diff options
Diffstat (limited to 'erts/emulator/sys')
-rw-r--r-- | erts/emulator/sys/common/erl_mseg.c | 315 | ||||
-rw-r--r-- | erts/emulator/sys/common/erl_mseg.h | 9 | ||||
-rw-r--r-- | erts/emulator/sys/common/erl_sys_common_misc.c | 204 | ||||
-rw-r--r-- | erts/emulator/sys/common/erl_util_queue.h | 77 | ||||
-rwxr-xr-x | erts/emulator/sys/win32/sys.c | 82 |
5 files changed, 438 insertions, 249 deletions
diff --git a/erts/emulator/sys/common/erl_mseg.c b/erts/emulator/sys/common/erl_mseg.c index 538eea88d1..bd8ba82a5f 100644 --- a/erts/emulator/sys/common/erl_mseg.c +++ b/erts/emulator/sys/common/erl_mseg.c @@ -39,6 +39,7 @@ #include "erl_alloc.h" #include "big.h" #include "erl_thr_progress.h" +#include "erl_util_queue.h" #if HAVE_ERTS_MSEG @@ -175,6 +176,7 @@ struct cache_t_ { Uint size; void *seg; cache_t *next; + cache_t *prev; }; @@ -183,9 +185,9 @@ typedef struct ErtsMsegAllctr_t_ ErtsMsegAllctr_t; struct mem_kind_t { cache_t cache[MAX_CACHE_SIZE]; - cache_t *cache_unpowered; - cache_t *cache_area[CACHE_AREAS]; - cache_t *cache_free; + cache_t cache_unpowered_node; + cache_t cache_powered_node[CACHE_AREAS]; + cache_t cache_free; Sint cache_size; Uint cache_hits; @@ -516,67 +518,94 @@ do { \ #define ERTS_DBG_MK_CHK_THR_ACCESS(MK) #endif -/* NEW CACHE interface */ +/* Cache interface */ -static ERTS_INLINE cache_t *mseg_cache_alloc_descriptor(MemKind *mk) { - cache_t *c = mk->cache_free; - ERTS_DBG_MK_CHK_THR_ACCESS(mk); - if (c) - mk->cache_free = c->next; - - return c; -} - -static ERTS_INLINE void mseg_cache_free_descriptor(MemKind *mk, cache_t *c) { - ERTS_DBG_MK_CHK_THR_ACCESS(mk); - ASSERT(c); - - c->seg = NULL; +static ERTS_INLINE void mseg_cache_clear_node(cache_t *c) { + c->seg = NULL; c->size = 0; - c->next = mk->cache_free; - mk->cache_free = c; + c->next = c; + c->prev = c; } -static ERTS_INLINE int cache_bless_segment(MemKind *mk, void *seg, Uint size) { +static ERTS_INLINE int cache_bless_segment(MemKind *mk, void *seg, Uint size, Uint flags) { cache_t *c; ERTS_DBG_MK_CHK_THR_ACCESS(mk); - if (mk->cache_free && MAP_IS_ALIGNED(seg)) { - if (IS_2POW(size)) { + ASSERT(!MSEG_FLG_IS_2POW(flags) || (MSEG_FLG_IS_2POW(flags) && MAP_IS_ALIGNED(seg) && IS_2POW(size))); + + /* The idea is that sbc caching is prefered over mbc caching. + * Blocks are normally allocated in mb carriers and thus cached there. + * Large blocks has no such cache and it is up to mseg to cache them to speed things up. + */ + + if (!erts_circleq_is_empty(&(mk->cache_free))) { + + /* We have free slots, use one to cache the segment */ + + c = erts_circleq_head(&(mk->cache_free)); + erts_circleq_remove(c); + + c->seg = seg; + c->size = size; + + if (MSEG_FLG_IS_2POW(flags)) { int ix = SIZE_TO_CACHE_AREA_IDX(size); ASSERT(ix < CACHE_AREAS); ASSERT((1 << (ix + MSEG_ALIGN_BITS)) == size); - /* unlink from free cache list */ - c = mseg_cache_alloc_descriptor(mk); + erts_circleq_push_head(&(mk->cache_powered_node[ix]), c); - /* link to cache area */ - c->seg = seg; - c->size = size; - c->next = mk->cache_area[ix]; + } else + erts_circleq_push_head(&(mk->cache_unpowered_node), c); - mk->cache_area[ix] = c; - mk->cache_size++; + mk->cache_size++; + ASSERT(mk->cache_size <= mk->ma->max_cache_size); - ASSERT(mk->cache_size <= mk->ma->max_cache_size); + return 1; + } else if (!MSEG_FLG_IS_2POW(flags) && !erts_circleq_is_empty(&(mk->cache_unpowered_node))) { - return 1; - } else { - /* unlink from free cache list */ - c = mseg_cache_alloc_descriptor(mk); + /* No free slots. + * Evict oldest slot from unpowered cache so we can cache an unpowered (sbc) segment */ + + c = erts_circleq_tail(&(mk->cache_unpowered_node)); + erts_circleq_remove(c); + + mseg_destroy(mk->ma, mk, c->seg, c->size); + mseg_cache_clear_node(c); + + c->seg = seg; + c->size = size; + + erts_circleq_push_head(&(mk->cache_unpowered_node), c); + + return 1; + } else if (!MSEG_FLG_IS_2POW(flags)) { + + /* No free slots and no unpowered (sbc) slots. + * Evict smallest slot from powered cache so we can cache an unpowered (sbc) segment. + * Note: Though this is the wanted policy I don't think it is used significantly. + * This branch could probably be removed without serious impact. + */ + + int i; + + for( i = 0; i < CACHE_AREAS; i++) { + if (erts_circleq_is_empty(&(mk->cache_powered_node[i]))) + continue; + + c = erts_circleq_tail(&(mk->cache_powered_node[i])); + erts_circleq_remove(c); + + mseg_destroy(mk->ma, mk, c->seg, c->size); + mseg_cache_clear_node(c); - /* link to cache area */ c->seg = seg; c->size = size; - c->next = mk->cache_unpowered; - - mk->cache_unpowered = c; - mk->cache_size++; - ASSERT(mk->cache_size <= mk->ma->max_cache_size); + erts_circleq_push_head(&(mk->cache_unpowered_node), c); return 1; } @@ -585,90 +614,110 @@ static ERTS_INLINE int cache_bless_segment(MemKind *mk, void *seg, Uint size) { return 0; } -static ERTS_INLINE void *cache_get_segment(MemKind *mk, Uint *size_p) { +static ERTS_INLINE void *cache_get_segment(MemKind *mk, Uint *size_p, Uint flags) { Uint size = *size_p; ERTS_DBG_MK_CHK_THR_ACCESS(mk); - if (IS_2POW(size)) { + + if (MSEG_FLG_IS_2POW(flags)) { int i, ix = SIZE_TO_CACHE_AREA_IDX(size); void *seg; cache_t *c; Uint csize; + ASSERT(IS_2POW(size)); + for( i = ix; i < CACHE_AREAS; i++) { - if ((c = mk->cache_area[i]) == NULL) + if (erts_circleq_is_empty(&(mk->cache_powered_node[i]))) continue; + c = erts_circleq_head(&(mk->cache_powered_node[i])); + erts_circleq_remove(c); + ASSERT(IS_2POW(c->size)); + ASSERT(MAP_IS_ALIGNED(c->seg)); + + csize = c->size; + seg = c->seg; - /* unlink from cache area */ - csize = c->size; - seg = c->seg; - mk->cache_area[i] = c->next; - c->next = NULL; mk->cache_size--; mk->cache_hits++; /* link to free cache list */ - mseg_cache_free_descriptor(mk, c); + mseg_cache_clear_node(c); + erts_circleq_push_head(&(mk->cache_free), c); ASSERT(!(mk->cache_size < 0)); - /* divvy up the cache - if needed */ - while( i > ix) { - csize = csize >> 1; - /* try to cache half of it */ - if (!cache_bless_segment(mk, (char *)seg + csize, csize)) { - /* wouldn't cache .. destroy it instead */ - mseg_destroy(mk->ma, mk, (char *)seg + csize, csize); - } - i--; + if (csize != size) { + mseg_destroy(mk->ma, mk, (char *)seg + size, csize - size); } - ASSERT(csize == size); + return seg; } } - else if (mk->cache_unpowered) { + else if (!erts_circleq_is_empty(&(mk->cache_unpowered_node))) { void *seg; - cache_t *c, *pc; + cache_t *c; + cache_t *best = NULL; + Uint bdiff = 0; Uint csize; Uint bad_max_abs = mk->ma->abs_max_cache_bad_fit; Uint bad_max_rel = mk->ma->rel_max_cache_bad_fit; - c = mk->cache_unpowered; - pc = c; - - while (c) { + erts_circleq_foreach(c, &(mk->cache_unpowered_node)) { csize = c->size; - if (csize >= size && - ((csize - size)*100 < bad_max_rel*size) && - (csize - size) < bad_max_abs ) { + if (csize >= size) { + if (((csize - size)*100 < bad_max_rel*size) && (csize - size) < bad_max_abs ) { - /* unlink from cache area */ - seg = c->seg; + seg = c->seg; - if (pc == c) { - mk->cache_unpowered = c->next; - } else { - pc->next = c->next; - } + erts_circleq_remove(c); + + mk->cache_size--; + mk->cache_hits++; - c->next = NULL; - mk->cache_size--; - mk->cache_hits++; + mseg_cache_clear_node(c); + erts_circleq_push_head(&(mk->cache_free), c); - /* link to free cache list */ - mseg_cache_free_descriptor(mk, c); - *size_p = csize; + *size_p = csize; - return seg; + return seg; + + } else if (!best || (csize - size) < bdiff) { + best = c; + bdiff = csize - size; + } } + } + + /* No cached segment met our criteria, use the best one found and trim it */ + + if (best) { + + seg = best->seg; + csize = best->size; + + ASSERT(best->seg); + ASSERT(best->size > 0); + + mk->cache_hits++; + + /* Use current cache placement for remaining segment space */ + + best->seg = seg + size; + best->size = csize - size; + + ASSERT((size % GET_PAGE_SIZE) == 0); + ASSERT((best->size % GET_PAGE_SIZE) == 0); + + *size_p = size; + + return seg; - pc = c; - c = c->next; } } return NULL; @@ -679,20 +728,18 @@ static ERTS_INLINE void *cache_get_segment(MemKind *mk, Uint *size_p) { * using callbacks from aux-work in the scheduler. */ -static ERTS_INLINE Uint mseg_drop_one_memkind_cache_size(MemKind *mk, cache_t **head) { +static ERTS_INLINE Uint mseg_drop_one_memkind_cache_size(MemKind *mk, cache_t *head) { cache_t *c = NULL; - c = *head; - - ASSERT( c != NULL ); - - *head = c->next; + c = erts_circleq_tail(head); + erts_circleq_remove(c); if (erts_mtrace_enabled) erts_mtrace_crr_free(SEGTYPE, SEGTYPE, c->seg); mseg_destroy(mk->ma, mk, c->seg, c->size); - mseg_cache_free_descriptor(mk, c); + mseg_cache_clear_node(c); + erts_circleq_push_head(&(mk->cache_free), c); mk->segments.current.watermark--; mk->cache_size--; @@ -702,30 +749,27 @@ static ERTS_INLINE Uint mseg_drop_one_memkind_cache_size(MemKind *mk, cache_t ** return mk->cache_size; } -static ERTS_INLINE Uint mseg_drop_memkind_cache_size(MemKind *mk, cache_t **head) { - cache_t *c = NULL, *next = NULL; - - c = *head; - ASSERT( c != NULL ); +static ERTS_INLINE Uint mseg_drop_memkind_cache_size(MemKind *mk, cache_t *head) { + cache_t *c = NULL; - while (c) { + while (!erts_circleq_is_empty(head)) { - next = c->next; + c = erts_circleq_tail(head); + erts_circleq_remove(c); if (erts_mtrace_enabled) erts_mtrace_crr_free(SEGTYPE, SEGTYPE, c->seg); mseg_destroy(mk->ma, mk, c->seg, c->size); - mseg_cache_free_descriptor(mk, c); + + mseg_cache_clear_node(c); + erts_circleq_push_head(&(mk->cache_free), c); mk->segments.current.watermark--; mk->cache_size--; - c = next; } - *head = NULL; - ASSERT( mk->cache_size >= 0 ); return mk->cache_size; @@ -743,12 +787,12 @@ static Uint mseg_check_memkind_cache(MemKind *mk) { ERTS_DBG_MK_CHK_THR_ACCESS(mk); for (i = 0; i < CACHE_AREAS; i++) { - if (mk->cache_area[i] != NULL) - return mseg_drop_one_memkind_cache_size(mk, &(mk->cache_area[i])); + if (!erts_circleq_is_empty(&(mk->cache_powered_node[i]))) + return mseg_drop_one_memkind_cache_size(mk, &(mk->cache_powered_node[i])); } - if (mk->cache_unpowered) - return mseg_drop_one_memkind_cache_size(mk, &(mk->cache_unpowered)); + if (!erts_circleq_is_empty(&(mk->cache_unpowered_node))) + return mseg_drop_one_memkind_cache_size(mk, &(mk->cache_unpowered_node)); return 0; } @@ -804,17 +848,17 @@ static void mseg_clear_memkind_cache(MemKind *mk) { /* drop pow2 caches */ for (i = 0; i < CACHE_AREAS; i++) { - if (mk->cache_area[i] == NULL) + if (erts_circleq_is_empty(&(mk->cache_powered_node[i]))) continue; - mseg_drop_memkind_cache_size(mk, &(mk->cache_area[i])); - ASSERT(mk->cache_area[i] == NULL); + mseg_drop_memkind_cache_size(mk, &(mk->cache_powered_node[i])); + ASSERT(erts_circleq_is_empty(&(mk->cache_powered_node[i]))); } /* drop varied caches */ - if(mk->cache_unpowered) - mseg_drop_memkind_cache_size(mk, &(mk->cache_unpowered)); + if (!erts_circleq_is_empty(&(mk->cache_unpowered_node))) + mseg_drop_memkind_cache_size(mk, &(mk->cache_unpowered_node)); - ASSERT(mk->cache_unpowered == NULL); + ASSERT(erts_circleq_is_empty(&(mk->cache_unpowered_node))); ASSERT(mk->cache_size == 0); } @@ -873,7 +917,7 @@ mseg_alloc(ErtsMsegAllctr_t *ma, ErtsAlcType_t atype, Uint *size_p, ma->min_seg_size = size; #endif - if (opt->cache && mk->cache_size > 0 && (seg = cache_get_segment(mk, &size)) != NULL) + if (opt->cache && mk->cache_size > 0 && (seg = cache_get_segment(mk, &size, flags)) != NULL) goto done; if ((seg = mseg_create(ma, mk, size)) == NULL) @@ -894,14 +938,13 @@ done: static void mseg_dealloc(ErtsMsegAllctr_t *ma, ErtsAlcType_t atype, void *seg, Uint size, - const ErtsMsegOpt_t *opt) + Uint flags, const ErtsMsegOpt_t *opt) { MemKind* mk = memkind(ma, opt); - ERTS_MSEG_DEALLOC_STAT(mk,size); - if (opt->cache && cache_bless_segment(mk, seg, size)) { + if (opt->cache && cache_bless_segment(mk, seg, size, flags)) { schedule_cache_check(ma); goto done; } @@ -934,7 +977,7 @@ mseg_realloc(ErtsMsegAllctr_t *ma, ErtsAlcType_t atype, void *seg, /* Dealloc old segment if new segment is of size 0 */ if (!(*new_size_p)) { - mseg_dealloc(ma, atype, seg, old_size, opt); + mseg_dealloc(ma, atype, seg, old_size, flags, opt); DEC_CC(ma, dealloc); return NULL; } @@ -993,7 +1036,7 @@ mseg_realloc(ErtsMsegAllctr_t *ma, ErtsAlcType_t atype, void *seg, else { if (!opt->preserv) { - mseg_dealloc(ma, atype, seg, old_size, opt); + mseg_dealloc(ma, atype, seg, old_size, flags, opt); new_seg = mseg_alloc(ma, atype, &new_size, flags, opt); ASSERT(MAP_IS_ALIGNED(new_seg) || !new_seg); } @@ -1020,7 +1063,7 @@ mseg_realloc(ErtsMsegAllctr_t *ma, ErtsAlcType_t atype, void *seg, new_size = old_size; else { sys_memcpy(((char *) new_seg), ((char *) seg), MIN(new_size, old_size)); - mseg_dealloc(ma, atype, seg, old_size, opt); + mseg_dealloc(ma, atype, seg, old_size, flags, opt); } #endif } @@ -1497,19 +1540,19 @@ erts_mseg_alloc(ErtsAlcType_t atype, Uint *size_p, Uint flags) void erts_mseg_dealloc_opt(ErtsAlcType_t atype, void *seg, - Uint size, const ErtsMsegOpt_t *opt) + Uint size, Uint flags, const ErtsMsegOpt_t *opt) { ErtsMsegAllctr_t *ma = ERTS_MSEG_ALLCTR_OPT(opt); ERTS_MSEG_LOCK(ma); ERTS_DBG_MA_CHK_THR_ACCESS(ma); - mseg_dealloc(ma, atype, seg, size, opt); + mseg_dealloc(ma, atype, seg, size, flags, opt); ERTS_MSEG_UNLOCK(ma); } void -erts_mseg_dealloc(ErtsAlcType_t atype, void *seg, Uint size) +erts_mseg_dealloc(ErtsAlcType_t atype, void *seg, Uint size, Uint flags) { - erts_mseg_dealloc_opt(atype, seg, size, &erts_mseg_default_opt); + erts_mseg_dealloc_opt(atype, seg, size, flags, &erts_mseg_default_opt); } void * @@ -1556,27 +1599,28 @@ erts_mseg_unit_size(void) return MSEG_ALIGNED_SIZE; } + static void mem_kind_init(ErtsMsegAllctr_t *ma, MemKind* mk, const char* name) { int i; + /* Clear all cache headers */ + mseg_cache_clear_node(&(mk->cache_free)); + mseg_cache_clear_node(&(mk->cache_unpowered_node)); + for (i = 0; i < CACHE_AREAS; i++) { - mk->cache_area[i] = NULL; + mseg_cache_clear_node(&(mk->cache_powered_node[i])); } - mk->cache_free = NULL; + /* Populate cache free list */ ASSERT(ma->max_cache_size <= MAX_CACHE_SIZE); for (i = 0; i < ma->max_cache_size; i++) { - mk->cache[i].seg = NULL; - mk->cache[i].size = 0; - mk->cache[i].next = mk->cache_free; - mk->cache_free = &(mk->cache[i]); + mseg_cache_clear_node(&(mk->cache[i])); + erts_circleq_push_head(&(mk->cache_free), &(mk->cache[i])); } - mk->cache_unpowered = NULL; - mk->cache_size = 0; mk->cache_hits = 0; @@ -1594,9 +1638,6 @@ static void mem_kind_init(ErtsMsegAllctr_t *ma, MemKind* mk, const char* name) ma->mk_list = mk; } - - - void erts_mseg_init(ErtsMsegInit_t *init) { @@ -1721,7 +1762,7 @@ erts_mseg_test(unsigned long op, case 0x401: return (unsigned long) erts_mseg_alloc(ERTS_ALC_A_INVALID, (Uint *) a1, (Uint) 0); case 0x402: - erts_mseg_dealloc(ERTS_ALC_A_INVALID, (void *) a1, (Uint) a2); + erts_mseg_dealloc(ERTS_ALC_A_INVALID, (void *) a1, (Uint) a2, (Uint) 0); return (unsigned long) 0; case 0x403: return (unsigned long) erts_mseg_realloc(ERTS_ALC_A_INVALID, diff --git a/erts/emulator/sys/common/erl_mseg.h b/erts/emulator/sys/common/erl_mseg.h index 3d0b0f0355..3cab9e18da 100644 --- a/erts/emulator/sys/common/erl_mseg.h +++ b/erts/emulator/sys/common/erl_mseg.h @@ -74,7 +74,7 @@ typedef struct { { \ 4*1024*1024, /* amcbf: Absolute max cache bad fit */ \ 20, /* rmcbf: Relative max cache bad fit */ \ - 5, /* mcs: Max cache size */ \ + 10, /* mcs: Max cache size */ \ 1000 /* cci: Cache check interval */ \ } @@ -93,11 +93,10 @@ extern const ErtsMsegOpt_t erts_mseg_default_opt; void *erts_mseg_alloc(ErtsAlcType_t, Uint *, Uint); void *erts_mseg_alloc_opt(ErtsAlcType_t, Uint *, Uint, const ErtsMsegOpt_t *); -void erts_mseg_dealloc(ErtsAlcType_t, void *, Uint); -void erts_mseg_dealloc_opt(ErtsAlcType_t, void *, Uint, const ErtsMsegOpt_t *); +void erts_mseg_dealloc(ErtsAlcType_t, void *, Uint, Uint); +void erts_mseg_dealloc_opt(ErtsAlcType_t, void *, Uint, Uint, const ErtsMsegOpt_t *); void *erts_mseg_realloc(ErtsAlcType_t, void *, Uint, Uint *, Uint); -void *erts_mseg_realloc_opt(ErtsAlcType_t, void *, Uint, Uint *, - Uint, const ErtsMsegOpt_t *); +void *erts_mseg_realloc_opt(ErtsAlcType_t, void *, Uint, Uint *, Uint, const ErtsMsegOpt_t *); void erts_mseg_clear_cache(void); void erts_mseg_cache_check(void); Uint erts_mseg_no( const ErtsMsegOpt_t *); diff --git a/erts/emulator/sys/common/erl_sys_common_misc.c b/erts/emulator/sys/common/erl_sys_common_misc.c index 1bf5fa89f4..0b31c125e5 100644 --- a/erts/emulator/sys/common/erl_sys_common_misc.c +++ b/erts/emulator/sys/common/erl_sys_common_misc.c @@ -47,14 +47,16 @@ /* Written once and only once */ static int filename_encoding = ERL_FILENAME_UNKNOWN; +static int filename_warning = ERL_FILENAME_WARNING_WARNING; #if defined(__WIN32__) || defined(__DARWIN__) static int user_filename_encoding = ERL_FILENAME_UTF8; /* Default unicode on windows */ #else static int user_filename_encoding = ERL_FILENAME_LATIN1; #endif -void erts_set_user_requested_filename_encoding(int encoding) +void erts_set_user_requested_filename_encoding(int encoding, int warning) { user_filename_encoding = encoding; + filename_warning = warning; } int erts_get_user_requested_filename_encoding(void) @@ -62,6 +64,11 @@ int erts_get_user_requested_filename_encoding(void) return user_filename_encoding; } +int erts_get_filename_warning_type(void) +{ + return filename_warning; +} + void erts_init_sys_common_misc(void) { #if defined(__WIN32__) @@ -107,9 +114,9 @@ int erts_get_native_filename_encoding(void) } /* For internal use by sys_double_to_chars_fast() */ -static char* float_first_trailing_zero(char* p) +static char* find_first_trailing_zero(char* p) { - for (--p; *p == '0' && *(p-1) == '0'; --p); + for (; *(p-1) == '0'; --p); if (*(p-1) == '.') ++p; return p; } @@ -120,34 +127,83 @@ sys_double_to_chars(double fp, char *buffer, size_t buffer_size) return sys_double_to_chars_ext(fp, buffer, buffer_size, SYS_DEFAULT_FLOAT_DECIMALS); } +/* Convert float to string using fixed point notation. + * decimals must be >= 0 + * if compact != 0, the trailing 0's will be truncated + */ int -sys_double_to_chars_fast(double f, char *outbuf, int maxlen, int decimals, int compact) +sys_double_to_chars_fast(double f, char *buffer, int buffer_size, int decimals, + int compact) { - enum { - FRAC_SIZE = 52 - , EXP_SIZE = 11 - , EXP_MASK = (1ll << EXP_SIZE) - 1 - , FRAC_MASK = (1ll << FRAC_SIZE) - 1 - , FRAC_MASK2 = (1ll << (FRAC_SIZE + 1)) - 1 - , MAX_FLOAT = 1ll << (FRAC_SIZE+1) + /* Note that some C compilers don't support "static const" propagation + * so we use a defines */ + #define SYS_DOUBLE_RND_CONST 0.55555555555555555 + #define FRAC_SIZE 52 + #define EXP_SIZE 11 + #define EXP_MASK ((1ll << EXP_SIZE) - 1) + #define MAX_DECIMALS (sizeof(cs_sys_double_pow10) \ + / sizeof(cs_sys_double_pow10[0])) + #define FRAC_MASK ((1ll << FRAC_SIZE) - 1) + #define FRAC_MASK2 ((1ll << (FRAC_SIZE + 1)) - 1) + #define MAX_FLOAT (1ll << (FRAC_SIZE+1)) + + static const double cs_sys_double_pow10[] = { + SYS_DOUBLE_RND_CONST / 1ll, + SYS_DOUBLE_RND_CONST / 10ll, + SYS_DOUBLE_RND_CONST / 100ll, + SYS_DOUBLE_RND_CONST / 1000ll, + SYS_DOUBLE_RND_CONST / 10000ll, + SYS_DOUBLE_RND_CONST / 100000ll, + SYS_DOUBLE_RND_CONST / 1000000ll, + SYS_DOUBLE_RND_CONST / 10000000ll, + SYS_DOUBLE_RND_CONST / 100000000ll, + SYS_DOUBLE_RND_CONST / 1000000000ll, + SYS_DOUBLE_RND_CONST / 10000000000ll, + SYS_DOUBLE_RND_CONST / 100000000000ll, + SYS_DOUBLE_RND_CONST / 1000000000000ll, + SYS_DOUBLE_RND_CONST / 10000000000000ll, + SYS_DOUBLE_RND_CONST / 100000000000000ll, + SYS_DOUBLE_RND_CONST / 1000000000000000ll, + SYS_DOUBLE_RND_CONST / 10000000000000000ll, + SYS_DOUBLE_RND_CONST / 100000000000000000ll, + SYS_DOUBLE_RND_CONST / 1000000000000000000ll }; - long long mantissa, int_part, int_part2, frac_part; + long long mantissa, int_part = 0, frac_part = 0; short exp; - int sign, i, n, m, max; - double absf; + int max; + int neg; + double fr; union { long long L; double F; } x; - char c, *p = outbuf; - int digit, roundup; + char *p = buffer; + + if (decimals < 0) + return -1; - x.F = f; + /* Round the number to given decimal places. The number of 5's in the + * SYS_DOUBLE_RND_CONST constant is chosen such that adding any more 5's doesn't + * change the double precision of the number, i.e.: + * 1> term_to_binary(0.55555555555555555, [{minor_version, 1}]). + * <<131,70,63,225,199,28,113,199,28,114>> + * 2> term_to_binary(0.5555555555555555555, [{minor_version, 1}]). + * <<131,70,63,225,199,28,113,199,28,114>> + */ + if (f >= 0) { + neg = 0; + fr = decimals < MAX_DECIMALS ? (f + cs_sys_double_pow10[decimals]) : f; + x.F = fr; + } else { + neg = 1; + fr = decimals < MAX_DECIMALS ? (f - cs_sys_double_pow10[decimals]) : f; + x.F = -fr; + } exp = (x.L >> FRAC_SIZE) & EXP_MASK; mantissa = x.L & FRAC_MASK; - sign = x.L >= 0 ? 1 : -1; + if (exp == EXP_MASK) { if (mantissa == 0) { - if (sign == -1) + if (neg) *p++ = '-'; *p++ = 'i'; *p++ = 'n'; @@ -158,101 +214,79 @@ sys_double_to_chars_fast(double f, char *outbuf, int maxlen, int decimals, int c *p++ = 'n'; } *p = '\0'; - return p - outbuf; + return p - buffer; } exp -= EXP_MASK >> 1; mantissa |= (1ll << FRAC_SIZE); - frac_part = 0; - int_part = 0; - absf = f * sign; - - /* Don't bother with optimizing too large numbers and decimals */ - if (absf > MAX_FLOAT || decimals > maxlen-17) { - int len = erts_snprintf(outbuf, maxlen, "%.*f", decimals, f); - if (len >= maxlen) + + /* Don't bother with optimizing too large numbers or too large precision */ + if (x.F > MAX_FLOAT || decimals >= MAX_DECIMALS) { + int len = erts_snprintf(buffer, buffer_size, "%.*f", decimals, f); + char* p = buffer + len; + if (len >= buffer_size) return -1; - p = outbuf + len; /* Delete trailing zeroes */ if (compact) - p = float_first_trailing_zero(outbuf + len); + p = find_first_trailing_zero(p); *p = '\0'; - return p - outbuf; - } - - if (exp >= FRAC_SIZE) + return p - buffer; + } else if (exp >= FRAC_SIZE) { int_part = mantissa << (exp - FRAC_SIZE); - else if (exp >= 0) { + } else if (exp >= 0) { int_part = mantissa >> (FRAC_SIZE - exp); frac_part = (mantissa << (exp + 1)) & FRAC_MASK2; - } - else /* if (exp < 0) */ + } else /* if (exp < 0) */ { frac_part = (mantissa & FRAC_MASK2) >> -(exp + 1); + } - if (int_part == 0) { - if (sign == -1) + if (!int_part) { + if (neg) *p++ = '-'; *p++ = '0'; } else { - int ret; + int ret, i, n; while (int_part != 0) { - int_part2 = int_part / 10; - *p++ = (char)(int_part - ((int_part2 << 3) + (int_part2 << 1)) + '0'); - int_part = int_part2; + long long j = int_part / 10; + *p++ = (char)(int_part - ((j << 3) + (j << 1)) + '0'); + int_part = j; } - if (sign == -1) + if (neg) *p++ = '-'; /* Reverse string */ - ret = p - outbuf; + ret = p - buffer; for (i = 0, n = ret/2; i < n; i++) { - int j = ret - i - 1; - c = outbuf[i]; - outbuf[i] = outbuf[j]; - outbuf[j] = c; + int j = ret - i - 1; + char c = buffer[i]; + buffer[i] = buffer[j]; + buffer[j] = c; } } - if (decimals != 0) + + if (decimals > 0) { + int i; *p++ = '.'; - max = maxlen - (p - outbuf) - 1 /* leave room for trailing '\0' */; - if (max > decimals) + max = buffer_size - (p - buffer) - 1 /* leave room for trailing '\0' */; + + if (decimals > max) + return -1; /* the number is not large enough to fit in the buffer */ + max = decimals; - for (m = 0; m < max; m++) { - /* frac_part *= 10; */ - frac_part = (frac_part << 3) + (frac_part << 1); - *p++ = (char)((frac_part >> (FRAC_SIZE + 1)) + '0'); - frac_part &= FRAC_MASK2; - } + for (i = 0; i < max; i++) { + /* frac_part *= 10; */ + frac_part = (frac_part << 3) + (frac_part << 1); - roundup = 0; - /* Rounding - look at the next digit */ - frac_part = (frac_part << 3) + (frac_part << 1); - digit = (frac_part >> (FRAC_SIZE + 1)); - if (digit > 5) - roundup = 1; - else if (digit == 5) { - frac_part &= FRAC_MASK2; - if (frac_part != 0) roundup = 1; - } - if (roundup) { - char d; - int pos = p - outbuf - 1; - do { - d = outbuf[pos]; - if (d == '-') break; - if (d == '.') continue; - if (++d != ':') { - outbuf[pos] = d; - break; - } - outbuf[pos] = '0'; - } while (--pos); + *p++ = (char)((frac_part >> (FRAC_SIZE + 1)) + '0'); + frac_part &= FRAC_MASK2; + } + + /* Delete trailing zeroes */ + if (compact) + p = find_first_trailing_zero(p); } - /* Delete trailing zeroes */ - if (compact && *(p - 1) == '0') - p = float_first_trailing_zero(--p); *p = '\0'; - return p - outbuf; + return p - buffer; } diff --git a/erts/emulator/sys/common/erl_util_queue.h b/erts/emulator/sys/common/erl_util_queue.h new file mode 100644 index 0000000000..47925e2264 --- /dev/null +++ b/erts/emulator/sys/common/erl_util_queue.h @@ -0,0 +1,77 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2013. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifndef ERL_UTIL_QUEUE_H_ +#define ERL_UTIL_QUEUE_H_ + +#define erts_circleq_head(Q) ((Q)->next) +#define erts_circleq_tail(Q) ((Q)->prev) +#define erts_circleq_next(Q) ((Q)->next) +#define erts_circleq_prev(Q) ((Q)->prev) +#define erts_circleq_is_empty(Q) ((Q)->next == (void *)(Q)) + +#define erts_circleq_remove(N) \ + do { \ + (N)->next->prev = (N)->prev; \ + (N)->prev->next = (N)->next; \ + (N)->next = (N); \ + (N)->prev = (N); \ + } while(0) + +#define erts_circleq_pop_head(Q, N) \ + do { \ + (N) = (Q)->next; \ + (N)->next->prev = (N)->prev; \ + (N)->prev->next = (N)->next; \ + (N)->next = (N); \ + (N)->prev = (N); \ + } while(0) + +#define erts_circleq_pop_tail(Q, N) \ + do { \ + (N) = (Q)->prev; \ + (N)->next->prev = (N)->prev; \ + (N)->prev->next = (N)->next; \ + (N)->next = (N); \ + (N)->prev = (N); \ + } while(0) + +#define erts_circleq_push_head(Q, N) \ + do { \ + (N)->next = (Q)->next; \ + (N)->prev = (void *)(Q); \ + (Q)->next->prev = (N); \ + (Q)->next = (N); \ + } while(0) + +#define erts_circleq_push_tail(Q, N) \ + do { \ + (N)->prev = (Q)->prev; \ + (N)->next = (void *)(Q); \ + (Q)->prev->next = (N); \ + (Q)->prev = (N); \ + } while(0) + +#define erts_circleq_foreach(V, Q) \ + for ((V) = (Q)->next; (V) != (const void *)(Q); (V) = (V)->next) + +#define erts_circleq_foreach_reverse(V, Q) \ + for ((V) = (Q)->prev; (V) != (const void *)(Q); (V) = (V)->prev) + +#endif diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c index f7756f99bc..19dffd0ea4 100755 --- a/erts/emulator/sys/win32/sys.c +++ b/erts/emulator/sys/win32/sys.c @@ -57,11 +57,13 @@ extern void _dosmaperr(DWORD); #define __argv e_argv #endif +typedef struct driver_data DriverData; + static void init_console(); static int get_and_remove_option(int* argc, char** argv, const char* option); static char *get_and_remove_option2(int *argc, char **argv, const char *option); -static int init_async_io(struct async_io* aio, int use_threads); +static int init_async_io(DriverData *dp, struct async_io* aio, int use_threads); static void release_async_io(struct async_io* aio, ErlDrvPort); static void async_read_file(struct async_io* aio, LPVOID buf, DWORD numToRead); static int async_write_file(struct async_io* aio, LPVOID buf, DWORD numToWrite); @@ -96,7 +98,7 @@ static erts_smp_atomic_t pipe_creation_counter; static int driver_write(long, HANDLE, byte*, int); static int create_file_thread(struct async_io* aio, int mode); #ifdef ERTS_SMP -static void close_active_handle(ErlDrvPort, HANDLE handle); +static void close_active_handle(DriverData *, HANDLE handle); static DWORD WINAPI threaded_handle_closer(LPVOID param); #endif static DWORD WINAPI threaded_reader(LPVOID param); @@ -440,6 +442,8 @@ typedef struct async_io { DWORD bytesTransferred; /* Bytes read or write in the last operation. * Valid only when DF_OVR_READY is set. */ + DriverData *dp; /* Pointer to driver data struct which + this struct is part of */ } AsyncIo; @@ -458,7 +462,7 @@ static BOOL (WINAPI *fpSetHandleInformation)(HANDLE,DWORD,DWORD); * none of the file handles. */ -typedef struct driver_data { +struct driver_data { int totalNeeded; /* Total number of bytes needed to fill * up the packet header or packet. */ int bytesInBuffer; /* Number of bytes read so far in @@ -476,7 +480,8 @@ typedef struct driver_data { AsyncIo in; /* Control block for overlapped reading. */ AsyncIo out; /* Control block for overlapped writing. */ int report_exit; /* Do report exit status for the port */ -} DriverData; + erts_atomic32_t refc; /* References to this struct */ +}; /* Driver interfaces */ static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); @@ -581,6 +586,26 @@ struct erl_drv_entry vanilla_driver_entry = { stop_select }; +static ERTS_INLINE void +refer_driver_data(DriverData *dp) +{ +#ifdef DEBUG + erts_aint32_t refc = erts_atomic32_inc_read_nob(&dp->refc); + ASSERT(refc > 1); +#else + erts_atomic32_inc_nob(&dp->refc); +#endif +} + +static ERTS_INLINE void +unrefer_driver_data(DriverData *dp) +{ + erts_aint32_t refc = erts_atomic32_dec_read_mb(&dp->refc); + ASSERT(refc >= 0); + if (refc == 0) + driver_free(dp); +} + /* * Initialises a DriverData structure. * @@ -604,6 +629,7 @@ new_driver_data(ErlDrvPort port_num, int packet_bytes, int wait_objs_required, i * any more, since driver_select() can't fail. */ + erts_atomic32_init_nob(&dp->refc, 1); dp->bytesInBuffer = 0; dp->totalNeeded = packet_bytes; dp->inBufSize = PORT_BUFSIZ; @@ -616,9 +642,9 @@ new_driver_data(ErlDrvPort port_num, int packet_bytes, int wait_objs_required, i dp->port_num = port_num; dp->packet_bytes = packet_bytes; dp->port_pid = INVALID_HANDLE_VALUE; - if (init_async_io(&dp->in, use_threads) == -1) + if (init_async_io(dp, &dp->in, use_threads) == -1) goto async_io_error1; - if (init_async_io(&dp->out, use_threads) == -1) + if (init_async_io(dp, &dp->out, use_threads) == -1) goto async_io_error2; return dp; @@ -662,7 +688,7 @@ release_driver_data(DriverData* dp) dp->in.fd = INVALID_HANDLE_VALUE; DEBUGF(("Waiting for the in event thingie")); if (WaitForSingleObject(dp->in.ov.hEvent,timeout) == WAIT_TIMEOUT) { - close_active_handle(dp->port_num, dp->in.ov.hEvent); + close_active_handle(dp, dp->in.ov.hEvent); dp->in.ov.hEvent = NULL; timeout = 0; } @@ -673,7 +699,7 @@ release_driver_data(DriverData* dp) dp->out.fd = INVALID_HANDLE_VALUE; DEBUGF(("Waiting for the out event thingie")); if (WaitForSingleObject(dp->out.ov.hEvent,timeout) == WAIT_TIMEOUT) { - close_active_handle(dp->port_num, dp->out.ov.hEvent); + close_active_handle(dp, dp->out.ov.hEvent); dp->out.ov.hEvent = NULL; } DEBUGF(("...done\n")); @@ -719,7 +745,7 @@ release_driver_data(DriverData* dp) * the exit thread. */ - driver_free(dp); + unrefer_driver_data(dp); } #ifdef ERTS_SMP @@ -727,11 +753,12 @@ release_driver_data(DriverData* dp) struct handles_to_be_closed { HANDLE handles[MAXIMUM_WAIT_OBJECTS]; unsigned cnt; + DriverData *dp; }; static struct handles_to_be_closed* htbc_curr = NULL; CRITICAL_SECTION htbc_lock; -static void close_active_handle(ErlDrvPort port_num, HANDLE handle) +static void close_active_handle(DriverData *dp, HANDLE handle) { struct handles_to_be_closed* htbc; int i; @@ -745,11 +772,14 @@ static void close_active_handle(ErlDrvPort port_num, HANDLE handle) sizeof(*htbc)); htbc->handles[0] = CreateAutoEvent(FALSE); htbc->cnt = 1; + htbc->dp = dp; + refer_driver_data(dp); /* Need to keep driver data until we have + closed the event; outstanding operation + might write into it.. */ thread = (HANDLE *) _beginthreadex(NULL, 0, threaded_handle_closer, htbc, 0, &tid); CloseHandle(thread); } htbc->handles[htbc->cnt++] = handle; - driver_select(port_num, (ErlDrvEvent)handle, ERL_DRV_USE_NO_CALLBACK, 0); SetEvent(htbc->handles[0]); htbc_curr = htbc; LeaveCriticalSection(&htbc_lock); @@ -798,6 +828,7 @@ threaded_handle_closer(LPVOID param) } LeaveCriticalSection(&htbc_lock); CloseHandle(htbc->handles[0]); + unrefer_driver_data(htbc->dp); erts_free(ERTS_ALC_T_DRV_TAB, htbc); DEBUGF(("threaded_handle_closer %p terminating\r\n", htbc)); return 0; @@ -864,8 +895,9 @@ reuse_driver_data(DriverData *dp, HANDLE ifd, HANDLE ofd, int read_write, ErlDrv */ static int -init_async_io(AsyncIo* aio, int use_threads) +init_async_io(DriverData *dp, AsyncIo* aio, int use_threads) { + aio->dp = dp; aio->flags = 0; aio->thread = (HANDLE) -1; aio->fd = INVALID_HANDLE_VALUE; @@ -884,6 +916,8 @@ init_async_io(AsyncIo* aio, int use_threads) if (aio->ov.hEvent == NULL) return -1; if (use_threads) { + OV_BUFFER_PTR(aio) = NULL; + OV_NUM_TO_READ(aio) = 0; aio->ioAllowed = CreateAutoEvent(FALSE); if (aio->ioAllowed == NULL) return -1; @@ -914,12 +948,8 @@ release_async_io(AsyncIo* aio, ErlDrvPort port_num) CloseHandle(aio->fd); aio->fd = INVALID_HANDLE_VALUE; - if (aio->ov.hEvent != NULL) { - (void) driver_select(port_num, - (ErlDrvEvent)aio->ov.hEvent, - ERL_DRV_USE, 0); - /* was CloseHandle(aio->ov.hEvent); */ - } + if (aio->ov.hEvent != NULL) + CloseHandle(aio->ov.hEvent); aio->ov.hEvent = NULL; @@ -1287,12 +1317,15 @@ create_file_thread(AsyncIo* aio, int mode) { DWORD tid; /* Id for thread. */ + refer_driver_data(aio->dp); aio->thread = (HANDLE) _beginthreadex(NULL, 0, (mode & DO_WRITE) ? threaded_writer : threaded_reader, aio, 0, &tid); - - return aio->thread != (HANDLE) -1; + if (aio->thread != (HANDLE) -1) + return 1; + unrefer_driver_data(aio->dp); + return 0; } /* @@ -2078,6 +2111,7 @@ threaded_reader(LPVOID param) if (aio->flags & DF_EXIT_THREAD) break; } + unrefer_driver_data(aio->dp); return 0; } @@ -2157,6 +2191,7 @@ threaded_writer(LPVOID param) } CloseHandle(aio->fd); aio->fd = INVALID_HANDLE_VALUE; + unrefer_driver_data(aio->dp); return 0; } @@ -2297,6 +2332,7 @@ static void fd_stop(ErlDrvData data) (void) driver_select(dp->port_num, (ErlDrvEvent)dp->out.ov.hEvent, ERL_DRV_WRITE, 0); + ASSERT(dp->out.flushEvent); SetEvent(dp->out.flushEvent); WaitForSingleObject(dp->out.flushReplyEvent, INFINITE); } @@ -2349,12 +2385,12 @@ stop(ErlDrvData data) if (dp->in.ov.hEvent != NULL) { (void) driver_select(dp->port_num, (ErlDrvEvent)dp->in.ov.hEvent, - ERL_DRV_READ, 0); + ERL_DRV_READ|ERL_DRV_USE_NO_CALLBACK, 0); } if (dp->out.ov.hEvent != NULL) { (void) driver_select(dp->port_num, (ErlDrvEvent)dp->out.ov.hEvent, - ERL_DRV_WRITE, 0); + ERL_DRV_WRITE|ERL_DRV_USE_NO_CALLBACK, 0); } if (dp->out.thread == (HANDLE) -1 && dp->in.thread == (HANDLE) -1) { @@ -2366,6 +2402,8 @@ stop(ErlDrvData data) */ HANDLE thread; DWORD tid; + + /* threaded_exiter implicitly takes over refc from us... */ thread = (HANDLE *) _beginthreadex(NULL, 0, threaded_exiter, dp, 0, &tid); CloseHandle(thread); } |