On Tue, Jul 23, 2024 at 02:26:23PM +1000, Alexey Kardashevskiy wrote: > > > On 22/7/24 22:06, Jason Gunthorpe wrote: > > On Mon, Jul 22, 2024 at 08:19:23PM +1000, Alexey Kardashevskiy wrote: > > > > > If there is vIOMMU, then the driver in the VM can decide whether it wants > > > private or shared memory for DMA, pass that new flag to dma_map() and 1) > > > have DMA memory allocated from the private pool (== no page state changes) > > > and 2) have C-bit set in the vIOMMU page table (which is in the VM memory). > > > > Not all HW supports a flow like that. > > Fair point but still, under what imaginary circumstance a driver could > decide to flip T=0/1 when up and running? It seems some people are thinking they need to do T=0 stuff before doing device attestation. But that wasn't my point, the issue is that the translation is different depending on T=0/1. On those implementations T=0 means "all shared memory with no vIOMMU" and T=1 means "all memory with a vIOMMU". This is quite different from "the VM can decide whether it wants private or shared memory", because it kind of can't. The entire device is either T=0/1 and that is that. Jason
