On Wed, Jan 29 2003, Benjamin Herrenschmidt wrote:
> On Tue, 2003-01-28 at 23:06, Jens Axboe wrote:
> > On Tue, Jan 28 2003, Larry Sendlosky wrote:
> > > I was glad to see the physical page support in 2.4.20.
> > > (and also noticed that the current BK tree clobbered it
> > > on a patch set from Alan).
> > >
> > > One question,
> > >
> > > + lastdataend = bh_phys(bh) + bh->b_size;
> > >
> > > bh_phys(x) uses bh->b_page. Does it make a difference
> > > if bh->b_page is zero? What if someone combines virt and phys
> > > buffer addresses in bh list?
> >
> > Yes good catch! New version attached.
>
> That's interesting. I wasn't awaye you could have a request
> containing such a "mixed" set of bh without valid pages.
> Actually, I though b_page was always valid. Looking at
> other drivers (typically the the csiss.c driver), it also
> unconditionally use b_page & bh_phys(). So either we are
> looking at a false problem, or that driver need fixing as
> well.
b_page is not always valid for IDE, this is a special case. ide-scsi
fabricates its own buffer_heads. cciss etc can rely on valid b_page
always.
> Now assuming that mix can happen, I don't like the fact that
> your new version will use lastdataend to compare against both
> physical and virtual addresses.
They should not be mixed in one call of build_sglist().
> Maybe the solution is to have an additional variable indicating
> if the last bh was virtual or physical, and reset lastdataend
> to ~0 when the current one is different...
That should be a BUG(), if anything.
-- Jens Axboe- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Jan 31 2003 - 22:00:21 EST