RfL branch for PCI bindings

RfL branch for PCI bindings

[Danilo Krummrich]

Hi Andreas and Miguel,

I'd like to come back to our discussion from the last "Rust for DRM" meeting regarding a
common "rust-pci" branch.

As discussed, in the context of the Nova project we are interested in using those bindings,
contribute some more (also in the context of Philipp's recent PCI work) and help moving the
code upstream.

With Nova we should have a driver project which should justify further efforts to upstream
the code.

- Danilo

Re: RfL branch for PCI bindings

[Greg KH]

On Thu, Feb 01, 2024 at 04:02:17PM +0100, Danilo Krummrich wrote:
> Hi Andreas and Miguel,
> 
> I'd like to come back to our discussion from the last "Rust for DRM" meeting regarding a
> common "rust-pci" branch.
> 
> As discussed, in the context of the Nova project we are interested in using those bindings,
> contribute some more (also in the context of Philipp's recent PCI work) and help moving the
> code upstream.
> 
> With Nova we should have a driver project which should justify further efforts to upstream
> the code.

Patches anywhere?  Can them be submitted again please as I think the
last time I saw them, they needed some work.

thanks,

greg k-h

Re: RfL branch for PCI bindings

[Miguel Ojeda]

On Thu, Feb 1, 2024 at 4:02 PM Danilo Krummrich <dakr@redhat.com> wrote:
>
> I'd like to come back to our discussion from the last "Rust for DRM" meeting regarding a
> common "rust-pci" branch.
>
> As discussed, in the context of the Nova project we are interested in using those bindings,
> contribute some more (also in the context of Philipp's recent PCI work) and help moving the
> code upstream.
>
> With Nova we should have a driver project which should justify further efforts to upstream
> the code.

Thanks Danilo!

My understanding from the meeting was that Andreas kindly agreed to
maintain a `rust-pci` branch in a compilable state (i.e. rebasing it
on top of e.g. `rust-next`). I am happy to create the branch already
if Andreas is ready, add it to the website, etc.

Or if you (Danilo) prefer to drive the branch instead (e.g. because
you may have more bandwidth), that is also fine.

Cheers,
Miguel

Re: RfL branch for PCI bindings

[Andreas Hindborg]

Miguel Ojeda <miguel.ojeda.sandonis@gmail.com> writes:

> On Thu, Feb 1, 2024 at 4:02 PM Danilo Krummrich <dakr@redhat.com> wrote:
>>
>> I'd like to come back to our discussion from the last "Rust for DRM" meeting regarding a
>> common "rust-pci" branch.
>>
>> As discussed, in the context of the Nova project we are interested in using those bindings,
>> contribute some more (also in the context of Philipp's recent PCI work) and help moving the
>> code upstream.
>>
>> With Nova we should have a driver project which should justify further efforts to upstream
>> the code.
>
> Thanks Danilo!
>
> My understanding from the meeting was that Andreas kindly agreed to
> maintain a `rust-pci` branch in a compilable state (i.e. rebasing it
> on top of e.g. `rust-next`). I am happy to create the branch already
> if Andreas is ready, add it to the website, etc.
>
> Or if you (Danilo) prefer to drive the branch instead (e.g. because
> you may have more bandwidth), that is also fine.

Exactly. I am currently keeping the patches alive for the Rust NVMe
driver that lives here [1], latest rebase here [2].

I am not the original author of the Rust PCI patches. As far as I know,
no one else is maintaining a copy of these patches on top of mainline
(if there was, I would just use their work).

I volunteer to extract these patches from the nvme tree into a separate
pci tree and keep that rebased on mainline + accept fixes and features
into that tree.

We can also do it the other way round and I would just rebase nvme on
top of your pci tree, that is up to you.

I don't have any plans of sending these patches upstream for now, since
the NVMe driver is not welcome upstream. If/when Nova or another Rust
project depending on pci is ready to merge, I would be happy to help
upstream the pci patches, if that makes sense. Again, you could also do
it yourself, that is completely up to you. Or perhaps the original
authors are interested in helping with this - I do not know since it
never came up.

We could start out in one configuration and then change later down the
road if it becomes clear that another config would be more beneficial.

Best regards,
Andreas


[1] https://rust-for-linux.com/nvme-driver
[2] https://github.com/metaspace/linux/commits/nvme-6.7/

Re: RfL branch for PCI bindings

[Miguel Ojeda]

On Fri, Feb 2, 2024 at 11:40 AM Andreas Hindborg <a.hindborg@samsung.com> wrote:
>
> We could start out in one configuration and then change later down the
> road if it becomes clear that another config would be more beneficial.

+1, let's start with what we discussed -- we can change the
arrangement whenever needed.

I have created the branch in our repository:

    https://github.com/Rust-for-Linux/linux/tree/rust-pci

And added it to the website:

    https://rust-for-linux.com/branches#rust-pci

Please feel free to push to the branch as you see fit -- thanks Andreas!

Cheers,
Miguel

Re: RfL branch for PCI bindings

[Danilo Krummrich]

On 2/2/24 11:40, Andreas Hindborg wrote:
> 
> Miguel Ojeda <miguel.ojeda.sandonis@gmail.com> writes:
> 
>> On Thu, Feb 1, 2024 at 4:02 PM Danilo Krummrich <dakr@redhat.com> wrote:
>>>
>>> I'd like to come back to our discussion from the last "Rust for DRM" meeting regarding a
>>> common "rust-pci" branch.
>>>
>>> As discussed, in the context of the Nova project we are interested in using those bindings,
>>> contribute some more (also in the context of Philipp's recent PCI work) and help moving the
>>> code upstream.
>>>
>>> With Nova we should have a driver project which should justify further efforts to upstream
>>> the code.
>>
>> Thanks Danilo!
>>
>> My understanding from the meeting was that Andreas kindly agreed to
>> maintain a `rust-pci` branch in a compilable state (i.e. rebasing it
>> on top of e.g. `rust-next`). I am happy to create the branch already
>> if Andreas is ready, add it to the website, etc.
>>
>> Or if you (Danilo) prefer to drive the branch instead (e.g. because
>> you may have more bandwidth), that is also fine.
> 
> Exactly. I am currently keeping the patches alive for the Rust NVMe
> driver that lives here [1], latest rebase here [2].
> 
> I am not the original author of the Rust PCI patches. As far as I know,
> no one else is maintaining a copy of these patches on top of mainline
> (if there was, I would just use their work).
> 
> I volunteer to extract these patches from the nvme tree into a separate
> pci tree and keep that rebased on mainline + accept fixes and features
> into that tree.

Thanks Andreas, very much appreciated! I think Philipp will start working
with this code soon.

> 
> We can also do it the other way round and I would just rebase nvme on
> top of your pci tree, that is up to you.
> 
> I don't have any plans of sending these patches upstream for now, since
> the NVMe driver is not welcome upstream. If/when Nova or another Rust
> project depending on pci is ready to merge, I would be happy to help
> upstream the pci patches, if that makes sense. Again, you could also do
> it yourself, that is completely up to you. Or perhaps the original
> authors are interested in helping with this - I do not know since it
> never came up.

The idea is to get an early version of Nova upstream and continue the
development upstream. Hence, once  ready, we do intend to upstream the code
in this context. Of course, your help with that will be very welcome.

- Danilo

> 
> We could start out in one configuration and then change later down the
> road if it becomes clear that another config would be more beneficial.
> 
> Best regards,
> Andreas
> 
> 
> [1] https://rust-for-linux.com/nvme-driver
> [2] https://github.com/metaspace/linux/commits/nvme-6.7/
> 

Re: RfL branch for PCI bindings

[Danilo Krummrich]

On 2/3/24 19:31, Miguel Ojeda wrote:
> On Fri, Feb 2, 2024 at 11:40 AM Andreas Hindborg <a.hindborg@samsung.com> wrote:
>>
>> We could start out in one configuration and then change later down the
>> road if it becomes clear that another config would be more beneficial.
> 
> +1, let's start with what we discussed -- we can change the
> arrangement whenever needed.
> 
> I have created the branch in our repository:
> 
>      https://github.com/Rust-for-Linux/linux/tree/rust-pci
> 
> And added it to the website:
> 
>      https://rust-for-linux.com/branches#rust-pci
> 
> Please feel free to push to the branch as you see fit -- thanks Andreas!

+1

And thanks Miguel for setting things up!

@Andreas: Please let us know once you pushed the patches.

- Danilo

> 
> Cheers,
> Miguel
> 

Re: RfL branch for PCI bindings

[Andreas Hindborg (Samsung)]

Danilo Krummrich <dakr@redhat.com> writes:

> On 2/3/24 19:31, Miguel Ojeda wrote:
>> On Fri, Feb 2, 2024 at 11:40 AM Andreas Hindborg <a.hindborg@samsung.com> wrote:
>>>
>>> We could start out in one configuration and then change later down the
>>> road if it becomes clear that another config would be more beneficial.
>> +1, let's start with what we discussed -- we can change the
>> arrangement whenever needed.
>> I have created the branch in our repository:
>>      https://github.com/Rust-for-Linux/linux/tree/rust-pci
>> And added it to the website:
>>      https://rust-for-linux.com/branches#rust-pci
>> Please feel free to push to the branch as you see fit -- thanks Andreas!
>
> +1
>
> And thanks Miguel for setting things up!
>
> @Andreas: Please let us know once you pushed the patches.

They are up. Please not that they are not high quality, lots of TODO and
missing safety comments.

NVMe rebased on top is here [1]

Let me know if you have any questions!

Best regards,
Andreas


[1] https://github.com/metaspace/linux/tree/rnvme-for-6.8