Re: [Qemu-devel] [PATCH 6/7] s390x/pci: move the memory region write from pcistg

[Pierre Morel <pmorel@linux.vnet.ibm.com>]

On 10/11/2017 10:51, Cornelia Huck wrote:
> On Fri, 10 Nov 2017 17:40:12 +0800
> Yi Min Zhao <zyimin@linux.vnet.ibm.com> wrote:
> 
>> 在 2017/11/10 上午3:23, Cornelia Huck 写道:
>>> On Tue,  7 Nov 2017 18:24:38 +0100
>>> Pierre Morel <pmorel@linux.vnet.ibm.com> wrote:
>>>   
>>>> Let's move the memory region write from pcistg into a dedicated
>>>> function.
>>>> This allows us to prepare a later patch searching for subregions
>>>> inside of the memory region.
>>> OK, so here is the memory region write. Do we have any sleeping
>>> endianness bugs in there for when we wire up tcg? I'm not sure how this
>>> plays with the bswaps (see patch 1).
>>>
>>> But maybe I've just gotten lost somewhere.
>> I think there's no error. For PCI bars' MRs, we got the little-endian data
>> that is exactly fit to the byte ordering of pcilg instruction. For PCI
>> config
>> space, the data has been swapped according to the cpu byte ordering.
> 
> Host or target cpu?
> 
>> So we use zpci_swap_endian() to swap the data back to the little-endian
>> ordering.
> 


I do not see where we use the zpci_swap_endian() function in this patch 
or in the zpci_write_bar() function.



> That swap is unconditional. If we were running on a little-endian host,
> it would be wrong, wouldn't it?

I think there is no problem here, we do not use the swap function but we 
use the memory_region_dispatch_write() to access a subregion of the PCI 
device which is defined as DEVICE_LITTLE_ENDIAN

AFAIU The memory access process the endianness correctly.


> 
>>>   
>>>> Signed-off-by: Pierre Morel <pmorel@linux.vnet.ibm.com>
>>>> Reviewed-by: Yi Min Zhao <zyimin@linux.vnet.ibm.com>
>>>> ---
>>>>    hw/s390x/s390-pci-inst.c | 27 +++++++++++++++++----------
>>>>    1 file changed, 17 insertions(+), 10 deletions(-)
> 


-- 
Pierre Morel
Linux/KVM/QEMU in Böblingen - Germany