On 8/2/2021 9:20 PM, Joerg Roedel wrote:

On Wed, Jul 28, 2021 at 10:52:28AM -0400, Tianyu Lan wrote:

quoted

In Isolation VM, all shared memory with host needs to mark visible
to host via hvcall. vmbus_establish_gpadl() has already done it for
storvsc rx/tx ring buffer. The page buffer used by vmbus_sendpacket_
mpb_desc() still need to handle. Use DMA API to map/umap these
memory during sending/receiving packet and Hyper-V DMA ops callback
will use swiotlb function to allocate bounce buffer and copy data
from/to bounce buffer.

I am wondering why you dont't use DMA-API unconditionally? It provides
enough abstraction to do the right thing for isolated and legacy VMs.

In VMbus, there is already a similar bounce buffer design and so there 
is no need to call DMA-API for such buffer. Calling DMA-API is to use
swiotlb bounce buffer for those buffer which hasn't been covered. This
is why need to conditionally call DMA API.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu