lispobj
alloc_code_object (unsigned boxed, unsigned unboxed) {
struct code * code;
- /* 4 == trace_table_offset offset in words */
- boxed = make_fixnum(boxed + 1 + 4);
+ /* Coming in, boxed is the number of boxed words requested.
+ * Converting it to a fixnum makes it measured in bytes. It's also
+ * rounded up to double word along the way. */
+ boxed = make_fixnum(boxed + 1 +
+ (offsetof(struct code, trace_table_offset) >>
+ WORD_SHIFT));
boxed &= ~LOWTAG_MASK;
+ /* Unboxed is in bytes, round it up to double word boundary. Now
+ * it's also a fixnum containing the number of unboxed words. */
unboxed += LOWTAG_MASK;
unboxed &= ~LOWTAG_MASK;
- code = (struct code *)pa_alloc(ALIGNED_SIZE((boxed + unboxed) *
- sizeof(lispobj)),
- CODE_PAGE_FLAG);
+ code = (struct code *)pa_alloc(boxed + unboxed, CODE_PAGE_FLAG);
+ /* It used to be that even on gencgc builds the
+ * ALLOCATE-CODE-OBJECT VOP did all this initialization within
+ * pseudo atomic. Here, we rely on gc being inhibited. */
+ if (SymbolValue(GC_INHIBIT, arch_os_get_current_thread()) == NIL)
+ lose("alloc_code_object called with GC enabled.");
boxed = boxed << (N_WIDETAG_BITS - WORD_SHIFT);
code->header = boxed | CODE_HEADER_WIDETAG;
code->code_size = unboxed;
;;; checkins which aren't released. (And occasionally for internal
;;; versions, especially for internal versions off the main CVS
;;; branch, it gets hairier, e.g. "0.pre7.14.flaky4.13".)
-"1.0.26.1"
+"1.0.26.2"