t *tsdn, arena_stats_t *arena_stats, + atomic_zu_t *p, size_t x) { #ifdef JEMALLOC_ATOMIC_U64 - UNUSED size_t r = atomic_fetch_sub_zu(p, x, ATOMIC_RELAXED); + size_t r = atomic_fetch_sub_zu(p, x, ATOMIC_RELAXED); assert(r - x <= r); #else malloc_mutex_assert_owner(tsdn, &arena_stats->mtx); @@ -218,11 +252,12 @@ arena_stats_accum_zu(atomic_zu_t *dst, size_t src) { } static inline void -arena_stats_large_nrequests_add(tsdn_t *tsdn, arena_stats_t *arena_stats, +arena_stats_large_flush_nrequests_add(tsdn_t *tsdn, arena_stats_t *arena_stats, szind_t szind, uint64_t nrequests) { arena_stats_lock(tsdn, arena_stats); - arena_stats_add_u64(tsdn, arena_stats, &arena_stats->lstats[szind - - NBINS].nrequests, nrequests); + arena_stats_large_t *lstats = &arena_stats->lstats[szind - SC_NBINS]; + arena_stats_add_u64(tsdn, arena_stats, &lstats->nrequests, nrequests); + arena_stats_add_u64(tsdn, arena_stats, &lstats->nflushes, 1); arena_stats_unlock(tsdn, arena_stats); } @@ -233,5 +268,4 @@ arena_stats_mapped_add(tsdn_t *tsdn, arena_stats_t *arena_stats, size_t size) { arena_stats_unlock(tsdn, arena_stats); } - #endif /* JEMALLOC_INTERNAL_ARENA_STATS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/arena_structs_b.h b/deps/jemalloc/include/jemalloc/internal/arena_structs_b.h index 38bc95962..eeab57fd6 100644 --- a/deps/jemalloc/include/jemalloc/internal/arena_structs_b.h +++ b/deps/jemalloc/include/jemalloc/internal/arena_structs_b.h @@ -10,7 +10,7 @@ #include "jemalloc/internal/mutex.h" #include "jemalloc/internal/nstime.h" #include "jemalloc/internal/ql.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/smoothstep.h" #include "jemalloc/internal/ticker.h" @@ -90,6 +90,9 @@ struct arena_s { */ atomic_u_t nthreads[2]; + /* Next bin shard for binding new threads. Synchronization: atomic. */ + atomic_u_t binshard_next; + /* * When percpu_arena is enabled, to amortize the cost of reading / * updating the current CPU id, track the most recent thread accessing @@ -113,7 +116,6 @@ struct arena_s { /* Synchronization: internal. */ prof_accum_t prof_accum; - uint64_t prof_accumbytes; /* * PRNG state for cache index randomization of large allocation base @@ -196,6 +198,7 @@ struct arena_s { * Synchronization: extent_avail_mtx. */ extent_tree_t extent_avail; + atomic_zu_t extent_avail_cnt; malloc_mutex_t extent_avail_mtx; /* @@ -203,7 +206,7 @@ struct arena_s { * * Synchronization: internal. */ - bin_t bins[NBINS]; + bins_t bins[SC_NBINS]; /* * Base allocator, from which arena metadata are allocated. diff --git a/deps/jemalloc/include/jemalloc/internal/arena_types.h b/deps/jemalloc/include/jemalloc/internal/arena_types.h index 70001b5f1..624937e4f 100644 --- a/deps/jemalloc/include/jemalloc/internal/arena_types.h +++ b/deps/jemalloc/include/jemalloc/internal/arena_types.h @@ -1,13 +1,15 @@ #ifndef JEMALLOC_INTERNAL_ARENA_TYPES_H #define JEMALLOC_INTERNAL_ARENA_TYPES_H +#include "jemalloc/internal/sc.h" + /* Maximum number of regions in one slab. */ -#define LG_SLAB_MAXREGS (LG_PAGE - LG_TINY_MIN) +#define LG_SLAB_MAXREGS (LG_PAGE - SC_LG_TINY_MIN) #define SLAB_MAXREGS (1U << LG_SLAB_MAXREGS) /* Default decay times in milliseconds. */ #define DIRTY_DECAY_MS_DEFAULT ZD(10 * 1000) -#define MUZZY_DECAY_MS_DEFAULT ZD(10 * 1000) +#define MUZZY_DECAY_MS_DEFAULT (0) /* Number of event ticks between time checks. */ #define DECAY_NTICKS_PER_UPDATE 1000 @@ -40,4 +42,10 @@ typedef enum { #define PERCPU_ARENA_ENABLED(m) ((m) >= percpu_arena_mode_enabled_base) #define PERCPU_ARENA_DEFAULT percpu_arena_disabled +/* + * When allocation_size >= oversize_threshold, use the dedicated huge arena + * (unless have explicitly spicified arena index). 0 disables the feature. + */ +#define OVERSIZE_THRESHOLD_DEFAULT (8 << 20) + #endif /* JEMALLOC_INTERNAL_ARENA_TYPES_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/atomic.h b/deps/jemalloc/include/jemalloc/internal/atomic.h index adadb1a3a..a76f54cee 100644 --- a/deps/jemalloc/include/jemalloc/internal/atomic.h +++ b/deps/jemalloc/include/jemalloc/internal/atomic.h @@ -1,12 +1,19 @@ #ifndef JEMALLOC_INTERNAL_ATOMIC_H #define JEMALLOC_INTERNAL_ATOMIC_H -#define ATOMIC_INLINE static inline +#define ATOMIC_INLINE JEMALLOC_ALWAYS_INLINE +#define JEMALLOC_U8_ATOMICS #if defined(JEMALLOC_GCC_ATOMIC_ATOMICS) # include "jemalloc/internal/atomic_gcc_atomic.h" +# if !defined(JEMALLOC_GCC_U8_ATOMIC_ATOMICS) +# undef JEMALLOC_U8_ATOMICS +# endif #elif defined(JEMALLOC_GCC_SYNC_ATOMICS) # include "jemalloc/internal/atomic_gcc_sync.h" +# if !defined(JEMALLOC_GCC_U8_SYNC_ATOMICS) +# undef JEMALLOC_U8_ATOMICS +# endif #elif defined(_MSC_VER) # include "jemalloc/internal/atomic_msvc.h" #elif defined(JEMALLOC_C11_ATOMICS) @@ -66,6 +73,8 @@ JEMALLOC_GENERATE_INT_ATOMICS(size_t, zu, LG_SIZEOF_PTR) JEMALLOC_GENERATE_INT_ATOMICS(ssize_t, zd, LG_SIZEOF_PTR) +JEMALLOC_GENERATE_INT_ATOMICS(uint8_t, u8, 0) + JEMALLOC_GENERATE_INT_ATOMICS(uint32_t, u32, 2) #ifdef JEMALLOC_ATOMIC_U64 diff --git a/deps/jemalloc/include/jemalloc/internal/atomic_gcc_atomic.h b/deps/jemalloc/include/jemalloc/internal/atomic_gcc_atomic.h index 6b73a14f8..471515e82 100644 --- a/deps/jemalloc/include/jemalloc/internal/atomic_gcc_atomic.h +++ b/deps/jemalloc/include/jemalloc/internal/atomic_gcc_atomic.h @@ -67,7 +67,8 @@ atomic_exchange_##short_type(atomic_##short_type##_t *a, type val, \ \ ATOMIC_INLINE bool \ atomic_compare_exchange_weak_##short_type(atomic_##short_type##_t *a, \ - type *expected, type desired, atomic_memory_order_t success_mo, \ + UNUSED type *expected, type desired, \ + atomic_memory_order_t success_mo, \ atomic_memory_order_t failure_mo) { \ return __atomic_compare_exchange(&a->repr, expected, &desired, \ true, atomic_enum_to_builtin(success_mo), \ @@ -76,7 +77,8 @@ atomic_compare_exchange_weak_##short_type(atomic_##short_type##_t *a, \ \ ATOMIC_INLINE bool \ atomic_compare_exchange_strong_##short_type(atomic_##short_type##_t *a, \ - type *expected, type desired, atomic_memory_order_t success_mo, \ + UNUSED type *expected, type desired, \ + atomic_memory_order_t success_mo, \ atomic_memory_order_t failure_mo) { \ return __atomic_compare_exchange(&a->repr, expected, &desired, \ false, \ diff --git a/deps/jemalloc/include/jemalloc/internal/atomic_gcc_sync.h b/deps/jemalloc/include/jemalloc/internal/atomic_gcc_sync.h index 30846e4d2..e02b7cbe3 100644 --- a/deps/jemalloc/include/jemalloc/internal/atomic_gcc_sync.h +++ b/deps/jemalloc/include/jemalloc/internal/atomic_gcc_sync.h @@ -27,8 +27,10 @@ atomic_fence(atomic_memory_order_t mo) { asm volatile("" ::: "memory"); # if defined(__i386__) || defined(__x86_64__) /* This is implicit on x86. */ -# elif defined(__ppc__) +# elif defined(__ppc64__) asm volatile("lwsync"); +# elif defined(__ppc__) + asm volatile("sync"); # elif defined(__sparc__) && defined(__arch64__) if (mo == atomic_memory_order_acquire) { asm volatile("membar #LoadLoad | #LoadStore"); @@ -113,8 +115,8 @@ atomic_store_##short_type(atomic_##short_type##_t *a, \ } \ \ ATOMIC_INLINE type \ -atomic_exchange_##short_type(atomic_##short_type##_t *a, type val, \ - atomic_memory_order_t mo) { \ +atomic_exchange_##short_type(atomic_##short_type##_t *a, type val, \ + atomic_memory_order_t mo) { \ /* \ * Because of FreeBSD, we care about gcc 4.2, which doesn't have\ * an atomic exchange builtin. We fake it with a CAS loop. \ @@ -129,8 +131,9 @@ atomic_exchange_##short_type(atomic_##short_type##_t *a, type val, \ \ ATOMIC_INLINE bool \ atomic_compare_exchange_weak_##short_type(atomic_##short_type##_t *a, \ - type *expected, type desired, atomic_memory_order_t success_mo, \ - atomic_memory_order_t failure_mo) { \ + type *expected, type desired, \ + atomic_memory_order_t success_mo, \ + atomic_memory_order_t failure_mo) { \ type prev = __sync_val_compare_and_swap(&a->repr, *expected, \ desired); \ if (prev == *expected) { \ @@ -142,8 +145,9 @@ atomic_compare_exchange_weak_##short_type(atomic_##short_type##_t *a, \ } \ ATOMIC_INLINE bool \ atomic_compare_exchange_strong_##short_type(atomic_##short_type##_t *a, \ - type *expected, type desired, atomic_memory_order_t success_mo, \ - atomic_memory_order_t failure_mo) { \ + type *expected, type desired, \ + atomic_memory_order_t success_mo, \ + atomic_memory_order_t failure_mo) { \ type prev = __sync_val_compare_and_swap(&a->repr, *expected, \ desired); \ if (prev == *expected) { \ diff --git a/deps/jemalloc/include/jemalloc/internal/background_thread_externs.h b/deps/jemalloc/include/jemalloc/internal/background_thread_externs.h index 3209aa49f..0f997e18b 100644 --- a/deps/jemalloc/include/jemalloc/internal/background_thread_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/background_thread_externs.h @@ -8,7 +8,6 @@ extern atomic_b_t background_thread_enabled_state; extern size_t n_background_threads; extern size_t max_background_threads; extern background_thread_info_t *background_thread_info; -extern bool can_enable_background_thread; bool background_thread_create(tsd_t *tsd, unsigned arena_ind); bool background_threads_enable(tsd_t *tsd); diff --git a/deps/jemalloc/include/jemalloc/internal/background_thread_inlines.h b/deps/jemalloc/include/jemalloc/internal/background_thread_inlines.h index ef50231e8..f85e86fa3 100644 --- a/deps/jemalloc/include/jemalloc/internal/background_thread_inlines.h +++ b/deps/jemalloc/include/jemalloc/internal/background_thread_inlines.h @@ -15,7 +15,12 @@ background_thread_enabled_set(tsdn_t *tsdn, bool state) { JEMALLOC_ALWAYS_INLINE background_thread_info_t * arena_background_thread_info_get(arena_t *arena) { unsigned arena_ind = arena_ind_get(arena); - return &background_thread_info[arena_ind % ncpus]; + return &background_thread_info[arena_ind % max_background_threads]; +} + +JEMALLOC_ALWAYS_INLINE background_thread_info_t * +background_thread_info_get(size_t ind) { + return &background_thread_info[ind % max_background_threads]; } JEMALLOC_ALWAYS_INLINE uint64_t diff --git a/deps/jemalloc/include/jemalloc/internal/background_thread_structs.h b/deps/jemalloc/include/jemalloc/internal/background_thread_structs.h index c1107dfe9..c02aa434c 100644 --- a/deps/jemalloc/include/jemalloc/internal/background_thread_structs.h +++ b/deps/jemalloc/include/jemalloc/internal/background_thread_structs.h @@ -9,6 +9,7 @@ #define BACKGROUND_THREAD_INDEFINITE_SLEEP UINT64_MAX #define MAX_BACKGROUND_THREAD_LIMIT MALLOCX_ARENA_LIMIT +#define DEFAULT_NUM_BACKGROUND_THREAD 4 typedef enum { background_thread_stopped, diff --git a/deps/jemalloc/include/jemalloc/internal/base_structs.h b/deps/jemalloc/include/jemalloc/internal/base_structs.h index 2102247ac..07f214eb2 100644 --- a/deps/jemalloc/include/jemalloc/internal/base_structs.h +++ b/deps/jemalloc/include/jemalloc/internal/base_structs.h @@ -3,7 +3,7 @@ #include "jemalloc/internal/jemalloc_internal_types.h" #include "jemalloc/internal/mutex.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" /* Embedded at the beginning of every block of base-managed virtual memory. */ struct base_block_s { @@ -46,7 +46,7 @@ struct base_s { base_block_t *blocks; /* Heap of extents that track unused trailing space within blocks. */ - extent_heap_t avail[NSIZES]; + extent_heap_t avail[SC_NSIZES]; /* Stats, only maintained if config_stats. */ size_t allocated; diff --git a/deps/jemalloc/include/jemalloc/internal/bin.h b/deps/jemalloc/include/jemalloc/internal/bin.h index 9b416ada7..8547e8930 100644 --- a/deps/jemalloc/include/jemalloc/internal/bin.h +++ b/deps/jemalloc/include/jemalloc/internal/bin.h @@ -1,10 +1,12 @@ #ifndef JEMALLOC_INTERNAL_BIN_H #define JEMALLOC_INTERNAL_BIN_H +#include "jemalloc/internal/bin_stats.h" +#include "jemalloc/internal/bin_types.h" #include "jemalloc/internal/extent_types.h" #include "jemalloc/internal/extent_structs.h" #include "jemalloc/internal/mutex.h" -#include "jemalloc/internal/bin_stats.h" +#include "jemalloc/internal/sc.h" /* * A bin contains a set of extents that are currently being used for slab @@ -41,6 +43,9 @@ struct bin_info_s { /* Total number of regions in a slab for this bin's size class. */ uint32_t nregs; + /* Number of sharded bins in each arena for this size class. */ + uint32_t n_shards; + /* * Metadata used to manipulate bitmaps for slabs associated with this * bin. @@ -48,8 +53,7 @@ struct bin_info_s { bitmap_info_t bitmap_info; }; -extern const bin_info_t bin_infos[NBINS]; - +extern bin_info_t bin_infos[SC_NBINS]; typedef struct bin_s bin_t; struct bin_s { @@ -78,6 +82,18 @@ struct bin_s { bin_stats_t stats; }; +/* A set of sharded bins of the same size class. */ +typedef struct bins_s bins_t; +struct bins_s { + /* Sharded bins. Dynamically sized. */ + bin_t *bin_shards; +}; + +void bin_shard_sizes_boot(unsigned bin_shards[SC_NBINS]); +bool bin_update_shard_size(unsigned bin_shards[SC_NBINS], size_t start_size, + size_t end_size, size_t nshards); +void bin_boot(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS]); + /* Initializes a bin to empty. Returns true on error. */ bool bin_init(bin_t *bin); @@ -90,7 +106,7 @@ void bin_postfork_child(tsdn_t *tsdn, bin_t *bin); static inline void bin_stats_merge(tsdn_t *tsdn, bin_stats_t *dst_bin_stats, bin_t *bin) { malloc_mutex_lock(tsdn, &bin->lock); - malloc_mutex_prof_read(tsdn, &dst_bin_stats->mutex_data, &bin->lock); + malloc_mutex_prof_accum(tsdn, &dst_bin_stats->mutex_data, &bin->lock); dst_bin_stats->nmalloc += bin->stats.nmalloc; dst_bin_stats->ndalloc += bin->stats.ndalloc; dst_bin_stats->nrequests += bin->stats.nrequests; @@ -100,6 +116,7 @@ bin_stats_merge(tsdn_t *tsdn, bin_stats_t *dst_bin_stats, bin_t *bin) { dst_bin_stats->nslabs += bin->stats.nslabs; dst_bin_stats->reslabs += bin->stats.reslabs; dst_bin_stats->curslabs += bin->stats.curslabs; + dst_bin_stats->nonfull_slabs += bin->stats.nonfull_slabs; malloc_mutex_unlock(tsdn, &bin->lock); } diff --git a/deps/jemalloc/include/jemalloc/internal/bin_stats.h b/deps/jemalloc/include/jemalloc/internal/bin_stats.h index 86e673ec4..d04519c82 100644 --- a/deps/jemalloc/include/jemalloc/internal/bin_stats.h +++ b/deps/jemalloc/include/jemalloc/internal/bin_stats.h @@ -45,6 +45,9 @@ struct bin_stats_s { /* Current number of slabs in this bin. */ size_t curslabs; + /* Current size of nonfull slabs heap in this bin. */ + size_t nonfull_slabs; + mutex_prof_data_t mutex_data; }; diff --git a/deps/jemalloc/include/jemalloc/internal/bin_types.h b/deps/jemalloc/include/jemalloc/internal/bin_types.h new file mode 100644 index 000000000..3533606b9 --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/bin_types.h @@ -0,0 +1,17 @@ +#ifndef JEMALLOC_INTERNAL_BIN_TYPES_H +#define JEMALLOC_INTERNAL_BIN_TYPES_H + +#include "jemalloc/internal/sc.h" + +#define BIN_SHARDS_MAX (1 << EXTENT_BITS_BINSHARD_WIDTH) +#define N_BIN_SHARDS_DEFAULT 1 + +/* Used in TSD static initializer only. Real init in arena_bind(). */ +#define TSD_BINSHARDS_ZERO_INITIALIZER {{UINT8_MAX}} + +typedef struct tsd_binshards_s tsd_binshards_t; +struct tsd_binshards_s { + uint8_t binshard[SC_NBINS]; +}; + +#endif /* JEMALLOC_INTERNAL_BIN_TYPES_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/bit_util.h b/deps/jemalloc/include/jemalloc/internal/bit_util.h index 8d078a8a3..c045eb868 100644 --- a/deps/jemalloc/include/jemalloc/internal/bit_util.h +++ b/deps/jemalloc/include/jemalloc/internal/bit_util.h @@ -27,6 +27,25 @@ ffs_u(unsigned bitmap) { return JEMALLOC_INTERNAL_FFS(bitmap); } +#ifdef JEMALLOC_INTERNAL_POPCOUNTL +BIT_UTIL_INLINE unsigned +popcount_lu(unsigned long bitmap) { + return JEMALLOC_INTERNAL_POPCOUNTL(bitmap); +} +#endif + +/* + * Clears first unset bit in bitmap, and returns + * place of bit. bitmap *must not* be 0. + */ + +BIT_UTIL_INLINE size_t +cfs_lu(unsigned long* bitmap) { + size_t bit = ffs_lu(*bitmap) - 1; + *bitmap ^= ZU(1) << bit; + return bit; +} + BIT_UTIL_INLINE unsigned ffs_zu(size_t bitmap) { #if LG_SIZEOF_PTR == LG_SIZEOF_INT @@ -63,6 +82,22 @@ ffs_u32(uint32_t bitmap) { BIT_UTIL_INLINE uint64_t pow2_ceil_u64(uint64_t x) { +#if (defined(__amd64__) || defined(__x86_64__) || defined(JEMALLOC_HAVE_BUILTIN_CLZ)) + if(unlikely(x <= 1)) { + return x; + } + size_t msb_on_index; +#if (defined(__amd64__) || defined(__x86_64__)) + asm ("bsrq %1, %0" + : "=r"(msb_on_index) // Outputs. + : "r"(x-1) // Inputs. + ); +#elif (defined(JEMALLOC_HAVE_BUILTIN_CLZ)) + msb_on_index = (63 ^ __builtin_clzll(x - 1)); +#endif + assert(msb_on_index < 63); + return 1ULL << (msb_on_index + 1); +#else x--; x |= x >> 1; x |= x >> 2; @@ -72,10 +107,27 @@ pow2_ceil_u64(uint64_t x) { x |= x >> 32; x++; return x; +#endif } BIT_UTIL_INLINE uint32_t pow2_ceil_u32(uint32_t x) { +#if ((defined(__i386__) || defined(JEMALLOC_HAVE_BUILTIN_CLZ)) && (!defined(__s390__))) + if(unlikely(x <= 1)) { + return x; + } + size_t msb_on_index; +#if (defined(__i386__)) + asm ("bsr %1, %0" + : "=r"(msb_on_index) // Outputs. + : "r"(x-1) // Inputs. + ); +#elif (defined(JEMALLOC_HAVE_BUILTIN_CLZ)) + msb_on_index = (31 ^ __builtin_clz(x - 1)); +#endif + assert(msb_on_index < 31); + return 1U << (msb_on_index + 1); +#else x--; x |= x >> 1; x |= x >> 2; @@ -84,6 +136,7 @@ pow2_ceil_u32(uint32_t x) { x |= x >> 16; x++; return x; +#endif } /* Compute the smallest power of 2 that is >= x. */ @@ -160,6 +213,27 @@ lg_floor(size_t x) { } #endif +BIT_UTIL_INLINE unsigned +lg_ceil(size_t x) { + return lg_floor(x) + ((x & (x - 1)) == 0 ? 0 : 1); +} + #undef BIT_UTIL_INLINE +/* A compile-time version of lg_floor and lg_ceil. */ +#define LG_FLOOR_1(x) 0 +#define LG_FLOOR_2(x) (x < (1ULL << 1) ? LG_FLOOR_1(x) : 1 + LG_FLOOR_1(x >> 1)) +#define LG_FLOOR_4(x) (x < (1ULL << 2) ? LG_FLOOR_2(x) : 2 + LG_FLOOR_2(x >> 2)) +#define LG_FLOOR_8(x) (x < (1ULL << 4) ? LG_FLOOR_4(x) : 4 + LG_FLOOR_4(x >> 4)) +#define LG_FLOOR_16(x) (x < (1ULL << 8) ? LG_FLOOR_8(x) : 8 + LG_FLOOR_8(x >> 8)) +#define LG_FLOOR_32(x) (x < (1ULL << 16) ? LG_FLOOR_16(x) : 16 + LG_FLOOR_16(x >> 16)) +#define LG_FLOOR_64(x) (x < (1ULL << 32) ? LG_FLOOR_32(x) : 32 + LG_FLOOR_32(x >> 32)) +#if LG_SIZEOF_PTR == 2 +# define LG_FLOOR(x) LG_FLOOR_32((x)) +#else +# define LG_FLOOR(x) LG_FLOOR_64((x)) +#endif + +#define LG_CEIL(x) (LG_FLOOR(x) + (((x) & ((x) - 1)) == 0 ? 0 : 1)) + #endif /* JEMALLOC_INTERNAL_BIT_UTIL_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/bitmap.h b/deps/jemalloc/include/jemalloc/internal/bitmap.h index ac990290a..c3f9cb490 100644 --- a/deps/jemalloc/include/jemalloc/internal/bitmap.h +++ b/deps/jemalloc/include/jemalloc/internal/bitmap.h @@ -3,18 +3,18 @@ #include "jemalloc/internal/arena_types.h" #include "jemalloc/internal/bit_util.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" typedef unsigned long bitmap_t; #define LG_SIZEOF_BITMAP LG_SIZEOF_LONG /* Maximum bitmap bit count is 2^LG_BITMAP_MAXBITS. */ -#if LG_SLAB_MAXREGS > LG_CEIL_NSIZES +#if LG_SLAB_MAXREGS > LG_CEIL(SC_NSIZES) /* Maximum bitmap bit count is determined by maximum regions per slab. */ # define LG_BITMAP_MAXBITS LG_SLAB_MAXREGS #else /* Maximum bitmap bit count is determined by number of extent size classes. */ -# define LG_BITMAP_MAXBITS LG_CEIL_NSIZES +# define LG_BITMAP_MAXBITS LG_CEIL(SC_NSIZES) #endif #define BITMAP_MAXBITS (ZU(1) << LG_BITMAP_MAXBITS) diff --git a/deps/jemalloc/include/jemalloc/internal/cache_bin.h b/deps/jemalloc/include/jemalloc/internal/cache_bin.h index 12f3ef2dd..d14556a3d 100644 --- a/deps/jemalloc/include/jemalloc/internal/cache_bin.h +++ b/deps/jemalloc/include/jemalloc/internal/cache_bin.h @@ -88,11 +88,21 @@ JEMALLOC_ALWAYS_INLINE void * cache_bin_alloc_easy(cache_bin_t *bin, bool *success) { void *ret; - if (unlikely(bin->ncached == 0)) { - bin->low_water = -1; - *success = false; - return NULL; + bin->ncached--; + + /* + * Check for both bin->ncached == 0 and ncached < low_water + * in a single branch. + */ + if (unlikely(bin->ncached <= bin->low_water)) { + bin->low_water = bin->ncached; + if (bin->ncached == -1) { + bin->ncached = 0; + *success = false; + return NULL; + } } + /* * success (instead of ret) should be checked upon the return of this * function. We avoid checking (ret == NULL) because there is never a @@ -101,14 +111,21 @@ cache_bin_alloc_easy(cache_bin_t *bin, bool *success) { * cacheline). */ *success = true; - ret = *(bin->avail - bin->ncached); - bin->ncached--; + ret = *(bin->avail - (bin->ncached + 1)); - if (unlikely(bin->ncached < bin->low_water)) { - bin->low_water = bin->ncached; + return ret; +} + +JEMALLOC_ALWAYS_INLINE bool +cache_bin_dalloc_easy(cache_bin_t *bin, cache_bin_info_t *bin_info, void *ptr) { + if (unlikely(bin->ncached == bin_info->ncached_max)) { + return false; } + assert(bin->ncached < bin_info->ncached_max); + bin->ncached++; + *(bin->avail - bin->ncached) = ptr; - return ret; + return true; } #endif /* JEMALLOC_INTERNAL_CACHE_BIN_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/ctl.h b/deps/jemalloc/include/jemalloc/internal/ctl.h index d927d9480..1d1aacc6f 100644 --- a/deps/jemalloc/include/jemalloc/internal/ctl.h +++ b/deps/jemalloc/include/jemalloc/internal/ctl.h @@ -5,7 +5,7 @@ #include "jemalloc/internal/malloc_io.h" #include "jemalloc/internal/mutex_prof.h" #include "jemalloc/internal/ql.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/stats.h" /* Maximum ctl tree depth. */ @@ -39,9 +39,12 @@ typedef struct ctl_arena_stats_s { uint64_t nmalloc_small; uint64_t ndalloc_small; uint64_t nrequests_small; + uint64_t nfills_small; + uint64_t nflushes_small; - bin_stats_t bstats[NBINS]; - arena_stats_large_t lstats[NSIZES - NBINS]; + bin_stats_t bstats[SC_NBINS]; + arena_stats_large_t lstats[SC_NSIZES - SC_NBINS]; + arena_stats_extents_t estats[SC_NPSIZES]; } ctl_arena_stats_t; typedef struct ctl_stats_s { diff --git a/deps/jemalloc/include/jemalloc/internal/emitter.h b/deps/jemalloc/include/jemalloc/internal/emitter.h index 3a2b2f7f2..542bc79c3 100644 --- a/deps/jemalloc/include/jemalloc/internal/emitter.h +++ b/deps/jemalloc/include/jemalloc/internal/emitter.h @@ -45,7 +45,9 @@ struct emitter_col_s { int int_val; unsigned unsigned_val; uint32_t uint32_val; + uint32_t uint32_t_val; uint64_t uint64_val; + uint64_t uint64_t_val; size_t size_val; ssize_t ssize_val; const char *str_val; @@ -60,17 +62,6 @@ struct emitter_row_s { ql_head(emitter_col_t) cols; }; -static inline void -emitter_row_init(emitter_row_t *row) { - ql_new(&row->cols); -} - -static inline void -emitter_col_init(emitter_col_t *col, emitter_row_t *row) { - ql_elm_new(col, link); - ql_tail_insert(&row->cols, col, link); -} - typedef struct emitter_s emitter_t; struct emitter_s { emitter_output_t output; @@ -80,18 +71,10 @@ struct emitter_s { int nesting_depth; /* True if we've already emitted a value at the given depth. */ bool item_at_depth; + /* True if we emitted a key and will emit corresponding value next. */ + bool emitted_key; }; -static inline void -emitter_init(emitter_t *emitter, emitter_output_t emitter_output, - void (*write_cb)(void *, const char *), void *cbopaque) { - emitter->output = emitter_output; - emitter->write_cb = write_cb; - emitter->cbopaque = cbopaque; - emitter->item_at_depth = false; - emitter->nesting_depth = 0; -} - /* Internal convenience function. Write to the emitter the given string. */ JEMALLOC_FORMAT_PRINTF(2, 3) static inline void @@ -103,22 +86,11 @@ emitter_printf(emitter_t *emitter, const char *format, ...) { va_end(ap); } -/* Write to the emitter the given string, but only in table mode. */ -JEMALLOC_FORMAT_PRINTF(2, 3) -static inline void -emitter_table_printf(emitter_t *emitter, const char *format, ...) { - if (emitter->output == emitter_output_table) { - va_list ap; - va_start(ap, format); - malloc_vcprintf(emitter->write_cb, emitter->cbopaque, format, ap); - va_end(ap); - } -} - -static inline void +static inline const char * JEMALLOC_FORMAT_ARG(3) emitter_gen_fmt(char *out_fmt, size_t out_size, const char *fmt_specifier, emitter_justify_t justify, int width) { size_t written; + fmt_specifier++; if (justify == emitter_justify_none) { written = malloc_snprintf(out_fmt, out_size, "%%%s", fmt_specifier); @@ -131,6 +103,7 @@ emitter_gen_fmt(char *out_fmt, size_t out_size, const char *fmt_specifier, } /* Only happens in case of bad format string, which *we* choose. */ assert(written < out_size); + return out_fmt; } /* @@ -156,26 +129,27 @@ emitter_print_value(emitter_t *emitter, emitter_justify_t justify, int width, char buf[BUF_SIZE]; #define EMIT_SIMPLE(type, format) \ - emitter_gen_fmt(fmt, FMT_SIZE, format, justify, width); \ - emitter_printf(emitter, fmt, *(const type *)value); \ + emitter_printf(emitter, \ + emitter_gen_fmt(fmt, FMT_SIZE, format, justify, width), \ + *(const type *)value); switch (value_type) { case emitter_type_bool: - emitter_gen_fmt(fmt, FMT_SIZE, "s", justify, width); - emitter_printf(emitter, fmt, *(const bool *)value ? - "true" : "false"); + emitter_printf(emitter, + emitter_gen_fmt(fmt, FMT_SIZE, "%s", justify, width), + *(const bool *)value ? "true" : "false"); break; case emitter_type_int: - EMIT_SIMPLE(int, "d") + EMIT_SIMPLE(int, "%d") break; case emitter_type_unsigned: - EMIT_SIMPLE(unsigned, "u") + EMIT_SIMPLE(unsigned, "%u") break; case emitter_type_ssize: - EMIT_SIMPLE(ssize_t, "zd") + EMIT_SIMPLE(ssize_t, "%zd") break; case emitter_type_size: - EMIT_SIMPLE(size_t, "zu") + EMIT_SIMPLE(size_t, "%zu") break; case emitter_type_string: str_written = malloc_snprintf(buf, BUF_SIZE, "\"%s\"", @@ -185,17 +159,17 @@ emitter_print_value(emitter_t *emitter, emitter_justify_t justify, int width, * anywhere near the fmt size. */ assert(str_written < BUF_SIZE); - emitter_gen_fmt(fmt, FMT_SIZE, "s", justify, width); - emitter_printf(emitter, fmt, buf); + emitter_printf(emitter, + emitter_gen_fmt(fmt, FMT_SIZE, "%s", justify, width), buf); break; case emitter_type_uint32: - EMIT_SIMPLE(uint32_t, FMTu32) + EMIT_SIMPLE(uint32_t, "%" FMTu32) break; case emitter_type_uint64: - EMIT_SIMPLE(uint64_t, FMTu64) + EMIT_SIMPLE(uint64_t, "%" FMTu64) break; case emitter_type_title: - EMIT_SIMPLE(char *const, "s"); + EMIT_SIMPLE(char *const, "%s"); break; default: unreachable(); @@ -235,201 +209,278 @@ emitter_indent(emitter_t *emitter) { static inline void emitter_json_key_prefix(emitter_t *emitter) { + if (emitter->emitted_key) { + emitter->emitted_key = false; + return; + } emitter_printf(emitter, "%s\n", emitter->item_at_depth ? "," : ""); emitter_indent(emitter); } +/******************************************************************************/ +/* Public functions for emitter_t. */ + static inline void -emitter_begin(emitter_t *emitter) { - if (emitter->output == emitter_output_json) { - assert(emitter->nesting_depth == 0); - emitter_printf(emitter, "{"); - emitter_nest_inc(emitter); - } else { - // tabular init - emitter_printf(emitter, "%s", ""); - } +emitter_init(emitter_t *emitter, emitter_output_t emitter_output, + void (*write_cb)(void *, const char *), void *cbopaque) { + emitter->output = emitter_output; + emitter->write_cb = write_cb; + emitter->cbopaque = cbopaque; + emitter->item_at_depth = false; + emitter->emitted_key = false; + emitter->nesting_depth = 0; } +/******************************************************************************/ +/* JSON public API. */ + +/* + * Emits a key (e.g. as appears in an object). The next json entity emitted will + * be the corresponding value. + */ static inline void -emitter_end(emitter_t *emitter) { +emitter_json_key(emitter_t *emitter, const char *json_key) { if (emitter->output == emitter_output_json) { - assert(emitter->nesting_depth == 1); - emitter_nest_dec(emitter); - emitter_printf(emitter, "\n}\n"); + emitter_json_key_prefix(emitter); + emitter_printf(emitter, "\"%s\": ", json_key); + emitter->emitted_key = true; } } -/* - * Note emits a different kv pair as well, but only in table mode. Omits the - * note if table_note_key is NULL. - */ static inline void -emitter_kv_note(emitter_t *emitter, const char *json_key, const char *table_key, - emitter_type_t value_type, const void *value, - const char *table_note_key, emitter_type_t table_note_value_type, - const void *table_note_value) { +emitter_json_value(emitter_t *emitter, emitter_type_t value_type, + const void *value) { if (emitter->output == emitter_output_json) { - assert(emitter->nesting_depth > 0); emitter_json_key_prefix(emitter); - emitter_printf(emitter, "\"%s\": ", json_key); - emitter_print_value(emitter, emitter_justify_none, -1, - value_type, value); - } else { - emitter_indent(emitter); - emitter_printf(emitter, "%s: ", table_key); emitter_print_value(emitter, emitter_justify_none, -1, value_type, value); - if (table_note_key != NULL) { - emitter_printf(emitter, " (%s: ", table_note_key); - emitter_print_value(emitter, emitter_justify_none, -1, - table_note_value_type, table_note_value); - emitter_printf(emitter, ")"); - } - emitter_printf(emitter, "\n"); + emitter->item_at_depth = true; } - emitter->item_at_depth = true; } +/* Shorthand for calling emitter_json_key and then emitter_json_value. */ static inline void -emitter_kv(emitter_t *emitter, const char *json_key, const char *table_key, +emitter_json_kv(emitter_t *emitter, const char *json_key, emitter_type_t value_type, const void *value) { - emitter_kv_note(emitter, json_key, table_key, value_type, value, NULL, - emitter_type_bool, NULL); + emitter_json_key(emitter, json_key); + emitter_json_value(emitter, value_type, value); } static inline void -emitter_json_kv(emitter_t *emitter, const char *json_key, - emitter_type_t value_type, const void *value) { +emitter_json_array_begin(emitter_t *emitter) { if (emitter->output == emitter_output_json) { - emitter_kv(emitter, json_key, NULL, value_type, value); + emitter_json_key_prefix(emitter); + emitter_printf(emitter, "["); + emitter_nest_inc(emitter); } } +/* Shorthand for calling emitter_json_key and then emitter_json_array_begin. */ static inline void -emitter_table_kv(emitter_t *emitter, const char *table_key, - emitter_type_t value_type, const void *value) { - if (emitter->output == emitter_output_table) { - emitter_kv(emitter, NULL, table_key, value_type, value); +emitter_json_array_kv_begin(emitter_t *emitter, const char *json_key) { + emitter_json_key(emitter, json_key); + emitter_json_array_begin(emitter); +} + +static inline void +emitter_json_array_end(emitter_t *emitter) { + if (emitter->output == emitter_output_json) { + assert(emitter->nesting_depth > 0); + emitter_nest_dec(emitter); + emitter_printf(emitter, "\n"); + emitter_indent(emitter); + emitter_printf(emitter, "]"); } } static inline void -emitter_dict_begin(emitter_t *emitter, const char *json_key, - const char *table_header) { +emitter_json_object_begin(emitter_t *emitter) { if (emitter->output == emitter_output_json) { emitter_json_key_prefix(emitter); - emitter_printf(emitter, "\"%s\": {", json_key); - emitter_nest_inc(emitter); - } else { - emitter_indent(emitter); - emitter_printf(emitter, "%s\n", table_header); + emitter_printf(emitter, "{"); emitter_nest_inc(emitter); } } +/* Shorthand for calling emitter_json_key and then emitter_json_object_begin. */ static inline void -emitter_dict_end(emitter_t *emitter) { +emitter_json_object_kv_begin(emitter_t *emitter, const char *json_key) { + emitter_json_key(emitter, json_key); + emitter_json_object_begin(emitter); +} + +static inline void +emitter_json_object_end(emitter_t *emitter) { if (emitter->output == emitter_output_json) { assert(emitter->nesting_depth > 0); emitter_nest_dec(emitter); emitter_printf(emitter, "\n"); emitter_indent(emitter); emitter_printf(emitter, "}"); - } else { - emitter_nest_dec(emitter); } } + +/******************************************************************************/ +/* Table public API. */ + static inline void -emitter_json_dict_begin(emitter_t *emitter, const char *json_key) { - if (emitter->output == emitter_output_json) { - emitter_dict_begin(emitter, json_key, NULL); +emitter_table_dict_begin(emitter_t *emitter, const char *table_key) { + if (emitter->output == emitter_output_table) { + emitter_indent(emitter); + emitter_printf(emitter, "%s\n", table_key); + emitter_nest_inc(emitter); } } static inline void -emitter_json_dict_end(emitter_t *emitter) { - if (emitter->output == emitter_output_json) { - emitter_dict_end(emitter); +emitter_table_dict_end(emitter_t *emitter) { + if (emitter->output == emitter_output_table) { + emitter_nest_dec(emitter); } } static inline void -emitter_table_dict_begin(emitter_t *emitter, const char *table_key) { +emitter_table_kv_note(emitter_t *emitter, const char *table_key, + emitter_type_t value_type, const void *value, + const char *table_note_key, emitter_type_t table_note_value_type, + const void *table_note_value) { if (emitter->output == emitter_output_table) { - emitter_dict_begin(emitter, NULL, table_key); + emitter_indent(emitter); + emitter_printf(emitter, "%s: ", table_key); + emitter_print_value(emitter, emitter_justify_none, -1, + value_type, value); + if (table_note_key != NULL) { + emitter_printf(emitter, " (%s: ", table_note_key); + emitter_print_value(emitter, emitter_justify_none, -1, + table_note_value_type, table_note_value); + emitter_printf(emitter, ")"); + } + emitter_printf(emitter, "\n"); } + emitter->item_at_depth = true; } static inline void -emitter_table_dict_end(emitter_t *emitter) { +emitter_table_kv(emitter_t *emitter, const char *table_key, + emitter_type_t value_type, const void *value) { + emitter_table_kv_note(emitter, table_key, value_type, value, NULL, + emitter_type_bool, NULL); +} + + +/* Write to the emitter the given string, but only in table mode. */ +JEMALLOC_FORMAT_PRINTF(2, 3) +static inline void +emitter_table_printf(emitter_t *emitter, const char *format, ...) { if (emitter->output == emitter_output_table) { - emitter_dict_end(emitter); + va_list ap; + va_start(ap, format); + malloc_vcprintf(emitter->write_cb, emitter->cbopaque, format, ap); + va_end(ap); } } static inline void -emitter_json_arr_begin(emitter_t *emitter, const char *json_key) { - if (emitter->output == emitter_output_json) { - emitter_json_key_prefix(emitter); - emitter_printf(emitter, "\"%s\": [", json_key); - emitter_nest_inc(emitter); +emitter_table_row(emitter_t *emitter, emitter_row_t *row) { + if (emitter->output != emitter_output_table) { + return; + } + emitter_col_t *col; + ql_foreach(col, &row->cols, link) { + emitter_print_value(emitter, col->justify, col->width, + col->type, (const void *)&col->bool_val); } + emitter_table_printf(emitter, "\n"); +} + +static inline void +emitter_row_init(emitter_row_t *row) { + ql_new(&row->cols); } static inline void -emitter_json_arr_end(emitter_t *emitter) { +emitter_col_init(emitter_col_t *col, emitter_row_t *row) { + ql_elm_new(col, link); + ql_tail_insert(&row->cols, col, link); +} + + +/******************************************************************************/ +/* + * Generalized public API. Emits using either JSON or table, according to + * settings in the emitter_t. */ + +/* + * Note emits a different kv pair as well, but only in table mode. Omits the + * note if table_note_key is NULL. + */ +static inline void +emitter_kv_note(emitter_t *emitter, const char *json_key, const char *table_key, + emitter_type_t value_type, const void *value, + const char *table_note_key, emitter_type_t table_note_value_type, + const void *table_note_value) { if (emitter->output == emitter_output_json) { - assert(emitter->nesting_depth > 0); - emitter_nest_dec(emitter); - emitter_printf(emitter, "\n"); - emitter_indent(emitter); - emitter_printf(emitter, "]"); + emitter_json_key(emitter, json_key); + emitter_json_value(emitter, value_type, value); + } else { + emitter_table_kv_note(emitter, table_key, value_type, value, + table_note_key, table_note_value_type, table_note_value); } + emitter->item_at_depth = true; } static inline void -emitter_json_arr_obj_begin(emitter_t *emitter) { +emitter_kv(emitter_t *emitter, const char *json_key, const char *table_key, + emitter_type_t value_type, const void *value) { + emitter_kv_note(emitter, json_key, table_key, value_type, value, NULL, + emitter_type_bool, NULL); +} + +static inline void +emitter_dict_begin(emitter_t *emitter, const char *json_key, + const char *table_header) { if (emitter->output == emitter_output_json) { - emitter_json_key_prefix(emitter); - emitter_printf(emitter, "{"); - emitter_nest_inc(emitter); + emitter_json_key(emitter, json_key); + emitter_json_object_begin(emitter); + } else { + emitter_table_dict_begin(emitter, table_header); } } static inline void -emitter_json_arr_obj_end(emitter_t *emitter) { +emitter_dict_end(emitter_t *emitter) { if (emitter->output == emitter_output_json) { - assert(emitter->nesting_depth > 0); - emitter_nest_dec(emitter); - emitter_printf(emitter, "\n"); - emitter_indent(emitter); - emitter_printf(emitter, "}"); + emitter_json_object_end(emitter); + } else { + emitter_table_dict_end(emitter); } } static inline void -emitter_json_arr_value(emitter_t *emitter, emitter_type_t value_type, - const void *value) { +emitter_begin(emitter_t *emitter) { if (emitter->output == emitter_output_json) { - emitter_json_key_prefix(emitter); - emitter_print_value(emitter, emitter_justify_none, -1, - value_type, value); + assert(emitter->nesting_depth == 0); + emitter_printf(emitter, "{"); + emitter_nest_inc(emitter); + } else { + /* + * This guarantees that we always call write_cb at least once. + * This is useful if some invariant is established by each call + * to write_cb, but doesn't hold initially: e.g., some buffer + * holds a null-terminated string. + */ + emitter_printf(emitter, "%s", ""); } } static inline void -emitter_table_row(emitter_t *emitter, emitter_row_t *row) { - if (emitter->output != emitter_output_table) { - return; - } - emitter_col_t *col; - ql_foreach(col, &row->cols, link) { - emitter_print_value(emitter, col->justify, col->width, - col->type, (const void *)&col->bool_val); +emitter_end(emitter_t *emitter) { + if (emitter->output == emitter_output_json) { + assert(emitter->nesting_depth == 1); + emitter_nest_dec(emitter); + emitter_printf(emitter, "\n}\n"); } - emitter_table_printf(emitter, "\n"); } #endif /* JEMALLOC_INTERNAL_EMITTER_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/extent_externs.h b/deps/jemalloc/include/jemalloc/internal/extent_externs.h index b8a4d026c..8aba57633 100644 --- a/deps/jemalloc/include/jemalloc/internal/extent_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/extent_externs.h @@ -24,13 +24,17 @@ size_t extent_size_quantize_floor(size_t size); size_t extent_size_quantize_ceil(size_t size); #endif -rb_proto(, extent_avail_, extent_tree_t, extent_t) +ph_proto(, extent_avail_, extent_tree_t, extent_t) ph_proto(, extent_heap_, extent_heap_t, extent_t) bool extents_init(tsdn_t *tsdn, extents_t *extents, extent_state_t state, bool delay_coalesce); extent_state_t extents_state_get(const extents_t *extents); size_t extents_npages_get(extents_t *extents); +/* Get the number of extents in the given page size index. */ +size_t extents_nextents_get(extents_t *extents, pszind_t ind); +/* Get the sum total bytes of the extents in the given page size index. */ +size_t extents_nbytes_get(extents_t *extents, pszind_t ind); extent_t *extents_alloc(tsdn_t *tsdn, arena_t *arena, extent_hooks_t **r_extent_hooks, extents_t *extents, void *new_addr, size_t size, size_t pad, size_t alignment, bool slab, szind_t szind, @@ -70,4 +74,10 @@ bool extent_merge_wrapper(tsdn_t *tsdn, arena_t *arena, bool extent_boot(void); +void extent_util_stats_get(tsdn_t *tsdn, const void *ptr, + size_t *nfree, size_t *nregs, size_t *size); +void extent_util_stats_verbose_get(tsdn_t *tsdn, const void *ptr, + size_t *nfree, size_t *nregs, size_t *size, + size_t *bin_nfree, size_t *bin_nregs, void **slabcur_addr); + #endif /* JEMALLOC_INTERNAL_EXTENT_EXTERNS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/extent_inlines.h b/deps/jemalloc/include/jemalloc/internal/extent_inlines.h index 77181df8d..77fa4c4a2 100644 --- a/deps/jemalloc/include/jemalloc/internal/extent_inlines.h +++ b/deps/jemalloc/include/jemalloc/internal/extent_inlines.h @@ -6,6 +6,7 @@ #include "jemalloc/internal/pages.h" #include "jemalloc/internal/prng.h" #include "jemalloc/internal/ql.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/sz.h" static inline void @@ -34,18 +35,19 @@ extent_unlock2(tsdn_t *tsdn, extent_t *extent1, extent_t *extent2) { (uintptr_t)extent2); } -static inline arena_t * -extent_arena_get(const extent_t *extent) { +static inline unsigned +extent_arena_ind_get(const extent_t *extent) { unsigned arena_ind = (unsigned)((extent->e_bits & EXTENT_BITS_ARENA_MASK) >> EXTENT_BITS_ARENA_SHIFT); - /* - * The following check is omitted because we should never actually read - * a NULL arena pointer. - */ - if (false && arena_ind >= MALLOCX_ARENA_LIMIT) { - return NULL; - } assert(arena_ind < MALLOCX_ARENA_LIMIT); + + return arena_ind; +} + +static inline arena_t * +extent_arena_get(const extent_t *extent) { + unsigned arena_ind = extent_arena_ind_get(extent); + return (arena_t *)atomic_load_p(&arenas[arena_ind], ATOMIC_ACQUIRE); } @@ -53,14 +55,14 @@ static inline szind_t extent_szind_get_maybe_invalid(const extent_t *extent) { szind_t szind = (szind_t)((extent->e_bits & EXTENT_BITS_SZIND_MASK) >> EXTENT_BITS_SZIND_SHIFT); - assert(szind <= NSIZES); + assert(szind <= SC_NSIZES); return szind; } static inline szind_t extent_szind_get(const extent_t *extent) { szind_t szind = extent_szind_get_maybe_invalid(extent); - assert(szind < NSIZES); /* Never call when "invalid". */ + assert(szind < SC_NSIZES); /* Never call when "invalid". */ return szind; } @@ -69,6 +71,14 @@ extent_usize_get(const extent_t *extent) { return sz_index2size(extent_szind_get(extent)); } +static inline unsigned +extent_binshard_get(const extent_t *extent) { + unsigned binshard = (unsigned)((extent->e_bits & + EXTENT_BITS_BINSHARD_MASK) >> EXTENT_BITS_BINSHARD_SHIFT); + assert(binshard < bin_infos[extent_szind_get(extent)].n_shards); + return binshard; +} + static inline size_t extent_sn_get(const extent_t *extent) { return (size_t)((extent->e_bits & EXTENT_BITS_SN_MASK) >> @@ -176,6 +186,11 @@ extent_prof_tctx_get(const extent_t *extent) { ATOMIC_ACQUIRE); } +static inline nstime_t +extent_prof_alloc_time_get(const extent_t *extent) { + return extent->e_alloc_time; +} + static inline void extent_arena_set(extent_t *extent, arena_t *arena) { unsigned arena_ind = (arena != NULL) ? arena_ind_get(arena) : ((1U << @@ -185,12 +200,20 @@ extent_arena_set(extent_t *extent, arena_t *arena) { } static inline void +extent_binshard_set(extent_t *extent, unsigned binshard) { + /* The assertion assumes szind is set already. */ + assert(binshard < bin_infos[extent_szind_get(extent)].n_shards); + extent->e_bits = (extent->e_bits & ~EXTENT_BITS_BINSHARD_MASK) | + ((uint64_t)binshard << EXTENT_BITS_BINSHARD_SHIFT); +} + +static inline void extent_addr_set(extent_t *extent, void *addr) { extent->e_addr = addr; } static inline void -extent_addr_randomize(UNUSED tsdn_t *tsdn, extent_t *extent, size_t alignment) { +extent_addr_randomize(tsdn_t *tsdn, extent_t *extent, size_t alignment) { assert(extent_base_get(extent) == extent_addr_get(extent)); if (alignment < PAGE) { @@ -234,7 +257,7 @@ extent_bsize_set(extent_t *extent, size_t bsize) { static inline void extent_szind_set(extent_t *extent, szind_t szind) { - assert(szind <= NSIZES); /* NSIZES means "invalid". */ + assert(szind <= SC_NSIZES); /* SC_NSIZES means "invalid". */ extent->e_bits = (extent->e_bits & ~EXTENT_BITS_SZIND_MASK) | ((uint64_t)szind << EXTENT_BITS_SZIND_SHIFT); } @@ -247,6 +270,16 @@ extent_nfree_set(extent_t *extent, unsigned nfree) { } static inline void +extent_nfree_binshard_set(extent_t *extent, unsigned nfree, unsigned binshard) { + /* The assertion assumes szind is set already. */ + assert(binshard < bin_infos[extent_szind_get(extent)].n_shards); + extent->e_bits = (extent->e_bits & + (~EXTENT_BITS_NFREE_MASK & ~EXTENT_BITS_BINSHARD_MASK)) | + ((uint64_t)binshard << EXTENT_BITS_BINSHARD_SHIFT) | + ((uint64_t)nfree << EXTENT_BITS_NFREE_SHIFT); +} + +static inline void extent_nfree_inc(extent_t *extent) { assert(extent_slab_get(extent)); extent->e_bits += ((uint64_t)1U << EXTENT_BITS_NFREE_SHIFT); @@ -259,6 +292,12 @@ extent_nfree_dec(extent_t *extent) { } static inline void +extent_nfree_sub(extent_t *extent, uint64_t n) { + assert(extent_slab_get(extent)); + extent->e_bits -= (n << EXTENT_BITS_NFREE_SHIFT); +} + +static inline void extent_sn_set(extent_t *extent, size_t sn) { extent->e_bits = (extent->e_bits & ~EXTENT_BITS_SN_MASK) | ((uint64_t)sn << EXTENT_BITS_SN_SHIFT); @@ -300,9 +339,34 @@ extent_prof_tctx_set(extent_t *extent, prof_tctx_t *tctx) { } static inline void +extent_prof_alloc_time_set(extent_t *extent, nstime_t t) { + nstime_copy(&extent->e_alloc_time, &t); +} + +static inline bool +extent_is_head_get(extent_t *extent) { + if (maps_coalesce) { + not_reached(); + } + + return (bool)((extent->e_bits & EXTENT_BITS_IS_HEAD_MASK) >> + EXTENT_BITS_IS_HEAD_SHIFT); +} + +static inline void +extent_is_head_set(extent_t *extent, bool is_head) { + if (maps_coalesce) { + not_reached(); + } + + extent->e_bits = (extent->e_bits & ~EXTENT_BITS_IS_HEAD_MASK) | + ((uint64_t)is_head << EXTENT_BITS_IS_HEAD_SHIFT); +} + +static inline void extent_init(extent_t *extent, arena_t *arena, void *addr, size_t size, bool slab, szind_t szind, size_t sn, extent_state_t state, bool zeroed, - bool committed, bool dumpable) { + bool committed, bool dumpable, extent_head_state_t is_head) { assert(addr == PAGE_ADDR2BASE(addr) || !slab); extent_arena_set(extent, arena); @@ -316,6 +380,10 @@ extent_init(extent_t *extent, arena_t *arena, void *addr, size_t size, extent_committed_set(extent, committed); extent_dumpable_set(extent, dumpable); ql_elm_new(extent, ql_link); + if (!maps_coalesce) { + extent_is_head_set(extent, (is_head == EXTENT_IS_HEAD) ? true : + false); + } if (config_prof) { extent_prof_tctx_set(extent, NULL); } @@ -327,7 +395,7 @@ extent_binit(extent_t *extent, void *addr, size_t bsize, size_t sn) { extent_addr_set(extent, addr); extent_bsize_set(extent, bsize); extent_slab_set(extent, false); - extent_szind_set(extent, NSIZES); + extent_szind_set(extent, SC_NSIZES); extent_sn_set(extent, sn); extent_state_set(extent, extent_state_active); extent_zeroed_set(extent, true); diff --git a/deps/jemalloc/include/jemalloc/internal/extent_structs.h b/deps/jemalloc/include/jemalloc/internal/extent_structs.h index 4873b9e9e..767cd8930 100644 --- a/deps/jemalloc/include/jemalloc/internal/extent_structs.h +++ b/deps/jemalloc/include/jemalloc/internal/extent_structs.h @@ -2,11 +2,12 @@ #define JEMALLOC_INTERNAL_EXTENT_STRUCTS_H #include "jemalloc/internal/atomic.h" +#include "jemalloc/internal/bit_util.h" #include "jemalloc/internal/bitmap.h" #include "jemalloc/internal/mutex.h" #include "jemalloc/internal/ql.h" #include "jemalloc/internal/ph.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" typedef enum { extent_state_active = 0, @@ -28,9 +29,10 @@ struct extent_s { * t: state * i: szind * f: nfree + * s: bin_shard * n: sn * - * nnnnnnnn ... nnnnffff ffffffii iiiiiitt zdcbaaaa aaaaaaaa + * nnnnnnnn ... nnnnnnss ssssffff ffffffii iiiiiitt zdcbaaaa aaaaaaaa * * arena_ind: Arena from which this extent came, or all 1 bits if * unassociated. @@ -75,6 +77,8 @@ struct extent_s { * * nfree: Number of free regions in slab. * + * bin_shard: the shard of the bin from which this extent came. + * * sn: Serial number (potentially non-unique). * * Serial numbers may wrap around if !opt_retain, but as long as @@ -112,7 +116,7 @@ struct extent_s { #define EXTENT_BITS_STATE_SHIFT (EXTENT_BITS_ZEROED_WIDTH + EXTENT_BITS_ZEROED_SHIFT) #define EXTENT_BITS_STATE_MASK MASK(EXTENT_BITS_STATE_WIDTH, EXTENT_BITS_STATE_SHIFT) -#define EXTENT_BITS_SZIND_WIDTH LG_CEIL_NSIZES +#define EXTENT_BITS_SZIND_WIDTH LG_CEIL(SC_NSIZES) #define EXTENT_BITS_SZIND_SHIFT (EXTENT_BITS_STATE_WIDTH + EXTENT_BITS_STATE_SHIFT) #define EXTENT_BITS_SZIND_MASK MASK(EXTENT_BITS_SZIND_WIDTH, EXTENT_BITS_SZIND_SHIFT) @@ -120,7 +124,15 @@ struct extent_s { #define EXTENT_BITS_NFREE_SHIFT (EXTENT_BITS_SZIND_WIDTH + EXTENT_BITS_SZIND_SHIFT) #define EXTENT_BITS_NFREE_MASK MASK(EXTENT_BITS_NFREE_WIDTH, EXTENT_BITS_NFREE_SHIFT) -#define EXTENT_BITS_SN_SHIFT (EXTENT_BITS_NFREE_WIDTH + EXTENT_BITS_NFREE_SHIFT) +#define EXTENT_BITS_BINSHARD_WIDTH 6 +#define EXTENT_BITS_BINSHARD_SHIFT (EXTENT_BITS_NFREE_WIDTH + EXTENT_BITS_NFREE_SHIFT) +#define EXTENT_BITS_BINSHARD_MASK MASK(EXTENT_BITS_BINSHARD_WIDTH, EXTENT_BITS_BINSHARD_SHIFT) + +#define EXTENT_BITS_IS_HEAD_WIDTH 1 +#define EXTENT_BITS_IS_HEAD_SHIFT (EXTENT_BITS_BINSHARD_WIDTH + EXTENT_BITS_BINSHARD_SHIFT) +#define EXTENT_BITS_IS_HEAD_MASK MASK(EXTENT_BITS_IS_HEAD_WIDTH, EXTENT_BITS_IS_HEAD_SHIFT) + +#define EXTENT_BITS_SN_SHIFT (EXTENT_BITS_IS_HEAD_WIDTH + EXTENT_BITS_IS_HEAD_SHIFT) #define EXTENT_BITS_SN_MASK (UINT64_MAX << EXTENT_BITS_SN_SHIFT) /* Pointer to the extent that this structure is responsible for. */ @@ -160,11 +172,13 @@ struct extent_s { /* Small region slab metadata. */ arena_slab_data_t e_slab_data; - /* - * Profile counters, used for large objects. Points to a - * prof_tctx_t. - */ - atomic_p_t e_prof_tctx; + /* Profiling data, used for large objects. */ + struct { + /* Time when this was allocated. */ + nstime_t e_alloc_time; + /* Points to a prof_tctx_t. */ + atomic_p_t e_prof_tctx; + }; }; }; typedef ql_head(extent_t) extent_list_t; @@ -180,14 +194,16 @@ struct extents_s { * * Synchronization: mtx. */ - extent_heap_t heaps[NPSIZES+1]; + extent_heap_t heaps[SC_NPSIZES + 1]; + atomic_zu_t nextents[SC_NPSIZES + 1]; + atomic_zu_t nbytes[SC_NPSIZES + 1]; /* * Bitmap for which set bits correspond to non-empty heaps. * * Synchronization: mtx. */ - bitmap_t bitmap[BITMAP_GROUPS(NPSIZES+1)]; + bitmap_t bitmap[BITMAP_GROUPS(SC_NPSIZES + 1)]; /* * LRU of all extents in heaps. @@ -216,4 +232,25 @@ struct extents_s { bool delay_coalesce; }; +/* + * The following two structs are for experimental purposes. See + * experimental_utilization_query_ctl and + * experimental_utilization_batch_query_ctl in src/ctl.c. + */ + +struct extent_util_stats_s { + size_t nfree; + size_t nregs; + size_t size; +}; + +struct extent_util_stats_verbose_s { + void *slabcur_addr; + size_t nfree; + size_t nregs; + size_t size; + size_t bin_nfree; + size_t bin_nregs; +}; + #endif /* JEMALLOC_INTERNAL_EXTENT_STRUCTS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/extent_types.h b/deps/jemalloc/include/jemalloc/internal/extent_types.h index c0561d99f..96925cf95 100644 --- a/deps/jemalloc/include/jemalloc/internal/extent_types.h +++ b/deps/jemalloc/include/jemalloc/internal/extent_types.h @@ -4,9 +4,10 @@ typedef struct extent_s extent_t; typedef struct extents_s extents_t; -#define EXTENT_HOOKS_INITIALIZER NULL +typedef struct extent_util_stats_s extent_util_stats_t; +typedef struct extent_util_stats_verbose_s extent_util_stats_verbose_t; -#define EXTENT_GROW_MAX_PIND (NPSIZES - 1) +#define EXTENT_HOOKS_INITIALIZER NULL /* * When reuse (and split) an active extent, (1U << opt_lg_extent_max_active_fit) @@ -14,4 +15,9 @@ typedef struct extents_s extents_t; */ #define LG_EXTENT_MAX_ACTIVE_FIT_DEFAULT 6 +typedef enum { + EXTENT_NOT_HEAD, + EXTENT_IS_HEAD /* Only relevant for Windows && opt.retain. */ +} extent_head_state_t; + #endif /* JEMALLOC_INTERNAL_EXTENT_TYPES_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/hash.h b/deps/jemalloc/include/jemalloc/internal/hash.h index dcfc992df..0270034e8 100644 --- a/deps/jemalloc/include/jemalloc/internal/hash.h +++ b/deps/jemalloc/include/jemalloc/internal/hash.h @@ -104,8 +104,8 @@ hash_x86_32(const void *key, int len, uint32_t seed) { uint32_t k1 = 0; switch (len & 3) { - case 3: k1 ^= tail[2] << 16; - case 2: k1 ^= tail[1] << 8; + case 3: k1 ^= tail[2] << 16; JEMALLOC_FALLTHROUGH + case 2: k1 ^= tail[1] << 8; JEMALLOC_FALLTHROUGH case 1: k1 ^= tail[0]; k1 *= c1; k1 = hash_rotl_32(k1, 15); k1 *= c2; h1 ^= k1; } @@ -119,7 +119,7 @@ hash_x86_32(const void *key, int len, uint32_t seed) { return h1; } -UNUSED static inline void +static inline void hash_x86_128(const void *key, const int len, uint32_t seed, uint64_t r_out[2]) { const uint8_t * data = (const uint8_t *) key; @@ -177,28 +177,29 @@ hash_x86_128(const void *key, const int len, uint32_t seed, uint32_t k4 = 0; switch (len & 15) { - case 15: k4 ^= tail[14] << 16; - case 14: k4 ^= tail[13] << 8; + case 15: k4 ^= tail[14] << 16; JEMALLOC_FALLTHROUGH + case 14: k4 ^= tail[13] << 8; JEMALLOC_FALLTHROUGH case 13: k4 ^= tail[12] << 0; k4 *= c4; k4 = hash_rotl_32(k4, 18); k4 *= c1; h4 ^= k4; - - case 12: k3 ^= tail[11] << 24; - case 11: k3 ^= tail[10] << 16; - case 10: k3 ^= tail[ 9] << 8; + JEMALLOC_FALLTHROUGH + case 12: k3 ^= tail[11] << 24; JEMALLOC_FALLTHROUGH + case 11: k3 ^= tail[10] << 16; JEMALLOC_FALLTHROUGH + case 10: k3 ^= tail[ 9] << 8; JEMALLOC_FALLTHROUGH case 9: k3 ^= tail[ 8] << 0; k3 *= c3; k3 = hash_rotl_32(k3, 17); k3 *= c4; h3 ^= k3; - - case 8: k2 ^= tail[ 7] << 24; - case 7: k2 ^= tail[ 6] << 16; - case 6: k2 ^= tail[ 5] << 8; + JEMALLOC_FALLTHROUGH + case 8: k2 ^= tail[ 7] << 24; JEMALLOC_FALLTHROUGH + case 7: k2 ^= tail[ 6] << 16; JEMALLOC_FALLTHROUGH + case 6: k2 ^= tail[ 5] << 8; JEMALLOC_FALLTHROUGH case 5: k2 ^= tail[ 4] << 0; k2 *= c2; k2 = hash_rotl_32(k2, 16); k2 *= c3; h2 ^= k2; - - case 4: k1 ^= tail[ 3] << 24; - case 3: k1 ^= tail[ 2] << 16; - case 2: k1 ^= tail[ 1] << 8; + JEMALLOC_FALLTHROUGH + case 4: k1 ^= tail[ 3] << 24; JEMALLOC_FALLTHROUGH + case 3: k1 ^= tail[ 2] << 16; JEMALLOC_FALLTHROUGH + case 2: k1 ^= tail[ 1] << 8; JEMALLOC_FALLTHROUGH case 1: k1 ^= tail[ 0] << 0; k1 *= c1; k1 = hash_rotl_32(k1, 15); k1 *= c2; h1 ^= k1; + JEMALLOC_FALLTHROUGH } } @@ -220,7 +221,7 @@ hash_x86_128(const void *key, const int len, uint32_t seed, r_out[1] = (((uint64_t) h4) << 32) | h3; } -UNUSED static inline void +static inline void hash_x64_128(const void *key, const int len, const uint32_t seed, uint64_t r_out[2]) { const uint8_t *data = (const uint8_t *) key; @@ -260,22 +261,22 @@ hash_x64_128(const void *key, const int len, const uint32_t seed, uint64_t k2 = 0; switch (len & 15) { - case 15: k2 ^= ((uint64_t)(tail[14])) << 48; /* falls through */ - case 14: k2 ^= ((uint64_t)(tail[13])) << 40; /* falls through */ - case 13: k2 ^= ((uint64_t)(tail[12])) << 32; /* falls through */ - case 12: k2 ^= ((uint64_t)(tail[11])) << 24; /* falls through */ - case 11: k2 ^= ((uint64_t)(tail[10])) << 16; /* falls through */ - case 10: k2 ^= ((uint64_t)(tail[ 9])) << 8; /* falls through */ + case 15: k2 ^= ((uint64_t)(tail[14])) << 48; JEMALLOC_FALLTHROUGH + case 14: k2 ^= ((uint64_t)(tail[13])) << 40; JEMALLOC_FALLTHROUGH + case 13: k2 ^= ((uint64_t)(tail[12])) << 32; JEMALLOC_FALLTHROUGH + case 12: k2 ^= ((uint64_t)(tail[11])) << 24; JEMALLOC_FALLTHROUGH + case 11: k2 ^= ((uint64_t)(tail[10])) << 16; JEMALLOC_FALLTHROUGH + case 10: k2 ^= ((uint64_t)(tail[ 9])) << 8; JEMALLOC_FALLTHROUGH case 9: k2 ^= ((uint64_t)(tail[ 8])) << 0; k2 *= c2; k2 = hash_rotl_64(k2, 33); k2 *= c1; h2 ^= k2; - /* falls through */ - case 8: k1 ^= ((uint64_t)(tail[ 7])) << 56; /* falls through */ - case 7: k1 ^= ((uint64_t)(tail[ 6])) << 48; /* falls through */ - case 6: k1 ^= ((uint64_t)(tail[ 5])) << 40; /* falls through */ - case 5: k1 ^= ((uint64_t)(tail[ 4])) << 32; /* falls through */ - case 4: k1 ^= ((uint64_t)(tail[ 3])) << 24; /* falls through */ - case 3: k1 ^= ((uint64_t)(tail[ 2])) << 16; /* falls through */ - case 2: k1 ^= ((uint64_t)(tail[ 1])) << 8; /* falls through */ + JEMALLOC_FALLTHROUGH + case 8: k1 ^= ((uint64_t)(tail[ 7])) << 56; JEMALLOC_FALLTHROUGH + case 7: k1 ^= ((uint64_t)(tail[ 6])) << 48; JEMALLOC_FALLTHROUGH + case 6: k1 ^= ((uint64_t)(tail[ 5])) << 40; JEMALLOC_FALLTHROUGH + case 5: k1 ^= ((uint64_t)(tail[ 4])) << 32; JEMALLOC_FALLTHROUGH + case 4: k1 ^= ((uint64_t)(tail[ 3])) << 24; JEMALLOC_FALLTHROUGH + case 3: k1 ^= ((uint64_t)(tail[ 2])) << 16; JEMALLOC_FALLTHROUGH + case 2: k1 ^= ((uint64_t)(tail[ 1])) << 8; JEMALLOC_FALLTHROUGH case 1: k1 ^= ((uint64_t)(tail[ 0])) << 0; k1 *= c1; k1 = hash_rotl_64(k1, 31); k1 *= c2; h1 ^= k1; } diff --git a/deps/jemalloc/include/jemalloc/internal/hook.h b/deps/jemalloc/include/jemalloc/internal/hook.h new file mode 100644 index 000000000..ee246b1e0 --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/hook.h @@ -0,0 +1,163 @@ +#ifndef JEMALLOC_INTERNAL_HOOK_H +#define JEMALLOC_INTERNAL_HOOK_H + +#include "jemalloc/internal/tsd.h" + +/* + * This API is *extremely* experimental, and may get ripped out, changed in API- + * and ABI-incompatible ways, be insufficiently or incorrectly documented, etc. + * + * It allows hooking the stateful parts of the API to see changes as they + * happen. + * + * Allocation hooks are called after the allocation is done, free hooks are + * called before the free is done, and expand hooks are called after the + * allocation is expanded. + * + * For realloc and rallocx, if the expansion happens in place, the expansion + * hook is called. If it is moved, then the alloc hook is called on the new + * location, and then the free hook is called on the old location (i.e. both + * hooks are invoked in between the alloc and the dalloc). + * + * If we return NULL from OOM, then usize might not be trustworthy. Calling + * realloc(NULL, size) only calls the alloc hook, and calling realloc(ptr, 0) + * only calls the free hook. (Calling realloc(NULL, 0) is treated as malloc(0), + * and only calls the alloc hook). + * + * Reentrancy: + * Reentrancy is guarded against from within the hook implementation. If you + * call allocator functions from within a hook, the hooks will not be invoked + * again. + * Threading: + * The installation of a hook synchronizes with all its uses. If you can + * prove the installation of a hook happens-before a jemalloc entry point, + * then the hook will get invoked (unless there's a racing removal). + * + * Hook insertion appears to be atomic at a per-thread level (i.e. if a thread + * allocates and has the alloc hook invoked, then a subsequent free on the + * same thread will also have the free hook invoked). + * + * The *removal* of a hook does *not* block until all threads are done with + * the hook. Hook authors have to be resilient to this, and need some + * out-of-band mechanism for cleaning up any dynamically allocated memory + * associated with their hook. + * Ordering: + * Order of hook execution is unspecified, and may be different than insertion + * order. + */ + +#define HOOK_MAX 4 + +enum hook_alloc_e { + hook_alloc_malloc, + hook_alloc_posix_memalign, + hook_alloc_aligned_alloc, + hook_alloc_calloc, + hook_alloc_memalign, + hook_alloc_valloc, + hook_alloc_mallocx, + + /* The reallocating functions have both alloc and dalloc variants */ + hook_alloc_realloc, + hook_alloc_rallocx, +}; +/* + * We put the enum typedef after the enum, since this file may get included by + * jemalloc_cpp.cpp, and C++ disallows enum forward declarations. + */ +typedef enum hook_alloc_e hook_alloc_t; + +enum hook_dalloc_e { + hook_dalloc_free, + hook_dalloc_dallocx, + hook_dalloc_sdallocx, + + /* + * The dalloc halves of reallocation (not called if in-place expansion + * happens). + */ + hook_dalloc_realloc, + hook_dalloc_rallocx, +}; +typedef enum hook_dalloc_e hook_dalloc_t; + + +enum hook_expand_e { + hook_expand_realloc, + hook_expand_rallocx, + hook_expand_xallocx, +}; +typedef enum hook_expand_e hook_expand_t; + +typedef void (*hook_alloc)( + void *extra, hook_alloc_t type, void *result, uintptr_t result_raw, + uintptr_t args_raw[3]); + +typedef void (*hook_dalloc)( + void *extra, hook_dalloc_t type, void *address, uintptr_t args_raw[3]); + +typedef void (*hook_expand)( + void *extra, hook_expand_t type, void *address, size_t old_usize, + size_t new_usize, uintptr_t result_raw, uintptr_t args_raw[4]); + +typedef struct hooks_s hooks_t; +struct hooks_s { + hook_alloc alloc_hook; + hook_dalloc dalloc_hook; + hook_expand expand_hook; + void *extra; +}; + +/* + * Begin implementation details; everything above this point might one day live + * in a public API. Everything below this point never will. + */ + +/* + * The realloc pathways haven't gotten any refactoring love in a while, and it's + * fairly difficult to pass information from the entry point to the hooks. We + * put the informaiton the hooks will need into a struct to encapsulate + * everything. + * + * Much of these pathways are force-inlined, so that the compiler can avoid + * materializing this struct until we hit an extern arena function. For fairly + * goofy reasons, *many* of the realloc paths hit an extern arena function. + * These paths are cold enough that it doesn't matter; eventually, we should + * rewrite the realloc code to make the expand-in-place and the + * free-then-realloc paths more orthogonal, at which point we don't need to + * spread the hook logic all over the place. + */ +typedef struct hook_ralloc_args_s hook_ralloc_args_t; +struct hook_ralloc_args_s { + /* I.e. as opposed to rallocx. */ + bool is_realloc; + /* + * The expand hook takes 4 arguments, even if only 3 are actually used; + * we add an extra one in case the user decides to memcpy without + * looking too closely at the hooked function. + */ + uintptr_t args[4]; +}; + +/* + * Returns an opaque handle to be used when removing the hook. NULL means that + * we couldn't install the hook. + */ +bool hook_boot(); + +void *hook_install(tsdn_t *tsdn, hooks_t *hooks); +/* Uninstalls the hook with the handle previously returned from hook_install. */ +void hook_remove(tsdn_t *tsdn, void *opaque); + +/* Hooks */ + +void hook_invoke_alloc(hook_alloc_t type, void *result, uintptr_t result_raw, + uintptr_t args_raw[3]); + +void hook_invoke_dalloc(hook_dalloc_t type, void *address, + uintptr_t args_raw[3]); + +void hook_invoke_expand(hook_expand_t type, void *address, size_t old_usize, + size_t new_usize, uintptr_t result_raw, uintptr_t args_raw[4]); + +#endif /* JEMALLOC_INTERNAL_HOOK_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/hooks.h b/deps/jemalloc/include/jemalloc/internal/hooks.h deleted file mode 100644 index cd49afcb0..000000000 --- a/deps/jemalloc/include/jemalloc/internal/hooks.h +++ /dev/null @@ -1,19 +0,0 @@ -#ifndef JEMALLOC_INTERNAL_HOOKS_H -#define JEMALLOC_INTERNAL_HOOKS_H - -extern JEMALLOC_EXPORT void (*hooks_arena_new_hook)(); -extern JEMALLOC_EXPORT void (*hooks_libc_hook)(); - -#define JEMALLOC_HOOK(fn, hook) ((void)(hook != NULL && (hook(), 0)), fn) - -#define open JEMALLOC_HOOK(open, hooks_libc_hook) -#define read JEMALLOC_HOOK(read, hooks_libc_hook) -#define write JEMALLOC_HOOK(write, hooks_libc_hook) -#define readlink JEMALLOC_HOOK(readlink, hooks_libc_hook) -#define close JEMALLOC_HOOK(close, hooks_libc_hook) -#define creat JEMALLOC_HOOK(creat, hooks_libc_hook) -#define secure_getenv JEMALLOC_HOOK(secure_getenv, hooks_libc_hook) -/* Note that this is undef'd and re-define'd in src/prof.c. */ -#define _Unwind_Backtrace JEMALLOC_HOOK(_Unwind_Backtrace, hooks_libc_hook) - -#endif /* JEMALLOC_INTERNAL_HOOKS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_decls.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_decls.h index be70df510..7d6053e21 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_decls.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_decls.h @@ -31,6 +31,9 @@ # include <sys/uio.h> # endif # include <pthread.h> +# ifdef __FreeBSD__ +# include <pthread_np.h> +# endif # include <signal.h> # ifdef JEMALLOC_OS_UNFAIR_LOCK # include <os/lock.h> diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_defs.h.in b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_defs.h.in index 8dad9a1db..c442a2191 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_defs.h.in +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_defs.h.in @@ -48,25 +48,13 @@ /* Defined if GCC __atomic atomics are available. */ #undef JEMALLOC_GCC_ATOMIC_ATOMICS +/* and the 8-bit variant support. */ +#undef JEMALLOC_GCC_U8_ATOMIC_ATOMICS /* Defined if GCC __sync atomics are available. */ #undef JEMALLOC_GCC_SYNC_ATOMICS - -/* - * Defined if __sync_add_and_fetch(uint32_t *, uint32_t) and - * __sync_sub_and_fetch(uint32_t *, uint32_t) are available, despite - * __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 not being defined (which means the - * functions are defined in libgcc instead of being inlines). - */ -#undef JE_FORCE_SYNC_COMPARE_AND_SWAP_4 - -/* - * Defined if __sync_add_and_fetch(uint64_t *, uint64_t) and - * __sync_sub_and_fetch(uint64_t *, uint64_t) are available, despite - * __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 not being defined (which means the - * functions are defined in libgcc instead of being inlines). - */ -#undef JE_FORCE_SYNC_COMPARE_AND_SWAP_8 +/* and the 8-bit variant support. */ +#undef JEMALLOC_GCC_U8_SYNC_ATOMICS /* * Defined if __builtin_clz() and __builtin_clzl() are available. @@ -78,12 +66,6 @@ */ #undef JEMALLOC_OS_UNFAIR_LOCK -/* - * Defined if OSSpin*() functions are available, as provided by Darwin, and - * documented in the spinlock(3) manual page. - */ -#undef JEMALLOC_OSSPIN - /* Defined if syscall(2) is usable. */ #undef JEMALLOC_USE_SYSCALL @@ -153,6 +135,9 @@ /* JEMALLOC_STATS enables statistics calculation. */ #undef JEMALLOC_STATS +/* JEMALLOC_EXPERIMENTAL_SMALLOCX_API enables experimental smallocx API. */ +#undef JEMALLOC_EXPERIMENTAL_SMALLOCX_API + /* JEMALLOC_PROF enables allocation profiling. */ #undef JEMALLOC_PROF @@ -234,6 +219,12 @@ #undef JEMALLOC_INTERNAL_FFS /* + * popcount*() functions to use for bitmapping. + */ +#undef JEMALLOC_INTERNAL_POPCOUNTL +#undef JEMALLOC_INTERNAL_POPCOUNT + +/* * If defined, explicitly attempt to more uniformly distribute large allocation * pointer alignments across all cache indices. */ @@ -246,6 +237,12 @@ #undef JEMALLOC_LOG /* + * If defined, use readlinkat() (instead of readlink()) to follow + * /etc/malloc_conf. + */ +#undef JEMALLOC_READLINKAT + +/* * Darwin (OS X) uses zones to work around Mach-O symbol override shortcomings. */ #undef JEMALLOC_ZONE @@ -363,4 +360,7 @@ */ #undef JEMALLOC_STRERROR_R_RETURNS_CHAR_WITH_GNU_SOURCE +/* Performs additional safety checks when defined. */ +#undef JEMALLOC_OPT_SAFETY_CHECKS + #endif /* JEMALLOC_INTERNAL_DEFS_H_ */ diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_externs.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_externs.h index e10fb275d..d291170be 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_externs.h @@ -2,7 +2,6 @@ #define JEMALLOC_INTERNAL_EXTERNS_H #include "jemalloc/internal/atomic.h" -#include "jemalloc/internal/size_classes.h" #include "jemalloc/internal/tsd_types.h" /* TSD checks this to set thread local slow state accordingly. */ @@ -11,6 +10,7 @@ extern bool malloc_slow; /* Run-time options. */ extern bool opt_abort; extern bool opt_abort_conf; +extern bool opt_confirm_conf; extern const char *opt_junk; extern bool opt_junk_alloc; extern bool opt_junk_free; @@ -25,6 +25,9 @@ extern unsigned ncpus; /* Number of arenas used for automatic multiplexing of threads and arenas. */ extern unsigned narenas_auto; +/* Base index for manual arenas. */ +extern unsigned manual_arena_base; + /* * Arenas that are used to service external requests. Not all elements of the * arenas array are necessarily used; arenas are created lazily as needed. @@ -49,5 +52,6 @@ void jemalloc_prefork(void); void jemalloc_postfork_parent(void); void jemalloc_postfork_child(void); bool malloc_initialized(void); +void je_sdallocx_noflags(void *ptr, size_t size); #endif /* JEMALLOC_INTERNAL_EXTERNS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_a.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_a.h index c6a1f7eb2..ddde9b4e6 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_a.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_a.h @@ -4,13 +4,15 @@ #include "jemalloc/internal/atomic.h" #include "jemalloc/internal/bit_util.h" #include "jemalloc/internal/jemalloc_internal_types.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/ticker.h" JEMALLOC_ALWAYS_INLINE malloc_cpuid_t malloc_getcpu(void) { assert(have_percpu_arena); -#if defined(JEMALLOC_HAVE_SCHED_GETCPU) +#if defined(_WIN32) + return GetCurrentProcessorNumber(); +#elif defined(JEMALLOC_HAVE_SCHED_GETCPU) return (malloc_cpuid_t)sched_getcpu(); #else not_reached(); @@ -108,14 +110,14 @@ decay_ticker_get(tsd_t *tsd, unsigned ind) { JEMALLOC_ALWAYS_INLINE cache_bin_t * tcache_small_bin_get(tcache_t *tcache, szind_t binind) { - assert(binind < NBINS); + assert(binind < SC_NBINS); return &tcache->bins_small[binind]; } JEMALLOC_ALWAYS_INLINE cache_bin_t * tcache_large_bin_get(tcache_t *tcache, szind_t binind) { - assert(binind >= NBINS &&binind < nhbins); - return &tcache->bins_large[binind - NBINS]; + assert(binind >= SC_NBINS &&binind < nhbins); + return &tcache->bins_large[binind - SC_NBINS]; } JEMALLOC_ALWAYS_INLINE bool @@ -156,7 +158,7 @@ pre_reentrancy(tsd_t *tsd, arena_t *arena) { if (fast) { /* Prepare slow path for reentrancy. */ tsd_slow_update(tsd); - assert(tsd->state == tsd_state_nominal_slow); + assert(tsd_state_get(tsd) == tsd_state_nominal_slow); } } diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_b.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_b.h index 2e76e5d8f..70d6e5788 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_b.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_b.h @@ -71,7 +71,8 @@ arena_ichoose(tsd_t *tsd, arena_t *arena) { static inline bool arena_is_auto(arena_t *arena) { assert(narenas_auto > 0); - return (arena_ind_get(arena) < narenas_auto); + + return (arena_ind_get(arena) < manual_arena_base); } JEMALLOC_ALWAYS_INLINE extent_t * diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_c.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_c.h index b19a94207..adae014a1 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_c.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_inlines_c.h @@ -1,6 +1,7 @@ #ifndef JEMALLOC_INTERNAL_INLINES_C_H #define JEMALLOC_INTERNAL_INLINES_C_H +#include "jemalloc/internal/hook.h" #include "jemalloc/internal/jemalloc_internal_types.h" #include "jemalloc/internal/sz.h" #include "jemalloc/internal/witness.h" @@ -42,7 +43,6 @@ iallocztm(tsdn_t *tsdn, size_t size, szind_t ind, bool zero, tcache_t *tcache, bool is_internal, arena_t *arena, bool slow_path) { void *ret; - assert(size != 0); assert(!is_internal || tcache == NULL); assert(!is_internal || arena == NULL || arena_is_auto(arena)); if (!tsdn_null(tsdn) && tsd_reentrancy_level_get(tsdn_tsd(tsdn)) == 0) { @@ -133,31 +133,20 @@ isdalloct(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, JEMALLOC_ALWAYS_INLINE void * iralloct_realign(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, - size_t extra, size_t alignment, bool zero, tcache_t *tcache, - arena_t *arena) { + size_t alignment, bool zero, tcache_t *tcache, arena_t *arena, + hook_ralloc_args_t *hook_args) { witness_assert_depth_to_rank(tsdn_witness_tsdp_get(tsdn), WITNESS_RANK_CORE, 0); void *p; size_t usize, copysize; - usize = sz_sa2u(size + extra, alignment); - if (unlikely(usize == 0 || usize > LARGE_MAXCLASS)) { + usize = sz_sa2u(size, alignment); + if (unlikely(usize == 0 || usize > SC_LARGE_MAXCLASS)) { return NULL; } p = ipalloct(tsdn, usize, alignment, zero, tcache, arena); if (p == NULL) { - if (extra == 0) { - return NULL; - } - /* Try again, without extra this time. */ - usize = sz_sa2u(size, alignment); - if (unlikely(usize == 0 || usize > LARGE_MAXCLASS)) { - return NULL; - } - p = ipalloct(tsdn, usize, alignment, zero, tcache, arena); - if (p == NULL) { - return NULL; - } + return NULL; } /* * Copy at most size bytes (not size+extra), since the caller has no @@ -165,13 +154,26 @@ iralloct_realign(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, */ copysize = (size < oldsize) ? size : oldsize; memcpy(p, ptr, copysize); + hook_invoke_alloc(hook_args->is_realloc + ? hook_alloc_realloc : hook_alloc_rallocx, p, (uintptr_t)p, + hook_args->args); + hook_invoke_dalloc(hook_args->is_realloc + ? hook_dalloc_realloc : hook_dalloc_rallocx, ptr, hook_args->args); isdalloct(tsdn, ptr, oldsize, tcache, NULL, true); return p; } +/* + * is_realloc threads through the knowledge of whether or not this call comes + * from je_realloc (as opposed to je_rallocx); this ensures that we pass the + * correct entry point into any hooks. + * Note that these functions are all force-inlined, so no actual bool gets + * passed-around anywhere. + */ JEMALLOC_ALWAYS_INLINE void * iralloct(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, size_t alignment, - bool zero, tcache_t *tcache, arena_t *arena) { + bool zero, tcache_t *tcache, arena_t *arena, hook_ralloc_args_t *hook_args) +{ assert(ptr != NULL); assert(size != 0); witness_assert_depth_to_rank(tsdn_witness_tsdp_get(tsdn), @@ -183,24 +185,24 @@ iralloct(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, size_t alignment, * Existing object alignment is inadequate; allocate new space * and copy. */ - return iralloct_realign(tsdn, ptr, oldsize, size, 0, alignment, - zero, tcache, arena); + return iralloct_realign(tsdn, ptr, oldsize, size, alignment, + zero, tcache, arena, hook_args); } return arena_ralloc(tsdn, arena, ptr, oldsize, size, alignment, zero, - tcache); + tcache, hook_args); } JEMALLOC_ALWAYS_INLINE void * iralloc(tsd_t *tsd, void *ptr, size_t oldsize, size_t size, size_t alignment, - bool zero) { + bool zero, hook_ralloc_args_t *hook_args) { return iralloct(tsd_tsdn(tsd), ptr, oldsize, size, alignment, zero, - tcache_get(tsd), NULL); + tcache_get(tsd), NULL, hook_args); } JEMALLOC_ALWAYS_INLINE bool ixalloc(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, size_t extra, - size_t alignment, bool zero) { + size_t alignment, bool zero, size_t *newsize) { assert(ptr != NULL); assert(size != 0); witness_assert_depth_to_rank(tsdn_witness_tsdp_get(tsdn), @@ -209,10 +211,12 @@ ixalloc(tsdn_t *tsdn, void *ptr, size_t oldsize, size_t size, size_t extra, if (alignment != 0 && ((uintptr_t)ptr & ((uintptr_t)alignment-1)) != 0) { /* Existing object alignment is inadequate. */ + *newsize = oldsize; return true; } - return arena_ralloc_no_move(tsdn, ptr, oldsize, size, extra, zero); + return arena_ralloc_no_move(tsdn, ptr, oldsize, size, extra, zero, + newsize); } JEMALLOC_ALWAYS_INLINE int @@ -228,7 +232,8 @@ iget_defrag_hint(tsdn_t *tsdn, void* ptr) { extent_t *slab = iealloc(tsdn, ptr); arena_t *arena = extent_arena_get(slab); szind_t binind = extent_szind_get(slab); - bin_t *bin = &arena->bins[binind]; + unsigned binshard = extent_binshard_get(slab); + bin_t *bin = &arena->bins[binind].bin_shards[binshard]; malloc_mutex_lock(tsdn, &bin->lock); /* don't bother moving allocations from the slab currently used for new allocations */ if (slab != bin->slabcur) { diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_macros.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_macros.h index ed75d3768..d8ea06f6d 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_macros.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_macros.h @@ -30,7 +30,7 @@ # define restrict #endif -/* Various function pointers are statick and immutable except during testing. */ +/* Various function pointers are static and immutable except during testing. */ #ifdef JEMALLOC_JET # define JET_MUTABLE #else @@ -40,4 +40,75 @@ #define JEMALLOC_VA_ARGS_HEAD(head, ...) head #define JEMALLOC_VA_ARGS_TAIL(head, ...) __VA_ARGS__ +#if (defined(__GNUC__) || defined(__GNUG__)) && !defined(__clang__) \ + && defined(JEMALLOC_HAVE_ATTR) && (__GNUC__ >= 7) +#define JEMALLOC_FALLTHROUGH JEMALLOC_ATTR(fallthrough); +#else +#define JEMALLOC_FALLTHROUGH /* falls through */ +#endif + +/* Diagnostic suppression macros */ +#if defined(_MSC_VER) && !defined(__clang__) +# define JEMALLOC_DIAGNOSTIC_PUSH __pragma(warning(push)) +# define JEMALLOC_DIAGNOSTIC_POP __pragma(warning(pop)) +# define JEMALLOC_DIAGNOSTIC_IGNORE(W) __pragma(warning(disable:W)) +# define JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS +# define JEMALLOC_DIAGNOSTIC_IGNORE_TYPE_LIMITS +# define JEMALLOC_DIAGNOSTIC_IGNORE_ALLOC_SIZE_LARGER_THAN +# define JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS +/* #pragma GCC diagnostic first appeared in gcc 4.6. */ +#elif (defined(__GNUC__) && ((__GNUC__ > 4) || ((__GNUC__ == 4) && \ + (__GNUC_MINOR__ > 5)))) || defined(__clang__) +/* + * The JEMALLOC_PRAGMA__ macro is an implementation detail of the GCC and Clang + * diagnostic suppression macros and should not be used anywhere else. + */ +# define JEMALLOC_PRAGMA__(X) _Pragma(#X) +# define JEMALLOC_DIAGNOSTIC_PUSH JEMALLOC_PRAGMA__(GCC diagnostic push) +# define JEMALLOC_DIAGNOSTIC_POP JEMALLOC_PRAGMA__(GCC diagnostic pop) +# define JEMALLOC_DIAGNOSTIC_IGNORE(W) \ + JEMALLOC_PRAGMA__(GCC diagnostic ignored W) + +/* + * The -Wmissing-field-initializers warning is buggy in GCC versions < 5.1 and + * all clang versions up to version 7 (currently trunk, unreleased). This macro + * suppresses the warning for the affected compiler versions only. + */ +# if ((defined(__GNUC__) && !defined(__clang__)) && (__GNUC__ < 5)) || \ + defined(__clang__) +# define JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS \ + JEMALLOC_DIAGNOSTIC_IGNORE("-Wmissing-field-initializers") +# else +# define JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS +# endif + +# define JEMALLOC_DIAGNOSTIC_IGNORE_TYPE_LIMITS \ + JEMALLOC_DIAGNOSTIC_IGNORE("-Wtype-limits") +# define JEMALLOC_DIAGNOSTIC_IGNORE_UNUSED_PARAMETER \ + JEMALLOC_DIAGNOSTIC_IGNORE("-Wunused-parameter") +# if defined(__GNUC__) && !defined(__clang__) && (__GNUC__ >= 7) +# define JEMALLOC_DIAGNOSTIC_IGNORE_ALLOC_SIZE_LARGER_THAN \ + JEMALLOC_DIAGNOSTIC_IGNORE("-Walloc-size-larger-than=") +# else +# define JEMALLOC_DIAGNOSTIC_IGNORE_ALLOC_SIZE_LARGER_THAN +# endif +# define JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS \ + JEMALLOC_DIAGNOSTIC_PUSH \ + JEMALLOC_DIAGNOSTIC_IGNORE_UNUSED_PARAMETER +#else +# define JEMALLOC_DIAGNOSTIC_PUSH +# define JEMALLOC_DIAGNOSTIC_POP +# define JEMALLOC_DIAGNOSTIC_IGNORE(W) +# define JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS +# define JEMALLOC_DIAGNOSTIC_IGNORE_TYPE_LIMITS +# define JEMALLOC_DIAGNOSTIC_IGNORE_ALLOC_SIZE_LARGER_THAN +# define JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS +#endif + +/* + * Disables spurious diagnostics for all headers. Since these headers are not + * included by users directly, it does not affect their diagnostic settings. + */ +JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS + #endif /* JEMALLOC_INTERNAL_MACROS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_types.h b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_types.h index 1b750b122..e296c5a7e 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_types.h +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_internal_types.h @@ -1,6 +1,8 @@ #ifndef JEMALLOC_INTERNAL_TYPES_H #define JEMALLOC_INTERNAL_TYPES_H +#include "jemalloc/internal/quantum.h" + /* Page size index type. */ typedef unsigned pszind_t; @@ -50,79 +52,6 @@ typedef int malloc_cpuid_t; /* Smallest size class to support. */ #define TINY_MIN (1U << LG_TINY_MIN) -/* - * Minimum allocation alignment is 2^LG_QUANTUM bytes (ignoring tiny size - * classes). - */ -#ifndef LG_QUANTUM -# if (defined(__i386__) || defined(_M_IX86)) -# define LG_QUANTUM 4 -# endif -# ifdef __ia64__ -# define LG_QUANTUM 4 -# endif -# ifdef __alpha__ -# define LG_QUANTUM 4 -# endif -# if (defined(__sparc64__) || defined(__sparcv9) || defined(__sparc_v9__)) -# define LG_QUANTUM 4 -# endif -# if (defined(__amd64__) || defined(__x86_64__) || defined(_M_X64)) -# define LG_QUANTUM 4 -# endif -# ifdef __arm__ -# define LG_QUANTUM 3 -# endif -# ifdef __aarch64__ -# define LG_QUANTUM 4 -# endif -# ifdef __hppa__ -# define LG_QUANTUM 4 -# endif -# ifdef __m68k__ -# define LG_QUANTUM 3 -# endif -# ifdef __mips__ -# define LG_QUANTUM 3 -# endif -# ifdef __nios2__ -# define LG_QUANTUM 3 -# endif -# ifdef __or1k__ -# define LG_QUANTUM 3 -# endif -# ifdef __powerpc__ -# define LG_QUANTUM 4 -# endif -# if defined(__riscv) || defined(__riscv__) -# define LG_QUANTUM 4 -# endif -# ifdef __s390__ -# define LG_QUANTUM 4 -# endif -# if (defined (__SH3E__) || defined(__SH4_SINGLE__) || defined(__SH4__) || \ - defined(__SH4_SINGLE_ONLY__)) -# define LG_QUANTUM 4 -# endif -# ifdef __tile__ -# define LG_QUANTUM 4 -# endif -# ifdef __le32__ -# define LG_QUANTUM 4 -# endif -# ifndef LG_QUANTUM -# error "Unknown minimum alignment for architecture; specify via " - "--with-lg-quantum" -# endif -#endif - -#define QUANTUM ((size_t)(1U << LG_QUANTUM)) -#define QUANTUM_MASK (QUANTUM - 1) - -/* Return the smallest quantum multiple that is >= a. */ -#define QUANTUM_CEILING(a) \ - (((a) + QUANTUM_MASK) & ~QUANTUM_MASK) - #define LONG ((size_t)(1U << LG_SIZEOF_LONG)) #define LONG_MASK (LONG - 1) diff --git a/deps/jemalloc/include/jemalloc/internal/jemalloc_preamble.h.in b/deps/jemalloc/include/jemalloc/internal/jemalloc_preamble.h.in index e621fbc85..3418cbfa2 100644 --- a/deps/jemalloc/include/jemalloc/internal/jemalloc_preamble.h.in +++ b/deps/jemalloc/include/jemalloc/internal/jemalloc_preamble.h.in @@ -21,7 +21,7 @@ # include "../jemalloc@install_suffix@.h" #endif -#if (defined(JEMALLOC_OSATOMIC) || defined(JEMALLOC_OSSPIN)) +#if defined(JEMALLOC_OSATOMIC) #include <libkern/OSAtomic.h> #endif @@ -45,7 +45,7 @@ # include "jemalloc/internal/private_namespace_jet.h" # endif #endif -#include "jemalloc/internal/hooks.h" +#include "jemalloc/internal/test_hooks.h" #ifdef JEMALLOC_DEFINE_MADVISE_FREE # define JEMALLOC_MADV_FREE 8 @@ -161,7 +161,26 @@ static const bool config_log = false #endif ; -#ifdef JEMALLOC_HAVE_SCHED_GETCPU +/* + * Are extra safety checks enabled; things like checking the size of sized + * deallocations, double-frees, etc. + */ +static const bool config_opt_safety_checks = +#ifdef JEMALLOC_OPT_SAFETY_CHECKS + true +#elif defined(JEMALLOC_DEBUG) + /* + * This lets us only guard safety checks by one flag instead of two; fast + * checks can guard solely by config_opt_safety_checks and run in debug mode + * too. + */ + true +#else + false +#endif + ; + +#if defined(_WIN32) || defined(JEMALLOC_HAVE_SCHED_GETCPU) /* Currently percpu_arena depends on sched_getcpu. */ #define JEMALLOC_PERCPU_ARENA #endif diff --git a/deps/jemalloc/include/jemalloc/internal/large_externs.h b/deps/jemalloc/include/jemalloc/internal/large_externs.h index 3f36282cd..a05019e8a 100644 --- a/deps/jemalloc/include/jemalloc/internal/large_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/large_externs.h @@ -1,13 +1,16 @@ #ifndef JEMALLOC_INTERNAL_LARGE_EXTERNS_H #define JEMALLOC_INTERNAL_LARGE_EXTERNS_H +#include "jemalloc/internal/hook.h" + void *large_malloc(tsdn_t *tsdn, arena_t *arena, size_t usize, bool zero); void *large_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, size_t alignment, bool zero); bool large_ralloc_no_move(tsdn_t *tsdn, extent_t *extent, size_t usize_min, size_t usize_max, bool zero); -void *large_ralloc(tsdn_t *tsdn, arena_t *arena, extent_t *extent, size_t usize, - size_t alignment, bool zero, tcache_t *tcache); +void *large_ralloc(tsdn_t *tsdn, arena_t *arena, void *ptr, size_t usize, + size_t alignment, bool zero, tcache_t *tcache, + hook_ralloc_args_t *hook_args); typedef void (large_dalloc_junk_t)(void *, size_t); extern large_dalloc_junk_t *JET_MUTABLE large_dalloc_junk; @@ -23,4 +26,7 @@ prof_tctx_t *large_prof_tctx_get(tsdn_t *tsdn, const extent_t *extent); void large_prof_tctx_set(tsdn_t *tsdn, extent_t *extent, prof_tctx_t *tctx); void large_prof_tctx_reset(tsdn_t *tsdn, extent_t *extent); +nstime_t large_prof_alloc_time_get(const extent_t *extent); +void large_prof_alloc_time_set(extent_t *extent, nstime_t time); + #endif /* JEMALLOC_INTERNAL_LARGE_EXTERNS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/malloc_io.h b/deps/jemalloc/include/jemalloc/internal/malloc_io.h index bfe556b52..1d1a414e0 100644 --- a/deps/jemalloc/include/jemalloc/internal/malloc_io.h +++ b/deps/jemalloc/include/jemalloc/internal/malloc_io.h @@ -54,7 +54,7 @@ size_t malloc_vsnprintf(char *str, size_t size, const char *format, size_t malloc_snprintf(char *str, size_t size, const char *format, ...) JEMALLOC_FORMAT_PRINTF(3, 4); /* - * The caller can set write_cb and cbopaque to null to choose to print with the + * The caller can set write_cb to null to choose to print with the * je_malloc_message hook. */ void malloc_vcprintf(void (*write_cb)(void *, const char *), void *cbopaque, diff --git a/deps/jemalloc/include/jemalloc/internal/mutex.h b/deps/jemalloc/include/jemalloc/internal/mutex.h index 6520c2512..7c24f0725 100644 --- a/deps/jemalloc/include/jemalloc/internal/mutex.h +++ b/deps/jemalloc/include/jemalloc/internal/mutex.h @@ -37,14 +37,17 @@ struct malloc_mutex_s { # endif #elif (defined(JEMALLOC_OS_UNFAIR_LOCK)) os_unfair_lock lock; -#elif (defined(JEMALLOC_OSSPIN)) - OSSpinLock lock; #elif (defined(JEMALLOC_MUTEX_INIT_CB)) pthread_mutex_t lock; malloc_mutex_t *postponed_next; #else pthread_mutex_t lock; #endif + /* + * Hint flag to avoid exclusive cache line contention + * during spin waiting + */ + atomic_b_t locked; }; /* * We only touch witness when configured w/ debug. However we @@ -84,10 +87,6 @@ struct malloc_mutex_s { # define MALLOC_MUTEX_LOCK(m) os_unfair_lock_lock(&(m)->lock) # define MALLOC_MUTEX_UNLOCK(m) os_unfair_lock_unlock(&(m)->lock) # define MALLOC_MUTEX_TRYLOCK(m) (!os_unfair_lock_trylock(&(m)->lock)) -#elif (defined(JEMALLOC_OSSPIN)) -# define MALLOC_MUTEX_LOCK(m) OSSpinLockLock(&(m)->lock) -# define MALLOC_MUTEX_UNLOCK(m) OSSpinLockUnlock(&(m)->lock) -# define MALLOC_MUTEX_TRYLOCK(m) (!OSSpinLockTry(&(m)->lock)) #else # define MALLOC_MUTEX_LOCK(m) pthread_mutex_lock(&(m)->lock) # define MALLOC_MUTEX_UNLOCK(m) pthread_mutex_unlock(&(m)->lock) @@ -101,22 +100,37 @@ struct malloc_mutex_s { #ifdef _WIN32 # define MALLOC_MUTEX_INITIALIZER #elif (defined(JEMALLOC_OS_UNFAIR_LOCK)) -# define MALLOC_MUTEX_INITIALIZER \ - {{{LOCK_PROF_DATA_INITIALIZER, OS_UNFAIR_LOCK_INIT}}, \ - WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} -#elif (defined(JEMALLOC_OSSPIN)) -# define MALLOC_MUTEX_INITIALIZER \ - {{{LOCK_PROF_DATA_INITIALIZER, 0}}, \ +# if defined(JEMALLOC_DEBUG) +# define MALLOC_MUTEX_INITIALIZER \ + {{{LOCK_PROF_DATA_INITIALIZER, OS_UNFAIR_LOCK_INIT, ATOMIC_INIT(false)}}, \ + WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT), 0} +# else +# define MALLOC_MUTEX_INITIALIZER \ + {{{LOCK_PROF_DATA_INITIALIZER, OS_UNFAIR_LOCK_INIT, ATOMIC_INIT(false)}}, \ WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} +# endif #elif (defined(JEMALLOC_MUTEX_INIT_CB)) -# define MALLOC_MUTEX_INITIALIZER \ - {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL}}, \ - WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} +# if (defined(JEMALLOC_DEBUG)) +# define MALLOC_MUTEX_INITIALIZER \ + {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL, ATOMIC_INIT(false)}}, \ + WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT), 0} +# else +# define MALLOC_MUTEX_INITIALIZER \ + {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL, ATOMIC_INIT(false)}}, \ + WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} +# endif + #else # define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_DEFAULT +# if defined(JEMALLOC_DEBUG) # define MALLOC_MUTEX_INITIALIZER \ - {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER}}, \ - WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} + {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, ATOMIC_INIT(false)}}, \ + WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT), 0} +# else +# define MALLOC_MUTEX_INITIALIZER \ + {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, ATOMIC_INIT(false)}}, \ + WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)} +# endif #endif #ifdef JEMALLOC_LAZY_LOCK @@ -139,6 +153,7 @@ void malloc_mutex_lock_slow(malloc_mutex_t *mutex); static inline void malloc_mutex_lock_final(malloc_mutex_t *mutex) { MALLOC_MUTEX_LOCK(mutex); + atomic_store_b(&mutex->locked, true, ATOMIC_RELAXED); } static inline bool @@ -164,6 +179,7 @@ malloc_mutex_trylock(tsdn_t *tsdn, malloc_mutex_t *mutex) { witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness); if (isthreaded) { if (malloc_mutex_trylock_final(mutex)) { + atomic_store_b(&mutex->locked, true, ATOMIC_RELAXED); return true; } mutex_owner_stats_update(tsdn, mutex); @@ -203,6 +219,7 @@ malloc_mutex_lock(tsdn_t *tsdn, malloc_mutex_t *mutex) { if (isthreaded) { if (malloc_mutex_trylock_final(mutex)) { malloc_mutex_lock_slow(mutex); + atomic_store_b(&mutex->locked, true, ATOMIC_RELAXED); } mutex_owner_stats_update(tsdn, mutex); } @@ -211,6 +228,7 @@ malloc_mutex_lock(tsdn_t *tsdn, malloc_mutex_t *mutex) { static inline void malloc_mutex_unlock(tsdn_t *tsdn, malloc_mutex_t *mutex) { + atomic_store_b(&mutex->locked, false, ATOMIC_RELAXED); witness_unlock(tsdn_witness_tsdp_get(tsdn), &mutex->witness); if (isthreaded) { MALLOC_MUTEX_UNLOCK(mutex); @@ -245,4 +263,26 @@ malloc_mutex_prof_read(tsdn_t *tsdn, mutex_prof_data_t *data, atomic_store_u32(&data->n_waiting_thds, 0, ATOMIC_RELAXED); } +static inline void +malloc_mutex_prof_accum(tsdn_t *tsdn, mutex_prof_data_t *data, + malloc_mutex_t *mutex) { + mutex_prof_data_t *source = &mutex->prof_data; + /* Can only read holding the mutex. */ + malloc_mutex_assert_owner(tsdn, mutex); + + nstime_add(&data->tot_wait_time, &source->tot_wait_time); + if (nstime_compare(&source->max_wait_time, &data->max_wait_time) > 0) { + nstime_copy(&data->max_wait_time, &source->max_wait_time); + } + data->n_wait_times += source->n_wait_times; + data->n_spin_acquired += source->n_spin_acquired; + if (data->max_n_thds < source->max_n_thds) { + data->max_n_thds = source->max_n_thds; + } + /* n_wait_thds is not reported. */ + atomic_store_u32(&data->n_waiting_thds, 0, ATOMIC_RELAXED); + data->n_owner_switches += source->n_owner_switches; + data->n_lock_ops += source->n_lock_ops; +} + #endif /* JEMALLOC_INTERNAL_MUTEX_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/mutex_prof.h b/deps/jemalloc/include/jemalloc/internal/mutex_prof.h index ce183d335..2cb8fb0cb 100644 --- a/deps/jemalloc/include/jemalloc/internal/mutex_prof.h +++ b/deps/jemalloc/include/jemalloc/internal/mutex_prof.h @@ -35,22 +35,31 @@ typedef enum { mutex_prof_num_arena_mutexes } mutex_prof_arena_ind_t; +/* + * The forth parameter is a boolean value that is true for derived rate counters + * and false for real ones. + */ #define MUTEX_PROF_UINT64_COUNTERS \ - OP(num_ops, uint64_t, "n_lock_ops") \ - OP(num_wait, uint64_t, "n_waiting") \ - OP(num_spin_acq, uint64_t, "n_spin_acq") \ - OP(num_owner_switch, uint64_t, "n_owner_switch") \ - OP(total_wait_time, uint64_t, "total_wait_ns") \ - OP(max_wait_time, uint64_t, "max_wait_ns") + OP(num_ops, uint64_t, "n_lock_ops", false, num_ops) \ + OP(num_ops_ps, uint64_t, "(#/sec)", true, num_ops) \ + OP(num_wait, uint64_t, "n_waiting", false, num_wait) \ + OP(num_wait_ps, uint64_t, "(#/sec)", true, num_wait) \ + OP(num_spin_acq, uint64_t, "n_spin_acq", false, num_spin_acq) \ + OP(num_spin_acq_ps, uint64_t, "(#/sec)", true, num_spin_acq) \ + OP(num_owner_switch, uint64_t, "n_owner_switch", false, num_owner_switch) \ + OP(num_owner_switch_ps, uint64_t, "(#/sec)", true, num_owner_switch) \ + OP(total_wait_time, uint64_t, "total_wait_ns", false, total_wait_time) \ + OP(total_wait_time_ps, uint64_t, "(#/sec)", true, total_wait_time) \ + OP(max_wait_time, uint64_t, "max_wait_ns", false, max_wait_time) #define MUTEX_PROF_UINT32_COUNTERS \ - OP(max_num_thds, uint32_t, "max_n_thds") + OP(max_num_thds, uint32_t, "max_n_thds", false, max_num_thds) #define MUTEX_PROF_COUNTERS \ MUTEX_PROF_UINT64_COUNTERS \ MUTEX_PROF_UINT32_COUNTERS -#define OP(counter, type, human) mutex_counter_##counter, +#define OP(counter, type, human, derived, base_counter) mutex_counter_##counter, #define COUNTER_ENUM(counter_list, t) \ typedef enum { \ diff --git a/deps/jemalloc/include/jemalloc/internal/prof_externs.h b/deps/jemalloc/include/jemalloc/internal/prof_externs.h index 04348696f..094f3e170 100644 --- a/deps/jemalloc/include/jemalloc/internal/prof_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/prof_externs.h @@ -14,6 +14,7 @@ extern bool opt_prof_gdump; /* High-water memory dumping. */ extern bool opt_prof_final; /* Final profile dumping. */ extern bool opt_prof_leak; /* Dump leak summary at exit. */ extern bool opt_prof_accum; /* Report cumulative bytes. */ +extern bool opt_prof_log; /* Turn logging on at boot. */ extern char opt_prof_prefix[ /* Minimize memory bloat for non-prof builds. */ #ifdef JEMALLOC_PROF @@ -45,7 +46,8 @@ extern size_t lg_prof_sample; void prof_alloc_rollback(tsd_t *tsd, prof_tctx_t *tctx, bool updated); void prof_malloc_sample_object(tsdn_t *tsdn, const void *ptr, size_t usize, prof_tctx_t *tctx); -void prof_free_sampled_object(tsd_t *tsd, size_t usize, prof_tctx_t *tctx); +void prof_free_sampled_object(tsd_t *tsd, const void *ptr, size_t usize, + prof_tctx_t *tctx); void bt_init(prof_bt_t *bt, void **vec); void prof_backtrace(prof_bt_t *bt); prof_tctx_t *prof_lookup(tsd_t *tsd, prof_bt_t *bt); @@ -89,4 +91,15 @@ void prof_postfork_parent(tsdn_t *tsdn); void prof_postfork_child(tsdn_t *tsdn); void prof_sample_threshold_update(prof_tdata_t *tdata); +bool prof_log_start(tsdn_t *tsdn, const char *filename); +bool prof_log_stop(tsdn_t *tsdn); +#ifdef JEMALLOC_JET +size_t prof_log_bt_count(void); +size_t prof_log_alloc_count(void); +size_t prof_log_thr_count(void); +bool prof_log_is_logging(void); +bool prof_log_rep_check(void); +void prof_log_dummy_set(bool new_value); +#endif + #endif /* JEMALLOC_INTERNAL_PROF_EXTERNS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/prof_inlines_a.h b/deps/jemalloc/include/jemalloc/internal/prof_inlines_a.h index a6efb4851..471d9853c 100644 --- a/deps/jemalloc/include/jemalloc/internal/prof_inlines_a.h +++ b/deps/jemalloc/include/jemalloc/internal/prof_inlines_a.h @@ -4,7 +4,8 @@ #include "jemalloc/internal/mutex.h" static inline bool -prof_accum_add(tsdn_t *tsdn, prof_accum_t *prof_accum, uint64_t accumbytes) { +prof_accum_add(tsdn_t *tsdn, prof_accum_t *prof_accum, + uint64_t accumbytes) { cassert(config_prof); bool overflow; @@ -42,7 +43,8 @@ prof_accum_add(tsdn_t *tsdn, prof_accum_t *prof_accum, uint64_t accumbytes) { } static inline void -prof_accum_cancel(tsdn_t *tsdn, prof_accum_t *prof_accum, size_t usize) { +prof_accum_cancel(tsdn_t *tsdn, prof_accum_t *prof_accum, + size_t usize) { cassert(config_prof); /* @@ -55,15 +57,15 @@ prof_accum_cancel(tsdn_t *tsdn, prof_accum_t *prof_accum, size_t usize) { #ifdef JEMALLOC_ATOMIC_U64 a0 = atomic_load_u64(&prof_accum->accumbytes, ATOMIC_RELAXED); do { - a1 = (a0 >= LARGE_MINCLASS - usize) ? a0 - (LARGE_MINCLASS - - usize) : 0; + a1 = (a0 >= SC_LARGE_MINCLASS - usize) + ? a0 - (SC_LARGE_MINCLASS - usize) : 0; } while (!atomic_compare_exchange_weak_u64(&prof_accum->accumbytes, &a0, a1, ATOMIC_RELAXED, ATOMIC_RELAXED)); #else malloc_mutex_lock(tsdn, &prof_accum->mtx); a0 = prof_accum->accumbytes; - a1 = (a0 >= LARGE_MINCLASS - usize) ? a0 - (LARGE_MINCLASS - usize) : - 0; + a1 = (a0 >= SC_LARGE_MINCLASS - usize) + ? a0 - (SC_LARGE_MINCLASS - usize) : 0; prof_accum->accumbytes = a1; malloc_mutex_unlock(tsdn, &prof_accum->mtx); #endif diff --git a/deps/jemalloc/include/jemalloc/internal/prof_inlines_b.h b/deps/jemalloc/include/jemalloc/internal/prof_inlines_b.h index 6ff465ad7..8ba8a1e1f 100644 --- a/deps/jemalloc/include/jemalloc/internal/prof_inlines_b.h +++ b/deps/jemalloc/include/jemalloc/internal/prof_inlines_b.h @@ -1,6 +1,7 @@ #ifndef JEMALLOC_INTERNAL_PROF_INLINES_B_H #define JEMALLOC_INTERNAL_PROF_INLINES_B_H +#include "jemalloc/internal/safety_check.h" #include "jemalloc/internal/sz.h" JEMALLOC_ALWAYS_INLINE bool @@ -61,13 +62,54 @@ prof_tctx_reset(tsdn_t *tsdn, const void *ptr, prof_tctx_t *tctx) { arena_prof_tctx_reset(tsdn, ptr, tctx); } +JEMALLOC_ALWAYS_INLINE nstime_t +prof_alloc_time_get(tsdn_t *tsdn, const void *ptr, alloc_ctx_t *alloc_ctx) { + cassert(config_prof); + assert(ptr != NULL); + + return arena_prof_alloc_time_get(tsdn, ptr, alloc_ctx); +} + +JEMALLOC_ALWAYS_INLINE void +prof_alloc_time_set(tsdn_t *tsdn, const void *ptr, alloc_ctx_t *alloc_ctx, + nstime_t t) { + cassert(config_prof); + assert(ptr != NULL); + + arena_prof_alloc_time_set(tsdn, ptr, alloc_ctx, t); +} + +JEMALLOC_ALWAYS_INLINE bool +prof_sample_check(tsd_t *tsd, size_t usize, bool update) { + ssize_t check = update ? 0 : usize; + + int64_t bytes_until_sample = tsd_bytes_until_sample_get(tsd); + if (update) { + bytes_until_sample -= usize; + if (tsd_nominal(tsd)) { + tsd_bytes_until_sample_set(tsd, bytes_until_sample); + } + } + if (likely(bytes_until_sample >= check)) { + return true; + } + + return false; +} + JEMALLOC_ALWAYS_INLINE bool prof_sample_accum_update(tsd_t *tsd, size_t usize, bool update, - prof_tdata_t **tdata_out) { + prof_tdata_t **tdata_out) { prof_tdata_t *tdata; cassert(config_prof); + /* Fastpath: no need to load tdata */ + if (likely(prof_sample_check(tsd, usize, update))) { + return true; + } + + bool booted = tsd_prof_tdata_get(tsd); tdata = prof_tdata_get(tsd, true); if (unlikely((uintptr_t)tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)) { tdata = NULL; @@ -81,21 +123,23 @@ prof_sample_accum_update(tsd_t *tsd, size_t usize, bool update, return true; } - if (likely(tdata->bytes_until_sample >= usize)) { - if (update) { - tdata->bytes_until_sample -= usize; - } + /* + * If this was the first creation of tdata, then + * prof_tdata_get() reset bytes_until_sample, so decrement and + * check it again + */ + if (!booted && prof_sample_check(tsd, usize, update)) { return true; - } else { - if (tsd_reentrancy_level_get(tsd) > 0) { - return true; - } - /* Compute new sample threshold. */ - if (update) { - prof_sample_threshold_update(tdata); - } - return !tdata->active; } + + if (tsd_reentrancy_level_get(tsd) > 0) { + return true; + } + /* Compute new sample threshold. */ + if (update) { + prof_sample_threshold_update(tdata); + } + return !tdata->active; } JEMALLOC_ALWAYS_INLINE prof_tctx_t * @@ -187,7 +231,7 @@ prof_realloc(tsd_t *tsd, const void *ptr, size_t usize, prof_tctx_t *tctx, * counters. */ if (unlikely(old_sampled)) { - prof_free_sampled_object(tsd, old_usize, old_tctx); + prof_free_sampled_object(tsd, ptr, old_usize, old_tctx); } } @@ -199,7 +243,7 @@ prof_free(tsd_t *tsd, const void *ptr, size_t usize, alloc_ctx_t *alloc_ctx) { assert(usize == isalloc(tsd_tsdn(tsd), ptr)); if (unlikely((uintptr_t)tctx > (uintptr_t)1U)) { - prof_free_sampled_object(tsd, usize, tctx); + prof_free_sampled_object(tsd, ptr, usize, tctx); } } diff --git a/deps/jemalloc/include/jemalloc/internal/prof_structs.h b/deps/jemalloc/include/jemalloc/internal/prof_structs.h index 0d58ae100..34ed4822b 100644 --- a/deps/jemalloc/include/jemalloc/internal/prof_structs.h +++ b/deps/jemalloc/include/jemalloc/internal/prof_structs.h @@ -169,7 +169,6 @@ struct prof_tdata_s { /* Sampling state. */ uint64_t prng_state; - uint64_t bytes_until_sample; /* State used to avoid dumping while operating on prof internals. */ bool enq; diff --git a/deps/jemalloc/include/jemalloc/internal/quantum.h b/deps/jemalloc/include/jemalloc/internal/quantum.h new file mode 100644 index 000000000..821086e99 --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/quantum.h @@ -0,0 +1,77 @@ +#ifndef JEMALLOC_INTERNAL_QUANTUM_H +#define JEMALLOC_INTERNAL_QUANTUM_H + +/* + * Minimum allocation alignment is 2^LG_QUANTUM bytes (ignoring tiny size + * classes). + */ +#ifndef LG_QUANTUM +# if (defined(__i386__) || defined(_M_IX86)) +# define LG_QUANTUM 4 +# endif +# ifdef __ia64__ +# define LG_QUANTUM 4 +# endif +# ifdef __alpha__ +# define LG_QUANTUM 4 +# endif +# if (defined(__sparc64__) || defined(__sparcv9) || defined(__sparc_v9__)) +# define LG_QUANTUM 4 +# endif +# if (defined(__amd64__) || defined(__x86_64__) || defined(_M_X64)) +# define LG_QUANTUM 4 +# endif +# ifdef __arm__ +# define LG_QUANTUM 3 +# endif +# ifdef __aarch64__ +# define LG_QUANTUM 4 +# endif +# ifdef __hppa__ +# define LG_QUANTUM 4 +# endif +# ifdef __m68k__ +# define LG_QUANTUM 3 +# endif +# ifdef __mips__ +# define LG_QUANTUM 3 +# endif +# ifdef __nios2__ +# define LG_QUANTUM 3 +# endif +# ifdef __or1k__ +# define LG_QUANTUM 3 +# endif +# ifdef __powerpc__ +# define LG_QUANTUM 4 +# endif +# if defined(__riscv) || defined(__riscv__) +# define LG_QUANTUM 4 +# endif +# ifdef __s390__ +# define LG_QUANTUM 4 +# endif +# if (defined (__SH3E__) || defined(__SH4_SINGLE__) || defined(__SH4__) || \ + defined(__SH4_SINGLE_ONLY__)) +# define LG_QUANTUM 4 +# endif +# ifdef __tile__ +# define LG_QUANTUM 4 +# endif +# ifdef __le32__ +# define LG_QUANTUM 4 +# endif +# ifndef LG_QUANTUM +# error "Unknown minimum alignment for architecture; specify via " + "--with-lg-quantum" +# endif +#endif + +#define QUANTUM ((size_t)(1U << LG_QUANTUM)) +#define QUANTUM_MASK (QUANTUM - 1) + +/* Return the smallest quantum multiple that is >= a. */ +#define QUANTUM_CEILING(a) \ + (((a) + QUANTUM_MASK) & ~QUANTUM_MASK) + +#endif /* JEMALLOC_INTERNAL_QUANTUM_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/rtree.h b/deps/jemalloc/include/jemalloc/internal/rtree.h index b59d33a80..16ccbebee 100644 --- a/deps/jemalloc/include/jemalloc/internal/rtree.h +++ b/deps/jemalloc/include/jemalloc/internal/rtree.h @@ -4,7 +4,7 @@ #include "jemalloc/internal/atomic.h" #include "jemalloc/internal/mutex.h" #include "jemalloc/internal/rtree_tsd.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/tsd.h" /* @@ -31,7 +31,7 @@ # error Unsupported number of significant virtual address bits #endif /* Use compact leaf representation if virtual address encoding allows. */ -#if RTREE_NHIB >= LG_CEIL_NSIZES +#if RTREE_NHIB >= LG_CEIL(SC_NSIZES) # define RTREE_LEAF_COMPACT #endif @@ -170,8 +170,8 @@ rtree_subkey(uintptr_t key, unsigned level) { */ # ifdef RTREE_LEAF_COMPACT JEMALLOC_ALWAYS_INLINE uintptr_t -rtree_leaf_elm_bits_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, - bool dependent) { +rtree_leaf_elm_bits_read(tsdn_t *tsdn, rtree_t *rtree, + rtree_leaf_elm_t *elm, bool dependent) { return (uintptr_t)atomic_load_p(&elm->le_bits, dependent ? ATOMIC_RELAXED : ATOMIC_ACQUIRE); } @@ -208,7 +208,7 @@ rtree_leaf_elm_bits_slab_get(uintptr_t bits) { # endif JEMALLOC_ALWAYS_INLINE extent_t * -rtree_leaf_elm_extent_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_extent_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, bool dependent) { #ifdef RTREE_LEAF_COMPACT uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, dependent); @@ -221,7 +221,7 @@ rtree_leaf_elm_extent_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } JEMALLOC_ALWAYS_INLINE szind_t -rtree_leaf_elm_szind_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, bool dependent) { #ifdef RTREE_LEAF_COMPACT uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, dependent); @@ -233,7 +233,7 @@ rtree_leaf_elm_szind_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } JEMALLOC_ALWAYS_INLINE bool -rtree_leaf_elm_slab_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, bool dependent) { #ifdef RTREE_LEAF_COMPACT uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, dependent); @@ -245,7 +245,7 @@ rtree_leaf_elm_slab_read(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } static inline void -rtree_leaf_elm_extent_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_extent_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, extent_t *extent) { #ifdef RTREE_LEAF_COMPACT uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, true); @@ -259,9 +259,9 @@ rtree_leaf_elm_extent_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } static inline void -rtree_leaf_elm_szind_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_szind_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, szind_t szind) { - assert(szind <= NSIZES); + assert(szind <= SC_NSIZES); #ifdef RTREE_LEAF_COMPACT uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, @@ -277,7 +277,7 @@ rtree_leaf_elm_szind_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } static inline void -rtree_leaf_elm_slab_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, +rtree_leaf_elm_slab_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, bool slab) { #ifdef RTREE_LEAF_COMPACT uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, @@ -292,8 +292,8 @@ rtree_leaf_elm_slab_write(UNUSED tsdn_t *tsdn, UNUSED rtree_t *rtree, } static inline void -rtree_leaf_elm_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, - extent_t *extent, szind_t szind, bool slab) { +rtree_leaf_elm_write(tsdn_t *tsdn, rtree_t *rtree, + rtree_leaf_elm_t *elm, extent_t *extent, szind_t szind, bool slab) { #ifdef RTREE_LEAF_COMPACT uintptr_t bits = ((uintptr_t)szind << LG_VADDR) | ((uintptr_t)extent & (((uintptr_t)0x1 << LG_VADDR) - 1)) | @@ -313,7 +313,7 @@ rtree_leaf_elm_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, static inline void rtree_leaf_elm_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm, szind_t szind, bool slab) { - assert(!slab || szind < NBINS); + assert(!slab || szind < SC_NBINS); /* * The caller implicitly assures that it is the only writer to the szind @@ -429,7 +429,7 @@ rtree_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key, dependent); if (!dependent && elm == NULL) { - return NSIZES; + return SC_NSIZES; } return rtree_leaf_elm_szind_read(tsdn, rtree, elm, dependent); } @@ -452,6 +452,42 @@ rtree_extent_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, return false; } +/* + * Try to read szind_slab from the L1 cache. Returns true on a hit, + * and fills in r_szind and r_slab. Otherwise returns false. + * + * Key is allowed to be NULL in order to save an extra branch on the + * fastpath. returns false in this case. + */ +JEMALLOC_ALWAYS_INLINE bool +rtree_szind_slab_read_fast(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, + uintptr_t key, szind_t *r_szind, bool *r_slab) { + rtree_leaf_elm_t *elm; + + size_t slot = rtree_cache_direct_map(key); + uintptr_t leafkey = rtree_leafkey(key); + assert(leafkey != RTREE_LEAFKEY_INVALID); + + if (likely(rtree_ctx->cache[slot].leafkey == leafkey)) { + rtree_leaf_elm_t *leaf = rtree_ctx->cache[slot].leaf; + assert(leaf != NULL); + uintptr_t subkey = rtree_subkey(key, RTREE_HEIGHT-1); + elm = &leaf[subkey]; + +#ifdef RTREE_LEAF_COMPACT + uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, + elm, true); + *r_szind = rtree_leaf_elm_bits_szind_get(bits); + *r_slab = rtree_leaf_elm_bits_slab_get(bits); +#else + *r_szind = rtree_leaf_elm_szind_read(tsdn, rtree, elm, true); + *r_slab = rtree_leaf_elm_slab_read(tsdn, rtree, elm, true); +#endif + return true; + } else { + return false; + } +} JEMALLOC_ALWAYS_INLINE bool rtree_szind_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent, szind_t *r_szind, bool *r_slab) { @@ -474,7 +510,7 @@ rtree_szind_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, static inline void rtree_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, uintptr_t key, szind_t szind, bool slab) { - assert(!slab || szind < NBINS); + assert(!slab || szind < SC_NBINS); rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key, true); rtree_leaf_elm_szind_slab_update(tsdn, rtree, elm, szind, slab); @@ -486,7 +522,7 @@ rtree_clear(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key, true); assert(rtree_leaf_elm_extent_read(tsdn, rtree, elm, false) != NULL); - rtree_leaf_elm_write(tsdn, rtree, elm, NULL, NSIZES, false); + rtree_leaf_elm_write(tsdn, rtree, elm, NULL, SC_NSIZES, false); } #endif /* JEMALLOC_INTERNAL_RTREE_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/rtree_tsd.h b/deps/jemalloc/include/jemalloc/internal/rtree_tsd.h index 93a75173a..562e29297 100644 --- a/deps/jemalloc/include/jemalloc/internal/rtree_tsd.h +++ b/deps/jemalloc/include/jemalloc/internal/rtree_tsd.h @@ -26,7 +26,7 @@ * Zero initializer required for tsd initialization only. Proper initialization * done via rtree_ctx_data_init(). */ -#define RTREE_CTX_ZERO_INITIALIZER {{{0}}, {{0}}} +#define RTREE_CTX_ZERO_INITIALIZER {{{0, 0}}, {{0, 0}}} typedef struct rtree_leaf_elm_s rtree_leaf_elm_t; diff --git a/deps/jemalloc/include/jemalloc/internal/safety_check.h b/deps/jemalloc/include/jemalloc/internal/safety_check.h new file mode 100644 index 000000000..53339ac12 --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/safety_check.h @@ -0,0 +1,26 @@ +#ifndef JEMALLOC_INTERNAL_SAFETY_CHECK_H +#define JEMALLOC_INTERNAL_SAFETY_CHECK_H + +void safety_check_fail(const char *format, ...); +/* Can set to NULL for a default. */ +void safety_check_set_abort(void (*abort_fn)()); + +JEMALLOC_ALWAYS_INLINE void +safety_check_set_redzone(void *ptr, size_t usize, size_t bumped_usize) { + assert(usize < bumped_usize); + for (size_t i = usize; i < bumped_usize && i < usize + 32; ++i) { + *((unsigned char *)ptr + i) = 0xBC; + } +} + +JEMALLOC_ALWAYS_INLINE void +safety_check_verify_redzone(const void *ptr, size_t usize, size_t bumped_usize) +{ + for (size_t i = usize; i < bumped_usize && i < usize + 32; ++i) { + if (unlikely(*((unsigned char *)ptr + i) != 0xBC)) { + safety_check_fail("Use after free error\n"); + } + } +} + +#endif /*JEMALLOC_INTERNAL_SAFETY_CHECK_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/sc.h b/deps/jemalloc/include/jemalloc/internal/sc.h new file mode 100644 index 000000000..9a099d8b6 --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/sc.h @@ -0,0 +1,333 @@ +#ifndef JEMALLOC_INTERNAL_SC_H +#define JEMALLOC_INTERNAL_SC_H + +#include "jemalloc/internal/jemalloc_internal_types.h" + +/* + * Size class computations: + * + * These are a little tricky; we'll first start by describing how things + * generally work, and then describe some of the details. + * + * Ignore the first few size classes for a moment. We can then split all the + * remaining size classes into groups. The size classes in a group are spaced + * such that they cover allocation request sizes in a power-of-2 range. The + * power of two is called the base of the group, and the size classes in it + * satisfy allocations in the half-open range (base, base * 2]. There are + * SC_NGROUP size classes in each group, equally spaced in the range, so that + * each one covers allocations for base / SC_NGROUP possible allocation sizes. + * We call that value (base / SC_NGROUP) the delta of the group. Each size class + * is delta larger than the one before it (including the initial size class in a + * group, which is delta larger than base, the largest size class in the + * previous group). + * To make the math all work out nicely, we require that SC_NGROUP is a power of + * two, and define it in terms of SC_LG_NGROUP. We'll often talk in terms of + * lg_base and lg_delta. For each of these groups then, we have that + * lg_delta == lg_base - SC_LG_NGROUP. + * The size classes in a group with a given lg_base and lg_delta (which, recall, + * can be computed from lg_base for these groups) are therefore: + * base + 1 * delta + * which covers allocations in (base, base + 1 * delta] + * base + 2 * delta + * which covers allocations in (base + 1 * delta, base + 2 * delta]. + * base + 3 * delta + * which covers allocations in (base + 2 * delta, base + 3 * delta]. + * ... + * base + SC_NGROUP * delta ( == 2 * base) + * which covers allocations in (base + (SC_NGROUP - 1) * delta, 2 * base]. + * (Note that currently SC_NGROUP is always 4, so the "..." is empty in + * practice.) + * Note that the last size class in the group is the next power of two (after + * base), so that we've set up the induction correctly for the next group's + * selection of delta. + * + * Now, let's start considering the first few size classes. Two extra constants + * come into play here: LG_QUANTUM and SC_LG_TINY_MIN. LG_QUANTUM ensures + * correct platform alignment; all objects of size (1 << LG_QUANTUM) or larger + * are at least (1 << LG_QUANTUM) aligned; this can be used to ensure that we + * never return improperly aligned memory, by making (1 << LG_QUANTUM) equal the + * highest required alignment of a platform. For allocation sizes smaller than + * (1 << LG_QUANTUM) though, we can be more relaxed (since we don't support + * platforms with types with alignment larger than their size). To allow such + * allocations (without wasting space unnecessarily), we introduce tiny size + * classes; one per power of two, up until we hit the quantum size. There are + * therefore LG_QUANTUM - SC_LG_TINY_MIN such size classes. + * + * Next, we have a size class of size (1 << LG_QUANTUM). This can't be the + * start of a group in the sense we described above (covering a power of two + * range) since, if we divided into it to pick a value of delta, we'd get a + * delta smaller than (1 << LG_QUANTUM) for sizes >= (1 << LG_QUANTUM), which + * is against the rules. + * + * The first base we can divide by SC_NGROUP while still being at least + * (1 << LG_QUANTUM) is SC_NGROUP * (1 << LG_QUANTUM). We can get there by + * having SC_NGROUP size classes, spaced (1 << LG_QUANTUM) apart. These size + * classes are: + * 1 * (1 << LG_QUANTUM) + * 2 * (1 << LG_QUANTUM) + * 3 * (1 << LG_QUANTUM) + * ... (although, as above, this "..." is empty in practice) + * SC_NGROUP * (1 << LG_QUANTUM). + * + * There are SC_NGROUP of these size classes, so we can regard it as a sort of + * pseudo-group, even though it spans multiple powers of 2, is divided + * differently, and both starts and ends on a power of 2 (as opposed to just + * ending). SC_NGROUP is itself a power of two, so the first group after the + * pseudo-group has the power-of-two base SC_NGROUP * (1 << LG_QUANTUM), for a + * lg_base of LG_QUANTUM + SC_LG_NGROUP. We can divide this base into SC_NGROUP + * sizes without violating our LG_QUANTUM requirements, so we can safely set + * lg_delta = lg_base - SC_LG_GROUP (== LG_QUANTUM). + * + * So, in order, the size classes are: + * + * Tiny size classes: + * - Count: LG_QUANTUM - SC_LG_TINY_MIN. + * - Sizes: + * 1 << SC_LG_TINY_MIN + * 1 << (SC_LG_TINY_MIN + 1) + * 1 << (SC_LG_TINY_MIN + 2) + * ... + * 1 << (LG_QUANTUM - 1) + * + * Initial pseudo-group: + * - Count: SC_NGROUP + * - Sizes: + * 1 * (1 << LG_QUANTUM) + * 2 * (1 << LG_QUANTUM) + * 3 * (1 << LG_QUANTUM) + * ... + * SC_NGROUP * (1 << LG_QUANTUM) + * + * Regular group 0: + * - Count: SC_NGROUP + * - Sizes: + * (relative to lg_base of LG_QUANTUM + SC_LG_NGROUP and lg_delta of + * lg_base - SC_LG_NGROUP) + * (1 << lg_base) + 1 * (1 << lg_delta) + * (1 << lg_base) + 2 * (1 << lg_delta) + * (1 << lg_base) + 3 * (1 << lg_delta) + * ... + * (1 << lg_base) + SC_NGROUP * (1 << lg_delta) [ == (1 << (lg_base + 1)) ] + * + * Regular group 1: + * - Count: SC_NGROUP + * - Sizes: + * (relative to lg_base of LG_QUANTUM + SC_LG_NGROUP + 1 and lg_delta of + * lg_base - SC_LG_NGROUP) + * (1 << lg_base) + 1 * (1 << lg_delta) + * (1 << lg_base) + 2 * (1 << lg_delta) + * (1 << lg_base) + 3 * (1 << lg_delta) + * ... + * (1 << lg_base) + SC_NGROUP * (1 << lg_delta) [ == (1 << (lg_base + 1)) ] + * + * ... + * + * Regular group N: + * - Count: SC_NGROUP + * - Sizes: + * (relative to lg_base of LG_QUANTUM + SC_LG_NGROUP + N and lg_delta of + * lg_base - SC_LG_NGROUP) + * (1 << lg_base) + 1 * (1 << lg_delta) + * (1 << lg_base) + 2 * (1 << lg_delta) + * (1 << lg_base) + 3 * (1 << lg_delta) + * ... + * (1 << lg_base) + SC_NGROUP * (1 << lg_delta) [ == (1 << (lg_base + 1)) ] + * + * + * Representation of metadata: + * To make the math easy, we'll mostly work in lg quantities. We record lg_base, + * lg_delta, and ndelta (i.e. number of deltas above the base) on a + * per-size-class basis, and maintain the invariant that, across all size + * classes, size == (1 << lg_base) + ndelta * (1 << lg_delta). + * + * For regular groups (i.e. those with lg_base >= LG_QUANTUM + SC_LG_NGROUP), + * lg_delta is lg_base - SC_LG_NGROUP, and ndelta goes from 1 to SC_NGROUP. + * + * For the initial tiny size classes (if any), lg_base is lg(size class size). + * lg_delta is lg_base for the first size class, and lg_base - 1 for all + * subsequent ones. ndelta is always 0. + * + * For the pseudo-group, if there are no tiny size classes, then we set + * lg_base == LG_QUANTUM, lg_delta == LG_QUANTUM, and have ndelta range from 0 + * to SC_NGROUP - 1. (Note that delta == base, so base + (SC_NGROUP - 1) * delta + * is just SC_NGROUP * base, or (1 << (SC_LG_NGROUP + LG_QUANTUM)), so we do + * indeed get a power of two that way). If there *are* tiny size classes, then + * the first size class needs to have lg_delta relative to the largest tiny size + * class. We therefore set lg_base == LG_QUANTUM - 1, + * lg_delta == LG_QUANTUM - 1, and ndelta == 1, keeping the rest of the + * pseudo-group the same. + * + * + * Other terminology: + * "Small" size classes mean those that are allocated out of bins, which is the + * same as those that are slab allocated. + * "Large" size classes are those that are not small. The cutoff for counting as + * large is page size * group size. + */ + +/* + * Size class N + (1 << SC_LG_NGROUP) twice the size of size class N. + */ +#define SC_LG_NGROUP 2 +#define SC_LG_TINY_MIN 3 + +#if SC_LG_TINY_MIN == 0 +/* The div module doesn't support division by 1, which this would require. */ +#error "Unsupported LG_TINY_MIN" +#endif + +/* + * The definitions below are all determined by the above settings and system + * characteristics. + */ +#define SC_NGROUP (1ULL << SC_LG_NGROUP) +#define SC_PTR_BITS ((1ULL << LG_SIZEOF_PTR) * 8) +#define SC_NTINY (LG_QUANTUM - SC_LG_TINY_MIN) +#define SC_LG_TINY_MAXCLASS (LG_QUANTUM > SC_LG_TINY_MIN ? LG_QUANTUM - 1 : -1) +#define SC_NPSEUDO SC_NGROUP +#define SC_LG_FIRST_REGULAR_BASE (LG_QUANTUM + SC_LG_NGROUP) +/* + * We cap allocations to be less than 2 ** (ptr_bits - 1), so the highest base + * we need is 2 ** (ptr_bits - 2). (This also means that the last group is 1 + * size class shorter than the others). + * We could probably save some space in arenas by capping this at LG_VADDR size. + */ +#define SC_LG_BASE_MAX (SC_PTR_BITS - 2) +#define SC_NREGULAR (SC_NGROUP * \ + (SC_LG_BASE_MAX - SC_LG_FIRST_REGULAR_BASE + 1) - 1) +#define SC_NSIZES (SC_NTINY + SC_NPSEUDO + SC_NREGULAR) + +/* The number of size classes that are a multiple of the page size. */ +#define SC_NPSIZES ( \ + /* Start with all the size classes. */ \ + SC_NSIZES \ + /* Subtract out those groups with too small a base. */ \ + - (LG_PAGE - 1 - SC_LG_FIRST_REGULAR_BASE) * SC_NGROUP \ + /* And the pseudo-group. */ \ + - SC_NPSEUDO \ + /* And the tiny group. */ \ + - SC_NTINY \ + /* Sizes where ndelta*delta is not a multiple of the page size. */ \ + - (SC_LG_NGROUP * SC_NGROUP)) +/* + * Note that the last line is computed as the sum of the second column in the + * following table: + * lg(base) | count of sizes to exclude + * ------------------------------|----------------------------- + * LG_PAGE - 1 | SC_NGROUP - 1 + * LG_PAGE | SC_NGROUP - 1 + * LG_PAGE + 1 | SC_NGROUP - 2 + * LG_PAGE + 2 | SC_NGROUP - 4 + * ... | ... + * LG_PAGE + (SC_LG_NGROUP - 1) | SC_NGROUP - (SC_NGROUP / 2) + */ + +/* + * We declare a size class is binnable if size < page size * group. Or, in other + * words, lg(size) < lg(page size) + lg(group size). + */ +#define SC_NBINS ( \ + /* Sub-regular size classes. */ \ + SC_NTINY + SC_NPSEUDO \ + /* Groups with lg_regular_min_base <= lg_base <= lg_base_max */ \ + + SC_NGROUP * (LG_PAGE + SC_LG_NGROUP - SC_LG_FIRST_REGULAR_BASE) \ + /* Last SC of the last group hits the bound exactly; exclude it. */ \ + - 1) + +/* + * The size2index_tab lookup table uses uint8_t to encode each bin index, so we + * cannot support more than 256 small size classes. + */ +#if (SC_NBINS > 256) +# error "Too many small size classes" +#endif + +/* The largest size class in the lookup table. */ +#define SC_LOOKUP_MAXCLASS ((size_t)1 << 12) + +/* Internal, only used for the definition of SC_SMALL_MAXCLASS. */ +#define SC_SMALL_MAX_BASE ((size_t)1 << (LG_PAGE + SC_LG_NGROUP - 1)) +#define SC_SMALL_MAX_DELTA ((size_t)1 << (LG_PAGE - 1)) + +/* The largest size class allocated out of a slab. */ +#define SC_SMALL_MAXCLASS (SC_SMALL_MAX_BASE \ + + (SC_NGROUP - 1) * SC_SMALL_MAX_DELTA) + +/* The smallest size class not allocated out of a slab. */ +#define SC_LARGE_MINCLASS ((size_t)1ULL << (LG_PAGE + SC_LG_NGROUP)) +#define SC_LG_LARGE_MINCLASS (LG_PAGE + SC_LG_NGROUP) + +/* Internal; only used for the definition of SC_LARGE_MAXCLASS. */ +#define SC_MAX_BASE ((size_t)1 << (SC_PTR_BITS - 2)) +#define SC_MAX_DELTA ((size_t)1 << (SC_PTR_BITS - 2 - SC_LG_NGROUP)) + +/* The largest size class supported. */ +#define SC_LARGE_MAXCLASS (SC_MAX_BASE + (SC_NGROUP - 1) * SC_MAX_DELTA) + +typedef struct sc_s sc_t; +struct sc_s { + /* Size class index, or -1 if not a valid size class. */ + int index; + /* Lg group base size (no deltas added). */ + int lg_base; + /* Lg delta to previous size class. */ + int lg_delta; + /* Delta multiplier. size == 1<<lg_base + ndelta<<lg_delta */ + int ndelta; + /* + * True if the size class is a multiple of the page size, false + * otherwise. + */ + bool psz; + /* + * True if the size class is a small, bin, size class. False otherwise. + */ + bool bin; + /* The slab page count if a small bin size class, 0 otherwise. */ + int pgs; + /* Same as lg_delta if a lookup table size class, 0 otherwise. */ + int lg_delta_lookup; +}; + +typedef struct sc_data_s sc_data_t; +struct sc_data_s { + /* Number of tiny size classes. */ + unsigned ntiny; + /* Number of bins supported by the lookup table. */ + int nlbins; + /* Number of small size class bins. */ + int nbins; + /* Number of size classes. */ + int nsizes; + /* Number of bits required to store NSIZES. */ + int lg_ceil_nsizes; + /* Number of size classes that are a multiple of (1U << LG_PAGE). */ + unsigned npsizes; + /* Lg of maximum tiny size class (or -1, if none). */ + int lg_tiny_maxclass; + /* Maximum size class included in lookup table. */ + size_t lookup_maxclass; + /* Maximum small size class. */ + size_t small_maxclass; + /* Lg of minimum large size class. */ + int lg_large_minclass; + /* The minimum large size class. */ + size_t large_minclass; + /* Maximum (large) size class. */ + size_t large_maxclass; + /* True if the sc_data_t has been initialized (for debugging only). */ + bool initialized; + + sc_t sc[SC_NSIZES]; +}; + +void sc_data_init(sc_data_t *data); +/* + * Updates slab sizes in [begin, end] to be pgs pages in length, if possible. + * Otherwise, does its best to accomodate the request. + */ +void sc_data_update_slab_size(sc_data_t *data, size_t begin, size_t end, + int pgs); +void sc_boot(sc_data_t *data); + +#endif /* JEMALLOC_INTERNAL_SC_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/seq.h b/deps/jemalloc/include/jemalloc/internal/seq.h new file mode 100644 index 000000000..ef2df4c6e --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/seq.h @@ -0,0 +1,55 @@ +#ifndef JEMALLOC_INTERNAL_SEQ_H +#define JEMALLOC_INTERNAL_SEQ_H + +#include "jemalloc/internal/atomic.h" + +/* + * A simple seqlock implementation. + */ + +#define seq_define(type, short_type) \ +typedef struct { \ + atomic_zu_t seq; \ + atomic_zu_t data[ \ + (sizeof(type) + sizeof(size_t) - 1) / sizeof(size_t)]; \ +} seq_##short_type##_t; \ + \ +/* \ + * No internal synchronization -- the caller must ensure that there's \ + * only a single writer at a time. \ + */ \ +static inline void \ +seq_store_##short_type(seq_##short_type##_t *dst, type *src) { \ + size_t buf[sizeof(dst->data) / sizeof(size_t)]; \ + buf[sizeof(buf) / sizeof(size_t) - 1] = 0; \ + memcpy(buf, src, sizeof(type)); \ + size_t old_seq = atomic_load_zu(&dst->seq, ATOMIC_RELAXED); \ + atomic_store_zu(&dst->seq, old_seq + 1, ATOMIC_RELAXED); \ + atomic_fence(ATOMIC_RELEASE); \ + for (size_t i = 0; i < sizeof(buf) / sizeof(size_t); i++) { \ + atomic_store_zu(&dst->data[i], buf[i], ATOMIC_RELAXED); \ + } \ + atomic_store_zu(&dst->seq, old_seq + 2, ATOMIC_RELEASE); \ +} \ + \ +/* Returns whether or not the read was consistent. */ \ +static inline bool \ +seq_try_load_##short_type(type *dst, seq_##short_type##_t *src) { \ + size_t buf[sizeof(src->data) / sizeof(size_t)]; \ + size_t seq1 = atomic_load_zu(&src->seq, ATOMIC_ACQUIRE); \ + if (seq1 % 2 != 0) { \ + return false; \ + } \ + for (size_t i = 0; i < sizeof(buf) / sizeof(size_t); i++) { \ + buf[i] = atomic_load_zu(&src->data[i], ATOMIC_RELAXED); \ + } \ + atomic_fence(ATOMIC_ACQUIRE); \ + size_t seq2 = atomic_load_zu(&src->seq, ATOMIC_RELAXED); \ + if (seq1 != seq2) { \ + return false; \ + } \ + memcpy(dst, buf, sizeof(type)); \ + return true; \ +} + +#endif /* JEMALLOC_INTERNAL_SEQ_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/size_classes.sh b/deps/jemalloc/include/jemalloc/internal/size_classes.sh deleted file mode 100755 index 998994d09..000000000 --- a/deps/jemalloc/include/jemalloc/internal/size_classes.sh +++ /dev/null @@ -1,361 +0,0 @@ -#!/bin/sh -# -# Usage: size_classes.sh <lg_qarr> <lg_tmin> <lg_parr> <lg_g> - -# The following limits are chosen such that they cover all supported platforms. - -# Pointer sizes. -lg_zarr="2 3" - -# Quanta. -lg_qarr=$1 - -# The range of tiny size classes is [2^lg_tmin..2^(lg_q-1)]. -lg_tmin=$2 - -# Maximum lookup size. -lg_kmax=12 - -# Page sizes. -lg_parr=`echo $3 | tr ',' ' '` - -# Size class group size (number of size classes for each size doubling). -lg_g=$4 - -pow2() { - e=$1 - pow2_result=1 - while [ ${e} -gt 0 ] ; do - pow2_result=$((${pow2_result} + ${pow2_result})) - e=$((${e} - 1)) - done -} - -lg() { - x=$1 - lg_result=0 - while [ ${x} -gt 1 ] ; do - lg_result=$((${lg_result} + 1)) - x=$((${x} / 2)) - done -} - -lg_ceil() { - y=$1 - lg ${y}; lg_floor=${lg_result} - pow2 ${lg_floor}; pow2_floor=${pow2_result} - if [ ${pow2_floor} -lt ${y} ] ; then - lg_ceil_result=$((${lg_floor} + 1)) - else - lg_ceil_result=${lg_floor} - fi -} - -reg_size_compute() { - lg_grp=$1 - lg_delta=$2 - ndelta=$3 - - pow2 ${lg_grp}; grp=${pow2_result} - pow2 ${lg_delta}; delta=${pow2_result} - reg_size=$((${grp} + ${delta}*${ndelta})) -} - -slab_size() { - lg_p=$1 - lg_grp=$2 - lg_delta=$3 - ndelta=$4 - - pow2 ${lg_p}; p=${pow2_result} - reg_size_compute ${lg_grp} ${lg_delta} ${ndelta} - - # Compute smallest slab size that is an integer multiple of reg_size. - try_slab_size=${p} - try_nregs=$((${try_slab_size} / ${reg_size})) - perfect=0 - while [ ${perfect} -eq 0 ] ; do - perfect_slab_size=${try_slab_size} - perfect_nregs=${try_nregs} - - try_slab_size=$((${try_slab_size} + ${p})) - try_nregs=$((${try_slab_size} / ${reg_size})) - if [ ${perfect_slab_size} -eq $((${perfect_nregs} * ${reg_size})) ] ; then - perfect=1 - fi - done - - slab_size_pgs=$((${perfect_slab_size} / ${p})) -} - -size_class() { - index=$1 - lg_grp=$2 - lg_delta=$3 - ndelta=$4 - lg_p=$5 - lg_kmax=$6 - - if [ ${lg_delta} -ge ${lg_p} ] ; then - psz="yes" - else - pow2 ${lg_p}; p=${pow2_result} - pow2 ${lg_grp}; grp=${pow2_result} - pow2 ${lg_delta}; delta=${pow2_result} - sz=$((${grp} + ${delta} * ${ndelta})) - npgs=$((${sz} / ${p})) - if [ ${sz} -eq $((${npgs} * ${p})) ] ; then - psz="yes" - else - psz="no" - fi - fi - - lg ${ndelta}; lg_ndelta=${lg_result}; pow2 ${lg_ndelta} - if [ ${pow2_result} -lt ${ndelta} ] ; then - rem="yes" - else - rem="no" - fi - - lg_size=${lg_grp} - if [ $((${lg_delta} + ${lg_ndelta})) -eq ${lg_grp} ] ; then - lg_size=$((${lg_grp} + 1)) - else - lg_size=${lg_grp} - rem="yes" - fi - - if [ ${lg_size} -lt $((${lg_p} + ${lg_g})) ] ; then - bin="yes" - slab_size ${lg_p} ${lg_grp} ${lg_delta} ${ndelta}; pgs=${slab_size_pgs} - else - bin="no" - pgs=0 - fi - if [ ${lg_size} -lt ${lg_kmax} \ - -o ${lg_size} -eq ${lg_kmax} -a ${rem} = "no" ] ; then - lg_delta_lookup=${lg_delta} - else - lg_delta_lookup="no" - fi - printf ' SC(%3d, %6d, %8d, %6d, %3s, %3s, %3d, %2s) \\\n' ${index} ${lg_grp} ${lg_delta} ${ndelta} ${psz} ${bin} ${pgs} ${lg_delta_lookup} - # Defined upon return: - # - psz ("yes" or "no") - # - bin ("yes" or "no") - # - pgs - # - lg_delta_lookup (${lg_delta} or "no") -} - -sep_line() { - echo " \\" -} - -size_classes() { - lg_z=$1 - lg_q=$2 - lg_t=$3 - lg_p=$4 - lg_g=$5 - - pow2 $((${lg_z} + 3)); ptr_bits=${pow2_result} - pow2 ${lg_g}; g=${pow2_result} - - echo "#define SIZE_CLASSES \\" - echo " /* index, lg_grp, lg_delta, ndelta, psz, bin, pgs, lg_delta_lookup */ \\" - - ntbins=0 - nlbins=0 - lg_tiny_maxclass='"NA"' - nbins=0 - npsizes=0 - - # Tiny size classes. - ndelta=0 - index=0 - lg_grp=${lg_t} - lg_delta=${lg_grp} - while [ ${lg_grp} -lt ${lg_q} ] ; do - size_class ${index} ${lg_grp} ${lg_delta} ${ndelta} ${lg_p} ${lg_kmax} - if [ ${lg_delta_lookup} != "no" ] ; then - nlbins=$((${index} + 1)) - fi - if [ ${psz} = "yes" ] ; then - npsizes=$((${npsizes} + 1)) - fi - if [ ${bin} != "no" ] ; then - nbins=$((${index} + 1)) - fi - ntbins=$((${ntbins} + 1)) - lg_tiny_maxclass=${lg_grp} # Final written value is correct. - index=$((${index} + 1)) - lg_delta=${lg_grp} - lg_grp=$((${lg_grp} + 1)) - done - - # First non-tiny group. - if [ ${ntbins} -gt 0 ] ; then - sep_line - # The first size class has an unusual encoding, because the size has to be - # split between grp and delta*ndelta. - lg_grp=$((${lg_grp} - 1)) - ndelta=1 - size_class ${index} ${lg_grp} ${lg_delta} ${ndelta} ${lg_p} ${lg_kmax} - index=$((${index} + 1)) - lg_grp=$((${lg_grp} + 1)) - lg_delta=$((${lg_delta} + 1)) - if [ ${psz} = "yes" ] ; then - npsizes=$((${npsizes} + 1)) - fi - fi - while [ ${ndelta} -lt ${g} ] ; do - size_class ${index} ${lg_grp} ${lg_delta} ${ndelta} ${lg_p} ${lg_kmax} - index=$((${index} + 1)) - ndelta=$((${ndelta} + 1)) - if [ ${psz} = "yes" ] ; then - npsizes=$((${npsizes} + 1)) - fi - done - - # All remaining groups. - lg_grp=$((${lg_grp} + ${lg_g})) - while [ ${lg_grp} -lt $((${ptr_bits} - 1)) ] ; do - sep_line - ndelta=1 - if [ ${lg_grp} -eq $((${ptr_bits} - 2)) ] ; then - ndelta_limit=$((${g} - 1)) - else - ndelta_limit=${g} - fi - while [ ${ndelta} -le ${ndelta_limit} ] ; do - size_class ${index} ${lg_grp} ${lg_delta} ${ndelta} ${lg_p} ${lg_kmax} - if [ ${lg_delta_lookup} != "no" ] ; then - nlbins=$((${index} + 1)) - # Final written value is correct: - lookup_maxclass="((((size_t)1) << ${lg_grp}) + (((size_t)${ndelta}) << ${lg_delta}))" - fi - if [ ${psz} = "yes" ] ; then - npsizes=$((${npsizes} + 1)) - fi - if [ ${bin} != "no" ] ; then - nbins=$((${index} + 1)) - # Final written value is correct: - small_maxclass="((((size_t)1) << ${lg_grp}) + (((size_t)${ndelta}) << ${lg_delta}))" - if [ ${lg_g} -gt 0 ] ; then - lg_large_minclass=$((${lg_grp} + 1)) - else - lg_large_minclass=$((${lg_grp} + 2)) - fi - fi - # Final written value is correct: - large_maxclass="((((size_t)1) << ${lg_grp}) + (((size_t)${ndelta}) << ${lg_delta}))" - index=$((${index} + 1)) - ndelta=$((${ndelta} + 1)) - done - lg_grp=$((${lg_grp} + 1)) - lg_delta=$((${lg_delta} + 1)) - done - echo - nsizes=${index} - lg_ceil ${nsizes}; lg_ceil_nsizes=${lg_ceil_result} - - # Defined upon completion: - # - ntbins - # - nlbins - # - nbins - # - nsizes - # - lg_ceil_nsizes - # - npsizes - # - lg_tiny_maxclass - # - lookup_maxclass - # - small_maxclass - # - lg_large_minclass - # - large_maxclass -} - -cat <<EOF -#ifndef JEMALLOC_INTERNAL_SIZE_CLASSES_H -#define JEMALLOC_INTERNAL_SIZE_CLASSES_H - -/* This file was automatically generated by size_classes.sh. */ - -#include "jemalloc/internal/jemalloc_internal_types.h" - -/* - * This header file defines: - * - * LG_SIZE_CLASS_GROUP: Lg of size class count for each size doubling. - * LG_TINY_MIN: Lg of minimum size class to support. - * SIZE_CLASSES: Complete table of SC(index, lg_grp, lg_delta, ndelta, psz, - * bin, pgs, lg_delta_lookup) tuples. - * index: Size class index. - * lg_grp: Lg group base size (no deltas added). - * lg_delta: Lg delta to previous size class. - * ndelta: Delta multiplier. size == 1<<lg_grp + ndelta<<lg_delta - * psz: 'yes' if a multiple of the page size, 'no' otherwise. - * bin: 'yes' if a small bin size class, 'no' otherwise. - * pgs: Slab page count if a small bin size class, 0 otherwise. - * lg_delta_lookup: Same as lg_delta if a lookup table size class, 'no' - * otherwise. - * NTBINS: Number of tiny bins. - * NLBINS: Number of bins supported by the lookup table. - * NBINS: Number of small size class bins. - * NSIZES: Number of size classes. - * LG_CEIL_NSIZES: Number of bits required to store NSIZES. - * NPSIZES: Number of size classes that are a multiple of (1U << LG_PAGE). - * LG_TINY_MAXCLASS: Lg of maximum tiny size class. - * LOOKUP_MAXCLASS: Maximum size class included in lookup table. - * SMALL_MAXCLASS: Maximum small size class. - * LG_LARGE_MINCLASS: Lg of minimum large size class. - * LARGE_MAXCLASS: Maximum (large) size class. - */ - -#define LG_SIZE_CLASS_GROUP ${lg_g} -#define LG_TINY_MIN ${lg_tmin} - -EOF - -for lg_z in ${lg_zarr} ; do - for lg_q in ${lg_qarr} ; do - lg_t=${lg_tmin} - while [ ${lg_t} -le ${lg_q} ] ; do - # Iterate through page sizes and compute how many bins there are. - for lg_p in ${lg_parr} ; do - echo "#if (LG_SIZEOF_PTR == ${lg_z} && LG_TINY_MIN == ${lg_t} && LG_QUANTUM == ${lg_q} && LG_PAGE == ${lg_p})" - size_classes ${lg_z} ${lg_q} ${lg_t} ${lg_p} ${lg_g} - echo "#define SIZE_CLASSES_DEFINED" - echo "#define NTBINS ${ntbins}" - echo "#define NLBINS ${nlbins}" - echo "#define NBINS ${nbins}" - echo "#define NSIZES ${nsizes}" - echo "#define LG_CEIL_NSIZES ${lg_ceil_nsizes}" - echo "#define NPSIZES ${npsizes}" - echo "#define LG_TINY_MAXCLASS ${lg_tiny_maxclass}" - echo "#define LOOKUP_MAXCLASS ${lookup_maxclass}" - echo "#define SMALL_MAXCLASS ${small_maxclass}" - echo "#define LG_LARGE_MINCLASS ${lg_large_minclass}" - echo "#define LARGE_MINCLASS (ZU(1) << LG_LARGE_MINCLASS)" - echo "#define LARGE_MAXCLASS ${large_maxclass}" - echo "#endif" - echo - done - lg_t=$((${lg_t} + 1)) - done - done -done - -cat <<EOF -#ifndef SIZE_CLASSES_DEFINED -# error "No size class definitions match configuration" -#endif -#undef SIZE_CLASSES_DEFINED -/* - * The size2index_tab lookup table uses uint8_t to encode each bin index, so we - * cannot support more than 256 small size classes. - */ -#if (NBINS > 256) -# error "Too many small size classes" -#endif - -#endif /* JEMALLOC_INTERNAL_SIZE_CLASSES_H */ -EOF diff --git a/deps/jemalloc/include/jemalloc/internal/stats.h b/deps/jemalloc/include/jemalloc/internal/stats.h index 852e34269..3b9e0eac1 100644 --- a/deps/jemalloc/include/jemalloc/internal/stats.h +++ b/deps/jemalloc/include/jemalloc/internal/stats.h @@ -10,7 +10,8 @@ OPTION('a', unmerged, config_stats, false) \ OPTION('b', bins, true, false) \ OPTION('l', large, true, false) \ - OPTION('x', mutex, true, false) + OPTION('x', mutex, true, false) \ + OPTION('e', extents, true, false) enum { #define OPTION(o, v, d, s) stats_print_option_num_##v, diff --git a/deps/jemalloc/include/jemalloc/internal/sz.h b/deps/jemalloc/include/jemalloc/internal/sz.h index 979462898..68e558abf 100644 --- a/deps/jemalloc/include/jemalloc/internal/sz.h +++ b/deps/jemalloc/include/jemalloc/internal/sz.h @@ -3,7 +3,7 @@ #include "jemalloc/internal/bit_util.h" #include "jemalloc/internal/pages.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/util.h" /* @@ -26,18 +26,18 @@ * sz_pind2sz_tab encodes the same information as could be computed by * sz_pind2sz_compute(). */ -extern size_t const sz_pind2sz_tab[NPSIZES+1]; +extern size_t sz_pind2sz_tab[SC_NPSIZES + 1]; /* * sz_index2size_tab encodes the same information as could be computed (at * unacceptable cost in some code paths) by sz_index2size_compute(). */ -extern size_t const sz_index2size_tab[NSIZES]; +extern size_t sz_index2size_tab[SC_NSIZES]; /* * sz_size2index_tab is a compact lookup table that rounds request sizes up to * size classes. In order to reduce cache footprint, the table is compressed, * and all accesses are via sz_size2index(). */ -extern uint8_t const sz_size2index_tab[]; +extern uint8_t sz_size2index_tab[]; static const size_t sz_large_pad = #ifdef JEMALLOC_CACHE_OBLIVIOUS @@ -47,49 +47,47 @@ static const size_t sz_large_pad = #endif ; +extern void sz_boot(const sc_data_t *sc_data); + JEMALLOC_ALWAYS_INLINE pszind_t sz_psz2ind(size_t psz) { - if (unlikely(psz > LARGE_MAXCLASS)) { - return NPSIZES; + if (unlikely(psz > SC_LARGE_MAXCLASS)) { + return SC_NPSIZES; } - { - pszind_t x = lg_floor((psz<<1)-1); - pszind_t shift = (x < LG_SIZE_CLASS_GROUP + LG_PAGE) ? 0 : x - - (LG_SIZE_CLASS_GROUP + LG_PAGE); - pszind_t grp = shift << LG_SIZE_CLASS_GROUP; + pszind_t x = lg_floor((psz<<1)-1); + pszind_t shift = (x < SC_LG_NGROUP + LG_PAGE) ? + 0 : x - (SC_LG_NGROUP + LG_PAGE); + pszind_t grp = shift << SC_LG_NGROUP; - pszind_t lg_delta = (x < LG_SIZE_CLASS_GROUP + LG_PAGE + 1) ? - LG_PAGE : x - LG_SIZE_CLASS_GROUP - 1; + pszind_t lg_delta = (x < SC_LG_NGROUP + LG_PAGE + 1) ? + LG_PAGE : x - SC_LG_NGROUP - 1; - size_t delta_inverse_mask = ZU(-1) << lg_delta; - pszind_t mod = ((((psz-1) & delta_inverse_mask) >> lg_delta)) & - ((ZU(1) << LG_SIZE_CLASS_GROUP) - 1); + size_t delta_inverse_mask = ZU(-1) << lg_delta; + pszind_t mod = ((((psz-1) & delta_inverse_mask) >> lg_delta)) & + ((ZU(1) << SC_LG_NGROUP) - 1); - pszind_t ind = grp + mod; - return ind; - } + pszind_t ind = grp + mod; + return ind; } static inline size_t sz_pind2sz_compute(pszind_t pind) { - if (unlikely(pind == NPSIZES)) { - return LARGE_MAXCLASS + PAGE; + if (unlikely(pind == SC_NPSIZES)) { + return SC_LARGE_MAXCLASS + PAGE; } - { - size_t grp = pind >> LG_SIZE_CLASS_GROUP; - size_t mod = pind & ((ZU(1) << LG_SIZE_CLASS_GROUP) - 1); + size_t grp = pind >> SC_LG_NGROUP; + size_t mod = pind & ((ZU(1) << SC_LG_NGROUP) - 1); - size_t grp_size_mask = ~((!!grp)-1); - size_t grp_size = ((ZU(1) << (LG_PAGE + - (LG_SIZE_CLASS_GROUP-1))) << grp) & grp_size_mask; + size_t grp_size_mask = ~((!!grp)-1); + size_t grp_size = ((ZU(1) << (LG_PAGE + (SC_LG_NGROUP-1))) << grp) + & grp_size_mask; - size_t shift = (grp == 0) ? 1 : grp; - size_t lg_delta = shift + (LG_PAGE-1); - size_t mod_size = (mod+1) << lg_delta; + size_t shift = (grp == 0) ? 1 : grp; + size_t lg_delta = shift + (LG_PAGE-1); + size_t mod_size = (mod+1) << lg_delta; - size_t sz = grp_size + mod_size; - return sz; - } + size_t sz = grp_size + mod_size; + return sz; } static inline size_t @@ -101,70 +99,70 @@ sz_pind2sz_lookup(pszind_t pind) { static inline size_t sz_pind2sz(pszind_t pind) { - assert(pind < NPSIZES+1); + assert(pind < SC_NPSIZES + 1); return sz_pind2sz_lookup(pind); } static inline size_t sz_psz2u(size_t psz) { - if (unlikely(psz > LARGE_MAXCLASS)) { - return LARGE_MAXCLASS + PAGE; - } - { - size_t x = lg_floor((psz<<1)-1); - size_t lg_delta = (x < LG_SIZE_CLASS_GROUP + LG_PAGE + 1) ? - LG_PAGE : x - LG_SIZE_CLASS_GROUP - 1; - size_t delta = ZU(1) << lg_delta; - size_t delta_mask = delta - 1; - size_t usize = (psz + delta_mask) & ~delta_mask; - return usize; + if (unlikely(psz > SC_LARGE_MAXCLASS)) { + return SC_LARGE_MAXCLASS + PAGE; } + size_t x = lg_floor((psz<<1)-1); + size_t lg_delta = (x < SC_LG_NGROUP + LG_PAGE + 1) ? + LG_PAGE : x - SC_LG_NGROUP - 1; + size_t delta = ZU(1) << lg_delta; + size_t delta_mask = delta - 1; + size_t usize = (psz + delta_mask) & ~delta_mask; + return usize; } static inline szind_t sz_size2index_compute(size_t size) { - if (unlikely(size > LARGE_MAXCLASS)) { - return NSIZES; + if (unlikely(size > SC_LARGE_MAXCLASS)) { + return SC_NSIZES; + } + + if (size == 0) { + return 0; } -#if (NTBINS != 0) - if (size <= (ZU(1) << LG_TINY_MAXCLASS)) { - szind_t lg_tmin = LG_TINY_MAXCLASS - NTBINS + 1; +#if (SC_NTINY != 0) + if (size <= (ZU(1) << SC_LG_TINY_MAXCLASS)) { + szind_t lg_tmin = SC_LG_TINY_MAXCLASS - SC_NTINY + 1; szind_t lg_ceil = lg_floor(pow2_ceil_zu(size)); return (lg_ceil < lg_tmin ? 0 : lg_ceil - lg_tmin); } #endif { szind_t x = lg_floor((size<<1)-1); - szind_t shift = (x < LG_SIZE_CLASS_GROUP + LG_QUANTUM) ? 0 : - x - (LG_SIZE_CLASS_GROUP + LG_QUANTUM); - szind_t grp = shift << LG_SIZE_CLASS_GROUP; + szind_t shift = (x < SC_LG_NGROUP + LG_QUANTUM) ? 0 : + x - (SC_LG_NGROUP + LG_QUANTUM); + szind_t grp = shift << SC_LG_NGROUP; - szind_t lg_delta = (x < LG_SIZE_CLASS_GROUP + LG_QUANTUM + 1) - ? LG_QUANTUM : x - LG_SIZE_CLASS_GROUP - 1; + szind_t lg_delta = (x < SC_LG_NGROUP + LG_QUANTUM + 1) + ? LG_QUANTUM : x - SC_LG_NGROUP - 1; size_t delta_inverse_mask = ZU(-1) << lg_delta; szind_t mod = ((((size-1) & delta_inverse_mask) >> lg_delta)) & - ((ZU(1) << LG_SIZE_CLASS_GROUP) - 1); + ((ZU(1) << SC_LG_NGROUP) - 1); - szind_t index = NTBINS + grp + mod; + szind_t index = SC_NTINY + grp + mod; return index; } } JEMALLOC_ALWAYS_INLINE szind_t sz_size2index_lookup(size_t size) { - assert(size <= LOOKUP_MAXCLASS); - { - szind_t ret = (sz_size2index_tab[(size-1) >> LG_TINY_MIN]); - assert(ret == sz_size2index_compute(size)); - return ret; - } + assert(size <= SC_LOOKUP_MAXCLASS); + szind_t ret = (sz_size2index_tab[(size + (ZU(1) << SC_LG_TINY_MIN) - 1) + >> SC_LG_TINY_MIN]); + assert(ret == sz_size2index_compute(size)); + return ret; } JEMALLOC_ALWAYS_INLINE szind_t sz_size2index(size_t size) { - assert(size > 0); - if (likely(size <= LOOKUP_MAXCLASS)) { + if (likely(size <= SC_LOOKUP_MAXCLASS)) { return sz_size2index_lookup(size); } return sz_size2index_compute(size); @@ -172,20 +170,20 @@ sz_size2index(size_t size) { static inline size_t sz_index2size_compute(szind_t index) { -#if (NTBINS > 0) - if (index < NTBINS) { - return (ZU(1) << (LG_TINY_MAXCLASS - NTBINS + 1 + index)); +#if (SC_NTINY > 0) + if (index < SC_NTINY) { + return (ZU(1) << (SC_LG_TINY_MAXCLASS - SC_NTINY + 1 + index)); } #endif { - size_t reduced_index = index - NTBINS; - size_t grp = reduced_index >> LG_SIZE_CLASS_GROUP; - size_t mod = reduced_index & ((ZU(1) << LG_SIZE_CLASS_GROUP) - + size_t reduced_index = index - SC_NTINY; + size_t grp = reduced_index >> SC_LG_NGROUP; + size_t mod = reduced_index & ((ZU(1) << SC_LG_NGROUP) - 1); size_t grp_size_mask = ~((!!grp)-1); size_t grp_size = ((ZU(1) << (LG_QUANTUM + - (LG_SIZE_CLASS_GROUP-1))) << grp) & grp_size_mask; + (SC_LG_NGROUP-1))) << grp) & grp_size_mask; size_t shift = (grp == 0) ? 1 : grp; size_t lg_delta = shift + (LG_QUANTUM-1); @@ -205,18 +203,22 @@ sz_index2size_lookup(szind_t index) { JEMALLOC_ALWAYS_INLINE size_t sz_index2size(szind_t index) { - assert(index < NSIZES); + assert(index < SC_NSIZES); return sz_index2size_lookup(index); } JEMALLOC_ALWAYS_INLINE size_t sz_s2u_compute(size_t size) { - if (unlikely(size > LARGE_MAXCLASS)) { + if (unlikely(size > SC_LARGE_MAXCLASS)) { return 0; } -#if (NTBINS > 0) - if (size <= (ZU(1) << LG_TINY_MAXCLASS)) { - size_t lg_tmin = LG_TINY_MAXCLASS - NTBINS + 1; + + if (size == 0) { + size++; + } +#if (SC_NTINY > 0) + if (size <= (ZU(1) << SC_LG_TINY_MAXCLASS)) { + size_t lg_tmin = SC_LG_TINY_MAXCLASS - SC_NTINY + 1; size_t lg_ceil = lg_floor(pow2_ceil_zu(size)); return (lg_ceil < lg_tmin ? (ZU(1) << lg_tmin) : (ZU(1) << lg_ceil)); @@ -224,8 +226,8 @@ sz_s2u_compute(size_t size) { #endif { size_t x = lg_floor((size<<1)-1); - size_t lg_delta = (x < LG_SIZE_CLASS_GROUP + LG_QUANTUM + 1) - ? LG_QUANTUM : x - LG_SIZE_CLASS_GROUP - 1; + size_t lg_delta = (x < SC_LG_NGROUP + LG_QUANTUM + 1) + ? LG_QUANTUM : x - SC_LG_NGROUP - 1; size_t delta = ZU(1) << lg_delta; size_t delta_mask = delta - 1; size_t usize = (size + delta_mask) & ~delta_mask; @@ -247,8 +249,7 @@ sz_s2u_lookup(size_t size) { */ JEMALLOC_ALWAYS_INLINE size_t sz_s2u(size_t size) { - assert(size > 0); - if (likely(size <= LOOKUP_MAXCLASS)) { + if (likely(size <= SC_LOOKUP_MAXCLASS)) { return sz_s2u_lookup(size); } return sz_s2u_compute(size); @@ -265,7 +266,7 @@ sz_sa2u(size_t size, size_t alignment) { assert(alignment != 0 && ((alignment - 1) & alignment) == 0); /* Try for a small size class. */ - if (size <= SMALL_MAXCLASS && alignment < PAGE) { + if (size <= SC_SMALL_MAXCLASS && alignment < PAGE) { /* * Round size up to the nearest multiple of alignment. * @@ -281,20 +282,20 @@ sz_sa2u(size_t size, size_t alignment) { * 192 | 11000000 | 64 */ usize = sz_s2u(ALIGNMENT_CEILING(size, alignment)); - if (usize < LARGE_MINCLASS) { + if (usize < SC_LARGE_MINCLASS) { return usize; } } /* Large size class. Beware of overflow. */ - if (unlikely(alignment > LARGE_MAXCLASS)) { + if (unlikely(alignment > SC_LARGE_MAXCLASS)) { return 0; } /* Make sure result is a large size class. */ - if (size <= LARGE_MINCLASS) { - usize = LARGE_MINCLASS; + if (size <= SC_LARGE_MINCLASS) { + usize = SC_LARGE_MINCLASS; } else { usize = sz_s2u(size); if (usize < size) { diff --git a/deps/jemalloc/include/jemalloc/internal/tcache_externs.h b/deps/jemalloc/include/jemalloc/internal/tcache_externs.h index 790367bd4..d63eafde8 100644 --- a/deps/jemalloc/include/jemalloc/internal/tcache_externs.h +++ b/deps/jemalloc/include/jemalloc/internal/tcache_externs.h @@ -1,15 +1,13 @@ #ifndef JEMALLOC_INTERNAL_TCACHE_EXTERNS_H #define JEMALLOC_INTERNAL_TCACHE_EXTERNS_H -#include "jemalloc/internal/size_classes.h" - extern bool opt_tcache; extern ssize_t opt_lg_tcache_max; extern cache_bin_info_t *tcache_bin_info; /* - * Number of tcache bins. There are NBINS small-object bins, plus 0 or more + * Number of tcache bins. There are SC_NBINS small-object bins, plus 0 or more * large-object bins. */ extern unsigned nhbins; diff --git a/deps/jemalloc/include/jemalloc/internal/tcache_inlines.h b/deps/jemalloc/include/jemalloc/internal/tcache_inlines.h index 0f6ab8cb5..5eca20e89 100644 --- a/deps/jemalloc/include/jemalloc/internal/tcache_inlines.h +++ b/deps/jemalloc/include/jemalloc/internal/tcache_inlines.h @@ -3,7 +3,7 @@ #include "jemalloc/internal/bin.h" #include "jemalloc/internal/jemalloc_internal_types.h" -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/sz.h" #include "jemalloc/internal/ticker.h" #include "jemalloc/internal/util.h" @@ -40,13 +40,13 @@ tcache_event(tsd_t *tsd, tcache_t *tcache) { JEMALLOC_ALWAYS_INLINE void * tcache_alloc_small(tsd_t *tsd, arena_t *arena, tcache_t *tcache, - UNUSED size_t size, szind_t binind, bool zero, bool slow_path) { + size_t size, szind_t binind, bool zero, bool slow_path) { void *ret; cache_bin_t *bin; bool tcache_success; size_t usize JEMALLOC_CC_SILENCE_INIT(0); - assert(binind < NBINS); + assert(binind < SC_NBINS); bin = tcache_small_bin_get(tcache, binind); ret = cache_bin_alloc_easy(bin, &tcache_success); assert(tcache_success == (ret != NULL)); @@ -107,7 +107,7 @@ tcache_alloc_large(tsd_t *tsd, arena_t *arena, tcache_t *tcache, size_t size, cache_bin_t *bin; bool tcache_success; - assert(binind >= NBINS &&binind < nhbins); + assert(binind >= SC_NBINS &&binind < nhbins); bin = tcache_large_bin_get(tcache, binind); ret = cache_bin_alloc_easy(bin, &tcache_success); assert(tcache_success == (ret != NULL)); @@ -166,7 +166,8 @@ tcache_dalloc_small(tsd_t *tsd, tcache_t *tcache, void *ptr, szind_t binind, cache_bin_t *bin; cache_bin_info_t *bin_info; - assert(tcache_salloc(tsd_tsdn(tsd), ptr) <= SMALL_MAXCLASS); + assert(tcache_salloc(tsd_tsdn(tsd), ptr) + <= SC_SMALL_MAXCLASS); if (slow_path && config_fill && unlikely(opt_junk_free)) { arena_dalloc_junk_small(ptr, &bin_infos[binind]); @@ -174,13 +175,12 @@ tcache_dalloc_small(tsd_t *tsd, tcache_t *tcache, void *ptr, szind_t binind, bin = tcache_small_bin_get(tcache, binind); bin_info = &tcache_bin_info[binind]; - if (unlikely(bin->ncached == bin_info->ncached_max)) { + if (unlikely(!cache_bin_dalloc_easy(bin, bin_info, ptr))) { tcache_bin_flush_small(tsd, tcache, bin, binind, (bin_info->ncached_max >> 1)); + bool ret = cache_bin_dalloc_easy(bin, bin_info, ptr); + assert(ret); } - assert(bin->ncached < bin_info->ncached_max); - bin->ncached++; - *(bin->avail - bin->ncached) = ptr; tcache_event(tsd, tcache); } @@ -191,7 +191,8 @@ tcache_dalloc_large(tsd_t *tsd, tcache_t *tcache, void *ptr, szind_t binind, cache_bin_t *bin; cache_bin_info_t *bin_info; - assert(tcache_salloc(tsd_tsdn(tsd), ptr) > SMALL_MAXCLASS); + assert(tcache_salloc(tsd_tsdn(tsd), ptr) + > SC_SMALL_MAXCLASS); assert(tcache_salloc(tsd_tsdn(tsd), ptr) <= tcache_maxclass); if (slow_path && config_fill && unlikely(opt_junk_free)) { @@ -215,6 +216,9 @@ JEMALLOC_ALWAYS_INLINE tcache_t * tcaches_get(tsd_t *tsd, unsigned ind) { tcaches_t *elm = &tcaches[ind]; if (unlikely(elm->tcache == NULL)) { + malloc_printf("<jemalloc>: invalid tcache id (%u).\n", ind); + abort(); + } else if (unlikely(elm->tcache == TCACHES_ELM_NEED_REINIT)) { elm->tcache = tcache_create_explicit(tsd); } return elm->tcache; diff --git a/deps/jemalloc/include/jemalloc/internal/tcache_structs.h b/deps/jemalloc/include/jemalloc/internal/tcache_structs.h index 07b738705..172ef9040 100644 --- a/deps/jemalloc/include/jemalloc/internal/tcache_structs.h +++ b/deps/jemalloc/include/jemalloc/internal/tcache_structs.h @@ -1,10 +1,14 @@ #ifndef JEMALLOC_INTERNAL_TCACHE_STRUCTS_H #define JEMALLOC_INTERNAL_TCACHE_STRUCTS_H -#include "jemalloc/internal/ql.h" -#include "jemalloc/internal/size_classes.h" #include "jemalloc/internal/cache_bin.h" +#include "jemalloc/internal/ql.h" +#include "jemalloc/internal/sc.h" #include "jemalloc/internal/ticker.h" +#include "jemalloc/internal/tsd_types.h" + +/* Various uses of this struct need it to be a named type. */ +typedef ql_elm(tsd_t) tsd_link_t; struct tcache_s { /* @@ -21,7 +25,7 @@ struct tcache_s { * During tcache initialization, the avail pointer in each element of * tbins is initialized to point to the proper offset within this array. */ - cache_bin_t bins_small[NBINS]; + cache_bin_t bins_small[SC_NBINS]; /* * This data is less hot; we can be a little less careful with our @@ -29,6 +33,11 @@ struct tcache_s { */ /* Lets us track all the tcaches in an arena. */ ql_elm(tcache_t) link; + + /* Logically scoped to tsd, but put here for cache layout reasons. */ + ql_elm(tsd_t) tsd_link; + bool in_hook; + /* * The descriptor lets the arena find our cache bins without seeing the * tcache definition. This enables arenas to aggregate stats across @@ -41,13 +50,13 @@ struct tcache_s { /* Next bin to GC. */ szind_t next_gc_bin; /* For small bins, fill (ncached_max >> lg_fill_div). */ - uint8_t lg_fill_div[NBINS]; + uint8_t lg_fill_div[SC_NBINS]; /* * We put the cache bins for large size classes at the end of the * struct, since some of them might not get used. This might end up * letting us avoid touching an extra page if we don't have to. */ - cache_bin_t bins_large[NSIZES-NBINS]; + cache_bin_t bins_large[SC_NSIZES-SC_NBINS]; }; /* Linkage for list of available (previously used) explicit tcache IDs. */ diff --git a/deps/jemalloc/include/jemalloc/internal/tcache_types.h b/deps/jemalloc/include/jemalloc/internal/tcache_types.h index e49bc9d79..dce69382e 100644 --- a/deps/jemalloc/include/jemalloc/internal/tcache_types.h +++ b/deps/jemalloc/include/jemalloc/internal/tcache_types.h @@ -1,7 +1,7 @@ #ifndef JEMALLOC_INTERNAL_TCACHE_TYPES_H #define JEMALLOC_INTERNAL_TCACHE_TYPES_H -#include "jemalloc/internal/size_classes.h" +#include "jemalloc/internal/sc.h" typedef struct tcache_s tcache_t; typedef struct tcaches_s tcaches_t; @@ -45,7 +45,7 @@ typedef struct tcaches_s tcaches_t; /* Number of tcache allocation/deallocation events between incremental GCs. */ #define TCACHE_GC_INCR \ - ((TCACHE_GC_SWEEP / NBINS) + ((TCACHE_GC_SWEEP / NBINS == 0) ? 0 : 1)) + ((TCACHE_GC_SWEEP / SC_NBINS) + ((TCACHE_GC_SWEEP / SC_NBINS == 0) ? 0 : 1)) /* Used in TSD static initializer only. Real init in tcache_data_init(). */ #define TCACHE_ZERO_INITIALIZER {0} @@ -53,4 +53,7 @@ typedef struct tcaches_s tcaches_t; /* Used in TSD static initializer only. Will be initialized to opt_tcache. */ #define TCACHE_ENABLED_ZERO_INITIALIZER false +/* Used for explicit tcache only. Means flushed but not destroyed. */ +#define TCACHES_ELM_NEED_REINIT ((tcache_t *)(uintptr_t)1) + #endif /* JEMALLOC_INTERNAL_TCACHE_TYPES_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/test_hooks.h b/deps/jemalloc/include/jemalloc/internal/test_hooks.h new file mode 100644 index 000000000..a6351e59a --- /dev/null +++ b/deps/jemalloc/include/jemalloc/internal/test_hooks.h @@ -0,0 +1,19 @@ +#ifndef JEMALLOC_INTERNAL_TEST_HOOKS_H +#define JEMALLOC_INTERNAL_TEST_HOOKS_H + +extern JEMALLOC_EXPORT void (*test_hooks_arena_new_hook)(); +extern JEMALLOC_EXPORT void (*test_hooks_libc_hook)(); + +#define JEMALLOC_HOOK(fn, hook) ((void)(hook != NULL && (hook(), 0)), fn) + +#define open JEMALLOC_HOOK(open, test_hooks_libc_hook) +#define read JEMALLOC_HOOK(read, test_hooks_libc_hook) +#define write JEMALLOC_HOOK(write, test_hooks_libc_hook) +#define readlink JEMALLOC_HOOK(readlink, test_hooks_libc_hook) +#define close JEMALLOC_HOOK(close, test_hooks_libc_hook) +#define creat JEMALLOC_HOOK(creat, test_hooks_libc_hook) +#define secure_getenv JEMALLOC_HOOK(secure_getenv, test_hooks_libc_hook) +/* Note that this is undef'd and re-define'd in src/prof.c. */ +#define _Unwind_Backtrace JEMALLOC_HOOK(_Unwind_Backtrace, test_hooks_libc_hook) + +#endif /* JEMALLOC_INTERNAL_TEST_HOOKS_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/ticker.h b/deps/jemalloc/include/jemalloc/internal/ticker.h index 4b3604708..52d0db4c8 100644 --- a/deps/jemalloc/include/jemalloc/internal/ticker.h +++ b/deps/jemalloc/include/jemalloc/internal/ticker.h @@ -75,4 +75,17 @@ ticker_tick(ticker_t *ticker) { return ticker_ticks(ticker, 1); } +/* + * Try to tick. If ticker would fire, return true, but rely on + * slowpath to reset ticker. + */ +static inline bool +ticker_trytick(ticker_t *ticker) { + --ticker->tick; + if (unlikely(ticker->tick < 0)) { + return true; + } + return false; +} + #endif /* JEMALLOC_INTERNAL_TICKER_H */ diff --git a/deps/jemalloc/include/jemalloc/internal/tsd.h b/deps/jemalloc/include/jemalloc/internal/tsd.h index 0b9841aa7..9ba260045 100644 --- a/deps/jemalloc/include/jemalloc/internal/tsd.h +++ b/deps/jemalloc/include/jemalloc/internal/tsd.h @@ -3,6 +3,7 @@ #include "jemalloc/internal/arena_types.h" #include "jemalloc/internal/assert.h" +#include "jemalloc/internal/bin_types.h" #include "jemalloc/internal/jemalloc_internal_externs.h" #include "jemalloc/internal/prof_types.h" #include "jemalloc/internal/ql.h" @@ -68,17 +69,19 @@ typedef void (*test_callback_t)(int *); O(offset_state, uint64_t, uint64_t) \ O(thread_allocated, uint64_t, uint64_t) \ O(thread_deallocated, uint64_t, uint64_t) \ + O(bytes_until_sample, int64_t, int64_t) \ O(prof_tdata, prof_tdata_t *, prof_tdata_t *) \ O(rtree_ctx, rtree_ctx_t, rtree_ctx_t) \ O(iarena, arena_t *, arena_t *) \ O(arena, arena_t *, arena_t *) \ O(arenas_tdata, arena_tdata_t *, arena_tdata_t *)\ + O(binshards, tsd_binshards_t, tsd_binshards_t)\ O(tcache, tcache_t, tcache_t) \ O(witness_tsd, witness_tsd_t, witness_tsdn_t) \ MALLOC_TEST_TSD #define TSD_INITIALIZER { \ - tsd_state_uninitialized, \ + ATOMIC_INIT(tsd_state_uninitialized), \ TCACHE_ENABLED_ZERO_INITIALIZER, \ false, \ 0, \ @@ -86,29 +89,97 @@ typedef void (*test_callback_t)(int *); 0, \ 0, \ 0, \ + 0, \ NULL, \ RTREE_CTX_ZERO_INITIALIZER, \ NULL, \ NULL, \ NULL, \ + TSD_BINSHARDS_ZERO_INITIALIZER, \ TCACHE_ZERO_INITIALIZER, \ WITNESS_TSD_INITIALIZER \ MALLOC_TEST_TSD_INITIALIZER \ } +void *malloc_tsd_malloc(size_t size); +void malloc_tsd_dalloc(void *wrapper); +void malloc_tsd_cleanup_register(bool (*f)(void)); +tsd_t *malloc_tsd_boot0(void); +void malloc_tsd_boot1(void); +void tsd_cleanup(void *arg); +tsd_t *tsd_fetch_slow(tsd_t *tsd, bool internal); +void tsd_state_set(tsd_t *tsd, uint8_t new_state); +void tsd_slow_update(tsd_t *tsd); +void tsd_prefork(tsd_t *tsd); +void tsd_postfork_parent(tsd_t *tsd); +void tsd_postfork_child(tsd_t *tsd); + +/* + * Call ..._inc when your module wants to take all threads down the slow paths, + * and ..._dec when it no longer needs to. + */ +void tsd_global_slow_inc(tsdn_t *tsdn); +void tsd_global_slow_dec(tsdn_t *tsdn); +bool tsd_global_slow(); + enum { - tsd_state_nominal = 0, /* Common case --> jnz. */ - tsd_state_nominal_slow = 1, /* Initialized but on slow path. */ - /* the above 2 nominal states should be lower values. */ - tsd_state_nominal_max = 1, /* used for comparison only. */ - tsd_state_minimal_initialized = 2, - tsd_state_purgatory = 3, - tsd_state_reincarnated = 4, - tsd_state_uninitialized = 5 + /* Common case --> jnz. */ + tsd_state_nominal = 0, + /* Initialized but on slow path. */ + tsd_state_nominal_slow = 1, + /* + * Some thread has changed global state in such a way that all nominal + * threads need to recompute their fast / slow status the next time they + * get a chance. + * + * Any thread can change another thread's status *to* recompute, but + * threads are the only ones who can change their status *from* + * recompute. + */ + tsd_state_nominal_recompute = 2, + /* + * The above nominal states should be lower values. We use + * tsd_nominal_max to separate nominal states from threads in the + * process of being born / dying. + */ + tsd_state_nominal_max = 2, + + /* + * A thread might free() during its death as its only allocator action; + * in such scenarios, we need tsd, but set up in such a way that no + * cleanup is necessary. + */ + tsd_state_minimal_initialized = 3, + /* States during which we know we're in thread death. */ + tsd_state_purgatory = 4, + tsd_state_reincarnated = 5, + /* + * What it says on the tin; tsd that hasn't been initialized. Note + * that even when the tsd struct lives in TLS, when need to keep track + * of stuff like whether or not our pthread destructors have been + * scheduled, so this really truly is different than the nominal state. + */ + tsd_state_uninitialized = 6 }; -/* Manually limit tsd_state_t to a single byte. */ -typedef uint8_t tsd_state_t; +/* + * Some TSD accesses can only be done in a nominal state. To enforce this, we + * wrap TSD member access in a function that asserts on TSD state, and mangle + * field names to prevent touching them accidentally. + */ +#define TSD_MANGLE(n) cant_access_tsd_items_directly_use_a_getter_or_setter_##n + +#ifdef JEMALLOC_U8_ATOMICS +# define tsd_state_t atomic_u8_t +# define tsd_atomic_load atomic_load_u8 +# define tsd_atomic_store atomic_store_u8 +# define tsd_atomic_exchange atomic_exchange_u8 +#else +# define tsd_state_t atomic_u32_t +# define tsd_atomic_load atomic_load_u32 +# define tsd_atomic_store atomic_store_u32 +# define tsd_atomic_exchange atomic_exchange_u32 +#endif /* The actual tsd. */ struct tsd_s { @@ -117,13 +188,29 @@ struct tsd_s { * module. Access any thread-local state through the getters and * setters below. */ - tsd_state_t state; + + /* + * We manually limit the state to just a single byte. Unless the 8-bit + * atomics are unavailable (which is rare). + */ + tsd_state_t state; #define O(n, t, nt) \ - t use_a_getter_or_setter_instead_##n; + t TSD_MANGLE(n); MALLOC_TSD #undef O }; +JEMALLOC_ALWAYS_INLINE uint8_t +tsd_state_get(tsd_t *tsd) { + /* + * This should be atomic. Unfortunately, compilers right now can't tell + * that this can be done as a memory comparison, and forces a load into + * a register that hurts fast-path performance. + */ + /* return atomic_load_u8(&tsd->state, ATOMIC_RELAXED); */ + return *(uint8_t *)&tsd->state; +} + /* * Wrapper around tsd_t that makes it possible to avoid implicit conversion * between tsd_t and tsdn_t, where tsdn_t is "nullable" and has to be @@ -150,15 +237,6 @@ tsdn_tsd(tsdn_t *tsdn) { return &tsdn->tsd; } -void *malloc_tsd_malloc(size_t size); -void malloc_tsd_dalloc(void *wrapper); -void malloc_tsd_cleanup_register(bool (*f)(void)); -tsd_t *malloc_tsd_boot0(void); -void malloc_tsd_boot1(void); -void tsd_cleanup(void *arg); -tsd_t *tsd_fetch_slow(tsd_t *tsd, bool internal); -void tsd_slow_update(tsd_t *tsd); - /* * We put the platform-specific data declarations and inlines into their own * header files to avoid cluttering this file. They define tsd_boot0, @@ -182,7 +260,7 @@ void tsd_slow_update(tsd_t *tsd); #define O(n, t, nt) \ JEMALLOC_ALWAYS_INLINE t * \ tsd_##n##p_get_unsafe(tsd_t *tsd) { \ - return &tsd->use_a_getter_or_setter_instead_##n; \ + return &tsd->TSD_MANGLE(n); \ } MALLOC_TSD #undef O @@ -191,10 +269,16 @@ MALLOC_TSD #define O(n, t, nt) \ JEMALLOC_ALWAYS_INLINE t * \ tsd_##n##p_get(tsd_t *tsd) { \ - assert(tsd->state == tsd_state_nominal || \ - tsd->state == tsd_state_nominal_slow || \ - tsd->state == tsd_state_reincarnated || \ - tsd->state == tsd_state_minimal_initialized); \ + /* \ + * Because the state might change asynchronously if it's \ + * nominal, we need to make sure that we only read it once. \ + */ \ + uint8_t state = tsd_state_get(tsd); \ + assert(state == tsd_state_nominal || \ + state == tsd_state_nominal_slow || \ + state == tsd_state_nominal_recompute || \ + state == tsd_state_reincarnated || \ + state == tsd_state_minimal_initialized); \ return tsd_##n##p_get_unsafe(tsd); \ } MALLOC_TSD @@ -229,8 +313,8 @@ MALLOC_TSD #define O(n, t, nt) \ JEMALLOC_ALWAYS_INLINE void \ tsd_##n##_set(tsd_t *tsd, t val) { \ - assert(tsd->state != tsd_state_reincarnated && \ - tsd->state != tsd_state_minimal_initialized); \ + assert(tsd_state_get(tsd) != tsd_state_reincarnated && \ + tsd_state_get(tsd) != tsd_state_minimal_initialized); \ *tsd_##n##p_get(tsd) = val; \ } MALLOC_TSD @@ -238,13 +322,18 @@ MALLOC_TSD JEMALLOC_ALWAYS_INLINE void tsd_assert_fast(tsd_t *tsd) { + /* + * Note that our fastness assertion does *not* include global slowness + * counters; it's not in general possible to ensure that they won't + * change asynchronously from underneath us. + */ assert(!malloc_slow && tsd_tcache_enabled_get(tsd) && tsd_reentrancy_level_get(tsd) == 0); } JEMALLOC_ALWAYS_INLINE bool tsd_fast(tsd_t *tsd) { - bool fast = (tsd->state == tsd_state_nominal); + bool fast = (tsd_state_get(tsd) == tsd_state_nominal); if (fast) { tsd_assert_fast(tsd); } @@ -261,7 +350,7 @@ tsd_fetch_impl(bool init, bool minimal) { } assert(tsd != NULL); - if (unlikely(tsd->state != tsd_state_nominal)) { + if (unlikely(tsd_state_get(tsd) != tsd_state_nominal)) { return tsd_fetch_slow(tsd, minimal); } assert(tsd_fast(tsd)); @@ -281,7 +370,7 @@ JEMALLOC_ALWAYS_INLINE tsd_t * tsd_internal_fetch(void) { tsd_t *tsd = tsd_fetch_min(); /* Use reincarnated state to prevent full initialization. */ - tsd->state = tsd_state_reincarnated; + tsd_state_set(tsd, tsd_state_reincarnated); return tsd; } @@ -293,7 +382,7 @@ tsd_fetch(void) { static inline bool tsd_nominal(tsd_t *tsd) { - return (tsd->state <= tsd_state_nominal_max); + return (tsd_state_get(tsd) <= tsd_state_nominal_max); } JEMALLOC_ALWAYS_INLINE tsdn_t * diff --git a/deps/jemalloc/include/jemalloc/internal/tsd_generic.h b/deps/jemalloc/include/jemalloc/internal/tsd_generic.h index 1e52ef767..cf73c0c71 100644 --- a/deps/jemalloc/include/jemalloc/internal/tsd_generic.h +++ b/deps/jemalloc/include/jemalloc/internal/tsd_generic.h @@ -77,7 +77,10 @@ tsd_wrapper_get(bool init) { abort(); } else { wrapper->initialized = false; + JEMALLOC_DIAGNOSTIC_PUSH + JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS tsd_t initializer = TSD_INITIALIZER; + JEMALLOC_DIAGNOSTIC_POP wrapper->val = initializer; } tsd_wrapper_set(wrapper); @@ -107,7 +110,10 @@ tsd_boot1(void) { tsd_boot_wrapper.initialized = false; tsd_cleanup(&tsd_boot_wrapper.val); wrapper->initialized = false; + JEMALLOC_DIAGNOSTIC_PUSH + JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS tsd_t initializer = TSD_INITIALIZER; + JEMALLOC_DIAGNOSTIC_POP wrapper->val = initializer; tsd_wrapper_set(wrapper); } diff --git a/deps/jemalloc/include/jemalloc/internal/tsd_malloc_thread_cleanup.h b/deps/jemalloc/include/jemalloc/internal/tsd_malloc_thread_cleanup.h index beb467a67..65852d5c1 100644 --- a/deps/jemalloc/include/jemalloc/internal/tsd_malloc_thread_cleanup.h +++ b/deps/jemalloc/include/jemalloc/internal/tsd_malloc_thread_cleanup.h @@ -3,8 +3,10 @@ #endif #define JEMALLOC_INTERNAL_TSD_MALLOC_THREAD_CLEANUP_H -extern __thread tsd_t tsd_tls; -extern __thread bool tsd_initialized; +#define JEMALLOC_TSD_TYPE_ATTR(type) __thread type JEMALLOC_TLS_MODEL + +extern JEMALLOC_TSD_TYPE_ATTR(tsd_t) tsd_tls; +extern JEMALLOC_TSD_TYPE_ATTR(bool) tsd_initialized; extern bool tsd_booted; /* Initialization/cleanup. */ @@ -47,7 +49,6 @@ tsd_get_allocates(void) { /* Get/set. */ JEMALLOC_ALWAYS_INLINE tsd_t * tsd_get(bool init) { - assert(tsd_booted); return &tsd_tls; } JEMALLOC_ALWAYS_INLINE void diff --git a/deps/jemalloc/include/jemalloc/internal/tsd_tls.h b/deps/jemalloc/include/jemalloc/internal/tsd_tls.h index 0de64b7b8..7d6c805be 100644 --- a/deps/jemalloc/include/jemalloc/internal/tsd_tls.h +++ b/deps/jemalloc/include/jemalloc/internal/tsd_tls.h @@ -3,7 +3,9 @@ #endif #define JEMALLOC_INTERNAL_TSD_TLS_H -extern __thread tsd_t tsd_tls; +#define JEMALLOC_TSD_TYPE_ATTR(type) __thread type JEMALLOC_TLS_MODEL + +extern JEMALLOC_TSD_TYPE_ATTR(tsd_t) tsd_tls; extern pthread_key_t tsd_tsd; extern bool tsd_booted; @@ -39,8 +41,7 @@ tsd_get_allocates(void) { /* Get/set. */ JEMALLOC_ALWAYS_INLINE tsd_t * -tsd_get(UNUSED bool init) { - assert(tsd_booted); +tsd_get(bool init) { return &tsd_tls; } diff --git a/deps/jemalloc/include/jemalloc/internal/witness.h b/deps/jemalloc/include/jemalloc/internal/witness.h index 7ace8ae4a..fff9e98cb 100644 --- a/deps/jemalloc/include/jemalloc/internal/witness.h +++ b/deps/jemalloc/include/jemalloc/internal/witness.h @@ -27,9 +27,9 @@ #define WITNESS_RANK_PROF_BT2GCTX 6U #define WITNESS_RANK_PROF_TDATAS 7U #define WITNESS_RANK_PROF_TDATA 8U -#define WITNESS_RANK_PROF_GCTX 9U - -#define WITNESS_RANK_BACKGROUND_THREAD 10U +#define WITNESS_RANK_PROF_LOG 9U +#define WITNESS_RANK_PROF_GCTX 10U +#define WITNESS_RANK_BACKGROUND_THREAD 11U /* * Used as an argument to witness_assert_depth_to_rank() in order to validate @@ -37,18 +37,19 @@ * witness_assert_depth_to_rank() is inclusive rather than exclusive, this * definition can have the same value as the minimally ranked core lock. */ -#define WITNESS_RANK_CORE 11U - -#define WITNESS_RANK_DECAY 11U -#define WITNESS_RANK_TCACHE_QL 12U -#define WITNESS_RANK_EXTENT_GROW 13U -#define WITNESS_RANK_EXTENTS 14U -#define WITNESS_RANK_EXTENT_AVAIL 15U - -#define WITNESS_RANK_EXTENT_POOL 16U -#define WITNESS_RANK_RTREE 17U -#define WITNESS_RANK_BASE 18U -#define WITNESS_RANK_ARENA_LARGE 19U +#define WITNESS_RANK_CORE 12U + +#define WITNESS_RANK_DECAY 12U +#define WITNESS_RANK_TCACHE_QL 13U +#define WITNESS_RANK_EXTENT_GROW 14U +#define WITNESS_RANK_EXTENTS 15U +#define WITNESS_RANK_EXTENT_AVAIL 16U + +#define WITNESS_RANK_EXTENT_POOL 17U +#define WITNESS_RANK_RTREE 18U +#define WITNESS_RANK_BASE 19U +#define WITNESS_RANK_ARENA_LARGE 20U +#define WITNESS_RANK_HOOK 21U #define WITNESS_RANK_LEAF 0xffffffffU #define WITNESS_RANK_BIN WITNESS_RANK_LEAF |