Re: [PATCH 1/2] rust: add initial scatterlist bindings

["Alexandre Courbot" <acourbot@nvidia.com>]

On Fri May 30, 2025 at 11:50 PM JST, Jason Gunthorpe wrote:
> On Fri, May 30, 2025 at 11:44:26PM +0900, Alexandre Courbot wrote:
>
>> I would be fully on board with a simpler design, definitely. The reason
>> why I've tried to keep some doors open is that as you mentioned
>> scatterlist is used in many different ways, and I am not familiar enough
>> with all these uses to draw a line and say "we will never ever need to
>> do that".
>
> I think it would be better to grow as needed. It is hard to speculate.
>
> We also have the new two step DMA API, so it may very well be the only
> use for this stuff is very simple mappings of VVec like things for
> DMA, and maybe this all gets rewritten to use the new DMA API and not
> scatterlist.
>
> Having a rust user facing API that allows for that would be a great
> thing.
>
> IOW I would maybe reframe the task here, it is not to create simple
> naive wrappers around scatterlist but to provide a nice rust API to go
> from VVec/etc to DMA mapping of that VVec/etc.

I like this focus on the practical instead of abstracting the C APIs as
closely as possible. Maybe we have been too focused on the tool rather
than the goal.

So if I understood your idea correctly, this would mean creating the
SGTable and mapping it in one call, eschewing the typestate entirely?
And the `SGTable` would own the backing data, and only release it upon
destruction and unmapping?

I guess the `SGTablePages` (or some renamed variant) would still be useful
to build the list and make sure the core types (e.g. `VVec`) are
ready-to-use with this new API.

One interesting thing to look at after a first version is available
would be a mechanism to ensure only one device (or only the CPU) can
access a buffer that has multiple mappings at any given time, with the
required synchronization performed transparently.

But for now I agree the simple use-case of single-device mapping is a
good way to get started.