Re: [resend/standalone PATCH v4] Add auxiliary bus support

[Dan Williams]

On Fri, Dec 4, 2020 at 3:41 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Dec 02, 2020 at 04:54:24PM -0800, Dan Williams wrote:
> > From: Dave Ertman <david.m.ertman@xxxxxxxxx>
> >
> > Add support for the Auxiliary Bus, auxiliary_device and auxiliary_driver.
> > It enables drivers to create an auxiliary_device and bind an
> > auxiliary_driver to it.
> >
> > The bus supports probe/remove shutdown and suspend/resume callbacks.
> > Each auxiliary_device has a unique string based id; driver binds to
> > an auxiliary_device based on this id through the bus.
> >
> > Co-developed-by: Kiran Patil <kiran.patil@xxxxxxxxx>
> > Co-developed-by: Ranjani Sridharan <ranjani.sridharan@xxxxxxxxxxxxxxx>
> > Co-developed-by: Fred Oh <fred.oh@xxxxxxxxxxxxxxx>
> > Co-developed-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > Signed-off-by: Kiran Patil <kiran.patil@xxxxxxxxx>
> > Signed-off-by: Ranjani Sridharan <ranjani.sridharan@xxxxxxxxxxxxxxx>
> > Signed-off-by: Fred Oh <fred.oh@xxxxxxxxxxxxxxx>
> > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > Signed-off-by: Dave Ertman <david.m.ertman@xxxxxxxxx>
> > Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@xxxxxxxxxxxxxxx>
> > Reviewed-by: Shiraz Saleem <shiraz.saleem@xxxxxxxxx>
> > Reviewed-by: Parav Pandit <parav@xxxxxxxxxxxx>
> > Reviewed-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > Reviewed-by: Martin Habets <mhabets@xxxxxxxxxxxxxx>
> > Link: https://lore.kernel.org/r/20201113161859.1775473-2-david.m.ertman@xxxxxxxxx
> > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > ---
> > This patch is "To:" the maintainers that have a pending backlog of
> > driver updates dependent on this facility, and "Cc:" Greg. Greg, I
> > understand you have asked for more time to fully review this and apply
> > it to driver-core.git, likely for v5.12, but please consider Acking it
> > for v5.11 instead. It looks good to me and several other stakeholders.
> > Namely, stakeholders that have pressure building up behind this facility
> > in particular Mellanox RDMA, but also SOF, Intel Ethernet, and later on
> > Compute Express Link.
> >
> > I will take the blame for the 2 months of silence that made this awkward
> > to take through driver-core.git, but at the same time I do not want to
> > see that communication mistake inconvenience other parties that
> > reasonably thought this was shaping up to land in v5.11.
> >
> > I am willing to host this version at:
> >
> > git://git.kernel.org/pub/scm/linux/kernel/git/djbw/linux tags/auxiliary-bus-for-5.11
> >
> > ...for all the independent drivers to have a common commit baseline. It
> > is not there yet pending Greg's Ack.
> >
> > For example implementations incorporating this patch, see Dave Ertman's
> > SOF series:
> >
> > https://lore.kernel.org/r/20201113161859.1775473-2-david.m.ertman@xxxxxxxxx
> >
> > ...and Leon's mlx5 series:
> >
> > http://lore.kernel.org/r/20201026111849.1035786-1-leon@xxxxxxxxxx
> >
> > PS: Greg I know I promised some review on newcomer patches to help with
> > your queue, unfortunately Intel-internal review is keeping my plate
> > full. Again, I do not want other stakeholder to be waiting on me to
> > resolve that backlog.
>
> Ok, I spent some hours today playing around with this.  I wrote up a
> small test-patch for this (how did anyone test this thing???) and while
> it feels awkward in places, and it feels like there is still way too
> much "boilerplate" code that a user has to write and manage, I don't
> have the time myself to fix it up right now.
>
> So I'll go apply this to my tree, and provide a tag for everyone else to
> be able to pull from for their different development trees so they can
> work on.
>
> I do have 3 follow-on patches that I will send to the list in response
> to this message that I will be applying on top of this patch.  They do
> some minor code formatting changes, as well as change the return type of
> the remove function to make it more future-proof.  That last change will
> require users of this code to change their implementations, but it will
> be obvious what to do as you will get a build warning.
>
> Note, I'm still not comfortable with a few things here.  The
> documentation feels odd, and didn't really help me out in writing any
> test code, which doesn't seem right.  Also the use of strings and '.' as
> part of the api feels awkward, and messy, and of course, totally
> undocumented.
>
> But, as the use of '.' is undocumented, that means we can change it in
> the future!  Because no driver or device name should ever be a user api
> reliant thing, if we come up with a better way to do all of this in the
> future, that shouldn't be a problem to change existing users over to
> this.  So this is a warning to everyone, you CAN NOT depend on the sysfs
> name of a device or bus name for any tool.  If so, your userspace tool
> is broken.
>
> Thanks for everyone in sticking with this, I know it's been a long slog,
> hopefully this will help some driver authors move forward with their
> crazy complex devices :)

To me, the documentation was written, and reviewed, more from the
perspective of "why not open code a custom bus instead". So I can see
after the fact how that is a bit too much theory and justification and
not enough practical application. Before the fact though this was a
bold mechanism to propose and it was not clear that everyone was
grokking the "why" and the tradeoffs.

I also think it was a bit early to identify consistent design patterns
across the implementations and codify those. I expect this to evolve
convenience macros just like other parts of the driver-core gained
over time. Now that it is in though, another pass through the
documentation to pull in more examples seems warranted.