On Wed, 17 May 2000 17:08:51 +0900, suzukis@xxxxxxxxxxxxxxxxxxxxxx wrote:
>Sorry for a bit too late report.
>
>Now drivers/block/ll_rw_blk.c of linux-2.3-xfs has
>kiobuf-based I/O request funcstions:
> req_new_io(),
> req_finished_io().
>
>It seems that SCSI driver calls these functions,
>but ll_rw_blk.c does not export them.
>Thus, when I modularize SCSI driver, scsi_mod.o
>includes unresolvable symbol: req_finished_io.
>
>If it's not harmful, I wish if additional EXPORT
>of symbols is added to the end of ll_rw_brk.c,
>aslike:
>
>#if CONFIG_SCSI == m
>EXPORT_SYMBOL(req_new_io);
>EXPORT_SYMBOL(req_finished_io);
>#endif
>
>Best wishes,
>
>suzuki
How does this integrate with the path thats already there for
doing kiobuf-based I/O requests for scsi devices? See the code
for ll_rw_kio() and __make_kio_request() thats in ll_rw_blk.c.
>From what I understand of the code, the new functions you're
talking about (req_new_io and req_finished_io) enhance the elevator
timing and disk utilization statistics (general I/O accounting).
They don't actually enable a kiobuf-based request to be passed to
the scsi mid-layers (i.e. fucntions scsi_lib.c and scsi_merge.c). Am
I right?
Thanks,
-Chait.
|