still memset puts a core on glibc cvs

[Steve Munroe]

Kaoru Fukui writes:

> I did step trace.then memset was almost OK.
> But when start ,I got sigsegv always.

> I don't know who set r8 with 0x38d00.
> they were (r3 = 0x38cf4,r8 = 0x38d00,r6 = 0x38d00) at sigsegv.

The code looks correct but the value r3 (0x38cf4) looks to be outside the
normal range for a process. A process map (cat /proc/<pid#>/maps) may be
useful.

The instruction that fails is still in the process of getting aligned to a
32-byte boundary (which is where the value in r8 and r6 (0x38d00) is
derived from. I would like to see all the gprs and a dump of the stack.

At the moment it looks like memset was passed a bad address ...