Changeset 14754

Timestamp:

Apr 29, 2011, 11:23:52 AM (14 years ago)

Author:

Gary Byers

Message:

In pc_luser_xp(), recognize that the alloc_trap is now immediately
preceded by a branch around it; the COMPARE_ALLOCPTR_TO_RM that we're
looking for precedes the branch. Note that the LinuxARM heap image
in the trunk currently does things this way; older images may not.

This seems to fix ticket:855 (the test case from ticket:717 runs
to completion, and other things that cons hysterically while creating
threads for no apparent reason - like SLIME - will hopefully not
be affected by this.)

File:

• trunk/source/lisp-kernel/arm-exceptions.c (modified) (1 diff)

trunk/source/lisp-kernel/arm-exceptions.c

@@ -1658 +1658 @@
         */
       } else {
+        Boolean ok = false;
         update_bytes_allocated(tcr, (void *) ptr_from_lispobj(cur_allocptr + disp));
         xpGPR(xp, allocptr) = VOID_ALLOCPTR + disp;
         instr = program_counter[-1];
-        if (IS_COMPARE_ALLOCPTR_TO_RM(instr)){
-          xpGPR(xp,RM_field(instr)) = VOID_ALLOCPTR;
+        if (IS_BRANCH_AROUND_ALLOC_TRAP(instr)) {
+          instr = program_counter[-2];
+          if (IS_COMPARE_ALLOCPTR_TO_RM(instr)){
+            xpGPR(xp,RM_field(instr)) = VOID_ALLOCPTR;
+            ok = true;
+          }
+        }
+        if (ok) {
+        /* Clear the carry bit, so that the trap will be taken. */
+        xpPSR(xp) &= ~PSR_C_MASK;
         } else {
           Bug(NULL, "unexpected instruction preceding alloc trap.");
         }
-        /* Clear the carry bit, so that the trap will be taken. */
-        xpPSR(xp) &= ~PSR_C_MASK;
       }
     } else {