_IO_IS_ISA question

[linas at austin.ibm.com]

On Fri, Mar 26, 2004 at 08:25:17AM +1100, Paul Mackerras wrote:
> > Hmmm. include/asm-ppc/io.h   ... twi ... isync ...
> >
> > I hope that we can do something that doesn't require either twi or isync,
> > or sync for that matter, as the cure seems worse than the disease.
>
> The twi/isync sequence has nothing to do with exceptions.  It is there
> to ensure that subsequent instructions are not executed until the data
> has come back from the device.  It would be nice not to have to do
> that, but there are quite a few drivers in the kernel that assume that
> I/O operations are synchronous, and we avoid a lot of subtle bugs by
> doing this.

Dang, I could have sworn that the ppc64 macros didn't do that, but now
that I really did look at them, I see that indeed they do, so objections
withdrawn & I am feeling slightly foolish.

Moving off-topic: anyone ever experiment with removing these isyncs
from the ppc64 code, and seeing if at least our most common
adapters/controllers still work?  Are these broken drivers for things
that no one would ever plug into ppc64?

--linas


** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/