On Fri, Jul 03, 2026 at 08:15:49PM +1000, Alexey Kardashevskiy wrote:

On 3/7/26 00:47, Jason Gunthorpe wrote:

quoted

On Thu, Jul 02, 2026 at 10:25:16AM +1000, Alexey Kardashevskiy wrote:

quoted

quoted

quoted

not externally available so I'll have to trick the DMA layer into
using SWIOTLB (which is still all shared, right?) as I specifically
want to skip page conversions. Setting low DMA mask won't guarantee
that the DMA layer won't allocate a page outside of SWIOTLB and
convert it. Manually do

Why so particular?

aahhh I missed "pre-". I need a way to get pre-shared pages for my
sev-guest activities.

It sounds to me like you don't, what you are worried about is
optimizing the dma_alloc_coherent flow to avoid fragmentation and all
CC architetures require this optimization.

It is not about fragmentation, it is about unwanted page state
changes when few pages are shared for a very short time.

That's really basically the same thing. The responsiblity for
optimizing the shared/private conversion lies with the DMA API not the
callers.

When I needed these pages, I could: get a private page, convert,
(possibly trigger RMP and IO PDE smashing if that 4K is backed with
2M), do the VM<->HV communication, convert back to private (and
possibly unsmash the page).

IMHO it would be better to use something like dma_pool to amortize the
coherent allocations instead of trying to hack with swiotlb.

Jason