[openib-general] Re: [PATCH 0/7] AMSO1100 RNIC Driver

Tom Tucker tom at opengridcomputing.com
Thu Mar 16 16:17:41 PST 2006


On Thu, 2006-03-16 at 14:50 -0800, Roland Dreier wrote:
>  > - BE issue resolved by using __raw_write[lw](cpu_to_be[32|16]...
> 
> I think maybe you sent out the wrong patchset (or maybe I screwed up
> my mail folder and have an old patchset).  But I see stuff like:
> 
>  > +#define C2_SET_CUR_RX(c2dev, cur_rx) \
>  > +	writel(cpu_to_be32(cur_rx), c2dev->mmio_txp_ring + 4092)
> 
This looks like an oversight...

> which is going to be bad on a big-endian arch.

But there are other writel's in the code that I thought were correct.
For example, if the byte ordering was excpected to be PCI/LE:
- write[wl](0,...) 		benign, but arguably bogus
- write[wl](LE anything, ...)	should be correct, yes?
- write[wl](cpu_xxx, ... )	any that remain are oversights.

> 
> BTW, I'm setting up my PPC 4xx board again soon, so I'll be able to
> try this driver on a big-endian, non-cache-coherent DMA system.
>  - R.




More information about the general mailing list