It''s really old code to let us keep a handle on the last request_queue
we
stuffed a request onto. This is because we plug that queue until we have
finished pulling down a batch of blkfront work, then unplug it.
Is this the best way to do that? Is there a benefit to doing this in the
first place on 2.6 kernels? To be honest I do not know. We''d need to
ask
someone like Jens Axboe, and/or just kill that code and measure resulting
disc throughput performance. My guess is that it''s merely legacy cruft
from
the 2.4 era of blkback (also I wrote that code back in 2004 and I remember
not having much of a clue at the time about Linux 2.4 block interfaces!).
-- Keir
On 6/3/08 10:49, "Jan Beulich" <jbeulich@novell.com> wrote:
> While these symbols are no longer available to modules in 2.6.25-rc4, it
> seems suspicious that blkback is the only user of them (outside of the
> block/ directory) even in 2.6.18 (and starting with 2.6.17, where all the
> other blk_put_queue() users were converted to blk_cleanup_queue(),
> which isn''t what blkback wants; there were no other users of
> blk_get_queue() even in 2.6.5 - didn''t check older code). Would it
be
> possible for someone familiar with the block layer and blkback''s
> interface to it to determine whether these calls aren''t really
left-overs?
>
> Thanks, Jan
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel