Re: [RFC PATCH v2 5/6] tpm: add SNP SVSM vTPM driver

[Jason Gunthorpe <jgg@ziepe.ca>]

On Mon, Mar 03, 2025 at 05:19:05PM +0100, Stefano Garzarella wrote:
> On Fri, Feb 28, 2025 at 08:28:19PM -0400, Jason Gunthorpe wrote:
> > On Fri, Feb 28, 2025 at 06:07:19PM +0100, Stefano Garzarella wrote:
> > > +/*
> > > + * tpm_svsm_remove() lives in .exit.text. For drivers registered via
> > > + * module_platform_driver_probe() this is ok because they cannot get unbound
> > > + * at runtime. So mark the driver struct with __refdata to prevent modpost
> > > + * triggering a section mismatch warning.
> > > + */
> > 
> > ??? Is that really true? I didn't know that
> 
> I initially followed drivers/virt/coco/sev-guest/sev-guest.c to figure out
> how to clean a driver registered with module_platform_driver_probe(), then I
> saw that pattern with the same comment is used in several other drivers.
> 
> > 
> > I thought you could unbind anything using /sys/../unbind?
> 
> I can't see `unbind` for this driver:
> 
>   $ ls /sys/bus/platform/drivers/tpm-svsm/
>   module	tpm-svsm  uevent
> 
> While I can see it for example for others like fw_cfg:

Wow, I didn't know that could be done

> BTW I can unload the `tpm-svsm` module.

Unload the module and implicitly unbound the driver

But not manually unbind the driver?? Huh? That seems pretty wrong..

> Loading it again will cause issues if I don't have a remove function
> that calls tpm_chip_unregister().

You definately need the remove function call doing what you have, it
is just surprising to me that there is a case where you can statically
know it is not callable..

Jason