Re: [PATCH v2 1/2] rust: add initial scatterlist bindings

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

Hi Abdiel,

On Wed Jul 9, 2025 at 8:59 PM JST, Abdiel Janulgue wrote:
> Hi,
>
> On 03/07/2025 10:03, Alexandre Courbot wrote:
>> On Wed Jul 2, 2025 at 11:37 AM JST, Alexandre Courbot wrote:
>>> Thoughts? If Abdiel is comfortable with this I can submit a v3 with this design
>>> for review (putting myself as co-developer), on which Abdiel could then keep
>>> iterating, as I suspect this would be easier to understand than this long email
>>> :).
>> 
>> Figured I could just as well share the code with you and save both of us
>> some time. ^_^;
>> 
>> The top commit of this branch contains the proposal discussed:
>> 
>> https://github.com/Gnurou/linux/tree/scatterlists
>> 
>> The sample code has been updated to add dummy examples for the 3
>> use-cases discussed (reference to an existing `sg_table`, refcounted
>> reference, and owned data).
>> 
>> There are still things missing, including the typestate on `SGEntry`, as
>> it wasn't necessary to demonstrate the basic idea.
>> 
>> Note also that if we decide to only support DMA-mapped SG-entries, we
>> can remove a bunch of code, including the one that maps a `SGTable` if
>> the backing type implements `BorrowMut`.
>> 
>> For your consideration. :) Please feel free to take and use anything you
>> find useful.
>
> Sorry for the delay, just came back from a week vacation.

Welcome back!

>
> Regarding your question regarding why I dropped type-state, I thought 
> the general consensus here is to drop this approach:
>
> https://lore.kernel.org/lkml/DAC20AXGABW2.X147X4JPMRBS@nvidia.com/
>
> I think I might have misunderstood after re-reading it now. What you 
> probably meant was to have no intermediate state from initialized to 
> mapped to create the sg_table?

Sorry, that was poorly phrased - and yes this is what I meant. Since we
have no use for unmapped SG tables at the moment, let's start with this
simple and working approach. We can always add more states/stuff later
if the need arises.

>
> Anyways, thanks for the code above I'll look into this in detail and 
> pick bits to integrate into v3 :)

Looking forward to it! Please don't hesitate to reach out if anything is
unclear.