Any complications of using CONFIG_MIPS_UNCACHED..?

Any complications of using CONFIG_MIPS_UNCACHED..?

[Ashish anand]

Hello,

I saw a good bit of discussion about CONFIG_MIPS_UNCACHED but
still i am yet to know...

If I want to use CONFIG_MIPS_UNCACHED (ignoring performance)
what all the side-effects and any restrictions that linux should
take care in software ?

I observed something surprising on my R4k mips system(virtually indexed
caches), after i use this option my driver never got status updation by
device in transmit and receive decriptors in system memory , Irrespective
of I (flushed+invalidate) caches or not...

if i don't use CONFIG_MIPS_UNCACHED then before checking status I need to 
(flush+invalidate) cache and whole thing works great...

Regards,
Ashish

Re: Any complications of using CONFIG_MIPS_UNCACHED..?

[Ralf Baechle]

On Wed, May 21, 2003 at 11:16:49AM +0530, Ashish anand wrote:

> I saw a good bit of discussion about CONFIG_MIPS_UNCACHED but
> still i am yet to know...
> 
> If I want to use CONFIG_MIPS_UNCACHED (ignoring performance)
> what all the side-effects and any restrictions that linux should
> take care in software ?
> 
> I observed something surprising on my R4k mips system(virtually indexed
> caches), after i use this option my driver never got status updation by
> device in transmit and receive decriptors in system memory , Irrespective
> of I (flushed+invalidate) caches or not...

Descriptors sounds like networking.  Network descriptors should usually
be stored in uncached or otherwise coherent space that is allocated
through pci_alloc_consistent anyway.  If you do that CONFIG_MIPS_UNCACHED
should not make any difference anymore.

> if i don't use CONFIG_MIPS_UNCACHED then before checking status I need to 
> (flush+invalidate) cache and whole thing works great...

That's a funny bug.  Normally the opposite behaviour would be expected for
a bug ...

  Ralf