2007-05-02 17:58:34

by john stultz

[permalink] [raw]
Subject: Re: [PATCH 1/3] ia64: convert to use clocksource code

On Fri, 2007-04-27 at 12:11 -0400, Peter Keilty wrote:
> Daniel Walker wrote:
> >On Fri, 2007-04-27 at 11:42 -0400, Peter Keilty wrote:
> >>>There is a read(), and a vread() did you modify the slow syscall path to
> >>>use the vread()?
> >>>
> >John mentioned that he thought fsys_mmio_ptr could be held in the vread
> >pointer. vread() is used in x86 for vsyscalls. It looks like you've used
> >the update_vsyscall() which is also used for vsyscalls. So vread could
> >also be used .. Have you considered that at all?
> >
> >
> No, but yes it can be done, overloading the meaning.

Yea. I'm not really psyched about overloading the vread pointer's use. I
mentioned it could be done if the #ifdef was objected to, but it seems a
bit abusive. The #ifdef isn't great, but I think its something I can
live with for now. At least its explicit.

> It would need to change in the future if vread was needed.
> I have no strong argument against using it.

Yea. I'd hold off on that for now.

thanks
-john



2007-05-02 19:10:09

by Daniel Walker

[permalink] [raw]
Subject: Re: [PATCH 1/3] ia64: convert to use clocksource code

On Wed, 2007-05-02 at 10:58 -0700, john stultz wrote:
> On Fri, 2007-04-27 at 12:11 -0400, Peter Keilty wrote:
> > Daniel Walker wrote:
> > >On Fri, 2007-04-27 at 11:42 -0400, Peter Keilty wrote:
> > >>>There is a read(), and a vread() did you modify the slow syscall path to
> > >>>use the vread()?
> > >>>
> > >John mentioned that he thought fsys_mmio_ptr could be held in the vread
> > >pointer. vread() is used in x86 for vsyscalls. It looks like you've used
> > >the update_vsyscall() which is also used for vsyscalls. So vread could
> > >also be used .. Have you considered that at all?
> > >
> > >
> > No, but yes it can be done, overloading the meaning.
>
> Yea. I'm not really psyched about overloading the vread pointer's use. I
> mentioned it could be done if the #ifdef was objected to, but it seems a
> bit abusive. The #ifdef isn't great, but I think its something I can
> live with for now. At least its explicit.

Use of config options like that is a bad precedence I think, which is
why I commented on it .. Since we have a vread pointer that exists
already, and it's used a very similar purpose it seems like bloat to
just add another pointer..

We could change the vread to be a plain void pointer, then let each
architecture use it however they want.

Daniel