On Tue, 2006-03-28 at 17:13 -0500, Alan Cox wrote: > On Tue, Mar 28, 2006 at 05:04:40PM -0500, Jeff Spaleta wrote: > > Or to ask it another way... if you use ide=nodma and mediacheck > > succeeds when before it failed.. does that mean that you should then > > use ide=nodma when doing the actual install on that system? > > Nope. The problem is a CD-R track looks something like this: > > <track> > [ISO fs] > [0-150K or so of undefined] > <end of track> > > and the catalogue (track listing) contains various bits of magic, your drive > serial number and start/end of each track. But that is imprecise. mediacheck reads isofs, not just the table of contents. At no time does it perform a read(2), nor any other operation, which would result in an access of any block after the end of the isofs. > On the other hand the mounted isofs image has a known size so the isofs itself > won't try and read or prefetch outside it. I don't think that adequately explains what's going on here. In mediacheck, we never try to read the lead-out, much less the possibly non-de-iced part of the disk. At all. Also, we see the same errors on pressed CDs and DVDs, which IIRC do not have the less well defined tracks that CD-Rs do. The block layer (or below) does the bogus prefetch. What inhibits this behavior when you're accessing through a filesystem driver? I think we have exactly the same problem there, and people have witnessed accesses through the FS fail near the end of the disk with our install media. To me, it looks like the same behavior. -- Peter -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
