Re: [PATCH v4 01/14] iommufd/viommu: Introduce IOMMUFD_OBJ_VDEVICE and its related struct

From: Jason Gunthorpe
Date: Fri Oct 25 2024 - 09:21:09 EST


On Fri, Oct 25, 2024 at 06:53:01PM +1100, Alexey Kardashevskiy wrote:
> > +#define iommufd_vdevice_alloc(ictx, drv_struct, member) \
> > + ({ \
> > + static_assert( \
> > + __same_type(struct iommufd_vdevice, \
> > + ((struct drv_struct *)NULL)->member)); \
> > + static_assert(offsetof(struct drv_struct, member.obj) == 0); \
> > + container_of(_iommufd_object_alloc(ictx, \
> > + sizeof(struct drv_struct), \
> > + IOMMUFD_OBJ_VDEVICE), \
> > + struct drv_struct, member.obj); \
> > + })
> > #endif
>
> A nit: it hurts eyes to read:
>
> mock_vdev = iommufd_vdevice_alloc(viommu->ictx, mock_vdevice, core);
>
> vs.
>
> mock_vdev = iommufd_vdevice_alloc(viommu->ictx, struct mock_vdevice, core);
>
> as for the former I go searching for a "mock_vdevice" variable and for the
> latter it is clear it is 1) a macro 2) which does some type checking.
>
> also, it makes it impossible to pass things like typeof(..) or a type from
> typedef. Thanks,

Makes sense to me

And the container_of() should not be used in these macros, the point
was to avoid it to make the PTR_ERR behavior cleraer. Just put a force
type cast

Jason