Re: [PATCH] powerpc/pseries: Add shutdown() to vio_driver and vio_bus

From: Michael Ellerman
Date: Mon Apr 19 2021 - 23:42:48 EST

Tyrel Datwyler <tyreld@xxxxxxxxxxxxx> writes:
> On 4/17/21 5:30 AM, Michael Ellerman wrote:
>> Tyrel Datwyler <tyreld@xxxxxxxxxxxxx> writes:
>>> On 4/1/21 5:13 PM, Tyrel Datwyler wrote:
>>>> Currently, neither the vio_bus or vio_driver structures provide support
>>>> for a shutdown() routine.
>>>> Add support for shutdown() by allowing drivers to provide a
>>>> implementation via function pointer in their vio_driver struct and
>>>> provide a proper implementation in the driver template for the vio_bus
>>>> that calls a vio drivers shutdown() if defined.
>>>> In the case that no shutdown() is defined by a vio driver and a kexec is
>>>> in progress we implement a big hammer that calls remove() to ensure no
>>>> further DMA for the devices is possible.
>>>> Signed-off-by: Tyrel Datwyler <tyreld@xxxxxxxxxxxxx>
>>>> ---
>>> Ping... any comments, problems with this approach?
>> The kexec part seems like a bit of a hack.
>> It also doesn't help for kdump, when none of the shutdown code is run.
> If I understand correctly for kdump we have a reserved memory space where the
> kdump kernel is loaded, but for kexec the memory region isn't reserved ahead of
> time meaning we can try and load the kernel over potential memory used for DMA
> by the current kernel.

That's correct.

>> How many drivers do we have? Can we just implement a proper shutdown for
>> them?
> Well that is the end goal. I just don't currently have the bandwidth to do each
> driver myself with a proper shutdown sequence, and thought this was a launching
> off point to at least introduce the shutdown callback to the VIO bus.

Fair enough.

> Off the top of my head we have 3 storage drivers, 2 network drivers, vtpm, vmc,
> pseries_rng, nx, nx842, hvcs, hvc_vio.
> I can drop the kexec_in_progress hammer and just have each driver call remove()
> themselves in their shutdown function. Leave it to each maintainer to decide if
> remove() is enough or if there is a more lightweight quiesce sequence they
> choose to implement.

That's OK, you've convinced me. I'll take it as-is.

Eventually it would be good for drivers to implement shutdown in the
optimal way for their device, but that can be done incrementally.