X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fruntime%2Fgencgc.c;h=175be61c3cabdb2e259bc7f06e5aee39eb74a493;hb=dd54f9e004a0a83d1328e94648f48dcc27e0be5b;hp=fea3194a267a5bcbedaccc9da6e43b7952bbcc72;hpb=137ba2db2d362f03754ccd080ddbe96f7e3c5dc7;p=sbcl.git diff --git a/src/runtime/gencgc.c b/src/runtime/gencgc.c old mode 100644 new mode 100755 index fea3194..175be61 --- a/src/runtime/gencgc.c +++ b/src/runtime/gencgc.c @@ -86,6 +86,9 @@ os_vm_size_t large_object_size = 4 * GENCGC_CARD_BYTES; os_vm_size_t large_object_size = 4 * PAGE_BYTES; #endif +/* Largest allocation seen since last GC. */ +os_vm_size_t large_allocation = 0; + /* * debugging @@ -165,7 +168,7 @@ boolean gc_active_p = 0; static boolean conservative_stack = 1; /* An array of page structures is allocated on gc initialization. - * This helps quickly map between an address its page structure. + * This helps to quickly map between an address and its page structure. * page_table_pages is set from the size of the dynamic space. */ page_index_t page_table_pages; struct page *page_table; @@ -907,15 +910,15 @@ struct new_area { size_t size; }; static struct new_area (*new_areas)[]; -static long new_areas_index; -long max_new_areas; +static size_t new_areas_index; +size_t max_new_areas; /* Add a new area to new_areas. */ static void add_new_area(page_index_t first_page, size_t offset, size_t size) { - unsigned long new_area_start,c; - long i; + size_t new_area_start, c; + ssize_t i; /* Ignore if full. */ if (new_areas_index >= NUM_NEW_AREAS) @@ -939,7 +942,7 @@ add_new_area(page_index_t first_page, size_t offset, size_t size) /* Search backwards for a prior area that this follows from. If found this will save adding a new area. */ for (i = new_areas_index-1, c = 0; (i >= 0) && (c < 8); i--, c++) { - unsigned long area_end = + size_t area_end = npage_bytes((*new_areas)[i].page) + (*new_areas)[i].offset + (*new_areas)[i].size; @@ -1454,8 +1457,9 @@ general_copy_large_object(lispobj object, long nwords, boolean boxedp) page_index_t next_page; page_bytes_t old_bytes_used; - /* BOXED ONLY? */ - /* Note: Any page write-protection must be removed, else a + /* FIXME: This comment is somewhat stale. + * + * Note: Any page write-protection must be removed, else a * later scavenge_newspace may incorrectly not scavenge these * pages. This would not be necessary if they are added to the * new areas, but let's do it for them all (they'll probably @@ -1511,13 +1515,17 @@ general_copy_large_object(lispobj object, long nwords, boolean boxedp) next_page++; while ((old_bytes_used == GENCGC_CARD_BYTES) && (page_table[next_page].gen == from_space) && + /* FIXME: It is not obvious to me why this is necessary + * as a loop condition: it seems to me that the + * region_start_offset test should be sufficient, but + * experimentally that is not the case. --NS + * 2011-11-28 */ + (boxedp ? + page_boxed_p(next_page) : + page_allocated_no_region_p(next_page)) && page_table[next_page].large_object && (page_table[next_page].region_start_offset == npage_bytes(next_page - first_page))) { - if (boxedp) - gc_assert(page_boxed_p(next_page)); - else - gc_assert(page_allocated_no_region_p(next_page)); /* Checks out OK, free the page. Don't need to both zeroing * pages as this should have been done before shrinking the * object. These pages shouldn't be write-protected, even if @@ -1581,23 +1589,7 @@ copy_large_unboxed_object(lispobj object, long nwords) lispobj copy_unboxed_object(lispobj object, long nwords) { - long tag; - lispobj *new; - - gc_assert(is_lisp_pointer(object)); - gc_assert(from_space_p(object)); - gc_assert((nwords & 0x01) == 0); - - /* Get tag of object. */ - tag = lowtag_of(object); - - /* Allocate space. */ - new = gc_quick_alloc_unboxed(nwords*N_WORD_BYTES); - - memcpy(new,native_pointer(object),nwords*N_WORD_BYTES); - - /* Return Lisp pointer of new object. */ - return ((lispobj) new) | tag; + return gc_general_copy_object(object, nwords, UNBOXED_PAGE_FLAG); } @@ -1619,13 +1611,13 @@ static lispobj trans_boxed(lispobj object); * Currently only absolute fixups to the constant vector, or to the * code area are checked. */ void -sniff_code_object(struct code *code, unsigned long displacement) +sniff_code_object(struct code *code, os_vm_size_t displacement) { #ifdef LISP_FEATURE_X86 long nheader_words, ncode_words, nwords; - void *p; - void *constants_start_addr = NULL, *constants_end_addr; - void *code_start_addr, *code_end_addr; + os_vm_address_t constants_start_addr = NULL, constants_end_addr, p; + os_vm_address_t code_start_addr, code_end_addr; + os_vm_address_t code_addr = (os_vm_address_t)code; int fixup_found = 0; if (!check_code_fixups) @@ -1637,10 +1629,10 @@ sniff_code_object(struct code *code, unsigned long displacement) nheader_words = HeaderValue(*(lispobj *)code); nwords = ncode_words + nheader_words; - constants_start_addr = (void *)code + 5*N_WORD_BYTES; - constants_end_addr = (void *)code + nheader_words*N_WORD_BYTES; - code_start_addr = (void *)code + nheader_words*N_WORD_BYTES; - code_end_addr = (void *)code + nwords*N_WORD_BYTES; + constants_start_addr = code_addr + 5*N_WORD_BYTES; + constants_end_addr = code_addr + nheader_words*N_WORD_BYTES; + code_start_addr = code_addr + nheader_words*N_WORD_BYTES; + code_end_addr = code_addr + nwords*N_WORD_BYTES; /* Work through the unboxed code. */ for (p = code_start_addr; p < code_end_addr; p++) { @@ -1657,8 +1649,8 @@ sniff_code_object(struct code *code, unsigned long displacement) /* Check for code references. */ /* Check for a 32 bit word that looks like an absolute reference to within the code adea of the code object. */ - if ((data >= (code_start_addr-displacement)) - && (data < (code_end_addr-displacement))) { + if ((data >= (void*)(code_start_addr-displacement)) + && (data < (void*)(code_end_addr-displacement))) { /* function header */ if ((d4 == 0x5e) && (((unsigned)p - 4 - 4*HeaderValue(*((unsigned *)p-1))) == @@ -1700,8 +1692,8 @@ sniff_code_object(struct code *code, unsigned long displacement) /* Check for a 32 bit word that looks like an absolute reference to within the constant vector. Constant references will be aligned. */ - if ((data >= (constants_start_addr-displacement)) - && (data < (constants_end_addr-displacement)) + if ((data >= (void*)(constants_start_addr-displacement)) + && (data < (void*)(constants_end_addr-displacement)) && (((unsigned)data & 0x3) == 0)) { /* Mov eax,m32 */ if (d1 == 0xa1) { @@ -1799,11 +1791,12 @@ gencgc_apply_code_fixups(struct code *old_code, struct code *new_code) /* x86-64 uses pc-relative addressing instead of this kludge */ #ifndef LISP_FEATURE_X86_64 long nheader_words, ncode_words, nwords; - void *constants_start_addr, *constants_end_addr; - void *code_start_addr, *code_end_addr; + os_vm_address_t constants_start_addr, constants_end_addr; + os_vm_address_t code_start_addr, code_end_addr; + os_vm_address_t code_addr = (os_vm_address_t)new_code; + os_vm_address_t old_addr = (os_vm_address_t)old_code; + os_vm_size_t displacement = code_addr - old_addr; lispobj fixups = NIL; - unsigned long displacement = - (unsigned long)new_code - (unsigned long)old_code; struct vector *fixups_vector; ncode_words = fixnum_value(new_code->code_size); @@ -1812,10 +1805,10 @@ gencgc_apply_code_fixups(struct code *old_code, struct code *new_code) /* FSHOW((stderr, "/compiled code object at %x: header words = %d, code words = %d\n", new_code, nheader_words, ncode_words)); */ - constants_start_addr = (void *)new_code + 5*N_WORD_BYTES; - constants_end_addr = (void *)new_code + nheader_words*N_WORD_BYTES; - code_start_addr = (void *)new_code + nheader_words*N_WORD_BYTES; - code_end_addr = (void *)new_code + nwords*N_WORD_BYTES; + constants_start_addr = code_addr + 5*N_WORD_BYTES; + constants_end_addr = code_addr + nheader_words*N_WORD_BYTES; + code_start_addr = code_addr + nheader_words*N_WORD_BYTES; + code_end_addr = code_addr + nwords*N_WORD_BYTES; /* FSHOW((stderr, "/const start = %x, end = %x\n", @@ -1863,24 +1856,22 @@ gencgc_apply_code_fixups(struct code *old_code, struct code *new_code) long length = fixnum_value(fixups_vector->length); long i; for (i = 0; i < length; i++) { - unsigned long offset = fixups_vector->data[i]; + long offset = fixups_vector->data[i]; /* Now check the current value of offset. */ - unsigned long old_value = - *(unsigned long *)((unsigned long)code_start_addr + offset); + os_vm_address_t old_value = *(os_vm_address_t *)(code_start_addr + offset); /* If it's within the old_code object then it must be an * absolute fixup (relative ones are not saved) */ - if ((old_value >= (unsigned long)old_code) - && (old_value < ((unsigned long)old_code - + nwords*N_WORD_BYTES))) + if ((old_value >= old_addr) + && (old_value < (old_addr + nwords*N_WORD_BYTES))) /* So add the dispacement. */ - *(unsigned long *)((unsigned long)code_start_addr + offset) = + *(os_vm_address_t *)(code_start_addr + offset) = old_value + displacement; else /* It is outside the old code object so it must be a * relative fixup (absolute fixups are not saved). So * subtract the displacement. */ - *(unsigned long *)((unsigned long)code_start_addr + offset) = + *(os_vm_address_t *)(code_start_addr + offset) = old_value - displacement; } } else { @@ -2615,15 +2606,15 @@ scavenge_newspace_generation_one_scan(generation_index_t generation) static void scavenge_newspace_generation(generation_index_t generation) { - long i; + size_t i; /* the new_areas array currently being written to by gc_alloc() */ struct new_area (*current_new_areas)[] = &new_areas_1; - long current_new_areas_index; + size_t current_new_areas_index; /* the new_areas created by the previous scavenge cycle */ struct new_area (*previous_new_areas)[] = NULL; - long previous_new_areas_index; + size_t previous_new_areas_index; /* Flush the current regions updating the tables. */ gc_alloc_update_all_page_tables(); @@ -2857,8 +2848,8 @@ print_ptr(lispobj *addr) page_index_t pi1 = find_page_index((void*)addr); if (pi1 != -1) - fprintf(stderr," %x: page %d alloc %d gen %d bytes_used %d offset %lu dont_move %d\n", - (unsigned long) addr, + fprintf(stderr," %p: page %d alloc %d gen %d bytes_used %d offset %lu dont_move %d\n", + addr, pi1, page_table[pi1].allocated, page_table[pi1].gen, @@ -3311,19 +3302,6 @@ write_protect_generation_pages(generation_index_t generation) } } -#if !defined(LISP_FEATURE_X86) && !defined(LISP_FEATURE_X86_64) -static void -scavenge_control_stack(struct thread *th) -{ - lispobj *control_stack = - (lispobj *)(th->control_stack_start); - unsigned long control_stack_size = - access_control_stack_pointer(th) - control_stack; - - scavenge(control_stack, control_stack_size); -} -#endif - #if defined(LISP_FEATURE_SB_THREAD) && (defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64)) static void preserve_context_registers (os_context_t *c) @@ -3445,7 +3423,41 @@ garbage_collect_generation(generation_index_t generation, int raise) for_each_thread(th) { void **ptr; void **esp=(void **)-1; -#ifdef LISP_FEATURE_SB_THREAD + if (th->state == STATE_DEAD) + continue; +# if defined(LISP_FEATURE_SB_SAFEPOINT) + /* Conservative collect_garbage is always invoked with a + * foreign C call or an interrupt handler on top of every + * existing thread, so the stored SP in each thread + * structure is valid, no matter which thread we are looking + * at. For threads that were running Lisp code, the pitstop + * and edge functions maintain this value within the + * interrupt or exception handler. */ + esp = os_get_csp(th); + assert_on_stack(th, esp); + + /* In addition to pointers on the stack, also preserve the + * return PC, the only value from the context that we need + * in addition to the SP. The return PC gets saved by the + * foreign call wrapper, and removed from the control stack + * into a register. */ + preserve_pointer(th->pc_around_foreign_call); + + /* And on platforms with interrupts: scavenge ctx registers. */ + + /* Disabled on Windows, because it does not have an explicit + * stack of `interrupt_contexts'. The reported CSP has been + * chosen so that the current context on the stack is + * covered by the stack scan. See also set_csp_from_context(). */ +# ifndef LISP_FEATURE_WIN32 + if (th != arch_os_get_current_thread()) { + long k = fixnum_value( + SymbolValue(FREE_INTERRUPT_CONTEXT_INDEX,th)); + while (k > 0) + preserve_context_registers(th->interrupt_contexts[--k]); + } +# endif +# elif defined(LISP_FEATURE_SB_THREAD) long i,free; if(th==arch_os_get_current_thread()) { /* Somebody is going to burn in hell for this, but casting @@ -3464,9 +3476,12 @@ garbage_collect_generation(generation_index_t generation, int raise) } } } -#else +# else esp = (void **)((void *)&raise); -#endif +# endif + if (!esp || esp == (void*) -1) + lose("garbage_collect: no SP known for thread %x (OS %x)", + th, th->os_thread); for (ptr = ((void **)th->control_stack_end)-1; ptr >= esp; ptr--) { preserve_pointer(*ptr); } @@ -3743,7 +3758,7 @@ void collect_garbage(generation_index_t last_gen) { generation_index_t gen = 0, i; - int raise; + int raise, more = 0; int gen_to_wp; /* The largest value of last_free_page seen since the time * remap_free_pages was called. */ @@ -3776,13 +3791,23 @@ collect_garbage(generation_index_t last_gen) do { /* Collect the generation. */ - if (gen >= gencgc_oldest_gen_to_gc) { - /* Never raise the oldest generation. */ + if (more || (gen >= gencgc_oldest_gen_to_gc)) { + /* Never raise the oldest generation. Never raise the extra generation + * collected due to more-flag. */ raise = 0; + more = 0; } else { raise = (gen < last_gen) || (generations[gen].num_gc >= generations[gen].number_of_gcs_before_promotion); + /* If we would not normally raise this one, but we're + * running low on space in comparison to the object-sizes + * we've been seeing, raise it and collect the next one + * too. */ + if (!raise && gen == last_gen) { + more = (2*large_allocation) >= (dynamic_space_size - bytes_allocated); + raise = more; + } } if (gencgc_verbose > 1) { @@ -3815,8 +3840,8 @@ collect_garbage(generation_index_t last_gen) gen++; } while ((gen <= gencgc_oldest_gen_to_gc) && ((gen < last_gen) - || ((gen <= gencgc_oldest_gen_to_gc) - && raise + || more + || (raise && (generations[gen].bytes_allocated > generations[gen].gc_trigger) && (generation_average_age(gen) @@ -3858,7 +3883,13 @@ collect_garbage(generation_index_t last_gen) update_dynamic_space_free_pointer(); - auto_gc_trigger = bytes_allocated + bytes_consed_between_gcs; + /* Update auto_gc_trigger. Make sure we trigger the next GC before + * running out of heap! */ + if (bytes_consed_between_gcs <= (dynamic_space_size - bytes_allocated)) + auto_gc_trigger = bytes_allocated + bytes_consed_between_gcs; + else + auto_gc_trigger = bytes_allocated + (dynamic_space_size - bytes_allocated)/2; + if(gencgc_verbose) fprintf(stderr,"Next gc when %"OS_VM_SIZE_FMT" bytes have been consed\n", auto_gc_trigger); @@ -3874,6 +3905,7 @@ collect_garbage(generation_index_t last_gen) } gc_active_p = 0; + large_allocation = 0; log_generation_stats(gc_logfile, "=== GC End ==="); SHOW("returning from collect_garbage"); @@ -4043,7 +4075,8 @@ gc_init(void) generations[i].num_gc = 0; generations[i].cum_sum_bytes_allocated = 0; /* the tune-able parameters */ - generations[i].bytes_consed_between_gc = bytes_consed_between_gcs; + generations[i].bytes_consed_between_gc + = bytes_consed_between_gcs/(os_vm_size_t)HIGHEST_NORMAL_GENERATION; generations[i].number_of_gcs_before_promotion = 1; generations[i].minimum_age_before_gc = 0.75; } @@ -4131,6 +4164,7 @@ general_alloc_internal(long nbytes, int page_type_flag, struct alloc_region *reg #endif void *new_obj; void *new_free_pointer; + os_vm_size_t trigger_bytes = 0; gc_assert(nbytes>0); @@ -4141,6 +4175,9 @@ general_alloc_internal(long nbytes, int page_type_flag, struct alloc_region *reg /* Must be inside a PA section. */ gc_assert(get_pseudo_atomic_atomic(thread)); + if (nbytes > large_allocation) + large_allocation = nbytes; + /* maybe we can do this quickly ... */ new_free_pointer = region->free_pointer + nbytes; if (new_free_pointer <= region->end_addr) { @@ -4149,10 +4186,19 @@ general_alloc_internal(long nbytes, int page_type_flag, struct alloc_region *reg return(new_obj); /* yup */ } + /* We don't want to count nbytes against auto_gc_trigger unless we + * have to: it speeds up the tenuring of objects and slows down + * allocation. However, unless we do so when allocating _very_ + * large objects we are in danger of exhausting the heap without + * running sufficient GCs. + */ + if (nbytes >= bytes_consed_between_gcs) + trigger_bytes = nbytes; + /* we have to go the long way around, it seems. Check whether we * should GC in the near future */ - if (auto_gc_trigger && bytes_allocated > auto_gc_trigger) { + if (auto_gc_trigger && (bytes_allocated+trigger_bytes > auto_gc_trigger)) { /* Don't flood the system with interrupts if the need to gc is * already noted. This can happen for example when SUB-GC * allocates or after a gc triggered in a WITHOUT-GCING. */ @@ -4161,6 +4207,9 @@ general_alloc_internal(long nbytes, int page_type_flag, struct alloc_region *reg * section */ SetSymbolValue(GC_PENDING,T,thread); if (SymbolValue(GC_INHIBIT,thread) == NIL) { +#ifdef LISP_FEATURE_SB_SAFEPOINT + thread_register_gc_trigger(); +#else set_pseudo_atomic_interrupted(thread); #ifdef LISP_FEATURE_PPC /* PPC calls alloc() from a trap or from pa_alloc(), @@ -4174,12 +4223,14 @@ general_alloc_internal(long nbytes, int page_type_flag, struct alloc_region *reg #else maybe_save_gc_mask_and_block_deferrables(NULL); #endif +#endif } } } new_obj = gc_alloc_with_region(nbytes, page_type_flag, region, 0); #ifndef LISP_FEATURE_WIN32 + /* for sb-prof, and not supported on Windows yet */ alloc_signal = SymbolValue(ALLOC_SIGNAL,thread); if ((alloc_signal & FIXNUM_TAG_MASK) == 0) { if ((signed long) alloc_signal <= 0) { @@ -4242,7 +4293,17 @@ void unhandled_sigmemoryfault(void* addr); * * Return true if this signal is a normal generational GC thing that * we were able to handle, or false if it was abnormal and control - * should fall through to the general SIGSEGV/SIGBUS/whatever logic. */ + * should fall through to the general SIGSEGV/SIGBUS/whatever logic. + * + * We have two control flags for this: one causes us to ignore faults + * on unprotected pages completely, and the second complains to stderr + * but allows us to continue without losing. + */ +extern boolean ignore_memoryfaults_on_unprotected_pages; +boolean ignore_memoryfaults_on_unprotected_pages = 0; + +extern boolean continue_after_memoryfault_on_unprotected_pages; +boolean continue_after_memoryfault_on_unprotected_pages = 0; int gencgc_handle_wp_violation(void* fault_addr) @@ -4273,17 +4334,39 @@ gencgc_handle_wp_violation(void* fault_addr) os_protect(page_address(page_index), GENCGC_CARD_BYTES, OS_VM_PROT_ALL); page_table[page_index].write_protected_cleared = 1; page_table[page_index].write_protected = 0; - } else { + } else if (!ignore_memoryfaults_on_unprotected_pages) { /* The only acceptable reason for this signal on a heap * access is that GENCGC write-protected the page. * However, if two CPUs hit a wp page near-simultaneously, * we had better not have the second one lose here if it * does this test after the first one has already set wp=0 */ - if(page_table[page_index].write_protected_cleared != 1) - lose("fault in heap page %d not marked as write-protected\nboxed_region.first_page: %d, boxed_region.last_page %d\n", - page_index, boxed_region.first_page, - boxed_region.last_page); + if(page_table[page_index].write_protected_cleared != 1) { + void lisp_backtrace(int frames); + lisp_backtrace(10); + fprintf(stderr, + "Fault @ %p, page %"PAGE_INDEX_FMT" not marked as write-protected:\n" + " boxed_region.first_page: %"PAGE_INDEX_FMT"," + " boxed_region.last_page %"PAGE_INDEX_FMT"\n" + " page.region_start_offset: %"OS_VM_SIZE_FMT"\n" + " page.bytes_used: %"PAGE_BYTES_FMT"\n" + " page.allocated: %d\n" + " page.write_protected: %d\n" + " page.write_protected_cleared: %d\n" + " page.generation: %d\n", + fault_addr, + page_index, + boxed_region.first_page, + boxed_region.last_page, + page_table[page_index].region_start_offset, + page_table[page_index].bytes_used, + page_table[page_index].allocated, + page_table[page_index].write_protected, + page_table[page_index].write_protected_cleared, + page_table[page_index].gen); + if (!continue_after_memoryfault_on_unprotected_pages) + lose("Feh.\n"); + } } ret = thread_mutex_unlock(&free_pages_lock); gc_assert(ret == 0);