On 5/25/2023 1:18 AM, Alex Williamson wrote:
On Wed, 24 May 2023 10:45:29 -0600
Alex Williamson <alex.williamson@xxxxxxxxxx> wrote:
On Wed, 17 May 2023 15:27:18 +0530
Nipun Gupta <nipun.gupta@xxxxxxx> wrote:
<snip>
+
+MODULE_DEVICE_TABLE(cdx, vfio_cdx_table);
+
+static struct cdx_driver vfio_cdx_driver = {
+ .probe = vfio_cdx_probe,
+ .remove = vfio_cdx_remove,
+ .match_id_table = vfio_cdx_table,
+ .driver = {
+ .name = "vfio-cdx",
+ .owner = THIS_MODULE,
+ },
+ .driver_managed_dma = true,
Hmm, looks like cdx bus is broken here, there's no actual
implementation of a dma_configure callback and no setup of the IOMMU
default domain for theoretical cdx drivers that might want to use the
DMA API. Without that, this driver_manged_dma flag doesn't provide any
guarantees to a vfio driver that we have exclusive ownership of the
group. Please fix, this flag needs to actually have some meaning on
cdx. Thanks,
Alex
Agree, this change was missed on CDX bus and we are working on fixing
this. Shall I send this fix as a commit with this patch?
Thanks,
Nipun