Re: [RFC PATCH 1/3] content: Add VIRTIO_F_SWIOTLB to negotiate use of SWIOTLB bounce buffers

[David Woodhouse]

On 4 April 2025 09:23:24 BST, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
>On Fri, Apr 04, 2025 at 08:50:47AM +0100, David Woodhouse wrote:
>> I do agree, this is fundamentally a system issue. In a CoCo model, it's
>> non-trivial for the system to allow *virtual* devices to do "DMA"
>> because that actually means allowing the VMM to access arbitrary guest
>> memory.
>
>
>> So "for the emulated devices, just use a device model that doesn't do
>> arbitrary DMA to system memory" is a nice simple answer, and keeps the
>> guest support restricted to its *own* standalone driver.
>
>It's also one that completely breaks the abstraction. 

Hm? Having a device that simply doesn't *do* any DMA surely doesn't break any system DMA / IOMMU abstractions because it's no longer even relevant to them, which is kind of the point.

Which abstraction did you mean?


> I still don't
>understand what the problem is with having the paravirtualized devices
>on a different part of the virtual PCIe topology so that the stage2
>IOMMU isn't used for them, but instead just the direct mapping or a
>stub viommu that blocks all access.

It can't have a direct mapping because the VMM can't access guest memory. It has to be blocked, which is fine. But that's only part of the picture — then how do you actually get data in/out of the device?

By having on-device memory and not attempting DMA to system memory, perhaps...? :)

>> What's annoying is that this should work out of the box *already* with
>> virtio-mmio and a `restricted-dma-pool` — for systems which aren't
>> afflicted by UEFI/ACPI/PCI as their discovery mechanisms.
>
>Yes.  And the fix is to get the equivalent to restricted-dma-pool into
>UEFI/ACPI.  That gives you a portable and device-independent way to
>describe this limitation, which is much better than hacking around it
>using an odd device model.

It still has the problem that existing drivers in all operating systems produced before 2030 will see the device and try to use it as-is, with no comprehension of this new thing.