Hi Vikas,
On 11/6/20 3:54 AM, Vikas Gupta wrote:
> Hi Alex,
>
> On Thu, Nov 5, 2020 at 12:38 PM Alex Williamson
> <alex.williamson@xxxxxxxxxx> wrote:
>>
>> On Thu, 5 Nov 2020 11:32:55 +0530
>> Vikas Gupta <vikas.gupta@xxxxxxxxxxxx> wrote:
>>
>>> diff --git a/include/uapi/linux/vfio.h b/include/uapi/linux/vfio.h
>>> index 2f313a238a8f..aab051e8338d 100644
>>> --- a/include/uapi/linux/vfio.h
>>> +++ b/include/uapi/linux/vfio.h
>>> @@ -203,6 +203,7 @@ struct vfio_device_info {
>>> #define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */
>>> #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6) /* vfio-fsl-mc device */
>>> #define VFIO_DEVICE_FLAGS_CAPS (1 << 7) /* Info supports caps */
>>> +#define VFIO_DEVICE_FLAGS_MSI (1 << 8) /* Device supports msi */
>>> __u32 num_regions; /* Max region index + 1 */
>>> __u32 num_irqs; /* Max IRQ index + 1 */
>>> __u32 cap_offset; /* Offset within info struct of first cap */
>>
>> This doesn't make any sense to me, MSIs are just edge triggered
>> interrupts to userspace, so why isn't this fully described via
>> VFIO_DEVICE_GET_IRQ_INFO? If we do need something new to describe it,
>> this seems incomplete, which indexes are MSI (IRQ_INFO can describe
>> that)? We also already support MSI with vfio-pci, so a global flag for
>> the device advertising this still seems wrong. Thanks,
>>
>> Alex
>>
> Since VFIO platform uses indexes for IRQ numbers so I think MSI(s)
> cannot be described using indexes.
> In the patch set there is no difference between MSI and normal
> interrupt for VFIO_DEVICE_GET_IRQ_INFO.
in vfio_platform_irq_init() we first iterate on normal interrupts using
get_irq(). Can't we add an MSI index at the end of this list with
vdev->irqs[i].count > 1 and set vdev->num_irqs accordingly?
Thanks
Eric
> The patch set adds MSI(s), say as an extension, to the normal
> interrupts and handled accordingly. Do you see this is a violation? If
> yes, then we`ll think of other possible ways to support MSI for the
> platform devices.
> Macro VFIO_DEVICE_FLAGS_MSI can be changed to any other name if it
> collides with an already supported vfio-pci or if not necessary, we
> can remove this flag.
>
> Thanks,
> Vikas
>
