Re: [PATCH v2] xen/pci: Refactor PCI MSI interrupts related code

[Jan Beulich <jbeulich@xxxxxxxx>]

On 21.04.2021 11:15, Rahul Singh wrote:
> Hi Roger,
> 
>> On 21 Apr 2021, at 9:16 am, Roger Pau Monné <roger.pau@xxxxxxxxxx> wrote:
>>
>> On Wed, Apr 21, 2021 at 08:07:08AM +0000, Rahul Singh wrote:
>>> Hi Jan,
>>>
>>>> On 20 Apr 2021, at 4:36 pm, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>
>>>> On 20.04.2021 15:45, Rahul Singh wrote:
>>>>>> On 19 Apr 2021, at 1:33 pm, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>>> On 19.04.2021 13:54, Julien Grall wrote:
>>>>>>> For the time being, I think move this code in x86 is a lot better than 
>>>>>>> #ifdef or keep the code in common code.
>>>>>>
>>>>>> Well, I would perhaps agree if it ended up being #ifdef CONFIG_X86.
>>>>>> I would perhaps not agree if there was a new CONFIG_* which other
>>>>>> (future) arch-es could select if desired.
>>>>>
>>>>> I agree with Julien moving the code to x86 file as currently it is 
>>>>> referenced only in x86 code
>>>>> and as of now we are not sure how other architecture will implement the 
>>>>> Interrupt remapping
>>>>> (via IOMMU or any other means).  
>>>>>
>>>>> Let me know if you are ok with moving the code to x86.
>>>>
>>>> I can't answer this with "yes" or "no" without knowing what the alternative
>>>> would be. As said, if the alternative is CONFIG_X86 #ifdef-ary, then yes.
>>>> If a separate CONFIG_* gets introduced (and selected by X86), then a
>>>> separate file (getting built only when that new setting is y) would seem
>>>> better to me.
>>>
>>> I just made a quick patch. Please let me know if below patch is ok. I move 
>>> the definition to  "passthrough/x86/iommu.c” file.
>>>
>>> diff --git a/xen/drivers/passthrough/pci.c b/xen/drivers/passthrough/pci.c
>>> index 705137f8be..199ce08612 100644
>>> --- a/xen/drivers/passthrough/pci.c
>>> +++ b/xen/drivers/passthrough/pci.c
>>> @@ -1303,13 +1303,6 @@ static int __init setup_dump_pcidevs(void)
>>> }
>>> __initcall(setup_dump_pcidevs);
>>>
>>> -int iommu_update_ire_from_msi(
>>> -    struct msi_desc *msi_desc, struct msi_msg *msg)
>>> -{
>>> -    return iommu_intremap
>>> -           ? iommu_call(&iommu_ops, update_ire_from_msi, msi_desc, msg) : 
>>> 0;
>>> -}
>>> -
>>> static int iommu_add_device(struct pci_dev *pdev)
>>> {
>>>     const struct domain_iommu *hd;
>>> diff --git a/xen/drivers/passthrough/x86/iommu.c 
>>> b/xen/drivers/passthrough/x86/iommu.c
>>> index b90bb31bfe..cf51dec564 100644
>>> --- a/xen/drivers/passthrough/x86/iommu.c
>>> +++ b/xen/drivers/passthrough/x86/iommu.c
>>> @@ -340,6 +340,13 @@ bool arch_iommu_use_permitted(const struct domain *d)
>>>             likely(!p2m_get_hostp2m(d)->global_logdirty));
>>> }
>>>
>>> +int iommu_update_ire_from_msi(
>>> +    struct msi_desc *msi_desc, struct msi_msg *msg)
>>> +{
>>> +    return iommu_intremap
>>> +           ? iommu_call(&iommu_ops, update_ire_from_msi, msi_desc, msg) : 
>>> 0;
>>> +}
>>> +
>>> /*
>>>  * Local variables:
>>>  * mode: C
>>> diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
>>> index ea0cd0f1a2..bd42d87b72 100644
>>> --- a/xen/include/xen/iommu.h
>>> +++ b/xen/include/xen/iommu.h
>>> @@ -243,7 +243,6 @@ struct iommu_ops {
>>>                            u8 devfn, device_t *dev);
>>> #ifdef CONFIG_HAS_PCI
>>>     int (*get_device_group_id)(u16 seg, u8 bus, u8 devfn);
>>> -    int (*update_ire_from_msi)(struct msi_desc *msi_desc, struct msi_msg 
>>> *msg);
>>> #endif /* HAS_PCI */
>>>
>>>     void (*teardown)(struct domain *d);
>>> @@ -272,6 +271,7 @@ struct iommu_ops {
>>>     int (*adjust_irq_affinities)(void);
>>>     void (*sync_cache)(const void *addr, unsigned int size);
>>>     void (*clear_root_pgtable)(struct domain *d);
>>> +    int (*update_ire_from_msi)(struct msi_desc *msi_desc, struct msi_msg 
>>> *msg);
>>
>> You also need to move the function prototype
>> (iommu_update_ire_from_msi) from iommu.h into asm-x86/iommu.h, or
>> maybe you could just define the function itself as static inline in
>> asm-x86/iommu.h?
> 
> 
> Ok. I will move function prototype (iommu_update_ire_from_msi) from 
> iommu.h into asm-x86/iommu.h.
> 
> I first tried to make the function as static inline in "asm-x86/iommu.h" but 
> after 
> that I observe the compilation error for debug build because "asm/iommu.h” 
> is included in "xen/iommu.h” before iommu_call() is defined in "xen/iommu.h”.
> That's why I decided to move it to the "passthrough/x86/iommu.c” file.

Which in turn would be an argument for keeping it in xen/iommu.h and
wrap it in an #ifdef.

Jan