On 7/1/22 16:07, Tian, Kevin wrote:
From: Alexey Kardashevskiy <a...@ozlabs.ru>
Sent: Friday, July 1, 2022 12:58 PM

On 4/8/22 01:23, Jason Gunthorpe via iommu wrote:
IOMMU_CACHE means that normal DMAs do not require any additional
mechanism and is the basic uAPI that VFIO exposes to userspace. For
instance VFIO applications like DPDK will not work if additional coherency
operations are required.

Therefore check IOMMU_CAP_CACHE_COHERENCY like vdpa & usnic do
allowing an IOMMU backed VFIO device to be created.

This just broke VFIO on POWER which does not use iommu_ops.

In this case below check is more reasonable to be put in type1
attach_group(). Do a iommu_group_for_each_dev() to verify
CACHE_COHERENCY similar to what Robin did for INTR_REMAP.

ah makes sense. I've posted an RFC with another problem - "[RFC PATCH kernel] vfio: Skip checking for IOMMU_CAP_CACHE_COHERENCY on POWER and more", would be great if both addressed, or I try moving those next week :) Thanks,

(sorry no access to my build machine now but I suppose Jason
can soon work out a fix once he sees this. 😊)

Signed-off-by: Jason Gunthorpe <j...@nvidia.com>
   drivers/vfio/vfio.c | 7 +++++++
   1 file changed, 7 insertions(+)

diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
index a4555014bd1e72..9edad767cfdad3 100644
--- a/drivers/vfio/vfio.c
+++ b/drivers/vfio/vfio.c
@@ -815,6 +815,13 @@ static int __vfio_register_dev(struct vfio_device

   int vfio_register_group_dev(struct vfio_device *device)
+       /*
+        * VFIO always sets IOMMU_CACHE because we offer no way for
userspace to
+        * restore cache coherency.
+        */
+       if (!iommu_capable(device->dev->bus,
+               return -EINVAL;
        return __vfio_register_dev(device,


iommu mailing list

Reply via email to