diff options
Diffstat (limited to 'erts/emulator/beam/beam_load.c')
-rw-r--r-- | erts/emulator/beam/beam_load.c | 2373 |
1 files changed, 1479 insertions, 894 deletions
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index 41c1b5d2c2..23258dbe9c 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -1,18 +1,19 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2014. All Rights Reserved. + * Copyright Ericsson AB 1996-2017. All Rights Reserved. * - * The contents of this file are subject to the Erlang Public License, - * Version 1.1, (the "License"); you may not use this file except in - * compliance with the License. You should have received a copy of the - * Erlang Public License along with this software. If not, it can be - * retrieved online at http://www.erlang.org/. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at * - * Software distributed under the License is distributed on an "AS IS" - * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See - * the License for the specific language governing rights and limitations - * under the License. + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. * * %CopyrightEnd% */ @@ -31,16 +32,20 @@ #include "bif.h" #include "external.h" #include "beam_load.h" +#include "beam_bp.h" #include "big.h" #include "erl_bits.h" #include "beam_catches.h" #include "erl_binary.h" #include "erl_zlib.h" +#include "erl_map.h" +#include "erl_process_dict.h" #ifdef HIPE #include "hipe_bif0.h" #include "hipe_mode_switch.h" #include "hipe_arch.h" +#include "hipe_load.h" #endif ErlDrvBinary* erts_gzinflate_buffer(char*, int); @@ -78,7 +83,7 @@ ErlDrvBinary* erts_gzinflate_buffer(char*, int); typedef struct { Uint value; /* Value of label (NULL if not known yet). */ - Uint patches; /* Index (into code buffer) to first location + Sint patches; /* Index (into code buffer) to first location * which must be patched with the value of this label. */ #ifdef ERTS_SMP @@ -102,7 +107,7 @@ typedef struct { */ typedef struct genop { - int op; /* Opcode. */ + unsigned int op; /* Opcode. */ int arity; /* Number of arguments. */ GenOpArg def_args[MAX_OPARGS]; /* Default buffer for arguments. */ GenOpArg* a; /* The arguments. */ @@ -152,13 +157,15 @@ typedef struct { #define STR_CHUNK 2 #define IMP_CHUNK 3 #define EXP_CHUNK 4 -#define NUM_MANDATORY 5 +#define MIN_MANDATORY 1 +#define MAX_MANDATORY 5 #define LAMBDA_CHUNK 5 #define LITERAL_CHUNK 6 #define ATTR_CHUNK 7 #define COMPILE_CHUNK 8 #define LINE_CHUNK 9 +#define UTF8_ATOM_CHUNK 10 #define NUM_CHUNK_TYPES (sizeof(chunk_types)/sizeof(chunk_types[0])) @@ -168,9 +175,13 @@ typedef struct { static Uint chunk_types[] = { /* - * Mandatory chunk types -- these MUST be present. + * Atom chunk types -- Atom or AtU8 MUST be present. */ MakeIffId('A', 't', 'o', 'm'), /* 0 */ + + /* + * Mandatory chunk types -- these MUST be present. + */ MakeIffId('C', 'o', 'd', 'e'), /* 1 */ MakeIffId('S', 't', 'r', 'T'), /* 2 */ MakeIffId('I', 'm', 'p', 'T'), /* 3 */ @@ -184,6 +195,7 @@ static Uint chunk_types[] = { MakeIffId('A', 't', 't', 'r'), /* 7 */ MakeIffId('C', 'I', 'n', 'f'), /* 8 */ MakeIffId('L', 'i', 'n', 'e'), /* 9 */ + MakeIffId('A', 't', 'U', '8'), /* 10 */ }; /* @@ -204,10 +216,7 @@ typedef struct { typedef struct { Eterm term; /* The tagged term (in the heap). */ - Uint heap_size; /* (Exact) size on the heap. */ - SWord offset; /* Offset from temporary location to final. */ - ErlOffHeap off_heap; /* Start of linked list of ProcBins. */ - Eterm* heap; /* Heap for term. */ + ErlHeapFragment* heap_frags; } Literal; /* @@ -245,7 +254,7 @@ typedef struct { /* * This structure contains all information about the module being loaded. */ - +#define MD5_SIZE 16 typedef struct LoaderState { /* * The current logical file within the binary. @@ -283,16 +292,17 @@ typedef struct LoaderState { byte* code_start; /* Start of code file. */ unsigned code_size; /* Size of code file. */ int specific_op; /* Specific opcode (-1 if not found). */ - int num_functions; /* Number of functions in module. */ - int num_labels; /* Number of labels. */ - int code_buffer_size; /* Size of code buffer in words. */ - BeamInstr* code; /* Loaded code. */ - int ci; /* Current index into loaded code. */ + unsigned int num_functions; /* Number of functions in module. */ + unsigned int num_labels; /* Number of labels. */ + BeamCodeHeader* hdr; /* Loaded code header */ + BeamInstr* codev; /* Loaded code buffer */ + int codev_size; /* Size of code buffer in words. */ + int ci; /* Current index into loaded code buffer. */ Label* labels; StringPatch* string_patches; /* Linked list of position into string table to patch. */ BeamInstr catches; /* Linked list of catch_yf instructions. */ unsigned loaded_size; /* Final size of code when loaded. */ - byte mod_md5[16]; /* MD5 for module code. */ + byte mod_md5[MD5_SIZE]; /* MD5 for module code. */ int may_load_nif; /* true if NIFs may later be loaded for this module */ int on_load; /* Index in the code for the on_load function * (or 0 if there is no on_load function) @@ -302,13 +312,13 @@ typedef struct LoaderState { * Atom table. */ - int num_atoms; /* Number of atoms in atom table. */ + unsigned int num_atoms; /* Number of atoms in atom table. */ Eterm* atom; /* Atom table. */ - int num_exps; /* Number of exports. */ + unsigned int num_exps; /* Number of exports. */ ExportEntry* export; /* Pointer to export table. */ - int num_imports; /* Number of imports. */ + unsigned int num_imports; /* Number of imports. */ ImportEntry* import; /* Import entry (translated information). */ /* @@ -322,8 +332,8 @@ typedef struct LoaderState { * Lambda table. */ - int num_lambdas; /* Number of lambdas in table. */ - int lambdas_allocated; /* Size of allocated lambda table. */ + unsigned int num_lambdas; /* Number of lambdas in table. */ + unsigned int lambdas_allocated; /* Size of allocated lambda table. */ Lambda* lambdas; /* Pointer to lambdas. */ Lambda def_lambdas[16]; /* Default storage for lambda table. */ char* lambda_error; /* Delayed missing 'FunT' error. */ @@ -332,8 +342,8 @@ typedef struct LoaderState { * Literals (constant pool). */ - int num_literals; /* Number of literals in table. */ - int allocated_literals; /* Number of literal entries allocated. */ + unsigned int num_literals; /* Number of literals in table. */ + unsigned int allocated_literals; /* Number of literal entries allocated. */ Literal* literals; /* Array of literals. */ LiteralPatch* literal_patches; /* Operands that need to be patched. */ Uint total_literal_size; /* Total heap size for all literals. */ @@ -342,13 +352,13 @@ typedef struct LoaderState { * Line table. */ BeamInstr* line_item; /* Line items from the BEAM file. */ - int num_line_items; /* Number of line items. */ + unsigned int num_line_items;/* Number of line items. */ LineInstr* line_instr; /* Line instructions */ - int num_line_instrs; /* Maximum number of line instructions */ - int current_li; /* Current line instruction */ - int* func_line; /* Mapping from function to first line instr */ + unsigned int num_line_instrs; /* Maximum number of line instructions */ + unsigned int current_li; /* Current line instruction */ + unsigned int* func_line; /* Mapping from function to first line instr */ Eterm* fname; /* List of file names */ - int num_fnames; /* Number of filenames in fname table */ + unsigned int num_fnames; /* Number of filenames in fname table */ int loc_size; /* Size of location info in bytes (2/4) */ } LoaderState; @@ -476,15 +486,20 @@ typedef struct LoaderState { static void free_loader_state(Binary* magic); -static void loader_state_dtor(Binary* magic); -static Eterm insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, - Eterm group_leader, Eterm module, - BeamInstr* code, Uint size); +static ErlHeapFragment* new_literal_fragment(Uint size); +static void free_literal_fragment(ErlHeapFragment*); +static int loader_state_dtor(Binary* magic); +#ifdef HIPE +static Eterm stub_insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, + Eterm group_leader, Eterm module, + BeamCodeHeader* code_hdr, Uint size, + HipeModule *hipe_code); +#endif static int init_iff_file(LoaderState* stp, byte* code, Uint size); static int scan_iff_file(LoaderState* stp, Uint* chunk_types, - Uint num_types, Uint num_mandatory); + Uint num_types); static int verify_chunks(LoaderState* stp); -static int load_atom_table(LoaderState* stp); +static int load_atom_table(LoaderState* stp, ErtsAtomEncoding enc); static int load_import_table(LoaderState* stp); static int read_export_table(LoaderState* stp); static int is_bif(Eterm mod, Eterm func, unsigned arity); @@ -511,7 +526,7 @@ static GenOp* gen_get_map_element(LoaderState* stp, GenOpArg Fail, GenOpArg Src, static int freeze_code(LoaderState* stp); -static void final_touch(LoaderState* stp); +static void final_touch(LoaderState* stp, struct erl_module_instance* inst_p); static void short_file(int line, LoaderState* stp, unsigned needed); static void load_printf(int line, LoaderState* context, char *fmt, ...); static int transform_engine(LoaderState* st); @@ -524,13 +539,16 @@ static void new_literal_patch(LoaderState* stp, int pos); static void new_string_patch(LoaderState* stp, int pos); static Uint new_literal(LoaderState* stp, Eterm** hpp, Uint heap_size); static int genopargcompare(GenOpArg* a, GenOpArg* b); -static Eterm exported_from_module(Process* p, Eterm mod); -static Eterm functions_in_module(Process* p, Eterm mod); -static Eterm attributes_for_module(Process* p, Eterm mod); -static Eterm compilation_info_for_module(Process* p, Eterm mod); -static Eterm native_addresses(Process* p, Eterm mod); -int patch_funentries(Eterm Patchlist); -int patch(Eterm Addresses, Uint fe); +static Eterm get_module_info(Process* p, ErtsCodeIndex code_ix, + BeamCodeHeader*, Eterm module, Eterm what); +static Eterm exported_from_module(Process* p, ErtsCodeIndex code_ix, + Eterm mod); +static Eterm functions_in_module(Process* p, BeamCodeHeader*); +static Eterm attributes_for_module(Process* p, BeamCodeHeader*); +static Eterm compilation_info_for_module(Process* p, BeamCodeHeader*); +static Eterm md5_of_module(Process* p, BeamCodeHeader*); +static Eterm has_native(BeamCodeHeader*); +static Eterm native_addresses(Process* p, BeamCodeHeader*); static int safe_mul(UWord a, UWord b, UWord* resp); static int must_swap_floats; @@ -595,6 +613,7 @@ extern void check_allocated_block(Uint type, void *blk); #define CHKBLK(TYPE,BLK) /* nothing */ #endif + Eterm erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, Eterm* modp, byte* code, Uint unloaded_size) @@ -617,7 +636,7 @@ erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, CHKALLOC(); CHKBLK(ERTS_ALC_T_CODE,stp->code); if (!init_iff_file(stp, code, unloaded_size) || - !scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES, NUM_MANDATORY) || + !scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES) || !verify_chunks(stp)) { goto load_error; } @@ -635,28 +654,43 @@ erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, /* * Initialize code area. */ - stp->code_buffer_size = 2048 + stp->num_functions; - stp->code = (BeamInstr *) erts_alloc(ERTS_ALC_T_CODE, - sizeof(BeamInstr) * stp->code_buffer_size); + stp->codev_size = 2048 + stp->num_functions; + stp->hdr = (BeamCodeHeader*) erts_alloc(ERTS_ALC_T_CODE, + (offsetof(BeamCodeHeader,functions) + + sizeof(BeamInstr) * stp->codev_size)); - stp->code[MI_NUM_FUNCTIONS] = stp->num_functions; - stp->ci = MI_FUNCTIONS + stp->num_functions + 1; + stp->hdr->num_functions = stp->num_functions; - stp->code[MI_ATTR_PTR] = 0; - stp->code[MI_ATTR_SIZE] = 0; - stp->code[MI_ATTR_SIZE_ON_HEAP] = 0; - stp->code[MI_COMPILE_PTR] = 0; - stp->code[MI_COMPILE_SIZE] = 0; - stp->code[MI_COMPILE_SIZE_ON_HEAP] = 0; + /* Let the codev array start at functions[0] in order to index + * both function pointers and the loaded code itself that follows. + */ + stp->codev = (BeamInstr*) &stp->hdr->functions; + stp->ci = stp->num_functions + 1; + + stp->hdr->attr_ptr = NULL; + stp->hdr->attr_size = 0; + stp->hdr->attr_size_on_heap = 0; + stp->hdr->compile_ptr = NULL; + stp->hdr->compile_size = 0; + stp->hdr->compile_size_on_heap = 0; + stp->hdr->literal_area = NULL; + stp->hdr->md5_ptr = NULL; /* * Read the atom table. */ CHKBLK(ERTS_ALC_T_CODE,stp->code); - define_file(stp, "atom table", ATOM_CHUNK); - if (!load_atom_table(stp)) { - goto load_error; + if (stp->chunks[UTF8_ATOM_CHUNK].size > 0) { + define_file(stp, "utf8 atom table", UTF8_ATOM_CHUNK); + if (!load_atom_table(stp, ERTS_ATOM_ENC_UTF8)) { + goto load_error; + } + } else { + define_file(stp, "atom table", ATOM_CHUNK); + if (!load_atom_table(stp, ERTS_ATOM_ENC_LATIN1)) { + goto load_error; + } } /* @@ -750,8 +784,11 @@ Eterm erts_finish_loading(Binary* magic, Process* c_p, ErtsProcLocks c_p_locks, Eterm* modp) { - Eterm retval; + Eterm retval = NIL; LoaderState* stp = ERTS_MAGIC_BIN_DATA(magic); + Module* mod_tab_p; + struct erl_module_instance* inst_p; + Uint size; /* * No other process may run since we will update the export @@ -760,27 +797,87 @@ erts_finish_loading(Binary* magic, Process* c_p, ERTS_SMP_LC_ASSERT(erts_initialized == 0 || erts_has_code_write_permission() || erts_smp_thr_progress_is_blocking()); - /* * Make current code for the module old and insert the new code * as current. This will fail if there already exists old code * for the module. */ + mod_tab_p = erts_put_module(stp->module); CHKBLK(ERTS_ALC_T_CODE,stp->code); - retval = insert_new_code(c_p, c_p_locks, stp->group_leader, stp->module, - stp->code, stp->loaded_size); - if (retval != NIL) { - goto load_error; + if (!stp->on_load) { + /* + * Normal case -- no -on_load() function. + */ + retval = beam_make_current_old(c_p, c_p_locks, stp->module); + ASSERT(retval == NIL); + } else { + ErtsCodeIndex code_ix = erts_staging_code_ix(); + Eterm module = stp->module; + int i, num_exps; + + /* + * There is an -on_load() function. We will keep the current + * code, but we must turn off any tracing. + */ + num_exps = export_list_size(code_ix); + for (i = 0; i < num_exps; i++) { + Export *ep = export_list(i, code_ix); + if (ep == NULL || ep->info.mfa.module != module) { + continue; + } + if (ep->addressv[code_ix] == ep->beam) { + if (ep->beam[0] == (BeamInstr) em_apply_bif) { + continue; + } else if (ep->beam[0] == + (BeamInstr) BeamOp(op_i_generic_breakpoint)) { + ERTS_SMP_LC_ASSERT(erts_smp_thr_progress_is_blocking()); + ASSERT(mod_tab_p->curr.num_traced_exports > 0); + erts_clear_export_break(mod_tab_p, &ep->info); + ep->addressv[code_ix] = (BeamInstr *) ep->beam[1]; + ep->beam[1] = 0; + } + ASSERT(ep->beam[1] == 0); + } + } + ASSERT(mod_tab_p->curr.num_breakpoints == 0); + ASSERT(mod_tab_p->curr.num_traced_exports == 0); } /* + * Update module table. + */ + + size = stp->loaded_size; + erts_total_code_size += size; + + if (!stp->on_load) { + inst_p = &mod_tab_p->curr; + } else { + mod_tab_p->on_load = + (struct erl_module_instance *) + erts_alloc(ERTS_ALC_T_PREPARED_CODE, + sizeof(struct erl_module_instance)); + inst_p = mod_tab_p->on_load; + erts_module_instance_init(inst_p); + } + + inst_p->code_hdr = stp->hdr; + inst_p->code_length = size; + + /* + * Update ranges (used for finding a function from a PC value). + */ + + erts_update_ranges((BeamInstr*)inst_p->code_hdr, size); + + /* * Ready for the final touch: fixing the export table entries for * exported and imported functions. This can't fail. */ CHKBLK(ERTS_ALC_T_CODE,stp->code); - final_touch(stp); + final_touch(stp, inst_p); /* * Loading succeded. @@ -792,7 +889,8 @@ erts_finish_loading(Binary* magic, Process* c_p, debug_dump_code(stp->code,stp->ci); #endif #endif - stp->code = NULL; /* Prevent code from being freed. */ + stp->hdr = NULL; /* Prevent code from being freed. */ + stp->codev = NULL; *modp = stp->module; /* @@ -803,7 +901,6 @@ erts_finish_loading(Binary* magic, Process* c_p, retval = am_on_load; } - load_error: free_loader_state(magic); return retval; } @@ -816,7 +913,7 @@ erts_alloc_loader_state(void) magic = erts_create_magic_binary(sizeof(LoaderState), loader_state_dtor); - erts_refc_inc(&magic->refc, 1); + erts_refc_inc(&magic->intern.refc, 1); stp = ERTS_MAGIC_BIN_DATA(magic); stp->bin = NULL; stp->function = THE_NON_VALUE; /* Function not known yet */ @@ -824,7 +921,8 @@ erts_alloc_loader_state(void) stp->specific_op = -1; stp->genop = NULL; stp->atom = NULL; - stp->code = NULL; + stp->hdr = NULL; + stp->codev = NULL; stp->labels = NULL; stp->import = NULL; stp->export = NULL; @@ -860,29 +958,73 @@ erts_module_for_prepared_code(Binary* magic) LoaderState* stp; if (ERTS_MAGIC_BIN_DESTRUCTOR(magic) != loader_state_dtor) { +#ifdef HIPE + HipeLoaderState *hipe_stp; + if ((hipe_stp = hipe_get_loader_state(magic)) + && hipe_stp->text_segment != 0) { + return hipe_stp->module; + } +#endif return NIL; } stp = ERTS_MAGIC_BIN_DATA(magic); - if (stp->code != 0) { + if (stp->hdr != 0) { return stp->module; } else { return NIL; } } +/* + * Return a non-zero value if the module has an on_load function, + * or 0 if it does not. + */ + +Eterm +erts_has_code_on_load(Binary* magic) +{ + LoaderState* stp; + + if (ERTS_MAGIC_BIN_DESTRUCTOR(magic) != loader_state_dtor) { + return NIL; + } + stp = ERTS_MAGIC_BIN_DATA(magic); + return stp->on_load ? am_true : am_false; +} + static void free_loader_state(Binary* magic) { loader_state_dtor(magic); - if (erts_refc_dectest(&magic->refc, 0) == 0) { - erts_bin_free(magic); - } + erts_bin_release(magic); +} + +static ErlHeapFragment* new_literal_fragment(Uint size) +{ + ErlHeapFragment* bp; + bp = (ErlHeapFragment*) ERTS_HEAP_ALLOC(ERTS_ALC_T_PREPARED_CODE, + ERTS_HEAP_FRAG_SIZE(size)); + ERTS_INIT_HEAP_FRAG(bp, size, size); + return bp; +} + +static void free_literal_fragment(ErlHeapFragment* bp) +{ + ASSERT(bp != NULL); + do { + ErlHeapFragment* next_bp = bp->next; + + erts_cleanup_offheap(&bp->off_heap); + ERTS_HEAP_FREE(ERTS_ALC_T_PREPARED_CODE, (void *) bp, + ERTS_HEAP_FRAG_SIZE(bp->size)); + bp = next_bp; + }while (bp != NULL); } /* * This destructor function can safely be called multiple times. */ -static void +static int loader_state_dtor(Binary* magic) { LoaderState* stp = ERTS_MAGIC_BIN_DATA(magic); @@ -891,9 +1033,14 @@ loader_state_dtor(Binary* magic) driver_free_binary(stp->bin); stp->bin = 0; } - if (stp->code != 0) { - erts_free(ERTS_ALC_T_CODE, stp->code); - stp->code = 0; + if (stp->hdr != 0) { + if (stp->hdr->literal_area) { + erts_release_literal_area(stp->hdr->literal_area); + stp->hdr->literal_area = NULL; + } + erts_free(ERTS_ALC_T_CODE, stp->hdr); + stp->hdr = 0; + stp->codev = 0; } if (stp->labels != 0) { erts_free(ERTS_ALC_T_PREPARED_CODE, (void *) stp->labels); @@ -918,10 +1065,9 @@ loader_state_dtor(Binary* magic) if (stp->literals != 0) { int i; for (i = 0; i < stp->num_literals; i++) { - if (stp->literals[i].heap != 0) { - erts_free(ERTS_ALC_T_PREPARED_CODE, - (void *) stp->literals[i].heap); - stp->literals[i].heap = 0; + if (stp->literals[i].heap_frags != 0) { + free_literal_fragment(stp->literals[i].heap_frags); + stp->literals[i].heap_frags = 0; } } erts_free(ERTS_ALC_T_PREPARED_CODE, (void *) stp->literals); @@ -963,12 +1109,15 @@ loader_state_dtor(Binary* magic) */ ASSERT(stp->genop_blocks == 0); + return 1; } +#ifdef HIPE static Eterm -insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, - Eterm group_leader, Eterm module, BeamInstr* code, - Uint size) +stub_insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, + Eterm group_leader, Eterm module, + BeamCodeHeader* code_hdr, Uint size, + HipeModule *hipe_code) { Module* modp; Eterm retval; @@ -988,17 +1137,21 @@ insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, erts_total_code_size += size; modp = erts_put_module(module); - modp->curr.code = code; + modp->curr.code_hdr = code_hdr; modp->curr.code_length = size; modp->curr.catches = BEAM_CATCHES_NIL; /* Will be filled in later. */ + DBG_TRACE_MFA(make_atom(modp->module), 0, 0, "insert_new_code " + "first_hipe_ref = %p", hipe_code->first_hipe_ref); + modp->curr.hipe_code = hipe_code; /* * Update ranges (used for finding a function from a PC value). */ - erts_update_ranges(code, size); + erts_update_ranges((BeamInstr*)modp->curr.code_hdr, size); return NIL; } +#endif static int init_iff_file(LoaderState* stp, byte* code, Uint size) @@ -1072,7 +1225,7 @@ init_iff_file(LoaderState* stp, byte* code, Uint size) * Scan the IFF file. The header should have been verified by init_iff_file(). */ static int -scan_iff_file(LoaderState* stp, Uint* chunk_types, Uint num_types, Uint num_mandatory) +scan_iff_file(LoaderState* stp, Uint* chunk_types, Uint num_types) { Uint count; Uint id; @@ -1151,7 +1304,16 @@ verify_chunks(LoaderState* stp) MD5_CTX context; MD5Init(&context); - for (i = 0; i < NUM_MANDATORY; i++) { + + if (stp->chunks[UTF8_ATOM_CHUNK].start != NULL) { + MD5Update(&context, stp->chunks[UTF8_ATOM_CHUNK].start, stp->chunks[UTF8_ATOM_CHUNK].size); + } else if (stp->chunks[ATOM_CHUNK].start != NULL) { + MD5Update(&context, stp->chunks[ATOM_CHUNK].start, stp->chunks[ATOM_CHUNK].size); + } else { + LoadError0(stp, "mandatory chunk of type 'Atom' or 'AtU8' not found\n"); + } + + for (i = MIN_MANDATORY; i < MAX_MANDATORY; i++) { if (stp->chunks[i].start != NULL) { MD5Update(&context, stp->chunks[i].start, stp->chunks[i].size); } else { @@ -1212,9 +1374,9 @@ verify_chunks(LoaderState* stp) } static int -load_atom_table(LoaderState* stp) +load_atom_table(LoaderState* stp, ErtsAtomEncoding enc) { - int i; + unsigned int i; GetInt(stp, 4, stp->num_atoms); stp->num_atoms++; @@ -1231,7 +1393,7 @@ load_atom_table(LoaderState* stp) GetByte(stp, n); GetString(stp, atom, n); - stp->atom[i] = erts_atom_put(atom, n, ERTS_ATOM_ENC_LATIN1, 1); + stp->atom[i] = erts_atom_put(atom, n, enc, 1); } /* @@ -1259,13 +1421,13 @@ load_atom_table(LoaderState* stp) static int load_import_table(LoaderState* stp) { - int i; + unsigned int i; GetInt(stp, 4, stp->num_imports); stp->import = erts_alloc(ERTS_ALC_T_PREPARED_CODE, stp->num_imports * sizeof(ImportEntry)); for (i = 0; i < stp->num_imports; i++) { - int n; + unsigned int n; Eterm mod; Eterm func; Uint arity; @@ -1273,17 +1435,17 @@ load_import_table(LoaderState* stp) GetInt(stp, 4, n); if (n >= stp->num_atoms) { - LoadError2(stp, "import entry %d: invalid atom number %d", i, n); + LoadError2(stp, "import entry %u: invalid atom number %u", i, n); } mod = stp->import[i].module = stp->atom[n]; GetInt(stp, 4, n); if (n >= stp->num_atoms) { - LoadError2(stp, "import entry %d: invalid atom number %d", i, n); + LoadError2(stp, "import entry %u: invalid atom number %u", i, n); } func = stp->import[i].function = stp->atom[n]; GetInt(stp, 4, arity); if (arity > MAX_REG) { - LoadError2(stp, "import entry %d: invalid arity %d", i, arity); + LoadError2(stp, "import entry %u: invalid arity %d", i, arity); } stp->import[i].arity = arity; stp->import[i].patches = 0; @@ -1294,8 +1456,8 @@ load_import_table(LoaderState* stp) * the BIF function. */ if ((e = erts_active_export_entry(mod, func, arity)) != NULL) { - if (e->code[3] == (BeamInstr) em_apply_bif) { - stp->import[i].bf = (BifFunction) e->code[4]; + if (e->beam[0] == (BeamInstr) em_apply_bif) { + stp->import[i].bf = (BifFunction) e->beam[1]; if (func == am_load_nif && mod == am_erlang && arity == 2) { stp->may_load_nif = 1; } @@ -1311,12 +1473,12 @@ load_import_table(LoaderState* stp) static int read_export_table(LoaderState* stp) { - int i; + unsigned int i; BeamInstr* address; GetInt(stp, 4, stp->num_exps); if (stp->num_exps > stp->num_functions) { - LoadError2(stp, "%d functions exported; only %d functions defined", + LoadError2(stp, "%u functions exported; only %u functions defined", stp->num_exps, stp->num_functions); } stp->export @@ -1334,18 +1496,18 @@ read_export_table(LoaderState* stp) stp->export[i].function = func; GetInt(stp, 4, arity); if (arity > MAX_REG) { - LoadError2(stp, "export table entry %d: absurdly high arity %d", i, arity); + LoadError2(stp, "export table entry %u: absurdly high arity %u", i, arity); } stp->export[i].arity = arity; GetInt(stp, 4, n); if (n >= stp->num_labels) { - LoadError3(stp, "export table entry %d: invalid label %d (highest defined label is %d)", i, n, stp->num_labels); + LoadError3(stp, "export table entry %u: invalid label %u (highest defined label is %u)", i, n, stp->num_labels); } value = stp->labels[n].value; if (value == 0) { - LoadError2(stp, "export table entry %d: label %d not resolved", i, n); + LoadError2(stp, "export table entry %u: label %u not resolved", i, n); } - stp->export[i].address = address = stp->code + value; + stp->export[i].address = address = stp->codev + value; /* * Find out if there is a BIF with the same name. @@ -1364,7 +1526,7 @@ read_export_table(LoaderState* stp) * any other functions that walk through all local functions. */ - if (stp->labels[n].patches) { + if (stp->labels[n].patches >= 0) { LoadError3(stp, "there are local calls to the stub for " "the BIF %T:%T/%d", stp->module, func, arity); @@ -1388,7 +1550,7 @@ is_bif(Eterm mod, Eterm func, unsigned arity) if (e == NULL) { return 0; } - if (e->code[3] != (BeamInstr) em_apply_bif) { + if (e->beam[0] != (BeamInstr) em_apply_bif) { return 0; } if (mod == am_erlang && func == am_apply && arity == 3) { @@ -1404,7 +1566,7 @@ is_bif(Eterm mod, Eterm func, unsigned arity) static int read_lambda_table(LoaderState* stp) { - int i; + unsigned int i; GetInt(stp, 4, stp->num_lambdas); if (stp->num_lambdas > stp->lambdas_allocated) { @@ -1424,12 +1586,12 @@ read_lambda_table(LoaderState* stp) GetAtom(stp, n, stp->lambdas[i].function); GetInt(stp, 4, arity); if (arity > MAX_REG) { - LoadError2(stp, "lambda entry %d: absurdly high arity %d", i, arity); + LoadError2(stp, "lambda entry %u: absurdly high arity %u", i, arity); } stp->lambdas[i].arity = arity; GetInt(stp, 4, n); if (n >= stp->num_labels) { - LoadError3(stp, "lambda entry %d: invalid label %d (highest defined label is %d)", + LoadError3(stp, "lambda entry %u: invalid label %u (highest defined label is %u)", i, n, stp->num_labels); } stp->lambdas[i].label = n; @@ -1446,10 +1608,11 @@ read_lambda_table(LoaderState* stp) return 0; } + static int read_literal_table(LoaderState* stp) { - int i; + unsigned int i; uLongf uncompressed_sz; byte* uncompressed = 0; @@ -1467,36 +1630,46 @@ read_literal_table(LoaderState* stp) stp->allocated_literals = stp->num_literals; for (i = 0; i < stp->num_literals; i++) { - stp->literals[i].heap = 0; + stp->literals[i].heap_frags = 0; } for (i = 0; i < stp->num_literals; i++) { - int sz; + Uint sz; Sint heap_size; byte* p; Eterm val; - Eterm* hp; + ErtsHeapFactory factory; GetInt(stp, 4, sz); /* Size of external term format. */ GetString(stp, p, sz); if ((heap_size = erts_decode_ext_size(p, sz)) < 0) { - LoadError1(stp, "literal %d: bad external format", i); - } - hp = stp->literals[i].heap = erts_alloc(ERTS_ALC_T_PREPARED_CODE, - heap_size*sizeof(Eterm)); - stp->literals[i].off_heap.first = 0; - stp->literals[i].off_heap.overhead = 0; - val = erts_decode_ext(&hp, &stp->literals[i].off_heap, &p); - stp->literals[i].heap_size = hp - stp->literals[i].heap; - if (stp->literals[i].heap_size > heap_size) { - erl_exit(1, "overrun by %d word(s) for literal heap, term %d", - stp->literals[i].heap_size - heap_size, i); + LoadError1(stp, "literal %u: bad external format", i); } - if (is_non_value(val)) { - LoadError1(stp, "literal %d: bad external format", i); - } - stp->literals[i].term = val; - stp->total_literal_size += stp->literals[i].heap_size; + + if (heap_size > 0) { + erts_factory_heap_frag_init(&factory, + new_literal_fragment(heap_size)); + factory.alloc_type = ERTS_ALC_T_PREPARED_CODE; + val = erts_decode_ext(&factory, &p, 0); + + if (is_non_value(val)) { + LoadError1(stp, "literal %u: bad external format", i); + } + erts_factory_close(&factory); + stp->literals[i].heap_frags = factory.heap_frags; + stp->total_literal_size += erts_used_frag_sz(factory.heap_frags); + } + else { + erts_factory_dummy_init(&factory); + val = erts_decode_ext(&factory, &p, 0); + if (is_non_value(val)) { + LoadError1(stp, "literal %u: bad external format", i); + } + ASSERT(is_immed(val)); + stp->literals[i].heap_frags = NULL; + } + stp->literals[i].term = val; + } erts_free(ERTS_ALC_T_TMP, uncompressed); return 1; @@ -1513,9 +1686,9 @@ read_line_table(LoaderState* stp) { unsigned version; ERTS_DECLARE_DUMMY(unsigned flags); - int num_line_items; + unsigned int num_line_items; BeamInstr* lp; - int i; + unsigned int i; BeamInstr fname_index; BeamInstr tag; @@ -1594,7 +1767,7 @@ read_line_table(LoaderState* stp) } } else if (tag == TAG_a) { if (val > stp->num_fnames) { - LoadError2(stp, "file index overflow (%d/%d)", + LoadError2(stp, "file index overflow (%u/%u)", val, stp->num_fnames); } fname_index = val; @@ -1630,9 +1803,9 @@ read_line_table(LoaderState* stp) stp->num_line_instrs * sizeof(LineInstr)); stp->current_li = 0; - stp->func_line = (int *) erts_alloc(ERTS_ALC_T_PREPARED_CODE, - stp->num_functions * - sizeof(int)); + stp->func_line = (unsigned int *) erts_alloc(ERTS_ALC_T_PREPARED_CODE, + stp->num_functions * + sizeof(int)); return 1; @@ -1656,6 +1829,10 @@ read_code_header(LoaderState* stp) */ GetInt(stp, 4, head_size); + if (head_size > stp->file_left) { + LoadError2(stp, "invalid code header size %u; bytes left %u", + head_size, stp->file_left); + } stp->code_start = stp->file_p + head_size; stp->code_size = stp->file_left - head_size; stp->file_left = head_size; @@ -1696,7 +1873,7 @@ read_code_header(LoaderState* stp) stp->num_labels * sizeof(Label)); for (i = 0; i < stp->num_labels; i++) { stp->labels[i].value = 0; - stp->labels[i].patches = 0; + stp->labels[i].patches = -1; #ifdef ERTS_SMP stp->labels[i].looprec_targeted = 0; #endif @@ -1715,13 +1892,14 @@ read_code_header(LoaderState* stp) } else {} #define CodeNeed(w) do { \ - ASSERT(ci <= code_buffer_size); \ - if (code_buffer_size < ci+(w)) { \ - code_buffer_size = 2*ci+(w); \ - stp->code = code = \ - (BeamInstr *) erts_realloc(ERTS_ALC_T_CODE, \ - (void *) code, \ - code_buffer_size * sizeof(BeamInstr)); \ + ASSERT(ci <= codev_size); \ + if (codev_size < ci+(w)) { \ + codev_size = 2*ci+(w); \ + stp->hdr = (BeamCodeHeader*) erts_realloc(ERTS_ALC_T_CODE, \ + (void *) stp->hdr, \ + (offsetof(BeamCodeHeader,functions) \ + + codev_size * sizeof(BeamInstr))); \ + code = stp->codev = (BeamInstr*) &stp->hdr->functions; \ } \ } while (0) @@ -1737,7 +1915,7 @@ load_code(LoaderState* stp) int arg; /* Number of current argument. */ int num_specific; /* Number of specific ops for current. */ BeamInstr* code; - int code_buffer_size; + int codev_size; int specific; Uint last_label = 0; /* Number of last label. */ Uint function_number = 0; @@ -1751,26 +1929,26 @@ load_code(LoaderState* stp) * by both the nif functionality and line instructions. */ enum { - FUNC_INFO_SZ = 5 + FUNC_INFO_SZ = sizeof(ErtsCodeInfo) / sizeof(Eterm) }; - code = stp->code; - code_buffer_size = stp->code_buffer_size; + code = stp->codev; + codev_size = stp->codev_size; ci = stp->ci; for (;;) { - int new_op; + unsigned int new_op; GenOp* tmp_op; - ASSERT(ci <= code_buffer_size); + ASSERT(ci <= codev_size); get_next_instr: GetByte(stp, new_op); if (new_op >= NUM_GENERIC_OPS) { - LoadError1(stp, "invalid opcode %d", new_op); + LoadError1(stp, "invalid opcode %u", new_op); } if (gen_opc[new_op].name[0] == '\0') { - LoadError1(stp, "invalid opcode %d", new_op); + LoadError1(stp, "invalid opcode %u", new_op); } @@ -1808,9 +1986,7 @@ load_code(LoaderState* stp) case TAG_o: break; case TAG_x: - if (last_op->a[arg].val == 0) { - last_op->a[arg].type = TAG_r; - } else if (last_op->a[arg].val >= MAX_REG) { + if (last_op->a[arg].val >= MAX_REG) { LoadError1(stp, "invalid x register number: %u", last_op->a[arg].val); } @@ -1855,15 +2031,14 @@ load_code(LoaderState* stp) */ { Eterm* hp; -/* XXX:PaN - Halfword should use ARCH_64 variant instead */ -#if !defined(ARCH_64) || HALFWORD_HEAP +#if !defined(ARCH_64) Uint high, low; # endif last_op->a[arg].val = new_literal(stp, &hp, FLOAT_SIZE_OBJECT); hp[0] = HEADER_FLONUM; last_op->a[arg].type = TAG_q; -#if defined(ARCH_64) && !HALFWORD_HEAP +#if defined(ARCH_64) GetInt(stp, 8, hp[1]); # else GetInt(stp, 4, high); @@ -1959,42 +2134,47 @@ load_code(LoaderState* stp) ASSERT(arity == last_op->arity); do_transform: - if (stp->genop == NULL) { - last_op_next = NULL; - goto get_next_instr; - } - + ASSERT(stp->genop != NULL); if (gen_opc[stp->genop->op].transform != -1) { - int need; - tmp_op = stp->genop; - - for (need = gen_opc[stp->genop->op].min_window-1; need > 0; need--) { - if (tmp_op == NULL) { - goto get_next_instr; - } - tmp_op = tmp_op->next; + if (stp->genop->next == NULL) { + /* + * Simple heuristic: Most transformations requires + * at least two instructions, so make sure that + * there are. That will reduce the number of + * TE_SHORT_WINDOWs. + */ + goto get_next_instr; } switch (transform_engine(stp)) { case TE_FAIL: - last_op_next = NULL; - last_op = NULL; + /* + * No transformation found. stp->genop != NULL and + * last_op_next is still valid. Go ahead and load + * the instruction. + */ break; case TE_OK: + /* + * Some transformation was applied. last_op_next is + * no longer valid and stp->genop may be NULL. + * Try to transform again. + */ + if (stp->genop == NULL) { + last_op_next = &stp->genop; + goto get_next_instr; + } last_op_next = NULL; - last_op = NULL; goto do_transform; case TE_SHORT_WINDOW: - last_op_next = NULL; - last_op = NULL; + /* + * No transformation applied. stp->genop != NULL and + * last_op_next is still valid. Fetch a new instruction + * before trying the transformation again. + */ goto get_next_instr; } } - if (stp->genop == NULL) { - last_op_next = NULL; - goto get_next_instr; - } - /* * From the collected generic instruction, find the specific * instruction. @@ -2017,7 +2197,42 @@ load_code(LoaderState* stp) if (((opc[specific].mask[0] & mask[0]) == mask[0]) && ((opc[specific].mask[1] & mask[1]) == mask[1]) && ((opc[specific].mask[2] & mask[2]) == mask[2])) { - break; + + if (!opc[specific].involves_r) { + break; /* No complications - match */ + } + + /* + * The specific operation uses the 'r' operand, + * which is shorthand for x(0). Now things + * get complicated. First we must check whether + * all operands that should be of type 'r' use + * x(0) (as opposed to some other X register). + */ + for (arg = 0; arg < arity; arg++) { + if (opc[specific].involves_r & (1 << arg) && + tmp_op->a[arg].type == TAG_x) { + if (tmp_op->a[arg].val != 0) { + break; /* Other X register than 0 */ + } + } + } + + if (arg == arity) { + /* + * All 'r' operands use x(0) in the generic + * operation. That means a match. Now we + * will need to rewrite the generic instruction + * to actually use 'r' instead of 'x(0)'. + */ + for (arg = 0; arg < arity; arg++) { + if (opc[specific].involves_r & (1 << arg) && + tmp_op->a[arg].type == TAG_x) { + tmp_op->a[arg].type = TAG_r; + } + } + break; /* Match */ + } } specific++; } @@ -2128,14 +2343,11 @@ load_code(LoaderState* stp) break; case 's': /* Any source (tagged constant or register) */ switch (tag) { - case TAG_r: - code[ci++] = make_rreg(); - break; case TAG_x: - code[ci++] = make_xreg(tmp_op->a[arg].val); + code[ci++] = make_loader_x_reg(tmp_op->a[arg].val); break; case TAG_y: - code[ci++] = make_yreg(tmp_op->a[arg].val); + code[ci++] = make_loader_y_reg(tmp_op->a[arg].val); break; case TAG_i: code[ci++] = (BeamInstr) make_small((Uint)tmp_op->a[arg].val); @@ -2146,6 +2358,10 @@ load_code(LoaderState* stp) case TAG_n: code[ci++] = NIL; break; + case TAG_q: + new_literal_patch(stp, ci); + code[ci++] = tmp_op->a[arg].val; + break; default: LoadError1(stp, "bad tag %d for general source", tmp_op->a[arg].type); @@ -2154,14 +2370,11 @@ load_code(LoaderState* stp) break; case 'd': /* Destination (x(0), x(N), y(N) */ switch (tag) { - case TAG_r: - code[ci++] = make_rreg(); - break; case TAG_x: - code[ci++] = make_xreg(tmp_op->a[arg].val); + code[ci++] = tmp_op->a[arg].val * sizeof(Eterm); break; case TAG_y: - code[ci++] = make_yreg(tmp_op->a[arg].val); + code[ci++] = tmp_op->a[arg].val * sizeof(Eterm) + 1; break; default: LoadError1(stp, "bad tag %d for destination", @@ -2205,7 +2418,7 @@ load_code(LoaderState* stp) VerifyTag(stp, tag, TAG_u); last_label = tmp_op->a[arg].val; if (!(0 < last_label && last_label < stp->num_labels)) { - LoadError2(stp, "invalid label num %d (0 < label < %d)", + LoadError2(stp, "invalid label num %u (0 < label < %u)", tmp_op->a[arg].val, stp->num_labels); } if (stp->labels[last_label].value != 0) { @@ -2318,20 +2531,13 @@ load_code(LoaderState* stp) stp->labels[tmp_op->a[arg].val].patches = ci; ci++; break; - case TAG_r: - CodeNeed(1); - code[ci++] = (R_REG_DEF << _TAG_PRIMARY_SIZE) | - TAG_PRIMARY_HEADER; - break; case TAG_x: CodeNeed(1); - code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) | - (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER; + code[ci++] = make_loader_x_reg(tmp_op->a[arg].val); break; case TAG_y: CodeNeed(1); - code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) | - (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER; + code[ci++] = make_loader_y_reg(tmp_op->a[arg].val); break; case TAG_n: CodeNeed(1); @@ -2354,24 +2560,18 @@ load_code(LoaderState* stp) switch (stp->specific_op) { case op_i_func_info_IaaI: { - Uint offset; - + Sint offset; if (function_number >= stp->num_functions) { - LoadError1(stp, "too many functions in module (header said %d)", + LoadError1(stp, "too many functions in module (header said %u)", stp->num_functions); } if (stp->may_load_nif) { const int finfo_ix = ci - FUNC_INFO_SZ; -#ifdef ERTS_DIRTY_SCHEDULERS - enum { MIN_FUNC_SZ = 4 }; -#else - enum { MIN_FUNC_SZ = 3 }; -#endif - if (finfo_ix - last_func_start < MIN_FUNC_SZ && last_func_start) { + if (finfo_ix - last_func_start < BEAM_NIF_MIN_FUNC_SZ && last_func_start) { /* Must make room for call_nif op */ - int pad = MIN_FUNC_SZ - (finfo_ix - last_func_start); - ASSERT(pad > 0 && pad < MIN_FUNC_SZ); + int pad = BEAM_NIF_MIN_FUNC_SZ - (finfo_ix - last_func_start); + ASSERT(pad > 0 && pad < BEAM_NIF_MIN_FUNC_SZ); CodeNeed(pad); sys_memmove(&code[finfo_ix+pad], &code[finfo_ix], FUNC_INFO_SZ*sizeof(BeamInstr)); @@ -2396,16 +2596,19 @@ load_code(LoaderState* stp) stp->function = code[ci-2]; stp->arity = code[ci-1]; + /* When this assert is triggered, it is normally a sign that + the size of the ops.tab i_func_info instruction is not + the same as FUNC_INFO_SZ */ ASSERT(stp->labels[last_label].value == ci - FUNC_INFO_SZ); - offset = MI_FUNCTIONS + function_number; - code[offset] = stp->labels[last_label].patches; + stp->hdr->functions[function_number] = (ErtsCodeInfo*) stp->labels[last_label].patches; + offset = function_number; stp->labels[last_label].patches = offset; function_number++; if (stp->arity > MAX_ARG) { LoadError1(stp, "too many arguments: %d", stp->arity); } #ifdef DEBUG - ASSERT(stp->labels[0].patches == 0); /* Should not be referenced. */ + ASSERT(stp->labels[0].patches < 0); /* Should not be referenced. */ for (i = 1; i < stp->num_labels; i++) { ASSERT(stp->labels[i].patches < ci); } @@ -2419,7 +2622,6 @@ load_code(LoaderState* stp) stp->on_load = ci; break; case op_bs_put_string_II: - case op_i_bs_match_string_rfII: case op_i_bs_match_string_xfII: new_string_patch(stp, ci-1); break; @@ -2437,14 +2639,14 @@ load_code(LoaderState* stp) if (stp->line_item) { BeamInstr item = code[ci-1]; BeamInstr loc; - int li; + unsigned int li; if (item >= stp->num_line_items) { - LoadError2(stp, "line instruction index overflow (%d/%d)", + LoadError2(stp, "line instruction index overflow (%u/%u)", item, stp->num_line_items); } li = stp->current_li; if (li >= stp->num_line_instrs) { - LoadError2(stp, "line instruction table overflow (%d/%d)", + LoadError2(stp, "line instruction table overflow (%u/%u)", li, stp->num_line_instrs); } loc = stp->line_item[item]; @@ -2476,7 +2678,11 @@ load_code(LoaderState* stp) * End of code found. */ case op_int_code_end: - stp->code_buffer_size = code_buffer_size; + if (function_number != stp->num_functions) { + LoadError2(stp, "too few functions (%u) in module (header said %u)", + function_number, stp->num_functions); + } + stp->codev_size = codev_size; stp->ci = ci; stp->function = THE_NON_VALUE; stp->genop = NULL; @@ -2491,7 +2697,10 @@ load_code(LoaderState* stp) { GenOp* next = stp->genop->next; FREE_GENOP(stp, stp->genop); - stp->genop = next; + if ((stp->genop = next) == NULL) { + last_op_next = &stp->genop; + goto get_next_instr; + } goto do_transform; } } @@ -2635,12 +2844,18 @@ mixed_types(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) } static int -same_label(LoaderState* stp, GenOpArg Target, GenOpArg Label) +is_killed_apply(LoaderState* stp, GenOpArg Reg, GenOpArg Live) { - return Target.type = TAG_f && Label.type == TAG_u && - Target.val == Label.val; + return Reg.type == TAG_x && Live.type == TAG_u && + Live.val+2 <= Reg.val; } +static int +is_killed(LoaderState* stp, GenOpArg Reg, GenOpArg Live) +{ + return Reg.type == TAG_x && Live.type == TAG_u && + Live.val <= Reg.val; +} /* * Generate an instruction for element/2. @@ -2657,17 +2872,17 @@ gen_element(LoaderState* stp, GenOpArg Fail, GenOpArg Index, op->next = NULL; if (Index.type == TAG_i && Index.val > 0 && - (Tuple.type == TAG_r || Tuple.type == TAG_x || Tuple.type == TAG_y)) { + (Tuple.type == TAG_x || Tuple.type == TAG_y)) { op->op = genop_i_fast_element_4; - op->a[0] = Tuple; - op->a[1] = Fail; + op->a[0] = Fail; + op->a[1] = Tuple; op->a[2].type = TAG_u; op->a[2].val = Index.val; op->a[3] = Dst; } else { op->op = genop_i_element_4; - op->a[0] = Tuple; - op->a[1] = Fail; + op->a[0] = Fail; + op->a[1] = Tuple; op->a[2] = Index; op->a[3] = Dst; } @@ -2793,23 +3008,16 @@ gen_get_integer2(LoaderState* stp, GenOpArg Fail, GenOpArg Ms, GenOpArg Live, goto generic; } } else { - GenOp* op2; - NEW_GENOP(stp, op2); - - op->op = genop_i_fetch_2; - op->arity = 2; - op->a[0] = Ms; - op->a[1] = Size; - op->next = op2; - - op2->op = genop_i_bs_get_integer_4; - op2->arity = 4; - op2->a[0] = Fail; - op2->a[1] = Live; - op2->a[2].type = TAG_u; - op2->a[2].val = (Unit.val << 3) | Flags.val; - op2->a[3] = Dst; - op2->next = NULL; + op->op = genop_i_bs_get_integer_6; + op->arity = 6; + op->a[0] = Fail; + op->a[1] = Live; + op->a[2].type = TAG_u; + op->a[2].val = (Unit.val << 3) | Flags.val; + op->a[3] = Ms; + op->a[4] = Size; + op->a[5] = Dst; + op->next = NULL; return op; } op->next = NULL; @@ -3169,7 +3377,11 @@ gen_increment_from_minus(LoaderState* stp, GenOpArg Reg, GenOpArg Integer, static int negation_is_small(LoaderState* stp, GenOpArg Int) { - return Int.type == TAG_i && IS_SSMALL(-Int.val); + /* Check for the rare case of overflow in BeamInstr (UWord) -> Sint + * Cast to the correct type before using IS_SSMALL (Sint) */ + return Int.type == TAG_i && + !(Int.val & ~((((BeamInstr)1) << ((sizeof(Sint)*8)-1))-1)) && + IS_SSMALL(-((Sint)Int.val)); } @@ -3229,14 +3441,14 @@ gen_literal_timeout(LoaderState* stp, GenOpArg Fail, GenOpArg Time) op->a[1].type = TAG_u; if (Time.type == TAG_i && (timeout = Time.val) >= 0 && -#if defined(ARCH_64) && !HALFWORD_HEAP +#if defined(ARCH_64) (timeout >> 32) == 0 #else 1 #endif ) { op->a[1].val = timeout; -#if !defined(ARCH_64) || HALFWORD_HEAP +#if !defined(ARCH_64) } else if (Time.type == TAG_q) { Eterm big; @@ -3253,7 +3465,7 @@ gen_literal_timeout(LoaderState* stp, GenOpArg Fail, GenOpArg Time) } #endif } else { -#if !defined(ARCH_64) || HALFWORD_HEAP +#if !defined(ARCH_64) error: #endif op->op = genop_i_wait_error_0; @@ -3276,14 +3488,14 @@ gen_literal_timeout_locked(LoaderState* stp, GenOpArg Fail, GenOpArg Time) op->a[1].type = TAG_u; if (Time.type == TAG_i && (timeout = Time.val) >= 0 && -#if defined(ARCH_64) && !HALFWORD_HEAP +#if defined(ARCH_64) (timeout >> 32) == 0 #else 1 #endif ) { op->a[1].val = timeout; -#if !defined(ARCH_64) || HALFWORD_HEAP +#if !defined(ARCH_64) } else if (Time.type == TAG_q) { Eterm big; @@ -3300,7 +3512,7 @@ gen_literal_timeout_locked(LoaderState* stp, GenOpArg Fail, GenOpArg Time) } #endif } else { -#if !defined(ARCH_64) || HALFWORD_HEAP +#if !defined(ARCH_64) error: #endif op->op = genop_i_wait_error_locked_0; @@ -3319,9 +3531,10 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; /* * Verify the validity of the list. @@ -3336,9 +3549,37 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, } /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_tuple_arity2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = Rest[0].val; + op->a[3].type = TAG_u; + op->a[3].val = Rest[2].val; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + } + + /* * Generate the generic instruction. + * Assumption: + * Few different tuple arities to select on (fewer than 20). + * Use linear scan approach. */ + align = 1; + + arity += 2*align; + size += align; + NEW_GENOP(stp, op); op->next = NULL; op->op = genop_i_select_tuple_arity_3; @@ -3346,39 +3587,36 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; - op->a[2].val = Size.val / 2; - for (i = 0; i < Size.val; i += 2) { - op->a[i+3].type = TAG_v; - op->a[i+3].val = make_arityval(Rest[i].val); - op->a[i+4] = Rest[i+1]; - } + op->a[2].val = size; - /* - * Sort the values to make them useful for a binary search. - */ + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + for (i = 3; i < arity - 2*align; i+=2) { + tmp[i-3].type = TAG_v; + tmp[i-3].val = make_arityval(Rest[i-3].val); + tmp[i-2] = Rest[i-2]; } -#endif /* - * Use a special-cased instruction if there are only two values. + * Sort the values to make them useful for a sentinel search */ - if (size == 2) { - op->op = genop_i_select_tuple_arity2_6; - op->arity--; - op->a[2].type = TAG_u; - op->a[2].val = arityval(op->a[3].val); - op->a[3] = op->a[4]; - op->a[4].type = TAG_u; - op->a[4].val = arityval(op->a[5].val); - op->a[5] = op->a[6]; + + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j + size] = tmp[i-2]; + j++; } + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + return op; } @@ -3600,45 +3838,109 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; + + if (size == 2) { + + /* + * Use a special-cased instruction if there are only two values. + */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2] = Rest[0]; + op->a[3] = Rest[2]; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + + } else if (size > 10) { + + /* binary search instruction */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val_bins_3; + GENOP_ARITY(op, arity); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = size; + for (i = 3; i < arity; i++) { + op->a[i] = Rest[i-3]; + } + + /* + * Sort the values to make them useful for a binary search. + */ + + qsort(op->a+3, size, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); +#ifdef DEBUG + for (i = 3; i < arity-2; i += 2) { + ASSERT(op->a[i].val < op->a[i+2].val); + } +#endif + return op; + } + + /* linear search instruction */ + + align = 1; + + arity += 2*align; + size += align; NEW_GENOP(stp, op); op->next = NULL; - op->op = genop_i_select_val_3; + op->op = genop_i_select_val_lins_3; GENOP_ARITY(op, arity); op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; op->a[2].val = size; - for (i = 3; i < arity; i++) { - op->a[i] = Rest[i-3]; + + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); + + for (i = 3; i < arity - 2*align; i++) { + tmp[i-3] = Rest[i-3]; } /* - * Sort the values to make them useful for a binary search. + * Sort the values to make them useful for a sentinel search */ - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j+size] = tmp[i-2]; + j++; } -#endif - /* - * Use a special-cased instruction if there are only two values. - */ - if (size == 2) { - op->op = genop_i_select_val2_6; - op->arity--; - op->a[2] = op->a[3]; - op->a[3] = op->a[4]; - op->a[4] = op->a[5]; - op->a[5] = op->a[6]; + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + /* add sentinel */ + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + +#ifdef DEBUG + for (i = 0; i < size - 1; i++) { + ASSERT(op->a[i+3].val <= op->a[i+4].val); } +#endif return op; } @@ -3763,150 +4065,97 @@ gen_make_fun2(LoaderState* stp, GenOpArg idx) op->next = NULL; return op; } + +static GenOp* +translate_gc_bif(LoaderState* stp, GenOp* op, GenOpArg Bif) +{ + const ErtsGcBif* p; + BifFunction bf; + + bf = stp->import[Bif.val].bf; + for (p = erts_gc_bifs; p->bif != 0; p++) { + if (p->bif == bf) { + op->a[1].type = TAG_u; + op->a[1].val = (BeamInstr) p->gc_bif; + return op; + } + } + + op->op = genop_unsupported_guard_bif_3; + op->arity = 3; + op->a[0].type = TAG_a; + op->a[0].val = stp->import[Bif.val].module; + op->a[1].type = TAG_a; + op->a[1].val = stp->import[Bif.val].function; + op->a[2].type = TAG_u; + op->a[2].val = stp->import[Bif.val].arity; + return op; +} + /* - * Rewrite gc_bifs with one parameter (the common case). Utilized - * in ops.tab to rewrite instructions calling bif's in guards - * to use a garbage collecting implementation. The instructions - * are sometimes once again rewritten to handle literals (putting the - * parameter in the mostly unused r[0] before the instruction is executed). + * Rewrite gc_bifs with one parameter (the common case). */ static GenOp* gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, GenOpArg Src, GenOpArg Dst) { GenOp* op; - BifFunction bf; NEW_GENOP(stp, op); op->next = NULL; - bf = stp->import[Bif.val].bf; - /* The translations here need to have a reverse counterpart in - beam_emu.c:translate_gc_bif for error handling to work properly. */ - if (bf == length_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_length_1; - } else if (bf == size_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_size_1; - } else if (bf == bit_size_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_bit_size_1; - } else if (bf == byte_size_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_byte_size_1; - } else if (bf == map_size_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_map_size_1; - } else if (bf == abs_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_abs_1; - } else if (bf == float_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_float_1; - } else if (bf == round_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_round_1; - } else if (bf == trunc_1) { - op->a[1].val = (BeamInstr) (void *) erts_gc_trunc_1; - } else { - op->op = genop_unsupported_guard_bif_3; - op->arity = 3; - op->a[0].type = TAG_a; - op->a[0].val = stp->import[Bif.val].module; - op->a[1].type = TAG_a; - op->a[1].val = stp->import[Bif.val].function; - op->a[2].type = TAG_u; - op->a[2].val = stp->import[Bif.val].arity; - return op; - } op->op = genop_i_gc_bif1_5; op->arity = 5; op->a[0] = Fail; - op->a[1].type = TAG_u; + /* op->a[1] is set by translate_gc_bif() */ op->a[2] = Src; op->a[3] = Live; op->a[4] = Dst; - return op; + return translate_gc_bif(stp, op, Bif); } /* * This is used by the ops.tab rule that rewrites gc_bifs with two parameters. - * The instruction returned is then again rewritten to an i_load instruction - * followed by i_gc_bif2_jIId, to handle literals properly. - * As opposed to the i_gc_bif1_jIsId, the instruction i_gc_bif2_jIId is - * always rewritten, regardless of if there actually are any literals. */ static GenOp* gen_guard_bif2(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, GenOpArg S1, GenOpArg S2, GenOpArg Dst) { GenOp* op; - BifFunction bf; NEW_GENOP(stp, op); op->next = NULL; - bf = stp->import[Bif.val].bf; - /* The translations here need to have a reverse counterpart in - beam_emu.c:translate_gc_bif for error handling to work properly. */ - if (bf == binary_part_2) { - op->a[1].val = (BeamInstr) (void *) erts_gc_binary_part_2; - } else { - op->op = genop_unsupported_guard_bif_3; - op->arity = 3; - op->a[0].type = TAG_a; - op->a[0].val = stp->import[Bif.val].module; - op->a[1].type = TAG_a; - op->a[1].val = stp->import[Bif.val].function; - op->a[2].type = TAG_u; - op->a[2].val = stp->import[Bif.val].arity; - return op; - } - op->op = genop_ii_gc_bif2_6; + op->op = genop_i_gc_bif2_6; op->arity = 6; op->a[0] = Fail; - op->a[1].type = TAG_u; - op->a[2] = S1; - op->a[3] = S2; - op->a[4] = Live; + /* op->a[1] is set by translate_gc_bif() */ + op->a[2] = Live; + op->a[3] = S1; + op->a[4] = S2; op->a[5] = Dst; - return op; + return translate_gc_bif(stp, op, Bif); } /* * This is used by the ops.tab rule that rewrites gc_bifs with three parameters. - * The instruction returned is then again rewritten to a move instruction that - * uses r[0] for temp storage, followed by an i_load instruction, - * followed by i_gc_bif3_jIsId, to handle literals properly. Rewriting - * always occur, as with the gc_bif2 counterpart. */ static GenOp* gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, GenOpArg S1, GenOpArg S2, GenOpArg S3, GenOpArg Dst) { GenOp* op; - BifFunction bf; NEW_GENOP(stp, op); op->next = NULL; - bf = stp->import[Bif.val].bf; - /* The translations here need to have a reverse counterpart in - beam_emu.c:translate_gc_bif for error handling to work properly. */ - if (bf == binary_part_3) { - op->a[1].val = (BeamInstr) (void *) erts_gc_binary_part_3; - } else { - op->op = genop_unsupported_guard_bif_3; - op->arity = 3; - op->a[0].type = TAG_a; - op->a[0].val = stp->import[Bif.val].module; - op->a[1].type = TAG_a; - op->a[1].val = stp->import[Bif.val].function; - op->a[2].type = TAG_u; - op->a[2].val = stp->import[Bif.val].arity; - return op; - } op->op = genop_ii_gc_bif3_7; op->arity = 7; op->a[0] = Fail; - op->a[1].type = TAG_u; - op->a[2] = S1; - op->a[3] = S2; - op->a[4] = S3; - op->a[5] = Live; + /* op->a[1] is set by translate_gc_bif() */ + op->a[2] = Live; + op->a[3] = S1; + op->a[4] = S2; + op->a[5] = S3; op->a[6] = Dst; - op->next = NULL; - return op; + return translate_gc_bif(stp, op, Bif); } static GenOp* @@ -3959,8 +4208,139 @@ tuple_append_put(LoaderState* stp, GenOpArg Arity, GenOpArg Dst, } /* + * Predicate to test whether the given literal is a map. + */ + +static int +literal_is_map(LoaderState* stp, GenOpArg Lit) +{ + Eterm term; + + ASSERT(Lit.type == TAG_q); + term = stp->literals[Lit.val].term; + return is_map(term); +} + +/* + * Predicate to test whether the given literal is an empty map. + */ + +static int +is_empty_map(LoaderState* stp, GenOpArg Lit) +{ + Eterm term; + + if (Lit.type != TAG_q) { + return 0; + } + term = stp->literals[Lit.val].term; + return is_flatmap(term) && flatmap_get_size(flatmap_val(term)) == 0; +} + +/* + * Pseudo predicate map_key_sort that will sort the Rest operand for + * map instructions as a side effect. + */ + +typedef struct SortGenOpArg { + Eterm term; /* Term to use for comparing */ + GenOpArg arg; /* Original data */ +} SortGenOpArg; + +static int +genopargtermcompare(SortGenOpArg* a, SortGenOpArg* b) +{ + return CMP_TERM(a->term, b->term); +} + +static int +map_key_sort(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) +{ + SortGenOpArg* t; + unsigned size = Size.val; + unsigned i; + + if (size == 2) { + return 1; /* Already sorted. */ + } + + + t = (SortGenOpArg *) erts_alloc(ERTS_ALC_T_TMP, size*sizeof(SortGenOpArg)); + + /* + * Copy original data and sort keys to a temporary array. + */ + for (i = 0; i < size; i += 2) { + t[i].arg = Rest[i]; + switch (Rest[i].type) { + case TAG_a: + t[i].term = Rest[i].val; + ASSERT(is_atom(t[i].term)); + break; + case TAG_i: + t[i].term = make_small(Rest[i].val); + break; + case TAG_n: + t[i].term = NIL; + break; + case TAG_q: + t[i].term = stp->literals[Rest[i].val].term; + break; + default: + /* + * Not a literal key. Not allowed. Only a single + * variable key is allowed in each map instruction. + */ + erts_free(ERTS_ALC_T_TMP, (void *) t); + return 0; + } +#ifdef DEBUG + t[i+1].term = THE_NON_VALUE; +#endif + t[i+1].arg = Rest[i+1]; + } + + /* + * Sort the temporary array. + */ + qsort((void *) t, size / 2, 2 * sizeof(SortGenOpArg), + (int (*)(const void *, const void *)) genopargtermcompare); + + /* + * Copy back the sorted, original data. + */ + for (i = 0; i < size; i++) { + Rest[i] = t[i].arg; + } + + erts_free(ERTS_ALC_T_TMP, (void *) t); + return 1; +} + +static int +hash_genop_arg(LoaderState* stp, GenOpArg Key, Uint32* hx) +{ + switch (Key.type) { + case TAG_a: + *hx = hashmap_make_hash(Key.val); + return 1; + case TAG_i: + *hx = hashmap_make_hash(make_small(Key.val)); + return 1; + case TAG_n: + *hx = hashmap_make_hash(NIL); + return 1; + case TAG_q: + *hx = hashmap_make_hash(stp->literals[Key.val].term); + return 1; + default: + return 0; + } +} + +/* * Replace a get_map_elements with one key to an instruction with one - * element + * element. */ static GenOp* @@ -3968,37 +4348,150 @@ gen_get_map_element(LoaderState* stp, GenOpArg Fail, GenOpArg Src, GenOpArg Size, GenOpArg* Rest) { GenOp* op; + GenOpArg Key; + Uint32 hx = 0; ASSERT(Size.type == TAG_u); NEW_GENOP(stp, op); op->next = NULL; - op->op = genop_get_map_element_4; - op->arity = 4; - op->a[0] = Fail; op->a[1] = Src; op->a[2] = Rest[0]; - op->a[3] = Rest[1]; + + Key = Rest[0]; + if (hash_genop_arg(stp, Key, &hx)) { + op->arity = 5; + op->op = genop_i_get_map_element_hash_5; + op->a[3].type = TAG_u; + op->a[3].val = (BeamInstr) hx; + op->a[4] = Rest[1]; + } else { + op->arity = 4; + op->op = genop_i_get_map_element_4; + op->a[3] = Rest[1]; + } return op; } +static int +hash_internal_genop_arg(LoaderState* stp, GenOpArg Key, Uint32* hx) +{ + Eterm key_term; + switch (Key.type) { + case TAG_a: + key_term = Key.val; + break; + case TAG_i: + key_term = make_small(Key.val); + break; + case TAG_n: + key_term = NIL; + break; + case TAG_q: + key_term = stp->literals[Key.val].term; + break; + default: + return 0; + } + *hx = erts_pd_make_hx(key_term); + return 1; +} + + static GenOp* -gen_has_map_field(LoaderState* stp, GenOpArg Fail, GenOpArg Src, - GenOpArg Size, GenOpArg* Rest) +gen_get(LoaderState* stp, GenOpArg Src, GenOpArg Dst) +{ + GenOp* op; + Uint32 hx = 0; + + NEW_GENOP(stp, op); + op->next = NULL; + if (hash_internal_genop_arg(stp, Src, &hx)) { + op->arity = 3; + op->op = genop_i_get_hash_3; + op->a[0] = Src; + op->a[1].type = TAG_u; + op->a[1].val = (BeamInstr) hx; + op->a[2] = Dst; + } else { + op->arity = 2; + op->op = genop_i_get_2; + op->a[0] = Src; + op->a[1] = Dst; + } + return op; +} + + +static GenOp* +gen_get_map_elements(LoaderState* stp, GenOpArg Fail, GenOpArg Src, + GenOpArg Size, GenOpArg* Rest) { GenOp* op; + Uint32 hx; + Uint i; + GenOpArg* dst; +#ifdef DEBUG + int good_hash; +#endif + ERTS_UNDEF(hx, 0); ASSERT(Size.type == TAG_u); NEW_GENOP(stp, op); + op->op = genop_i_get_map_elements_3; + GENOP_ARITY(op, 3 + 3*(Size.val/2)); op->next = NULL; - op->op = genop_has_map_field_3; - op->arity = 4; + op->a[0] = Fail; + op->a[1] = Src; + op->a[2].type = TAG_u; + op->a[2].val = 3*(Size.val/2); + + dst = op->a+3; + for (i = 0; i < Size.val / 2; i++) { + dst[0] = Rest[2*i]; + dst[1] = Rest[2*i+1]; +#ifdef DEBUG + good_hash = +#endif + hash_genop_arg(stp, dst[0], &hx); +#ifdef DEBUG + ASSERT(good_hash); +#endif + dst[2].type = TAG_u; + dst[2].val = (BeamInstr) hx; + dst += 3; + } + return op; +} + +static GenOp* +gen_has_map_fields(LoaderState* stp, GenOpArg Fail, GenOpArg Src, + GenOpArg Size, GenOpArg* Rest) +{ + GenOp* op; + Uint i; + Uint n; + + ASSERT(Size.type == TAG_u); + n = Size.val; + + NEW_GENOP(stp, op); + GENOP_ARITY(op, 3 + 2*n); + op->next = NULL; + op->op = genop_get_map_elements_3; op->a[0] = Fail; op->a[1] = Src; - op->a[2] = Rest[0]; + op->a[2].type = TAG_u; + op->a[2].val = 2*n; + + for (i = 0; i < n; i++) { + op->a[3+2*i] = Rest[i]; + op->a[3+2*i+1].type = TAG_x; + op->a[3+2*i+1].val = SCRATCH_X_REG; /* Ignore result */ + } return op; } @@ -4010,8 +4503,8 @@ gen_has_map_field(LoaderState* stp, GenOpArg Fail, GenOpArg Src, static int freeze_code(LoaderState* stp) { - BeamInstr* code = stp->code; - Uint *literal_end = NULL; + BeamCodeHeader* code_hdr = stp->hdr; + BeamInstr* codev = (BeamInstr*) &stp->hdr->functions; int i; byte* str_table; unsigned strtab_size = stp->chunks[STR_CHUNK].size; @@ -4036,128 +4529,80 @@ freeze_code(LoaderState* stp) if (stp->line_instr == 0) { line_size = 0; } else { - line_size = (MI_LINE_FUNC_TAB + (stp->num_functions + 1) + - (stp->current_li+1) + stp->num_fnames) * - sizeof(Eterm) + (stp->current_li+1) * stp->loc_size; + line_size = (offsetof(BeamCodeLineTab,func_tab) + + (stp->num_functions + 1) * sizeof(BeamInstr**) /* func_tab */ + + (stp->current_li + 1) * sizeof(BeamInstr*) /* line items */ + + stp->num_fnames * sizeof(Eterm) /* fname table */ + + (stp->current_li + 1) * stp->loc_size); /* loc_tab */ } - size = (stp->ci * sizeof(BeamInstr)) + - (stp->total_literal_size * sizeof(Eterm)) + - strtab_size + attr_size + compile_size + line_size; + size = offsetof(BeamCodeHeader,functions) + (stp->ci * sizeof(BeamInstr)) + + strtab_size + attr_size + compile_size + MD5_SIZE + line_size; /* * Move the code to its final location. */ - code = (BeamInstr *) erts_realloc(ERTS_ALC_T_CODE, (void *) code, size); - CHKBLK(ERTS_ALC_T_CODE,code); + code_hdr = (BeamCodeHeader*) erts_realloc(ERTS_ALC_T_CODE, (void *) code_hdr, size); + codev = (BeamInstr*) &code_hdr->functions; + CHKBLK(ERTS_ALC_T_CODE,code_hdr); /* * Place a pointer to the op_int_code_end instruction in the * function table in the beginning of the file. */ - code[MI_FUNCTIONS+stp->num_functions] = (BeamInstr) (code + stp->ci - 1); - CHKBLK(ERTS_ALC_T_CODE,code); + code_hdr->functions[stp->num_functions] = (ErtsCodeInfo*)(codev + stp->ci - 1); + CHKBLK(ERTS_ALC_T_CODE,code_hdr); /* * Store the pointer to the on_load function. */ if (stp->on_load) { - code[MI_ON_LOAD_FUNCTION_PTR] = (BeamInstr) (code + stp->on_load); + code_hdr->on_load_function_ptr = codev + stp->on_load; } else { - code[MI_ON_LOAD_FUNCTION_PTR] = 0; + code_hdr->on_load_function_ptr = NULL; } - CHKBLK(ERTS_ALC_T_CODE,code); + CHKBLK(ERTS_ALC_T_CODE,code_hdr); - literal_end = (Uint *) (code+stp->ci); /* - * Place the literal heap directly after the code and fix up all - * instructions that refer to it. + * Place the literals in their own allocated heap (for fast range check) + * and fix up all instructions that refer to it. */ { - Uint* ptr; - Uint* low; - Uint* high; + Eterm* ptr; LiteralPatch* lp; - struct erl_off_heap_header* off_heap = 0; - struct erl_off_heap_header** off_heap_last = &off_heap; - - low = (Uint *) (code+stp->ci); - high = low + stp->total_literal_size; - code[MI_LITERALS_START] = (BeamInstr) low; - code[MI_LITERALS_END] = (BeamInstr) high; - ptr = low; - for (i = 0; i < stp->num_literals; i++) { - SWord offset; - struct erl_off_heap_header* t_off_heap; - - sys_memcpy(ptr, stp->literals[i].heap, - stp->literals[i].heap_size*sizeof(Eterm)); - offset = ptr - stp->literals[i].heap; - stp->literals[i].offset = offset; - high = ptr + stp->literals[i].heap_size; - while (ptr < high) { - Eterm val = *ptr; - switch (primary_tag(val)) { - case TAG_PRIMARY_LIST: - case TAG_PRIMARY_BOXED: - *ptr++ = offset_ptr(val, offset); - break; - case TAG_PRIMARY_HEADER: - if (header_is_transparent(val)) { - ptr++; - } else { - if (thing_subtag(val) == REFC_BINARY_SUBTAG) { - struct erl_off_heap_header* oh; - - oh = (struct erl_off_heap_header*) ptr; - if (oh->next) { - Eterm** uptr = (Eterm **) (void *) &oh->next; - *uptr += offset; - } - } - ptr += 1 + thing_arityval(val); - } - break; - default: - ptr++; - break; - } - } - ASSERT(ptr == high); + ErlOffHeap code_off_heap; + ErtsLiteralArea *literal_area; + Uint lit_asize; - /* - * Re-link the off_heap list for this term onto the - * off_heap list for the entire module. - */ - t_off_heap = stp->literals[i].off_heap.first; - if (t_off_heap) { - t_off_heap = (struct erl_off_heap_header *) - offset_ptr((UWord) t_off_heap, offset); - while (t_off_heap) { - *off_heap_last = t_off_heap; - off_heap_last = &t_off_heap->next; - t_off_heap = t_off_heap->next; - } - } + ERTS_INIT_OFF_HEAP(&code_off_heap); + + lit_asize = ERTS_LITERAL_AREA_ALLOC_SIZE(stp->total_literal_size); + literal_area = erts_alloc(ERTS_ALC_T_LITERAL, lit_asize); + ptr = &literal_area->start[0]; + literal_area->end = ptr + stp->total_literal_size; + + for (i = 0; i < stp->num_literals; i++) { + if (is_not_immed(stp->literals[i].term)) { + erts_move_multi_frags(&ptr, &code_off_heap, + stp->literals[i].heap_frags, + &stp->literals[i].term, 1, 1); + ASSERT(erts_is_literal(stp->literals[i].term, + ptr_val(stp->literals[i].term))); + } } - code[MI_LITERALS_OFF_HEAP] = (BeamInstr) off_heap; + literal_area->off_heap = code_off_heap.first; lp = stp->literal_patches; while (lp != 0) { BeamInstr* op_ptr; - Uint literal; Literal* lit; - op_ptr = code + lp->pos; + op_ptr = codev + lp->pos; lit = &stp->literals[op_ptr[0]]; - literal = lit->term; - if (is_boxed(literal) || is_list(literal)) { - literal = offset_ptr(literal, lit->offset); - } - op_ptr[0] = literal; + op_ptr[0] = lit->term; lp = lp->next; } - literal_end += stp->total_literal_size; + code_hdr->literal_area = literal_area; } CHKBLK(ERTS_ALC_T_CODE,code); @@ -4165,52 +4610,49 @@ freeze_code(LoaderState* stp) * If there is line information, place it here. */ if (stp->line_instr == 0) { - code[MI_LINE_TABLE] = (BeamInstr) 0; - str_table = (byte *) literal_end; + code_hdr->line_table = NULL; + str_table = (byte *) (codev + stp->ci); } else { - Eterm* line_tab = (Eterm *) literal_end; - Eterm* p; - int ftab_size = stp->num_functions; - int num_instrs = stp->current_li; - Eterm* first_line_item; + BeamCodeLineTab* const line_tab = (BeamCodeLineTab *) (codev+stp->ci); + const unsigned int ftab_size = stp->num_functions; + const unsigned int num_instrs = stp->current_li; + const BeamInstr** const line_items = + (const BeamInstr**) &line_tab->func_tab[ftab_size + 1]; - code[MI_LINE_TABLE] = (BeamInstr) line_tab; - p = line_tab + MI_LINE_FUNC_TAB; + code_hdr->line_table = line_tab; - first_line_item = (p + ftab_size + 1); for (i = 0; i < ftab_size; i++) { - *p++ = (Eterm) (BeamInstr) (first_line_item + stp->func_line[i]); + line_tab->func_tab[i] = line_items + stp->func_line[i]; } - *p++ = (Eterm) (BeamInstr) (first_line_item + num_instrs); - ASSERT(p == first_line_item); + line_tab->func_tab[i] = line_items + num_instrs; + for (i = 0; i < num_instrs; i++) { - *p++ = (Eterm) (BeamInstr) (code + stp->line_instr[i].pos); + line_items[i] = codev + stp->line_instr[i].pos; } - *p++ = (Eterm) (BeamInstr) (code + stp->ci - 1); + line_items[i] = codev + stp->ci - 1; - line_tab[MI_LINE_FNAME_PTR] = (Eterm) (BeamInstr) p; - memcpy(p, stp->fname, stp->num_fnames*sizeof(Eterm)); - p += stp->num_fnames; + line_tab->fname_ptr = (Eterm*) &line_items[i + 1]; + memcpy(line_tab->fname_ptr, stp->fname, stp->num_fnames*sizeof(Eterm)); - line_tab[MI_LINE_LOC_TAB] = (Eterm) (BeamInstr) p; - line_tab[MI_LINE_LOC_SIZE] = stp->loc_size; + line_tab->loc_size = stp->loc_size; if (stp->loc_size == 2) { - Uint16* locp = (Uint16 *) p; - for (i = 0; i < num_instrs; i++) { + Uint16* locp = (Uint16 *) &line_tab->fname_ptr[stp->num_fnames]; + line_tab->loc_tab.p2 = locp; + for (i = 0; i < num_instrs; i++) { *locp++ = (Uint16) stp->line_instr[i].loc; - } - *locp++ = LINE_INVALID_LOCATION; - str_table = (byte *) locp; + } + *locp++ = LINE_INVALID_LOCATION; + str_table = (byte *) locp; } else { - Uint32* locp = (Uint32 *) p; - ASSERT(stp->loc_size == 4); + Uint32* locp = (Uint32 *) &line_tab->fname_ptr[stp->num_fnames]; + ASSERT(stp->loc_size == 4); + line_tab->loc_tab.p4 = locp; for (i = 0; i < num_instrs; i++) { *locp++ = stp->line_instr[i].loc; } *locp++ = LINE_INVALID_LOCATION; - str_table = (byte *) locp; + str_table = (byte *) locp; } - CHKBLK(ERTS_ALC_T_CODE,code); } @@ -4222,13 +4664,13 @@ freeze_code(LoaderState* stp) if (attr_size) { byte* attr = str_table + strtab_size; sys_memcpy(attr, stp->chunks[ATTR_CHUNK].start, stp->chunks[ATTR_CHUNK].size); - code[MI_ATTR_PTR] = (BeamInstr) attr; - code[MI_ATTR_SIZE] = (BeamInstr) stp->chunks[ATTR_CHUNK].size; + code_hdr->attr_ptr = attr; + code_hdr->attr_size = (BeamInstr) stp->chunks[ATTR_CHUNK].size; decoded_size = erts_decode_ext_size(attr, attr_size); if (decoded_size < 0) { LoadError0(stp, "bad external term representation of module attributes"); } - code[MI_ATTR_SIZE_ON_HEAP] = decoded_size; + code_hdr->attr_size_on_heap = decoded_size; } CHKBLK(ERTS_ALC_T_CODE,code); if (compile_size) { @@ -4238,9 +4680,9 @@ freeze_code(LoaderState* stp) stp->chunks[COMPILE_CHUNK].size); CHKBLK(ERTS_ALC_T_CODE,code); - code[MI_COMPILE_PTR] = (BeamInstr) compile_info; + code_hdr->compile_ptr = compile_info; CHKBLK(ERTS_ALC_T_CODE,code); - code[MI_COMPILE_SIZE] = (BeamInstr) stp->chunks[COMPILE_CHUNK].size; + code_hdr->compile_size = (BeamInstr) stp->chunks[COMPILE_CHUNK].size; CHKBLK(ERTS_ALC_T_CODE,code); decoded_size = erts_decode_ext_size(compile_info, compile_size); CHKBLK(ERTS_ALC_T_CODE,code); @@ -4248,15 +4690,24 @@ freeze_code(LoaderState* stp) LoadError0(stp, "bad external term representation of compilation information"); } CHKBLK(ERTS_ALC_T_CODE,code); - code[MI_COMPILE_SIZE_ON_HEAP] = decoded_size; + code_hdr->compile_size_on_heap = decoded_size; + } + CHKBLK(ERTS_ALC_T_CODE,code); + { + byte* md5_sum = str_table + strtab_size + attr_size + compile_size; + CHKBLK(ERTS_ALC_T_CODE,code); + sys_memcpy(md5_sum, stp->mod_md5, MD5_SIZE); + CHKBLK(ERTS_ALC_T_CODE,code); + code_hdr->md5_ptr = md5_sum; + CHKBLK(ERTS_ALC_T_CODE,code); } CHKBLK(ERTS_ALC_T_CODE,code); /* * Make sure that we have not overflowed the allocated code space. */ - ASSERT(str_table + strtab_size + attr_size + compile_size == - ((byte *) code) + size); + ASSERT(str_table + strtab_size + attr_size + compile_size + MD5_SIZE == + ((byte *) code_hdr) + size); /* * Patch all instructions that refer to the string table. @@ -4268,46 +4719,47 @@ freeze_code(LoaderState* stp) BeamInstr* op_ptr; byte* strp; - op_ptr = code + sp->pos; + op_ptr = codev + sp->pos; strp = str_table + op_ptr[0]; op_ptr[0] = (BeamInstr) strp; sp = sp->next; } } - CHKBLK(ERTS_ALC_T_CODE,code); + CHKBLK(ERTS_ALC_T_CODE,code_hdr); /* * Resolve all labels. */ for (i = 0; i < stp->num_labels; i++) { - Uint this_patch; - Uint next_patch; + Sint this_patch; + Sint next_patch; Uint value = stp->labels[i].value; - if (value == 0 && stp->labels[i].patches != 0) { + if (value == 0 && stp->labels[i].patches >= 0) { LoadError1(stp, "label %d not resolved", i); } ASSERT(value < stp->ci); this_patch = stp->labels[i].patches; - while (this_patch != 0) { + while (this_patch >= 0) { ASSERT(this_patch < stp->ci); - next_patch = code[this_patch]; + next_patch = codev[this_patch]; ASSERT(next_patch < stp->ci); - code[this_patch] = (BeamInstr) (code + value); + codev[this_patch] = (BeamInstr) (codev + value); this_patch = next_patch; } } - CHKBLK(ERTS_ALC_T_CODE,code); + CHKBLK(ERTS_ALC_T_CODE,code_hdr); /* * Save the updated code pointer and code size. */ - stp->code = code; + stp->hdr = code_hdr; + stp->codev = codev; stp->loaded_size = size; - CHKBLK(ERTS_ALC_T_CODE,code); + CHKBLK(ERTS_ALC_T_CODE,code_hdr); return 1; load_error: @@ -4315,19 +4767,19 @@ freeze_code(LoaderState* stp) * Make sure that the caller frees the newly reallocated block, and * not the old one (in case it has moved). */ - stp->code = code; + stp->hdr = code_hdr; + stp->codev = codev; return 0; } static void -final_touch(LoaderState* stp) +final_touch(LoaderState* stp, struct erl_module_instance* inst_p) { - int i; + unsigned int i; int on_load = stp->on_load; unsigned catches; Uint index; - BeamInstr* code = stp->code; - Module* modp; + BeamInstr* codev = stp->codev; /* * Allocate catch indices and fix up all catch_yf instructions. @@ -4336,14 +4788,13 @@ final_touch(LoaderState* stp) index = stp->catches; catches = BEAM_CATCHES_NIL; while (index != 0) { - BeamInstr next = code[index]; - code[index] = BeamOpCode(op_catch_yf); - catches = beam_catches_cons((BeamInstr *)code[index+2], catches); - code[index+2] = make_catch(catches); + BeamInstr next = codev[index]; + codev[index] = BeamOpCode(op_catch_yf); + catches = beam_catches_cons((BeamInstr *)codev[index+2], catches); + codev[index+2] = make_catch(catches); index = next; } - modp = erts_put_module(stp->module); - modp->curr.catches = catches; + inst_p->catches = catches; /* * Export functions. @@ -4359,16 +4810,16 @@ final_touch(LoaderState* stp) } ep = erts_export_put(stp->module, stp->export[i].function, stp->export[i].arity); - if (!on_load) { - ep->addressv[erts_staging_code_ix()] = address; - } else { + if (on_load) { /* - * Don't make any of the exported functions - * callable yet. + * on_load: Don't make any of the exported functions + * callable yet. Keep any function in the current + * code callable. */ - ep->addressv[erts_staging_code_ix()] = ep->code+3; - ep->code[4] = (BeamInstr) address; + ep->beam[1] = (BeamInstr) address; } + else + ep->addressv[erts_staging_code_ix()] = address; } /* @@ -4390,8 +4841,8 @@ final_touch(LoaderState* stp) current = stp->import[i].patches; while (current != 0) { ASSERT(current < stp->ci); - next = stp->code[current]; - stp->code[current] = import; + next = stp->codev[current]; + stp->codev[current] = import; current = next; } } @@ -4404,17 +4855,17 @@ final_touch(LoaderState* stp) for (i = 0; i < stp->num_lambdas; i++) { unsigned entry_label = stp->lambdas[i].label; ErlFunEntry* fe = stp->lambdas[i].fe; - BeamInstr* code_ptr = (BeamInstr *) (stp->code + stp->labels[entry_label].value); + BeamInstr* code_ptr = stp->codev + stp->labels[entry_label].value; if (fe->address[0] != 0) { /* * We are hiding a pointer into older code. */ - erts_refc_dec(&fe->refc, 1); + erts_smp_refc_dec(&fe->refc, 1); } fe->address = code_ptr; #ifdef HIPE - hipe_set_closure_stub(fe, stp->lambdas[i].num_free); + hipe_set_closure_stub(fe); #endif } } @@ -4425,32 +4876,26 @@ transform_engine(LoaderState* st) { Uint op; int ap; /* Current argument. */ - Uint* restart; /* Where to restart if current match fails. */ - GenOpArg def_vars[TE_MAX_VARS]; /* Default buffer for variables. */ - GenOpArg* var = def_vars; - int num_vars = 0; + const Uint* restart; /* Where to restart if current match fails. */ + GenOpArg var[TE_MAX_VARS]; /* Buffer for variables. */ + GenOpArg* rest_args = NULL; + int num_rest_args = 0; int i; /* General index. */ Uint mask; GenOp* instr; - Uint* pc; - int rval; + GenOp* first = st->genop; + GenOp* keep = NULL; + const Uint* pc; static Uint restart_fail[1] = {TOP_fail}; - ASSERT(gen_opc[st->genop->op].transform != -1); - pc = op_transform + gen_opc[st->genop->op].transform; - restart = pc; + ASSERT(gen_opc[first->op].transform != -1); + restart = op_transform + gen_opc[first->op].transform; restart: - if (var != def_vars) { - erts_free(ERTS_ALC_T_LOADER_TMP, (void *) var); - var = def_vars; - } ASSERT(restart != NULL); pc = restart; ASSERT(*pc < NUM_TOPS); /* Valid instruction? */ - instr = st->genop; - -#define RETURN(r) rval = (r); goto do_return; + instr = first; #ifdef DEBUG restart = NULL; @@ -4468,7 +4913,7 @@ transform_engine(LoaderState* st) * We'll need at least one more instruction to decide whether * this combination matches or not. */ - RETURN(TE_SHORT_WINDOW); + return TE_SHORT_WINDOW; } if (*pc++ != instr->op) goto restart; @@ -4517,7 +4962,8 @@ transform_engine(LoaderState* st) if (var[i].type != instr->a[ap].type) goto restart; switch (var[i].type) { - case TAG_r: case TAG_n: break; + case TAG_n: + break; default: if (var[i].val != instr->a[ap].val) goto restart; @@ -4550,7 +4996,7 @@ transform_engine(LoaderState* st) if (i >= st->num_imports || st->import[i].bf == NULL) goto restart; if (bif_number != -1 && - bif_export[bif_number]->code[4] != (BeamInstr) st->import[i].bf) { + bif_export[bif_number]->beam[1] != (BeamInstr) st->import[i].bf) { goto restart; } } @@ -4629,19 +5075,9 @@ transform_engine(LoaderState* st) #if defined(TOP_rest_args) case TOP_rest_args: { - int n = *pc++; int formal_arity = gen_opc[instr->op].arity; - int j = formal_arity; - - num_vars = n + (instr->arity - formal_arity); - var = erts_alloc(ERTS_ALC_T_LOADER_TMP, - num_vars * sizeof(GenOpArg)); - for (i = 0; i < n; i++) { - var[i] = def_vars[i]; - } - while (i < num_vars) { - var[i++] = instr->a[j++]; - } + num_rest_args = instr->arity - formal_arity; + rest_args = instr->a + formal_arity; } break; #endif @@ -4650,21 +5086,22 @@ transform_engine(LoaderState* st) break; case TOP_commit: instr = instr->next; /* The next_instr was optimized away. */ - - /* - * The left-hand side of this transformation matched. - * Delete all matched instructions. - */ - while (st->genop != instr) { - GenOp* next = st->genop->next; - FREE_GENOP(st, st->genop); - st->genop = next; - } + keep = instr; + st->genop = instr; #ifdef DEBUG instr = 0; #endif break; - +#if defined(TOP_keep) + case TOP_keep: + /* Keep the current instruction unchanged. */ + keep = instr; + st->genop = instr; +#ifdef DEBUG + instr = 0; +#endif + break; +#endif #if defined(TOP_call_end) case TOP_call_end: { @@ -4689,22 +5126,19 @@ transform_engine(LoaderState* st) lastp = &((*lastp)->next); } - instr = instr->next; /* The next_instr was optimized away. */ - - /* - * The left-hand side of this transformation matched. - * Delete all matched instructions. - */ - while (st->genop != instr) { - GenOp* next = st->genop->next; - FREE_GENOP(st, st->genop); - st->genop = next; - } - *lastp = st->genop; + keep = instr->next; /* The next_instr was optimized away. */ + *lastp = keep; st->genop = new_instr; } - RETURN(TE_OK); + /* FALLTHROUGH */ #endif + case TOP_end: + while (first != keep) { + GenOp* next = first->next; + FREE_GENOP(st, first); + first = next; + } + return TE_OK; case TOP_new_instr: /* * Note that the instructions are generated in reverse order. @@ -4716,6 +5150,12 @@ transform_engine(LoaderState* st) instr->arity = gen_opc[op].arity; ap = 0; break; +#ifdef TOP_rename + case TOP_rename: + instr->op = op = *pc++; + instr->arity = gen_opc[op].arity; + return TE_OK; +#endif case TOP_store_type: i = *pc++; instr->a[ap].type = i; @@ -4735,14 +5175,10 @@ transform_engine(LoaderState* st) #if defined(TOP_store_rest_args) case TOP_store_rest_args: { - int n = *pc++; - int num_extra = num_vars - n; - - ASSERT(n <= num_vars); - GENOP_ARITY(instr, instr->arity+num_extra); + GENOP_ARITY(instr, instr->arity+num_rest_args); memcpy(instr->a, instr->def_args, ap*sizeof(GenOpArg)); - memcpy(instr->a+ap, var+n, num_extra*sizeof(GenOpArg)); - ap += num_extra; + memcpy(instr->a+ap, rest_args, num_rest_args*sizeof(GenOpArg)); + ap += num_rest_args; } break; #endif @@ -4754,21 +5190,12 @@ transform_engine(LoaderState* st) case TOP_try_me_else_fail: restart = restart_fail; break; - case TOP_end: - RETURN(TE_OK); case TOP_fail: - RETURN(TE_FAIL); + return TE_FAIL; default: ASSERT(0); } } -#undef RETURN - - do_return: - if (var != def_vars) { - erts_free(ERTS_ALC_T_LOADER_TMP, (void *) var); - } - return rval; } static void @@ -5020,14 +5447,14 @@ new_genop(LoaderState* stp) static int new_label(LoaderState* stp) { - int num = stp->num_labels; + unsigned int num = stp->num_labels; stp->num_labels++; stp->labels = (Label *) erts_realloc(ERTS_ALC_T_PREPARED_CODE, (void *) stp->labels, stp->num_labels * sizeof(Label)); stp->labels[num].value = 0; - stp->labels[num].patches = 0; + stp->labels[num].patches = -1; return num; } @@ -5076,19 +5503,18 @@ new_literal(LoaderState* stp, Eterm** hpp, Uint heap_size) stp->total_literal_size += heap_size; lit = stp->literals + stp->num_literals; - lit->offset = 0; - lit->heap_size = heap_size; - lit->heap = erts_alloc(ERTS_ALC_T_PREPARED_CODE, heap_size*sizeof(Eterm)); - lit->term = make_boxed(lit->heap); - lit->off_heap.first = 0; - lit->off_heap.overhead = 0; - *hpp = lit->heap; + lit->heap_frags = new_literal_fragment(heap_size); + lit->term = make_boxed(lit->heap_frags->mem); + *hpp = lit->heap_frags->mem; return stp->num_literals++; } Eterm erts_module_info_0(Process* p, Eterm module) { + Module* modp; + ErtsCodeIndex code_ix = erts_active_code_ix(); + BeamCodeHeader* code_hdr; Eterm *hp; Eterm list = NIL; Eterm tup; @@ -5097,21 +5523,31 @@ erts_module_info_0(Process* p, Eterm module) return THE_NON_VALUE; } - if (erts_get_module(module, erts_active_code_ix()) == NULL) { + modp = erts_get_module(module, code_ix); + if (modp == NULL) { return THE_NON_VALUE; } + code_hdr = modp->curr.code_hdr; + if (code_hdr == NULL) { + return THE_NON_VALUE; + } + #define BUILD_INFO(What) \ - tup = erts_module_info_1(p, module, What); \ + tup = get_module_info(p, code_ix, code_hdr, module, What); \ hp = HAlloc(p, 5); \ tup = TUPLE2(hp, What, tup); \ hp += 3; \ list = CONS(hp, tup, list) + BUILD_INFO(am_md5); +#ifdef HIPE + BUILD_INFO(am_native); +#endif BUILD_INFO(am_compile); BUILD_INFO(am_attributes); - BUILD_INFO(am_imports); BUILD_INFO(am_exports); + BUILD_INFO(am_module); #undef BUILD_INFO return list; } @@ -5119,20 +5555,47 @@ erts_module_info_0(Process* p, Eterm module) Eterm erts_module_info_1(Process* p, Eterm module, Eterm what) { + Module* modp; + ErtsCodeIndex code_ix = erts_active_code_ix(); + BeamCodeHeader* code_hdr; + + if (is_not_atom(module)) { + return THE_NON_VALUE; + } + + modp = erts_get_module(module, code_ix); + if (modp == NULL) { + return THE_NON_VALUE; + } + + code_hdr = modp->curr.code_hdr; + if (code_hdr == NULL) { + return THE_NON_VALUE; + } + + return get_module_info(p, code_ix, code_hdr, module, what); +} + +static Eterm +get_module_info(Process* p, ErtsCodeIndex code_ix, BeamCodeHeader* code_hdr, + Eterm module, Eterm what) +{ if (what == am_module) { return module; - } else if (what == am_imports) { - return NIL; + } else if (what == am_md5) { + return md5_of_module(p, code_hdr); } else if (what == am_exports) { - return exported_from_module(p, module); + return exported_from_module(p, code_ix, module); } else if (what == am_functions) { - return functions_in_module(p, module); + return functions_in_module(p, code_hdr); } else if (what == am_attributes) { - return attributes_for_module(p, module); + return attributes_for_module(p, code_hdr); } else if (what == am_compile) { - return compilation_info_for_module(p, module); + return compilation_info_for_module(p, code_hdr); } else if (what == am_native_addresses) { - return native_addresses(p, module); + return native_addresses(p, code_hdr); + } else if (what == am_native) { + return has_native(code_hdr); } return THE_NON_VALUE; } @@ -5140,16 +5603,12 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) /* * Builds a list of all functions in the given module: * [{Name, Arity},...] - * - * Returns a tagged term, or 0 on error. */ Eterm functions_in_module(Process* p, /* Process whose heap to use. */ - Eterm mod) /* Tagged atom for module. */ + BeamCodeHeader* code_hdr) { - Module* modp; - BeamInstr* code; int i; Uint num_functions; Uint need; @@ -5157,32 +5616,21 @@ functions_in_module(Process* p, /* Process whose heap to use. */ Eterm* hp_end; Eterm result = NIL; - if (is_not_atom(mod)) { - return THE_NON_VALUE; - } - - modp = erts_get_module(mod, erts_active_code_ix()); - if (modp == NULL) { - return THE_NON_VALUE; - } - code = modp->curr.code; - num_functions = code[MI_NUM_FUNCTIONS]; + num_functions = code_hdr->num_functions; need = 5*num_functions; hp = HAlloc(p, need); hp_end = hp + need; for (i = num_functions-1; i >= 0 ; i--) { - BeamInstr* func_info = (BeamInstr *) code[MI_FUNCTIONS+i]; - Eterm name = (Eterm) func_info[3]; - int arity = (int) func_info[4]; + ErtsCodeInfo* ci = code_hdr->functions[i]; Eterm tuple; /* * If the function name is [], this entry is a stub for * a BIF that should be ignored. */ - ASSERT(is_atom(name) || is_nil(name)); - if (is_atom(name)) { - tuple = TUPLE2(hp, name, make_small(arity)); + ASSERT(is_atom(ci->mfa.function) || is_nil(ci->mfa.function)); + if (is_atom(ci->mfa.function)) { + tuple = TUPLE2(hp, ci->mfa.function, make_small(ci->mfa.arity)); hp += 3; result = CONS(hp, tuple, result); hp += 2; @@ -5193,87 +5641,136 @@ functions_in_module(Process* p, /* Process whose heap to use. */ } /* + * Returns 'true' if mod has any native compiled functions, otherwise 'false' + */ + +static Eterm +has_native(BeamCodeHeader *code_hdr) +{ + Eterm result = am_false; +#ifdef HIPE + if (erts_is_module_native(code_hdr)) { + result = am_true; + } +#endif + return result; +} + +void +erts_release_literal_area(ErtsLiteralArea* literal_area) +{ + struct erl_off_heap_header* oh; + + if (!literal_area) + return; + + oh = literal_area->off_heap; + + while (oh) { + Binary* bptr; + ASSERT(thing_subtag(oh->thing_word) == REFC_BINARY_SUBTAG); + bptr = ((ProcBin*)oh)->val; + erts_bin_release(bptr); + oh = oh->next; + } + erts_free(ERTS_ALC_T_LITERAL, literal_area); +} + +int +erts_is_module_native(BeamCodeHeader* code_hdr) +{ + Uint i, num_functions; + + /* Check NativeAdress of first real function in module */ + if (code_hdr != NULL) { + num_functions = code_hdr->num_functions; + for (i=0; i<num_functions; i++) { + ErtsCodeInfo* ci = code_hdr->functions[i]; + if (is_atom(ci->mfa.function)) { + return erts_is_function_native(ci); + } + else ASSERT(is_nil(ci->mfa.function)); /* ignore BIF stubs */ + } + } + return 0; +} + +int +erts_is_function_native(ErtsCodeInfo *ci) +{ +#ifdef HIPE + ASSERT(ci->op == (BeamInstr) BeamOp(op_i_func_info_IaaI)); + return erts_codeinfo_to_code(ci)[0] == (BeamInstr) BeamOp(op_hipe_trap_call) + || erts_codeinfo_to_code(ci)[0] == (BeamInstr) BeamOp(op_hipe_trap_call_closure); +#else + return 0; +#endif +} + +/* * Builds a list of all functions including native addresses. * [{Name,Arity,NativeAddress},...] - * - * Returns a tagged term, or 0 on error. */ static Eterm -native_addresses(Process* p, Eterm mod) +native_addresses(Process* p, BeamCodeHeader* code_hdr) { - Module* modp; - BeamInstr* code; + Eterm result = NIL; +#ifdef HIPE int i; Eterm* hp; Uint num_functions; Uint need; Eterm* hp_end; - Eterm result = NIL; - - if (is_not_atom(mod)) { - return THE_NON_VALUE; - } - - modp = erts_get_module(mod, erts_active_code_ix()); - if (modp == NULL) { - return THE_NON_VALUE; - } - code = modp->curr.code; - num_functions = code[MI_NUM_FUNCTIONS]; + num_functions = code_hdr->num_functions; need = (6+BIG_UINT_HEAP_SIZE)*num_functions; hp = HAlloc(p, need); hp_end = hp + need; for (i = num_functions-1; i >= 0 ; i--) { - BeamInstr* func_info = (BeamInstr *) code[MI_FUNCTIONS+i]; - Eterm name = (Eterm) func_info[3]; - int arity = (int) func_info[4]; + ErtsCodeInfo *ci = code_hdr->functions[i]; Eterm tuple; - ASSERT(is_atom(name) || is_nil(name)); /* [] if BIF stub */ - if (func_info[1] != 0) { - Eterm addr; - ASSERT(is_atom(name)); - addr = erts_bld_uint(&hp, NULL, func_info[1]); - tuple = erts_bld_tuple(&hp, NULL, 3, name, make_small(arity), addr); + ASSERT(is_atom(ci->mfa.function) + || is_nil(ci->mfa.function)); /* [] if BIF stub */ + if (ci->u.ncallee != NULL) { + Eterm addr; + ASSERT(is_atom(ci->mfa.function)); + addr = erts_bld_uint(&hp, NULL, (Uint)ci->u.ncallee); + tuple = erts_bld_tuple(&hp, NULL, 3, ci->mfa.function, + make_small(ci->mfa.arity), addr); result = erts_bld_cons(&hp, NULL, tuple, result); } } HRelease(p, hp_end, hp); +#endif return result; } /* * Builds a list of all exported functions in the given module: * [{Name, Arity},...] - * - * Returns a tagged term, or 0 on error. */ Eterm exported_from_module(Process* p, /* Process whose heap to use. */ + ErtsCodeIndex code_ix, Eterm mod) /* Tagged atom for module. */ { - int i; + int i, num_exps; Eterm* hp = NULL; Eterm* hend = NULL; Eterm result = NIL; - ErtsCodeIndex code_ix; - - if (is_not_atom(mod)) { - return THE_NON_VALUE; - } - code_ix = erts_active_code_ix(); - for (i = 0; i < export_list_size(code_ix); i++) { + num_exps = export_list_size(code_ix); + for (i = 0; i < num_exps; i++) { Export* ep = export_list(i,code_ix); - if (ep->code[0] == mod) { + if (ep->info.mfa.module == mod) { Eterm tuple; - if (ep->addressv[code_ix] == ep->code+3 && - ep->code[3] == (BeamInstr) em_call_error_handler) { + if (ep->addressv[code_ix] == ep->beam && + ep->beam[0] == (BeamInstr) em_call_error_handler) { /* There is a call to the function, but it does not exist. */ continue; } @@ -5283,7 +5780,8 @@ exported_from_module(Process* p, /* Process whose heap to use. */ hp = HAlloc(p, need); hend = hp + need; } - tuple = TUPLE2(hp, ep->code[1], make_small(ep->code[2])); + tuple = TUPLE2(hp, ep->info.mfa.function, + make_small(ep->info.mfa.arity)); hp += 3; result = CONS(hp, tuple, result); hp += 2; @@ -5295,91 +5793,68 @@ exported_from_module(Process* p, /* Process whose heap to use. */ /* * Returns a list of all attributes for the module. - * - * Returns a tagged term, or 0 on error. */ Eterm attributes_for_module(Process* p, /* Process whose heap to use. */ - Eterm mod) /* Tagged atom for module. */ - + BeamCodeHeader* code_hdr) { - Module* modp; - BeamInstr* code; - Eterm* hp; byte* ext; Eterm result = NIL; - Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { - return THE_NON_VALUE; - } - - modp = erts_get_module(mod, erts_active_code_ix()); - if (modp == NULL) { - return THE_NON_VALUE; - } - code = modp->curr.code; - ext = (byte *) code[MI_ATTR_PTR]; + ext = code_hdr->attr_ptr; if (ext != NULL) { - hp = HAlloc(p, code[MI_ATTR_SIZE_ON_HEAP]); - end = hp + code[MI_ATTR_SIZE_ON_HEAP]; - result = erts_decode_ext(&hp, &MSO(p), &ext); + ErtsHeapFactory factory; + erts_factory_proc_prealloc_init(&factory, p, code_hdr->attr_size_on_heap); + result = erts_decode_ext(&factory, &ext, 0); if (is_value(result)) { - ASSERT(hp <= end); + erts_factory_close(&factory); } - HRelease(p,end,hp); } return result; } /* * Returns a list containing compilation information. - * - * Returns a tagged term, or 0 on error. */ Eterm compilation_info_for_module(Process* p, /* Process whose heap to use. */ - Eterm mod) /* Tagged atom for module. */ + BeamCodeHeader* code_hdr) { - Module* modp; - BeamInstr* code; - Eterm* hp; byte* ext; Eterm result = NIL; - Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { - return THE_NON_VALUE; - } - - modp = erts_get_module(mod, erts_active_code_ix()); - if (modp == NULL) { - return THE_NON_VALUE; - } - code = modp->curr.code; - ext = (byte *) code[MI_COMPILE_PTR]; + ext = code_hdr->compile_ptr; if (ext != NULL) { - hp = HAlloc(p, code[MI_COMPILE_SIZE_ON_HEAP]); - end = hp + code[MI_COMPILE_SIZE_ON_HEAP]; - result = erts_decode_ext(&hp, &MSO(p), &ext); + ErtsHeapFactory factory; + erts_factory_proc_prealloc_init(&factory, p, code_hdr->compile_size_on_heap); + result = erts_decode_ext(&factory, &ext, 0); if (is_value(result)) { - ASSERT(hp <= end); + erts_factory_close(&factory); } - HRelease(p,end,hp); } return result; } /* + * Returns the MD5 checksum for a module + */ + +Eterm +md5_of_module(Process* p, /* Process whose heap to use. */ + BeamCodeHeader* code_hdr) +{ + return new_binary(p, code_hdr->md5_ptr, MD5_SIZE); +} + +/* * Build a single {M,F,A,Loction} item to be part of * a stack trace. */ Eterm* erts_build_mfa_item(FunctionInfo* fi, Eterm* hp, Eterm args, Eterm* mfa_p) { - BeamInstr* current = fi->current; Eterm loc = NIL; if (fi->loc != LINE_INVALID_LOCATION) { @@ -5389,7 +5864,7 @@ erts_build_mfa_item(FunctionInfo* fi, Eterm* hp, Eterm args, Eterm* mfa_p) Eterm file_term = NIL; if (file == 0) { - Atom* ap = atom_tab(atom_val(fi->current[0])); + Atom* ap = atom_tab(atom_val(fi->mfa->module)); file_term = buf_to_intlist(&hp, ".erl", 4, NIL); file_term = buf_to_intlist(&hp, (char*)ap->name, ap->len, file_term); } else { @@ -5408,10 +5883,12 @@ erts_build_mfa_item(FunctionInfo* fi, Eterm* hp, Eterm args, Eterm* mfa_p) } if (is_list(args) || is_nil(args)) { - *mfa_p = TUPLE4(hp, current[0], current[1], args, loc); + *mfa_p = TUPLE4(hp, fi->mfa->module, fi->mfa->function, + args, loc); } else { - Eterm arity = make_small(current[2]); - *mfa_p = TUPLE4(hp, current[0], current[1], arity, loc); + Eterm arity = make_small(fi->mfa->arity); + *mfa_p = TUPLE4(hp, fi->mfa->module, fi->mfa->function, + arity, loc); } return hp + 5; } @@ -5422,9 +5899,9 @@ erts_build_mfa_item(FunctionInfo* fi, Eterm* hp, Eterm args, Eterm* mfa_p) * the function. */ void -erts_set_current_function(FunctionInfo* fi, BeamInstr* current) +erts_set_current_function(FunctionInfo* fi, ErtsCodeMFA* mfa) { - fi->current = current; + fi->mfa = mfa; fi->needed = 5; fi->loc = LINE_INVALID_LOCATION; } @@ -5433,13 +5910,13 @@ erts_set_current_function(FunctionInfo* fi, BeamInstr* current) /* * Returns a pointer to {module, function, arity}, or NULL if not found. */ -BeamInstr* +ErtsCodeMFA* find_function_from_pc(BeamInstr* pc) { FunctionInfo fi; erts_lookup_function_info(&fi, pc, 0); - return fi.current; + return fi.mfa; } /* @@ -5494,7 +5971,7 @@ code_get_chunk_2(BIF_ALIST_2) goto error; } if (!init_iff_file(stp, start, binary_size(Bin)) || - !scan_iff_file(stp, &chunk, 1, 1) || + !scan_iff_file(stp, &chunk, 1) || stp->chunks[0].start == NULL) { res = am_undefined; goto done; @@ -5543,12 +6020,12 @@ code_module_md5_1(BIF_ALIST_1) } stp->module = THE_NON_VALUE; /* Suppress diagnostiscs */ if (!init_iff_file(stp, bytes, binary_size(Bin)) || - !scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES, NUM_MANDATORY) || + !scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES) || !verify_chunks(stp)) { res = am_undefined; goto done; } - res = new_binary(p, stp->mod_md5, sizeof(stp->mod_md5)); + res = new_binary(p, stp->mod_md5, MD5_SIZE); done: erts_free_aligned_binary_bytes(temp_alloc); @@ -5556,31 +6033,28 @@ code_module_md5_1(BIF_ALIST_1) return res; } -#define WORDS_PER_FUNCTION 6 +#ifdef HIPE +#define WORDS_PER_FUNCTION (sizeof(ErtsCodeInfo) / sizeof(UWord) + 1) static BeamInstr* -make_stub(BeamInstr* fp, Eterm mod, Eterm func, Uint arity, Uint native, BeamInstr OpCode) +make_stub(ErtsCodeInfo* info, Eterm mod, Eterm func, Uint arity, Uint native, BeamInstr OpCode) { - fp[0] = (BeamInstr) BeamOp(op_i_func_info_IaaI); - fp[1] = native; - fp[2] = mod; - fp[3] = func; - fp[4] = arity; -#ifdef HIPE - if (native) { - fp[5] = BeamOpCode(op_move_return_nr); - hipe_mfa_save_orig_beam_op(mod, func, arity, fp+5); - } -#endif - fp[5] = OpCode; - return fp + WORDS_PER_FUNCTION; + DBG_TRACE_MFA(mod,func,arity,"make beam stub at %p", erts_codeinfo_to_code(info)); + ASSERT(WORDS_PER_FUNCTION == 6); + info->op = (BeamInstr) BeamOp(op_i_func_info_IaaI); + info->u.ncallee = (void (*)(void)) native; + info->mfa.module = mod; + info->mfa.function = func; + info->mfa.arity = arity; + erts_codeinfo_to_code(info)[0] = OpCode; + return erts_codeinfo_to_code(info)+1; } static byte* stub_copy_info(LoaderState* stp, int chunk, /* Chunk: ATTR_CHUNK or COMPILE_CHUNK */ byte* info, /* Where to store info. */ - BeamInstr* ptr_word, /* Where to store pointer into info. */ + byte** ptr_word, /* Where to store pointer into info. */ BeamInstr* size_word, /* Where to store size into info. */ BeamInstr* size_on_heap_word) /* Where to store size on heap. */ { @@ -5588,7 +6062,7 @@ stub_copy_info(LoaderState* stp, Uint size = stp->chunks[chunk].size; if (size != 0) { memcpy(info, stp->chunks[chunk].start, size); - *ptr_word = (BeamInstr) info; + *ptr_word = info; decoded_size = erts_decode_ext_size(info, size); if (decoded_size < 0) { return 0; @@ -5602,11 +6076,11 @@ stub_copy_info(LoaderState* stp, static int stub_read_export_table(LoaderState* stp) { - int i; + unsigned int i; GetInt(stp, 4, stp->num_exps); if (stp->num_exps > stp->num_functions) { - LoadError2(stp, "%d functions exported; only %d functions defined", + LoadError2(stp, "%u functions exported; only %u functions defined", stp->num_exps, stp->num_functions); } stp->export @@ -5620,7 +6094,7 @@ stub_read_export_table(LoaderState* stp) GetAtom(stp, n, stp->export[i].function); GetInt(stp, 4, n); if (n > MAX_REG) { - LoadError2(stp, "export table entry %d: absurdly high arity %d", i, n); + LoadError2(stp, "export table entry %u: absurdly high arity %u", i, n); } stp->export[i].arity = n; GetInt(stp, 4, n); /* Ignore label */ @@ -5632,22 +6106,17 @@ stub_read_export_table(LoaderState* stp) } static void -stub_final_touch(LoaderState* stp, BeamInstr* fp) +stub_final_touch(LoaderState* stp, ErtsCodeInfo* ci) { - int i; - int n = stp->num_exps; - Eterm mod = fp[2]; - Eterm function = fp[3]; - int arity = fp[4]; -#ifdef HIPE + unsigned int i; + unsigned int n = stp->num_exps; Lambda* lp; -#endif - if (is_bif(mod, function, arity)) { - fp[1] = 0; - fp[2] = 0; - fp[3] = 0; - fp[4] = 0; + if (is_bif(ci->mfa.module, ci->mfa.function, ci->mfa.arity)) { + ci->u.ncallee = NULL; + ci->mfa.module = 0; + ci->mfa.function = 0; + ci->mfa.arity = 0; return; } @@ -5656,9 +6125,14 @@ stub_final_touch(LoaderState* stp, BeamInstr* fp) */ for (i = 0; i < n; i++) { - if (stp->export[i].function == function && stp->export[i].arity == arity) { - Export* ep = erts_export_put(mod, function, arity); - ep->addressv[erts_staging_code_ix()] = fp+5; + if (stp->export[i].function == ci->mfa.function && + stp->export[i].arity == ci->mfa.arity) { + Export* ep = erts_export_put(ci->mfa.module, + ci->mfa.function, + ci->mfa.arity); + ep->addressv[erts_staging_code_ix()] = erts_codeinfo_to_code(ci); + DBG_TRACE_MFA_P(&ci->mfa,"set beam stub at %p in export at %p (code_ix=%d)", + erts_codeinfo_to_code(ci), ep, erts_staging_code_ix()); return; } } @@ -5668,16 +6142,14 @@ stub_final_touch(LoaderState* stp, BeamInstr* fp) * Search the lambda table to find out which. */ -#ifdef HIPE n = stp->num_lambdas; for (i = 0, lp = stp->lambdas; i < n; i++, lp++) { ErlFunEntry* fe = stp->lambdas[i].fe; - if (lp->function == function && lp->arity == arity) { - fp[5] = (Eterm) BeamOpCode(op_hipe_trap_call_closure); - fe->address = &(fp[5]); + if (lp->function == ci->mfa.function && lp->arity == ci->mfa.arity) { + *erts_codeinfo_to_code(ci) = (Eterm) BeamOpCode(op_hipe_trap_call_closure); + fe->address = erts_codeinfo_to_code(ci); } } -#endif return; } @@ -5686,10 +6158,9 @@ stub_final_touch(LoaderState* stp, BeamInstr* fp) [{Adr, Patchtyppe} | Addresses] and the address of a fun_entry. */ -int +static int patch(Eterm Addresses, Uint fe) { -#ifdef HIPE Eterm* listp; Eterm tuple; Eterm* tp; @@ -5725,15 +6196,13 @@ patch(Eterm Addresses, Uint fe) } -#endif return 1; } -int +static int patch_funentries(Eterm Patchlist) { -#ifdef HIPE while (!is_nil(Patchlist)) { Eterm Info; Eterm MFA; @@ -5807,50 +6276,41 @@ patch_funentries(Eterm Patchlist) fe = erts_get_fun_entry(Mod, uniq, index); fe->native_address = (Uint *)native_address; - /* Deliberate MEMORY LEAK of native fun entries!!! - * - * Uncomment line below when hipe code upgrade and purging works correctly. - * Today we may get cases when old (leaked) native code of a purged module - * gets called and tries to create instances of a deleted fun entry. - * - * Reproduced on a debug emulator with stdlib_test/qlc_SUITE:join_merge - * - * erts_refc_dec(&fe->refc, 1); - */ + erts_smp_refc_dec(&fe->refc, 1); if (!patch(Addresses, (Uint) fe)) return 0; } -#endif return 1; /* Signal that all went well */ } - /* * Do a dummy load of a module. No threaded code will be loaded. * Used for loading native code. * Will also patch all references to fun_entries to point to * the new fun_entries created. */ - Eterm -erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) +erts_make_stub_module(Process* p, Eterm hipe_magic_bin, Eterm Beam, Eterm Info) { Binary* magic; + Binary* hipe_magic; LoaderState* stp; + HipeLoaderState* hipe_stp; + HipeModule *hipe_code; BeamInstr Funcs; BeamInstr Patchlist; + Eterm MD5Bin; Eterm* tp; - BeamInstr* code = NULL; - BeamInstr* ptrs; + BeamCodeHeader* code_hdr; + BeamInstr* code_base; BeamInstr* fp; byte* info; - Uint ci; - int n; + Sint n; int code_size; int rval; - int i; + Sint i; byte* temp_alloc = NULL; byte* bytes; Uint size; @@ -5861,20 +6321,27 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) */ magic = erts_alloc_loader_state(); stp = ERTS_MAGIC_BIN_DATA(magic); + hipe_code = erts_alloc(ERTS_ALC_T_HIPE_LL, sizeof(*hipe_code)); - if (is_not_atom(Mod)) { + if (!is_internal_magic_ref(hipe_magic_bin) || + !(hipe_magic = erts_magic_ref2bin(hipe_magic_bin), + hipe_stp = hipe_get_loader_state(hipe_magic)) || + hipe_stp->module == NIL || hipe_stp->text_segment == 0) { goto error; } if (is_not_tuple(Info)) { goto error; } tp = tuple_val(Info); - if (tp[0] != make_arityval(2)) { + if (tp[0] != make_arityval(3)) { goto error; } Funcs = tp[1]; - Patchlist = tp[2]; - + Patchlist = tp[2]; + MD5Bin = tp[3]; + if (is_not_binary(MD5Bin) || (binary_size(MD5Bin) != MD5_SIZE)) { + goto error; + } if ((n = erts_list_length(Funcs)) < 0) { goto error; } @@ -5887,13 +6354,13 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * Scan the Beam binary and read the interesting sections. */ - stp->module = Mod; + stp->module = hipe_stp->module; stp->group_leader = p->group_leader; stp->num_functions = n; if (!init_iff_file(stp, bytes, size)) { goto error; } - if (!scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES, NUM_MANDATORY) || + if (!scan_iff_file(stp, chunk_types, NUM_CHUNK_TYPES) || !verify_chunks(stp)) { goto error; } @@ -5901,9 +6368,16 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) if (!read_code_header(stp)) { goto error; } - define_file(stp, "atom table", ATOM_CHUNK); - if (!load_atom_table(stp)) { - goto error; + if (stp->chunks[UTF8_ATOM_CHUNK].size > 0) { + define_file(stp, "utf8 atom table", UTF8_ATOM_CHUNK); + if (!load_atom_table(stp, ERTS_ATOM_ENC_UTF8)) { + goto error; + } + } else { + define_file(stp, "atom table", ATOM_CHUNK); + if (!load_atom_table(stp, ERTS_ATOM_ENC_LATIN1)) { + goto error; + } } define_file(stp, "export table", EXP_CHUNK); if (!stub_read_export_table(stp)) { @@ -5921,37 +6395,38 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * Allocate memory for the stub module. */ - code_size = ((WORDS_PER_FUNCTION+1)*n + MI_FUNCTIONS + 2) * sizeof(BeamInstr); - code_size += stp->chunks[ATTR_CHUNK].size; - code_size += stp->chunks[COMPILE_CHUNK].size; - code = erts_alloc_fnf(ERTS_ALC_T_CODE, code_size); - if (!code) { + code_size = (offsetof(BeamCodeHeader,functions) + + ((n+1) * sizeof(BeamInstr*)) + + (WORDS_PER_FUNCTION*n + 1) * sizeof(BeamInstr) + + stp->chunks[ATTR_CHUNK].size + + stp->chunks[COMPILE_CHUNK].size + + MD5_SIZE); + code_hdr = erts_alloc_fnf(ERTS_ALC_T_CODE, code_size); + if (!code_hdr) { goto error; } /* - * Initialize code area. + * Initialize code header. */ - code[MI_NUM_FUNCTIONS] = n; - code[MI_ATTR_PTR] = 0; - code[MI_ATTR_SIZE] = 0; - code[MI_ATTR_SIZE_ON_HEAP] = 0; - code[MI_COMPILE_PTR] = 0; - code[MI_COMPILE_SIZE] = 0; - code[MI_COMPILE_SIZE_ON_HEAP] = 0; - code[MI_LITERALS_START] = 0; - code[MI_LITERALS_END] = 0; - code[MI_LITERALS_OFF_HEAP] = 0; - code[MI_ON_LOAD_FUNCTION_PTR] = 0; - ci = MI_FUNCTIONS + n + 1; + code_hdr->num_functions = n; + code_hdr->attr_ptr = NULL; + code_hdr->attr_size = 0; + code_hdr->attr_size_on_heap = 0; + code_hdr->compile_ptr = NULL; + code_hdr->compile_size = 0; + code_hdr->compile_size_on_heap = 0; + code_hdr->literal_area = NULL; + code_hdr->on_load_function_ptr = NULL; + code_hdr->line_table = NULL; + code_hdr->md5_ptr = NULL; /* * Make stubs for all functions. */ - ptrs = code + MI_FUNCTIONS; - fp = code + ci; + fp = code_base = (BeamInstr*) &code_hdr->functions[n+1]; for (i = 0; i < n; i++) { Eterm* listp; Eterm tuple; @@ -5994,20 +6469,17 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * Set the pointer and make the stub. Put a return instruction * as the body until we know what kind of trap we should put there. */ - ptrs[i] = (BeamInstr) fp; -#ifdef HIPE + code_hdr->functions[i] = (ErtsCodeInfo*)fp; op = (Eterm) BeamOpCode(op_hipe_trap_call); /* Might be changed later. */ -#else - op = (Eterm) BeamOpCode(op_move_return_nr); -#endif - fp = make_stub(fp, Mod, func, arity, (Uint)native_address, op); + fp = make_stub((ErtsCodeInfo*)fp, hipe_stp->module, func, arity, + (Uint)native_address, op); } /* * Insert the last pointer and the int_code_end instruction. */ - ptrs[i] = (BeamInstr) fp; + code_hdr->functions[i] = (ErtsCodeInfo*)fp; *fp++ = (BeamInstr) BeamOp(op_int_code_end); /* @@ -6016,25 +6488,44 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) info = (byte *) fp; info = stub_copy_info(stp, ATTR_CHUNK, info, - code+MI_ATTR_PTR, - code+MI_ATTR_SIZE, - code+MI_ATTR_SIZE_ON_HEAP); + &code_hdr->attr_ptr, + &code_hdr->attr_size, + &code_hdr->attr_size_on_heap); if (info == NULL) { goto error; } info = stub_copy_info(stp, COMPILE_CHUNK, info, - code+MI_COMPILE_PTR, - code+MI_COMPILE_SIZE, - code+MI_COMPILE_SIZE_ON_HEAP); + &code_hdr->compile_ptr, + &code_hdr->compile_size, + &code_hdr->compile_size_on_heap); if (info == NULL) { goto error; } + { + byte *tmp = NULL; + byte *md5 = NULL; + if ((md5 = erts_get_aligned_binary_bytes(MD5Bin, &tmp)) != NULL) { + sys_memcpy(info, md5, MD5_SIZE); + code_hdr->md5_ptr = info; + } + erts_free_aligned_binary_bytes(tmp); + } + + /* + * Initialise HiPE module + */ + 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; + hipe_code->first_hipe_ref = hipe_stp->new_hipe_refs; + hipe_code->first_hipe_sdesc = hipe_stp->new_hipe_sdesc; /* * Insert the module in the module table. */ - rval = insert_new_code(p, 0, p->group_leader, Mod, code, code_size); + rval = stub_insert_new_code(p, 0, p->group_leader, hipe_stp->module, + code_hdr, code_size, hipe_code); if (rval != NIL) { goto error; } @@ -6043,25 +6534,76 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * Export all stub functions and insert the correct type of HiPE trap. */ - fp = code + ci; + fp = code_base; for (i = 0; i < n; i++) { - stub_final_touch(stp, fp); + stub_final_touch(stp, (ErtsCodeInfo*)fp); fp += WORDS_PER_FUNCTION; } if (patch_funentries(Patchlist)) { + Eterm mod = hipe_stp->module; + /* Prevent code from being freed */ + hipe_stp->text_segment = 0; + hipe_stp->data_segment = 0; + hipe_stp->new_hipe_refs = NULL; + hipe_stp->new_hipe_sdesc = NULL; + erts_free_aligned_binary_bytes(temp_alloc); free_loader_state(magic); - return Mod; + hipe_free_loader_state(hipe_stp); + + return mod; } error: + erts_free(ERTS_ALC_T_HIPE_LL, hipe_code); erts_free_aligned_binary_bytes(temp_alloc); free_loader_state(magic); BIF_ERROR(p, BADARG); } +int erts_commit_hipe_patch_load(Eterm hipe_magic_bin) +{ + Binary* hipe_magic; + HipeLoaderState* hipe_stp; + HipeModule *hipe_code; + Module* modp; + + if (!is_internal_magic_ref(hipe_magic_bin) || + !(hipe_magic = erts_magic_ref2bin(hipe_magic_bin), + hipe_stp = hipe_get_loader_state(hipe_magic)) || + hipe_stp->module == NIL || hipe_stp->text_segment == 0) { + return 0; + } + + modp = erts_get_module(hipe_stp->module, erts_active_code_ix()); + if (!modp) + return 0; + + /* + * Initialise HiPE module + */ + 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; + hipe_code->first_hipe_ref = hipe_stp->new_hipe_refs; + hipe_code->first_hipe_sdesc = hipe_stp->new_hipe_sdesc; + + modp->curr.hipe_code = hipe_code; + + /* Prevent code from being freed */ + hipe_stp->text_segment = 0; + hipe_stp->data_segment = 0; + hipe_stp->new_hipe_refs = NULL; + hipe_stp->new_hipe_sdesc = NULL; + + return 1; +} + #undef WORDS_PER_FUNCTION +#endif /* HIPE */ + static int safe_mul(UWord a, UWord b, UWord* resp) { @@ -6075,3 +6617,46 @@ static int safe_mul(UWord a, UWord b, UWord* resp) } } +#ifdef ENABLE_DBG_TRACE_MFA + +#define MFA_MAX 10 +Eterm dbg_trace_m[MFA_MAX]; +Eterm dbg_trace_f[MFA_MAX]; +Uint dbg_trace_a[MFA_MAX]; +unsigned int dbg_trace_ix = 0; + +void dbg_set_traced_mfa(const char* m, const char* f, Uint a) +{ + unsigned i = dbg_trace_ix++; + ASSERT(i < MFA_MAX); + dbg_trace_m[i] = am_atom_put(m, strlen(m)); + dbg_trace_f[i] = am_atom_put(f, strlen(f)); + dbg_trace_a[i] = a; +} + +int dbg_is_traced_mfa(Eterm m, Eterm f, Uint a) +{ + unsigned int i; + for (i = 0; i < dbg_trace_ix; ++i) { + if (m == dbg_trace_m[i] && + (!f || (f == dbg_trace_f[i] && a == dbg_trace_a[i]))) { + + return i+1; + } + } + return 0; +} + +void dbg_vtrace_mfa(unsigned ix, const char* format, ...) +{ + va_list arglist; + va_start(arglist, format); + ASSERT(--ix < MFA_MAX); + erts_fprintf(stderr, "MFA TRACE %T:%T/%u: ", + dbg_trace_m[ix], dbg_trace_f[ix], (int)dbg_trace_a[ix]); + + erts_vfprintf(stderr, format, arglist); + va_end(arglist); +} + +#endif /* ENABLE_DBG_TRACE_MFA */ |