Re: [Qemu-devel] [PATCH v2 4/6] prep: move int-ack register from PReP to Raven PCI emulation

[Avi Kivity]

On 04/29/2012 11:12 PM, Andreas Färber wrote:
> Am 29.04.2012 10:32, schrieb Avi Kivity:
> > On 04/28/2012 09:46 PM, Andreas Färber wrote:
> >> Am 14.04.2012 22:48, schrieb Hervé Poussineau:
> >>> Register is one byte-wide (as per specification), so there is no need to 
> >>> specify endianness.
> >>
> >> The region was 4 bytes before, now it's 1. What happens when a 4-byte
> >> read is attempted at that address? Do we need to specify the valid
> >> widths for the MemoryRegion? Or is such a read constructed from this
> >> region and (assuming) the return value of an unassigned read?
> > 
> > This area of what happens during access that falls across region
> > boundaries is very underspecified in qemu; nor is it clear what happens
> > in real hardware (in all its variations).
>
> So, what's your conclusion here? Should we add .valid.max_access_size =
> 1? Or shall I apply as is?

If the specification says 1 byte and there are no known guests that
abuse it, I say change it.  Regardless, we need to collect requirements
for this shady area and implement them.

-- 
error compiling committee.c: too many arguments to function