Re: [PATCH] swiotlb: Allow swiotlb to live at pre-defined address

[Mark Rutland]

On Thu, Mar 26, 2020 at 06:11:31PM +0100, Alexander Graf wrote:
> On 26.03.20 18:05, Christoph Hellwig wrote:
> > 
> > On Thu, Mar 26, 2020 at 05:29:22PM +0100, Alexander Graf wrote:
> > > The swiotlb is a very convenient fallback mechanism for bounce buffering of
> > > DMAable data. It is usually used for the compatibility case where devices
> > > can only DMA to a "low region".
> > > 
> > > However, in some scenarios this "low region" may be bound even more
> > > heavily. For example, there are embedded system where only an SRAM region
> > > is shared between device and CPU. There are also heterogeneous computing
> > > scenarios where only a subset of RAM is cache coherent between the
> > > components of the system. There are partitioning hypervisors, where
> > > a "control VM" that implements device emulation has limited view into a
> > > partition's memory for DMA capabilities due to safety concerns.
> > > 
> > > This patch adds a command line driven mechanism to move all DMA memory into
> > > a predefined shared memory region which may or may not be part of the
> > > physical address layout of the Operating System.
> > > 
> > > Ideally, the typical path to set this configuration would be through Device
> > > Tree or ACPI, but neither of the two mechanisms is standardized yet. Also,
> > > in the x86 MicroVM use case, we have neither ACPI nor Device Tree, but
> > > instead configure the system purely through kernel command line options.
> > > 
> > > I'm sure other people will find the functionality useful going forward
> > > though and extend it to be triggered by DT/ACPI in the future.
> > 
> > I'm totally against hacking in a kernel parameter for this.  We'll need
> > a proper documented DT or ACPI way.
> 
> I'm with you on that sentiment, but in the environment I'm currently looking
> at, we have neither DT nor ACPI: The kernel gets purely configured via
> kernel command line. For other unenumerable artifacts on the system, such as
> virtio-mmio platform devices, that works well enough and also basically
> "hacks a kernel parameter" to specify the system layout.

On the arm64 front, you'd *have* to pass a DT to the kernel (as that's
where we get the command line from), and we *only* discover memory
from the DT or EFI memory map, so the arguments above aren't generally
applicable. You can enumerate virtio-mmio devices from DT, also.

Device-specific constraints on memory should really be described in a
per-device fashion in the FW tables so that the OS can decide how to
handle them. Just becuase one device can only access memory in a
specific 1MiB window doesn't mean all other should be forced to share
the same constraint. I think that's what Christoph was alluding to.

Thanks,
Mark.