Re: [Qemu-devel] Re: [PATCH] Always swap endianness in DBDMA

[Alexander Graf]

On 24.12.2009, at 01:08, Aurelien Jarno wrote:

> On Thu, Dec 24, 2009 at 12:39:36AM +0100, Laurent Vivier wrote:
>> Le mercredi 23 décembre 2009 à 10:55 +0000, Paul Brook a écrit :
>>>> The problem is that the whole define is just plain wrong which tells me
>>>> that the code is using the bswap functions incorrectly. This really
>>>> needs to be fixed by someone who knows the dbdma device. I don't see how
>>>> calling incorrect calls even more incorrect makes any difference.
>>> 
>>> The real problem is that devices shouldn't be doing byteswapping at all. 
>>> This 
>>> should be determined by the (currently non-existant) bus layers and 
>>> implemented in generic code before the device callback.
>>> 
>>> The current code[1] is a nasty hack that sort-of works for most of the 
>>> current 
>>> machines because all devices happen to be connected the same way.  However 
>>> there are other machines (e.g. ixp4xx) some peripherals are connected 
>>> natively, whereas others are cross-wired.
>>> 
>>> On a related note, I'm not sure what the author of mac_bdbma.c was 
>>> smoking[2]. 
>> 
>> I didn't smoke... just need some sleep.
>> 
>>> It appears to keep register values in big-endian form for no good reason. 
>>> Much 
>>> easier would be to store them in native form, and just do the byteswapping 
>>> when accessed by the CPU.
>>> 
> 
> I have just posted a patch that does that, and also simulate the bus is
> connected backward, using the same trick as in other devices. Alexander,
> could you please try if it works for you on a big endian host?

I just did and it works on PPC guest, PPC64 host.

Alex