[PATCH 0/8] Make SCMI transport as standalone drivers

[Etienne CARRIERE]

Hi Cristian, Peng,

On Thursday, July 11, 2024, Cristian Marussi worte:
> On Thu, Jul 11, 2024 at 01:26:16PM +0000, Peng Fan wrote:
> > > Subject: [PATCH 0/8] Make SCMI transport as standalone drivers
> >
> > You may need use V2 here :)
> 
> ...oh damn :P
> 
> > >
> > > Hi all,
> > >
> > > Till now the SCMI transport layer was being built embedded into in the
> > > core SCMI stack.
> > >
> > > Some of these transports, despite being currently part of the main
> > > SCMI module, are indeed also registered with different subsystems like
> > > optee or virtio, and actively probed also by those: this led to a few
> > > awkward and convoluted tricks to properly handle such interactions at
> > > boot time in the SCMI stack.
> > >
> > > Moreover some partner expressed the desire to be able to fully
> > > modularize the transports components.
> > >
> > > This series aim to make all such transports as standalone drivers that
> > > can be optionally loaded as modules.
> > >
> > > In order to do this, at first some new mechanism is introduced to
> > > support this new capability while maintaining, in parallel, the old
> > > legacy embedded transports; then each transport, one by one, is
> > > transitioned to be a standalone driver and finally the old legacy
> > > support for embedded transport is removed.
> > >
> > > Patch [1/8] is a mostly unrelated (but much needed) clean-up from
> > > Peng, which I included in this series to avoid conflicts at merge.
> > >
> > > Patch [2/8] simply collects the existing datagram manipulation helpers
> > > in a pair of function pointers structures, in preparation for later reworks.
> > >
> > > Patch [3/8] adds the bulk of the new logic to the core SCMI stack and
> > > then each existing transport is transitioned to be a standalone driver in
> > > patches 4,5,6,7 while shuffling around the compatibles. (no DT change
> > > is needed of curse for backward compatibility) While doing this I kept
> > > the module authorship in line with the main
> > > author(S) as spitted out by git-blame.
> > >
> > > Finally patch [8/8] removes all the legacy dead code from the core
> > > SCMI stack.
> > >
> > > No new symbol EXPORT has been added.
> > >
> > > The new transport drivers have been tested, as built-in and LKM, as
> > > follows:
> > >
> > > - mailbox on JUNO
> > > - virtio on emulation
> > > - optee on QEMU/optee using Linaro setup
> > >
> > > Exercised using the regular SCMI drivers stack and the SCMI ACS suite,
> > > testing commands, replies, delayed responses and notification.
> > >
> > > Multiple virtual SCMI instances support has been tested too.
> > >
> > > SMC has NOT been tested/exercised at run-time, only compile-tested.
> > > (due to lack of hardware)
> > >
> > > Note that in this new setup, all the probe deferral and retries between
> > > the SCMI core stack and the transports has been removed, since no
> > > more needed.
> > >
> > > Moreover the new drivers have been tested also with a fully
> > > modularized SCMI stack, i.e.:
> > >
> > >   scmi-core.ko + scmi-module.ko + scmi_transport_*.ko [ + vendor
> > > modules ]
> > >
> > > ToBeDone:
> > >  - completely remove any dependency at build time at the Kconfig level
> > > between
> > >    the SCMI core and the transport drivers: so that the transports will be
> > >    dependent only on the related subsystems (optee/virtio/mailbox/smc)
> > >    (easy to be done but maybe it is not worth...)
> > >  - integrate per-platform transport configuration capabilities
> > >    (max_rx_timeout_ms & friends..)
> > >
> > > Based on sudeep/for-next/scmi/updates.
> > >
> > > Any feedback, and especially testing (:D) is welcome.
> > >
> >
> > For the v2 patchset:
> > Tested-by: Peng Fan <peng.fan at nxp.com>  #i.MX95-19x19-EVK
> >
> 
> Thanks a lot for the review and testing,
> 
> Cristian


I've tested this v2 on stm32mp157c-scmi.dts. Using built-in modules
works perfectly.  I've tweaked my platform setup to test the .ko and
modprobe part. It works ok for the probe part but I faced kernel oops
when unloading scmi-module after transport is loaded, used, then unoaded.
The issue I saw is around calls to info->desc->ops->chan_free in
scmi_cleanup_channels(). I wonder if there are some ops that were not
unregistered when transport driver is unloaded.

Aside this, I'll sent few minor comments on few patches of your series.

Best regards,
Etienne