suzanne.soy/racket
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Chez Scheme GC: repair for locked object in space_new

Browse Source
This commit is contained in:
Matthew Flatt 2020-10-02 16:47:53 -06:00
parent cb50bab726
commit 629153c4e0
1 changed files with 23 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
racket/src/ChezScheme/c/gc.c
View File

@ -1213,6 +1213,9 @@ ptr GCENTRY(ptr tc, ptr count_roots_ls) {
/* coordinate with alloc.c */
(SYMVAL(sym) != sunbound || SYMPLIST(sym) != Snil || SYMSPLIST(sym) != Snil)) {
seginfo *sym_si = SegInfo(ptr_get_segment(sym));
#ifdef ENABLE_PARALLEL
thread_gc *tgc = sym_si->creator; /* shadows enclosing `tgc` binding */
#endif
if (!new_marked(sym_si, sym))
mark_or_copy_pure(&sym, sym, sym_si);
}
@ -1980,6 +1983,12 @@ static iptr sweep_generation_pass(thread_gc *tgc) {
relocate_impure_help(ppn, p, from_g, pp, 2 * ptr_bytes);
pp = ppn + 1;
}
} else if (s == space_new) { /* used by infer_space to mean "type ojbect" */
while (pp < nl) {
p = TYPE(TO_PTR(pp), type_typed_object);
sweep(tgc, p, from_g);
pp = TO_VOIDP((uptr)TO_PTR(pp) + size_object(p));
}
} else if (s == space_closure) {
while (pp < nl) {
p = TYPE(TO_PTR(pp), type_closure);
@ -2038,6 +2047,7 @@ static iptr sweep_generation_pass(thread_gc *tgc) {
size_record_inst(UNFIX(RECORDDESCSIZE(RECORDINSTTYPE(p)))));
}
} else {
printf(">> %d\n", s);
S_error_abort("dirty range sweep: unexpected space");
}
FLUSH_REMOTE_RANGE(tgc, s, from_g);
@ -2077,21 +2087,24 @@ void enlarge_sweep_stack(thread_gc *tgc) {
#ifdef ENABLE_PARALLEL
static ISPC infer_space(ptr p, seginfo *si) {
/* Certain kinds of values get allocated to more specific spaces by
parallel mode compared to non-parallel mode. Marking objects from
a previous collection can mean sweeping from the less-specific
space, however. We can synthesize an appropropriate space here,
since it will be used only by the handling of received ranges. */
parallel mode compared to non-parallel mode, and locked objects
may stay in a less-specific space like `space_new`. Marking
objects from a previous collection can mean sweeping from the
less-specific space. We can synthesize an appropropriate space
here, since it will be used only by the handling of received
ranges. */
if (si->marked_mask) {
ITYPE t = TYPEBITS(p);
if (t == type_typed_object) {
ptr tf = TYPEFIELD(p);
if ((iptr)tf == type_ratnum)
return space_pure;
else if ((iptr)tf == type_exactnum)
return space_pure;
if (t == type_typed_object)
return space_new; /* means "typed object" to sweep_generation_pass */
else if (t == type_pair) {
if (si->space == space_new)
return space_impure;
} else if (t == type_closure)
return space_closure;
else if (t == type_symbol)
return space_symbol;
}
return si->space;