On Fri, 24 Sep 2021 14:48:52 -0300
Jason Gunthorpe <jgg@xxxxxxxx> wrote:
> On Fri, Sep 24, 2021 at 05:57:03PM +0200, Christoph Hellwig wrote:
> > Ensure pgsize_bitmap is always valid by initializing it to ULONG_MAX
> > in vfio_iommu_type1_open and remove the now pointless update for
> > the external domain case in vfio_iommu_type1_attach_group, which was
> > just setting pgsize_bitmap to ULONG_MAX when only external domains
> > were attached.
> >
> > Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> > drivers/vfio/vfio_iommu_type1.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
> > index a48e9f597cb213..2c698e1a29a1d8 100644
> > +++ b/drivers/vfio/vfio_iommu_type1.c
> > @@ -2196,7 +2196,6 @@ static int vfio_iommu_type1_attach_group(void *iommu_data,
> > if (!iommu->external_domain) {
> > INIT_LIST_HEAD(&domain->group_list);
> > iommu->external_domain = domain;
> > - vfio_update_pgsize_bitmap(iommu);
> > } else {
> > kfree(domain);
> > }
> > @@ -2582,6 +2581,7 @@ static void *vfio_iommu_type1_open(unsigned long arg)
> > mutex_init(&iommu->lock);
> > BLOCKING_INIT_NOTIFIER_HEAD(&iommu->notifier);
> > init_waitqueue_head(&iommu->vaddr_wait);
> > + iommu->pgsize_bitmap = ULONG_MAX;
>
> I wonder if this needs the PAGE_MASK/SIZE stuff?
>
> iommu->pgsize_bitmap = ULONG_MASK & PAGE_MASK;
>
> ?
>
> vfio_update_pgsize_bitmap() goes to some trouble to avoid setting bits
> below the CPU page size here
Yep, though PAGE_MASK should already be UL, so just PAGE_MASK itself
should work. The ULONG_MAX in the update function just allows us to
detect sub-page, ex. if the IOMMU supports 2K we can expose 4K minimum,
but we can't if the min IOMMU page is 64K. Thanks,
Alex
