diff options
Diffstat (limited to 'erts/emulator/beam/beam_load.c')
-rw-r--r-- | erts/emulator/beam/beam_load.c | 1118 |
1 files changed, 644 insertions, 474 deletions
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index b70e5b9a2d..0afdedf6c2 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2014. All Rights Reserved. + * Copyright Ericsson AB 1996-2016. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -32,6 +32,7 @@ #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" @@ -80,7 +81,7 @@ 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 @@ -104,7 +105,7 @@ 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. */ @@ -282,11 +283,12 @@ 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. */ @@ -301,13 +303,13 @@ 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). */ /* @@ -321,8 +323,8 @@ 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. */ @@ -331,8 +333,8 @@ 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. */ @@ -341,13 +343,13 @@ 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; @@ -478,9 +480,9 @@ static void free_loader_state(Binary* magic); static ErlHeapFragment* new_literal_fragment(Uint size); static void free_literal_fragment(ErlHeapFragment*); 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 Eterm stub_insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, + Eterm group_leader, Eterm module, + BeamCodeHeader* code, Uint size); 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); @@ -512,7 +514,7 @@ 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); @@ -526,15 +528,15 @@ 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 get_module_info(Process* p, ErtsCodeIndex code_ix, - BeamInstr* code, Eterm module, Eterm what); + BeamCodeHeader*, Eterm module, Eterm what); static Eterm exported_from_module(Process* p, ErtsCodeIndex code_ix, Eterm mod); -static Eterm functions_in_module(Process* p, BeamInstr* code); -static Eterm attributes_for_module(Process* p, BeamInstr* code); -static Eterm compilation_info_for_module(Process* p, BeamInstr* code); -static Eterm md5_of_module(Process* p, BeamInstr* code); -static Eterm has_native(BeamInstr* code); -static Eterm native_addresses(Process* p, BeamInstr* code); +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*); int patch_funentries(Eterm Patchlist); int patch(Eterm Addresses, Uint fe); static int safe_mul(UWord a, UWord b, UWord* resp); @@ -601,6 +603,7 @@ 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) @@ -641,20 +644,27 @@ 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; - stp->code[MI_MD5_PTR] = 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. @@ -757,8 +767,11 @@ 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 @@ -767,27 +780,89 @@ 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; + + /* + * There is an -on_load() function. We will keep the current + * code, but we must turn off any tracing. + */ + + for (i = 0; i < export_list_size(code_ix); i++) { + Export *ep = export_list(i, code_ix); + if (ep == NULL || ep->code[0] != module) { + continue; + } + if (ep->addressv[code_ix] == ep->code+3) { + if (ep->code[3] == (BeamInstr) em_apply_bif) { + continue; + } else if (ep->code[3] == + (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->code+3); + ep->addressv[code_ix] = (BeamInstr *) ep->code[4]; + ep->code[4] = 0; + } + ASSERT(ep->code[4] == 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; + inst_p->nif = 0; + inst_p->num_breakpoints = 0; + inst_p->num_traced_exports = 0; } + 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. @@ -799,7 +874,8 @@ 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; /* @@ -810,7 +886,6 @@ erts_finish_loading(Binary* magic, Process* c_p, retval = am_on_load; } - load_error: free_loader_state(magic); return retval; } @@ -831,7 +906,8 @@ 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; @@ -870,13 +946,30 @@ erts_module_for_prepared_code(Binary* magic) 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) { @@ -891,7 +984,7 @@ 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); + ERTS_INIT_HEAP_FRAG(bp, size, size); return bp; } @@ -920,9 +1013,14 @@ 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); @@ -994,9 +1092,9 @@ 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) +stub_insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, + Eterm group_leader, Eterm module, + BeamCodeHeader* code_hdr, Uint size) { Module* modp; Eterm retval; @@ -1016,7 +1114,7 @@ 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. */ @@ -1024,7 +1122,7 @@ insert_new_code(Process *c_p, ErtsProcLocks c_p_locks, * 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; } @@ -1242,7 +1340,7 @@ verify_chunks(LoaderState* stp) static int load_atom_table(LoaderState* stp) { - int i; + unsigned int i; GetInt(stp, 4, stp->num_atoms); stp->num_atoms++; @@ -1287,13 +1385,13 @@ 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; @@ -1301,17 +1399,17 @@ 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; @@ -1339,12 +1437,12 @@ 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 @@ -1362,18 +1460,18 @@ 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. @@ -1392,7 +1490,7 @@ 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); @@ -1432,7 +1530,7 @@ 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) { @@ -1452,12 +1550,12 @@ 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; @@ -1478,7 +1576,7 @@ read_lambda_table(LoaderState* stp) static int read_literal_table(LoaderState* stp) { - int i; + unsigned int i; uLongf uncompressed_sz; byte* uncompressed = 0; @@ -1500,7 +1598,7 @@ read_literal_table(LoaderState* stp) } for (i = 0; i < stp->num_literals; i++) { - int sz; + Uint sz; Sint heap_size; byte* p; Eterm val; @@ -1509,17 +1607,17 @@ read_literal_table(LoaderState* stp) 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); + LoadError1(stp, "literal %u: bad external format", i); } if (heap_size > 0) { - erts_factory_message_init(&factory, NULL, NULL, - new_literal_fragment(heap_size)); + 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); + val = erts_decode_ext(&factory, &p, 0); if (is_non_value(val)) { - LoadError1(stp, "literal %d: bad external format", i); + LoadError1(stp, "literal %u: bad external format", i); } erts_factory_close(&factory); stp->literals[i].heap_frags = factory.heap_frags; @@ -1527,9 +1625,9 @@ read_literal_table(LoaderState* stp) } else { erts_factory_dummy_init(&factory); - val = erts_decode_ext(&factory, &p); + val = erts_decode_ext(&factory, &p, 0); if (is_non_value(val)) { - LoadError1(stp, "literal %d: bad external format", i); + LoadError1(stp, "literal %u: bad external format", i); } ASSERT(is_immed(val)); stp->literals[i].heap_frags = NULL; @@ -1552,9 +1650,9 @@ 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; @@ -1633,7 +1731,7 @@ 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; @@ -1669,9 +1767,9 @@ 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; @@ -1695,6 +1793,10 @@ 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; @@ -1735,7 +1837,7 @@ 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 @@ -1754,13 +1856,14 @@ 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) @@ -1776,7 +1879,7 @@ 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; @@ -1793,23 +1896,23 @@ load_code(LoaderState* stp) FUNC_INFO_SZ = 5 }; - 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); } @@ -1847,9 +1950,7 @@ 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); } @@ -1894,15 +1995,14 @@ 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); @@ -1998,42 +2098,47 @@ 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. @@ -2056,7 +2161,42 @@ 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++; } @@ -2167,14 +2307,11 @@ 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); @@ -2185,6 +2322,10 @@ 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); @@ -2193,14 +2334,11 @@ 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", @@ -2244,7 +2382,7 @@ 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) { @@ -2357,20 +2495,13 @@ 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); @@ -2393,10 +2524,9 @@ 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); } @@ -2436,15 +2566,15 @@ load_code(LoaderState* stp) stp->arity = code[ci-1]; 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] = (BeamInstr*) 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); } @@ -2458,7 +2588,6 @@ 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; @@ -2476,14 +2605,14 @@ 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]; @@ -2515,7 +2644,11 @@ 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; @@ -2530,7 +2663,10 @@ 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; } } @@ -2674,12 +2810,18 @@ 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. @@ -2696,17 +2838,17 @@ 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; } @@ -2832,23 +2974,16 @@ 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; @@ -3272,14 +3407,14 @@ 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; @@ -3296,7 +3431,7 @@ 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; @@ -3319,14 +3454,14 @@ 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; @@ -3343,7 +3478,7 @@ 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; @@ -3899,9 +4034,7 @@ gen_make_fun2(LoaderState* stp, GenOpArg idx) /* * 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). + * to use a garbage collecting implementation. */ static GenOp* gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, @@ -3956,10 +4089,6 @@ gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg 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, @@ -3986,23 +4115,19 @@ gen_guard_bif2(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, 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[2] = Live; + op->a[3] = S1; + op->a[4] = S2; op->a[5] = Dst; return op; } /* * 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, @@ -4033,10 +4158,10 @@ gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, 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[2] = Live; + op->a[3] = S1; + op->a[4] = S2; + op->a[5] = S3; op->a[6] = Dst; op->next = NULL; return op; @@ -4258,6 +4383,53 @@ gen_get_map_element(LoaderState* stp, GenOpArg Fail, GenOpArg Src, return op; } +static int +hash_internal_genop_arg(LoaderState* stp, GenOpArg Key, Uint32* hx) +{ + switch (Key.type) { + case TAG_a: + *hx = atom_tab(atom_val(Key.val))->slot.bucket.hvalue; + return 1; + case TAG_i: + *hx = Key.val; + return 1; + case TAG_n: + *hx = make_internal_hash(NIL); + return 1; + case TAG_q: + *hx = make_internal_hash(stp->literals[Key.val].term); + return 1; + default: + return 0; + } +} + + +static GenOp* +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) @@ -4270,6 +4442,7 @@ gen_get_map_elements(LoaderState* stp, GenOpArg Fail, GenOpArg Src, int good_hash; #endif + ERTS_UNDEF(hx, 0); ASSERT(Size.type == TAG_u); NEW_GENOP(stp, op); @@ -4323,7 +4496,7 @@ gen_has_map_fields(LoaderState* stp, GenOpArg Fail, GenOpArg Src, 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 = 0; /* x(0); normally not used */ + op->a[3+2*i+1].val = SCRATCH_X_REG; /* Ignore result */ } return op; } @@ -4336,8 +4509,8 @@ gen_has_map_fields(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; @@ -4362,77 +4535,80 @@ 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)) + + 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] = (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; ErlOffHeap code_off_heap; + ErtsLiteralArea *literal_area; + Uint lit_asize; ERTS_INIT_OFF_HEAP(&code_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; + 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 (stp->literals[i].heap_frags) { - move_multi_frags(&ptr, &code_off_heap, stp->literals[i].heap_frags, - &stp->literals[i].term, 1); + 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))); } - else ASSERT(is_immed(stp->literals[i].term)); } - code[MI_LITERALS_OFF_HEAP] = (BeamInstr) code_off_heap.first; + literal_area->off_heap = code_off_heap.first; lp = stp->literal_patches; while (lp != 0) { BeamInstr* op_ptr; Literal* lit; - op_ptr = code + lp->pos; + op_ptr = codev + lp->pos; lit = &stp->literals[op_ptr[0]]; 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); @@ -4440,52 +4616,49 @@ 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); } @@ -4497,13 +4670,13 @@ 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) { @@ -4513,9 +4686,9 @@ 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); @@ -4523,7 +4696,7 @@ 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); { @@ -4531,7 +4704,7 @@ freeze_code(LoaderState* stp) CHKBLK(ERTS_ALC_T_CODE,code); sys_memcpy(md5_sum, stp->mod_md5, MD5_SIZE); CHKBLK(ERTS_ALC_T_CODE,code); - code[MI_MD5_PTR] = (BeamInstr) md5_sum; + code_hdr->md5_ptr = md5_sum; CHKBLK(ERTS_ALC_T_CODE,code); } CHKBLK(ERTS_ALC_T_CODE,code); @@ -4540,7 +4713,7 @@ freeze_code(LoaderState* stp) * Make sure that we have not overflowed the allocated code space. */ ASSERT(str_table + strtab_size + attr_size + compile_size + MD5_SIZE == - ((byte *) code) + size); + ((byte *) code_hdr) + size); /* * Patch all instructions that refer to the string table. @@ -4552,46 +4725,47 @@ 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: @@ -4599,19 +4773,19 @@ 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. @@ -4620,14 +4794,13 @@ 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. @@ -4647,10 +4820,10 @@ final_touch(LoaderState* stp) ep->addressv[erts_staging_code_ix()] = address; } else { /* - * 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; } } @@ -4674,8 +4847,8 @@ 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; } } @@ -4688,7 +4861,7 @@ 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) { /* @@ -4710,31 +4883,25 @@ 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; + 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; + GenOp* first = st->genop; + GenOp* keep = NULL; Uint* pc; - int rval; 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; @@ -4752,7 +4919,7 @@ 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; @@ -4801,7 +4968,8 @@ 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; @@ -4913,19 +5081,9 @@ 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 @@ -4934,21 +5092,22 @@ 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: { @@ -4973,22 +5132,19 @@ 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. @@ -5000,6 +5156,12 @@ 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; @@ -5019,14 +5181,10 @@ 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 @@ -5038,21 +5196,12 @@ 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 @@ -5304,14 +5453,14 @@ 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; } @@ -5371,7 +5520,7 @@ erts_module_info_0(Process* p, Eterm module) { Module* modp; ErtsCodeIndex code_ix = erts_active_code_ix(); - BeamInstr* code; + BeamCodeHeader* code_hdr; Eterm *hp; Eterm list = NIL; Eterm tup; @@ -5385,13 +5534,13 @@ erts_module_info_0(Process* p, Eterm module) return THE_NON_VALUE; } - code = modp->curr.code; - if (code == NULL) { + code_hdr = modp->curr.code_hdr; + if (code_hdr == NULL) { return THE_NON_VALUE; } #define BUILD_INFO(What) \ - tup = get_module_info(p, code_ix, code, module, What); \ + tup = get_module_info(p, code_ix, code_hdr, module, What); \ hp = HAlloc(p, 5); \ tup = TUPLE2(hp, What, tup); \ hp += 3; \ @@ -5414,7 +5563,7 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) { Module* modp; ErtsCodeIndex code_ix = erts_active_code_ix(); - BeamInstr* code; + BeamCodeHeader* code_hdr; if (is_not_atom(module)) { return THE_NON_VALUE; @@ -5425,34 +5574,34 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) return THE_NON_VALUE; } - code = modp->curr.code; - if (code == NULL) { + code_hdr = modp->curr.code_hdr; + if (code_hdr == NULL) { return THE_NON_VALUE; } - return get_module_info(p, code_ix, code, module, what); + return get_module_info(p, code_ix, code_hdr, module, what); } static Eterm -get_module_info(Process* p, ErtsCodeIndex code_ix, BeamInstr* code, +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_md5) { - return md5_of_module(p, code); + return md5_of_module(p, code_hdr); } else if (what == am_exports) { return exported_from_module(p, code_ix, module); } else if (what == am_functions) { - return functions_in_module(p, code); + return functions_in_module(p, code_hdr); } else if (what == am_attributes) { - return attributes_for_module(p, code); + return attributes_for_module(p, code_hdr); } else if (what == am_compile) { - return compilation_info_for_module(p, code); + return compilation_info_for_module(p, code_hdr); } else if (what == am_native_addresses) { - return native_addresses(p, code); + return native_addresses(p, code_hdr); } else if (what == am_native) { - return has_native(code); + return has_native(code_hdr); } return THE_NON_VALUE; } @@ -5464,7 +5613,7 @@ get_module_info(Process* p, ErtsCodeIndex code_ix, BeamInstr* code, Eterm functions_in_module(Process* p, /* Process whose heap to use. */ - BeamInstr* code) + BeamCodeHeader* code_hdr) { int i; Uint num_functions; @@ -5473,12 +5622,12 @@ functions_in_module(Process* p, /* Process whose heap to use. */ Eterm* hp_end; Eterm result = NIL; - 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]; + BeamInstr* func_info = code_hdr->functions[i]; Eterm name = (Eterm) func_info[3]; int arity = (int) func_info[4]; Eterm tuple; @@ -5504,27 +5653,49 @@ functions_in_module(Process* p, /* Process whose heap to use. */ */ static Eterm -has_native(BeamInstr *code) +has_native(BeamCodeHeader *code_hdr) { Eterm result = am_false; #ifdef HIPE - if (erts_is_module_native(code)) { + 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; + if (erts_refc_dectest(&bptr->refc, 0) == 0) { + erts_bin_free(bptr); + } + oh = oh->next; + } + erts_free(ERTS_ALC_T_LITERAL, literal_area); +} + int -erts_is_module_native(BeamInstr* code) +erts_is_module_native(BeamCodeHeader* code_hdr) { Uint i, num_functions; /* Check NativeAdress of first real function in module */ - if (code != NULL) { - num_functions = code[MI_NUM_FUNCTIONS]; + if (code_hdr != NULL) { + num_functions = code_hdr->num_functions; for (i=0; i<num_functions; i++) { - BeamInstr* func_info = (BeamInstr *) code[MI_FUNCTIONS+i]; + BeamInstr* func_info = (BeamInstr *) code_hdr->functions[i]; Eterm name = (Eterm) func_info[3]; if (is_atom(name)) { return func_info[1] != 0; @@ -5541,7 +5712,7 @@ erts_is_module_native(BeamInstr* code) */ static Eterm -native_addresses(Process* p, BeamInstr* code) +native_addresses(Process* p, BeamCodeHeader* code_hdr) { int i; Eterm* hp; @@ -5550,12 +5721,12 @@ native_addresses(Process* p, BeamInstr* code) Eterm* hp_end; Eterm result = NIL; - 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]; + BeamInstr* func_info = code_hdr->functions[i]; Eterm name = (Eterm) func_info[3]; int arity = (int) func_info[4]; Eterm tuple; @@ -5621,16 +5792,16 @@ exported_from_module(Process* p, /* Process whose heap to use. */ Eterm attributes_for_module(Process* p, /* Process whose heap to use. */ - BeamInstr* code) + BeamCodeHeader* code_hdr) { byte* ext; Eterm result = NIL; - ext = (byte *) code[MI_ATTR_PTR]; + ext = code_hdr->attr_ptr; if (ext != NULL) { ErtsHeapFactory factory; - erts_factory_proc_prealloc_init(&factory, p, code[MI_ATTR_SIZE_ON_HEAP]); - result = erts_decode_ext(&factory, &ext); + erts_factory_proc_prealloc_init(&factory, p, code_hdr->attr_size_on_heap); + result = erts_decode_ext(&factory, &ext, 0); if (is_value(result)) { erts_factory_close(&factory); } @@ -5644,16 +5815,16 @@ attributes_for_module(Process* p, /* Process whose heap to use. */ Eterm compilation_info_for_module(Process* p, /* Process whose heap to use. */ - BeamInstr* code) + BeamCodeHeader* code_hdr) { byte* ext; Eterm result = NIL; - ext = (byte *) code[MI_COMPILE_PTR]; + ext = code_hdr->compile_ptr; if (ext != NULL) { ErtsHeapFactory factory; - erts_factory_proc_prealloc_init(&factory, p, code[MI_COMPILE_SIZE_ON_HEAP]); - result = erts_decode_ext(&factory, &ext); + erts_factory_proc_prealloc_init(&factory, p, code_hdr->compile_size_on_heap); + result = erts_decode_ext(&factory, &ext, 0); if (is_value(result)) { erts_factory_close(&factory); } @@ -5667,9 +5838,9 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ Eterm md5_of_module(Process* p, /* Process whose heap to use. */ - BeamInstr* code) + BeamCodeHeader* code_hdr) { - return new_binary(p, (byte *) code[MI_MD5_PTR], MD5_SIZE); + return new_binary(p, code_hdr->md5_ptr, MD5_SIZE); } /* @@ -5868,7 +6039,7 @@ make_stub(BeamInstr* fp, Eterm mod, Eterm func, Uint arity, Uint native, BeamIns fp[4] = arity; #ifdef HIPE if (native) { - fp[5] = BeamOpCode(op_move_return_nr); + fp[5] = BeamOpCode(op_move_return_n); hipe_mfa_save_orig_beam_op(mod, func, arity, fp+5); } #endif @@ -5880,7 +6051,7 @@ 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. */ { @@ -5888,7 +6059,7 @@ 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; @@ -5902,11 +6073,11 @@ 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 @@ -5920,7 +6091,7 @@ 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 */ @@ -5934,8 +6105,8 @@ stub_read_export_table(LoaderState* stp) static void stub_final_touch(LoaderState* stp, BeamInstr* fp) { - int i; - int n = stp->num_exps; + unsigned int i; + unsigned int n = stp->num_exps; Eterm mod = fp[2]; Eterm function = fp[3]; int arity = fp[4]; @@ -6143,15 +6314,14 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) 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; @@ -6225,39 +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_size += MD5_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; - code[MI_MD5_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; @@ -6300,11 +6469,11 @@ 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; + code_hdr->functions[i] = fp; #ifdef HIPE op = (Eterm) BeamOpCode(op_hipe_trap_call); /* Might be changed later. */ #else - op = (Eterm) BeamOpCode(op_move_return_nr); + op = (Eterm) BeamOpCode(op_move_return_n); #endif fp = make_stub(fp, Mod, func, arity, (Uint)native_address, op); } @@ -6313,7 +6482,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * Insert the last pointer and the int_code_end instruction. */ - ptrs[i] = (BeamInstr) fp; + code_hdr->functions[i] = fp; *fp++ = (BeamInstr) BeamOp(op_int_code_end); /* @@ -6322,16 +6491,16 @@ 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; } @@ -6340,7 +6509,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) byte *md5 = NULL; if ((md5 = erts_get_aligned_binary_bytes(MD5Bin, &tmp)) != NULL) { sys_memcpy(info, md5, MD5_SIZE); - code[MI_MD5_PTR] = (BeamInstr) info; + code_hdr->md5_ptr = info; } erts_free_aligned_binary_bytes(tmp); } @@ -6349,7 +6518,8 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) * 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, Mod, + code_hdr, code_size); if (rval != NIL) { goto error; } @@ -6358,7 +6528,7 @@ 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); fp += WORDS_PER_FUNCTION; |