On 2/17/2014 6:50 AM, Mark Cave-Ayland wrote:
> On 14/02/14 14:54, Peter Crosthwaite wrote:
>
>>> The short answer is "we don't know" because we don't have any
>>> documentation.
>>
>> Sigh.... This has happened quite a lot lately.
>>
>> If the kernel driver has macros, re-use them as much as possible. If
>> you have a vague idea on whats, what, a few well invented names would
>> help the device self-documentation.
>
> Okay. I now have a revised version which borrows macro names from the
> Linux and BSD drivers which I think should be more readable. I'll post
> the revised version to the list shortly.
>
>>>> Your hander switch statements stride in 4, are you only doing this for
>>>> your one exception case of that one-byte big-endian access I commented
>>>> earlier.
>>>
>>>
>>> Yes, that is correct.
>>>
>>
>> Should you trap misaligned accesses then?
>
> Over the weekend I found out that the non-BT458 accesses (addr >= 0x10)
> are done as byte accesses and so byte accesses do need to be allowed to
> these registers. My interpretation of reading the SBus documentation is
> that on real hardware the bus converts accesses for you, and so I don't
> think a trap would be suitable here. Also I've not found an image (yet)
> that attempts bad accesses in this way across my OpenBIOS ISO test suite...
When you create the memory region for the registers, try this:
.endianness = DEVICE_BIG_ENDIAN,
.impl = {
.min_access_size = 1,
.max_access_size = 1,
},
and ignore the bottom 2 address bits for the DAC accesses.
QEMU should then break any 32-bit accesses down to the correct 8-bit
accesses.
Bob
