Hi, >> I don't see any major disadvantages for fix 3.
On second thought... The frame info recorded needs to be cacheable by %rip, and will be reused purely by %rip regardless of stack layout. If we are recording the real sigcontext_t address, will that still work? The original format hoped to be cacheable by recording only a delta from the current CFA so it's not dependent on actual stack location - not so good an idea with alt stacks as you noticed. I might try experiment a bit with a few of our apps being profiled. Regards, Lassi _______________________________________________ Libunwind-devel mailing list [email protected] https://lists.nongnu.org/mailman/listinfo/libunwind-devel
