aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator/beam
diff options
context:
space:
mode:
Diffstat (limited to 'erts/emulator/beam')
-rw-r--r--erts/emulator/beam/beam_debug.c63
-rw-r--r--erts/emulator/beam/beam_emu.c17
-rw-r--r--erts/emulator/beam/beam_load.c6
-rw-r--r--erts/emulator/beam/bif.c126
-rw-r--r--erts/emulator/beam/bif.tab4
-rw-r--r--erts/emulator/beam/break.c2
-rw-r--r--erts/emulator/beam/copy.c5
-rw-r--r--erts/emulator/beam/erl_alloc.types6
-rw-r--r--erts/emulator/beam/erl_async.h8
-rw-r--r--erts/emulator/beam/erl_bif_info.c23
-rw-r--r--erts/emulator/beam/erl_bif_re.c35
-rw-r--r--erts/emulator/beam/erl_db.c2057
-rw-r--r--erts/emulator/beam/erl_db.h47
-rw-r--r--erts/emulator/beam/erl_db_hash.c2003
-rw-r--r--erts/emulator/beam/erl_db_hash.h12
-rw-r--r--erts/emulator/beam/erl_db_tree.c511
-rw-r--r--erts/emulator/beam/erl_db_util.c180
-rw-r--r--erts/emulator/beam/erl_db_util.h103
-rw-r--r--erts/emulator/beam/erl_init.c53
-rw-r--r--erts/emulator/beam/erl_lock_check.c6
-rw-r--r--erts/emulator/beam/erl_mtrace.c7
-rw-r--r--erts/emulator/beam/erl_nif.c37
-rw-r--r--erts/emulator/beam/erl_node_tables.c41
-rw-r--r--erts/emulator/beam/erl_process.c130
-rw-r--r--erts/emulator/beam/erl_process.h59
-rw-r--r--erts/emulator/beam/erl_rbtree.h66
-rw-r--r--erts/emulator/beam/erl_term.h15
-rw-r--r--erts/emulator/beam/global.h15
-rw-r--r--erts/emulator/beam/ops.tab14
-rw-r--r--erts/emulator/beam/sys.h8
-rw-r--r--erts/emulator/beam/utils.c47
31 files changed, 3626 insertions, 2080 deletions
diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 8326d348af..37d8699a4b 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -853,15 +853,23 @@ dirty_test(Process *c_p, Eterm type, Eterm arg1, Eterm arg2, UWord *I)
goto badarg;
esdp = erts_proc_sched_data(c_p);
if (!esdp)
- ERTS_BIF_PREP_RET(ret, am_error);
- else if (!ERTS_SCHEDULER_IS_DIRTY(esdp))
+ goto scheduler_type_error;
+
+ switch (esdp->type) {
+ case ERTS_SCHED_NORMAL:
ERTS_BIF_PREP_RET(ret, am_normal);
- else if (ERTS_SCHEDULER_IS_DIRTY_CPU(esdp))
+ break;
+ case ERTS_SCHED_DIRTY_CPU:
ERTS_BIF_PREP_RET(ret, am_dirty_cpu);
- else if (ERTS_SCHEDULER_IS_DIRTY_IO(esdp))
+ break;
+ case ERTS_SCHED_DIRTY_IO:
ERTS_BIF_PREP_RET(ret, am_dirty_io);
- else
+ break;
+ default:
+ scheduler_type_error:
ERTS_BIF_PREP_RET(ret, am_error);
+ break;
+ }
}
else if (am_error == arg1) {
switch (arg2) {
@@ -1118,3 +1126,48 @@ ms_wait(Process *c_p, Eterm etimeout, int busy)
}
#endif /* ERTS_DIRTY_SCHEDULERS */
+
+#ifdef ERTS_SMP
+# define ERTS_STACK_LIMIT ((char *) ethr_get_stacklimit())
+#else
+# define ERTS_STACK_LIMIT ((char *) erts_scheduler_stack_limit)
+#endif
+
+/*
+ * The below functions is for testing of the stack
+ * limit functionality. They are intentionally
+ * written body recursive in order to prevent
+ * last call optimization...
+ */
+
+UWord
+erts_check_stack_recursion_downwards(char *start_c)
+{
+ char *limit = ERTS_STACK_LIMIT;
+ char c;
+ UWord res;
+ if (erts_check_below_limit(&c, limit + 1024))
+ return (char *) erts_ptr_id(start_c) - (char *) erts_ptr_id(&c);
+ res = erts_check_stack_recursion_downwards(start_c);
+ erts_ptr_id(&c);
+ return res;
+}
+
+UWord
+erts_check_stack_recursion_upwards(char *start_c)
+{
+ char *limit = ERTS_STACK_LIMIT;
+ char c;
+ UWord res;
+ if (erts_check_above_limit(&c, limit - 1024))
+ return (char *) erts_ptr_id(&c) - (char *) erts_ptr_id(start_c);
+ res = erts_check_stack_recursion_upwards(start_c);
+ erts_ptr_id(&c);
+ return res;
+}
+
+int
+erts_is_above_stack_limit(char *ptr)
+{
+ return (char *) ptr > ERTS_STACK_LIMIT;
+}
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 8be0f58227..6010c17c17 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -850,6 +850,15 @@ do { \
} while (0)
#endif
+#define IsTaggedTuple(Src,Arityval,Tag,Fail) \
+ do { \
+ if (!(is_tuple(Src) && \
+ (tuple_val(Src))[0] == Arityval && \
+ (tuple_val(Src))[1] == Tag)) { \
+ Fail; \
+ } \
+ } while (0)
+
#define IsBoolean(X, Fail) if ((X) != am_true && (X) != am_false) { Fail; }
#define IsBinary(Src, Fail) \
@@ -3035,10 +3044,12 @@ do { \
GetArg2(2, Op1, Op2);
if (is_both_small(Op1, Op2)) {
/*
- * We could extract the tag from one argument, but a tag extraction
- * could mean a shift. Therefore, play it safe here.
+ * TAG ^ TAG == 0.
+ *
+ * Therefore, we perform the XOR operation on the tagged values,
+ * and OR in the tag bits.
*/
- Eterm result = make_small(signed_val(Op1) ^ signed_val(Op2));
+ Eterm result = (Op1 ^ Op2) | make_small(0);
StoreBifResult(4, result);
}
DO_OUTLINED_ARITH_2(bxor, Op1, Op2);
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 48206a75a8..6eea963016 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -6321,7 +6321,7 @@ erts_make_stub_module(Process* p, Eterm hipe_magic_bin, Eterm Beam, Eterm Info)
*/
magic = erts_alloc_loader_state();
stp = ERTS_MAGIC_BIN_DATA(magic);
- hipe_code = erts_alloc(ERTS_ALC_T_HIPE, sizeof(*hipe_code));
+ hipe_code = erts_alloc(ERTS_ALC_T_HIPE_LL, sizeof(*hipe_code));
if (!is_internal_magic_ref(hipe_magic_bin) ||
!(hipe_magic = erts_magic_ref2bin(hipe_magic_bin),
@@ -6556,7 +6556,7 @@ erts_make_stub_module(Process* p, Eterm hipe_magic_bin, Eterm Beam, Eterm Info)
}
error:
- erts_free(ERTS_ALC_T_HIPE, hipe_code);
+ erts_free(ERTS_ALC_T_HIPE_LL, hipe_code);
erts_free_aligned_binary_bytes(temp_alloc);
free_loader_state(magic);
BIF_ERROR(p, BADARG);
@@ -6583,7 +6583,7 @@ int erts_commit_hipe_patch_load(Eterm hipe_magic_bin)
/*
* Initialise HiPE module
*/
- hipe_code = erts_alloc(ERTS_ALC_T_HIPE, sizeof(*hipe_code));
+ hipe_code = erts_alloc(ERTS_ALC_T_HIPE_LL, sizeof(*hipe_code));
hipe_code->text_segment = hipe_stp->text_segment;
hipe_code->text_segment_size = hipe_stp->text_segment_size;
hipe_code->data_segment = hipe_stp->data_segment;
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index 614bedab12..2d37f977c0 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -4254,6 +4254,132 @@ BIF_RETTYPE list_to_pid_1(BIF_ALIST_1)
BIF_ERROR(BIF_P, BADARG);
}
+BIF_RETTYPE list_to_ref_1(BIF_ALIST_1)
+{
+ /*
+ * A valid reference is on the format
+ * "#Ref<N.X.Y.Z>" where N, X, Y, and Z are
+ * 32-bit integers (i.e., max 10 characters).
+ */
+ Eterm *hp;
+ Eterm res;
+ Uint32 refn[ERTS_MAX_REF_NUMBERS];
+ int n = 0;
+ Uint ints[1 + ERTS_MAX_REF_NUMBERS] = {0};
+ char* cp;
+ Sint i;
+ DistEntry *dep = NULL;
+ char buf[5 /* #Ref< */
+ + (1 + ERTS_MAX_REF_NUMBERS)*(10 + 1) /* N.X.Y.Z> */
+ + 1 /* \0 */];
+
+ /* walk down the list and create a C string */
+ if ((i = intlist_to_buf(BIF_ARG_1, buf, sizeof(buf)-1)) < 0)
+ goto bad;
+
+ buf[i] = '\0'; /* null terminal */
+
+ cp = &buf[0];
+ if (*cp++ != '#') goto bad;
+ if (*cp++ != 'R') goto bad;
+ if (*cp++ != 'e') goto bad;
+ if (*cp++ != 'f') goto bad;
+ if (*cp++ != '<') goto bad;
+
+ for (i = 0; i < sizeof(ints)/sizeof(Uint); i++) {
+ if (*cp < '0' || *cp > '9') goto bad;
+
+ while (*cp >= '0' && *cp <= '9') {
+ ints[i] = 10*ints[i] + (*cp - '0');
+ cp++;
+ }
+
+ n++;
+ if (ints[i] > ~((Uint32) 0)) goto bad;
+ if (*cp == '>') break;
+ if (*cp++ != '.') goto bad;
+ }
+
+ if (*cp++ != '>') goto bad;
+ if (*cp != '\0') goto bad;
+
+ if (n < 2) goto bad;
+
+ for (n = 0; i > 0; i--)
+ refn[n++] = (Uint32) ints[i];
+
+ ASSERT(n <= ERTS_MAX_REF_NUMBERS);
+
+ dep = erts_channel_no_to_dist_entry(ints[0]);
+
+ if (!dep)
+ goto bad;
+
+ if(dep == erts_this_dist_entry) {
+ ErtsMagicBinary *mb;
+ Uint32 sid;
+ if (refn[0] > MAX_REFERENCE) goto bad;
+ if (n != ERTS_REF_NUMBERS) goto bad;
+ sid = erts_get_ref_numbers_thr_id(refn);
+ if (sid > erts_no_schedulers) goto bad;
+ mb = erts_magic_ref_lookup_bin(refn);
+ if (mb) {
+ hp = HAlloc(BIF_P, ERTS_MAGIC_REF_THING_SIZE);
+ res = erts_mk_magic_ref(&hp, &BIF_P->off_heap,
+ (Binary *) mb);
+ }
+ else {
+ hp = HAlloc(BIF_P, ERTS_REF_THING_SIZE);
+ write_ref_thing(hp, refn[0], refn[1], refn[2]);
+ res = make_internal_ref(hp);
+ }
+ }
+ else {
+ ExternalThing *etp;
+ ErlNode *enp;
+ Uint hsz;
+ int j;
+
+ if (is_nil(dep->cid))
+ goto bad;
+
+ enp = erts_find_or_insert_node(dep->sysname, dep->creation);
+ ASSERT(enp != erts_this_node);
+
+ hsz = EXTERNAL_THING_HEAD_SIZE;
+#if defined(ARCH_64)
+ hsz += n/2 + 1;
+#else
+ hsz += n;
+#endif
+
+ etp = (ExternalThing *) HAlloc(BIF_P, hsz);
+ etp->header = make_external_ref_header(n/2);
+ etp->next = BIF_P->off_heap.first;
+ etp->node = enp;
+ i = 0;
+#if defined(ARCH_64)
+ etp->data.ui32[i] = n;
+#endif
+ for (j = 0; j < n; j++) {
+ etp->data.ui32[i] = refn[j];
+ i++;
+ }
+
+ BIF_P->off_heap.first = (struct erl_off_heap_header*) etp;
+ res = make_external_ref(etp);
+ }
+
+ erts_deref_dist_entry(dep);
+ BIF_RET(res);
+
+ bad:
+ if (dep)
+ erts_deref_dist_entry(dep);
+ BIF_ERROR(BIF_P, BADARG);
+}
+
+
/**********************************************************************/
BIF_RETTYPE group_leader_0(BIF_ALIST_0)
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index ce2cffa498..4140938210 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -88,6 +88,7 @@ bif erlang:list_to_binary/1
bif erlang:list_to_float/1
bif erlang:list_to_integer/1
bif erlang:list_to_pid/1
+bif erlang:list_to_ref/1
bif erlang:list_to_tuple/1
bif erlang:loaded/0
bif erlang:localtime/0
@@ -324,7 +325,7 @@ bif erlang:match_spec_test/3
# Bifs in ets module.
#
-bif ets:all/0
+bif ets:internal_request_all/0
bif ets:new/2
bif ets:delete/1
bif ets:delete/2
@@ -360,6 +361,7 @@ bif ets:select_reverse/1
bif ets:select_reverse/2
bif ets:select_reverse/3
bif ets:select_delete/2
+bif ets:select_replace/2
bif ets:match_spec_compile/1
bif ets:match_spec_run_r/3
diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c
index 4dad5736c5..0b40d70cb7 100644
--- a/erts/emulator/beam/break.c
+++ b/erts/emulator/beam/break.c
@@ -512,6 +512,8 @@ do_break(void)
erts_free_read_env(mode);
#endif /* __WIN32__ */
+ ASSERT(erts_smp_thr_progress_is_blocking());
+
erts_printf("\n"
"BREAK: (a)bort (c)ontinue (p)roc info (i)nfo (l)oaded\n"
" (v)ersion (k)ill (D)b-tables (d)istribution\n");
diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c
index e567eabc82..264ba89e8b 100644
--- a/erts/emulator/beam/copy.c
+++ b/erts/emulator/beam/copy.c
@@ -286,11 +286,6 @@ do { \
(dst) = result; \
} while(0)
-#define BOXED_VISITED_MASK ((Eterm) 3)
-#define BOXED_VISITED ((Eterm) 1)
-#define BOXED_SHARED_UNPROCESSED ((Eterm) 2)
-#define BOXED_SHARED_PROCESSED ((Eterm) 3)
-
#define COUNT_OFF_HEAP (0)
/*
diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types
index 3d5de72ee7..43f43f9034 100644
--- a/erts/emulator/beam/erl_alloc.types
+++ b/erts/emulator/beam/erl_alloc.types
@@ -227,6 +227,7 @@ type DB_DMC_ERROR ETS ETS db_dmc_error
type DB_DMC_ERR_INFO ETS ETS db_dmc_error_info
type DB_TERM ETS ETS db_term
type DB_PROC_CLEANUP SHORT_LIVED ETS db_proc_cleanup_state
+type ETS_ALL_REQ SHORT_LIVED ETS ets_all_request
type INSTR_INFO LONG_LIVED SYSTEM instr_info
type LOGGER_DSBUF TEMPORARY SYSTEM logger_dsbuf
type TMP_DSBUF TEMPORARY SYSTEM tmp_dsbuf
@@ -350,8 +351,9 @@ type SL_MPATHS SHORT_LIVED SYSTEM sl_migration_paths
+if hipe
-# Currently most hipe code use this type.
-type HIPE SYSTEM SYSTEM hipe_data
+type HIPE_LL LONG_LIVED SYSTEM hipe_long_lived
+type HIPE_SL SHORT_LIVED SYSTEM hipe_short_lived
+type HIPE_STK STANDARD SYSTEM hipe_nstack
+if exec_alloc
type HIPE_EXEC EXEC CODE hipe_code
diff --git a/erts/emulator/beam/erl_async.h b/erts/emulator/beam/erl_async.h
index 473c7686e5..c884a5040d 100644
--- a/erts/emulator/beam/erl_async.h
+++ b/erts/emulator/beam/erl_async.h
@@ -27,7 +27,6 @@ extern int erts_async_max_threads;
#define ERTS_ASYNC_THREAD_MAX_STACK_SIZE 8192 /* Kilo words */
extern int erts_async_thread_suggested_stack_size;
-#ifdef USE_THREADS
#ifdef ERTS_SMP
/*
@@ -47,6 +46,10 @@ extern int erts_async_thread_suggested_stack_size;
# define ERTS_USE_ASYNC_READY_Q 0
#endif
+#ifndef USE_THREADS
+# undef ERTS_USE_ASYNC_READY_Q
+# define ERTS_USE_ASYNC_READY_Q 0
+#endif /* !USE_THREADS */
#if ERTS_USE_ASYNC_READY_Q
int erts_check_async_ready(void *);
int erts_async_ready_clean(void *, void *);
@@ -58,10 +61,7 @@ void *erts_get_async_ready_queue(Uint sched_id);
#endif
#endif /* ERTS_USE_ASYNC_READY_Q */
-#endif /* USE_THREADS */
-
void erts_init_async(void);
void erts_exit_flush_async(void);
-
#endif /* ERL_ASYNC_H__ */
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index 06a73ffea5..3a8687dc59 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -3634,10 +3634,6 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1)
BIF_RET(TUPLE2(hp, make_small((Uint) words),
erts_ets_hash_sizeof_ext_segtab()));
}
- else if (ERTS_IS_ATOM_STR("DbTable_meta", BIF_ARG_1)) {
- /* Used by ets_SUITE (stdlib) */
- BIF_RET(erts_ets_get_meta_state(BIF_P));
- }
else if (ERTS_IS_ATOM_STR("check_io_debug", BIF_ARG_1)) {
/* Used by driver_SUITE (emulator) */
Uint sz, *szp;
@@ -3747,6 +3743,21 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1)
BIF_RET(erts_sint64_to_big(value, &hp));
}
}
+ else if (ERTS_IS_ATOM_STR("stack_check", BIF_ARG_1)) {
+ UWord size;
+ char c;
+ if (erts_is_above_stack_limit(&c))
+ size = erts_check_stack_recursion_downwards(&c);
+ else
+ size = erts_check_stack_recursion_upwards(&c);
+ if (IS_SSMALL(size))
+ BIF_RET(make_small(size));
+ else {
+ Uint hsz = BIG_UWORD_HEAP_SIZE(size);
+ Eterm *hp = HAlloc(BIF_P, hsz);
+ BIF_RET(uword_to_big(size, hp));
+ }
+ }
}
else if (is_tuple(BIF_ARG_1)) {
Eterm* tp = tuple_val(BIF_ARG_1);
@@ -4398,10 +4409,6 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2)
}
BIF_RET(am_ok);
}
- else if (ERTS_IS_ATOM_STR("DbTable_meta", BIF_ARG_1)) {
- /* Used by ets_SUITE (stdlib) */
- BIF_RET(erts_ets_restore_meta_state(BIF_P, BIF_ARG_2));
- }
else if (ERTS_IS_ATOM_STR("make", BIF_ARG_1)) {
if (ERTS_IS_ATOM_STR("magic_ref", BIF_ARG_2)) {
Binary *bin = erts_create_magic_binary(0, empty_magic_ref_destructor);
diff --git a/erts/emulator/beam/erl_bif_re.c b/erts/emulator/beam/erl_bif_re.c
index a66b05c6ff..925d99e0de 100644
--- a/erts/emulator/beam/erl_bif_re.c
+++ b/erts/emulator/beam/erl_bif_re.c
@@ -64,12 +64,47 @@ static void erts_erts_pcre_stack_free(void *ptr) {
erts_free(ERTS_ALC_T_RE_STACK,ptr);
}
+#define ERTS_PCRE_STACK_MARGIN (10*1024)
+
+#ifdef ERTS_SMP
+# define ERTS_STACK_LIMIT ((char *) ethr_get_stacklimit())
+#else
+# define ERTS_STACK_LIMIT ((char *) erts_scheduler_stack_limit)
+#endif
+
+static int
+stack_guard_downwards(void)
+{
+ char *limit = ERTS_STACK_LIMIT;
+ char c;
+
+ ASSERT(limit);
+
+ return erts_check_below_limit(&c, limit + ERTS_PCRE_STACK_MARGIN);
+}
+
+static int
+stack_guard_upwards(void)
+{
+ char *limit = ERTS_STACK_LIMIT;
+ char c;
+
+ ASSERT(limit);
+
+ return erts_check_above_limit(&c, limit - ERTS_PCRE_STACK_MARGIN);
+}
+
void erts_init_bif_re(void)
{
+ char c;
erts_pcre_malloc = &erts_erts_pcre_malloc;
erts_pcre_free = &erts_erts_pcre_free;
erts_pcre_stack_malloc = &erts_erts_pcre_stack_malloc;
erts_pcre_stack_free = &erts_erts_pcre_stack_free;
+ if ((char *) erts_ptr_id(&c) > ERTS_STACK_LIMIT)
+ erts_pcre_stack_guard = stack_guard_downwards;
+ else
+ erts_pcre_stack_guard = stack_guard_upwards;
default_table = NULL; /* ISO8859-1 default, forced into pcre */
max_loop_limit = CONTEXT_REDS * LOOP_FACTOR;
diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c
index 9f077dd407..378328856d 100644
--- a/erts/emulator/beam/erl_db.c
+++ b/erts/emulator/beam/erl_db.c
@@ -19,9 +19,7 @@
*/
/*
- * This file contains the bif interface functions and
- * the handling of the "meta tables" ie the tables of
- * db tables.
+ * This file contains the 'ets' bif interface functions.
*/
/*
@@ -43,6 +41,7 @@
#include "erl_db.h"
#include "bif.h"
#include "big.h"
+#include "erl_binary.h"
erts_smp_atomic_t erts_ets_misc_mem_size;
@@ -74,62 +73,226 @@ enum DbIterSafety {
#define DID_TRAP(P,Ret) (!is_value(Ret) && ((P)->freason == TRAP))
+/*
+ * "fixed_tabs": list of all fixed tables for a process
+ */
+#ifdef DEBUG
+static int fixed_tabs_find(DbFixation* first, DbFixation* fix);
+#endif
-/*
-** The main meta table, containing all ets tables.
-*/
-#ifdef ERTS_SMP
+static void fixed_tabs_insert(Process* p, DbFixation* fix)
+{
+ DbFixation* first = erts_psd_get(p, ERTS_PSD_ETS_FIXED_TABLES);
+
+ if (!first) {
+ fix->tabs.next = fix->tabs.prev = fix;
+ erts_psd_set(p, ERTS_PSD_ETS_FIXED_TABLES, fix);
+ }
+ else {
+ ASSERT(!fixed_tabs_find(first, fix));
+ fix->tabs.prev = first->tabs.prev;
+ fix->tabs.next = first;
+ fix->tabs.prev->tabs.next = fix;
+ first->tabs.prev = fix;
+ }
+}
+
+static void fixed_tabs_delete(Process *p, DbFixation* fix)
+{
+ if (fix->tabs.next == fix) {
+ DbFixation* old;
+ ASSERT(fix->tabs.prev == fix);
+ old = erts_psd_set(p, ERTS_PSD_ETS_FIXED_TABLES, NULL);
+ ASSERT(old == fix); (void)old;
+ }
+ else {
+ DbFixation *first = (DbFixation*) erts_psd_get(p, ERTS_PSD_ETS_FIXED_TABLES);
-#define ERTS_META_MAIN_TAB_LOCK_TAB_BITS 8
-#define ERTS_META_MAIN_TAB_LOCK_TAB_SIZE (1 << ERTS_META_MAIN_TAB_LOCK_TAB_BITS)
-#define ERTS_META_MAIN_TAB_LOCK_TAB_MASK (ERTS_META_MAIN_TAB_LOCK_TAB_SIZE - 1)
+ ASSERT(fixed_tabs_find(first, fix));
+ fix->tabs.prev->tabs.next = fix->tabs.next;
+ fix->tabs.next->tabs.prev = fix->tabs.prev;
-typedef union {
- erts_smp_rwmtx_t rwmtx;
- byte cache_line_align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(
- sizeof(erts_smp_rwmtx_t))];
-} erts_meta_main_tab_lock_t;
+ if (fix == first)
+ erts_psd_set(p, ERTS_PSD_ETS_FIXED_TABLES, fix->tabs.next);
+ }
+}
-static erts_meta_main_tab_lock_t *meta_main_tab_locks;
+#ifdef DEBUG
+static int fixed_tabs_find(DbFixation* first, DbFixation* fix)
+{
+ DbFixation* p;
+ if (!first) {
+ first = (DbFixation*) erts_psd_get(fix->procs.p, ERTS_PSD_ETS_FIXED_TABLES);
+ }
+ p = first;
+ do {
+ if (p == fix)
+ return 1;
+ ASSERT(p->procs.p == fix->procs.p);
+ ASSERT(p->tabs.next->tabs.prev == p);
+ p = p->tabs.next;
+ } while (p != first);
+ return 0;
+}
#endif
-static struct {
- union {
- DbTable *tb; /* Only directly readable if slot is ALIVE */
- UWord next_free; /* (index<<2)|1 if slot is FREE */
- }u;
-} *meta_main_tab;
-/* A slot in meta_main_tab can have three states:
- * FREE : Free to use for new table. Part of linked free-list.
- * ALIVE: Contains a table
- * DEAD : Contains a table that is being removed.
+
+/*
+ * fixing_procs: tree of all processes fixating a table
*/
-#define IS_SLOT_FREE(i) (meta_main_tab[(i)].u.next_free & 1)
-#define IS_SLOT_DEAD(i) (meta_main_tab[(i)].u.next_free & 2)
-#define IS_SLOT_ALIVE(i) (!(meta_main_tab[(i)].u.next_free & (1|2)))
-#define GET_NEXT_FREE_SLOT(i) (meta_main_tab[(i)].u.next_free >> 2)
-#define SET_NEXT_FREE_SLOT(i,next) (meta_main_tab[(i)].u.next_free = ((next)<<2)|1)
-#define MARK_SLOT_DEAD(i) (meta_main_tab[(i)].u.next_free |= 2)
-#define GET_ANY_SLOT_TAB(i) ((DbTable*)(meta_main_tab[(i)].u.next_free & ~(1|2))) /* dead or alive */
+#define ERTS_RBT_PREFIX fixing_procs
+#define ERTS_RBT_T DbFixation
+#define ERTS_RBT_KEY_T Process*
+#define ERTS_RBT_FLAGS_T int
+#define ERTS_RBT_INIT_EMPTY_TNODE(T) \
+ do { \
+ (T)->procs.parent = NULL; \
+ (T)->procs.right = NULL; \
+ (T)->procs.left = NULL; \
+ } while (0)
+#define ERTS_RBT_IS_RED(T) ((T)->procs.is_red)
+#define ERTS_RBT_SET_RED(T) ((T)->procs.is_red = 1)
+#define ERTS_RBT_IS_BLACK(T) (!(T)->procs.is_red)
+#define ERTS_RBT_SET_BLACK(T) ((T)->procs.is_red = 0)
+#define ERTS_RBT_GET_FLAGS(T) ((T)->procs.is_red)
+#define ERTS_RBT_SET_FLAGS(T, F) ((T)->procs.is_red = (F))
+#define ERTS_RBT_GET_PARENT(T) ((T)->procs.parent)
+#define ERTS_RBT_SET_PARENT(T, P) ((T)->procs.parent = (P))
+#define ERTS_RBT_GET_RIGHT(T) ((T)->procs.right)
+#define ERTS_RBT_SET_RIGHT(T, R) ((T)->procs.right = (R))
+#define ERTS_RBT_GET_LEFT(T) ((T)->procs.left)
+#define ERTS_RBT_SET_LEFT(T, L) ((T)->procs.left = (L))
+#define ERTS_RBT_GET_KEY(T) ((T)->procs.p)
+#define ERTS_RBT_IS_LT(KX, KY) ((KX) < (KY))
+#define ERTS_RBT_IS_EQ(KX, KY) ((KX) == (KY))
+
+#define ERTS_RBT_WANT_INSERT
+#define ERTS_RBT_WANT_LOOKUP
+#define ERTS_RBT_WANT_DELETE
+#define ERTS_RBT_WANT_FOREACH
+#define ERTS_RBT_WANT_FOREACH_DESTROY
+#ifdef DEBUG
+# define ERTS_RBT_WANT_LOOKUP
+#endif
+#define ERTS_RBT_UNDEF
+
+#include "erl_rbtree.h"
+
+#ifdef HARDDEBUG
+# error Do something useful with CHECK_TABLES maybe
+#else
+# define CHECK_TABLES()
+#endif
+
-static ERTS_INLINE erts_smp_rwmtx_t *
-get_meta_main_tab_lock(unsigned slot)
+static void
+send_ets_transfer_message(Process *c_p, Process *proc,
+ ErtsProcLocks *locks,
+ DbTable *tb, Eterm heir_data);
+static void schedule_free_dbtable(DbTable* tb);
+static void delete_sched_table(Process *c_p, DbTable *tb);
+
+static void table_dec_refc(DbTable *tb, erts_aint_t min_val)
+{
+ if (erts_smp_refc_dectest(&tb->common.refc, min_val) == 0)
+ schedule_free_dbtable(tb);
+}
+
+static int
+db_table_tid_destructor(Binary *unused)
+{
+ return 1;
+}
+
+static ERTS_INLINE void
+make_btid(DbTable *tb)
+{
+ Binary *btid = erts_create_magic_indirection(db_table_tid_destructor);
+ erts_smp_atomic_t *tbref = erts_smp_binary_to_magic_indirection(btid);
+ erts_smp_atomic_init_nob(tbref, (erts_aint_t) tb);
+ tb->common.btid = btid;
+ /*
+ * Table and magic indirection refer eachother,
+ * and table is refered once by being alive...
+ */
+ erts_smp_refc_init(&tb->common.refc, 2);
+ erts_refc_inc(&btid->refc, 1);
+}
+
+static ERTS_INLINE DbTable* btid2tab(Binary* btid)
+{
+ erts_smp_atomic_t *tbref = erts_smp_binary_to_magic_indirection(btid);
+ return (DbTable *) erts_smp_atomic_read_nob(tbref);
+}
+
+static DbTable *
+tid2tab(Eterm tid)
+{
+ DbTable *tb;
+ Binary *btid;
+ erts_smp_atomic_t *tbref;
+ if (!is_internal_magic_ref(tid))
+ return NULL;
+
+ btid = erts_magic_ref2bin(tid);
+ if (ERTS_MAGIC_BIN_DESTRUCTOR(btid) != db_table_tid_destructor)
+ return NULL;
+
+ tbref = erts_smp_binary_to_magic_indirection(btid);
+ tb = (DbTable *) erts_smp_atomic_read_nob(tbref);
+
+ ASSERT(!tb || tb->common.btid == btid);
+
+ return tb;
+}
+
+static ERTS_INLINE int
+is_table_alive(DbTable *tb)
+{
+ erts_smp_atomic_t *tbref;
+ DbTable *rtb;
+
+ tbref = erts_smp_binary_to_magic_indirection(tb->common.btid);
+ rtb = (DbTable *) erts_smp_atomic_read_nob(tbref);
+
+ ASSERT(!rtb || rtb == tb);
+
+ return !!rtb;
+}
+
+static ERTS_INLINE int
+is_table_named(DbTable *tb)
{
#ifdef ERTS_SMP
- return &meta_main_tab_locks[slot & ERTS_META_MAIN_TAB_LOCK_TAB_MASK].rwmtx;
+ return tb->common.type & DB_NAMED_TABLE;
#else
- return NULL;
+ return tb->common.status & DB_NAMED_TABLE;
#endif
}
-static erts_smp_spinlock_t meta_main_tab_main_lock;
-static Uint meta_main_tab_first_free; /* Index of first free slot */
-static int meta_main_tab_cnt; /* Number of active tables */
-static int meta_main_tab_top; /* Highest ever used slot + 1 */
-static Uint meta_main_tab_slot_mask; /* The slot index part of an unnamed table id */
-static Uint meta_main_tab_seq_incr;
-static Uint meta_main_tab_seq_cnt = 0; /* To give unique(-ish) table identifiers */
+
+static ERTS_INLINE void
+tid_clear(Process *c_p, DbTable *tb)
+{
+ DbTable *rtb;
+ Binary *btid = tb->common.btid;
+ erts_smp_atomic_t *tbref = erts_smp_binary_to_magic_indirection(btid);
+ rtb = (DbTable *) erts_smp_atomic_xchg_nob(tbref, (erts_aint_t) NULL);
+ ASSERT(!rtb || tb == rtb);
+ if (rtb) {
+ table_dec_refc(tb, 1);
+ delete_sched_table(c_p, tb);
+ }
+}
+
+static ERTS_INLINE Eterm
+make_tid(Process *c_p, DbTable *tb)
+{
+ Eterm *hp = HAlloc(c_p, ERTS_MAGIC_REF_THING_SIZE);
+ return erts_mk_magic_ref(&hp, &c_p->off_heap, tb->common.btid);
+}
+
/*
** The meta hash table of all NAMED ets tables
@@ -181,8 +344,6 @@ int user_requested_db_max_tabs;
int erts_ets_realloc_always_moves;
int erts_ets_always_compress;
static int db_max_tabs;
-static DbTable *meta_pid_to_tab; /* Pid mapped to owned tables */
-static DbTable *meta_pid_to_fixed_tab; /* Pid mapped to fixed tables */
static Eterm ms_delete_all;
static Eterm ms_delete_all_buff[8]; /* To compare with for deletion
of all objects */
@@ -195,15 +356,13 @@ static void fix_table_locked(Process* p, DbTable* tb);
static void unfix_table_locked(Process* p, DbTable* tb, db_lock_kind_t* kind);
static void set_heir(Process* me, DbTable* tb, Eterm heir, UWord heir_data);
static void free_heir_data(DbTable*);
-static void free_fixations_locked(DbTable *tb);
+static SWord free_fixations_locked(Process* p, DbTable *tb);
-static int free_table_cont(Process *p,
- DbTable *tb,
- int first,
- int clean_meta_tab);
+static SWord free_table_continue(Process *p, DbTable *tb, SWord reds);
static void print_table(fmtfn_t to, void *to_arg, int show, DbTable* tb);
static BIF_RETTYPE ets_select_delete_1(BIF_ALIST_1);
static BIF_RETTYPE ets_select_count_1(BIF_ALIST_1);
+static BIF_RETTYPE ets_select_replace_1(BIF_ALIST_1);
static BIF_RETTYPE ets_select_trap_1(BIF_ALIST_1);
static BIF_RETTYPE ets_delete_trap(BIF_ALIST_1);
static Eterm table_info(Process* p, DbTable* tb, Eterm What);
@@ -218,6 +377,7 @@ static BIF_RETTYPE ets_select3(Process* p, Eterm arg1, Eterm arg2, Eterm arg3);
*/
Export ets_select_delete_continue_exp;
Export ets_select_count_continue_exp;
+Export ets_select_replace_continue_exp;
Export ets_select_continue_exp;
/*
@@ -235,23 +395,16 @@ free_dbtable(void *vtb)
erts_smp_atomic_read_nob(&tb->common.memory_size)-sizeof(DbTable),
tb->common.fixations);
}
- erts_fprintf(stderr, "ets: free_dbtable(%T) deleted!!!\r\n",
- tb->common.id);
-
- erts_fprintf(stderr, "ets: free_dbtable: meta_pid_to_tab common.memory_size = %ld\n",
- erts_smp_atomic_read_nob(&meta_pid_to_tab->common.memory_size));
- print_table(ERTS_PRINT_STDOUT, NULL, 1, meta_pid_to_tab);
-
-
- erts_fprintf(stderr, "ets: free_dbtable: meta_pid_to_fixed_tab common.memory_size = %ld\n",
- erts_smp_atomic_read_nob(&meta_pid_to_fixed_tab->common.memory_size));
- print_table(ERTS_PRINT_STDOUT, NULL, 1, meta_pid_to_fixed_tab);
#endif
#ifdef ERTS_SMP
erts_smp_rwmtx_destroy(&tb->common.rwlock);
erts_smp_mtx_destroy(&tb->common.fixlock);
#endif
ASSERT(is_immed(tb->common.heir_data));
+
+ if (tb->common.btid && erts_refc_dectest(&tb->common.btid->refc, 0) == 0)
+ erts_bin_free(tb->common.btid);
+
erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable));
}
@@ -266,13 +419,163 @@ static void schedule_free_dbtable(DbTable* tb)
* Caller is *not* allowed to access the specialized part
* (hash or tree) of *tb after this function has returned.
*/
- ASSERT(erts_smp_refc_read(&tb->common.ref, 0) == 0);
+ ASSERT(erts_smp_refc_read(&tb->common.refc, 0) == 0);
+ ASSERT(erts_smp_refc_read(&tb->common.fix_count, 0) == 0);
erts_schedule_thr_prgr_later_cleanup_op(free_dbtable,
(void *) tb,
&tb->release.data,
sizeof(DbTable));
}
+static ERTS_INLINE void
+save_sched_table(Process *c_p, DbTable *tb)
+{
+ ErtsSchedulerData *esdp = erts_proc_sched_data(c_p);
+ DbTable *first;
+
+ ASSERT(esdp);
+ esdp->ets_tables.count++;
+ erts_smp_refc_inc(&tb->common.refc, 1);
+
+ first = esdp->ets_tables.clist;
+ if (!first) {
+ tb->common.all.next = tb->common.all.prev = tb;
+ esdp->ets_tables.clist = tb;
+ }
+ else {
+ tb->common.all.prev = first->common.all.prev;
+ tb->common.all.next = first;
+ tb->common.all.prev->common.all.next = tb;
+ first->common.all.prev = tb;
+ }
+}
+
+static ERTS_INLINE void
+remove_sched_table(ErtsSchedulerData *esdp, DbTable *tb)
+{
+ ErtsEtsAllYieldData *eaydp;
+ ASSERT(esdp);
+ ASSERT(erts_get_ref_numbers_thr_id(ERTS_MAGIC_BIN_REFN(tb->common.btid))
+ == (Uint32) esdp->no);
+
+ ASSERT(esdp->ets_tables.count > 0);
+ esdp->ets_tables.count--;
+
+ eaydp = ERTS_SCHED_AUX_YIELD_DATA(esdp, ets_all);
+ if (eaydp->ongoing) {
+ /* ets:all() op process list from last to first... */
+ if (eaydp->tab == tb) {
+ if (eaydp->tab == esdp->ets_tables.clist)
+ eaydp->tab = NULL;
+ else
+ eaydp->tab = tb->common.all.prev;
+ }
+ }
+
+ if (tb->common.all.next == tb) {
+ ASSERT(tb->common.all.prev == tb);
+ ASSERT(esdp->ets_tables.clist == tb);
+ esdp->ets_tables.clist = NULL;
+ }
+ else {
+#ifdef DEBUG
+ DbTable *tmp = esdp->ets_tables.clist;
+ do {
+ if (tmp == tb) break;
+ tmp = tmp->common.all.next;
+ } while (tmp != esdp->ets_tables.clist);
+ ASSERT(tmp == tb);
+#endif
+ tb->common.all.prev->common.all.next = tb->common.all.next;
+ tb->common.all.next->common.all.prev = tb->common.all.prev;
+
+ if (esdp->ets_tables.clist == tb)
+ esdp->ets_tables.clist = tb->common.all.next;
+
+ }
+
+ table_dec_refc(tb, 0);
+}
+
+static void
+scheduled_remove_sched_table(void *vtb)
+{
+ remove_sched_table(erts_get_scheduler_data(), (DbTable *) vtb);
+}
+
+static void
+delete_sched_table(Process *c_p, DbTable *tb)
+{
+ ErtsSchedulerData *esdp = erts_proc_sched_data(c_p);
+ Uint32 sid;
+
+ ASSERT(esdp);
+
+ ASSERT(tb->common.btid);
+ sid = erts_get_ref_numbers_thr_id(ERTS_MAGIC_BIN_REFN(tb->common.btid));
+ ASSERT(1 <= sid && sid <= erts_no_schedulers);
+ if (sid == (Uint32) esdp->no)
+ remove_sched_table(esdp, tb);
+ else
+ erts_schedule_misc_aux_work((int) sid, scheduled_remove_sched_table, tb);
+}
+
+static ERTS_INLINE void
+save_owned_table(Process *c_p, DbTable *tb)
+{
+ DbTable *first;
+
+ erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_STATUS);
+
+ first = (DbTable*) erts_psd_get(c_p, ERTS_PSD_ETS_OWNED_TABLES);
+
+ erts_smp_refc_inc(&tb->common.refc, 1);
+
+ if (!first) {
+ tb->common.owned.next = tb->common.owned.prev = tb;
+ erts_psd_set(c_p, ERTS_PSD_ETS_OWNED_TABLES, tb);
+ }
+ else {
+ tb->common.owned.prev = first->common.owned.prev;
+ tb->common.owned.next = first;
+ tb->common.owned.prev->common.owned.next = tb;
+ first->common.owned.prev = tb;
+ }
+ erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_STATUS);
+}
+
+static ERTS_INLINE void
+delete_owned_table(Process *p, DbTable *tb)
+{
+ erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS);
+ if (tb->common.owned.next == tb) {
+ DbTable* old;
+ ASSERT(tb->common.owned.prev == tb);
+ old = erts_psd_set(p, ERTS_PSD_ETS_OWNED_TABLES, NULL);
+ ASSERT(old == tb); (void)old;
+ }
+ else {
+ DbTable *first = (DbTable*) erts_psd_get(p, ERTS_PSD_ETS_OWNED_TABLES);
+#ifdef DEBUG
+ DbTable *tmp = first;
+ do {
+ if (tmp == tb) break;
+ tmp = tmp->common.owned.next;
+ } while (tmp != first);
+ ASSERT(tmp == tb);
+#endif
+ tb->common.owned.prev->common.owned.next = tb->common.owned.next;
+ tb->common.owned.next->common.owned.prev = tb->common.owned.prev;
+
+ if (tb == first)
+ erts_psd_set(p, ERTS_PSD_ETS_OWNED_TABLES, tb->common.owned.next);
+ }
+ erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS);
+
+ table_dec_refc(tb, 1);
+}
+
+
static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock,
char *rwname, char* fixname)
{
@@ -294,7 +597,6 @@ static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock,
static ERTS_INLINE void db_lock(DbTable* tb, db_lock_kind_t kind)
{
#ifdef ERTS_SMP
- ASSERT(tb != meta_pid_to_tab && tb != meta_pid_to_fixed_tab);
if (tb->common.type & DB_FINE_LOCKED) {
if (kind == LCK_WRITE) {
erts_smp_rwmtx_rwlock(&tb->common.rwlock);
@@ -327,8 +629,6 @@ static ERTS_INLINE void db_unlock(DbTable* tb, db_lock_kind_t kind)
* to follow the tb pointer!
*/
#ifdef ERTS_SMP
- ASSERT(tb != meta_pid_to_tab && tb != meta_pid_to_fixed_tab);
-
if (tb->common.type & DB_FINE_LOCKED) {
if (kind == LCK_WRITE) {
ASSERT(tb->common.is_thread_safe);
@@ -354,20 +654,6 @@ static ERTS_INLINE void db_unlock(DbTable* tb, db_lock_kind_t kind)
#endif
}
-
-static ERTS_INLINE void db_meta_lock(DbTable* tb, db_lock_kind_t kind)
-{
- ASSERT(tb == meta_pid_to_tab || tb == meta_pid_to_fixed_tab);
- ASSERT(kind != LCK_WRITE);
- /* As long as we only lock for READ we don't have to lock at all. */
-}
-
-static ERTS_INLINE void db_meta_unlock(DbTable* tb, db_lock_kind_t kind)
-{
- ASSERT(tb == meta_pid_to_tab || tb == meta_pid_to_fixed_tab);
- ASSERT(kind != LCK_WRITE);
-}
-
static ERTS_INLINE
DbTable* db_get_table_aux(Process *p,
Eterm id,
@@ -375,7 +661,7 @@ DbTable* db_get_table_aux(Process *p,
db_lock_kind_t kind,
int meta_already_locked)
{
- DbTable *tb = NULL;
+ DbTable *tb;
erts_smp_rwmtx_t *mtl = NULL;
/*
@@ -385,23 +671,7 @@ DbTable* db_get_table_aux(Process *p,
*/
ASSERT(erts_get_scheduler_data());
- if (is_small(id)) {
- Uint slot = unsigned_val(id) & meta_main_tab_slot_mask;
- if (!meta_already_locked) {
- mtl = get_meta_main_tab_lock(slot);
- erts_smp_rwmtx_rlock(mtl);
- }
-#if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_CHECK)
- else {
- erts_smp_rwmtx_t *test_mtl = get_meta_main_tab_lock(slot);
- ERTS_SMP_LC_ASSERT(erts_lc_rwmtx_is_rlocked(test_mtl)
- || erts_lc_rwmtx_is_rwlocked(test_mtl));
- }
-#endif
- if (slot < db_max_tabs && IS_SLOT_ALIVE(slot))
- tb = meta_main_tab[slot].u.tb;
- }
- else if (is_atom(id)) {
+ if (is_atom(id)) {
struct meta_name_tab_entry* bucket = meta_name_tab_bucket(id,&mtl);
if (!meta_already_locked)
erts_smp_rwmtx_rlock(mtl);
@@ -410,7 +680,7 @@ DbTable* db_get_table_aux(Process *p,
|| erts_lc_rwmtx_is_rwlocked(mtl));
mtl = NULL;
}
-
+ tb = NULL;
if (bucket->pu.tb != NULL) {
if (is_atom(bucket->u.name_atom)) { /* single */
if (bucket->u.name_atom == id)
@@ -428,11 +698,13 @@ DbTable* db_get_table_aux(Process *p,
}
}
}
+ else
+ tb = tid2tab(id);
+
if (tb) {
db_lock(tb, kind);
- if (tb->common.id != id
- || ((tb->common.status & what) == 0
- && p->common.id != tb->common.owner)) {
+ if ((tb->common.status & what) == 0
+ && p->common.id != tb->common.owner) {
db_unlock(tb, kind);
tb = NULL;
}
@@ -451,18 +723,6 @@ DbTable* db_get_table(Process *p,
return db_get_table_aux(p, id, what, kind, 0);
}
-/* Requires meta_main_tab_locks[slot] locked.
-*/
-static ERTS_INLINE void free_slot(int slot)
-{
- ASSERT(!IS_SLOT_FREE(slot));
- erts_smp_spin_lock(&meta_main_tab_main_lock);
- SET_NEXT_FREE_SLOT(slot,meta_main_tab_first_free);
- meta_main_tab_first_free = slot;
- meta_main_tab_cnt--;
- erts_smp_spin_unlock(&meta_main_tab_main_lock);
-}
-
static int insert_named_tab(Eterm name_atom, DbTable* tb, int have_lock)
{
int ret = 0;
@@ -527,9 +787,10 @@ static int remove_named_tab(DbTable *tb, int have_lock)
{
int ret = 0;
erts_smp_rwmtx_t* rwlock;
- Eterm name_atom = tb->common.id;
+ Eterm name_atom = tb->common.the_name;
struct meta_name_tab_entry* bucket = meta_name_tab_bucket(name_atom,
&rwlock);
+ ASSERT(is_table_named(tb));
#ifdef ERTS_SMP
if (!have_lock && erts_smp_rwmtx_tryrwlock(rwlock) == EBUSY) {
db_unlock(tb, LCK_WRITE);
@@ -600,11 +861,11 @@ done:
*/
static ERTS_INLINE void local_fix_table(DbTable* tb)
{
- erts_smp_refc_inc(&tb->common.ref, 1);
+ erts_smp_refc_inc(&tb->common.fix_count, 1);
}
static ERTS_INLINE void local_unfix_table(DbTable* tb)
{
- if (erts_smp_refc_dectest(&tb->common.ref, 0) == 0) {
+ if (erts_smp_refc_dectest(&tb->common.fix_count, 0) == 0) {
ASSERT(IS_HASH_TABLE(tb->common.status));
db_unfix_table_hash(&(tb->hash));
}
@@ -1244,6 +1505,7 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
{
DbTable* tb;
Eterm ret;
+ Eterm old_name;
erts_smp_rwmtx_t *lck1, *lck2;
#ifdef HARDDEBUG
@@ -1260,12 +1522,10 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
(void) meta_name_tab_bucket(BIF_ARG_2, &lck1);
- if (is_small(BIF_ARG_1)) {
- Uint slot = unsigned_val(BIF_ARG_1) & meta_main_tab_slot_mask;
- lck2 = get_meta_main_tab_lock(slot);
- }
- else if (is_atom(BIF_ARG_1)) {
- (void) meta_name_tab_bucket(BIF_ARG_1, &lck2);
+ if (is_atom(BIF_ARG_1)) {
+ old_name = BIF_ARG_1;
+ named_tab:
+ (void) meta_name_tab_bucket(old_name, &lck2);
if (lck1 == lck2)
lck2 = NULL;
else if (lck1 > lck2) {
@@ -1275,7 +1535,16 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
}
}
else {
- BIF_ERROR(BIF_P, BADARG);
+ tb = tid2tab(BIF_ARG_1);
+ if (!tb)
+ BIF_ERROR(BIF_P, BADARG);
+ else {
+ if (is_table_named(tb)) {
+ old_name = tb->common.the_name;
+ goto named_tab;
+ }
+ lck2 = NULL;
+ }
}
erts_smp_rwmtx_rwlock(lck1);
@@ -1286,21 +1555,19 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
if (!tb)
goto badarg;
- if (is_not_atom(tb->common.id)) { /* Not a named table */
- tb->common.the_name = BIF_ARG_2;
- goto done;
- }
+ if (is_table_named(tb)) {
+ if (!insert_named_tab(BIF_ARG_2, tb, 1))
+ goto badarg;
- if (!insert_named_tab(BIF_ARG_2, tb, 1))
- goto badarg;
-
- if (!remove_named_tab(tb, 1))
- erts_exit(ERTS_ERROR_EXIT,"Could not find named tab %s", tb->common.id);
-
- tb->common.id = tb->common.the_name = BIF_ARG_2;
+ if (!remove_named_tab(tb, 1))
+ erts_exit(ERTS_ERROR_EXIT,"Could not find named tab %s", tb->common.the_name);
+ ret = BIF_ARG_2;
+ }
+ else { /* Not a named table */
+ ret = BIF_ARG_1;
+ }
+ tb->common.the_name = BIF_ARG_2;
- done:
- ret = tb->common.id;
db_unlock(tb, LCK_WRITE);
erts_smp_rwmtx_rwunlock(lck1);
if (lck2)
@@ -1324,7 +1591,6 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
BIF_RETTYPE ets_new_2(BIF_ALIST_2)
{
DbTable* tb = NULL;
- int slot;
Eterm list;
Eterm val;
Eterm ret;
@@ -1339,9 +1605,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
#ifdef DEBUG
int cret;
#endif
- DeclareTmpHeap(meta_tuple,3,BIF_P);
DbTableMethod* meth;
- erts_smp_rwmtx_t *mmtl;
if (is_not_atom(BIF_ARG_1)) {
BIF_ERROR(BIF_P, BADARG);
@@ -1350,7 +1614,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
BIF_ERROR(BIF_P, BADARG);
}
- status = DB_NORMAL | DB_SET | DB_PROTECTED;
+ status = DB_SET | DB_PROTECTED;
keypos = 1;
is_named = 0;
#ifdef ERTS_SMP
@@ -1433,6 +1697,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
}
else if (val == am_named_table) {
is_named = 1;
+ status |= DB_NAMED_TABLE;
}
else if (val == am_compressed) {
is_compressed = 1;
@@ -1487,7 +1752,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
tb->common.type = status & ERTS_ETS_TABLE_TYPES;
/* Note, 'type' is *read only* from now on... */
#endif
- erts_smp_refc_init(&tb->common.ref, 0);
+ erts_smp_refc_init(&tb->common.fix_count, 0);
db_init_lock(tb, status & (DB_FINE_LOCKED|DB_FREQ_READ),
"db_tab", "db_tab_fix");
tb->common.keypos = keypos;
@@ -1496,7 +1761,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
erts_smp_atomic_init_nob(&tb->common.nitems, 0);
- tb->common.fixations = NULL;
+ tb->common.fixing_procs = NULL;
tb->common.compress = is_compressed;
#ifdef DEBUG
@@ -1505,87 +1770,36 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
meth->db_create(BIF_P, tb);
ASSERT(cret == DB_ERROR_NONE);
- erts_smp_spin_lock(&meta_main_tab_main_lock);
-
- if (meta_main_tab_cnt >= db_max_tabs) {
- erts_smp_spin_unlock(&meta_main_tab_main_lock);
- erts_send_error_to_logger_str(BIF_P->group_leader,
- "** Too many db tables **\n");
- free_heir_data(tb);
- tb->common.meth->db_free_table(tb);
- free_dbtable((void *) tb);
- BIF_ERROR(BIF_P, SYSTEM_LIMIT);
- }
-
- slot = meta_main_tab_first_free;
- ASSERT(slot>=0 && slot<db_max_tabs);
- meta_main_tab_first_free = GET_NEXT_FREE_SLOT(slot);
- meta_main_tab_cnt++;
- if (slot >= meta_main_tab_top) {
- ASSERT(slot == meta_main_tab_top);
- meta_main_tab_top = slot + 1;
- }
-
- if (is_named) {
- ret = BIF_ARG_1;
- }
- else {
- ret = make_small(slot | meta_main_tab_seq_cnt);
- meta_main_tab_seq_cnt += meta_main_tab_seq_incr;
- ASSERT((unsigned_val(ret) & meta_main_tab_slot_mask) == slot);
- }
- erts_smp_spin_unlock(&meta_main_tab_main_lock);
+ make_btid(tb);
- tb->common.id = ret;
- tb->common.slot = slot; /* store slot for erase */
+ if (is_named)
+ ret = BIF_ARG_1;
+ else
+ ret = make_tid(BIF_P, tb);
- mmtl = get_meta_main_tab_lock(slot);
- erts_smp_rwmtx_rwlock(mmtl);
- meta_main_tab[slot].u.tb = tb;
- ASSERT(IS_SLOT_ALIVE(slot));
- erts_smp_rwmtx_rwunlock(mmtl);
+ save_sched_table(BIF_P, tb);
if (is_named && !insert_named_tab(BIF_ARG_1, tb, 0)) {
- mmtl = get_meta_main_tab_lock(slot);
- erts_smp_rwmtx_rwlock(mmtl);
- free_slot(slot);
- erts_smp_rwmtx_rwunlock(mmtl);
+ tid_clear(BIF_P, tb);
db_lock(tb,LCK_WRITE);
free_heir_data(tb);
tb->common.meth->db_free_table(tb);
- schedule_free_dbtable(tb);
db_unlock(tb,LCK_WRITE);
+ table_dec_refc(tb, 0);
BIF_ERROR(BIF_P, BADARG);
}
BIF_P->flags |= F_USING_DB; /* So we can remove tb if p dies */
+ save_owned_table(BIF_P, tb);
#ifdef HARDDEBUG
erts_fprintf(stderr,
"ets:new(%T,%T)=%T; Process: %T, initial: %T:%T/%bpu\n",
BIF_ARG_1, BIF_ARG_2, ret, BIF_P->common.id,
BIF_P->u.initial[0], BIF_P->u.initial[1], BIF_P->u.initial[2]);
- erts_fprintf(stderr, "ets: new: meta_pid_to_tab common.memory_size = %ld\n",
- erts_smp_atomic_read_nob(&meta_pid_to_tab->common.memory_size));
- erts_fprintf(stderr, "ets: new: meta_pid_to_fixed_tab common.memory_size = %ld\n",
- erts_smp_atomic_read_nob(&meta_pid_to_fixed_tab->common.memory_size));
#endif
- UseTmpHeap(3,BIF_P);
-
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- if (db_put_hash(meta_pid_to_tab,
- TUPLE2(meta_tuple,
- BIF_P->common.id,
- make_small(slot)),
- 0) != DB_ERROR_NONE) {
- erts_exit(ERTS_ERROR_EXIT,"Could not update ets metadata.");
- }
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
-
- UnUseTmpHeap(3,BIF_P);
-
BIF_RET(ret);
}
@@ -1690,9 +1904,9 @@ BIF_RETTYPE ets_lookup_element_3(BIF_ALIST_3)
*/
BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
{
- int trap;
+ SWord initial_reds = ERTS_BIF_REDS_LEFT(BIF_P);
+ SWord reds = initial_reds;
DbTable* tb;
- erts_smp_rwmtx_t *mmtl;
#ifdef HARDDEBUG
erts_fprintf(stderr,
@@ -1715,7 +1929,6 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
tb->common.status |= DB_DELETE;
if (tb->common.owner != BIF_P->common.id) {
- DeclareTmpHeap(meta_tuple,3,BIF_P);
/*
* The table is being deleted by a process other than its owner.
@@ -1723,50 +1936,33 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
* current process will be killed (e.g. by an EXIT signal), we will
* now transfer the ownership to the current process.
*/
- UseTmpHeap(3,BIF_P);
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_tab, tb->common.owner,
- make_small(tb->common.slot));
-
- BIF_P->flags |= F_USING_DB;
- tb->common.owner = BIF_P->common.id;
-
- db_put_hash(meta_pid_to_tab,
- TUPLE2(meta_tuple,
- BIF_P->common.id,
- make_small(tb->common.slot)),
- 0);
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
- UnUseTmpHeap(3,BIF_P);
- }
- mmtl = get_meta_main_tab_lock(tb->common.slot);
-#ifdef ERTS_SMP
- if (erts_smp_rwmtx_tryrwlock(mmtl) == EBUSY) {
- /*
- * We keep our increased refc over this op in order to
- * prevent the table from disapearing.
- */
- db_unlock(tb, LCK_WRITE);
- erts_smp_rwmtx_rwlock(mmtl);
- db_lock(tb, LCK_WRITE);
+ Process *rp = erts_proc_lookup_raw(tb->common.owner);
+ /*
+ * Process 'rp' might be exiting, but our table lock prevents it
+ * from terminating as it cannot complete erts_db_process_exiting().
+ */
+ ASSERT(!(ERTS_PSFLG_FREE & erts_smp_atomic32_read_nob(&rp->state)));
+
+ delete_owned_table(rp, tb);
+ BIF_P->flags |= F_USING_DB;
+ tb->common.owner = BIF_P->common.id;
+ save_owned_table(BIF_P, tb);
}
-#endif
- /* We must keep the slot, to be found by db_proc_dead() if process dies */
- MARK_SLOT_DEAD(tb->common.slot);
- erts_smp_rwmtx_rwunlock(mmtl);
- if (is_atom(tb->common.id))
+
+ tid_clear(BIF_P, tb);
+
+ if (is_table_named(tb))
remove_named_tab(tb, 0);
/* disable inheritance */
free_heir_data(tb);
tb->common.heir = am_none;
- free_fixations_locked(tb);
-
- trap = free_table_cont(BIF_P, tb, 1, 1);
+ reds -= free_fixations_locked(BIF_P, tb);
db_unlock(tb, LCK_WRITE);
- if (trap) {
+
+ if (free_table_continue(BIF_P, tb, reds) < 0) {
/*
* Package the DbTable* pointer into a bignum so that it can be safely
* passed through a trap. We used to pass the DbTable* pointer directly
@@ -1776,9 +1972,11 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
Eterm *hp = HAlloc(BIF_P, 2);
hp[0] = make_pos_bignum_header(1);
hp[1] = (Eterm) tb;
+ BUMP_ALL_REDS(BIF_P);
BIF_TRAP1(&ets_delete_continue_exp, BIF_P, make_big(hp));
}
else {
+ BUMP_REDS(BIF_P, (initial_reds - reds));
BIF_RET(am_true);
}
}
@@ -1790,7 +1988,6 @@ BIF_RETTYPE ets_give_away_3(BIF_ALIST_3)
{
Process* to_proc = NULL;
ErtsProcLocks to_locks = ERTS_PROC_LOCK_MAIN;
- DeclareTmpHeap(buf,5,BIF_P);
Eterm to_pid = BIF_ARG_2;
Eterm from_pid;
DbTable* tb = NULL;
@@ -1812,26 +2009,14 @@ BIF_RETTYPE ets_give_away_3(BIF_ALIST_3)
goto badarg; /* or should we be idempotent? return false maybe */
}
- UseTmpHeap(5,BIF_P);
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_tab, tb->common.owner,
- make_small(tb->common.slot));
-
+ delete_owned_table(BIF_P, tb);
to_proc->flags |= F_USING_DB;
tb->common.owner = to_pid;
-
- db_put_hash(meta_pid_to_tab,
- TUPLE2(buf,to_pid,make_small(tb->common.slot)),
- 0);
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
+ save_owned_table(to_proc, tb);
db_unlock(tb,LCK_WRITE);
- erts_send_message(BIF_P, to_proc, &to_locks,
- TUPLE4(buf, am_ETS_TRANSFER,
- tb->common.id,
- from_pid,
- BIF_ARG_3),
- 0);
+ send_ets_transfer_message(BIF_P, to_proc, &to_locks,
+ tb, BIF_ARG_3);
erts_smp_proc_unlock(to_proc, to_locks);
UnUseTmpHeap(5,BIF_P);
BIF_RET(am_true);
@@ -2074,7 +2259,7 @@ BIF_RETTYPE ets_select_delete_2(BIF_ALIST_2)
if (safety == ITER_UNSAFE) {
local_fix_table(tb);
}
- cret = tb->common.meth->db_select_delete(BIF_P, tb, BIF_ARG_2, &ret);
+ cret = tb->common.meth->db_select_delete(BIF_P, tb, BIF_ARG_1, BIF_ARG_2, &ret);
if (DID_TRAP(BIF_P,ret) && safety != ITER_SAFE) {
fix_table_locked(BIF_P,tb);
@@ -2101,46 +2286,254 @@ BIF_RETTYPE ets_select_delete_2(BIF_ALIST_2)
return result;
}
-/*
-** Return a list of tables on this node
-*/
-BIF_RETTYPE ets_all_0(BIF_ALIST_0)
+/*
+ * ets:all/0
+ *
+ * ets:all() calls ets:internal_request_all/0 which
+ * requests information about all tables from
+ * each scheduler thread. Each scheduler replies
+ * to the calling process with information about
+ * existing tables created on that specific scheduler.
+ */
+
+struct ErtsEtsAllReq_ {
+ erts_smp_atomic32_t refc;
+ Process *proc;
+ ErtsOIRefStorage ref;
+ ErtsEtsAllReqList list[1]; /* one per scheduler */
+};
+
+#define ERTS_ETS_ALL_REQ_SIZE \
+ (sizeof(ErtsEtsAllReq) \
+ + (sizeof(ErtsEtsAllReqList) \
+ * (erts_no_schedulers - 1)))
+
+typedef struct {
+ ErtsEtsAllReq *ongoing;
+ ErlHeapFragment *hfrag;
+ DbTable *tab;
+ ErtsEtsAllReq *queue;
+} ErtsEtsAllData;
+
+/* Tables handled before yielding */
+#define ERTS_ETS_ALL_TB_YCNT 200
+/*
+ * Min yield count required before starting
+ * an operation that will require yield.
+ */
+#define ERTS_ETS_ALL_TB_YCNT_START 10
+
+#ifdef DEBUG
+/* Test yielding... */
+#undef ERTS_ETS_ALL_TB_YCNT
+#undef ERTS_ETS_ALL_TB_YCNT_START
+#define ERTS_ETS_ALL_TB_YCNT 10
+#define ERTS_ETS_ALL_TB_YCNT_START 1
+#endif
+
+static int
+ets_all_reply(ErtsSchedulerData *esdp, ErtsEtsAllReq **reqpp,
+ ErlHeapFragment **hfragpp, DbTable **tablepp,
+ int *yield_count_p)
{
- DbTable* tb;
- Eterm previous;
- int i;
- Eterm* hp;
- Eterm* hendp;
- int t_tabs_cnt;
- int t_top;
-
- erts_smp_spin_lock(&meta_main_tab_main_lock);
- t_tabs_cnt = meta_main_tab_cnt;
- t_top = meta_main_tab_top;
- erts_smp_spin_unlock(&meta_main_tab_main_lock);
-
- hp = HAlloc(BIF_P, 2*t_tabs_cnt);
- hendp = hp + 2*t_tabs_cnt;
-
- previous = NIL;
- for(i = 0; i < t_top; i++) {
- erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(i);
- erts_smp_rwmtx_rlock(mmtl);
- if (IS_SLOT_ALIVE(i)) {
- if (hp == hendp) {
- /* Racing table creator, grab some more heap space */
- t_tabs_cnt = 10;
- hp = HAlloc(BIF_P, 2*t_tabs_cnt);
- hendp = hp + 2*t_tabs_cnt;
- }
- tb = meta_main_tab[i].u.tb;
- previous = CONS(hp, tb->common.id, previous);
- hp += 2;
- }
- erts_smp_rwmtx_runlock(mmtl);
+ ErtsEtsAllReq *reqp = *reqpp;
+ ErlHeapFragment *hfragp = *hfragpp;
+ int ycount = *yield_count_p;
+ DbTable *tb, *first;
+ Uint sz;
+ Eterm list, msg, ref, *hp;
+ ErlOffHeap *ohp;
+ ErtsMessage *mp;
+
+ /*
+ * - save_sched_table() inserts at end of circular list.
+ *
+ * - This function scans from the end so we know that
+ * the amount of tables to scan wont grow even if we
+ * yield.
+ *
+ * - remove_sched_table() updates the table we yielded
+ * on if it removes it.
+ */
+
+ if (hfragp) {
+ /* Restart of a yielded operation... */
+ ASSERT(hfragp->used_size < hfragp->alloc_size);
+ ohp = &hfragp->off_heap;
+ hp = &hfragp->mem[hfragp->used_size];
+ list = *hp;
+ hfragp->used_size = hfragp->alloc_size;
+ first = esdp->ets_tables.clist;
+ tb = *tablepp;
+ }
+ else {
+ /* A new operation... */
+ ASSERT(!*tablepp);
+
+ /* Max heap size needed... */
+ sz = esdp->ets_tables.count;
+ sz *= ERTS_MAGIC_REF_THING_SIZE + 2;
+ sz += 3 + ERTS_REF_THING_SIZE;
+ hfragp = new_message_buffer(sz);
+
+ hp = &hfragp->mem[0];
+ ohp = &hfragp->off_heap;
+ list = NIL;
+ first = esdp->ets_tables.clist;
+ tb = first ? first->common.all.prev : NULL;
+ }
+
+ if (tb) {
+ while (1) {
+ if (is_table_alive(tb)) {
+ Eterm tid;
+ if (is_table_named(tb))
+ tid = tb->common.the_name;
+ else
+ tid = erts_mk_magic_ref(&hp, ohp, tb->common.btid);
+ list = CONS(hp, tid, list);
+ hp += 2;
+ }
+
+ if (tb == first)
+ break;
+
+ tb = tb->common.all.prev;
+
+ if (--ycount <= 0) {
+ sz = hp - &hfragp->mem[0];
+ ASSERT(hfragp->alloc_size > sz + 1);
+ *hp = list;
+ hfragp->used_size = sz;
+ *hfragpp = hfragp;
+ *reqpp = reqp;
+ *tablepp = tb;
+ *yield_count_p = 0;
+ return 1; /* Yield! */
+ }
+ }
}
- HRelease(BIF_P, hendp, hp);
- BIF_RET(previous);
+
+ ref = erts_oiref_storage_make_ref(&reqp->ref, &hp);
+ msg = TUPLE2(hp, ref, list);
+ hp += 3;
+
+ sz = hp - &hfragp->mem[0];
+ ASSERT(sz <= hfragp->alloc_size);
+
+ hfragp = erts_resize_message_buffer(hfragp, sz, &msg, 1);
+
+ mp = erts_alloc_message(0, NULL);
+ mp->data.heap_frag = hfragp;
+
+ erts_queue_message(reqp->proc, 0, mp, msg, am_system);
+
+ erts_proc_dec_refc(reqp->proc);
+
+ if (erts_smp_atomic32_dec_read_nob(&reqp->refc) == 0)
+ erts_free(ERTS_ALC_T_ETS_ALL_REQ, reqp);
+
+ *reqpp = NULL;
+ *hfragpp = NULL;
+ *tablepp = NULL;
+ *yield_count_p = ycount;
+
+ return 0;
+}
+
+int
+erts_handle_yielded_ets_all_request(ErtsSchedulerData *esdp,
+ ErtsEtsAllYieldData *eaydp)
+{
+ int ix = (int) esdp->no - 1;
+ int yc = ERTS_ETS_ALL_TB_YCNT;
+
+ while (1) {
+ if (!eaydp->ongoing) {
+ ErtsEtsAllReq *ongoing;
+
+ if (!eaydp->queue)
+ return 0; /* All work completed! */
+
+ if (yc < ERTS_ETS_ALL_TB_YCNT_START && yc > esdp->ets_tables.count)
+ return 1; /* Yield! */
+
+ eaydp->ongoing = ongoing = eaydp->queue;
+ if (ongoing->list[ix].next == ongoing)
+ eaydp->queue = NULL;
+ else {
+ ongoing->list[ix].next->list[ix].prev = ongoing->list[ix].prev;
+ ongoing->list[ix].prev->list[ix].next = ongoing->list[ix].next;
+ eaydp->queue = ongoing->list[ix].next;
+ }
+ ASSERT(!eaydp->hfrag);
+ ASSERT(!eaydp->tab);
+ }
+
+ if (ets_all_reply(esdp, &eaydp->ongoing, &eaydp->hfrag, &eaydp->tab, &yc))
+ return 1; /* Yield! */
+ }
+}
+
+static void
+handle_ets_all_request(void *vreq)
+{
+ ErtsSchedulerData *esdp = erts_get_scheduler_data();
+ ErtsEtsAllYieldData *eayp = ERTS_SCHED_AUX_YIELD_DATA(esdp, ets_all);
+ ErtsEtsAllReq *req = (ErtsEtsAllReq *) vreq;
+
+ if (!eayp->ongoing && !eayp->queue) {
+ /* No ets:all() operations ongoing... */
+ ErlHeapFragment *hf = NULL;
+ DbTable *tb = NULL;
+ int yc = ERTS_ETS_ALL_TB_YCNT;
+ if (ets_all_reply(esdp, &req, &hf, &tb, &yc)) {
+ /* Yielded... */
+ ASSERT(hf);
+ eayp->ongoing = req;
+ eayp->hfrag = hf;
+ eayp->tab = tb;
+ erts_notify_new_aux_yield_work(esdp);
+ }
+ }
+ else {
+ /* Ongoing ets:all() operations; queue up this request... */
+ int ix = (int) esdp->no - 1;
+ if (!eayp->queue) {
+ req->list[ix].next = req;
+ req->list[ix].prev = req;
+ eayp->queue = req;
+ }
+ else {
+ req->list[ix].next = eayp->queue;
+ req->list[ix].prev = eayp->queue->list[ix].prev;
+ eayp->queue->list[ix].prev = req;
+ req->list[ix].prev->list[ix].next = req;
+ }
+ }
+}
+
+BIF_RETTYPE ets_internal_request_all_0(BIF_ALIST_0)
+{
+ Eterm ref = erts_make_ref(BIF_P);
+ ErtsEtsAllReq *req = erts_alloc(ERTS_ALC_T_ETS_ALL_REQ,
+ ERTS_ETS_ALL_REQ_SIZE);
+ erts_smp_atomic32_init_nob(&req->refc,
+ (erts_aint32_t) erts_no_schedulers);
+ erts_oiref_storage_save(&req->ref, ref);
+ req->proc = BIF_P;
+ erts_proc_add_refc(BIF_P, (Sint) erts_no_schedulers);
+
+#ifdef ERTS_SMP
+ if (erts_no_schedulers > 1)
+ erts_schedule_multi_misc_aux_work(1,
+ erts_no_schedulers,
+ handle_ets_all_request,
+ (void *) req);
+#endif
+
+ handle_ets_all_request((void *) req);
+ BIF_RET(ref);
}
@@ -2245,7 +2638,7 @@ ets_select3(Process* p, Eterm arg1, Eterm arg2, Eterm arg3)
if (safety == ITER_UNSAFE) {
local_fix_table(tb);
}
- cret = tb->common.meth->db_select_chunk(p, tb,
+ cret = tb->common.meth->db_select_chunk(p, tb, arg1,
arg2, chunk_size,
0 /* not reversed */,
&ret);
@@ -2414,8 +2807,7 @@ ets_select2(Process* p, Eterm arg1, Eterm arg2)
local_fix_table(tb);
}
- cret = tb->common.meth->db_select(p, tb, arg2,
- 0, &ret);
+ cret = tb->common.meth->db_select(p, tb, arg1, arg2, 0, &ret);
if (DID_TRAP(p,ret) && safety != ITER_SAFE) {
fix_table_locked(p, tb);
@@ -2506,7 +2898,7 @@ BIF_RETTYPE ets_select_count_2(BIF_ALIST_2)
if (safety == ITER_UNSAFE) {
local_fix_table(tb);
}
- cret = tb->common.meth->db_select_count(BIF_P,tb,BIF_ARG_2, &ret);
+ cret = tb->common.meth->db_select_count(BIF_P,tb, BIF_ARG_1, BIF_ARG_2, &ret);
if (DID_TRAP(BIF_P,ret) && safety != ITER_SAFE) {
fix_table_locked(BIF_P, tb);
@@ -2532,6 +2924,103 @@ BIF_RETTYPE ets_select_count_2(BIF_ALIST_2)
return result;
}
+/*
+ ** This is for trapping, cannot be called directly.
+ */
+static BIF_RETTYPE ets_select_replace_1(BIF_ALIST_1)
+{
+ Process *p = BIF_P;
+ Eterm a1 = BIF_ARG_1;
+ BIF_RETTYPE result;
+ DbTable* tb;
+ int cret;
+ Eterm ret;
+ Eterm *tptr;
+ db_lock_kind_t kind = LCK_WRITE_REC;
+
+ CHECK_TABLES();
+ ASSERT(is_tuple(a1));
+ tptr = tuple_val(a1);
+ ASSERT(arityval(*tptr) >= 1);
+
+ if ((tb = db_get_table(p, tptr[1], DB_WRITE, kind)) == NULL) {
+ BIF_ERROR(p,BADARG);
+ }
+
+ cret = tb->common.meth->db_select_replace_continue(p,tb,a1,&ret);
+
+ if(!DID_TRAP(p,ret) && ITERATION_SAFETY(p,tb) != ITER_SAFE) {
+ unfix_table_locked(p, tb, &kind);
+ }
+
+ db_unlock(tb, kind);
+
+ switch (cret) {
+ case DB_ERROR_NONE:
+ ERTS_BIF_PREP_RET(result, ret);
+ break;
+ default:
+ ERTS_BIF_PREP_ERROR(result, p, BADARG);
+ break;
+ }
+ erts_match_set_release_result(p);
+
+ return result;
+}
+
+
+BIF_RETTYPE ets_select_replace_2(BIF_ALIST_2)
+{
+ BIF_RETTYPE result;
+ DbTable* tb;
+ int cret;
+ Eterm ret;
+ enum DbIterSafety safety;
+
+ CHECK_TABLES();
+
+ if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC)) == NULL) {
+ BIF_ERROR(BIF_P, BADARG);
+ }
+
+ if (tb->common.status & DB_BAG) {
+ /* Bag implementation presented both semantic consistency
+ and performance issues */
+ db_unlock(tb, LCK_WRITE_REC);
+ BIF_ERROR(BIF_P, BADARG);
+ }
+
+ safety = ITERATION_SAFETY(BIF_P,tb);
+ if (safety == ITER_UNSAFE) {
+ local_fix_table(tb);
+ }
+ cret = tb->common.meth->db_select_replace(BIF_P, tb, BIF_ARG_1, BIF_ARG_2, &ret);
+
+ if (DID_TRAP(BIF_P,ret) && safety != ITER_SAFE) {
+ fix_table_locked(BIF_P,tb);
+ }
+ if (safety == ITER_UNSAFE) {
+ local_unfix_table(tb);
+ }
+ db_unlock(tb, LCK_WRITE_REC);
+
+ switch (cret) {
+ case DB_ERROR_NONE:
+ ERTS_BIF_PREP_RET(result, ret);
+ break;
+ case DB_ERROR_SYSRES:
+ ERTS_BIF_PREP_ERROR(result, BIF_P, SYSTEM_LIMIT);
+ break;
+ default:
+ ERTS_BIF_PREP_ERROR(result, BIF_P, BADARG);
+ break;
+ }
+
+ erts_match_set_release_result(BIF_P);
+
+ return result;
+}
+
BIF_RETTYPE ets_select_reverse_3(BIF_ALIST_3)
{
@@ -2560,7 +3049,7 @@ BIF_RETTYPE ets_select_reverse_3(BIF_ALIST_3)
if (safety == ITER_UNSAFE) {
local_fix_table(tb);
}
- cret = tb->common.meth->db_select_chunk(BIF_P,tb,
+ cret = tb->common.meth->db_select_chunk(BIF_P,tb, BIF_ARG_1,
BIF_ARG_2, chunk_size,
1 /* reversed */, &ret);
if (DID_TRAP(BIF_P,ret) && safety != ITER_SAFE) {
@@ -2610,7 +3099,7 @@ BIF_RETTYPE ets_select_reverse_2(BIF_ALIST_2)
if (safety == ITER_UNSAFE) {
local_fix_table(tb);
}
- cret = tb->common.meth->db_select(BIF_P,tb,BIF_ARG_2,
+ cret = tb->common.meth->db_select(BIF_P,tb, BIF_ARG_1, BIF_ARG_2,
1 /*reversed*/, &ret);
if (DID_TRAP(BIF_P,ret) && safety != ITER_SAFE) {
@@ -2701,7 +3190,7 @@ BIF_RETTYPE ets_info_1(BIF_ALIST_1)
*/
if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_INFO, LCK_READ)) == NULL) {
- if (is_atom(BIF_ARG_1) || is_small(BIF_ARG_1)) {
+ if (is_atom(BIF_ARG_1) || is_ref(BIF_ARG_1)) {
BIF_RET(am_undefined);
}
BIF_ERROR(BIF_P, BADARG);
@@ -2763,7 +3252,7 @@ BIF_RETTYPE ets_info_2(BIF_ALIST_2)
Eterm ret = THE_NON_VALUE;
if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_INFO, LCK_READ)) == NULL) {
- if (is_atom(BIF_ARG_1) || is_small(BIF_ARG_1)) {
+ if (is_atom(BIF_ARG_1) || is_ref(BIF_ARG_1)) {
BIF_RET(am_undefined);
}
BIF_ERROR(BIF_P, BADARG);
@@ -2853,7 +3342,6 @@ int erts_ets_rwmtx_spin_count = -1;
void init_db(ErtsDbSpinCount db_spin_count)
{
- DbTable init_tb;
int i;
Eterm *hp;
unsigned bits;
@@ -2902,16 +3390,6 @@ void init_db(ErtsDbSpinCount db_spin_count)
if (erts_ets_rwmtx_spin_count >= 0)
rwmtx_opt.main_spincount = erts_ets_rwmtx_spin_count;
- meta_main_tab_locks =
- erts_alloc_permanent_cache_aligned(ERTS_ALC_T_DB_TABLES,
- sizeof(erts_meta_main_tab_lock_t)
- * ERTS_META_MAIN_TAB_LOCK_TAB_SIZE);
-
- for (i = 0; i < ERTS_META_MAIN_TAB_LOCK_TAB_SIZE; i++) {
- erts_smp_rwmtx_init_opt_x(&meta_main_tab_locks[i].rwmtx, &rwmtx_opt,
- "meta_main_tab_slot", make_small(i));
- }
- erts_smp_spinlock_init(&meta_main_tab_main_lock, "meta_main_tab_main");
for (i=0; i<META_NAME_TAB_LOCK_CNT; i++) {
erts_smp_rwmtx_init_opt_x(&meta_name_tab_rwlocks[i].lck, &rwmtx_opt,
"meta_name_tab", make_small(i));
@@ -2931,20 +3409,6 @@ void init_db(ErtsDbSpinCount db_spin_count)
erts_exit(ERTS_ERROR_EXIT,"Max limit for ets tabled too high %u (max %u).",
db_max_tabs, ((Uint)1)<<SMALL_BITS);
}
- meta_main_tab_slot_mask = (((Uint)1)<<bits) - 1;
- meta_main_tab_seq_incr = (((Uint)1)<<bits);
-
- size = sizeof(*meta_main_tab)*db_max_tabs;
- meta_main_tab = erts_db_alloc_nt(ERTS_ALC_T_DB_TABLES, size);
- ERTS_ETS_MISC_MEM_ADD(size);
-
- meta_main_tab_cnt = 0;
- meta_main_tab_top = 0;
- for (i=1; i<db_max_tabs; i++) {
- SET_NEXT_FREE_SLOT(i-1,i);
- }
- SET_NEXT_FREE_SLOT(db_max_tabs-1, (Uint)-1);
- meta_main_tab_first_free = 0;
meta_name_tab_mask = (((Uint) 1)<<(bits-1)) - 1; /* At least half the size of main tab */
size = sizeof(struct meta_name_tab_entry)*(meta_name_tab_mask+1);
@@ -2959,70 +3423,6 @@ void init_db(ErtsDbSpinCount db_spin_count)
db_initialize_hash();
db_initialize_tree();
- /*TT*/
- /* Create meta table invertion. */
- erts_smp_atomic_init_nob(&init_tb.common.memory_size, 0);
- meta_pid_to_tab = (DbTable*) erts_db_alloc(ERTS_ALC_T_DB_TABLE,
- &init_tb,
- sizeof(DbTable));
- erts_smp_atomic_init_nob(&meta_pid_to_tab->common.memory_size,
- erts_smp_atomic_read_nob(&init_tb.common.memory_size));
-
- meta_pid_to_tab->common.id = NIL;
- meta_pid_to_tab->common.the_name = am_true;
- meta_pid_to_tab->common.status = (DB_NORMAL | DB_BAG | DB_PUBLIC | DB_FINE_LOCKED);
-#ifdef ERTS_SMP
- meta_pid_to_tab->common.type
- = meta_pid_to_tab->common.status & ERTS_ETS_TABLE_TYPES;
- /* Note, 'type' is *read only* from now on... */
- meta_pid_to_tab->common.is_thread_safe = 0;
-#endif
- meta_pid_to_tab->common.keypos = 1;
- meta_pid_to_tab->common.owner = NIL;
- erts_smp_atomic_init_nob(&meta_pid_to_tab->common.nitems, 0);
- meta_pid_to_tab->common.slot = -1;
- meta_pid_to_tab->common.meth = &db_hash;
- meta_pid_to_tab->common.compress = 0;
-
- erts_smp_refc_init(&meta_pid_to_tab->common.ref, 0);
- /* Neither rwlock or fixlock used
- db_init_lock(meta_pid_to_tab, "meta_pid_to_tab", "meta_pid_to_tab_FIX");*/
-
- if (db_create_hash(NULL, meta_pid_to_tab) != DB_ERROR_NONE) {
- erts_exit(ERTS_ERROR_EXIT,"Unable to create ets metadata tables.");
- }
-
- erts_smp_atomic_set_nob(&init_tb.common.memory_size, 0);
- meta_pid_to_fixed_tab = (DbTable*) erts_db_alloc(ERTS_ALC_T_DB_TABLE,
- &init_tb,
- sizeof(DbTable));
- erts_smp_atomic_init_nob(&meta_pid_to_fixed_tab->common.memory_size,
- erts_smp_atomic_read_nob(&init_tb.common.memory_size));
-
- meta_pid_to_fixed_tab->common.id = NIL;
- meta_pid_to_fixed_tab->common.the_name = am_true;
- meta_pid_to_fixed_tab->common.status = (DB_NORMAL | DB_BAG | DB_PUBLIC | DB_FINE_LOCKED);
-#ifdef ERTS_SMP
- meta_pid_to_fixed_tab->common.type
- = meta_pid_to_fixed_tab->common.status & ERTS_ETS_TABLE_TYPES;
- /* Note, 'type' is *read only* from now on... */
- meta_pid_to_fixed_tab->common.is_thread_safe = 0;
-#endif
- meta_pid_to_fixed_tab->common.keypos = 1;
- meta_pid_to_fixed_tab->common.owner = NIL;
- erts_smp_atomic_init_nob(&meta_pid_to_fixed_tab->common.nitems, 0);
- meta_pid_to_fixed_tab->common.slot = -1;
- meta_pid_to_fixed_tab->common.meth = &db_hash;
- meta_pid_to_fixed_tab->common.compress = 0;
-
- erts_smp_refc_init(&meta_pid_to_fixed_tab->common.ref, 0);
- /* Neither rwlock or fixlock used
- db_init_lock(meta_pid_to_fixed_tab, "meta_pid_to_fixed_tab", "meta_pid_to_fixed_tab_FIX");*/
-
- if (db_create_hash(NULL, meta_pid_to_fixed_tab) != DB_ERROR_NONE) {
- erts_exit(ERTS_ERROR_EXIT,"Unable to create ets metadata tables.");
- }
-
/* Non visual BIF to trap to. */
erts_init_trap_export(&ets_select_delete_continue_exp,
am_ets, am_atom_put("delete_trap",11), 1,
@@ -3034,6 +3434,11 @@ void init_db(ErtsDbSpinCount db_spin_count)
&ets_select_count_1);
/* Non visual BIF to trap to. */
+ erts_init_trap_export(&ets_select_replace_continue_exp,
+ am_ets, am_atom_put("replace_trap",11), 1,
+ &ets_select_replace_1);
+
+ /* Non visual BIF to trap to. */
erts_init_trap_export(&ets_select_continue_exp,
am_ets, am_atom_put("select_trap",11), 1,
&ets_select_trap_1);
@@ -3051,81 +3456,18 @@ void init_db(ErtsDbSpinCount db_spin_count)
ms_delete_all = CONS(hp, ms_delete_all,NIL);
}
-#define ARRAY_CHUNK 100
-
-typedef enum {
- ErtsDbProcCleanupProgressTables,
- ErtsDbProcCleanupProgressFixations,
- ErtsDbProcCleanupProgressDone,
-} ErtsDbProcCleanupProgress;
-
-typedef enum {
- ErtsDbProcCleanupOpGetTables,
- ErtsDbProcCleanupOpDeleteTables,
- ErtsDbProcCleanupOpGetFixations,
- ErtsDbProcCleanupOpDeleteFixations,
- ErtsDbProcCleanupOpDone
-} ErtsDbProcCleanupOperation;
-
-typedef struct {
- ErtsDbProcCleanupProgress progress;
- ErtsDbProcCleanupOperation op;
- struct {
- Eterm arr[ARRAY_CHUNK];
- int size;
- int ix;
- int clean_ix;
- } slots;
-} ErtsDbProcCleanupState;
-
-
-static void
-proc_exit_cleanup_tables_meta_data(Eterm pid, ErtsDbProcCleanupState *state)
+void
+erts_ets_sched_spec_data_init(ErtsSchedulerData *esdp)
{
- ASSERT(state->slots.clean_ix <= state->slots.ix);
- if (state->slots.clean_ix < state->slots.ix) {
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- if (state->slots.size < ARRAY_CHUNK
- && state->slots.ix == state->slots.size) {
- Eterm dummy;
- db_erase_hash(meta_pid_to_tab,pid,&dummy);
- }
- else {
- int ix;
- /* Need to erase each explicitly */
- for (ix = state->slots.clean_ix; ix < state->slots.ix; ix++)
- db_erase_bag_exact2(meta_pid_to_tab,
- pid,
- state->slots.arr[ix]);
- }
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
- state->slots.clean_ix = state->slots.ix;
- }
+ ErtsEtsAllYieldData *eaydp = ERTS_SCHED_AUX_YIELD_DATA(esdp, ets_all);
+ eaydp->ongoing = NULL;
+ eaydp->hfrag = NULL;
+ eaydp->tab = NULL;
+ eaydp->queue = NULL;
+ esdp->ets_tables.clist = NULL;
+ esdp->ets_tables.count = 0;
}
-static void
-proc_exit_cleanup_fixations_meta_data(Eterm pid, ErtsDbProcCleanupState *state)
-{
- ASSERT(state->slots.clean_ix <= state->slots.ix);
- if (state->slots.clean_ix < state->slots.ix) {
- db_meta_lock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- if (state->slots.size < ARRAY_CHUNK
- && state->slots.ix == state->slots.size) {
- Eterm dummy;
- db_erase_hash(meta_pid_to_fixed_tab,pid,&dummy);
- }
- else {
- int ix;
- /* Need to erase each explicitly */
- for (ix = state->slots.clean_ix; ix < state->slots.ix; ix++)
- db_erase_bag_exact2(meta_pid_to_fixed_tab,
- pid,
- state->slots.arr[ix]);
- }
- db_meta_unlock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- state->slots.clean_ix = state->slots.ix;
- }
-}
/* In: Table LCK_WRITE
** Return TRUE : ok, table not mine and NOT locked anymore.
@@ -3135,7 +3477,6 @@ static int give_away_to_heir(Process* p, DbTable* tb)
{
Process* to_proc;
ErtsProcLocks to_locks = ERTS_PROC_LOCK_MAIN;
- DeclareTmpHeap(buf,5,p);
Eterm to_pid;
UWord heir_data;
@@ -3179,19 +3520,12 @@ retry:
erts_smp_proc_unlock(to_proc, to_locks);
return 0; /* heir dead and pid reused, table still mine */
}
- UseTmpHeap(5,p);
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_tab, tb->common.owner,
- make_small(tb->common.slot));
+ delete_owned_table(p, tb);
to_proc->flags |= F_USING_DB;
tb->common.owner = to_pid;
-
- db_put_hash(meta_pid_to_tab,
- TUPLE2(buf,to_pid,make_small(tb->common.slot)),
- 0);
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
- UnUseTmpHeap(5,p);
+ save_owned_table(to_proc, tb);
+
db_unlock(tb,LCK_WRITE);
heir_data = tb->common.heir_data;
if (!is_immed(heir_data)) {
@@ -3199,17 +3533,100 @@ retry:
ASSERT(arityval(*tpv) == 1);
heir_data = tpv[1];
}
- erts_send_message(p, to_proc, &to_locks,
- TUPLE4(buf,
- am_ETS_TRANSFER,
- tb->common.id,
- p->common.id,
- heir_data),
- 0);
+ send_ets_transfer_message(p, to_proc, &to_locks, tb, heir_data);
erts_smp_proc_unlock(to_proc, to_locks);
return !0;
}
+static void
+send_ets_transfer_message(Process *c_p, Process *proc,
+ ErtsProcLocks *locks,
+ DbTable *tb, Eterm heir_data)
+{
+ Uint hsz, hd_sz;
+ ErtsMessage *mp;
+ Eterm *hp;
+ ErlOffHeap *ohp;
+ Eterm tid, hd_copy, msg, sender;
+
+ hsz = 5;
+ if (!is_table_named(tb))
+ hsz += ERTS_MAGIC_REF_THING_SIZE;
+ if (is_immed(heir_data))
+ hd_sz = 0;
+ else {
+ hd_sz = size_object(heir_data);
+ hsz += hd_sz;
+ }
+
+ mp = erts_alloc_message_heap(proc, locks, hsz, &hp, &ohp);
+ if (is_table_named(tb))
+ tid = tb->common.the_name;
+ else
+ tid = erts_mk_magic_ref(&hp, ohp, tb->common.btid);
+ if (!hd_sz)
+ hd_copy = heir_data;
+ else
+ hd_copy = copy_struct(heir_data, hd_sz, &hp, ohp);
+ sender = c_p->common.id;
+ msg = TUPLE4(hp, am_ETS_TRANSFER, tid, sender, hd_copy);
+ erts_queue_message(proc, *locks, mp, msg, sender);
+}
+
+
+/* Auto-release fixation from exiting process */
+static SWord proc_cleanup_fixed_table(Process* p, DbFixation* fix)
+{
+ DbTable* tb = btid2tab(fix->tabs.btid);
+ SWord work = 0;
+
+ ASSERT(fix->procs.p == p); (void)p;
+ if (tb) {
+ db_lock(tb, LCK_WRITE_REC);
+ if (!(tb->common.status & DB_DELETE)) {
+ erts_aint_t diff;
+ #ifdef ERTS_SMP
+ erts_smp_mtx_lock(&tb->common.fixlock);
+ #endif
+
+ ASSERT(fixing_procs_rbt_lookup(tb->common.fixing_procs, p));
+
+ diff = -((erts_aint_t) fix->counter);
+ erts_smp_refc_add(&tb->common.fix_count,diff,0);
+ fix->counter = 0;
+
+ fixing_procs_rbt_delete(&tb->common.fixing_procs, fix);
+
+ #ifdef ERTS_SMP
+ erts_smp_mtx_unlock(&tb->common.fixlock);
+ #endif
+ if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) {
+ work += db_unfix_table_hash(&(tb->hash));
+ }
+
+ ASSERT(sizeof(DbFixation) == ERTS_ALC_DBG_BLK_SZ(fix));
+ ERTS_DB_ALC_MEM_UPDATE_(tb, sizeof(DbFixation), 0);
+ }
+ else {
+ ASSERT(fix->counter == 0);
+ }
+ db_unlock(tb, LCK_WRITE_REC);
+ }
+ else {
+ ASSERT(fix->counter == 0);
+ }
+
+ if (erts_refc_dectest(&fix->tabs.btid->refc, 0) == 0) {
+ erts_bin_free(fix->tabs.btid);
+ }
+ erts_free(ERTS_ALC_T_DB_FIXATION, fix);
+ ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
+ ++work;
+
+ return work;
+}
+
+
/*
* erts_db_process_exiting() is called when a process terminates.
* It returns 0 when completely done, and !0 when it wants to
@@ -3223,276 +3640,160 @@ retry:
int
erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks)
{
- ErtsDbProcCleanupState *state = (ErtsDbProcCleanupState *) c_p->u.terminate;
+ typedef struct {
+ enum {
+ GET_OWNED_TABLE,
+ FREE_OWNED_TABLE,
+ UNFIX_TABLES,
+ }op;
+ DbTable *tb;
+ } CleanupState;
+ CleanupState *state = (CleanupState *) c_p->u.terminate;
Eterm pid = c_p->common.id;
- ErtsDbProcCleanupState default_state;
- int ret;
+ CleanupState default_state;
+ SWord initial_reds = ERTS_BIF_REDS_LEFT(c_p);
+ SWord reds = initial_reds;
if (!state) {
state = &default_state;
- state->progress = ErtsDbProcCleanupProgressTables;
- state->op = ErtsDbProcCleanupOpGetTables;
+ state->op = GET_OWNED_TABLE;
+ state->tb = NULL;
}
- while (!0) {
+ do {
switch (state->op) {
- case ErtsDbProcCleanupOpGetTables:
- state->slots.size = ARRAY_CHUNK;
- db_meta_lock(meta_pid_to_tab, LCK_READ);
- ret = db_get_element_array(meta_pid_to_tab,
- pid,
- 2,
- state->slots.arr,
- &state->slots.size);
- db_meta_unlock(meta_pid_to_tab, LCK_READ);
- if (ret == DB_ERROR_BADKEY) {
- /* Done with tables; now fixations */
- state->progress = ErtsDbProcCleanupProgressFixations;
- state->op = ErtsDbProcCleanupOpGetFixations;
- break;
- } else if (ret != DB_ERROR_NONE) {
- ERTS_DB_INTERNAL_ERROR("Inconsistent ets table metadata");
- }
-
- state->slots.ix = 0;
- state->slots.clean_ix = 0;
- state->op = ErtsDbProcCleanupOpDeleteTables;
- /* Fall through */
-
- case ErtsDbProcCleanupOpDeleteTables:
-
- while (state->slots.ix < state->slots.size) {
- DbTable *tb = NULL;
- Sint ix = unsigned_val(state->slots.arr[state->slots.ix]);
- erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix);
- erts_smp_rwmtx_rlock(mmtl);
- if (!IS_SLOT_FREE(ix)) {
- tb = GET_ANY_SLOT_TAB(ix);
- ASSERT(tb);
- }
- erts_smp_rwmtx_runlock(mmtl);
- if (tb) {
- int do_yield;
- db_lock(tb, LCK_WRITE);
- /* Ownership may have changed since
- we looked up the table. */
- if (tb->common.owner != pid) {
- do_yield = 0;
- db_unlock(tb, LCK_WRITE);
- }
- else if (tb->common.heir != am_none
- && tb->common.heir != pid
- && give_away_to_heir(c_p, tb)) {
- do_yield = 0;
- }
- else {
- int first_call;
-#ifdef HARDDEBUG
- erts_fprintf(stderr,
- "erts_db_process_exiting(); Table: %T, "
- "Process: %T\n",
- tb->common.id, pid);
-#endif
- first_call = (tb->common.status & DB_DELETE) == 0;
- if (first_call) {
- /* Clear all access bits. */
- tb->common.status &= ~(DB_PROTECTED
- | DB_PUBLIC
- | DB_PRIVATE);
- tb->common.status |= DB_DELETE;
-
- if (is_atom(tb->common.id))
- remove_named_tab(tb, 0);
-
- free_heir_data(tb);
- free_fixations_locked(tb);
- }
-
- do_yield = free_table_cont(c_p, tb, first_call, 0);
- db_unlock(tb, LCK_WRITE);
- }
- if (do_yield)
- goto yield;
- }
- state->slots.ix++;
- if (ERTS_BIF_REDS_LEFT(c_p) <= 0)
- goto yield;
- }
+ case GET_OWNED_TABLE: {
+ DbTable* tb;
+ erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_STATUS);
+ tb = (DbTable*) erts_psd_get(c_p, ERTS_PSD_ETS_OWNED_TABLES);
+ erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_STATUS);
+
+ if (!tb) {
+ /* Done with owned tables; now fixations */
+ state->op = UNFIX_TABLES;
+ break;
+ }
- proc_exit_cleanup_tables_meta_data(pid, state);
- state->op = ErtsDbProcCleanupOpGetTables;
- break;
+ ASSERT(tb != state->tb);
+ state->tb = tb;
+ db_lock(tb, LCK_WRITE);
+ /*
+ * Ownership may have changed since we looked up the table.
+ */
+ if (tb->common.owner != pid) {
+ db_unlock(tb, LCK_WRITE);
+ break;
+ }
+ if (tb->common.heir != am_none
+ && tb->common.heir != pid
+ && give_away_to_heir(c_p, tb)) {
+ break;
+ }
+ tid_clear(c_p, tb);
+ /* Clear all access bits. */
+ tb->common.status &= ~(DB_PROTECTED | DB_PUBLIC | DB_PRIVATE);
+ tb->common.status |= DB_DELETE;
+
+ if (is_table_named(tb))
+ remove_named_tab(tb, 0);
+
+ free_heir_data(tb);
+ reds -= free_fixations_locked(c_p, tb);
+ db_unlock(tb, LCK_WRITE);
+ state->op = FREE_OWNED_TABLE;
+ break;
+ }
+ case FREE_OWNED_TABLE:
+ reds = free_table_continue(c_p, state->tb, reds);
+ if (reds < 0)
+ goto yield;
- case ErtsDbProcCleanupOpGetFixations:
- state->slots.size = ARRAY_CHUNK;
- db_meta_lock(meta_pid_to_fixed_tab, LCK_READ);
- ret = db_get_element_array(meta_pid_to_fixed_tab,
- pid,
- 2,
- state->slots.arr,
- &state->slots.size);
- db_meta_unlock(meta_pid_to_fixed_tab, LCK_READ);
-
- if (ret == DB_ERROR_BADKEY) {
- /* Done */
- state->progress = ErtsDbProcCleanupProgressDone;
- state->op = ErtsDbProcCleanupOpDone;
- break;
- } else if (ret != DB_ERROR_NONE) {
- ERTS_DB_INTERNAL_ERROR("Inconsistent ets fix table metadata");
- }
+ state->op = GET_OWNED_TABLE;
+ break;
- state->slots.ix = 0;
- state->slots.clean_ix = 0;
- state->op = ErtsDbProcCleanupOpDeleteFixations;
- /* Fall through */
+ case UNFIX_TABLES: {
+ DbFixation* fix;
- case ErtsDbProcCleanupOpDeleteFixations:
+ fix = (DbFixation*) erts_psd_get(c_p, ERTS_PSD_ETS_FIXED_TABLES);
- while (state->slots.ix < state->slots.size) {
- DbTable *tb = NULL;
- Sint ix = unsigned_val(state->slots.arr[state->slots.ix]);
- erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix);
- erts_smp_rwmtx_rlock(mmtl);
- if (IS_SLOT_ALIVE(ix)) {
- tb = meta_main_tab[ix].u.tb;
- ASSERT(tb);
- }
- erts_smp_rwmtx_runlock(mmtl);
- if (tb) {
- int reds = 0;
-
- db_lock(tb, LCK_WRITE_REC);
- if (!(tb->common.status & DB_DELETE)) {
- DbFixation** pp;
-
- #ifdef ERTS_SMP
- erts_smp_mtx_lock(&tb->common.fixlock);
- #endif
- reds = 10;
-
- for (pp = &tb->common.fixations; *pp != NULL;
- pp = &(*pp)->next) {
- if ((*pp)->pid == pid) {
- DbFixation* fix = *pp;
- erts_aint_t diff = -((erts_aint_t) fix->counter);
- erts_smp_refc_add(&tb->common.ref,diff,0);
- *pp = fix->next;
- erts_db_free(ERTS_ALC_T_DB_FIXATION,
- tb, fix, sizeof(DbFixation));
- ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
- break;
- }
- }
- #ifdef ERTS_SMP
- erts_smp_mtx_unlock(&tb->common.fixlock);
- #endif
- if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) {
- db_unfix_table_hash(&(tb->hash));
- reds += 40;
- }
- }
- db_unlock(tb, LCK_WRITE_REC);
- BUMP_REDS(c_p, reds);
- }
- state->slots.ix++;
- if (ERTS_BIF_REDS_LEFT(c_p) <= 0)
- goto yield;
- }
+ if (!fix) {
+ /* Done */
- proc_exit_cleanup_fixations_meta_data(pid, state);
- state->op = ErtsDbProcCleanupOpGetFixations;
- break;
+ if (state != &default_state)
+ erts_free(ERTS_ALC_T_DB_PROC_CLEANUP, state);
+ c_p->u.terminate = NULL;
- case ErtsDbProcCleanupOpDone:
+ BUMP_REDS(c_p, (initial_reds - reds));
+ return 0;
+ }
- if (state != &default_state)
- erts_free(ERTS_ALC_T_DB_PROC_CLEANUP, state);
- c_p->u.terminate = NULL;
- return 0;
+ fixed_tabs_delete(c_p, fix);
+ reds -= proc_cleanup_fixed_table(c_p, fix);
+ break;
+ }
default:
ERTS_DB_INTERNAL_ERROR("Bad internal state");
- }
- }
-
- yield:
+ }
- switch (state->progress) {
- case ErtsDbProcCleanupProgressTables:
- proc_exit_cleanup_tables_meta_data(pid, state);
- break;
- case ErtsDbProcCleanupProgressFixations:
- proc_exit_cleanup_fixations_meta_data(pid, state);
- break;
- default:
- break;
- }
+ } while (reds > 0);
- ASSERT(c_p->u.terminate == (void *) state
- || state == &default_state);
+ yield:
if (state == &default_state) {
c_p->u.terminate = erts_alloc(ERTS_ALC_T_DB_PROC_CLEANUP,
- sizeof(ErtsDbProcCleanupState));
- sys_memcpy(c_p->u.terminate,
- (void*) state,
- sizeof(ErtsDbProcCleanupState));
+ sizeof(CleanupState));
+ sys_memcpy(c_p->u.terminate, (void*) state, sizeof(CleanupState));
}
+ else
+ ASSERT(state == c_p->u.terminate);
return !0;
}
+
/* SMP note: table only need to be LCK_READ locked */
static void fix_table_locked(Process* p, DbTable* tb)
{
DbFixation *fix;
- DeclareTmpHeap(meta_tuple,3,p);
#ifdef ERTS_SMP
erts_smp_mtx_lock(&tb->common.fixlock);
#endif
- erts_smp_refc_inc(&tb->common.ref,1);
- fix = tb->common.fixations;
+ erts_smp_refc_inc(&tb->common.fix_count,1);
+ fix = tb->common.fixing_procs;
if (fix == NULL) {
tb->common.time.monotonic
= erts_get_monotonic_time(erts_proc_sched_data(p));
tb->common.time.offset = erts_get_time_offset();
}
else {
- for (; fix != NULL; fix = fix->next) {
- if (fix->pid == p->common.id) {
- ++(fix->counter);
+ fix = fixing_procs_rbt_lookup(fix, p);
+ if (fix) {
+ ASSERT(fixed_tabs_find(NULL, fix));
+ ++(fix->counter);
+
#ifdef ERTS_SMP
- erts_smp_mtx_unlock(&tb->common.fixlock);
+ erts_smp_mtx_unlock(&tb->common.fixlock);
#endif
- return;
- }
+ return;
}
}
fix = (DbFixation *) erts_db_alloc(ERTS_ALC_T_DB_FIXATION,
tb, sizeof(DbFixation));
ERTS_ETS_MISC_MEM_ADD(sizeof(DbFixation));
- fix->pid = p->common.id;
+ fix->tabs.btid = tb->common.btid;
+ erts_refc_inc(&fix->tabs.btid->refc, 2);
+ fix->procs.p = p;
fix->counter = 1;
- fix->next = tb->common.fixations;
- tb->common.fixations = fix;
+ fixing_procs_rbt_insert(&tb->common.fixing_procs, fix);
+
#ifdef ERTS_SMP
erts_smp_mtx_unlock(&tb->common.fixlock);
#endif
- p->flags |= F_USING_DB;
- UseTmpHeap(3,p);
- db_meta_lock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- if (db_put_hash(meta_pid_to_fixed_tab,
- TUPLE2(meta_tuple,
- p->common.id,
- make_small(tb->common.slot)),
- 0) != DB_ERROR_NONE) {
- UnUseTmpHeap(3,p);
- erts_exit(ERTS_ERROR_EXIT,"Could not insert ets metadata in safe_fixtable.");
- }
- UnUseTmpHeap(3,p);
- db_meta_unlock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
+ p->flags |= F_USING_DB;
+
+ fixed_tabs_insert(p, fix);
}
/* SMP note: May re-lock table
@@ -3500,28 +3801,26 @@ static void fix_table_locked(Process* p, DbTable* tb)
static void unfix_table_locked(Process* p, DbTable* tb,
db_lock_kind_t* kind_p)
{
- DbFixation** pp;
+ DbFixation* fix;
#ifdef ERTS_SMP
erts_smp_mtx_lock(&tb->common.fixlock);
#endif
- for (pp = &tb->common.fixations; *pp != NULL; pp = &(*pp)->next) {
- if ((*pp)->pid == p->common.id) {
- DbFixation* fix = *pp;
- erts_smp_refc_dec(&tb->common.ref,0);
- --(fix->counter);
- ASSERT(fix->counter >= 0);
- if (fix->counter > 0) {
- break;
- }
- *pp = fix->next;
+ fix = fixing_procs_rbt_lookup(tb->common.fixing_procs, p);
+
+ if (fix) {
+ erts_smp_refc_dec(&tb->common.fix_count,0);
+ --(fix->counter);
+ ASSERT(fix->counter >= 0);
+ if (fix->counter == 0) {
+ fixing_procs_rbt_delete(&tb->common.fixing_procs, fix);
#ifdef ERTS_SMP
erts_smp_mtx_unlock(&tb->common.fixlock);
#endif
- db_meta_lock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_fixed_tab,
- p->common.id, make_small(tb->common.slot));
- db_meta_unlock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
+ fixed_tabs_delete(p, fix);
+
+ erts_refc_dec(&fix->tabs.btid->refc, 1);
+
erts_db_free(ERTS_ALC_T_DB_FIXATION,
tb, (void *) fix, sizeof(DbFixation));
ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
@@ -3548,29 +3847,87 @@ unlocked:
}
}
-/* Assume that tb is WRITE locked */
-static void free_fixations_locked(DbTable *tb)
+struct free_fixations_ctx
{
- DbFixation *fix;
- DbFixation *next_fix;
+ Process* p;
+ DbTable* tb;
+ SWord cnt;
+};
+
+static void free_fixations_op(DbFixation* fix, void* vctx)
+{
+ struct free_fixations_ctx* ctx = (struct free_fixations_ctx*) vctx;
+ erts_aint_t diff;
+#ifdef DEBUG
+ DbTable* dbg_tb = btid2tab(fix->tabs.btid);
+#endif
+
+ ASSERT(!dbg_tb || dbg_tb == ctx->tb);
+ ASSERT(fix->counter > 0);
+ ASSERT(ctx->tb->common.status & DB_DELETE);
+
+ diff = -((erts_aint_t) fix->counter);
+ erts_smp_refc_add(&ctx->tb->common.fix_count, diff, 0);
+
+#ifdef ERTS_SMP
+ if (fix->procs.p != ctx->p) { /* Fixated by other process */
+ fix->counter = 0;
- fix = tb->common.fixations;
- while (fix != NULL) {
- erts_aint_t diff = -((erts_aint_t) fix->counter);
- erts_smp_refc_add(&tb->common.ref,diff,0);
- next_fix = fix->next;
- db_meta_lock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_fixed_tab,
- fix->pid,
- make_small(tb->common.slot));
- db_meta_unlock(meta_pid_to_fixed_tab, LCK_WRITE_REC);
- erts_db_free(ERTS_ALC_T_DB_FIXATION,
- tb, (void *) fix, sizeof(DbFixation));
- ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
+ /* Fake memory stats for table */
+ ASSERT(sizeof(DbFixation) == ERTS_ALC_DBG_BLK_SZ(fix));
+ ERTS_DB_ALC_MEM_UPDATE_(ctx->tb, sizeof(DbFixation), 0);
- fix = next_fix;
+ erts_schedule_ets_free_fixation(fix->procs.p->common.id, fix);
+ /*
+ * Either sys task is scheduled and erts_db_execute_free_fixation()
+ * will remove 'fix' or process will exit, drop sys task and
+ * proc_cleanup_fixed_table() will remove 'fix'.
+ */
}
- tb->common.fixations = NULL;
+ else
+#endif
+ {
+ fixed_tabs_delete(fix->procs.p, fix);
+
+ if (erts_refc_dectest(&fix->tabs.btid->refc, 0) == 0) {
+ erts_bin_free(fix->tabs.btid);
+ }
+
+ erts_db_free(ERTS_ALC_T_DB_FIXATION,
+ ctx->tb, (void *) fix, sizeof(DbFixation));
+ ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
+ }
+ ctx->cnt++;
+}
+
+#ifdef ERTS_SMP
+int erts_db_execute_free_fixation(Process* p, DbFixation* fix)
+{
+ ASSERT(fix->counter == 0);
+ fixed_tabs_delete(p, fix);
+
+ if (erts_refc_dectest(&fix->tabs.btid->refc, 0) == 0) {
+ erts_bin_free(fix->tabs.btid);
+ }
+ erts_free(ERTS_ALC_T_DB_FIXATION, fix);
+ ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation));
+ return 1;
+}
+#endif
+
+static SWord free_fixations_locked(Process* p, DbTable *tb)
+{
+ struct free_fixations_ctx ctx;
+
+ ERTS_SMP_LC_ASSERT(erts_smp_lc_rwmtx_is_rwlocked(&tb->common.rwlock));
+
+ ctx.p = p;
+ ctx.tb = tb;
+ ctx.cnt = 0;
+ fixing_procs_rbt_foreach_destroy(&tb->common.fixing_procs,
+ free_fixations_op, &ctx);
+ tb->common.fixing_procs = NULL;
+ return ctx.cnt;
}
static void set_heir(Process* me, DbTable* tb, Eterm heir, UWord heir_data)
@@ -3634,55 +3991,40 @@ static void free_heir_data(DbTable* tb)
static BIF_RETTYPE ets_delete_trap(BIF_ALIST_1)
{
- Process *p = BIF_P;
+ SWord initial_reds = ERTS_BIF_REDS_LEFT(BIF_P);
+ SWord reds = initial_reds;
Eterm cont = BIF_ARG_1;
- int trap;
Eterm* ptr = big_val(cont);
DbTable *tb = *((DbTable **) (UWord) (ptr + 1));
ASSERT(*ptr == make_pos_bignum_header(1));
- db_lock(tb, LCK_WRITE);
- trap = free_table_cont(p, tb, 0, 1);
- db_unlock(tb, LCK_WRITE);
-
- if (trap) {
- BIF_TRAP1(&ets_delete_continue_exp, p, cont);
+ if (free_table_continue(BIF_P, tb, reds) < 0) {
+ BUMP_ALL_REDS(BIF_P);
+ BIF_TRAP1(&ets_delete_continue_exp, BIF_P, cont);
}
else {
+ BUMP_REDS(BIF_P, (initial_reds - reds));
BIF_RET(am_true);
}
}
/*
- * free_table_cont() returns 0 when done and !0 when more work is needed.
+ * free_table_continue() returns reductions left
+ * done if >= 0
+ * yield if < 0
*/
-static int free_table_cont(Process *p,
- DbTable *tb,
- int first,
- int clean_meta_tab)
+static SWord free_table_continue(Process *p, DbTable *tb, SWord reds)
{
- Eterm result;
- erts_smp_rwmtx_t *mmtl;
-
-#ifdef HARDDEBUG
- if (!first) {
- erts_fprintf(stderr,"ets: free_table_cont %T (continue)\r\n",
- tb->common.id);
- }
-#endif
-
- result = tb->common.meth->db_free_table_continue(tb);
+ reds = tb->common.meth->db_free_table_continue(tb, reds);
- if (result == 0) {
+ if (reds < 0) {
#ifdef HARDDEBUG
erts_fprintf(stderr,"ets: free_table_cont %T (continue begin)\r\n",
tb->common.id);
#endif
/* More work to be done. Let other processes work and call us again. */
- BUMP_ALL_REDS(p);
- return !0;
}
else {
#ifdef HARDDEBUG
@@ -3690,27 +4032,28 @@ static int free_table_cont(Process *p,
tb->common.id);
#endif
/* Completely done - we will not get called again. */
- mmtl = get_meta_main_tab_lock(tb->common.slot);
-#ifdef ERTS_SMP
- if (erts_smp_rwmtx_tryrwlock(mmtl) == EBUSY) {
- erts_smp_rwmtx_rwunlock(&tb->common.rwlock);
- erts_smp_rwmtx_rwlock(mmtl);
- erts_smp_rwmtx_rwlock(&tb->common.rwlock);
- }
-#endif
- free_slot(tb->common.slot);
- erts_smp_rwmtx_rwunlock(mmtl);
-
- if (clean_meta_tab) {
- db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
- db_erase_bag_exact2(meta_pid_to_tab,tb->common.owner,
- make_small(tb->common.slot));
- db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
- }
- schedule_free_dbtable(tb);
- BUMP_REDS(p, 100);
- return 0;
+ delete_owned_table(p, tb);
+ table_dec_refc(tb, 0);
}
+ return reds;
+}
+
+struct fixing_procs_info_ctx
+{
+ Process* p;
+ Eterm list;
+};
+
+static void fixing_procs_info_op(DbFixation* fix, void* vctx)
+{
+ struct fixing_procs_info_ctx* ctx = (struct fixing_procs_info_ctx*) vctx;
+ Eterm* hp;
+ Eterm tpl;
+
+ hp = HAllocX(ctx->p, 5, 100);
+ tpl = TUPLE2(hp, fix->procs.p->common.id, make_small(fix->counter));
+ hp += 3;
+ ctx->list = CONS(hp, tpl, ctx->list);
}
static Eterm table_info(Process* p, DbTable* tb, Eterm What)
@@ -3759,7 +4102,7 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What)
} else if (What == am_node) {
ret = erts_this_dist_entry->sysname;
} else if (What == am_named_table) {
- ret = is_atom(tb->common.id) ? am_true : am_false;
+ ret = is_table_named(tb) ? am_true : am_false;
} else if (What == am_compressed) {
ret = tb->common.compress ? am_true : am_false;
}
@@ -3784,9 +4127,9 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What)
if (IS_FIXED(tb)) {
Uint need;
Eterm *hp;
- Eterm tpl, lst;
- DbFixation *fix;
+ Eterm time;
Sint64 mtime;
+ struct fixing_procs_info_ctx ctx;
need = 3;
if (use_monotonic) {
@@ -3799,19 +4142,15 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What)
mtime = 0;
need += 4;
}
- for (fix = tb->common.fixations; fix != NULL; fix = fix->next) {
- need += 5;
- }
+ ctx.p = p;
+ ctx.list = NIL;
+ fixing_procs_rbt_foreach(tb->common.fixing_procs,
+ fixing_procs_info_op,
+ &ctx);
+
hp = HAlloc(p, need);
- lst = NIL;
- for (fix = tb->common.fixations; fix != NULL; fix = fix->next) {
- tpl = TUPLE2(hp,fix->pid,make_small(fix->counter));
- hp += 3;
- lst = CONS(hp,tpl,lst);
- hp += 2;
- }
if (use_monotonic)
- tpl = (IS_SSMALL(mtime)
+ time = (IS_SSMALL(mtime)
? make_small(mtime)
: erts_sint64_to_big(mtime, &hp));
else {
@@ -3819,10 +4158,10 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What)
erts_make_timestamp_value(&ms, &s, &us,
tb->common.time.monotonic,
tb->common.time.offset);
- tpl = TUPLE3(hp, make_small(ms), make_small(s), make_small(us));
+ time = TUPLE3(hp, make_small(ms), make_small(s), make_small(us));
hp += 4;
}
- ret = TUPLE2(hp, tpl, lst);
+ ret = TUPLE2(hp, time, ctx.list);
} else {
ret = am_false;
}
@@ -3867,7 +4206,19 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What)
static void print_table(fmtfn_t to, void *to_arg, int show, DbTable* tb)
{
- erts_print(to, to_arg, "Table: %T\n", tb->common.id);
+ Eterm tid;
+ Eterm heap[ERTS_MAGIC_REF_THING_SIZE];
+
+ if (is_table_named(tb)) {
+ tid = tb->common.the_name;
+ } else {
+ ErlOffHeap oh;
+ ERTS_INIT_OFF_HEAP(&oh);
+ write_magic_ref_thing(heap, &oh, (ErtsMagicBinary *) tb->common.btid);
+ tid = make_internal_ref(heap);
+ }
+
+ erts_print(to, to_arg, "Table: %T\n", tid);
erts_print(to, to_arg, "Name: %T\n", tb->common.the_name);
tb->common.meth->db_print(to, to_arg, show, tb);
@@ -3885,21 +4236,30 @@ static void print_table(fmtfn_t to, void *to_arg, int show, DbTable* tb)
erts_print(to, to_arg, "Read Concurrency: %T\n", table_info(NULL, tb, am_read_concurrency));
}
+typedef struct {
+ fmtfn_t to;
+ void *to_arg;
+ int show;
+} ErtsPrintDbInfo;
+
+static void
+db_info_print(DbTable *tb, void *vpdbip)
+{
+ ErtsPrintDbInfo *pdbip = (ErtsPrintDbInfo *) vpdbip;
+ erts_print(pdbip->to, pdbip->to_arg, "=ets:%T\n", tb->common.owner);
+ erts_print(pdbip->to, pdbip->to_arg, "Slot: %bpu\n", (Uint) tb);
+ print_table(pdbip->to, pdbip->to_arg, pdbip->show, tb);
+}
+
void db_info(fmtfn_t to, void *to_arg, int show) /* Called by break handler */
{
- int i;
- for (i=0; i < db_max_tabs; i++)
- if (IS_SLOT_ALIVE(i)) {
- erts_print(to, to_arg, "=ets:%T\n", meta_main_tab[i].u.tb->common.owner);
- erts_print(to, to_arg, "Slot: %d\n", i);
- print_table(to, to_arg, show, meta_main_tab[i].u.tb);
- }
-#ifdef DEBUG
- erts_print(to, to_arg, "=internal_ets: Process to table index\n");
- print_table(to, to_arg, show, meta_pid_to_tab);
- erts_print(to, to_arg, "=internal_ets: Process to fixation index\n");
- print_table(to, to_arg, show, meta_pid_to_fixed_tab);
-#endif
+ ErtsPrintDbInfo pdbi;
+
+ pdbi.to = to;
+ pdbi.to_arg = to_arg;
+ pdbi.show = show;
+
+ erts_db_foreach_table(db_info_print, &pdbi);
}
Uint
@@ -3914,15 +4274,22 @@ erts_get_ets_misc_mem_size(void)
void
erts_db_foreach_table(void (*func)(DbTable *, void *), void *arg)
{
- int i, j;
- j = 0;
- for(i = 0; (i < db_max_tabs && j < meta_main_tab_cnt); i++) {
- if (IS_SLOT_ALIVE(i)) {
- j++;
- (*func)(meta_main_tab[i].u.tb, arg);
- }
+ int ix;
+
+ ASSERT(erts_smp_thr_progress_is_blocking());
+
+ for (ix = 0; ix < erts_no_schedulers; ix++) {
+ ErtsSchedulerData *esdp = ERTS_SCHEDULER_IX(ix);
+ DbTable *first = esdp->ets_tables.clist;
+ if (first) {
+ DbTable *tb = first;
+ do {
+ if (is_table_alive(tb))
+ (*func)(tb, arg);
+ tb = tb->common.all.next;
+ } while (tb != first);
+ }
}
- ASSERT(j == meta_main_tab_cnt);
}
/* SMP Note: May only be used when system is locked */
@@ -3972,53 +4339,3 @@ erts_ets_colliding_names(Process* p, Eterm name, Uint cnt)
return list;
}
-/*
- * For testing only
- * Retreive meta table size state
- */
-Eterm erts_ets_get_meta_state(Process* p)
-{
- Eterm* hp = HAlloc(p, 3);
- return TUPLE2(hp,
- erts_ets_hash_get_memstate(p, &meta_pid_to_tab->hash),
- erts_ets_hash_get_memstate(p, &meta_pid_to_fixed_tab->hash));
-}
-/*
- * For testing only
- * Restore a previously retrieved meta table size state.
- * We do this to suppress failed memory checks
- * caused by the hysteresis of meta tables grow/shrink limits.
- */
-Eterm erts_ets_restore_meta_state(Process* p, Eterm meta_state)
-{
- Eterm* tv;
- Eterm* hp;
- if (!is_tuple_arity(meta_state, 2))
- return am_badarg;
-
- tv = tuple_val(meta_state);
- hp = HAlloc(p, 3);
- return TUPLE2(hp,
- erts_ets_hash_restore_memstate(&meta_pid_to_tab->hash, tv[1]),
- erts_ets_hash_restore_memstate(&meta_pid_to_fixed_tab->hash, tv[2]));
-}
-
-#ifdef HARDDEBUG /* Here comes some debug functions */
-
-void db_check_tables(void)
-{
-#ifdef ERTS_SMP
- return;
-#else
- int i;
-
- for (i = 0; i < db_max_tabs; i++) {
- if (IS_SLOT_ALIVE(i)) {
- DbTable* tb = meta_main_tab[i].t;
- tb->common.meth->db_check_table(tb);
- }
- }
-#endif
-}
-
-#endif /* HARDDEBUG */
diff --git a/erts/emulator/beam/erl_db.h b/erts/emulator/beam/erl_db.h
index 852440ff31..4ff9f224e8 100644
--- a/erts/emulator/beam/erl_db.h
+++ b/erts/emulator/beam/erl_db.h
@@ -24,8 +24,37 @@
*
*/
-#ifndef __DB_H__
-#define __DB_H__
+#ifndef ERTS_DB_SCHED_SPEC_TYPES__
+#define ERTS_DB_SCHED_SPEC_TYPES__
+
+union db_table;
+typedef union db_table DbTable;
+
+typedef struct ErtsEtsAllReq_ ErtsEtsAllReq;
+
+typedef struct {
+ ErtsEtsAllReq *next;
+ ErtsEtsAllReq *prev;
+} ErtsEtsAllReqList;
+
+typedef struct {
+ ErtsEtsAllReq *ongoing;
+ ErlHeapFragment *hfrag;
+ DbTable *tab;
+ ErtsEtsAllReq *queue;
+} ErtsEtsAllYieldData;
+
+typedef struct {
+ Uint count;
+ DbTable *clist;
+} ErtsEtsTables;
+
+#endif /* ERTS_DB_SCHED_SPEC_TYPES__ */
+
+#ifndef ERTS_ONLY_SCHED_SPEC_ETS_DATA
+
+#ifndef ERL_DB_H__
+#define ERL_DB_H__
#include "sys.h"
#undef ERL_THR_PROGRESS_TSD_TYPE_ONLY
@@ -46,6 +75,12 @@ typedef struct {
ErtsThrPrgrLaterOp data;
} DbTableRelease;
+struct ErtsSchedulerData_;
+int erts_handle_yielded_ets_all_request(struct ErtsSchedulerData_ *esdp,
+ ErtsEtsAllYieldData *eadp);
+
+void erts_ets_sched_spec_data_init(struct ErtsSchedulerData_ *esdp);
+
/*
* So, the structure for a database table, NB this is only
* interesting in db.c.
@@ -74,6 +109,7 @@ typedef enum {
void init_db(ErtsDbSpinCount);
int erts_db_process_exiting(Process *, ErtsProcLocks);
+int erts_db_execute_free_fixation(Process*, DbFixation*);
void db_info(fmtfn_t, void *, int);
void erts_db_foreach_table(void (*)(DbTable *, void *), void *);
void erts_db_foreach_offheap(DbTable *,
@@ -86,16 +122,14 @@ extern int erts_ets_realloc_always_moves; /* set in erl_init */
extern int erts_ets_always_compress; /* set in erl_init */
extern Export ets_select_delete_continue_exp;
extern Export ets_select_count_continue_exp;
+extern Export ets_select_replace_continue_exp;
extern Export ets_select_continue_exp;
extern erts_smp_atomic_t erts_ets_misc_mem_size;
Eterm erts_ets_colliding_names(Process*, Eterm name, Uint cnt);
-Eterm erts_ets_get_meta_state(Process* p);
-Eterm erts_ets_restore_meta_state(Process* p, Eterm target_state);
-
Uint erts_db_get_max_tabs(void);
-#endif
+#endif /* ERL_DB_H__ */
#if defined(ERTS_WANT_DB_INTERNAL__) && !defined(ERTS_HAVE_DB_INTERNAL__)
#define ERTS_HAVE_DB_INTERNAL__
@@ -271,3 +305,4 @@ erts_db_free_nt(ErtsAlcType_t type, void *ptr, Uint size)
#endif /* #if defined(ERTS_WANT_DB_INTERNAL__) && !defined(ERTS_HAVE_DB_INTERNAL__) */
+#endif /* !ERTS_ONLY_SCHED_SPEC_ETS_DATA */
diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c
index bb29d56aa7..7ab27df00c 100644
--- a/erts/emulator/beam/erl_db_hash.c
+++ b/erts/emulator/beam/erl_db_hash.c
@@ -288,6 +288,9 @@ static ERTS_INLINE Sint next_slot_w(DbTableHash* tb, Uint ix,
#endif
}
+#ifndef MIN
+#define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
+#endif
/*
* Some special binary flags
@@ -352,7 +355,8 @@ static ERTS_INLINE void SET_SEGTAB(DbTableHash* tb,
erts_smp_atomic_set_nob(&tb->segtab, (erts_aint_t) segtab);
}
-
+/* Used by select_replace on analyze_pattern */
+typedef int (*extra_match_validator_t)(int keypos, Eterm match, Eterm guard, Eterm body);
/*
** Forward decl's (static functions)
@@ -368,8 +372,9 @@ static void shrink(DbTableHash* tb, int nitems);
static void grow(DbTableHash* tb, int nitems);
static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2,
Uint sz, DbTableHash*);
-static int analyze_pattern(DbTableHash *tb, Eterm pattern,
- struct mp_info *mpi);
+static int analyze_pattern(DbTableHash *tb, Eterm pattern,
+ extra_match_validator_t extra_validator, /* Optional callback */
+ struct mp_info *mpi);
/*
* Method interface functions
@@ -393,24 +398,29 @@ static int db_erase_object_hash(DbTable *tbl, Eterm object,Eterm *ret);
static int db_slot_hash(Process *p, DbTable *tbl,
Eterm slot_term, Eterm *ret);
-static int db_select_chunk_hash(Process *p, DbTable *tbl,
+static int db_select_chunk_hash(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Sint chunk_size,
int reverse, Eterm *ret);
-static int db_select_hash(Process *p, DbTable *tbl,
+static int db_select_hash(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, int reverse, Eterm *ret);
-static int db_select_count_hash(Process *p, DbTable *tbl,
- Eterm pattern, Eterm *ret);
-static int db_select_delete_hash(Process *p, DbTable *tbl,
- Eterm pattern, Eterm *ret);
-
-static int db_select_continue_hash(Process *p, DbTable *tbl,
+static int db_select_continue_hash(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
-static int db_select_count_continue_hash(Process *p, DbTable *tbl,
+static int db_select_count_hash(Process *p, DbTable *tbl, Eterm tid,
+ Eterm pattern, Eterm *ret);
+static int db_select_count_continue_hash(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
+static int db_select_delete_hash(Process *p, DbTable *tbl, Eterm tid,
+ Eterm pattern, Eterm *ret);
static int db_select_delete_continue_hash(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
+
+static int db_select_replace_hash(Process *p, DbTable *tbl, Eterm tid,
+ Eterm pattern, Eterm *ret);
+static int db_select_replace_continue_hash(Process *p, DbTable *tbl,
+ Eterm continuation, Eterm *ret);
+
static int db_take_hash(Process *, DbTable *, Eterm, Eterm *);
static void db_print_hash(fmtfn_t to,
void *to_arg,
@@ -418,7 +428,7 @@ static void db_print_hash(fmtfn_t to,
DbTable *tbl);
static int db_free_table_hash(DbTable *tbl);
-static int db_free_table_continue_hash(DbTable *tbl);
+static SWord db_free_table_continue_hash(DbTable *tbl, SWord reds);
static void db_foreach_offheap_hash(DbTable *,
@@ -523,17 +533,14 @@ DbTableMethod db_hash =
db_select_delete_continue_hash,
db_select_count_hash,
db_select_count_continue_hash,
+ db_select_replace_hash,
+ db_select_replace_continue_hash,
db_take_hash,
db_delete_all_objects_hash,
db_free_table_hash,
db_free_table_continue_hash,
db_print_hash,
db_foreach_offheap_hash,
-#ifdef HARDDEBUG
- db_check_table_hash,
-#else
- NULL,
-#endif
db_lookup_dbterm_hash,
db_finalize_dbterm_hash
};
@@ -581,9 +588,10 @@ static void restore_fixdel(DbTableHash* tb, FixedDeletion* fixdel)
** Table interface routines ie what's called by the bif's
*/
-void db_unfix_table_hash(DbTableHash *tb)
+SWord db_unfix_table_hash(DbTableHash *tb)
{
FixedDeletion* fixdel;
+ SWord work = 0;
ERTS_SMP_LC_ASSERT(erts_smp_lc_rwmtx_is_rwlocked(&tb->common.rwlock)
|| (erts_smp_lc_rwmtx_is_rlocked(&tb->common.rwlock)
@@ -604,7 +612,7 @@ restart:
if (!IS_FIXED(tb)) {
goto restart; /* unfixed again! */
}
- return;
+ return work;
}
if (ix < NACTIVE(tb)) {
bp = &BUCKET(tb, ix);
@@ -614,6 +622,7 @@ restart:
if (b->hvalue == INVALID_HASH) {
*bp = b->next;
free_term(tb, b);
+ work++;
b = *bp;
} else {
bp = &b->next;
@@ -629,9 +638,12 @@ restart:
(void *) fx,
sizeof(FixedDeletion));
ERTS_ETS_MISC_MEM_ADD(-sizeof(FixedDeletion));
+ work++;
}
/* ToDo: Maybe try grow/shrink the table as well */
+
+ return work;
}
int db_create_hash(Process *p, DbTable *tbl)
@@ -846,7 +858,6 @@ Lnew:
grow(tb, nitems);
}
}
- CHECK_TABLES();
return DB_ERROR_NONE;
Ldone:
@@ -871,7 +882,6 @@ get_term_list(Process *p, DbTableHash *tb, Eterm key, HashValue hval,
}
}
copy = build_term_list(p, b1, b2, sz, tb);
- CHECK_TABLES();
if (bend) {
*bend = b2;
}
@@ -903,70 +913,6 @@ done:
RUNLOCK_HASH(lck);
return DB_ERROR_NONE;
}
-
-int db_get_element_array(DbTable *tbl,
- Eterm key,
- int ndex,
- Eterm *ret,
- int *num_ret)
-{
- DbTableHash *tb = &tbl->hash;
- HashValue hval;
- int ix;
- HashDbTerm* b1;
- int num = 0;
- int retval;
- erts_smp_rwmtx_t* lck;
-
- ASSERT(!IS_FIXED(tbl)); /* no support for fixed tables here */
-
- hval = MAKE_HASH(key);
- lck = RLOCK_HASH(tb, hval);
- ix = hash_to_ix(tb, hval);
- b1 = BUCKET(tb, ix);
-
- while(b1 != 0) {
- if (has_live_key(tb,b1,key,hval)) {
- if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) {
- HashDbTerm* b;
- HashDbTerm* b2 = b1->next;
-
- while(b2 != NULL && has_live_key(tb,b2,key,hval)) {
- if (ndex > arityval(b2->dbterm.tpl[0])) {
- retval = DB_ERROR_BADITEM;
- goto done;
- }
- b2 = b2->next;
- }
-
- b = b1;
- while(b != b2) {
- if (num < *num_ret) {
- ret[num++] = b->dbterm.tpl[ndex];
- } else {
- retval = DB_ERROR_NONE;
- goto done;
- }
- b = b->next;
- }
- *num_ret = num;
- }
- else {
- ASSERT(*num_ret > 0);
- ret[0] = b1->dbterm.tpl[ndex];
- *num_ret = 1;
- }
- retval = DB_ERROR_NONE;
- goto done;
- }
- b1 = b1->next;
- }
- retval = DB_ERROR_BADKEY;
-done:
- RUNLOCK_HASH(lck);
- return retval;
-}
-
static int db_member_hash(DbTable *tbl, Eterm key, Eterm *ret)
{
@@ -1059,54 +1005,6 @@ done:
}
/*
- * Very internal interface, removes elements of arity two from
- * BAG. Used for the PID meta table
- */
-int db_erase_bag_exact2(DbTable *tbl, Eterm key, Eterm value)
-{
- DbTableHash *tb = &tbl->hash;
- HashValue hval;
- int ix;
- HashDbTerm** bp;
- HashDbTerm* b;
- erts_smp_rwmtx_t* lck;
- int found = 0;
-
- hval = MAKE_HASH(key);
- lck = WLOCK_HASH(tb,hval);
- ix = hash_to_ix(tb, hval);
- bp = &BUCKET(tb, ix);
- b = *bp;
-
- ASSERT(!IS_FIXED(tb));
- ASSERT((tb->common.status & DB_BAG));
- ASSERT(!tb->common.compress);
-
- while(b != 0) {
- if (has_live_key(tb,b,key,hval)) {
- found = 1;
- if ((arityval(b->dbterm.tpl[0]) == 2) &&
- EQ(value, b->dbterm.tpl[2])) {
- *bp = b->next;
- free_term(tb, b);
- erts_smp_atomic_dec_nob(&tb->common.nitems);
- b = *bp;
- break;
- }
- } else if (found) {
- break;
- }
- bp = &b->next;
- b = b->next;
- }
- WUNLOCK_HASH(lck);
- if (found) {
- try_shrink(tb);
- }
- return DB_ERROR_NONE;
-}
-
-/*
** NB, this is for the db_erase/2 bif.
*/
int db_erase_hash(DbTable *tbl, Eterm key, Eterm *ret)
@@ -1253,814 +1151,1104 @@ static BIF_RETTYPE bif_trap1(Export *bif,
{
BIF_TRAP1(bif, p, p1);
}
-
+
+
/*
- * Continue collecting select matches, this may happen either due to a trap
- * or when the user calls ets:select/1
+ * Match traversal callbacks
*/
-static int db_select_continue_hash(Process *p,
- DbTable *tbl,
- Eterm continuation,
- Eterm *ret)
-{
- DbTableHash *tb = &tbl->hash;
- Sint slot_ix;
- Sint save_slot_ix;
- Sint chunk_size;
- int all_objects;
- Binary *mp;
- int num_left = 1000;
- HashDbTerm *current = 0;
- Eterm match_list;
- Eterm *hp;
- Eterm match_res;
- Sint got;
- Eterm *tptr;
- erts_smp_rwmtx_t* lck;
-#define RET_TO_BIF(Term, State) do { *ret = (Term); return State; } while(0);
+/* Called when no match is possible.
+ * context_ptr: Pointer to context
+ * ret: Pointer to traversal function term return.
+ *
+ * Both the direct return value and 'ret' are used as the traversal function return values.
+ */
+typedef int (*mtraversal_on_nothing_can_match_t)(void* context_ptr, Eterm* ret);
+
+/* Called for each match result.
+ * context_ptr: Pointer to context
+ * slot_ix: Current slot index
+ * current_ptr_ptr: Triple pointer to either the bucket or the 'next' pointer in the previous element;
+ * can be (carefully) used to adjust iteration when deleting or replacing elements.
+ * match_res: The result of running the match program against the current term.
+ *
+ * Should return 1 for successful match, 0 otherwise.
+ */
+typedef int (*mtraversal_on_match_res_t)(void* context_ptr, Sint slot_ix, HashDbTerm*** current_ptr_ptr,
+ Eterm match_res);
+
+/* Called when either we've matched enough elements in this cycle or EOT was reached.
+ * context_ptr: Pointer to context
+ * slot_ix: Current slot index
+ * got: How many elements have been matched so far
+ * iterations_left: Number of intended iterations (down from an initial max.) left in this traversal cycle
+ * mpp: Double pointer to the compiled match program
+ * ret: Pointer to traversal function term return.
+ *
+ * Both the direct return value and 'ret' are used as the traversal function return values.
+ * If *mpp is set to NULL, it won't be deallocated (useful for trapping.)
+ */
+typedef int (*mtraversal_on_loop_ended_t)(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret);
+
+/* Called when it's time to trap
+ * context_ptr: Pointer to context
+ * slot_ix: Current slot index
+ * got: How many elements have been matched so far
+ * mpp: Double pointer to the compiled match program
+ * ret: Pointer to traversal function term return.
+ *
+ * Both the direct return value and 'ret' are used as the traversal function return values.
+ * If *mpp is set to NULL, it won't be deallocated (useful for trapping.)
+ */
+typedef int (*mtraversal_on_trap_t)(void* context_ptr, Sint slot_ix, Sint got, Binary** mpp, Eterm* ret);
- /* Decode continuation. We know it's a tuple but not the arity or anything else */
+/*
+ * Begin hash table match traversal
+ */
+static int match_traverse(Process* p, DbTableHash* tb,
+ Eterm pattern,
+ extra_match_validator_t extra_match_validator, /* Optional */
+ Sint chunk_size, /* If 0, no chunking */
+ Sint iterations_left, /* Nr. of iterations left */
+ Eterm** hpp, /* Heap */
+ int lock_for_write, /* Set to 1 if we're going to delete or
+ modify existing terms */
+ mtraversal_on_nothing_can_match_t on_nothing_can_match,
+ mtraversal_on_match_res_t on_match_res,
+ mtraversal_on_loop_ended_t on_loop_ended,
+ mtraversal_on_trap_t on_trap,
+ void* context_ptr, /* State for callbacks above */
+ Eterm* ret)
+{
+ Sint slot_ix; /* Slot index */
+ HashDbTerm** current_ptr; /* Refers to either the bucket pointer or
+ * the 'next' pointer in the previous term
+ */
+ HashDbTerm* saved_current; /* Helper to avoid double skip on match */
+ struct mp_info mpi;
+ unsigned current_list_pos = 0; /* Prefound buckets list index */
+ Eterm match_res;
+ Sint got = 0; /* Matched terms counter */
+ erts_smp_rwmtx_t* lck; /* Slot lock */
+ int ret_value;
+#ifdef ERTS_SMP
+ erts_smp_rwmtx_t* (*lock_hash_function)(DbTableHash*, HashValue)
+ = (lock_for_write ? WLOCK_HASH : RLOCK_HASH);
+ void (*unlock_hash_function)(erts_smp_rwmtx_t*)
+ = (lock_for_write ? WUNLOCK_HASH : RUNLOCK_HASH);
+#else
+ #define lock_hash_function(tb, hval) NULL
+ #define unlock_hash_function(lck) ((void)lck)
+#endif
+ Sint (*next_slot_function)(DbTableHash*, Uint, erts_smp_rwmtx_t**)
+ = (lock_for_write ? next_slot_w : next_slot);
- tptr = tuple_val(continuation);
+ if ((ret_value = analyze_pattern(tb, pattern, extra_match_validator, &mpi))
+ != DB_ERROR_NONE)
+ {
+ *ret = NIL;
+ goto done;
+ }
- if (arityval(*tptr) != 6)
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
-
- if (!is_small(tptr[2]) || !is_small(tptr[3]) ||
- !(is_list(tptr[5]) || tptr[5] == NIL) || !is_small(tptr[6]))
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
- if ((chunk_size = signed_val(tptr[3])) < 0)
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
- mp = erts_db_get_match_prog_binary(tptr[4]);
- if (!mp)
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
- all_objects = mp->flags & BIN_FLAG_ALL_OBJECTS;
- match_list = tptr[5];
- if ((got = signed_val(tptr[6])) < 0)
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
+ if (!mpi.something_can_match) {
+ /* Can't possibly match anything */
+ ret_value = on_nothing_can_match(context_ptr, ret);
+ goto done;
+ }
- slot_ix = signed_val(tptr[2]);
- if (slot_ix < 0 /* EOT */
- || (chunk_size && got >= chunk_size)) {
- goto done; /* Already got all or enough in the match_list */
+ if (mpi.all_objects) {
+ mpi.mp->flags |= BIN_FLAG_ALL_OBJECTS;
}
- lck = RLOCK_HASH(tb,slot_ix);
- if (slot_ix >= NACTIVE(tb)) {
- RUNLOCK_HASH(lck);
- RET_TO_BIF(NIL,DB_ERROR_BADPARAM);
+ /*
+ * Look for initial slot / bucket
+ */
+ if (!mpi.key_given) {
+ /* Run this code if pattern is variable or GETKEY(pattern) */
+ /* is a variable */
+ slot_ix = 0;
+ lck = lock_hash_function(tb,slot_ix);
+ for (;;) {
+ ASSERT(slot_ix < NACTIVE(tb));
+ if (*(current_ptr = &BUCKET(tb,slot_ix)) != NULL) {
+ break;
+ }
+ slot_ix = next_slot_function(tb,slot_ix,&lck);
+ if (slot_ix == 0) {
+ ret_value = on_loop_ended(context_ptr, slot_ix, got, iterations_left, &mpi.mp, ret);
+ goto done;
+ }
+ }
+ } else {
+ /* We have at least one */
+ slot_ix = mpi.lists[current_list_pos].ix;
+ lck = lock_hash_function(tb, slot_ix);
+ current_ptr = mpi.lists[current_list_pos].bucket;
+ ASSERT(*current_ptr == BUCKET(tb,slot_ix));
+ ++current_list_pos;
}
- while ((current = BUCKET(tb,slot_ix)) == NULL) {
- slot_ix = next_slot(tb, slot_ix, &lck);
- if (slot_ix == 0) {
- slot_ix = -1; /* EOT */
- goto done;
- }
- }
+ /*
+ * Execute traversal cycle
+ */
for(;;) {
- if (current->hvalue != INVALID_HASH &&
- (match_res = db_match_dbterm(&tb->common, p, mp, all_objects,
- &current->dbterm, &hp, 2),
- is_value(match_res))) {
+ if (*current_ptr != NULL) {
+ if ((*current_ptr)->hvalue != INVALID_HASH) {
+ match_res = db_match_dbterm(&tb->common, p, mpi.mp, 0,
+ &(*current_ptr)->dbterm, hpp, 2);
+ saved_current = *current_ptr;
+ if (on_match_res(context_ptr, slot_ix, &current_ptr, match_res)) {
+ ++got;
+ }
+ --iterations_left;
+ if (*current_ptr != saved_current) {
+ /* Don't advance to next, the callback did it already */
+ continue;
+ }
+ }
+ current_ptr = &((*current_ptr)->next);
+ }
+ else if (mpi.key_given) { /* Key is bound */
+ unlock_hash_function(lck);
+ if (current_list_pos == mpi.num_lists) {
+ ret_value = on_loop_ended(context_ptr, -1, got, iterations_left, &mpi.mp, ret);
+ goto done;
+ } else {
+ slot_ix = mpi.lists[current_list_pos].ix;
+ lck = lock_hash_function(tb, slot_ix);
+ current_ptr = mpi.lists[current_list_pos].bucket;
+ ASSERT(mpi.lists[current_list_pos].bucket == &BUCKET(tb,slot_ix));
+ ++current_list_pos;
+ }
+ }
+ else { /* Key is variable */
+ if ((slot_ix = next_slot_function(tb,slot_ix,&lck)) == 0) {
+ slot_ix = -1;
+ break;
+ }
+ if (chunk_size && got >= chunk_size) {
+ unlock_hash_function(lck);
+ break;
+ }
+ if (iterations_left <= 0 || MBUF(p)) {
+ /*
+ * We have either reached our limit, or just created some heap fragments.
+ * Since many heap fragments will make the GC slower, trap and GC now.
+ */
+ unlock_hash_function(lck);
+ ret_value = on_trap(context_ptr, slot_ix, got, &mpi.mp, ret);
+ goto done;
+ }
+ current_ptr = &BUCKET(tb,slot_ix);
+ }
+ }
- match_list = CONS(hp, match_res, match_list);
- ++got;
- }
+ ret_value = on_loop_ended(context_ptr, slot_ix, got, iterations_left, &mpi.mp, ret);
- --num_left;
- save_slot_ix = slot_ix;
- if ((current = next(tb, (Uint*)&slot_ix, &lck, current)) == NULL) {
- slot_ix = -1; /* EOT */
- break;
- }
- if (slot_ix != save_slot_ix) {
- if (chunk_size && got >= chunk_size) {
- RUNLOCK_HASH(lck);
- break;
- }
- if (num_left <= 0 || MBUF(p)) {
- /*
- * We have either reached our limit, or just created some heap fragments.
- * Since many heap fragments will make the GC slower, trap and GC now.
- */
- RUNLOCK_HASH(lck);
- goto trap;
- }
- }
- }
done:
- BUMP_REDS(p, 1000 - num_left);
- if (chunk_size) {
- Eterm continuation;
- Eterm rest = NIL;
- Sint rest_size = 0;
-
- if (got > chunk_size) { /* Cannot write destructively here,
- the list may have
- been in user space */
- rest = NIL;
- hp = HAlloc(p, (got - chunk_size) * 2);
- while (got-- > chunk_size) {
- rest = CONS(hp, CAR(list_val(match_list)), rest);
- hp += 2;
- match_list = CDR(list_val(match_list));
- ++rest_size;
- }
- }
- if (rest != NIL || slot_ix >= 0) {
- hp = HAlloc(p,3+7);
- continuation = TUPLE6(hp, tptr[1], make_small(slot_ix),
- tptr[3], tptr[4], rest,
- make_small(rest_size));
- hp += 7;
- RET_TO_BIF(TUPLE2(hp, match_list, continuation),DB_ERROR_NONE);
- } else {
- if (match_list != NIL) {
- hp = HAlloc(p, 3);
- RET_TO_BIF(TUPLE2(hp, match_list, am_EOT),DB_ERROR_NONE);
- } else {
- RET_TO_BIF(am_EOT, DB_ERROR_NONE);
- }
- }
+ /* We should only jump directly to this label if
+ * we've already called on_nothing_can_match / on_loop_ended / on_trap
+ */
+ if (mpi.mp != NULL) {
+ erts_bin_free(mpi.mp);
}
- RET_TO_BIF(match_list,DB_ERROR_NONE);
-
-trap:
- BUMP_ALL_REDS(p);
-
- hp = HAlloc(p,7);
- continuation = TUPLE6(hp, tptr[1], make_small(slot_ix), tptr[3],
- tptr[4], match_list, make_small(got));
- RET_TO_BIF(bif_trap1(&ets_select_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
-
-#undef RET_TO_BIF
-
-}
+ if (mpi.lists != mpi.dlists) {
+ erts_free(ERTS_ALC_T_DB_SEL_LIST,
+ (void *) mpi.lists);
+ }
+ return ret_value;
-static int db_select_hash(Process *p, DbTable *tbl,
- Eterm pattern, int reverse,
- Eterm *ret)
-{
- return db_select_chunk_hash(p, tbl, pattern, 0, reverse, ret);
+#ifndef SMP
+#undef lock_hash_function
+#undef unlock_hash_function
+#endif
}
-static int db_select_chunk_hash(Process *p, DbTable *tbl,
- Eterm pattern, Sint chunk_size,
- int reverse, /* not used */
- Eterm *ret)
-{
- DbTableHash *tb = &tbl->hash;
- struct mp_info mpi;
- Sint slot_ix;
- HashDbTerm *current = 0;
- unsigned current_list_pos = 0;
- Eterm match_list;
+/*
+ * Continue hash table match traversal
+ */
+static int match_traverse_continue(Process* p, DbTableHash* tb,
+ Sint chunk_size, /* If 0, no chunking */
+ Sint iterations_left, /* Nr. of iterations left */
+ Eterm** hpp, /* Heap */
+ Sint slot_ix, /* Slot index to resume traversal from */
+ Sint got, /* Matched terms counter */
+ Binary** mpp, /* Existing match program */
+ int lock_for_write, /* Set to 1 if we're going to delete or
+ modify existing terms */
+ mtraversal_on_match_res_t on_match_res,
+ mtraversal_on_loop_ended_t on_loop_ended,
+ mtraversal_on_trap_t on_trap,
+ void* context_ptr, /* For callbacks */
+ Eterm* ret)
+{
+ int all_objects = (*mpp)->flags & BIN_FLAG_ALL_OBJECTS;
+ HashDbTerm** current_ptr; /* Refers to either the bucket pointer or
+ * the 'next' pointer in the previous term
+ */
+ HashDbTerm* saved_current; /* Helper to avoid double skip on match */
Eterm match_res;
- Eterm *hp;
- int num_left = 1000;
- Uint got = 0;
- Eterm continuation;
- int errcode;
- Eterm mpb;
erts_smp_rwmtx_t* lck;
+ int ret_value;
+#ifdef ERTS_SMP
+ erts_smp_rwmtx_t* (*lock_hash_function)(DbTableHash*, HashValue)
+ = (lock_for_write ? WLOCK_HASH : RLOCK_HASH);
+ void (*unlock_hash_function)(erts_smp_rwmtx_t*)
+ = (lock_for_write ? WUNLOCK_HASH : RUNLOCK_HASH);
+#else
+ #define lock_hash_function(tb, hval) NULL
+ #define unlock_hash_function(lck) ((void)lck)
+#endif
+ Sint (*next_slot_function)(DbTableHash* tb, Uint ix, erts_smp_rwmtx_t** lck_ptr)
+ = (lock_for_write ? next_slot_w : next_slot);
-
-#define RET_TO_BIF(Term,RetVal) do { \
- if (mpi.mp != NULL) { \
- erts_bin_free(mpi.mp); \
- } \
- if (mpi.lists != mpi.dlists) { \
- erts_free(ERTS_ALC_T_DB_SEL_LIST, \
- (void *) mpi.lists); \
- } \
- *ret = (Term); \
- return RetVal; \
- } while(0)
-
-
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
- RET_TO_BIF(NIL,errcode);
+ if (got < 0) {
+ *ret = NIL;
+ return DB_ERROR_BADPARAM;
}
- if (!mpi.something_can_match) {
- if (chunk_size) {
- RET_TO_BIF(am_EOT, DB_ERROR_NONE); /* We're done */
- }
- RET_TO_BIF(NIL, DB_ERROR_NONE);
- /* can't possibly match anything */
+ if (slot_ix < 0 /* EOT */
+ || (chunk_size && got >= chunk_size))
+ {
+ /* Already got all or enough in the match_list */
+ ret_value = on_loop_ended(context_ptr, slot_ix, got, iterations_left, mpp, ret);
+ goto done;
}
- if (!mpi.key_given) {
- /* Run this code if pattern is variable or GETKEY(pattern) */
- /* is a variable */
- slot_ix = 0;
- lck = RLOCK_HASH(tb,slot_ix);
- for (;;) {
- ASSERT(slot_ix < NACTIVE(tb));
- if ((current = BUCKET(tb,slot_ix)) != NULL) {
- break;
- }
- slot_ix = next_slot(tb,slot_ix,&lck);
- if (slot_ix == 0) {
- if (chunk_size) {
- RET_TO_BIF(am_EOT, DB_ERROR_NONE); /* We're done */
- }
- RET_TO_BIF(NIL,DB_ERROR_NONE);
- }
- }
- } else {
- /* We have at least one */
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = RLOCK_HASH(tb, slot_ix);
- current = *(mpi.lists[current_list_pos].bucket);
- ASSERT(current == BUCKET(tb,slot_ix));
- ++current_list_pos;
+ lck = lock_hash_function(tb, slot_ix);
+ if (slot_ix >= NACTIVE(tb)) { /* Is this possible? */
+ unlock_hash_function(lck);
+ *ret = NIL;
+ ret_value = DB_ERROR_BADPARAM;
+ goto done;
}
- match_list = NIL;
-
+ /*
+ * Resume traversal cycle from where we left
+ */
+ current_ptr = &BUCKET(tb,slot_ix);
for(;;) {
- if (current != NULL) {
- if (current->hvalue != INVALID_HASH) {
- match_res = db_match_dbterm(&tb->common, p, mpi.mp, 0,
- &current->dbterm, &hp, 2);
- if (is_value(match_res)) {
- match_list = CONS(hp, match_res, match_list);
- ++got;
- }
- --num_left;
- }
- current = current->next;
- }
- else if (mpi.key_given) { /* Key is bound */
- RUNLOCK_HASH(lck);
- if (current_list_pos == mpi.num_lists) {
- slot_ix = -1; /* EOT */
- goto done;
- } else {
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = RLOCK_HASH(tb, slot_ix);
- current = *(mpi.lists[current_list_pos].bucket);
- ASSERT(mpi.lists[current_list_pos].bucket == &BUCKET(tb,slot_ix));
- ++current_list_pos;
- }
- }
- else { /* Key is variable */
-
- if ((slot_ix=next_slot(tb,slot_ix,&lck)) == 0) {
- slot_ix = -1;
- break;
- }
- if (chunk_size && got >= chunk_size) {
- RUNLOCK_HASH(lck);
- break;
- }
- if (num_left <= 0 || MBUF(p)) {
- /*
- * We have either reached our limit, or just created some heap fragments.
- * Since many heap fragments will make the GC slower, trap and GC now.
- */
- RUNLOCK_HASH(lck);
- goto trap;
- }
- current = BUCKET(tb,slot_ix);
+ if (*current_ptr != NULL) {
+ if ((*current_ptr)->hvalue != INVALID_HASH) {
+ match_res = db_match_dbterm(&tb->common, p, *mpp, all_objects,
+ &(*current_ptr)->dbterm, hpp, 2);
+ saved_current = *current_ptr;
+ if (on_match_res(context_ptr, slot_ix, &current_ptr, match_res)) {
+ ++got;
+ }
+ --iterations_left;
+ if (*current_ptr != saved_current) {
+ /* Don't advance to next, the callback did it already */
+ continue;
+ }
+ }
+ current_ptr = &((*current_ptr)->next);
+ }
+ else {
+ if ((slot_ix=next_slot_function(tb,slot_ix,&lck)) == 0) {
+ slot_ix = -1;
+ break;
+ }
+ if (chunk_size && got >= chunk_size) {
+ unlock_hash_function(lck);
+ break;
+ }
+ if (iterations_left <= 0 || MBUF(p)) {
+ /*
+ * We have either reached our limit, or just created some heap fragments.
+ * Since many heap fragments will make the GC slower, trap and GC now.
+ */
+ unlock_hash_function(lck);
+ ret_value = on_trap(context_ptr, slot_ix, got, mpp, ret);
+ goto done;
+ }
+ current_ptr = &BUCKET(tb,slot_ix);
}
}
-done:
- BUMP_REDS(p, 1000 - num_left);
- if (chunk_size) {
- Eterm continuation;
- Eterm rest = NIL;
- Sint rest_size = 0;
-
- if (mpi.all_objects)
- (mpi.mp)->flags |= BIN_FLAG_ALL_OBJECTS;
- if (got > chunk_size) { /* Split list in return value and 'rest' */
- Eterm tmp = match_list;
- rest = match_list;
- while (got-- > chunk_size + 1) {
- tmp = CDR(list_val(tmp));
- ++rest_size;
- }
- ++rest_size;
- match_list = CDR(list_val(tmp));
- CDR(list_val(tmp)) = NIL; /* Destructive, the list has never
- been in 'user space' */
- }
- if (rest != NIL || slot_ix >= 0) { /* Need more calls */
- hp = HAlloc(p,3+7+ERTS_MAGIC_REF_THING_SIZE);
- mpb = erts_db_make_match_prog_ref(p,(mpi.mp),&hp);
- if (mpi.all_objects)
- (mpi.mp)->flags |= BIN_FLAG_ALL_OBJECTS;
- continuation = TUPLE6(hp, tb->common.id,make_small(slot_ix),
- make_small(chunk_size),
- mpb, rest,
- make_small(rest_size));
- mpi.mp = NULL; /*otherwise the return macro will destroy it */
- hp += 7;
- RET_TO_BIF(TUPLE2(hp, match_list, continuation),DB_ERROR_NONE);
- } else { /* All data is exhausted */
- if (match_list != NIL) { /* No more data to search but still a
- result to return to the caller */
- hp = HAlloc(p, 3);
- RET_TO_BIF(TUPLE2(hp, match_list, am_EOT),DB_ERROR_NONE);
- } else { /* Reached the end of the ttable with no data to return */
- RET_TO_BIF(am_EOT, DB_ERROR_NONE);
- }
- }
- }
- RET_TO_BIF(match_list,DB_ERROR_NONE);
-trap:
- BUMP_ALL_REDS(p);
- if (mpi.all_objects)
- (mpi.mp)->flags |= BIN_FLAG_ALL_OBJECTS;
- hp = HAlloc(p,7+ERTS_MAGIC_REF_THING_SIZE);
- mpb = erts_db_make_match_prog_ref(p,(mpi.mp),&hp);
- continuation = TUPLE6(hp, tb->common.id, make_small(slot_ix),
- make_small(chunk_size),
- mpb, match_list,
- make_small(got));
- mpi.mp = NULL; /*otherwise the return macro will destroy it */
- RET_TO_BIF(bif_trap1(&ets_select_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
-#undef RET_TO_BIF
-
-}
+ ret_value = on_loop_ended(context_ptr, slot_ix, got, iterations_left, mpp, ret);
-static int db_select_count_hash(Process *p,
- DbTable *tbl,
- Eterm pattern,
- Eterm *ret)
-{
- DbTableHash *tb = &tbl->hash;
- struct mp_info mpi;
- Uint slot_ix = 0;
- HashDbTerm* current = NULL;
- unsigned current_list_pos = 0;
- Eterm *hp;
- int num_left = 1000;
- Uint got = 0;
- Eterm continuation;
- int errcode;
- Eterm egot;
- Eterm mpb;
- erts_smp_rwmtx_t* lck;
-
-#define RET_TO_BIF(Term,RetVal) do { \
- if (mpi.mp != NULL) { \
- erts_bin_free(mpi.mp); \
- } \
- if (mpi.lists != mpi.dlists) { \
- erts_free(ERTS_ALC_T_DB_SEL_LIST, \
- (void *) mpi.lists); \
- } \
- *ret = (Term); \
- return RetVal; \
- } while(0)
+done:
+ /* We should only jump directly to this label if
+ * we've already called on_loop_ended / on_trap
+ */
+ return ret_value;
+#ifndef SMP
+#undef lock_hash_function
+#undef unlock_hash_function
+#endif
+}
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
- RET_TO_BIF(NIL,errcode);
- }
- if (!mpi.something_can_match) {
- RET_TO_BIF(make_small(0), DB_ERROR_NONE);
- /* can't possibly match anything */
- }
+/*
+ * Common traversal trapping/continuation code;
+ * used by select_count, select_delete and select_replace,
+ * as well as their continuation-handling counterparts.
+ */
- if (!mpi.key_given) {
- /* Run this code if pattern is variable or GETKEY(pattern) */
- /* is a variable */
- slot_ix = 0;
- lck = RLOCK_HASH(tb,slot_ix);
- current = BUCKET(tb,slot_ix);
- } else {
- /* We have at least one */
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = RLOCK_HASH(tb, slot_ix);
- current = *(mpi.lists[current_list_pos].bucket);
- ASSERT(current == BUCKET(tb,slot_ix));
- ++current_list_pos;
- }
+static ERTS_INLINE int on_mtraversal_simple_trap(Export* trap_function,
+ Process* p,
+ DbTableHash* tb,
+ Eterm tid,
+ Eterm* prev_continuation_tptr,
+ Sint slot_ix,
+ Sint got,
+ Binary** mpp,
+ Eterm* ret)
+{
+ Eterm* hp;
+ Eterm egot;
+ Eterm mpb;
+ Eterm continuation;
+ int is_first_trap = (prev_continuation_tptr == NULL);
+ size_t base_halloc_sz = (is_first_trap ? ERTS_MAGIC_REF_THING_SIZE : 0);
- for(;;) {
- if (current != NULL) {
- if (current->hvalue != INVALID_HASH) {
- if (db_match_dbterm(&tb->common, p, mpi.mp, 0,
- &current->dbterm, NULL,0) == am_true) {
- ++got;
- }
- --num_left;
- }
- current = current->next;
- }
- else { /* next bucket */
- if (mpi.key_given) { /* Key is bound */
- RUNLOCK_HASH(lck);
- if (current_list_pos == mpi.num_lists) {
- goto done;
- } else {
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = RLOCK_HASH(tb, slot_ix);
- current = *(mpi.lists[current_list_pos].bucket);
- ASSERT(mpi.lists[current_list_pos].bucket == &BUCKET(tb,slot_ix));
- ++current_list_pos;
- }
- }
- else {
- if ((slot_ix=next_slot(tb,slot_ix,&lck)) == 0) {
- goto done;
- }
- if (num_left <= 0) {
- RUNLOCK_HASH(lck);
- goto trap;
- }
- current = BUCKET(tb,slot_ix);
- }
- }
- }
-done:
- BUMP_REDS(p, 1000 - num_left);
- RET_TO_BIF(erts_make_integer(got,p),DB_ERROR_NONE);
-trap:
BUMP_ALL_REDS(p);
if (IS_USMALL(0, got)) {
- hp = HAlloc(p, ERTS_MAGIC_REF_THING_SIZE + 5);
+ hp = HAlloc(p, base_halloc_sz + 5);
egot = make_small(got);
}
else {
- hp = HAlloc(p, BIG_UINT_HEAP_SIZE + ERTS_MAGIC_REF_THING_SIZE + 5);
+ hp = HAlloc(p, base_halloc_sz + BIG_UINT_HEAP_SIZE + 5);
egot = uint_to_big(got, hp);
hp += BIG_UINT_HEAP_SIZE;
}
- mpb = erts_db_make_match_prog_ref(p,mpi.mp,&hp);
- continuation = TUPLE4(hp, tb->common.id, make_small(slot_ix),
- mpb,
- egot);
- mpi.mp = NULL; /*otherwise the return macro will destroy it */
- RET_TO_BIF(bif_trap1(&ets_select_count_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
-#undef RET_TO_BIF
+ if (is_first_trap) {
+ mpb = erts_db_make_match_prog_ref(p, *mpp, &hp);
+ *mpp = NULL; /* otherwise the caller will destroy it */
+ }
+ else {
+ mpb = prev_continuation_tptr[3];
+ }
+
+ continuation = TUPLE4(
+ hp,
+ tid,
+ make_small(slot_ix),
+ mpb,
+ egot);
+ *ret = bif_trap1(trap_function, p, continuation);
+ return DB_ERROR_NONE;
}
-static int db_select_delete_hash(Process *p,
- DbTable *tbl,
- Eterm pattern,
- Eterm *ret)
+static ERTS_INLINE int unpack_simple_mtraversal_continuation(Eterm continuation,
+ Eterm** tptr_ptr,
+ Eterm* tid_ptr,
+ Sint* slot_ix_p,
+ Binary** mpp,
+ Sint* got_p)
{
- DbTableHash *tb = &tbl->hash;
- struct mp_info mpi;
- Uint slot_ix = 0;
- HashDbTerm **current = NULL;
- unsigned current_list_pos = 0;
- Eterm *hp;
- int num_left = 1000;
- Uint got = 0;
- Eterm continuation;
- int errcode;
- Uint last_pseudo_delete = (Uint)-1;
- Eterm mpb;
- Eterm egot;
-#ifdef ERTS_SMP
- erts_aint_t fixated_by_me = tb->common.is_thread_safe ? 0 : 1; /* ToDo: something nicer */
-#else
- erts_aint_t fixated_by_me = 0;
-#endif
- erts_smp_rwmtx_t* lck;
-
-#define RET_TO_BIF(Term,RetVal) do { \
- if (mpi.mp != NULL) { \
- erts_bin_free(mpi.mp); \
- } \
- if (mpi.lists != mpi.dlists) { \
- erts_free(ERTS_ALC_T_DB_SEL_LIST, \
- (void *) mpi.lists); \
- } \
- *ret = (Term); \
- return RetVal; \
- } while(0)
-
+ Eterm* tptr;
+ ASSERT(is_tuple(continuation));
+ tptr = tuple_val(continuation);
+ if (arityval(*tptr) != 4)
+ return 1;
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
- RET_TO_BIF(NIL,errcode);
+ if (! is_small(tptr[2]) || !(is_big(tptr[4]) || is_small(tptr[4]))) {
+ return 1;
}
- if (!mpi.something_can_match) {
- RET_TO_BIF(make_small(0), DB_ERROR_NONE);
- /* can't possibly match anything */
+ *tptr_ptr = tptr;
+ *tid_ptr = tptr[1];
+ *slot_ix_p = unsigned_val(tptr[2]);
+ *mpp = erts_db_get_match_prog_binary_unchecked(tptr[3]);
+ if (is_big(tptr[4])) {
+ *got_p = big_to_uint32(tptr[4]);
+ }
+ else {
+ *got_p = unsigned_val(tptr[4]);
}
+ return 0;
+}
- if (!mpi.key_given) {
- /* Run this code if pattern is variable or GETKEY(pattern) */
- /* is a variable */
- lck = WLOCK_HASH(tb,slot_ix);
- current = &BUCKET(tb,slot_ix);
- } else {
- /* We have at least one */
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = WLOCK_HASH(tb, slot_ix);
- current = mpi.lists[current_list_pos++].bucket;
- ASSERT(*current == BUCKET(tb,slot_ix));
+
+/*
+ *
+ * select / select_chunk match traversal
+ *
+ */
+
+#define MAX_SELECT_CHUNK_ITERATIONS 1000
+
+typedef struct {
+ Process* p;
+ DbTableHash* tb;
+ Eterm tid;
+ Eterm* hp;
+ Sint chunk_size;
+ Eterm match_list;
+ Eterm* prev_continuation_tptr;
+} mtraversal_select_chunk_context_t;
+
+static int mtraversal_select_chunk_on_nothing_can_match(void* context_ptr, Eterm* ret) {
+ mtraversal_select_chunk_context_t* sc_context_ptr = (mtraversal_select_chunk_context_t*) context_ptr;
+ *ret = (sc_context_ptr->chunk_size > 0 ? am_EOT : NIL);
+ return DB_ERROR_NONE;
+}
+
+static int mtraversal_select_chunk_on_match_res(void* context_ptr, Sint slot_ix,
+ HashDbTerm*** current_ptr_ptr,
+ Eterm match_res)
+{
+ mtraversal_select_chunk_context_t* sc_context_ptr = (mtraversal_select_chunk_context_t*) context_ptr;
+ if (is_value(match_res)) {
+ sc_context_ptr->match_list = CONS(sc_context_ptr->hp, match_res, sc_context_ptr->match_list);
+ return 1;
}
+ return 0;
+}
+static int mtraversal_select_chunk_on_loop_ended(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_chunk_context_t* sc_context_ptr = (mtraversal_select_chunk_context_t*) context_ptr;
+ Eterm mpb;
- for(;;) {
- if ((*current) == NULL) {
- if (mpi.key_given) { /* Key is bound */
- WUNLOCK_HASH(lck);
- if (current_list_pos == mpi.num_lists) {
- goto done;
- } else {
- slot_ix = mpi.lists[current_list_pos].ix;
- lck = WLOCK_HASH(tb, slot_ix);
- current = mpi.lists[current_list_pos].bucket;
- ASSERT(mpi.lists[current_list_pos].bucket == &BUCKET(tb,slot_ix));
- ++current_list_pos;
- }
- } else {
- if ((slot_ix=next_slot_w(tb,slot_ix,&lck)) == 0) {
- goto done;
- }
- if (num_left <= 0) {
- WUNLOCK_HASH(lck);
- goto trap;
- }
- current = &BUCKET(tb,slot_ix);
- }
- }
- else if ((*current)->hvalue == INVALID_HASH) {
- current = &((*current)->next);
- }
- else {
- int did_erase = 0;
- if (db_match_dbterm(&tb->common, p, mpi.mp, 0,
- &(*current)->dbterm, NULL, 0) == am_true) {
- HashDbTerm *del;
- if (NFIXED(tb) > fixated_by_me) { /* fixated by others? */
- if (slot_ix != last_pseudo_delete) {
- if (!add_fixed_deletion(tb, slot_ix, fixated_by_me))
- goto do_erase;
- last_pseudo_delete = slot_ix;
- }
- (*current)->hvalue = INVALID_HASH;
- } else {
- do_erase:
- del = *current;
- *current = (*current)->next;
- free_term(tb, del);
- did_erase = 1;
- }
- erts_smp_atomic_dec_nob(&tb->common.nitems);
- ++got;
- }
- --num_left;
- if (!did_erase) {
- current = &((*current)->next);
- }
- }
+ if (iterations_left == MAX_SELECT_CHUNK_ITERATIONS) {
+ /* We didn't get to iterate a single time, which means EOT */
+ ASSERT(sc_context_ptr->match_list == NIL);
+ *ret = (sc_context_ptr->chunk_size > 0 ? am_EOT : NIL);
+ return DB_ERROR_NONE;
}
-done:
- BUMP_REDS(p, 1000 - num_left);
- if (got) {
- try_shrink(tb);
+ else {
+ ASSERT(iterations_left < MAX_SELECT_CHUNK_ITERATIONS);
+ BUMP_REDS(sc_context_ptr->p, MAX_SELECT_CHUNK_ITERATIONS - iterations_left);
+ if (sc_context_ptr->chunk_size) {
+ Eterm continuation;
+ Eterm rest = NIL;
+ Sint rest_size = 0;
+
+ if (got > sc_context_ptr->chunk_size) { /* Split list in return value and 'rest' */
+ Eterm tmp = sc_context_ptr->match_list;
+ rest = sc_context_ptr->match_list;
+ while (got-- > sc_context_ptr->chunk_size + 1) {
+ tmp = CDR(list_val(tmp));
+ ++rest_size;
+ }
+ ++rest_size;
+ sc_context_ptr->match_list = CDR(list_val(tmp));
+ CDR(list_val(tmp)) = NIL; /* Destructive, the list has never
+ been in 'user space' */
+ }
+ if (rest != NIL || slot_ix >= 0) { /* Need more calls */
+ sc_context_ptr->hp = HAlloc(sc_context_ptr->p, 3 + 7 + ERTS_MAGIC_REF_THING_SIZE);
+ mpb = erts_db_make_match_prog_ref(sc_context_ptr->p, *mpp, &sc_context_ptr->hp);
+ continuation = TUPLE6(
+ sc_context_ptr->hp,
+ sc_context_ptr->tid,
+ make_small(slot_ix),
+ make_small(sc_context_ptr->chunk_size),
+ mpb, rest,
+ make_small(rest_size));
+ *mpp = NULL; /* Otherwise the caller will destroy it */
+ sc_context_ptr->hp += 7;
+ *ret = TUPLE2(sc_context_ptr->hp, sc_context_ptr->match_list, continuation);
+ return DB_ERROR_NONE;
+ } else { /* All data is exhausted */
+ if (sc_context_ptr->match_list != NIL) { /* No more data to search but still a
+ result to return to the caller */
+ sc_context_ptr->hp = HAlloc(sc_context_ptr->p, 3);
+ *ret = TUPLE2(sc_context_ptr->hp, sc_context_ptr->match_list, am_EOT);
+ return DB_ERROR_NONE;
+ } else { /* Reached the end of the ttable with no data to return */
+ *ret = am_EOT;
+ return DB_ERROR_NONE;
+ }
+ }
+ }
+ *ret = sc_context_ptr->match_list;
+ return DB_ERROR_NONE;
}
- RET_TO_BIF(erts_make_integer(got,p),DB_ERROR_NONE);
-trap:
- BUMP_ALL_REDS(p);
- if (IS_USMALL(0, got)) {
- hp = HAlloc(p, ERTS_MAGIC_REF_THING_SIZE + 5);
- egot = make_small(got);
+}
+
+static int mtraversal_select_chunk_on_trap(void* context_ptr, Sint slot_ix, Sint got,
+ Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_chunk_context_t* sc_context_ptr = (mtraversal_select_chunk_context_t*) context_ptr;
+ Eterm mpb;
+ Eterm continuation;
+ Eterm* hp;
+
+ BUMP_ALL_REDS(sc_context_ptr->p);
+
+ if (sc_context_ptr->prev_continuation_tptr == NULL) {
+ /* First time we're trapping */
+ hp = HAlloc(sc_context_ptr->p, 7 + ERTS_MAGIC_REF_THING_SIZE);
+ mpb = erts_db_make_match_prog_ref(sc_context_ptr->p, *mpp, &hp);
+ continuation = TUPLE6(
+ hp,
+ sc_context_ptr->tid,
+ make_small(slot_ix),
+ make_small(sc_context_ptr->chunk_size),
+ mpb,
+ sc_context_ptr->match_list,
+ make_small(got));
+ *mpp = NULL; /* otherwise the caller will destroy it */
}
else {
- hp = HAlloc(p, BIG_UINT_HEAP_SIZE + ERTS_MAGIC_REF_THING_SIZE + 5);
- egot = uint_to_big(got, hp);
- hp += BIG_UINT_HEAP_SIZE;
- }
- mpb = erts_db_make_match_prog_ref(p,mpi.mp,&hp);
- continuation = TUPLE4(hp, tb->common.id, make_small(slot_ix),
- mpb,
- egot);
- mpi.mp = NULL; /*otherwise the return macro will destroy it */
- RET_TO_BIF(bif_trap1(&ets_select_delete_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
+ /* Not the first time we're trapping; reuse continuation terms */
+ hp = HAlloc(sc_context_ptr->p, 7);
+ continuation = TUPLE6(
+ hp,
+ sc_context_ptr->prev_continuation_tptr[1],
+ make_small(slot_ix),
+ sc_context_ptr->prev_continuation_tptr[3],
+ sc_context_ptr->prev_continuation_tptr[4],
+ sc_context_ptr->match_list,
+ make_small(got));
+ }
+ *ret = bif_trap1(&ets_select_continue_exp, sc_context_ptr->p, continuation);
+ return DB_ERROR_NONE;
+}
-#undef RET_TO_BIF
+static int db_select_hash(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, int reverse, Eterm *ret) {
+ return db_select_chunk_hash(p, tbl, tid, pattern, 0, reverse, ret);
+}
+static int db_select_chunk_hash(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Sint chunk_size,
+ int reverse, Eterm *ret)
+{
+ mtraversal_select_chunk_context_t sc_context;
+ sc_context.p = p;
+ sc_context.tb = &tbl->hash;
+ sc_context.tid = tid;
+ sc_context.hp = NULL;
+ sc_context.chunk_size = chunk_size;
+ sc_context.match_list = NIL;
+ sc_context.prev_continuation_tptr = NULL;
+
+ return match_traverse(
+ sc_context.p, sc_context.tb,
+ pattern, NULL,
+ sc_context.chunk_size,
+ MAX_SELECT_CHUNK_ITERATIONS,
+ &sc_context.hp, 0,
+ mtraversal_select_chunk_on_nothing_can_match,
+ mtraversal_select_chunk_on_match_res,
+ mtraversal_select_chunk_on_loop_ended,
+ mtraversal_select_chunk_on_trap,
+ &sc_context, ret);
}
+
/*
-** This is called when select_delete traps
-*/
-static int db_select_delete_continue_hash(Process *p,
- DbTable *tbl,
- Eterm continuation,
- Eterm *ret)
+ *
+ * select_continue match traversal
+ *
+ */
+
+static int mtraversal_select_chunk_continue_on_loop_ended(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret)
{
- DbTableHash *tb = &tbl->hash;
- Uint slot_ix;
- Uint last_pseudo_delete = (Uint)-1;
- HashDbTerm **current = NULL;
- Eterm *hp;
- int num_left = 1000;
- Uint got;
- Eterm *tptr;
- Binary *mp;
- Eterm egot;
- int fixated_by_me = ONLY_WRITER(p,tb) ? 0 : 1; /* ToDo: something nicer */
- erts_smp_rwmtx_t* lck;
+ mtraversal_select_chunk_context_t* sc_context_ptr = (mtraversal_select_chunk_context_t*) context_ptr;
+ Eterm continuation;
+ Eterm rest = NIL;
+ Eterm* hp;
+
+ ASSERT(iterations_left <= MAX_SELECT_CHUNK_ITERATIONS);
+ BUMP_REDS(sc_context_ptr->p, MAX_SELECT_CHUNK_ITERATIONS - iterations_left);
+ if (sc_context_ptr->chunk_size) {
+ Sint rest_size = 0;
+ if (got > sc_context_ptr->chunk_size) {
+ /* Cannot write destructively here,
+ the list may have
+ been in user space */
+ hp = HAlloc(sc_context_ptr->p, (got - sc_context_ptr->chunk_size) * 2);
+ while (got-- > sc_context_ptr->chunk_size) {
+ rest = CONS(hp, CAR(list_val(sc_context_ptr->match_list)), rest);
+ hp += 2;
+ sc_context_ptr->match_list = CDR(list_val(sc_context_ptr->match_list));
+ ++rest_size;
+ }
+ }
+ if (rest != NIL || slot_ix >= 0) {
+ hp = HAlloc(sc_context_ptr->p, 3 + 7);
+ continuation = TUPLE6(
+ hp,
+ sc_context_ptr->prev_continuation_tptr[1],
+ make_small(slot_ix),
+ sc_context_ptr->prev_continuation_tptr[3],
+ sc_context_ptr->prev_continuation_tptr[4],
+ rest,
+ make_small(rest_size));
+ hp += 7;
+ *ret = TUPLE2(hp, sc_context_ptr->match_list, continuation);
+ return DB_ERROR_NONE;
+ } else {
+ if (sc_context_ptr->match_list != NIL) {
+ hp = HAlloc(sc_context_ptr->p, 3);
+ *ret = TUPLE2(hp, sc_context_ptr->match_list, am_EOT);
+ return DB_ERROR_NONE;
+ } else {
+ *ret = am_EOT;
+ return DB_ERROR_NONE;
+ }
+ }
+ }
+ *ret = sc_context_ptr->match_list;
+ return DB_ERROR_NONE;
+}
-#define RET_TO_BIF(Term,RetVal) do { \
- *ret = (Term); \
- return RetVal; \
- } while(0)
+/*
+ * This is called when select traps
+ */
+static int db_select_continue_hash(Process* p, DbTable* tbl, Eterm continuation, Eterm* ret) {
+ mtraversal_select_chunk_context_t sc_context = {0};
+ Eterm* tptr;
+ Eterm tid;
+ Binary* mp;
+ Sint got;
+ Sint slot_ix;
+ Sint chunk_size;
+ Eterm match_list;
+ Sint iterations_left = MAX_SELECT_CHUNK_ITERATIONS;
-
+ /* Decode continuation. We know it's a tuple but not the arity or anything else */
+ ASSERT(is_tuple(continuation));
tptr = tuple_val(continuation);
- slot_ix = unsigned_val(tptr[2]);
- mp = erts_db_get_match_prog_binary_unchecked(tptr[3]);
- if (is_big(tptr[4])) {
- got = big_to_uint32(tptr[4]);
- } else {
- got = unsigned_val(tptr[4]);
- }
-
- lck = WLOCK_HASH(tb,slot_ix);
- if (slot_ix >= NACTIVE(tb)) {
- WUNLOCK_HASH(lck);
- goto done;
- }
- current = &BUCKET(tb,slot_ix);
- for(;;) {
- if ((*current) == NULL) {
- if ((slot_ix=next_slot_w(tb,slot_ix,&lck)) == 0) {
- goto done;
- }
- if (num_left <= 0) {
- WUNLOCK_HASH(lck);
- goto trap;
- }
- current = &BUCKET(tb,slot_ix);
- }
- else if ((*current)->hvalue == INVALID_HASH) {
- current = &((*current)->next);
- }
- else {
- int did_erase = 0;
- if (db_match_dbterm(&tb->common, p, mp, 0,
- &(*current)->dbterm, NULL, 0) == am_true) {
- HashDbTerm *del;
- if (NFIXED(tb) > fixated_by_me) { /* fixated by others? */
- if (slot_ix != last_pseudo_delete) {
- if (!add_fixed_deletion(tb, slot_ix, fixated_by_me))
- goto do_erase;
- last_pseudo_delete = slot_ix;
- }
- (*current)->hvalue = INVALID_HASH;
- } else {
- do_erase:
- del = *current;
- *current = (*current)->next;
- free_term(tb, del);
- did_erase = 1;
- }
- erts_smp_atomic_dec_nob(&tb->common.nitems);
- ++got;
- }
-
- --num_left;
- if (!did_erase) {
- current = &((*current)->next);
- }
- }
- }
-done:
- BUMP_REDS(p, 1000 - num_left);
- if (got) {
- try_shrink(tb);
- }
- RET_TO_BIF(erts_make_integer(got,p),DB_ERROR_NONE);
-trap:
- BUMP_ALL_REDS(p);
- if (IS_USMALL(0, got)) {
- hp = HAlloc(p, 5);
- egot = make_small(got);
- }
- else {
- hp = HAlloc(p, BIG_UINT_HEAP_SIZE + 5);
- egot = uint_to_big(got, hp);
- hp += BIG_UINT_HEAP_SIZE;
- }
- continuation = TUPLE4(hp, tb->common.id, make_small(slot_ix),
- tptr[3],
- egot);
- RET_TO_BIF(bif_trap1(&ets_select_delete_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
+ if (arityval(*tptr) != 6)
+ goto badparam;
+
+ if (!is_small(tptr[2]) || !is_small(tptr[3]) ||
+ !(is_list(tptr[5]) || tptr[5] == NIL) || !is_small(tptr[6]))
+ goto badparam;
+ if ((chunk_size = signed_val(tptr[3])) < 0)
+ goto badparam;
+
+ mp = erts_db_get_match_prog_binary(tptr[4]);
+ if (mp == NULL)
+ goto badparam;
+
+ if ((got = signed_val(tptr[6])) < 0)
+ goto badparam;
-#undef RET_TO_BIF
+ tid = tptr[1];
+ slot_ix = signed_val(tptr[2]);
+ match_list = tptr[5];
+ /* Proceed */
+ sc_context.p = p;
+ sc_context.tb = &tbl->hash;
+ sc_context.tid = tid;
+ sc_context.hp = NULL;
+ sc_context.chunk_size = chunk_size;
+ sc_context.match_list = match_list;
+ sc_context.prev_continuation_tptr = tptr;
+
+ return match_traverse_continue(
+ sc_context.p, sc_context.tb, sc_context.chunk_size,
+ iterations_left, &sc_context.hp, slot_ix, got, &mp, 0,
+ mtraversal_select_chunk_on_match_res, /* Reuse callback */
+ mtraversal_select_chunk_continue_on_loop_ended,
+ mtraversal_select_chunk_on_trap, /* Reuse callback */
+ &sc_context, ret);
+
+badparam:
+ *ret = NIL;
+ return DB_ERROR_BADPARAM;
}
-
+
+#undef MAX_SELECT_CHUNK_ITERATIONS
+
+
/*
-** This is called when select_count traps
-*/
-static int db_select_count_continue_hash(Process *p,
- DbTable *tbl,
- Eterm continuation,
- Eterm *ret)
+ *
+ * select_count match traversal
+ *
+ */
+
+#define MAX_SELECT_COUNT_ITERATIONS 1000
+
+typedef struct {
+ Process* p;
+ DbTableHash* tb;
+ Eterm tid;
+ Eterm* hp;
+ Eterm* prev_continuation_tptr;
+} mtraversal_select_count_context_t;
+
+static int mtraversal_select_count_on_nothing_can_match(void* context_ptr, Eterm* ret) {
+ *ret = make_small(0);
+ return DB_ERROR_NONE;
+}
+
+static int mtraversal_select_count_on_match_res(void* context_ptr, Sint slot_ix,
+ HashDbTerm*** current_ptr_ptr,
+ Eterm match_res)
{
- DbTableHash *tb = &tbl->hash;
- Uint slot_ix;
- HashDbTerm* current;
- Eterm *hp;
- int num_left = 1000;
- Uint got;
- Eterm *tptr;
- Binary *mp;
- Eterm egot;
- erts_smp_rwmtx_t* lck;
+ return (match_res == am_true);
+}
-#define RET_TO_BIF(Term,RetVal) do { \
- *ret = (Term); \
- return RetVal; \
- } while(0)
+static int mtraversal_select_count_on_loop_ended(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_count_context_t* scnt_context_ptr = (mtraversal_select_count_context_t*) context_ptr;
+ ASSERT(iterations_left <= MAX_SELECT_COUNT_ITERATIONS);
+ BUMP_REDS(scnt_context_ptr->p, MAX_SELECT_COUNT_ITERATIONS - iterations_left);
+ *ret = erts_make_integer(got, scnt_context_ptr->p);
+ return DB_ERROR_NONE;
+}
-
- tptr = tuple_val(continuation);
- slot_ix = unsigned_val(tptr[2]);
- mp = erts_db_get_match_prog_binary_unchecked(tptr[3]);
- if (is_big(tptr[4])) {
- got = big_to_uint32(tptr[4]);
- } else {
- got = unsigned_val(tptr[4]);
- }
-
+static int mtraversal_select_count_on_trap(void* context_ptr, Sint slot_ix, Sint got,
+ Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_count_context_t* scnt_context_ptr = (mtraversal_select_count_context_t*) context_ptr;
+ return on_mtraversal_simple_trap(
+ &ets_select_count_continue_exp,
+ scnt_context_ptr->p,
+ scnt_context_ptr->tb,
+ scnt_context_ptr->tid,
+ scnt_context_ptr->prev_continuation_tptr,
+ slot_ix, got, mpp, ret);
+}
- lck = RLOCK_HASH(tb, slot_ix);
- if (slot_ix >= NACTIVE(tb)) { /* Is this posible? */
- RUNLOCK_HASH(lck);
- goto done;
- }
- current = BUCKET(tb,slot_ix);
-
- for(;;) {
- if (current != NULL) {
- if (current->hvalue == INVALID_HASH) {
- current = current->next;
- continue;
- }
- if (db_match_dbterm(&tb->common, p, mp, 0, &current->dbterm,
- NULL, 0) == am_true) {
- ++got;
- }
- --num_left;
- current = current->next;
- }
- else { /* next bucket */
- if ((slot_ix = next_slot(tb,slot_ix,&lck)) == 0) {
- goto done;
- }
- if (num_left <= 0) {
- RUNLOCK_HASH(lck);
- goto trap;
- }
- current = BUCKET(tb,slot_ix);
- }
- }
-done:
- BUMP_REDS(p, 1000 - num_left);
- RET_TO_BIF(erts_make_integer(got,p),DB_ERROR_NONE);
-trap:
- BUMP_ALL_REDS(p);
- if (IS_USMALL(0, got)) {
- hp = HAlloc(p, 5);
- egot = make_small(got);
+static int db_select_count_hash(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) {
+ mtraversal_select_count_context_t scnt_context = {0};
+ Sint iterations_left = MAX_SELECT_COUNT_ITERATIONS;
+ Sint chunk_size = 0;
+
+ scnt_context.p = p;
+ scnt_context.tb = &tbl->hash;
+ scnt_context.tid = tid;
+ scnt_context.hp = NULL;
+ scnt_context.prev_continuation_tptr = NULL;
+
+ return match_traverse(
+ scnt_context.p, scnt_context.tb,
+ pattern, NULL,
+ chunk_size, iterations_left, NULL, 0,
+ mtraversal_select_count_on_nothing_can_match,
+ mtraversal_select_count_on_match_res,
+ mtraversal_select_count_on_loop_ended,
+ mtraversal_select_count_on_trap,
+ &scnt_context, ret);
+}
+
+/*
+ * This is called when select_count traps
+ */
+static int db_select_count_continue_hash(Process* p, DbTable* tbl, Eterm continuation, Eterm* ret) {
+ mtraversal_select_count_context_t scnt_context = {0};
+ Eterm* tptr;
+ Eterm tid;
+ Binary* mp;
+ Sint got;
+ Sint slot_ix;
+ Sint chunk_size = 0;
+ *ret = NIL;
+
+ if (unpack_simple_mtraversal_continuation(continuation, &tptr, &tid, &slot_ix, &mp, &got)) {
+ *ret = NIL;
+ return DB_ERROR_BADPARAM;
+ }
+
+ scnt_context.p = p;
+ scnt_context.tb = &tbl->hash;
+ scnt_context.tid = tid;
+ scnt_context.hp = NULL;
+ scnt_context.prev_continuation_tptr = tptr;
+
+ return match_traverse_continue(
+ scnt_context.p, scnt_context.tb, chunk_size,
+ MAX_SELECT_COUNT_ITERATIONS,
+ NULL, slot_ix, got, &mp, 0,
+ mtraversal_select_count_on_match_res, /* Reuse callback */
+ mtraversal_select_count_on_loop_ended, /* Reuse callback */
+ mtraversal_select_count_on_trap, /* Reuse callback */
+ &scnt_context, ret);
+}
+
+#undef MAX_SELECT_COUNT_ITERATIONS
+
+
+/*
+ *
+ * select_delete match traversal
+ *
+ */
+
+#define MAX_SELECT_DELETE_ITERATIONS 1000
+
+typedef struct {
+ Process* p;
+ DbTableHash* tb;
+ Eterm tid;
+ Eterm* hp;
+ Eterm* prev_continuation_tptr;
+ erts_aint_t fixated_by_me;
+ Uint last_pseudo_delete;
+} mtraversal_select_delete_context_t;
+
+static int mtraversal_select_delete_on_nothing_can_match(void* context_ptr, Eterm* ret) {
+ *ret = make_small(0);
+ return DB_ERROR_NONE;
+}
+
+static int mtraversal_select_delete_on_match_res(void* context_ptr, Sint slot_ix,
+ HashDbTerm*** current_ptr_ptr,
+ Eterm match_res)
+{
+ HashDbTerm** current_ptr = *current_ptr_ptr;
+ mtraversal_select_delete_context_t* sd_context_ptr = (mtraversal_select_delete_context_t*) context_ptr;
+ HashDbTerm* del;
+ if (match_res != am_true)
+ return 0;
+
+ if (NFIXED(sd_context_ptr->tb) > sd_context_ptr->fixated_by_me) { /* fixated by others? */
+ if (slot_ix != sd_context_ptr->last_pseudo_delete) {
+ if (!add_fixed_deletion(sd_context_ptr->tb, slot_ix, sd_context_ptr->fixated_by_me))
+ goto do_erase;
+ sd_context_ptr->last_pseudo_delete = slot_ix;
+ }
+ (*current_ptr)->hvalue = INVALID_HASH;
}
else {
- hp = HAlloc(p, BIG_UINT_HEAP_SIZE + 5);
- egot = uint_to_big(got, hp);
- hp += BIG_UINT_HEAP_SIZE;
+ do_erase:
+ del = *current_ptr;
+ *current_ptr = (*current_ptr)->next; // replace pointer to term using next
+ free_term(sd_context_ptr->tb, del);
}
- continuation = TUPLE4(hp, tb->common.id, make_small(slot_ix),
- tptr[3],
- egot);
- RET_TO_BIF(bif_trap1(&ets_select_count_continue_exp, p,
- continuation),
- DB_ERROR_NONE);
+ erts_smp_atomic_dec_nob(&sd_context_ptr->tb->common.nitems);
-#undef RET_TO_BIF
+ return 1;
+}
+static int mtraversal_select_delete_on_loop_ended(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_delete_context_t* sd_context_ptr = (mtraversal_select_delete_context_t*) context_ptr;
+ ASSERT(iterations_left <= MAX_SELECT_DELETE_ITERATIONS);
+ BUMP_REDS(sd_context_ptr->p, MAX_SELECT_DELETE_ITERATIONS - iterations_left);
+ if (got) {
+ try_shrink(sd_context_ptr->tb);
+ }
+ *ret = erts_make_integer(got, sd_context_ptr->p);
+ return DB_ERROR_NONE;
}
-
+
+static int mtraversal_select_delete_on_trap(void* context_ptr, Sint slot_ix, Sint got,
+ Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_delete_context_t* sd_context_ptr = (mtraversal_select_delete_context_t*) context_ptr;
+ return on_mtraversal_simple_trap(
+ &ets_select_delete_continue_exp,
+ sd_context_ptr->p,
+ sd_context_ptr->tb,
+ sd_context_ptr->tid,
+ sd_context_ptr->prev_continuation_tptr,
+ slot_ix, got, mpp, ret);
+}
+
+static int db_select_delete_hash(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) {
+ mtraversal_select_delete_context_t sd_context = {0};
+ Sint chunk_size = 0;
+
+ sd_context.p = p;
+ sd_context.tb = &tbl->hash;
+ sd_context.tid = tid;
+ sd_context.hp = NULL;
+ sd_context.prev_continuation_tptr = NULL;
+#ifdef ERTS_SMP
+ sd_context.fixated_by_me = sd_context.tb->common.is_thread_safe ? 0 : 1; /* TODO: something nicer */
+#else
+ sd_context.fixated_by_me = 0;
+#endif
+ sd_context.last_pseudo_delete = (Uint) -1;
+
+ return match_traverse(
+ sd_context.p, sd_context.tb,
+ pattern, NULL,
+ chunk_size,
+ MAX_SELECT_DELETE_ITERATIONS, NULL, 1,
+ mtraversal_select_delete_on_nothing_can_match,
+ mtraversal_select_delete_on_match_res,
+ mtraversal_select_delete_on_loop_ended,
+ mtraversal_select_delete_on_trap,
+ &sd_context, ret);
+}
+
+/*
+ * This is called when select_delete traps
+ */
+static int db_select_delete_continue_hash(Process* p, DbTable* tbl, Eterm continuation, Eterm* ret) {
+ mtraversal_select_delete_context_t sd_context = {0};
+ Eterm* tptr;
+ Eterm tid;
+ Binary* mp;
+ Sint got;
+ Sint slot_ix;
+ Sint chunk_size = 0;
+
+ if (unpack_simple_mtraversal_continuation(continuation, &tptr, &tid, &slot_ix, &mp, &got)) {
+ *ret = NIL;
+ return DB_ERROR_BADPARAM;
+ }
+
+ sd_context.p = p;
+ sd_context.tb = &tbl->hash;
+ sd_context.tid = tid;
+ sd_context.hp = NULL;
+ sd_context.prev_continuation_tptr = tptr;
+ sd_context.fixated_by_me = ONLY_WRITER(p, sd_context.tb) ? 0 : 1; /* TODO: something nicer */
+ sd_context.last_pseudo_delete = (Uint) -1;
+
+ return match_traverse_continue(
+ sd_context.p, sd_context.tb, chunk_size,
+ MAX_SELECT_DELETE_ITERATIONS,
+ NULL, slot_ix, got, &mp, 1,
+ mtraversal_select_delete_on_match_res, /* Reuse callback */
+ mtraversal_select_delete_on_loop_ended, /* Reuse callback */
+ mtraversal_select_delete_on_trap, /* Reuse callback */
+ &sd_context, ret);
+}
+
+#undef MAX_SELECT_DELETE_ITERATIONS
+
+
+/*
+ *
+ * select_replace match traversal
+ *
+ */
+
+#define MAX_SELECT_REPLACE_ITERATIONS 1000
+
+typedef struct {
+ Process* p;
+ DbTableHash* tb;
+ Eterm tid;
+ Eterm* hp;
+ Eterm* prev_continuation_tptr;
+} mtraversal_select_replace_context_t;
+
+static int mtraversal_select_replace_on_nothing_can_match(void* context_ptr, Eterm* ret) {
+ *ret = make_small(0);
+ return DB_ERROR_NONE;
+}
+
+static int mtraversal_select_replace_on_match_res(void* context_ptr, Sint slot_ix,
+ HashDbTerm*** current_ptr_ptr,
+ Eterm match_res)
+{
+ mtraversal_select_replace_context_t* sr_context_ptr = (mtraversal_select_replace_context_t*) context_ptr;
+ DbTableHash* tb = sr_context_ptr->tb;
+ HashDbTerm* new;
+ HashDbTerm* next;
+ HashValue hval;
+
+ if (is_value(match_res)) {
+#ifdef DEBUG
+ Eterm key = db_getkey(tb->common.keypos, match_res);
+ ASSERT(is_value(key));
+ ASSERT(eq(key, GETKEY(tb, (**current_ptr_ptr)->dbterm.tpl)));
+#endif
+ next = (**current_ptr_ptr)->next;
+ hval = (**current_ptr_ptr)->hvalue;
+ new = new_dbterm(tb, match_res);
+ new->next = next;
+ new->hvalue = hval;
+ free_term(tb, **current_ptr_ptr);
+ **current_ptr_ptr = new; /* replace 'next' pointer in previous object */
+ *current_ptr_ptr = &((**current_ptr_ptr)->next); /* advance to next object */
+ return 1;
+ }
+ return 0;
+}
+
+static int mtraversal_select_replace_on_loop_ended(void* context_ptr, Sint slot_ix, Sint got,
+ Sint iterations_left, Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_replace_context_t* sr_context_ptr = (mtraversal_select_replace_context_t*) context_ptr;
+ ASSERT(iterations_left <= MAX_SELECT_REPLACE_ITERATIONS);
+ /* the more objects we've replaced, the more reductions we've consumed */
+ BUMP_REDS(sr_context_ptr->p,
+ MIN(MAX_SELECT_REPLACE_ITERATIONS * 2,
+ (MAX_SELECT_REPLACE_ITERATIONS - iterations_left) + (int)got));
+ *ret = erts_make_integer(got, sr_context_ptr->p);
+ return DB_ERROR_NONE;
+}
+
+static int mtraversal_select_replace_on_trap(void* context_ptr, Sint slot_ix, Sint got,
+ Binary** mpp, Eterm* ret)
+{
+ mtraversal_select_replace_context_t* sr_context_ptr = (mtraversal_select_replace_context_t*) context_ptr;
+ return on_mtraversal_simple_trap(
+ &ets_select_replace_continue_exp,
+ sr_context_ptr->p,
+ sr_context_ptr->tb,
+ sr_context_ptr->tid,
+ sr_context_ptr->prev_continuation_tptr,
+ slot_ix, got, mpp, ret);
+}
+
+static int db_select_replace_hash(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret)
+{
+ mtraversal_select_replace_context_t sr_context = {0};
+ Sint chunk_size = 0;
+
+ /* Bag implementation presented both semantic consistency and performance issues,
+ * unsupported for now
+ */
+ ASSERT(!(tbl->hash.common.status & DB_BAG));
+
+ sr_context.p = p;
+ sr_context.tb = &tbl->hash;
+ sr_context.tid = tid;
+ sr_context.hp = NULL;
+ sr_context.prev_continuation_tptr = NULL;
+
+ return match_traverse(
+ sr_context.p, sr_context.tb,
+ pattern, db_match_keeps_key,
+ chunk_size,
+ MAX_SELECT_REPLACE_ITERATIONS, NULL, 1,
+ mtraversal_select_replace_on_nothing_can_match,
+ mtraversal_select_replace_on_match_res,
+ mtraversal_select_replace_on_loop_ended,
+ mtraversal_select_replace_on_trap,
+ &sr_context, ret);
+}
+
+/*
+ * This is called when select_replace traps
+ */
+static int db_select_replace_continue_hash(Process* p, DbTable* tbl, Eterm continuation, Eterm* ret)
+{
+ mtraversal_select_replace_context_t sr_context = {0};
+ Eterm* tptr;
+ Eterm tid ;
+ Binary* mp;
+ Sint got;
+ Sint slot_ix;
+ Sint chunk_size = 0;
+ *ret = NIL;
+
+ if (unpack_simple_mtraversal_continuation(continuation, &tptr, &tid, &slot_ix, &mp, &got)) {
+ *ret = NIL;
+ return DB_ERROR_BADPARAM;
+ }
+
+ /* Proceed */
+ sr_context.p = p;
+ sr_context.tb = &tbl->hash;
+ sr_context.tid = tid;
+ sr_context.hp = NULL;
+ sr_context.prev_continuation_tptr = tptr;
+
+ return match_traverse_continue(
+ sr_context.p, sr_context.tb, chunk_size,
+ MAX_SELECT_REPLACE_ITERATIONS,
+ NULL, slot_ix, got, &mp, 1,
+ mtraversal_select_replace_on_match_res, /* Reuse callback */
+ mtraversal_select_replace_on_loop_ended, /* Reuse callback */
+ mtraversal_select_replace_on_trap, /* Reuse callback */
+ &sr_context, ret);
+}
+
+
static int db_take_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret)
{
DbTableHash *tb = &tbl->hash;
@@ -2101,6 +2289,7 @@ static int db_take_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret)
return DB_ERROR_NONE;
}
+
/*
** Other interface routines (not directly coupled to one bif)
*/
@@ -2197,19 +2386,17 @@ static void db_print_hash(fmtfn_t to, void *to_arg, int show, DbTable *tbl)
/* release all memory occupied by a single table */
static int db_free_table_hash(DbTable *tbl)
{
- while (!db_free_table_continue_hash(tbl))
+ while (db_free_table_continue_hash(tbl, ERTS_SWORD_MAX) < 0)
;
return 0;
}
-static int db_free_table_continue_hash(DbTable *tbl)
+static SWord db_free_table_continue_hash(DbTable *tbl, SWord reds)
{
DbTableHash *tb = &tbl->hash;
- int done;
FixedDeletion* fixdel = (FixedDeletion*) erts_smp_atomic_read_acqb(&tb->fixdel);
- ERTS_SMP_LC_ASSERT(IS_TAB_WLOCKED(tb));
+ ERTS_SMP_LC_ASSERT(IS_TAB_WLOCKED(tb) || (tb->common.status & DB_DELETE));
- done = 0;
while (fixdel != NULL) {
FixedDeletion *fx = fixdel;
@@ -2219,22 +2406,21 @@ static int db_free_table_continue_hash(DbTable *tbl)
(void *) fx,
sizeof(FixedDeletion));
ERTS_ETS_MISC_MEM_ADD(-sizeof(FixedDeletion));
- if (++done >= 2*DELETE_RECORD_LIMIT) {
+ if (--reds < 0) {
erts_smp_atomic_set_relb(&tb->fixdel, (erts_aint_t)fixdel);
- return 0; /* Not done */
+ return reds; /* Not done */
}
}
erts_smp_atomic_set_relb(&tb->fixdel, (erts_aint_t)NULL);
- done /= 2;
while(tb->nslots != 0) {
- done += 1 + EXT_SEGSZ/64 + free_seg(tb, 1);
+ reds -= EXT_SEGSZ/64 + free_seg(tb, 1);
/*
* If we have done enough work, get out here.
*/
- if (done >= DELETE_RECORD_LIMIT) {
- return 0; /* Not done */
+ if (reds < 0) {
+ return reds; /* Not done */
}
}
#ifdef ERTS_SMP
@@ -2249,7 +2435,7 @@ static int db_free_table_continue_hash(DbTable *tbl)
}
#endif
ASSERT(erts_smp_atomic_read_nob(&tb->common.memory_size) == sizeof(DbTable));
- return 1; /* Done */
+ return reds; /* Done */
}
@@ -2262,7 +2448,8 @@ static int db_free_table_continue_hash(DbTable *tbl)
** slots should be searched. Also compiles the match program
*/
static int analyze_pattern(DbTableHash *tb, Eterm pattern,
- struct mp_info *mpi)
+ extra_match_validator_t extra_validator, /* Optional callback */
+ struct mp_info *mpi)
{
Eterm *ptpl;
Eterm lst, tpl, ttpl;
@@ -2300,7 +2487,10 @@ static int analyze_pattern(DbTableHash *tb, Eterm pattern,
i = 0;
for(lst = pattern; is_list(lst); lst = CDR(list_val(lst))) {
- Eterm body;
+ Eterm match;
+ Eterm guard;
+ Eterm body;
+
ttpl = CAR(list_val(lst));
if (!is_tuple(ttpl)) {
if (buff != sbuff) {
@@ -2315,9 +2505,17 @@ static int analyze_pattern(DbTableHash *tb, Eterm pattern,
}
return DB_ERROR_BADPARAM;
}
- matches[i] = tpl = ptpl[1];
- guards[i] = ptpl[2];
+ matches[i] = match = tpl = ptpl[1];
+ guards[i] = guard = ptpl[2];
bodies[i] = body = ptpl[3];
+
+ if(extra_validator != NULL && !extra_validator(tb->common.keypos, match, guard, body)) {
+ if (buff != sbuff) {
+ erts_free(ERTS_ALC_T_DB_TMP, buff);
+ }
+ return DB_ERROR_BADPARAM;
+ }
+
if (!is_list(body) || CDR(list_val(body)) != NIL ||
CAR(list_val(body)) != am_DollarUnderscore) {
mpi->all_objects = 0;
@@ -3007,63 +3205,4 @@ Eterm erts_ets_hash_sizeof_ext_segtab(void)
{
return make_small(((SIZEOF_EXT_SEGTAB(0)-1) / sizeof(UWord)) + 1);
}
-/* For testing only */
-Eterm erts_ets_hash_get_memstate(Process* p, DbTableHash* tb)
-{
- Eterm seg_cnt;
- while (!begin_resizing(tb))
- /*spinn*/;
-
- seg_cnt = make_small(SLOT_IX_TO_SEG_IX(tb->nslots));
- done_resizing(tb);
- return seg_cnt;
-}
-/* For testing only */
-Eterm erts_ets_hash_restore_memstate(DbTableHash* tb, Eterm memstate)
-{
- int seg_cnt, target;
-
- if (!is_small(memstate))
- return make_small(__LINE__);
-
- target = signed_val(memstate);
- if (target < 1)
- return make_small(__LINE__);
- while (1) {
- while (!begin_resizing(tb))
- /*spin*/;
- seg_cnt = SLOT_IX_TO_SEG_IX(tb->nslots);
- done_resizing(tb);
-
- if (target == seg_cnt)
- return am_ok;
- if (IS_FIXED(tb))
- return make_small(__LINE__);
- if (target < seg_cnt)
- shrink(tb, 0);
- else
- grow(tb, INT_MAX);
- }
-}
-
-#ifdef HARDDEBUG
-
-void db_check_table_hash(DbTable *tbl)
-{
- DbTableHash *tb = &tbl->hash;
- HashDbTerm* list;
- int j;
-
- for (j = 0; j < NACTIVE(tb); j++) {
- if ((list = BUCKET(tb,j)) != 0) {
- while (list != 0) {
- if (!is_tuple(make_tuple(list->dbterm.tpl))) {
- erts_exit(ERTS_ERROR_EXIT, "Bad term in slot %d of ets table", j);
- }
- list = list->next;
- }
- }
- }
-}
-#endif
diff --git a/erts/emulator/beam/erl_db_hash.h b/erts/emulator/beam/erl_db_hash.h
index 6d25c73549..c340c72311 100644
--- a/erts/emulator/beam/erl_db_hash.h
+++ b/erts/emulator/beam/erl_db_hash.h
@@ -75,7 +75,7 @@ typedef struct db_table_hash {
** table types. The process is always an [in out] parameter.
*/
void db_initialize_hash(void);
-void db_unfix_table_hash(DbTableHash *tb /* [in out] */);
+SWord db_unfix_table_hash(DbTableHash *tb);
Uint db_kept_items_hash(DbTableHash *tb);
/* Interface for meta pid table */
@@ -88,14 +88,6 @@ int db_get_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret);
int db_erase_hash(DbTable *tbl, Eterm key, Eterm *ret);
-int db_get_element_array(DbTable *tbl,
- Eterm key,
- int ndex,
- Eterm *ret,
- int *num_ret);
-
-int db_erase_bag_exact2(DbTable *tbl, Eterm key, Eterm value);
-
/* not yet in method table */
int db_mark_all_deleted_hash(DbTable *tbl);
@@ -110,7 +102,5 @@ typedef struct {
void db_calc_stats_hash(DbTableHash* tb, DbHashStats*);
Eterm erts_ets_hash_sizeof_ext_segtab(void);
-Eterm erts_ets_hash_get_memstate(Process*, DbTableHash* tb);
-Eterm erts_ets_hash_restore_memstate(DbTableHash* tb, Eterm memstate);
#endif /* _DB_HASH_H */
diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c
index c4ecd2ba37..ab8da6ccf6 100644
--- a/erts/emulator/beam/erl_db_tree.c
+++ b/erts/emulator/beam/erl_db_tree.c
@@ -76,9 +76,18 @@
((Dtt->pos) ? \
(Dtt)->array[(Dtt)->pos - 1] : NULL)
-#define EMPTY_NODE(Dtt) (TOP_NODE(Dtt) == NULL)
+#define TOPN_NODE(Dtt, Pos) \
+ (((Pos) < Dtt->pos) ? \
+ (Dtt)->array[(Dtt)->pos - ((Pos) + 1)] : NULL)
+
+#define REPLACE_TOP_NODE(Dtt, Node) \
+ if ((Dtt)->pos) (Dtt)->array[(Dtt)->pos - 1] = (Node)
+#define EMPTY_NODE(Dtt) (TOP_NODE(Dtt) == NULL)
+#ifndef MIN
+#define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
+#endif
/* Obtain table static stack if available. NULL if not.
** Must be released with release_stack()
@@ -180,7 +189,6 @@ static ERTS_INLINE TreeDbTerm* replace_dbterm(DbTableTree *tb, TreeDbTerm* old,
static TreeDbTerm *traverse_until(TreeDbTerm *t, int *current, int to);
static void check_slot_pos(DbTableTree *tb);
static void check_saved_stack(DbTableTree *tb);
-static int check_table_tree(DbTableTree* tb, TreeDbTerm *t);
#define TREE_DEBUG
#endif
@@ -226,9 +234,9 @@ struct mp_info {
Eterm most; /* The highest matching key (possibly
* partially bound expression) */
- TreeDbTerm *save_term; /* If the key is completely bound, this
- * will be the Tree node we're searching
- * for, otherwise it will be useless */
+ TreeDbTerm **save_term; /* If the key is completely bound, this
+ * will be the Tree node we're searching
+ * for, otherwise it will be useless */
Binary *mp; /* The compiled match program */
};
@@ -278,12 +286,30 @@ struct select_delete_context {
};
/*
+ * Used by doit_select_replace
+ */
+struct select_replace_context {
+ Process *p;
+ DbTableTree *tb;
+ Binary *mp;
+ Eterm end_condition;
+ Eterm *lastobj;
+ Sint32 max;
+ int keypos;
+ int all_objects;
+ Sint replaced;
+};
+
+/* Used by select_replace on analyze_pattern */
+typedef int (*extra_match_validator_t)(int keypos, Eterm match, Eterm guard, Eterm body);
+
+/*
** Forward declarations
*/
static TreeDbTerm *linkout_tree(DbTableTree *tb, Eterm key);
static TreeDbTerm *linkout_object_tree(DbTableTree *tb,
Eterm object);
-static int do_free_tree_cont(DbTableTree *tb, int num_left);
+static SWord do_free_tree_continue(DbTableTree *tb, SWord reds);
static void free_term(DbTableTree *tb, TreeDbTerm* p);
static int balance_left(TreeDbTerm **this);
static int balance_right(TreeDbTerm **this);
@@ -291,6 +317,7 @@ static int delsub(TreeDbTerm **this);
static TreeDbTerm *slot_search(Process *p, DbTableTree *tb, Sint slot);
static TreeDbTerm *find_node(DbTableTree *tb, Eterm key);
static TreeDbTerm **find_node2(DbTableTree *tb, Eterm key);
+static TreeDbTerm **find_ptr(DbTableTree *tb, DbTreeStack*, TreeDbTerm *this);
static TreeDbTerm *find_next(DbTableTree *tb, DbTreeStack*, Eterm key);
static TreeDbTerm *find_prev(DbTableTree *tb, DbTreeStack*, Eterm key);
static TreeDbTerm *find_next_from_pb_key(DbTableTree *tb, DbTreeStack*,
@@ -312,14 +339,23 @@ static void traverse_forward(DbTableTree *tb,
TreeDbTerm *,
void *,
int),
- void *context);
-static int key_given(DbTableTree *tb, Eterm pattern, TreeDbTerm **ret,
+ void *context);
+static void traverse_update_backwards(DbTableTree *tb,
+ DbTreeStack*,
+ Eterm lastkey,
+ int (*doit)(DbTableTree *tb,
+ TreeDbTerm **, // out
+ void *,
+ int),
+ void *context);
+static int key_given(DbTableTree *tb, Eterm pattern, TreeDbTerm ***ret,
Eterm *partly_bound_key);
static Sint cmp_partly_bound(Eterm partly_bound_key, Eterm bound_key);
static Sint do_cmp_partly_bound(Eterm a, Eterm b, int *done);
static int analyze_pattern(DbTableTree *tb, Eterm pattern,
- struct mp_info *mpi);
+ extra_match_validator_t extra_validator, /* Optional callback */
+ struct mp_info *mpi);
static int doit_select(DbTableTree *tb,
TreeDbTerm *this,
void *ptr,
@@ -336,6 +372,10 @@ static int doit_select_delete(DbTableTree *tb,
TreeDbTerm *this,
void *ptr,
int forward);
+static int doit_select_replace(DbTableTree *tb,
+ TreeDbTerm **this_ptr,
+ void *ptr,
+ int forward);
static int partly_bound_can_match_lesser(Eterm partly_bound_1,
Eterm partly_bound_2);
@@ -369,27 +409,31 @@ static int db_erase_tree(DbTable *tbl, Eterm key, Eterm *ret);
static int db_erase_object_tree(DbTable *tbl, Eterm object,Eterm *ret);
static int db_slot_tree(Process *p, DbTable *tbl,
Eterm slot_term, Eterm *ret);
-static int db_select_tree(Process *p, DbTable *tbl,
+static int db_select_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, int reversed, Eterm *ret);
-static int db_select_count_tree(Process *p, DbTable *tbl,
+static int db_select_count_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Eterm *ret);
-static int db_select_chunk_tree(Process *p, DbTable *tbl,
+static int db_select_chunk_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Sint chunk_size,
int reversed, Eterm *ret);
static int db_select_continue_tree(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
static int db_select_count_continue_tree(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
-static int db_select_delete_tree(Process *p, DbTable *tbl,
+static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Eterm *ret);
static int db_select_delete_continue_tree(Process *p, DbTable *tbl,
Eterm continuation, Eterm *ret);
+static int db_select_replace_tree(Process *p, DbTable *tbl, Eterm tid,
+ Eterm pattern, Eterm *ret);
+static int db_select_replace_continue_tree(Process *p, DbTable *tbl,
+ Eterm continuation, Eterm *ret);
static int db_take_tree(Process *, DbTable *, Eterm, Eterm *);
static void db_print_tree(fmtfn_t to, void *to_arg,
int show, DbTable *tbl);
static int db_free_table_tree(DbTable *tbl);
-static int db_free_table_continue_tree(DbTable *tbl);
+static SWord db_free_table_continue_tree(DbTable *tbl, SWord);
static void db_foreach_offheap_tree(DbTable *,
void (*)(ErlOffHeap *, void *),
@@ -436,17 +480,14 @@ DbTableMethod db_tree =
db_select_delete_continue_tree,
db_select_count_tree,
db_select_count_continue_tree,
+ db_select_replace_tree,
+ db_select_replace_continue_tree,
db_take_tree,
db_delete_all_objects_tree,
db_free_table_tree,
db_free_table_continue_tree,
db_print_tree,
db_foreach_offheap_tree,
-#ifdef HARDDEBUG
- db_check_table_tree,
-#else
- NULL,
-#endif
db_lookup_dbterm_tree,
db_finalize_dbterm_tree
@@ -1058,7 +1099,7 @@ static int db_select_continue_tree(Process *p,
}
-static int db_select_tree(Process *p, DbTable *tbl,
+static int db_select_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, int reverse, Eterm *ret)
{
/* Strategy: Traverse backwards to build resulting list from tail to head */
@@ -1095,7 +1136,7 @@ static int db_select_tree(Process *p, DbTable *tbl,
sc.got = 0;
sc.chunk_size = 0;
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
+ if ((errcode = analyze_pattern(tb, pattern, NULL, &mpi)) != DB_ERROR_NONE) {
RET_TO_BIF(NIL,errcode);
}
@@ -1109,7 +1150,7 @@ static int db_select_tree(Process *p, DbTable *tbl,
if (!mpi.got_partial && mpi.some_limitation &&
CMP_EQ(mpi.least,mpi.most)) {
- doit_select(tb,mpi.save_term,&sc,0 /* direction doesn't matter */);
+ doit_select(tb,*(mpi.save_term),&sc,0 /* direction doesn't matter */);
RET_TO_BIF(sc.accum,DB_ERROR_NONE);
}
@@ -1151,7 +1192,7 @@ static int db_select_tree(Process *p, DbTable *tbl,
continuation = TUPLE8
(hp,
- tb->common.id,
+ tid,
key,
sc.end_condition, /* From the match program, needn't be copied */
make_small(0), /* Chunk size of zero means not chunked to the
@@ -1263,7 +1304,7 @@ static int db_select_count_continue_tree(Process *p,
}
-static int db_select_count_tree(Process *p, DbTable *tbl,
+static int db_select_count_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Eterm *ret)
{
DbTableTree *tb = &tbl->tree;
@@ -1298,7 +1339,7 @@ static int db_select_count_tree(Process *p, DbTable *tbl,
sc.keypos = tb->common.keypos;
sc.got = 0;
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
+ if ((errcode = analyze_pattern(tb, pattern, NULL, &mpi)) != DB_ERROR_NONE) {
RET_TO_BIF(NIL,errcode);
}
@@ -1312,7 +1353,7 @@ static int db_select_count_tree(Process *p, DbTable *tbl,
if (!mpi.got_partial && mpi.some_limitation &&
CMP_EQ(mpi.least,mpi.most)) {
- doit_select_count(tb,mpi.save_term,&sc,0 /* dummy */);
+ doit_select_count(tb,*(mpi.save_term),&sc,0 /* dummy */);
RET_TO_BIF(erts_make_integer(sc.got,p),DB_ERROR_NONE);
}
@@ -1349,7 +1390,7 @@ static int db_select_count_tree(Process *p, DbTable *tbl,
continuation = TUPLE5
(hp,
- tb->common.id,
+ tid,
key,
sc.end_condition, /* From the match program, needn't be copied */
mpb,
@@ -1363,7 +1404,7 @@ static int db_select_count_tree(Process *p, DbTable *tbl,
}
-static int db_select_chunk_tree(Process *p, DbTable *tbl,
+static int db_select_chunk_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Sint chunk_size,
int reverse,
Eterm *ret)
@@ -1401,7 +1442,7 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl,
sc.got = 0;
sc.chunk_size = chunk_size;
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
+ if ((errcode = analyze_pattern(tb, pattern, NULL, &mpi)) != DB_ERROR_NONE) {
RET_TO_BIF(NIL,errcode);
}
@@ -1415,7 +1456,7 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl,
if (!mpi.got_partial && mpi.some_limitation &&
CMP_EQ(mpi.least,mpi.most)) {
- doit_select(tb,mpi.save_term,&sc, 0 /* direction doesn't matter */);
+ doit_select(tb,*(mpi.save_term),&sc, 0 /* direction doesn't matter */);
if (sc.accum != NIL) {
hp=HAlloc(p, 3);
RET_TO_BIF(TUPLE2(hp,sc.accum,am_EOT),DB_ERROR_NONE);
@@ -1474,7 +1515,7 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl,
continuation = TUPLE8
(hp,
- tb->common.id,
+ tid,
key,
sc.end_condition, /* From the match program,
needn't be copied */
@@ -1499,7 +1540,7 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl,
mpb = erts_db_make_match_prog_ref(p,mpi.mp,&hp);
continuation = TUPLE8
(hp,
- tb->common.id,
+ tid,
key,
sc.end_condition, /* From the match program, needn't be copied */
make_small(chunk_size),
@@ -1605,7 +1646,7 @@ static int db_select_delete_continue_tree(Process *p,
#undef RET_TO_BIF
}
-static int db_select_delete_tree(Process *p, DbTable *tbl,
+static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm tid,
Eterm pattern, Eterm *ret)
{
DbTableTree *tb = &tbl->tree;
@@ -1643,7 +1684,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl,
sc.keypos = tb->common.keypos;
sc.tb = tb;
- if ((errcode = analyze_pattern(tb, pattern, &mpi)) != DB_ERROR_NONE) {
+ if ((errcode = analyze_pattern(tb, pattern, NULL, &mpi)) != DB_ERROR_NONE) {
RET_TO_BIF(0,errcode);
}
@@ -1656,7 +1697,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl,
if (!mpi.got_partial && mpi.some_limitation &&
CMP_EQ(mpi.least,mpi.most)) {
- doit_select_delete(tb,mpi.save_term,&sc, 0 /* direction doesn't
+ doit_select_delete(tb,*(mpi.save_term),&sc, 0 /* direction doesn't
matter */);
RET_TO_BIF(erts_make_integer(sc.accum,p),DB_ERROR_NONE);
}
@@ -1691,7 +1732,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl,
continuation = TUPLE5
(hp,
- tb->common.id,
+ tid,
key,
sc.end_condition, /* From the match program, needn't be copied */
mpb,
@@ -1708,6 +1749,208 @@ static int db_select_delete_tree(Process *p, DbTable *tbl,
}
+static int db_select_replace_continue_tree(Process *p,
+ DbTable *tbl,
+ Eterm continuation,
+ Eterm *ret)
+{
+ DbTableTree *tb = &tbl->tree;
+ DbTreeStack* stack;
+ struct select_replace_context sc;
+ unsigned sz;
+ Eterm *hp;
+ Eterm lastkey;
+ Eterm end_condition;
+ Binary *mp;
+ Eterm key;
+ Eterm *tptr;
+ Eterm ereplaced;
+ Sint prev_replaced;
+
+
+#define RET_TO_BIF(Term, State) do { *ret = (Term); return State; } while(0);
+
+ /* Decode continuation. We know it's a tuple and everything else as
+ this is only called by ourselves */
+
+ /* continuation:
+ {Table, Lastkey, EndCondition, MatchProgBin, HowManyReplaced}*/
+
+ tptr = tuple_val(continuation);
+
+ if (arityval(*tptr) != 5)
+ erts_exit(ERTS_ERROR_EXIT,"Internal error in ets:select_replace/1");
+
+ lastkey = tptr[2];
+ end_condition = tptr[3];
+ mp = erts_db_get_match_prog_binary_unchecked(tptr[4]);
+
+ sc.p = p;
+ sc.mp = mp;
+ sc.end_condition = NIL;
+ sc.lastobj = NULL;
+ sc.max = 1000;
+ sc.keypos = tb->common.keypos;
+ if (is_big(tptr[5])) {
+ sc.replaced = big_to_uint32(tptr[5]);
+ } else {
+ sc.replaced = unsigned_val(tptr[5]);
+ }
+ prev_replaced = sc.replaced;
+
+ stack = get_any_stack(tb);
+ traverse_update_backwards(tb, stack, lastkey, &doit_select_replace, &sc);
+ release_stack(tb,stack);
+
+ // the more objects we've replaced, the more reductions we've consumed
+ BUMP_REDS(p, MIN(2000, (1000 - sc.max) + (sc.replaced - prev_replaced)));
+
+ if (sc.max > 0) {
+ RET_TO_BIF(erts_make_integer(sc.replaced,p), DB_ERROR_NONE);
+ }
+ key = GETKEY(tb, sc.lastobj);
+ if (end_condition != NIL &&
+ (cmp_partly_bound(end_condition,key) > 0)) {
+ /* done anyway */
+ RET_TO_BIF(make_small(sc.replaced),DB_ERROR_NONE);
+ }
+ /* Not done yet, let's trap. */
+ sz = size_object(key);
+ if (IS_USMALL(0, sc.replaced)) {
+ hp = HAlloc(p, sz + 6);
+ ereplaced = make_small(sc.replaced);
+ }
+ else {
+ hp = HAlloc(p, BIG_UINT_HEAP_SIZE + sz + 6);
+ ereplaced = uint_to_big(sc.replaced, hp);
+ hp += BIG_UINT_HEAP_SIZE;
+ }
+ key = copy_struct(key, sz, &hp, &MSO(p));
+ continuation = TUPLE5
+ (hp,
+ tptr[1],
+ key,
+ tptr[3],
+ tptr[4],
+ ereplaced);
+ RET_TO_BIF(bif_trap1(&ets_select_replace_continue_exp, p, continuation),
+ DB_ERROR_NONE);
+
+#undef RET_TO_BIF
+}
+
+static int db_select_replace_tree(Process *p, DbTable *tbl, Eterm tid,
+ Eterm pattern, Eterm *ret)
+{
+ DbTableTree *tb = &tbl->tree;
+ DbTreeStack* stack;
+ struct select_replace_context sc;
+ struct mp_info mpi;
+ Eterm lastkey = THE_NON_VALUE;
+ Eterm key;
+ Eterm continuation;
+ unsigned sz;
+ Eterm *hp;
+ TreeDbTerm *this;
+ int errcode;
+ Eterm ereplaced;
+ Eterm mpb;
+
+
+#define RET_TO_BIF(Term,RetVal) do { \
+ if (mpi.mp != NULL) { \
+ erts_bin_free(mpi.mp); \
+ } \
+ *ret = (Term); \
+ return RetVal; \
+ } while(0)
+
+ mpi.mp = NULL;
+
+ sc.lastobj = NULL;
+ sc.p = p;
+ sc.tb = tb;
+ sc.max = 1000;
+ sc.end_condition = NIL;
+ sc.keypos = tb->common.keypos;
+ sc.replaced = 0;
+
+ if ((errcode = analyze_pattern(tb, pattern, db_match_keeps_key, &mpi)) != DB_ERROR_NONE) {
+ RET_TO_BIF(NIL,errcode);
+ }
+
+ if (!mpi.something_can_match) {
+ RET_TO_BIF(make_small(0),DB_ERROR_NONE);
+ /* can't possibly match anything */
+ }
+
+ sc.mp = mpi.mp;
+ sc.all_objects = mpi.all_objects;
+
+ stack = get_static_stack(tb);
+ if (!mpi.got_partial && mpi.some_limitation &&
+ CMP_EQ(mpi.least,mpi.most)) {
+ TreeDbTerm* term = *(mpi.save_term);
+ doit_select_replace(tb,mpi.save_term,&sc,0 /* dummy */);
+ if (stack != NULL) {
+ if (TOP_NODE(stack) == term)
+ // throw away potentially invalid reference
+ REPLACE_TOP_NODE(stack, *(mpi.save_term));
+ release_stack(tb, stack);
+ }
+ RET_TO_BIF(erts_make_integer(sc.replaced,p),DB_ERROR_NONE);
+ }
+
+ if (stack == NULL)
+ stack = get_any_stack(tb);
+
+ if (mpi.some_limitation) {
+ if ((this = find_next_from_pb_key(tb, stack, mpi.most)) != NULL) {
+ lastkey = GETKEY(tb, this->dbterm.tpl);
+ }
+ sc.end_condition = mpi.least;
+ }
+
+ traverse_update_backwards(tb, stack, lastkey, &doit_select_replace, &sc);
+ release_stack(tb,stack);
+ // the more objects we've replaced, the more reductions we've consumed
+ BUMP_REDS(p, MIN(2000, (1000 - sc.max) + sc.replaced));
+ if (sc.max > 0) {
+ RET_TO_BIF(erts_make_integer(sc.replaced,p),DB_ERROR_NONE);
+ }
+
+ key = GETKEY(tb, sc.lastobj);
+ sz = size_object(key);
+ if (IS_USMALL(0, sc.replaced)) {
+ hp = HAlloc(p, sz + ERTS_MAGIC_REF_THING_SIZE + 6);
+ ereplaced = make_small(sc.replaced);
+ }
+ else {
+ hp = HAlloc(p, BIG_UINT_HEAP_SIZE + sz + ERTS_MAGIC_REF_THING_SIZE + 6);
+ ereplaced = uint_to_big(sc.replaced, hp);
+ hp += BIG_UINT_HEAP_SIZE;
+ }
+ key = copy_struct(key, sz, &hp, &MSO(p));
+ if (mpi.all_objects)
+ (mpi.mp)->flags |= BIN_FLAG_ALL_OBJECTS;
+ mpb = erts_db_make_match_prog_ref(p,mpi.mp,&hp);
+
+ continuation = TUPLE5
+ (hp,
+ tid,
+ key,
+ sc.end_condition, /* From the match program, needn't be copied */
+ mpb,
+ ereplaced);
+
+ /* Don't free mpi.mp, so don't use macro */
+ *ret = bif_trap1(&ets_select_replace_continue_exp, p, continuation);
+ return DB_ERROR_NONE;
+
+#undef RET_TO_BIF
+
+}
+
static int db_take_tree(Process *p, DbTable *tbl, Eterm key, Eterm *ret)
{
DbTableTree *tb = &tbl->tree;
@@ -1757,23 +2000,22 @@ static void db_print_tree(fmtfn_t to, void *to_arg,
/* release all memory occupied by a single table */
static int db_free_table_tree(DbTable *tbl)
{
- while (!db_free_table_continue_tree(tbl))
+ while (db_free_table_continue_tree(tbl, ERTS_SWORD_MAX) < 0)
;
return 1;
}
-static int db_free_table_continue_tree(DbTable *tbl)
+static SWord db_free_table_continue_tree(DbTable *tbl, SWord reds)
{
DbTableTree *tb = &tbl->tree;
- int result;
if (!tb->deletion) {
tb->static_stack.pos = 0;
tb->deletion = 1;
PUSH_NODE(&tb->static_stack, tb->root);
}
- result = do_free_tree_cont(tb, DELETE_RECORD_LIMIT);
- if (result) { /* Completely done. */
+ reds = do_free_tree_continue(tb, reds);
+ if (reds >= 0) { /* Completely done. */
erts_db_free(ERTS_ALC_T_DB_STK,
(DbTable *) tb,
(void *) tb->static_stack.array,
@@ -1781,7 +2023,7 @@ static int db_free_table_continue_tree(DbTable *tbl)
ASSERT(erts_smp_atomic_read_nob(&tb->common.memory_size)
== sizeof(DbTable));
}
- return result;
+ return reds;
}
static int db_delete_all_objects_tree(Process* p, DbTable* tbl)
@@ -1954,8 +2196,9 @@ static TreeDbTerm *linkout_object_tree(DbTableTree *tb,
** For the select functions, analyzes the pattern and determines which
** part of the tree should be searched. Also compiles the match program
*/
-static int analyze_pattern(DbTableTree *tb, Eterm pattern,
- struct mp_info *mpi)
+static int analyze_pattern(DbTableTree *tb, Eterm pattern,
+ extra_match_validator_t extra_validator, /* Optional callback */
+ struct mp_info *mpi)
{
Eterm lst, tpl, ttpl;
Eterm *matches,*guards, *bodies;
@@ -1993,7 +2236,10 @@ static int analyze_pattern(DbTableTree *tb, Eterm pattern,
i = 0;
for(lst = pattern; is_list(lst); lst = CDR(list_val(lst))) {
- Eterm body;
+ Eterm match;
+ Eterm guard;
+ Eterm body;
+
ttpl = CAR(list_val(lst));
if (!is_tuple(ttpl)) {
if (buff != sbuff) {
@@ -2008,9 +2254,17 @@ static int analyze_pattern(DbTableTree *tb, Eterm pattern,
}
return DB_ERROR_BADPARAM;
}
- matches[i] = tpl = ptpl[1];
- guards[i] = ptpl[2];
+ matches[i] = match = tpl = ptpl[1];
+ guards[i] = guard = ptpl[2];
bodies[i] = body = ptpl[3];
+
+ if(extra_validator != NULL && !extra_validator(tb->common.keypos, match, guard, body)) {
+ if (buff != sbuff) {
+ erts_free(ERTS_ALC_T_DB_TMP, buff);
+ }
+ return DB_ERROR_BADPARAM;
+ }
+
if (!is_list(body) || CDR(list_val(body)) != NIL ||
CAR(list_val(body)) != am_DollarUnderscore) {
mpi->all_objects = 0;
@@ -2018,7 +2272,7 @@ static int analyze_pattern(DbTableTree *tb, Eterm pattern,
++i;
partly_bound = NIL;
- res = key_given(tb, tpl, &mpi->save_term, &partly_bound);
+ res = key_given(tb, tpl, &(mpi->save_term), &partly_bound);
if ( res >= 0 ) { /* Can match something */
key = 0;
mpi->something_can_match = 1;
@@ -2064,7 +2318,7 @@ static int analyze_pattern(DbTableTree *tb, Eterm pattern,
return DB_ERROR_NONE;
}
-static int do_free_tree_cont(DbTableTree *tb, int num_left)
+static SWord do_free_tree_continue(DbTableTree *tb, SWord reds)
{
TreeDbTerm *root;
TreeDbTerm *p;
@@ -2083,15 +2337,14 @@ static int do_free_tree_cont(DbTableTree *tb, int num_left)
root = p;
} else {
free_term(tb, root);
- if (--num_left > 0) {
- break;
- } else {
- return 0; /* Done enough for now */
- }
+ if (--reds < 0) {
+ return reds; /* Done enough for now */
+ }
+ break;
}
}
}
- return 1;
+ return reds;
}
/*
@@ -2522,6 +2775,58 @@ static TreeDbTerm **find_node2(DbTableTree *tb, Eterm key)
return this;
}
+/*
+ * Find node and return the address of the node pointer (NULL if not found)
+ * Tries to reuse the existing stack for performance.
+ */
+
+static TreeDbTerm **find_ptr(DbTableTree *tb, DbTreeStack *stack, TreeDbTerm *this) {
+ Eterm key = GETKEY(tb, this->dbterm.tpl);
+ TreeDbTerm *tmp;
+ TreeDbTerm *parent;
+ Sint c;
+
+ if(( tmp = TOP_NODE(stack)) != NULL) {
+ if (!cmp_key_eq(tb,key,tmp)) {
+ /* Start from the beginning */
+ stack->pos = stack->slot = 0;
+ }
+ }
+ if (EMPTY_NODE(stack)) { /* Have to rebuild the stack */
+ if (( tmp = tb->root ) == NULL)
+ return NULL;
+ for (;;) {
+ PUSH_NODE(stack, tmp);
+ if (( c = cmp_key(tb,key,tmp) ) < 0) {
+ if (tmp->left == NULL) /* We are at the next
+ and the element does
+ not exist */
+ break;
+ else
+ tmp = tmp->left;
+ } else if (c > 0) {
+ if (tmp->right == NULL) /* Done */
+ return NULL;
+ else
+ tmp = tmp->right;
+ } else
+ break;
+ }
+ }
+
+ if (TOP_NODE(stack) != this)
+ return NULL;
+
+ parent = TOPN_NODE(stack, 1);
+ if (parent == NULL)
+ return ((this != tb->root) ? NULL : &(tb->root));
+ if (parent->left == this)
+ return &(parent->left);
+ if (parent->right == this)
+ return &(parent->right);
+ return NULL;
+}
+
static int
db_lookup_dbterm_tree(Process *p, DbTable *tbl, Eterm key, Eterm obj,
DbUpdateHandle* handle)
@@ -2663,13 +2968,60 @@ static void traverse_forward(DbTableTree *tb,
}
/*
+ * Traverse the tree with an update callback function, used by db_select_replace
+ */
+static void traverse_update_backwards(DbTableTree *tb,
+ DbTreeStack* stack,
+ Eterm lastkey,
+ int (*doit)(DbTableTree*,
+ TreeDbTerm**,
+ void*,
+ int),
+ void* context)
+{
+ int res;
+ TreeDbTerm *this, *next, **this_ptr;
+
+ if (lastkey == THE_NON_VALUE) {
+ stack->pos = stack->slot = 0;
+ if (( this = tb->root ) == NULL) {
+ return;
+ }
+ while (this != NULL) {
+ PUSH_NODE(stack, this);
+ this = this->right;
+ }
+ this = TOP_NODE(stack);
+ this_ptr = find_ptr(tb, stack, this);
+ ASSERT(this_ptr != NULL);
+ res = (*doit)(tb, this_ptr, context, 0);
+ REPLACE_TOP_NODE(stack, *this_ptr);
+ next = find_prev(tb, stack, GETKEY(tb, (*this_ptr)->dbterm.tpl));
+ if (!res)
+ return;
+ } else {
+ next = find_prev(tb, stack, lastkey);
+ }
+
+ while ((this = next) != NULL) {
+ this_ptr = find_ptr(tb, stack, this);
+ ASSERT(this_ptr != NULL);
+ res = (*doit)(tb, this_ptr, context, 0);
+ REPLACE_TOP_NODE(stack, *this_ptr);
+ next = find_prev(tb, stack, GETKEY(tb, (*this_ptr)->dbterm.tpl));
+ if (!res)
+ return;
+ }
+}
+
+/*
* Returns 0 if not given 1 if given and -1 on no possible match
* if key is given; *ret is set to point to the object concerned.
*/
-static int key_given(DbTableTree *tb, Eterm pattern, TreeDbTerm **ret,
+static int key_given(DbTableTree *tb, Eterm pattern, TreeDbTerm ***ret,
Eterm *partly_bound)
{
- TreeDbTerm *this;
+ TreeDbTerm **this;
Eterm key;
ASSERT(ret != NULL);
@@ -2679,7 +3031,7 @@ static int key_given(DbTableTree *tb, Eterm pattern, TreeDbTerm **ret,
if (is_non_value(key))
return -1; /* can't possibly match anything */
if (!db_has_variable(key)) { /* Bound key */
- if (( this = find_node(tb, key) ) == NULL) {
+ if (( this = find_node2(tb, key) ) == NULL) {
return -1;
}
*ret = this;
@@ -3102,6 +3454,46 @@ static int doit_select_delete(DbTableTree *tb, TreeDbTerm *this, void *ptr,
return 1;
}
+static int doit_select_replace(DbTableTree *tb, TreeDbTerm **this, void *ptr,
+ int forward)
+{
+ struct select_replace_context *sc = (struct select_replace_context *) ptr;
+ Eterm ret;
+
+ sc->lastobj = (*this)->dbterm.tpl;
+
+ /* Always backwards traversing */
+ if (sc->end_condition != NIL &&
+ (cmp_partly_bound(sc->end_condition,
+ GETKEY_WITH_POS(sc->keypos, (*this)->dbterm.tpl)) > 0)) {
+ return 0;
+ }
+ ret = db_match_dbterm(&tb->common, sc->p, sc->mp, 0,
+ &(*this)->dbterm, NULL, 0);
+
+ if (is_value(ret)) {
+ TreeDbTerm* new;
+ TreeDbTerm* old = *this;
+#ifdef DEBUG
+ Eterm key = db_getkey(tb->common.keypos, ret);
+ ASSERT(is_value(key));
+ ASSERT(cmp_key(tb, key, old) == 0);
+#endif
+ new = new_dbterm(tb, ret);
+ new->left = old->left;
+ new->right = old->right;
+ new->balance = old->balance;
+ sc->lastobj = new->dbterm.tpl;
+ *this = new;
+ free_term(tb, old);
+ ++(sc->replaced);
+ }
+ if (--(sc->max) <= 0) {
+ return 0;
+ }
+ return 1;
+}
+
#ifdef TREE_DEBUG
static void do_dump_tree2(DbTableTree* tb, int to, void *to_arg, int show,
TreeDbTerm *t, int offset)
@@ -3130,6 +3522,9 @@ static void do_dump_tree2(DbTableTree* tb, int to, void *to_arg, int show,
#ifdef HARDDEBUG
+/*
+ * No called, but kept as it might come to use
+ */
void db_check_table_tree(DbTable *tbl)
{
DbTableTree *tb = &tbl->tree;
diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c
index 6f30b1d3dd..03cc11bdc4 100644
--- a/erts/emulator/beam/erl_db_util.c
+++ b/erts/emulator/beam/erl_db_util.c
@@ -1119,6 +1119,177 @@ error:
return NULL;
}
+/*
+ * Compare a matching term 'a' with a constructing term 'b' for equality.
+ *
+ * Returns true if 'b' is guaranteed to always construct
+ * the same term as 'a' has matched.
+ */
+static int db_match_eq_body(Eterm a, Eterm b)
+{
+ DECLARE_ESTACK(s);
+ Uint arity;
+ Eterm *ap, *bp;
+ int const_mode = 0;
+ const Eterm CONST_MODE_OFF = THE_NON_VALUE;
+
+ while (1) {
+ switch(b & _TAG_PRIMARY_MASK) {
+ case TAG_PRIMARY_LIST:
+ if (!is_list(a))
+ return 0;
+ ESTACK_PUSH2(s, CDR(list_val(a)), CDR(list_val(b)));
+ a = CAR(list_val(a));
+ b = CAR(list_val(b));
+ continue; /* loop without pop */
+
+ case TAG_PRIMARY_BOXED:
+ if (is_tuple(b)) {
+ bp = tuple_val(b);
+ if (!const_mode) {
+ if (bp[0] == make_arityval(1) && is_tuple(bp[1])) {
+ b = bp[1]; /* double-tuple syntax */
+ }
+ else if (bp[0] == make_arityval(2) && bp[1] == am_const) {
+ ESTACK_PUSH(s, CONST_MODE_OFF);
+ const_mode = 1; /* {const, term()} syntax */
+ b = bp[2];
+ continue; /* loop without pop */
+ }
+ else
+ return 0; /* function call or invalid tuple syntax */
+ }
+ if (!is_tuple(a))
+ return 0;
+
+ ap = tuple_val(a);
+ bp = tuple_val(b);
+ if (ap[0] != bp[0])
+ return 0;
+ arity = arityval(ap[0]);
+ if (arity > 0) {
+ a = *(++ap);
+ b = *(++bp);
+ while(--arity) {
+ ESTACK_PUSH2(s, *(++ap), *(++bp));
+ }
+ continue; /* loop without pop */
+ }
+ }
+ else if (is_map(b)) {
+ /* We don't know what other pairs the matched map may contain */
+ return 0;
+ }
+ else if (!eq(a,b)) /* other boxed */
+ return 0;
+ break;
+
+ case TAG_PRIMARY_IMMED1:
+ if (a != b || a == am_Underscore || a == am_DollarDollar
+ || a == am_DollarUnderscore
+ || (const_mode && db_is_variable(a) >= 0)) {
+
+ return 0;
+ }
+ break;
+ default:
+ erts_exit(ERTS_ABORT_EXIT, "db_compare: "
+ "Bad object on ESTACK: 0x%bex\n", b);
+ }
+
+pop_next:
+ if (ESTACK_ISEMPTY(s))
+ break; /* done */
+
+ b = ESTACK_POP(s);
+ if (b == CONST_MODE_OFF) {
+ ASSERT(const_mode);
+ const_mode = 0;
+ goto pop_next;
+ }
+ a = ESTACK_POP(s);
+ }
+
+ DESTROY_ESTACK(s);
+ return 1;
+}
+
+/* This is used by select_replace */
+int db_match_keeps_key(int keypos, Eterm match, Eterm guard, Eterm body)
+{
+ Eterm match_key;
+ Eterm* body_list;
+ Eterm single_body_term;
+ Eterm* single_body_term_tpl;
+ Eterm single_body_subterm;
+ Eterm single_body_subterm_key;
+ Eterm* single_body_subterm_key_tpl;
+
+ if (!is_list(body)) {
+ return 0;
+ }
+
+ body_list = list_val(body);
+ if (CDR(body_list) != NIL) {
+ return 0;
+ }
+
+ single_body_term = CAR(body_list);
+ if (single_body_term == am_DollarUnderscore) {
+ /* same tuple is returned */
+ return 1;
+ }
+
+ if (!is_tuple(single_body_term)) {
+ return 0;
+ }
+
+ single_body_term_tpl = tuple_val(single_body_term);
+ if (arityval(*single_body_term_tpl) != 1) {
+ // not the 1-element tuple we're expecting
+ return 0;
+ }
+
+ match_key = db_getkey(keypos, match);
+ if (!is_value(match_key)) {
+ // can't get key out of match
+ return 0;
+ }
+
+ single_body_subterm = single_body_term_tpl[1];
+ single_body_subterm_key = db_getkey(keypos, single_body_subterm);
+ if (!is_value(single_body_subterm_key)) {
+ // can't get key out of single body subterm
+ return 0;
+ }
+
+ if (db_match_eq_body(match_key, single_body_subterm_key)) {
+ /* tuple with same key is returned */
+ return 1;
+ }
+
+ if (!is_tuple(single_body_subterm_key)) {
+ /* can't possibly be an element instruction */
+ return 0;
+ }
+
+ single_body_subterm_key_tpl = tuple_val(single_body_subterm_key);
+ if (arityval(*single_body_subterm_key_tpl) != 3) {
+ /* can't possibly be an element instruction */
+ return 0;
+ }
+
+ if (single_body_subterm_key_tpl[1] == am_element &&
+ single_body_subterm_key_tpl[3] == am_DollarUnderscore &&
+ single_body_subterm_key_tpl[2] == make_small(keypos))
+ {
+ /* {element, KeyPos, '$_'} */
+ return 1;
+ }
+
+ return 0;
+}
+
/* This is used when tracing */
Eterm erts_match_set_lint(Process *p, Eterm matchexpr) {
return db_match_set_lint(p, matchexpr, DCOMP_TRACE);
@@ -2172,7 +2343,7 @@ restart:
}
}
else {
- *esp = term;
+ *esp++ = term;
}
break;
case matchPushArrayAsList:
@@ -5161,6 +5332,7 @@ void db_free_tmp_uncompressed(DbTerm* obj)
Eterm db_match_dbterm(DbTableCommon* tb, Process* c_p, Binary* bprog,
int all, DbTerm* obj, Eterm** hpp, Uint extra)
{
+ enum erts_pam_run_flags flags;
Uint32 dummy;
Eterm res;
@@ -5168,9 +5340,13 @@ Eterm db_match_dbterm(DbTableCommon* tb, Process* c_p, Binary* bprog,
obj = db_alloc_tmp_uncompressed(tb, obj);
}
+ flags = (hpp ?
+ ERTS_PAM_COPY_RESULT | ERTS_PAM_CONTIGUOUS_TUPLE :
+ ERTS_PAM_TMP_RESULT | ERTS_PAM_CONTIGUOUS_TUPLE);
+
res = db_prog_match(c_p, c_p,
bprog, make_tuple(obj->tpl), NULL, 0,
- ERTS_PAM_COPY_RESULT|ERTS_PAM_CONTIGUOUS_TUPLE, &dummy);
+ flags, &dummy);
if (is_value(res) && hpp!=NULL) {
*hpp = HAlloc(c_p, extra);
diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h
index 471fefe3cb..9be77fcefa 100644
--- a/erts/emulator/beam/erl_db_util.h
+++ b/erts/emulator/beam/erl_db_util.h
@@ -75,9 +75,6 @@ typedef struct db_term {
*/
} DbTerm;
-union db_table;
-typedef union db_table DbTable;
-
#define DB_MUST_RESIZE 1
#define DB_NEW_OBJECT 2
#define DB_INC_TRY_GROW 4
@@ -138,43 +135,56 @@ typedef struct db_table_method
Eterm slot,
Eterm* ret);
int (*db_select_chunk)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
+ Eterm tid,
Eterm pattern,
Sint chunk_size,
int reverse,
Eterm* ret);
int (*db_select)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
+ Eterm tid,
Eterm pattern,
int reverse,
Eterm* ret);
int (*db_select_delete)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
+ Eterm tid,
Eterm pattern,
Eterm* ret);
int (*db_select_continue)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
Eterm continuation,
Eterm* ret);
int (*db_select_delete_continue)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
Eterm continuation,
Eterm* ret);
int (*db_select_count)(Process* p,
- DbTable* tb, /* [in out] */
+ DbTable* tb, /* [in out] */
+ Eterm tid,
Eterm pattern,
Eterm* ret);
int (*db_select_count_continue)(Process* p,
DbTable* tb, /* [in out] */
Eterm continuation,
Eterm* ret);
+ int (*db_select_replace)(Process* p,
+ DbTable* tb, /* [in out] */
+ Eterm tid,
+ Eterm pattern,
+ Eterm* ret);
+ int (*db_select_replace_continue)(Process* p,
+ DbTable* tb, /* [in out] */
+ Eterm continuation,
+ Eterm* ret);
int (*db_take)(Process *, DbTable *, Eterm, Eterm *);
int (*db_delete_all_objects)(Process* p,
DbTable* db /* [in out] */ );
int (*db_free_table)(DbTable* db /* [in out] */ );
- int (*db_free_table_continue)(DbTable* db); /* [in out] */
+ SWord (*db_free_table_continue)(DbTable* db, SWord reds);
void (*db_print)(fmtfn_t to,
void* to_arg,
@@ -184,7 +194,6 @@ typedef struct db_table_method
void (*db_foreach_offheap)(DbTable* db, /* [in out] */
void (*func)(ErlOffHeap *, void *),
void *arg);
- void (*db_check_table)(DbTable* tb);
/* Lookup a dbterm for updating. Return false if not found. */
int (*db_lookup_dbterm)(Process *, DbTable *, Eterm key, Eterm obj,
@@ -198,11 +207,27 @@ typedef struct db_table_method
} DbTableMethod;
typedef struct db_fixation {
- Eterm pid;
+ /* Node in fixed_tabs list */
+ struct {
+ struct db_fixation *next, *prev;
+ Binary* btid;
+ } tabs;
+
+ /* Node in fixing_procs tree */
+ struct {
+ struct db_fixation *left, *right, *parent;
+ int is_red;
+ Process* p;
+ } procs;
+
Uint counter;
- struct db_fixation *next;
} DbFixation;
+typedef struct {
+ DbTable *next;
+ DbTable *prev;
+} DbTableList;
+
/*
* This structure contains data for all different types of database
* tables. Note that these fields must match the same fields
@@ -212,10 +237,13 @@ typedef struct db_fixation {
*/
typedef struct db_table_common {
- erts_smp_refc_t ref; /* fixation counter */
+ erts_smp_refc_t refc; /* reference count of table struct */
+ erts_smp_refc_t fix_count;/* fixation counter */
+ DbTableList all;
+ DbTableList owned;
#ifdef ERTS_SMP
erts_smp_rwmtx_t rwlock; /* rw lock on table */
- erts_smp_mtx_t fixlock; /* Protects fixations,megasec,sec,microsec */
+ erts_smp_mtx_t fixlock; /* Protects fixing_procs and time */
int is_thread_safe; /* No fine locking inside table needed */
Uint32 type; /* table type, *read only* after creation */
#endif
@@ -224,7 +252,7 @@ typedef struct db_table_common {
UWord heir_data; /* To send in ETS-TRANSFER (is_immed or (DbTerm*) */
Uint64 heir_started_interval; /* To further identify the heir */
Eterm the_name; /* an atom */
- Eterm id; /* atom | integer */
+ Binary *btid;
DbTableMethod* meth; /* table methods */
erts_smp_atomic_t nitems; /* Total number of items in table */
erts_smp_atomic_t memory_size;/* Total memory size. NOTE: in bytes! */
@@ -232,36 +260,35 @@ typedef struct db_table_common {
ErtsMonotonicTime monotonic;
ErtsMonotonicTime offset;
} time;
- DbFixation* fixations; /* List of processes who have done safe_fixtable,
+ DbFixation* fixing_procs; /* Tree of processes who have done safe_fixtable,
"local" fixations not included. */
/* All 32-bit fields */
Uint32 status; /* bit masks defined below */
- int slot; /* slot index in meta_main_tab */
int keypos; /* defaults to 1 */
int compress;
} DbTableCommon;
/* These are status bit patterns */
-#define DB_NORMAL (1 << 0)
-#define DB_PRIVATE (1 << 1)
-#define DB_PROTECTED (1 << 2)
-#define DB_PUBLIC (1 << 3)
-#define DB_BAG (1 << 4)
-#define DB_SET (1 << 5)
-/*#define DB_LHASH (1 << 6)*/
-#define DB_FINE_LOCKED (1 << 7) /* fine grained locking enabled */
-#define DB_DUPLICATE_BAG (1 << 8)
-#define DB_ORDERED_SET (1 << 9)
-#define DB_DELETE (1 << 10) /* table is being deleted */
-#define DB_FREQ_READ (1 << 11)
-
-#define ERTS_ETS_TABLE_TYPES (DB_BAG|DB_SET|DB_DUPLICATE_BAG|DB_ORDERED_SET|DB_FINE_LOCKED|DB_FREQ_READ)
+#define DB_PRIVATE (1 << 0)
+#define DB_PROTECTED (1 << 1)
+#define DB_PUBLIC (1 << 2)
+#define DB_DELETE (1 << 3) /* table is being deleted */
+#define DB_SET (1 << 4)
+#define DB_BAG (1 << 5)
+#define DB_DUPLICATE_BAG (1 << 6)
+#define DB_ORDERED_SET (1 << 7)
+#define DB_FINE_LOCKED (1 << 8) /* write_concurrency */
+#define DB_FREQ_READ (1 << 9) /* read_concurrency */
+#define DB_NAMED_TABLE (1 << 10)
+
+#define ERTS_ETS_TABLE_TYPES (DB_BAG|DB_SET|DB_DUPLICATE_BAG|DB_ORDERED_SET\
+ |DB_FINE_LOCKED|DB_FREQ_READ|DB_NAMED_TABLE)
#define IS_HASH_TABLE(Status) (!!((Status) & \
(DB_BAG | DB_SET | DB_DUPLICATE_BAG)))
#define IS_TREE_TABLE(Status) (!!((Status) & \
DB_ORDERED_SET))
-#define NFIXED(T) (erts_smp_refc_read(&(T)->common.ref,0))
+#define NFIXED(T) (erts_smp_refc_read(&(T)->common.fix_count,0))
#define IS_FIXED(T) (NFIXED(T) != 0)
/*
@@ -356,6 +383,7 @@ Eterm db_add_counter(Eterm** hpp, Wterm counter, Eterm incr);
Eterm db_match_set_lint(Process *p, Eterm matchexpr, Uint flags);
Binary *db_match_set_compile(Process *p, Eterm matchexpr,
Uint flags);
+int db_match_keeps_key(int keypos, Eterm match, Eterm guard, Eterm body);
int erts_db_match_prog_destructor(Binary *);
typedef struct match_prog {
@@ -506,14 +534,5 @@ erts_db_get_match_prog_binary(Eterm term)
#define Binary2MatchProg(BP) \
(ASSERT(IsMatchProgBinary((BP))), \
((MatchProg *) ERTS_MAGIC_BIN_DATA((BP))))
-/*
-** Debugging
-*/
-#ifdef HARDDEBUG
-void db_check_tables(void); /* in db.c */
-#define CHECK_TABLES() db_check_tables()
-#else
-#define CHECK_TABLES()
-#endif
#endif /* _DB_UTIL_H */
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index b2c307e826..56c0f19230 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -61,6 +61,9 @@
#define ERTS_DEFAULT_NO_ASYNC_THREADS 10
+#define ERTS_DEFAULT_SCHED_STACK_SIZE 256
+#define ERTS_MIN_SCHED_STACK_SIZE 20
+
/*
* The variables below (prefixed with etp_) are for erts/etc/unix/etp-commands
* only. Do not remove even though they aren't used elsewhere in the emulator!
@@ -125,6 +128,8 @@ const Eterm etp_hole_marker = ERTS_HOLE_MARKER;
const Eterm etp_hole_marker = 0;
#endif
+static int modified_sched_thread_suggested_stack_size = 0;
+
/*
* Note about VxWorks: All variables must be initialized by executable code,
* not by an initializer. Otherwise a new instance of the emulator will
@@ -1237,24 +1242,38 @@ early_init(int *argc, char **argv) /*
}
#ifndef ERTS_SMP
+
+void *erts_scheduler_stack_limit;
+
+
static void set_main_stack_size(void)
{
- if (erts_sched_thread_suggested_stack_size > 0) {
+ char c;
+ UWord stacksize;
# if HAVE_DECL_GETRLIMIT && HAVE_DECL_SETRLIMIT && HAVE_DECL_RLIMIT_STACK
- struct rlimit rl;
- int bytes = erts_sched_thread_suggested_stack_size * sizeof(Uint) * 1024;
- if (getrlimit(RLIMIT_STACK, &rl) != 0 ||
- (rl.rlim_cur = bytes, setrlimit(RLIMIT_STACK, &rl) != 0)) {
- erts_fprintf(stderr, "failed to set stack size for scheduler "
- "thread to %d bytes\n", bytes);
- erts_usage();
- }
+ struct rlimit rl;
+ int bytes;
+ stacksize = erts_sched_thread_suggested_stack_size * sizeof(Uint) * 1024;
+ /* Add some extra pages... neede by some systems... */
+ bytes = (int) stacksize + 3*erts_sys_get_page_size();
+ if (getrlimit(RLIMIT_STACK, &rl) != 0 ||
+ (rl.rlim_cur = bytes, setrlimit(RLIMIT_STACK, &rl) != 0)) {
+ erts_fprintf(stderr, "failed to set stack size for scheduler "
+ "thread to %d bytes\n", bytes);
+ erts_usage();
+ }
# else
+ if (modified_sched_thread_suggested_stack_size) {
erts_fprintf(stderr, "no OS support for dynamic stack size limit\n");
- erts_usage();
-# endif
+ erts_usage();
}
+ /* Be conservative and hope it is not more than 64 kWords... */
+ stacksize = 64*1024*sizeof(void *);
+# endif
+
+ erts_scheduler_stack_limit = erts_calc_stacklimit(&c, stacksize);
}
+
#endif
void
@@ -1299,12 +1318,11 @@ erl_start(int argc, char **argv)
port_tab_sz_ignore_files = 1;
}
-#if (defined(__APPLE__) && defined(__MACH__)) || defined(__DARWIN__)
/*
- * The default stack size on MacOS X is too small for pcre.
+ * A default stack size suitable for pcre which might use quite
+ * a lot of stack.
*/
- erts_sched_thread_suggested_stack_size = 256;
-#endif
+ erts_sched_thread_suggested_stack_size = ERTS_DEFAULT_SCHED_STACK_SIZE;
#ifdef DEBUG
verbose = DEBUG_DEFAULT;
@@ -1927,6 +1945,7 @@ erl_start(int argc, char **argv)
/* suggested stack size (Kilo Words) for scheduler threads */
arg = get_arg(sub_param+2, argv[i+1], &i);
erts_sched_thread_suggested_stack_size = atoi(arg);
+ modified_sched_thread_suggested_stack_size = 1;
if ((erts_sched_thread_suggested_stack_size
< ERTS_SCHED_THREAD_MIN_STACK_SIZE)
@@ -2236,6 +2255,9 @@ erl_start(int argc, char **argv)
boot_argc = argc - i; /* Number of arguments to init */
boot_argv = &argv[i];
+ if (erts_sched_thread_suggested_stack_size < ERTS_MIN_SCHED_STACK_SIZE)
+ erts_sched_thread_suggested_stack_size = ERTS_MIN_SCHED_STACK_SIZE;
+
erl_init(ncpu,
proc_tab_sz,
legacy_proc_tab,
@@ -2307,6 +2329,7 @@ erl_start(int argc, char **argv)
#endif
set_main_stack_size();
erts_sched_init_time_sup(esdp);
+ erts_ets_sched_spec_data_init(esdp);
process_main(esdp->x_reg_array, esdp->f_reg_array);
}
#endif
diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c
index 6ff9aea5ab..da73469516 100644
--- a/erts/emulator/beam/erl_lock_check.c
+++ b/erts/emulator/beam/erl_lock_check.c
@@ -100,14 +100,12 @@ static erts_lc_lock_order_t erts_lock_order[] = {
{ "dist_entry_links", "address" },
{ "code_write_permission", NULL },
{ "purge_state", NULL },
+ { "meta_name_tab", "address" },
+ { "db_tab", "address" },
{ "proc_status", "pid" },
{ "proc_trace", "pid" },
{ "ports_snapshot", NULL },
- { "meta_name_tab", "address" },
- { "meta_main_tab_slot", "address" },
- { "db_tab", "address" },
{ "db_tab_fix", "address" },
- { "meta_main_tab_main", NULL },
{ "db_hash_slot", "address" },
{ "node_table", NULL },
{ "dist_table", NULL },
diff --git a/erts/emulator/beam/erl_mtrace.c b/erts/emulator/beam/erl_mtrace.c
index e275867928..bb6f8660f1 100644
--- a/erts/emulator/beam/erl_mtrace.c
+++ b/erts/emulator/beam/erl_mtrace.c
@@ -572,7 +572,7 @@ void erts_mtrace_pre_init(void)
void erts_mtrace_init(char *receiver, char *nodename)
{
- char hostname[MAXHOSTNAMELEN];
+ char hostname[MAXHOSTNAMELEN + 1];
char pid[21]; /* enough for a 64 bit number */
socket_desc = ERTS_SOCK_INVALID_SOCKET;
@@ -613,9 +613,10 @@ void erts_mtrace_init(char *receiver, char *nodename)
}
tracep = trace_buffer;
endp = trace_buffer + TRACE_BUF_SZ;
- if (erts_sock_gethostname(hostname, MAXHOSTNAMELEN) != 0)
+ /* gethostname requires that the len is max(hostname) + 1 */
+ if (erts_sock_gethostname(hostname, MAXHOSTNAMELEN + 1) != 0)
hostname[0] = '\0';
- hostname[MAXHOSTNAMELEN-1] = '\0';
+ hostname[MAXHOSTNAMELEN] = '\0';
sys_get_pid(pid, sizeof(pid));
write_trace_header(nodename ? nodename : "", pid, hostname);
erts_mtrace_update_heap_size();
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index e6da4c1a76..f86b9739fa 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -998,16 +998,14 @@ int enif_inspect_binary(ErlNifEnv* env, Eterm bin_term, ErlNifBinary* bin)
byte* raw_ptr;
}u;
- if (is_boxed(bin_term) && *binary_val(bin_term) == HEADER_SUB_BIN) {
- ErlSubBin* sb = (ErlSubBin*) binary_val(bin_term);
- if (sb->is_writable) {
- ProcBin* pb = (ProcBin*) binary_val(sb->orig);
- ASSERT(pb->thing_word == HEADER_PROC_BIN);
- if (pb->flags) {
- erts_emasculate_writable_binary(pb);
- sb->is_writable = 0;
- }
- }
+ if (is_binary(bin_term)) {
+ ProcBin *pb = (ProcBin*) binary_val(bin_term);
+ if (pb->thing_word == HEADER_SUB_BIN) {
+ ErlSubBin* sb = (ErlSubBin*) pb;
+ pb = (ProcBin*) binary_val(sb->orig);
+ }
+ if (pb->thing_word == HEADER_PROC_BIN && pb->flags)
+ erts_emasculate_writable_binary(pb);
}
u.tmp = NULL;
bin->data = erts_get_aligned_binary_bytes_extra(bin_term, &u.raw_ptr, allocator,
@@ -1024,7 +1022,7 @@ int enif_inspect_binary(ErlNifEnv* env, Eterm bin_term, ErlNifBinary* bin)
bin->bin_term = bin_term;
bin->size = binary_size(bin_term);
bin->ref_bin = NULL;
- ADD_READONLY_CHECK(env, bin->data, bin->size);
+ ADD_READONLY_CHECK(env, bin->data, bin->size);
return 1;
}
@@ -2820,14 +2818,19 @@ enif_thread_type(void)
if (!esdp)
return ERL_NIF_THR_UNDEFINED;
- if (!ERTS_SCHEDULER_IS_DIRTY(esdp))
+ switch (esdp->type) {
+ case ERTS_SCHED_NORMAL:
return ERL_NIF_THR_NORMAL_SCHEDULER;
-
- if (ERTS_SCHEDULER_IS_DIRTY_CPU(esdp))
+#ifdef ERTS_DIRTY_SCHEDULERS
+ case ERTS_SCHED_DIRTY_CPU:
return ERL_NIF_THR_DIRTY_CPU_SCHEDULER;
-
- ASSERT(ERTS_SCHEDULER_IS_DIRTY_IO(esdp));
- return ERL_NIF_THR_DIRTY_IO_SCHEDULER;
+ case ERTS_SCHED_DIRTY_IO:
+ return ERL_NIF_THR_DIRTY_IO_SCHEDULER;
+#endif
+ default:
+ ERTS_INTERNAL_ERROR("Invalid scheduler type");
+ return -1;
+ }
}
/* Maps */
diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c
index e2072fe30f..89b627aaf5 100644
--- a/erts/emulator/beam/erl_node_tables.c
+++ b/erts/emulator/beam/erl_node_tables.c
@@ -850,14 +850,15 @@ static Eterm AM_timer;
static Eterm AM_delayed_delete_timer;
static void setup_reference_table(void);
-static Eterm reference_table_term(Uint **hpp, Uint *szp);
+static Eterm reference_table_term(Uint **hpp, ErlOffHeap *ohp, Uint *szp);
static void delete_reference_table(void);
-#if BIG_UINT_HEAP_SIZE > 3 /* 2-tuple */
-#define ID_HEAP_SIZE BIG_UINT_HEAP_SIZE
-#else
-#define ID_HEAP_SIZE 3 /* 2-tuple */
-#endif
+#undef ERTS_MAX__
+#define ERTS_MAX__(A, B) ((A) > (B) ? (A) : (B))
+
+#define ID_HEAP_SIZE \
+ ERTS_MAX__(ERTS_MAGIC_REF_THING_SIZE, \
+ ERTS_MAX__(BIG_UINT_HEAP_SIZE, 3))
typedef struct node_referrer_ {
struct node_referrer_ *next;
@@ -870,6 +871,7 @@ typedef struct node_referrer_ {
int system_ref;
Eterm id;
Uint id_heap[ID_HEAP_SIZE];
+ ErlOffHeap off_heap;
} NodeReferrer;
typedef struct {
@@ -942,7 +944,7 @@ erts_get_node_and_dist_references(struct process *proc)
/* Get term size */
size = 0;
- (void) reference_table_term(NULL, &size);
+ (void) reference_table_term(NULL, NULL, &size);
hp = HAlloc(proc, size);
#ifdef DEBUG
@@ -951,7 +953,7 @@ erts_get_node_and_dist_references(struct process *proc)
#endif
/* Write term */
- res = reference_table_term(&hp, NULL);
+ res = reference_table_term(&hp, &proc->off_heap, NULL);
ASSERT(endp == hp);
@@ -1048,13 +1050,14 @@ insert_node_referrer(ReferredNode *referred_node, int type, Eterm id)
nrp = (NodeReferrer *) erts_alloc(ERTS_ALC_T_NC_TMP,
sizeof(NodeReferrer));
nrp->next = referred_node->referrers;
+ ERTS_INIT_OFF_HEAP(&nrp->off_heap);
referred_node->referrers = nrp;
if(IS_CONST(id))
nrp->id = id;
else {
Uint *hp = &nrp->id_heap[0];
- ASSERT(is_big(id) || is_tuple(id));
- nrp->id = copy_struct(id, size_object(id), &hp, NULL);
+ ASSERT(is_big(id) || is_tuple(id) || is_internal_magic_ref(id));
+ nrp->id = copy_struct(id, size_object(id), &hp, &nrp->off_heap);
}
nrp->heap_ref = 0;
nrp->link_ref = 0;
@@ -1211,10 +1214,20 @@ insert_links2(ErtsLink *lnk, Eterm id)
static void
insert_ets_table(DbTable *tab, void *unused)
{
+ ErlOffHeap off_heap;
+ Eterm heap[ERTS_MAGIC_REF_THING_SIZE];
struct insert_offheap2_arg a;
a.type = ETS_REF;
- a.id = tab->common.id;
+ if (tab->common.status & DB_NAMED_TABLE)
+ a.id = tab->common.the_name;
+ else {
+ Eterm *hp = &heap[0];
+ ERTS_INIT_OFF_HEAP(&off_heap);
+ a.id = erts_mk_magic_ref(&hp, &off_heap, tab->common.btid);
+ }
erts_db_foreach_offheap(tab, insert_offheap2, (void *) &a);
+ if (is_not_atom(a.id))
+ erts_cleanup_offheap(&off_heap);
}
static void
@@ -1518,7 +1531,7 @@ setup_reference_table(void)
*/
static Eterm
-reference_table_term(Uint **hpp, Uint *szp)
+reference_table_term(Uint **hpp, ErlOffHeap *ohp, Uint *szp)
{
#undef MK_2TUP
#undef MK_3TUP
@@ -1573,12 +1586,11 @@ reference_table_term(Uint **hpp, Uint *szp)
nrid = nrp->id;
if (!IS_CONST(nrp->id)) {
-
Uint nrid_sz = size_object(nrp->id);
if (szp)
*szp += nrid_sz;
if (hpp)
- nrid = copy_struct(nrp->id, nrid_sz, hpp, NULL);
+ nrid = copy_struct(nrp->id, nrid_sz, hpp, ohp);
}
if (is_internal_pid(nrid) || nrid == am_error_logger) {
@@ -1713,6 +1725,7 @@ delete_reference_table(void)
NodeReferrer *tnrp;
nrp = referred_nodes[i].referrers;
while(nrp) {
+ erts_cleanup_offheap(&nrp->off_heap);
tnrp = nrp;
nrp = nrp->next;
erts_free(ERTS_ALC_T_NC_TMP, (void *) tnrp);
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 52d9f9ddf7..11a1ce625b 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -449,7 +449,8 @@ typedef enum {
ERTS_PSTT_CPC, /* Check Process Code */
ERTS_PSTT_CLA, /* Copy Literal Area */
ERTS_PSTT_COHMQ, /* Change off heap message queue */
- ERTS_PSTT_FTMQ /* Flush trace msg queue */
+ ERTS_PSTT_FTMQ, /* Flush trace msg queue */
+ ERTS_PSTT_ETS_FREE_FIXATION
} ErtsProcSysTaskType;
#define ERTS_MAX_PROC_SYS_TASK_ARGS 2
@@ -602,6 +603,7 @@ dbg_chk_aux_work_val(erts_aint32_t value)
valid |= ERTS_SSI_AUX_WORK_REAP_PORTS;
#endif
valid |= ERTS_SSI_AUX_WORK_DEBUG_WAIT_COMPLETED;
+ valid |= ERTS_SSI_AUX_WORK_YIELD;
if (~valid & value)
erts_exit(ERTS_ABORT_EXIT,
@@ -690,6 +692,8 @@ erts_pre_init_process(void)
= "SET_TMO";
erts_aux_work_flag_descr[ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK_IX]
= "MSEG_CACHE_CHECK";
+ erts_aux_work_flag_descr[ERTS_SSI_AUX_WORK_YIELD_IX]
+ = "YIELD";
erts_aux_work_flag_descr[ERTS_SSI_AUX_WORK_REAP_PORTS_IX]
= "REAP_PORTS";
erts_aux_work_flag_descr[ERTS_SSI_AUX_WORK_DEBUG_WAIT_COMPLETED_IX]
@@ -726,6 +730,16 @@ erts_pre_init_process(void)
= ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS;
erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].set_locks
= ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS;
+
+ erts_psd_required_locks[ERTS_PSD_ETS_OWNED_TABLES].get_locks
+ = ERTS_PSD_ETS_OWNED_TABLES_GET_LOCKS;
+ erts_psd_required_locks[ERTS_PSD_ETS_OWNED_TABLES].set_locks
+ = ERTS_PSD_ETS_OWNED_TABLES_SET_LOCKS;
+
+ erts_psd_required_locks[ERTS_PSD_ETS_FIXED_TABLES].get_locks
+ = ERTS_PSD_ETS_FIXED_TABLES_GET_LOCKS;
+ erts_psd_required_locks[ERTS_PSD_ETS_FIXED_TABLES].set_locks
+ = ERTS_PSD_ETS_FIXED_TABLES_SET_LOCKS;
#endif
}
@@ -2557,6 +2571,48 @@ handle_reap_ports(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waiting)
return aux_work & ~ERTS_SSI_AUX_WORK_REAP_PORTS;
}
+void
+erts_notify_new_aux_yield_work(ErtsSchedulerData *esdp)
+{
+ ASSERT(esdp == erts_get_scheduler_data());
+ /* Always called by the scheduler itself... */
+ set_aux_work_flags_wakeup_nob(esdp->ssi, ERTS_SSI_AUX_WORK_YIELD);
+}
+
+static ERTS_INLINE erts_aint32_t
+handle_yield(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waiting)
+{
+ int yield = 0;
+ /*
+ * Yield operations are always requested by the scheduler itself.
+ *
+ * The following handlers should *not* set the ERTS_SSI_AUX_WORK_YIELD
+ * flag in order to indicate more work. They should instead return
+ * information so this "main handler" can manipulate the flag...
+ *
+ * The following handlers should be able to handle being called
+ * even though no work is to be done...
+ */
+
+ /* Various yielding operations... */
+
+ yield |= erts_handle_yielded_ets_all_request(awdp->esdp,
+ &awdp->yield.ets_all);
+
+ /*
+ * Other yielding operations...
+ *
+ */
+
+ if (!yield) {
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_YIELD);
+ return aux_work & ~ERTS_SSI_AUX_WORK_YIELD;
+ }
+
+ return aux_work;
+}
+
+
#if HAVE_ERTS_MSEG
static ERTS_INLINE erts_aint32_t
@@ -2697,6 +2753,9 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
handle_mseg_cache_check);
#endif
+ HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_YIELD,
+ handle_yield);
+
HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_REAP_PORTS,
handle_reap_ports);
@@ -5888,14 +5947,30 @@ erts_sched_set_wake_cleanup_threshold(char *str)
static void
init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp)
{
- if (!esdp)
- awdp->sched_id = 0;
+ int id = 0;
+ if (esdp) {
+ switch (esdp->type) {
+ case ERTS_SCHED_NORMAL:
+ id = (int) esdp->no;
+ break;
#ifdef ERTS_DIRTY_SCHEDULERS
- else if (ERTS_SCHEDULER_IS_DIRTY(esdp))
- awdp->sched_id = (int) ERTS_DIRTY_SCHEDULER_NO(esdp);
+ case ERTS_SCHED_DIRTY_CPU:
+ id = (int) erts_no_schedulers;
+ id += (int) esdp->dirty_no;
+ break;
+ case ERTS_SCHED_DIRTY_IO:
+ id = (int) erts_no_schedulers;
+ id += (int) erts_no_dirty_cpu_schedulers;
+ id += (int) esdp->dirty_no;
+ break;
#endif
- else
- awdp->sched_id = (int) esdp->no;
+ default:
+ ERTS_INTERNAL_ERROR("Invalid scheduler type");
+ break;
+ }
+ }
+
+ awdp->sched_id = id;
awdp->esdp = esdp;
awdp->ssi = esdp ? esdp->ssi : NULL;
#ifdef ERTS_SMP
@@ -5968,7 +6043,7 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
ASSERT(runq == ERTS_DIRTY_IO_RUNQ);
esdp->type = ERTS_SCHED_DIRTY_IO;
}
- ERTS_DIRTY_SCHEDULER_NO(esdp) = (Uint) num;
+ esdp->dirty_no = (Uint) num;
if (num == 1) {
/*
* Multi-scheduling block functionality depends
@@ -5980,7 +6055,7 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
else {
esdp->type = ERTS_SCHED_NORMAL;
esdp->no = (Uint) num;
- ERTS_DIRTY_SCHEDULER_NO(esdp) = 0;
+ esdp->dirty_no = 0;
runq->scheduler = esdp;
}
esdp->dirty_shadow_process = shadow_proc;
@@ -7717,11 +7792,11 @@ suspend_scheduler(ErtsSchedulerData *esdp)
break;
case ERTS_SCHED_DIRTY_CPU:
online_flag = ERTS_SCHDLR_SSPND_CHNG_DCPU_ONLN;
- no = ERTS_DIRTY_SCHEDULER_NO(esdp);
+ no = esdp->dirty_no;
break;
case ERTS_SCHED_DIRTY_IO:
online_flag = 0;
- no = ERTS_DIRTY_SCHEDULER_NO(esdp);
+ no = esdp->dirty_no;
break;
default:
ERTS_INTERNAL_ERROR("Invalid scheduler type");
@@ -8727,6 +8802,8 @@ sched_thread_func(void *vesdp)
ERTS_VERIFY_UNUSED_TEMP_ALLOC(NULL);
#endif
+ erts_ets_sched_spec_data_init(esdp);
+
process_main(esdp->x_reg_array, esdp->f_reg_array);
/* No schedulers should *ever* terminate */
@@ -8743,8 +8820,7 @@ sched_dirty_cpu_thread_func(void *vesdp)
{
ErtsThrPrgrCallbacks callbacks;
ErtsSchedulerData *esdp = vesdp;
- Uint no = ERTS_DIRTY_SCHEDULER_NO(esdp);
- ERTS_DIRTY_SCHEDULER_TYPE(esdp) = ERTS_DIRTY_CPU_SCHEDULER;
+ Uint no = esdp->dirty_no;
ASSERT(no != 0);
ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch();
callbacks.arg = (void *) esdp->ssi;
@@ -8792,8 +8868,7 @@ sched_dirty_io_thread_func(void *vesdp)
{
ErtsThrPrgrCallbacks callbacks;
ErtsSchedulerData *esdp = vesdp;
- Uint no = ERTS_DIRTY_SCHEDULER_NO(esdp);
- ERTS_DIRTY_SCHEDULER_TYPE(esdp) = ERTS_DIRTY_IO_SCHEDULER;
+ Uint no = esdp->dirty_no;
ASSERT(no != 0);
ERTS_DIRTY_IO_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch();
callbacks.arg = (void *) esdp->ssi;
@@ -11148,6 +11223,12 @@ execute_sys_tasks(Process *c_p, erts_aint32_t *statep, int in_reds)
st_res = am_true;
break;
#endif
+#ifdef ERTS_SMP
+ case ERTS_PSTT_ETS_FREE_FIXATION:
+ reds -= erts_db_execute_free_fixation(c_p, (DbFixation*)st->arg[0]);
+ st_res = am_true;
+ break;
+#endif
default:
ERTS_INTERNAL_ERROR("Invalid process sys task type");
st_res = am_false;
@@ -11199,7 +11280,8 @@ cleanup_sys_tasks(Process *c_p, erts_aint32_t in_state, int in_reds)
case ERTS_PSTT_GC_MAJOR:
case ERTS_PSTT_GC_MINOR:
case ERTS_PSTT_CPC:
- case ERTS_PSTT_COHMQ:
+ case ERTS_PSTT_COHMQ:
+ case ERTS_PSTT_ETS_FREE_FIXATION:
st_res = am_false;
break;
case ERTS_PSTT_CLA:
@@ -11553,13 +11635,12 @@ erts_internal_request_system_task_4(BIF_ALIST_4)
}
static void
-erts_schedule_generic_sys_task(Eterm pid, ErtsProcSysTaskType type)
+erts_schedule_generic_sys_task(Eterm pid, ErtsProcSysTaskType type, void* arg)
{
Process *rp = erts_proc_lookup(pid);
if (rp) {
ErtsProcSysTask *st;
erts_aint32_t state, fail_state;
- int i;
st = erts_alloc(ERTS_ALC_T_PROC_SYS_TSK,
ERTS_PROC_SYS_TASK_SIZE(0));
@@ -11568,8 +11649,7 @@ erts_schedule_generic_sys_task(Eterm pid, ErtsProcSysTaskType type)
st->reply_tag = NIL;
st->req_id = NIL;
st->req_id_sz = 0;
- for (i = 0; i < ERTS_MAX_PROC_SYS_TASK_ARGS; i++)
- st->arg[i] = NIL;
+ st->arg[0] = (Eterm)arg;
ERTS_INIT_OFF_HEAP(&st->off_heap);
state = erts_smp_atomic32_read_nob(&rp->state);
@@ -11585,7 +11665,13 @@ erts_schedule_generic_sys_task(Eterm pid, ErtsProcSysTaskType type)
void
erts_schedule_complete_off_heap_message_queue_change(Eterm pid)
{
- erts_schedule_generic_sys_task(pid, ERTS_PSTT_COHMQ);
+ erts_schedule_generic_sys_task(pid, ERTS_PSTT_COHMQ, NULL);
+}
+
+void
+erts_schedule_ets_free_fixation(Eterm pid, DbFixation* fix)
+{
+ erts_schedule_generic_sys_task(pid, ERTS_PSTT_ETS_FREE_FIXATION, fix);
}
#ifdef ERTS_DIRTY_SCHEDULERS
@@ -11633,7 +11719,7 @@ erts_schedule_flush_trace_messages(Process *proc, int force_on_proc)
dhndl = erts_thr_progress_unmanaged_delay();
#endif
- erts_schedule_generic_sys_task(pid, ERTS_PSTT_FTMQ);
+ erts_schedule_generic_sys_task(pid, ERTS_PSTT_FTMQ, NULL);
#ifdef ERTS_SMP
erts_thr_progress_unmanaged_continue(dhndl);
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index 8bf372dad5..883d9f2a4c 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -63,6 +63,9 @@ typedef struct process Process;
#define ERTS_ONLY_INCLUDE_TRACE_FLAGS
#include "erl_trace.h"
#undef ERTS_ONLY_INCLUDE_TRACE_FLAGS
+#define ERTS_ONLY_SCHED_SPEC_ETS_DATA
+#include "erl_db.h"
+#undef ERTS_ONLY_SCHED_SPEC_ETS_DATA
#ifdef HIPE
#include "hipe_process.h"
@@ -312,6 +315,7 @@ typedef enum {
ERTS_SSI_AUX_WORK_PENDING_EXITERS_IX,
ERTS_SSI_AUX_WORK_SET_TMO_IX,
ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK_IX,
+ ERTS_SSI_AUX_WORK_YIELD_IX,
ERTS_SSI_AUX_WORK_REAP_PORTS_IX,
ERTS_SSI_AUX_WORK_DEBUG_WAIT_COMPLETED_IX, /* SHOULD be last flag index */
@@ -348,6 +352,8 @@ typedef enum {
(((erts_aint32_t) 1) << ERTS_SSI_AUX_WORK_SET_TMO_IX)
#define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK \
(((erts_aint32_t) 1) << ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK_IX)
+#define ERTS_SSI_AUX_WORK_YIELD \
+ (((erts_aint32_t) 1) << ERTS_SSI_AUX_WORK_YIELD_IX)
#define ERTS_SSI_AUX_WORK_REAP_PORTS \
(((erts_aint32_t) 1) << ERTS_SSI_AUX_WORK_REAP_PORTS_IX)
#define ERTS_SSI_AUX_WORK_DEBUG_WAIT_COMPLETED \
@@ -613,6 +619,10 @@ typedef struct {
} delayed_wakeup;
#endif
struct {
+ ErtsEtsAllYieldData ets_all;
+ /* Other yielding operations... */
+ } yield;
+ struct {
struct {
erts_aint32_t flags;
void (*callback)(void *);
@@ -621,19 +631,16 @@ typedef struct {
} debug;
} ErtsAuxWorkData;
+#define ERTS_SCHED_AUX_YIELD_DATA(ESDP, NAME) \
+ (&(ESDP)->aux_work_data.yield.NAME)
+void erts_notify_new_aux_yield_work(ErtsSchedulerData *esdp);
+
#ifdef ERTS_DIRTY_SCHEDULERS
typedef enum {
ERTS_DIRTY_CPU_SCHEDULER,
ERTS_DIRTY_IO_SCHEDULER
} ErtsDirtySchedulerType;
-typedef union {
- struct {
- ErtsDirtySchedulerType type: 1;
- Uint num: sizeof(Uint)*8 - 1;
- } s;
- Uint no;
-} ErtsDirtySchedId;
#endif
struct ErtsSchedulerData_ {
@@ -660,7 +667,7 @@ struct ErtsSchedulerData_ {
ErtsSchedType type;
Uint no; /* Scheduler number for normal schedulers */
#ifdef ERTS_DIRTY_SCHEDULERS
- ErtsDirtySchedId dirty_no; /* Scheduler number for dirty schedulers */
+ Uint dirty_no; /* Scheduler number for dirty schedulers */
Process *dirty_shadow_process;
#endif
Port *current_port;
@@ -688,7 +695,7 @@ struct ErtsSchedulerData_ {
ErtsSchedWallTime sched_wall_time;
ErtsGCInfo gc_info;
ErtsPortTaskHandle nosuspend_port_task_handle;
-
+ ErtsEtsTables ets_tables;
#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
erts_alloc_verify_func_t verify_unused_temp_alloc;
Allctr_t *verify_unused_temp_alloc_data;
@@ -822,14 +829,16 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi)
#define ERTS_PSD_CALL_TIME_BP 3
#define ERTS_PSD_DELAYED_GC_TASK_QS 4
#define ERTS_PSD_NIF_TRAP_EXPORT 5
-#define ERTS_PSD_SUSPENDED_SAVED_CALLS_BUF 6
+#define ERTS_PSD_ETS_OWNED_TABLES 6
+#define ERTS_PSD_ETS_FIXED_TABLES 7
+#define ERTS_PSD_SUSPENDED_SAVED_CALLS_BUF 8
-#define ERTS_PSD_SIZE 7
+#define ERTS_PSD_SIZE 9
#if !defined(HIPE)
# undef ERTS_PSD_SUSPENDED_SAVED_CALLS_BUF
# undef ERTS_PSD_SIZE
-# define ERTS_PSD_SIZE 6
+# define ERTS_PSD_SIZE 8
#endif
typedef struct {
@@ -857,6 +866,12 @@ typedef struct {
#define ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN
#define ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN
+#define ERTS_PSD_ETS_OWNED_TABLES_GET_LOCKS ERTS_PROC_LOCK_STATUS
+#define ERTS_PSD_ETS_OWNED_TABLES_SET_LOCKS ERTS_PROC_LOCK_STATUS
+
+#define ERTS_PSD_ETS_FIXED_TABLES_GET_LOCKS ERTS_PROC_LOCK_MAIN
+#define ERTS_PSD_ETS_FIXED_TABLES_SET_LOCKS ERTS_PROC_LOCK_MAIN
+
typedef struct {
ErtsProcLocks get_locks;
ErtsProcLocks set_locks;
@@ -1556,23 +1571,13 @@ extern int erts_system_profile_ts_type;
#define ERTS_DIRTY_IO_SCHEDULER_IX(IX) \
(ASSERT(0 <= (IX) && (IX) < erts_no_dirty_io_schedulers), \
&erts_aligned_dirty_io_scheduler_data[(IX)].esd)
-#define ERTS_DIRTY_SCHEDULER_NO(ESDP) \
- ((ESDP)->dirty_no.s.num)
-#define ERTS_DIRTY_SCHEDULER_TYPE(ESDP) \
- ((ESDP)->dirty_no.s.type)
-#ifdef ERTS_SMP
#define ERTS_SCHEDULER_IS_DIRTY(ESDP) \
- ((ESDP)->dirty_no.s.num != 0)
+ ((ESDP)->type != ERTS_SCHED_NORMAL)
#define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP) \
- (ERTS_SCHEDULER_IS_DIRTY((ESDP)) & ((ESDP)->dirty_no.s.type == 0))
+ ((ESDP)->type == ERTS_SCHED_DIRTY_CPU)
#define ERTS_SCHEDULER_IS_DIRTY_IO(ESDP) \
- (ERTS_SCHEDULER_IS_DIRTY((ESDP)) & ((ESDP)->dirty_no.s.type == 1))
-#else
-#define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0
-#define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP) 0
-#define ERTS_SCHEDULER_IS_DIRTY_IO(ESDP) 0
-#endif
-#else
+ ((ESDP)->type == ERTS_SCHED_DIRTY_IO)
+#else /* !ERTS_DIRTY_SCHEDULERS */
#define ERTS_RUNQ_IX_IS_DIRTY(IX) 0
#define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0
#define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP) 0
@@ -1782,6 +1787,8 @@ void erts_schedule_thr_prgr_later_cleanup_op(void (*)(void *),
ErtsThrPrgrLaterOp *,
UWord);
void erts_schedule_complete_off_heap_message_queue_change(Eterm pid);
+struct db_fixation;
+void erts_schedule_ets_free_fixation(Eterm pid, struct db_fixation*);
void erts_schedule_flush_trace_messages(Process *proc, int force_on_proc);
int erts_flush_trace_messages(Process *c_p, ErtsProcLocks locks);
diff --git a/erts/emulator/beam/erl_rbtree.h b/erts/emulator/beam/erl_rbtree.h
index 5fefaea978..6a42853957 100644
--- a/erts/emulator/beam/erl_rbtree.h
+++ b/erts/emulator/beam/erl_rbtree.h
@@ -105,7 +105,10 @@
* <ERTS_RBT_PREFIX>_rbt_yield_state_t.
*
* The yield state should be statically initialized by
- * ERTS_RBT_YIELD_STAT_INITER.
+ * ERTS_RBT_YIELD_STAT_INITER
+ *
+ * or dynamically initialized with
+ * ERTS_RBT_YIELD_STAT_INIT(<ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate)
*
*
* The following API functions are implemented if corresponding
@@ -178,8 +181,8 @@
* Operate by calling the operator 'op' on each element.
* Order is undefined.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed. The tree should not be
* modified until all of it has been processed.
*
@@ -195,8 +198,8 @@
* Order is undefined. Each element should be destroyed
* by 'op'.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed.
*
* 'arg' is passed as argument to 'op'.
@@ -228,8 +231,8 @@
* Operate by calling the operator 'op' on each element from
* smallest towards larger elements.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed. The tree should not be
* modified until all of it has been processed.
*
@@ -244,8 +247,8 @@
* Operate by calling the operator 'op' on each element from
* largest towards smaller elements.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed. The tree should not be
* modified until all of it has been processed.
*
@@ -296,8 +299,8 @@
* Note that elements are often destroyed in another order
* than the order that the elements are operated on.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed. The tree should not be
* modified until all of it has been processed.
*
@@ -318,8 +321,8 @@
* Note that elements are often destroyed in another order
* than the order that the elements are operated on.
*
- * Yield when 'ylimit' elements has been processed. Zero is
- * returned when yielding, and a non-zero value is returned when
+ * Yield when 'ylimit' elements has been processed. True is
+ * returned when yielding, and false is returned when
* the whole tree has been processed. The tree should not be
* modified until all of it has been processed.
*
@@ -422,6 +425,13 @@
#ifndef ERTS_RBT_YIELD_STAT_INITER
# define ERTS_RBT_YIELD_STAT_INITER {NULL, 0}
#endif
+#ifndef ERTS_RBT_YIELD_STAT_INIT
+# define ERTS_RBT_YIELD_STAT_INIT(YS) \
+ do { \
+ (YS)->x = NULL; \
+ (YS)->up = 0; \
+ } while (0)
+#endif
#define ERTS_RBT_CONCAT_MACRO_VALUES___(X, Y) \
X ## Y
@@ -476,12 +486,12 @@ typedef struct {
#if defined(ERTS_RBT_HARD_DEBUG) \
&& (defined(ERTS_RBT_WANT_DELETE) \
|| defined(ERTS_RBT_NEED_INSERT__))
-static void ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root);
+static void ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root, ERTS_RBT_T *node);
# define ERTS_RBT_NEED_HDBG_CHECK_TREE__
-# define ERTS_RBT_HDBG_CHECK_TREE__(R) \
- ERTS_RBT_FUNC__(hdbg_check_tree)((R))
+# define ERTS_RBT_HDBG_CHECK_TREE__(R,N) \
+ ERTS_RBT_FUNC__(hdbg_check_tree)((R),(N))
#else
-# define ERTS_RBT_HDBG_CHECK_TREE__(R) ((void) 1)
+# define ERTS_RBT_HDBG_CHECK_TREE__(R,N) ((void) 1)
#endif
#ifdef ERTS_RBT_NEED_ROTATE__
@@ -634,7 +644,7 @@ ERTS_RBT_FUNC__(delete)(ERTS_RBT_T **root, ERTS_RBT_T *n)
ERTS_RBT_T null_x; /* null_x is used to get the fixup started when we
splice out a node without children. */
- ERTS_RBT_HDBG_CHECK_TREE__(*root);
+ ERTS_RBT_HDBG_CHECK_TREE__(*root, n);
ERTS_RBT_INIT_EMPTY_TNODE(&null_x);
@@ -852,7 +862,7 @@ ERTS_RBT_FUNC__(delete)(ERTS_RBT_T **root, ERTS_RBT_T *n)
}
}
- ERTS_RBT_HDBG_CHECK_TREE__(*root);
+ ERTS_RBT_HDBG_CHECK_TREE__(*root, NULL);
}
@@ -982,7 +992,7 @@ ERTS_RBT_FUNC__(insert_aux__)(ERTS_RBT_T **root, ERTS_RBT_T *n, int lookup)
{
ERTS_RBT_KEY_T kn = ERTS_RBT_GET_KEY(n);
- ERTS_RBT_HDBG_CHECK_TREE__(*root);
+ ERTS_RBT_HDBG_CHECK_TREE__(*root, NULL);
ERTS_RBT_INIT_EMPTY_TNODE(n);
@@ -1004,7 +1014,7 @@ ERTS_RBT_FUNC__(insert_aux__)(ERTS_RBT_T **root, ERTS_RBT_T *n, int lookup)
if (lookup && ERTS_RBT_IS_EQ(kn, kx)) {
- ERTS_RBT_HDBG_CHECK_TREE__(*root);
+ ERTS_RBT_HDBG_CHECK_TREE__(*root, NULL);
return x;
}
@@ -1038,7 +1048,7 @@ ERTS_RBT_FUNC__(insert_aux__)(ERTS_RBT_T **root, ERTS_RBT_T *n, int lookup)
ERTS_RBT_FUNC__(insert_fixup__)(root, n);
}
- ERTS_RBT_HDBG_CHECK_TREE__(*root);
+ ERTS_RBT_HDBG_CHECK_TREE__(*root, n);
return NULL;
}
@@ -1364,7 +1374,7 @@ ERTS_RBT_FUNC__(foreach_ordered__)(ERTS_RBT_T **root,
ystate->x = NULL;
ystate->up = 0;
}
- return 1; /* Done */
+ return 0; /* Done */
}
x = p;
}
@@ -1579,15 +1589,17 @@ ERTS_RBT_FUNC__(debug_print)(FILE *filep, ERTS_RBT_T *x, int indent,
#ifdef ERTS_RBT_NEED_HDBG_CHECK_TREE__
static void
-ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root)
+ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root, ERTS_RBT_T *n)
{
int black_depth = -1, no_black = 0;
ERTS_RBT_T *c, *p, *x = root;
ERTS_RBT_KEY_T kx;
ERTS_RBT_KEY_T kc;
- if (!x)
+ if (!x) {
+ ERTS_RBT_ASSERT(!n);
return;
+ }
ERTS_RBT_ASSERT(!ERTS_RBT_GET_PARENT(x));
@@ -1597,6 +1609,9 @@ ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root)
while (1) {
+ if (x == n)
+ n = NULL;
+
if (ERTS_RBT_IS_BLACK(x))
no_black++;
else {
@@ -1668,6 +1683,7 @@ ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root)
if (!p) {
ERTS_RBT_ASSERT(root == x);
ERTS_RBT_ASSERT(no_black == 0);
+ ERTS_RBT_ASSERT(!n);
return; /* Done */
}
diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h
index a602a8f7c6..097d580d99 100644
--- a/erts/emulator/beam/erl_term.h
+++ b/erts/emulator/beam/erl_term.h
@@ -873,6 +873,12 @@ typedef union {
ErtsORefThing o;
} ErtsRefThing;
+/* for copy sharing */
+#define BOXED_VISITED_MASK ((Eterm) 3)
+#define BOXED_VISITED ((Eterm) 1)
+#define BOXED_SHARED_UNPROCESSED ((Eterm) 2)
+#define BOXED_SHARED_PROCESSED ((Eterm) 3)
+
#define ERTS_REF_THING_SIZE (sizeof(ErtsORefThing)/sizeof(Uint))
#define ERTS_MAGIC_REF_THING_SIZE (sizeof(ErtsMRefThing)/sizeof(Uint))
#define ERTS_MAX_INTERNAL_REF_SIZE (sizeof(ErtsRefThing)/sizeof(Uint))
@@ -888,9 +894,14 @@ typedef union {
# define is_ref_thing_header(x) ((x) == ERTS_REF_THING_HEADER)
-#define is_ordinary_ref_thing(x) \
- (ASSERT(is_ref_thing_header(*((Eterm *)(x)))), \
+#ifdef SHCOPY
+#define is_ordinary_ref_thing(x) \
+ (((ErtsRefThing *) (x))->o.marker == ERTS_ORDINARY_REF_MARKER)
+#else
+#define is_ordinary_ref_thing(x) \
+ (ASSERT(is_ref_thing_header((*((Eterm *)(x))) & ~BOXED_VISITED_MASK)), \
((ErtsRefThing *) (x))->o.marker == ERTS_ORDINARY_REF_MARKER)
+#endif
#define is_magic_ref_thing(x) \
(!is_ordinary_ref_thing((x)))
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index c4c848f49f..139394680a 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -1136,6 +1136,11 @@ void erts_short_init(void);
void erl_start(int, char**);
void erts_usage(void);
Eterm erts_preloaded(Process* p);
+
+#ifndef ERTS_SMP
+extern void *erts_scheduler_stack_limit;
+#endif
+
/* erl_md5.c */
typedef struct {
@@ -1198,6 +1203,11 @@ Uint64 erts_timestamp_millis(void);
Export* erts_find_function(Eterm, Eterm, unsigned int, ErtsCodeIndex);
+void *erts_calc_stacklimit(char *prev_c, UWord stacksize);
+int erts_check_below_limit(char *ptr, char *limit);
+int erts_check_above_limit(char *ptr, char *limit);
+void *erts_ptr_id(void *ptr);
+
Eterm store_external_or_ref_in_proc_(Process *, Eterm);
Eterm store_external_or_ref_(Uint **, ErlOffHeap*, Eterm);
@@ -1233,6 +1243,11 @@ void erts_init_external(void);
/* erl_map.c */
void erts_init_map(void);
+/* beam_debug.c */
+UWord erts_check_stack_recursion_downwards(char *start_c);
+UWord erts_check_stack_recursion_upwards(char *start_c);
+int erts_is_above_stack_limit(char *ptr);
+
/* erl_unicode.c */
void erts_init_unicode(void);
Sint erts_unicode_set_loop_limit(Sint limit);
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index ec36b23059..9b5bd7a749 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -620,6 +620,20 @@ test_heap Need u==1 | put_list Y=y x==0 x==0 => test_heap_1_put_list Need Y
%macro: test_heap_1_put_list TestHeapPutList -pack
test_heap_1_put_list I y
+#
+# is_tagged_tuple Fail=f Src=rxy Arity Atom=a
+#
+
+is_tagged_tuple Fail Literal=q Arity Atom => \
+ move Literal x | is_tagged_tuple Fail x Arity Atom
+is_tagged_tuple Fail=f c Arity Atom => jump Fail
+
+%macro:is_tagged_tuple IsTaggedTuple -fail_action
+
+is_tagged_tuple f r A a
+is_tagged_tuple f x A a
+is_tagged_tuple f y A a
+
# Test tuple & arity (head)
is_tuple Fail Literal=q => move Literal x | is_tuple Fail x
diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h
index c6ea8049c3..24de35696c 100644
--- a/erts/emulator/beam/sys.h
+++ b/erts/emulator/beam/sys.h
@@ -487,6 +487,12 @@ extern volatile int erts_break_requested;
void erts_do_break_handling(void);
#endif
+#if !defined(ERTS_SMP) && !defined(__WIN32__)
+extern volatile Uint erts_signal_state;
+#define ERTS_SIGNAL_STATE erts_signal_state
+void erts_handle_signal_state(void);
+#endif
+
#ifdef ERTS_SMP
extern erts_smp_atomic32_t erts_writing_erl_crash_dump;
extern erts_tsd_key_t erts_is_crash_dumping_key;
@@ -596,6 +602,8 @@ __decl_noreturn void __noreturn erts_exit(int n, char*, ...);
Eterm erts_check_io_info(void *p);
+UWord erts_sys_get_page_size(void);
+
/* Size of misc memory allocated from system dependent code */
Uint erts_sys_misc_mem_sz(void);
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 092a5320ba..8f3f48f38f 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -4835,6 +4835,53 @@ Uint64 erts_timestamp_millis(void)
#endif
}
+void *
+erts_calc_stacklimit(char *prev_c, UWord stacksize)
+{
+ /*
+ * We *don't* want this function inlined, i.e., it is
+ * risky to call this function from another function
+ * in utils.c
+ */
+
+ UWord pagesize = erts_sys_get_page_size();
+ char c;
+ char *start;
+ if (&c > prev_c) {
+ start = (char *) ((((UWord) prev_c) / pagesize) * pagesize);
+ return (void *) (start + stacksize);
+ }
+ else {
+ start = (char *) (((((UWord) prev_c) - 1) / pagesize + 1) * pagesize);
+ return (void *) (start - stacksize);
+ }
+}
+
+/*
+ * erts_check_below_limit() and
+ * erts_check_above_limit() are put
+ * in utils.c in order to prevent
+ * inlining.
+ */
+
+int
+erts_check_below_limit(char *ptr, char *limit)
+{
+ return ptr < limit;
+}
+
+int
+erts_check_above_limit(char *ptr, char *limit)
+{
+ return ptr > limit;
+}
+
+void *
+erts_ptr_id(void *ptr)
+{
+ return ptr;
+}
+
#ifdef DEBUG
/*
* Handy functions when using a debugger - don't use in the code!