>Hmmm.
>
>Can a signal be raised between the time a [34]86 SX processor writes
>the first and second halves of a 32 bit int?  If so, sig_atomic_t
>would have to be a "short".
>
>       --jtc

Take the situation where a short exists 2 bytes before a page boundary.

Write the first half of the long

Get page fault to enter VM code

Get some kind of signal from VM code

Question is.. What happens. Do you only restart the write instruction if you 
don't
raise the signal, or do you raise the signal, then continue?

 -- Alex