Messages in this thread | | | Date | Thu, 11 Jan 2001 18:04:43 +0100 | From | Jens Axboe <> | Subject | Re: KIOBUFS ?? |
| |
On Wed, Jan 10 2001, Andre Hedrick wrote: > > LT, > > Will this maddness insure that the granularity of the request will be > dependent to the k_dev_t? Specifically, can one make KIOBUFS do the > sizing of buffer to match the ideal or specified size limits imposed by a > given block device? Otherwise I will need to design an sub-request layer > to reduce the pain of restarting the entire request because of the huge > DMA-PRD-Chain that has no clue how to report error location and allow a > restart from NxPRD's before the error.
Take a look at the XFS tree. I wrote IDE kiobuf support for that, and it simply doesn't take the request the queue before it has been completed (which may take many start-ups for a huge request). But unless you can make the prd table bigger (which doesn't make much sense anyway), I don't see any harm in setting up sg for each iteration.
-- * Jens Axboe <axboe@suse.de> * SuSE Labs - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
| |