Re: [2.6.6-mm4-ff1] I/O context isolation

[Jens Axboe]

On Fri, May 21 2004, FabF wrote:
> On Fri, 2004-05-21 at 09:38, Nick Piggin wrote:
> > FabF wrote:
> > > Jens,
> > > 
> > > 	Here's ff1 patchset to have generic I/O context.
> > > ff1 : Export io context operations from blkdev/ll_rw_blk (ok)
> > > ff2 : Make io_context generic plateform by importing IO stuff from
> > > as_io.
> > > 
> > 
> > Can I just ask why you want as_io_context in generic code?
> > It is currently nicely hidden away in as-iosched.c where
> > nobody else needs to ever see it.
> I do want I/O context to be generic not the whole as_io.
> That export should bring:
> 	-All elevators to use io_context

For?

> 	-source tree to be more self-explanatory
> 	-have a stronger elevator interface

Those aren't real arguments :-)

> > > 	AFAICS, cfq_queue for instance could disappear when using io_context
> > > but I think elv_data should remain elevator side....
> > > I don't want to go in the wild here so if you've got suggestions, don't
> > > hesitate ;)
> > > 
> > 
> > cfq_queue is per-queue-per-process. io_context is just
> > per-process, so it isn't a trivial replacement.
> But I guess we can merge that stuff to have "a one way, one place" code
> rather than repeat code in all elv.

That's another pit fall, don't design for a need that isn't there. Using
io contexts in cfq does not get any easier with the code moved to
iocontext.c file, it's fine where it is.

-- 
Jens Axboe

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/