4 day week.

[VIRT-327 # Richard's upstream QEMU work ]

Bug hunting vs aa32 ldrex/strex. I had hoped it would be relatively easy to reproduce -- just run something from the .NET testsuite -- but even getting that far wasn't obvious. So I put that aside; let's see if Peter's request for an actual reproducer gets results.

Bug hunting vs aa64 gcc sync-4.c as reported by clyon. I determined that it's not the fault of the null-pointer dereference, and that something goes wrong somewhere in libgcc's exception unwind prior to the c++ throw. But it doesn't fail all of the time. And worse, the problem vanishes when randomize_va_space is disabled. So I can neither get a "good" vs "bad" trace without needless differences nor produce a failure under gdb. I should try again with rr and see if that works...

r~