Re: 2.4.7-pre9..

[Jens Axboe]

On Fri, Jul 20 2001, Linus Torvalds wrote:
> 
> 
> On Fri, 20 Jul 2001, Jens Axboe wrote:
> >
> > > The paging stuff doesn't use any of this. The paging stuff use the page
> > > cache lock bit, and always has.
> >
> > Paging still hits a request, I assumed that's what the (really really)
> > old comment meant to say.
> 
> No. Tha paging stuff (and _all_ regular IO) uses a regular request, with a
> NULL waiter. That's the normal path. That normal path depends on the
> buffer heads associated with the requests and their "bh->b_end_io()"
> function marking other state up-to-date, and then waits on the page being
> locked.

This is perfectly clear. I'm saying 'paging uses a request just like any
other I/O', and you seem to disagree and restate the same thing?! In
fact the lower layers have no way of knowing what is what, paging or
not.

> The req->sem (and now req->completion) thing is a very different thing: it
> is a "we are waiting for this particular request", and is used when it's
> not really IO and doesn't have a bh, but something that has side effects.
> Like an ioctl that causes a special command to the disk to change some
> parameters, or query the size of the disk or something.

Ditto! Are you reading what I write?

> So the comment has just always been wrong, I think. It may be that the
> original swapping code was doing raw requests instead of real IO, so maybe
> the comment was actually correct back in 1992 or something. My memory
> isn't good enough..

Good, so now you agree that the corrected comment (as per pre9) is crap,
and the patch I sent that changes the wording to:

"Ok, this is an expanded form so that we can use the same
request for paging requests."

is so much better than _you_ mixing ->waiting and ->sem into this paging
or non-paging pool?

But in fact the whole comment block should just be removed. It gives no
useful or additional information.

-- 
Jens Axboe