Re: [PATCH for-5.2] s390x/pci: fix endianness issues

[Thomas Huth <thuth@redhat.com>]

On 17/11/2020 19.49, Philippe Mathieu-Daudé wrote:
> On 11/17/20 7:39 PM, Thomas Huth wrote:
>> On 17/11/2020 19.30, Philippe Mathieu-Daudé wrote:
>>> On 11/17/20 7:19 PM, Matthew Rosato wrote:
>>>> On 11/17/20 12:59 PM, Philippe Mathieu-Daudé wrote:
>>>>> On 11/17/20 6:13 PM, Cornelia Huck wrote:
>>>>>> zPCI control blocks are big endian, we need to take care that we
>>>>>> do proper accesses in order not to break tcg guests on little endian
>>>>>> hosts.
>>>>>>
>>>>>> Fixes: 28dc86a07299 ("s390x/pci: use a PCI Group structure")
>>>>>> Fixes: 9670ee752727 ("s390x/pci: use a PCI Function structure")
>>>>>> Fixes: 1e7552ff5c34 ("s390x/pci: get zPCI function info from host")
>>>>>> Signed-off-by: Cornelia Huck <cohuck@redhat.com>
>>>>>> ---
>>>>>>
>>>>>> Works for me with virtio-pci devices for tcg on x86 and s390x, and
>>>>>> for kvm.
>>>>>> The vfio changes are not strictly needed; did not test them due to
>>>>>> lack of
>>>>>> hardware -- testing appreciated. >>
>>>>>> As this fixes a regression, I want this in 5.2.
>>>>>>
>>>>>> ---
>>>>>>   hw/s390x/s390-pci-bus.c  | 12 ++++++------
>>>>>>   hw/s390x/s390-pci-inst.c |  4 ++--
>>>>>>   hw/s390x/s390-pci-vfio.c |  8 ++++----
>>>>>>   3 files changed, 12 insertions(+), 12 deletions(-)
>>>>>>
>>>>>> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
>>>>>> index e0dc20ce4a56..17e64e0b1200 100644
>>>>>> --- a/hw/s390x/s390-pci-bus.c
>>>>>> +++ b/hw/s390x/s390-pci-bus.c
>>>>>> @@ -787,12 +787,12 @@ static void s390_pci_init_default_group(void)
>>>>>>     static void set_pbdev_info(S390PCIBusDevice *pbdev)
>>>>>>   {
>>>>>> -    pbdev->zpci_fn.sdma = ZPCI_SDMA_ADDR;
>>>>>> -    pbdev->zpci_fn.edma = ZPCI_EDMA_ADDR;
>>>>>> -    pbdev->zpci_fn.pchid = 0;
>>>>>> +    stq_p(&pbdev->zpci_fn.sdma, ZPCI_SDMA_ADDR);
>>>>>
>>>>> "zPCI control blocks are big endian" so don't we
>>>>> need the _be_ accessors? stq_be_p() etc...
>>>>>
>>>>
>>>> I don't think this is necessary.  This is only available for target
>>>> s390x, which is always big endian...  cpu-all.h should define stq_p as
>>>> stq_be_p for example inside the #if defined(TARGET_WORDS_BIGENDIAN).
>>>
>>> But if you run on little-endian host, you need to byte-swap that,
>>> isn't it?
>>
>> It's done by the macros. They depend on the target endianess. See cpu-all.h.
> 
> I'm confused because the description is about target endianness,
> but stq_p() is about host alignment.

stq_p() is apparently also about endianess. Why would it depend on
TARGET_WORDS_BIGENDIAN otherwise?

 Thomas