On Wed, Nov 16 2005, Bartlomiej Zolnierkiewicz wrote: > On 11/16/05, Jens Axboe <axboe@xxxxxxx> wrote: > > On Wed, Nov 16 2005, Bartlomiej Zolnierkiewicz wrote: > > > On 11/16/05, Jens Axboe <axboe@xxxxxxx> wrote: > > > > > > > I updated that patch, and converted IDE and SCSI to use it. See the > > > > results here: > > > > > > > > http://brick.kernel.dk/git/?p=linux-2.6-block.git;a=shortlog;h=blk-softirq > > > > > > I like it but: > > > > > > * "we know it's either an FS or PC request" assumption in > > > ide_softirq_done() is really wrong > > > > It used to be correct :-) > > Sorry but it has been always like that, > other requests also pass through ide_end_request() > (which of course needs fixing). You misunderstand, for calls to blk_complete_request() it wasn't true initially since it always obyed rq_all_done() (which returns 0 for non-fs and non-pc requests). > > Irk it's nasty, since it basically means we have to hold ide_lock over > > the entire functions looking at hwgroup->rq. > > > > It's ok for __ide_end_request() to be entered with the ide_lock held, > > the costly affair is usually completing the request. Which now happens > > outside of the lock. > > We should get rid of ide_preempt later. > > This will also allow us to remove ide_do_drive_cmd() > and use blk_execute_rq() exclusively. That would be very nice! Reminds me that there might still be a race with head insertion and REQ_STARTED request in front in the block layer, that needs inspection. But killing ide_do_drive_cmd() would be very nice. -- Jens Axboe - : 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
