Re: elevator priorities vs. full request queues

Jens Axboe wrote:

	switch (bio_data_dir(bio)) {

Ah yes, much nicer.

Maybe it would be more logical to assign 'default' priority in
bio_init(), in fact?

What made me shy away from this is that bio_init and bio_alloc
seem to get called from contexts where "current" wouldn't
necessarily have anything to do with the process that has
actually initiated the IO.

For example, I'm not sure if the calls from ide-scsi.c, when
it builds requests piece by piece, are safe in this regard.

The callers of submit_bio all seem to be more of the
"classical" variety.

- Werner

-- 
  _________________________________________________________________________
 / Werner Almesberger, Buenos Aires, Argentina         wa@almesberger.net /
/_http://www.almesberger.net/____________________________________________/

On Tue, Jul 13 2004, Werner Almesberger wrote:

quoted

Maybe it would be more logical to assign 'default' priority in
bio_init(), in fact?

What made me shy away from this is that bio_init and bio_alloc
seem to get called from contexts where "current" wouldn't
necessarily have anything to do with the process that has
actually initiated the IO.

For example, I'm not sure if the calls from ide-scsi.c, when
it builds requests piece by piece, are safe in this regard.

That's not really what I meant - 'default' priority isn't tied to the
process. It's just that if default io prio was something other than 0,
then it would make sense to set it there. Say if the range was 0-15, and
you wanted 7 to be 'in the middle' default priority.

But I suppose you just want io submission to set it, in which case it
doesn't matter what bio_init() sets it to.

-- 
Jens Axboe