Changeset 5273

Timestamp:

Sep 27, 2006, 4:39:32 AM (18 years ago)

Author:

Gary Byers

Message:

Still get very occasional SIGFPE on return from (e.g.) #_asin on
x86-64 Darwin. (#_asin often does a division-by-zero, but that should
be masked when in foreign code.) Don't reload the lisp mxcsr until
other registers have been popped, to make debugging easier in case
it's the act of reloading the mxcsr that causes the spurious trap.)

File:

• trunk/ccl/lisp-kernel/x86-spentry64.s (modified) (2 diffs)

trunk/ccl/lisp-kernel/x86-spentry64.s

@@ -4064 +4064 @@
         __(movq $TCR_STATE_LISP,%rcontext:tcr.valence)
         __(movq %rcontext:tcr.save_vsp,%rsp)
-        __(ldmxcsr %rcontext:tcr.lisp_mxcsr)
         __(pop %rbp)
         __(pop %ra0)
@@ -4077 +4076 @@
         __(pop %temp2)
         __(pop %temp1)
+        __(ldmxcsr %rcontext:tcr.lisp_mxcsr)
         __(check_pending_interrupt(%temp0))
         __(pop %temp0)