On Wed, 2013-12-04 at 09:15 -0800, Linus Torvalds wrote: > On Wed, Dec 4, 2013 at 8:54 AM, H. Peter Anvin <h...@zytor.com> wrote: > > > > That is why I talk about the atomic instruction word... most (but not > > *all*) architectures have a fundamental minimum unit of instructions > > which is aligned and can be atomically written. Typically this is 1, 2, > > or 4 bytes. > > Note that it's not just about the "atomically written", it's also > about the guarantee that it's atomically *read*. > > x86 can certainly atomically write a 4-byte instruction too, it's just > that there's no guarantee - even if the instruction is aligned etc - > that the actual instruction decoding always ends up reading it that > way. It might re-read an instruction after encountering a prefix byte > etc etc. So even if it's all properly aligned, the reading side might > do something odd.
The ARM architecture has similar issues. Even though the instruction size is mostly fixed, the architecture specification itself only guarantees a very tiny subset of instructions are safe to modify whilst there may be concurrent execution of that instruction. I'm quoting a discussion from a while ago: http://lkml.org/lkml/2012/12/10/346 -- Tixy -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
