On 11/16/05, Jens Axboe <axboe@xxxxxxx> wrote: > On Wed, Nov 16 2005, Bartlomiej Zolnierkiewicz wrote: > > > > which of course didn't work, so it was changed to the above which then > > > > broke the assumption of what type of requests we expect to see in > > > > ide_softirq_done(). We can't generically handle this case, so it's > > > > probably best to just add this logic to __ide_end_request() - it's just > > > > another case for _not_ using the blk_complete_request() path, just like > > > > the partial case. > > > > > > Sounds better but I honestly think that you simply cannot obtain > > > reliable nr_sectors to complete for FS/PC requests just from the > > > request type. Two examples are: failed disk flush requests and > > > cd noretry requests (both are of FS type). > > > > first example is bad :-) > > Both your examples are wrong - a flush request is non-fs/pc, and noretry > requests doesn't impact the type of the request at from this POV. I meant doing partial completions of the real request from the ->end_flush() and noretry doesn't impact the type but it does impact the nr_sectors to complete. However none of these matters any longer with your latest patch which simplifies things a lot. > > > IMO the best way to fix it is to actually move more (not less!) of > > > the logic from driver->end_request() paths to ide_softirq_done(). > > > > Your latest patch is also a good way to fix it > > (now the only thing left is rq->errors/rq->retries discussed earlier). > > Yeah, that is still pending... I updated the patch series so it's a > clean 1-2-3-4 step series now, there instead of this little additions. Excellent! Bartlomiej - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
