Re: [Openvortex-dev] Findings...

[Jeff Muizelaar]

On Wed, 23 Apr 2003, Shamus wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi!
>
> >I agree. I expected the card wouldn't be doing anything before it was
>
> >programmed to.
> >Imagine a soundcard capturing sound and writing it per DMA somewhere
> >into memory ... very bad.
> >Maybe there some few Vortex card badly designed with such odd
> behaviour.
> >Anyway,  our driver
> >has to deal with such situations.
> >
> >Any sugestions ?
>
> According "Linux Device Drivers, 2nd Ed.", the right way to handle IRQ
> grabbing is at the device's open() code--they also suggest that the
> IRQ should be released in the device's close() code as well which
> means you'd have to keep track of how many open()'s have happened and
> how many close()'s before releasing the IRQ (keep a reference count).

I disagree. According to "Linux Device Drivers" the reason for doing this
is so that devices don't hog interrupts, this is not a problem if the
device can share it. Additionally, I don't remember actually seeing any
drivers do a request_irq at open time. They all just do it at load time.

> In this case, it sounds like it would be a good idea to do things this
> way for the au88x0 driver. That way the card would (ideally) be fully
> initialized before an IRQ can occur. As for badly designed Vortex
> cards, mine's revision FA, so that is a distinct possibility.  ;-(  At
> least now there is the possibility of finding these problems...  :-p

The current version in cvs should not have this bug. Give it a try.

-Jeff