/* Part of SWI-Prolog Author: Jan Wielemaker E-mail: J.Wielemaker@vu.nl WWW: http://www.swi-prolog.org Copyright (c) 2007-2020, University of Amsterdam VU University Amsterdam CWI, Amsterdam All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /*#define O_DEBUG 1*/ #include "pl-incl.h" /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Segmented stack handling. A segmented stack is a stack that is allocated in segments, This means we cannot compare addresses otherwise then by identity. We use a segmented stack for cycle detection. Measurements with the chunksize on SuSE Linux 10.2 indicate there is no measurable performance change above approximately 256 bytes. We'll keep the figure on the safe side for systems with less efficient malloc implementations. Note that atom-gc requires completely asynchronous calling of scanSegStack() and therefore pushSegStack()/popSegStack() must push the data before updating the pointers. TBD: Avoid instruction/cache write reordering in push/pop. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */ int pushSegStack_(segstack *stack, void *data) { if ( stack->top + stack->unit_size <= stack->max ) { memcpy(stack->top, data, stack->unit_size); stack->top += stack->unit_size; return TRUE; } else { size_t chunksize = tmp_nalloc((size_t)1024<chunk_count++); segchunk *chunk = tmp_malloc(chunksize); if ( !chunk ) return FALSE; /* out of memory */ chunk->allocated = TRUE; chunk->size = chunksize; chunk->next = NULL; chunk->previous = stack->last; chunk->top = CHUNK_DATA(chunk); /* async scanning */ if ( stack->last ) { stack->last->next = chunk; stack->last->top = stack->top; stack->top = chunk->top; /* async scanning */ stack->last = chunk; } else { stack->top = chunk->top; /* async scanning */ stack->last = stack->first = chunk; } stack->base = CHUNK_DATA(chunk); stack->max = addPointer(chunk, chunk->size); memcpy(CHUNK_DATA(chunk), data, stack->unit_size); stack->top = CHUNK_DATA(chunk) + stack->unit_size; return TRUE; } } int pushRecordSegStack(segstack *stack, Record r) { if ( stack->top + sizeof(r) <= stack->max ) { Record *rp = (Record*)stack->top; *rp++ = r; stack->top = (char*)rp; return TRUE; } else { return pushSegStack_(stack, &r); } } /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Pop data. Note that we leave the first chunk associated with the stack to speedup frequent small usage. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */ int popSegStack_(segstack *stack, void *data) { again: if ( stack->top >= stack->base + stack->unit_size ) { stack->top -= stack->unit_size; memcpy(data, stack->top, stack->unit_size); return TRUE; } else { segchunk *chunk = stack->last; if ( chunk ) { if ( chunk->previous ) { stack->last = chunk->previous; stack->last->next = NULL; if ( chunk->allocated ) tmp_free(chunk); chunk = stack->last; stack->base = CHUNK_DATA(chunk); stack->max = addPointer(chunk, chunk->size); stack->top = chunk->top; stack->chunk_count--; goto again; } } return FALSE; } } void * topOfSegStack(segstack *stack) { segchunk *chunk; if ( stack->top >= stack->base + stack->unit_size ) { return stack->top - stack->unit_size; } else if ( stack->last && (chunk=stack->last->previous) ) { assert(chunk->top - stack->unit_size >= CHUNK_DATA(chunk)); return chunk->top - stack->unit_size; } return NULL; } void popTopOfSegStack_(segstack *stack) { again: if ( stack->top >= stack->base + stack->unit_size ) { stack->top -= stack->unit_size; } else { segchunk *chunk = stack->last; if ( chunk ) { if ( chunk->previous ) { segchunk *del = chunk; stack->chunk_count--; stack->last = chunk->previous; stack->last->next = NULL; chunk = stack->last; stack->top = chunk->top; MEMORY_BARRIER(); /* Sync with scanSegStack() */ stack->base = CHUNK_DATA(chunk); stack->max = addPointer(chunk, chunk->size); if ( del->allocated ) tmp_free(del); goto again; } } assert(0); } } /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - scanSegStack(segstack *stack, void (*func)(void *cell)) Walk along all living cells on the stack and call func on them. The stack is traversed last-to-first. This is used by markAtomsFindall(), which runs concurrent with findall/3 in the thread being scanned. New records pushed while AGC is in progress are marked by findall/3 itself. Findall/3 can concurrently pop this stack. It may do this quicker than the marking, but this is no problem because the pointers remain valid as all records are reclaimed at the end of findall/3 by calling clear_mem_pool(). It is only a problem if popTopOfSegStack_() pops a chunk because the popped chunk is freed. Therefore $collect_findall_bag/2 locks if it needs to call the expensive chunk-popping popTopOfSegStack_(). With thanks to Eugeniy Meshcheryakov for finding this and running many tests. The issue is triggered by Tests/thread_agc_findall.pl, notably on slow single core hardware. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */ static inline void scan_chunk(segstack *stack, char *top, char *base, void (*func)(void *cell)) { while(top >= base+stack->unit_size) { top -= stack->unit_size; (*func)((void*)top); } } void scanSegStack(segstack *stack, void (*func)(void *cell)) { segchunk *chunk; if ( (chunk=stack->last) ) /* something there */ { if ( stack->base == CHUNK_DATA(chunk) ) { char *top = stack->top; if ( !(top > (char*)chunk && top <= (char*)((intptr_t)chunk + chunk->size)) ) top = chunk->top; /* top has moved to new chunk */ scan_chunk(stack, top, CHUNK_DATA(chunk), func); chunk = chunk->previous; } for(; chunk; chunk=chunk->previous) scan_chunk(stack, chunk->top, CHUNK_DATA(chunk), func); } } void clearSegStack_(segstack *s) { segchunk *c = s->first; segchunk *n; if ( !c->allocated ) /* statically allocated first chunk */ { n = c->next; c->next = NULL; s->last = c; s->base = s->top = c->top; s->max = addPointer(c, c->size); s->chunk_count = 1; for(c=n; c; c = n) { n = c->next; tmp_free(c); } } else /* all dynamic chunks */ { for(; c; c = n) { n = c->next; tmp_free(c); } memset(s, 0, sizeof(*s)); } }