>Steve Harris wrote: > >>On Tue, Dec 10, 2002 at 02:03:42 +0100, Tim Goetze wrote: >>> >Not if it generates machine code. That way, it could (theoretically) >>> >>> in realtime? we'll need everybodies' spare cpu cycles! >> >>No, it really isn't that slow. >> >>SyncModular does a similar trick. > >it'd still be interesting to know how the sync problems this >method poses are solved: you cannot rely on executable code >modifications to be atomic. an indirect jump instruction is >not guaranteed to work ok: a pointer on x86 is 32 bits, and >atomic is 24 bits
no, atomic is 32 bits on x86. its only 24 bits on sparc, where the need to provide a spinlock to cover cache write-back effects forces it to 24 bits. you can do atomic exchange and compare-and-swap on pointers for x86. --p
