Re: [Openvortex-dev] ac97 codec read-write cleanup in cvs

[Jeff Muizelaar]

On Tue, 17 Jun 2003, Ryan Underwood wrote:

>
> Hi,
>
> Just a quick note on the patch --
>
> +       /* verify that the write succeeded */
> +       if (vortex_codec_read(codec, addr) != data) {
> +               printk(KERN_ERR "vortex: ac97 codec write failed\n");
>
> I'm not sure if that's correct.  The first read after the write will
> always return the previous data, not the written data.  That was the
> problem that I started with.  So, if this code actually works instead of
> bailing with an error every time, I would be very surprised.  I had to
> read twice before the data would match what I wrote to the codec.

Yeah, I know, it is commented out in cvs right now, I committed before
doing any testing at all...

I don't think this is pci caching issue. Afaik, all writes should be
committed on the first read from the mmio region, and prefetches are only
allowed when they are no side-effects.

Maybe it is a timing thing, who knows...

We actually should not drop the spinlock until we are sure that codec
write finished...

I'll look into this more when I get I chance....

-Jeff