On 10/06/2016 11:40 AM, Paolo Bonzini wrote:
>
>
> On 06/10/2016 10:37, David Gibson wrote:
>> On Thu, Oct 06, 2016 at 09:23:11AM +0200, Paolo Bonzini wrote:
>>>
>>>
>>> On 06/10/2016 05:45, David Gibson wrote:
>>>> qtest is about testing hardware, not the guest cpu, and whether you're
>>>> accessing MMIO space or buffers the hardware will read via DMA, it's
>>>> the hardware which determines the correct endianness, not the guest
>>>> cpu.
>>>
>>> Well, sort of... qtest writes are definitely injecting things from the
>>> point of view of the guest CPU. For example see Laurent's patches that
>>> add RTAS commands to qtest.
>>
>> Sure. My point is that any guest CPU can do both LE and BE accesses
>> pretty simply, so we might as well specify which we want.
>
> On this I do agree.
Yes. It is pretty confusing today from the test side. To clarify
things a little, I came up with this :
+static inline uint32_t make_be32(uint32_t data)
+{
+ return bswap32(data);
+}
...
+ writel(ASPEED_FLASH_BASE, make_be32(some_page_addr));
Which is what a Little Endian CPU would do to write a BE value. So
it should probably be called, make_b32_on_le_cpu() to be correct.
I am not sure there is a good interface for these accesses ...
C.
