Re: [PATCH v2 05/11] xen/arm: Mark device as PCI while creating one

[Jan Beulich <jbeulich@xxxxxxxx>]

On 28.09.2021 10:09, Oleksandr Andrushchenko wrote:
> 
> On 27.09.21 13:26, Jan Beulich wrote:
>> On 27.09.2021 12:04, Oleksandr Andrushchenko wrote:
>>> On 27.09.21 13:00, Jan Beulich wrote:
>>>> On 27.09.2021 11:35, Oleksandr Andrushchenko wrote:
>>>>> On 27.09.21 12:19, Jan Beulich wrote:
>>>>>> On 27.09.2021 10:45, Oleksandr Andrushchenko wrote:
>>>>>>> On 27.09.21 10:45, Jan Beulich wrote:
>>>>>>>> On 23.09.2021 14:54, Oleksandr Andrushchenko wrote:
>>>>>>>>> --- a/xen/drivers/passthrough/pci.c
>>>>>>>>> +++ b/xen/drivers/passthrough/pci.c
>>>>>>>>> @@ -328,6 +328,9 @@ static struct pci_dev *alloc_pdev(struct pci_seg 
>>>>>>>>> *pseg, u8 bus, u8 devfn)
>>>>>>>>>          *((u8*) &pdev->bus) = bus;
>>>>>>>>>          *((u8*) &pdev->devfn) = devfn;
>>>>>>>>>          pdev->domain = NULL;
>>>>>>>>> +#ifdef CONFIG_ARM
>>>>>>>>> +    pci_to_dev(pdev)->type = DEV_PCI;
>>>>>>>>> +#endif
>>>>>>>> I have to admit that I'm not happy about new CONFIG_<arch> conditionals
>>>>>>>> here. I'd prefer to see this done by a new arch helper, unless there 
>>>>>>>> are
>>>>>>>> obstacles I'm overlooking.
>>>>>>> Do you mean something like arch_pci_alloc_pdev(dev)?
>>>>>> I'd recommend against "alloc" in its name; "new" instead maybe?
>>>>> I am fine with arch_pci_new_pdev, but arch prefix points to the fact that
>>>>> this is just an architecture specific part of the pdev allocation rather 
>>>>> than
>>>>> actual pdev allocation itself, so with this respect arch_pci_alloc_pdev 
>>>>> seems
>>>>> more natural to me.
>>>> The bulk of the function is about populating the just allocated struct.
>>>> There's no arch-specific part of the allocation (so far, leaving aside
>>>> MSI-X), you only want and arch-specific part of the initialization. I
>>>> would agree with "alloc" in the name if further allocation was to
>>>> happen there.
>>> Hm, then arch_pci_init_pdev sounds more reasonable
>> Fine with me.
> 
> Do we want this to be void or returning an error code? If error code is 
> needed,
> then we would also need a roll-back function, e.g. arch_pci_free_pdev or
> arch_pci_release_pdev or arch_pci_fini_pdev or something, so it can be used in
> case of error or in free_pdev function.

I'd start with void and make it return an error (and deal with necessary
cleanup) only once a need arises.

Jan