[BUG] New arm scmi check in linux-next causing rk3568 not to boot due to firmware bug

[Sudeep Holla]

+ Cristian

Hi Nicolas,

Thanks for the formal report.

On Wed, May 04, 2022 at 02:49:07PM +0200, Nicolas Frattaroli wrote:
> Good day,
> 
> a user on the #linux-rockchip channel on the Libera.chat IRC network
> reported that their RK3568 was no longer getting a CPU and GPU clock
> from scmi and consequently not booting when using linux-next. This
> was bisected down to the following commit:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/drivers/firmware/arm_scmi/base.c?h=next-20220503&id=3b0041f6e10e5bdbb646d98172be43e88734ed62
> 
> The error message in the log is as follows:
> 
> arm-scmi firmware:scmi: Malformed reply - real_sz:8  calc_sz:4, t->rx.len is 12, sizeof(u32) is 4, loop_num_ret is 3
> 
> The rockchip firmware (bl31) being used was v1.32, from here:
> 
> https://github.com/JeffyCN/rockchip_mirrors/blob/rkbin/bin/rk35/rk3568_bl31_v1.32.elf
>

So this platform is not supported in upstream TF-A like its predecessors ?

> This seems like a non-fatal firmware bug, for which a kernel workaround is
> certainly possible, but it would be good if rockchip could fix this in their
> firmware.
>

Indeed, we added this check finding issue in one of our tests. Luckily
it helped to unearth the same issue on this platform, but due to the
nature of its f/w release, it is bit unfortunate that it can't be fixed
easily and quickly. But I really wish this gets fixed in the firmware.
Are there any other f/w bugs reported so far ? If so how are they fixed
as I don't expect all such bugs can be worked around in the kernel though
this might be. I would like to hear details there if possible.

> The user going by "amazingfate" reported that commenting out the
>   ret = -EPROTO; break;
> fixes the issue for them.
>

Sure, or we could relax the check as calc_sz <= real_sz or something so
that the reverse is still caught and handled as OS might read junk data in
the later case.

> I'm writing here to get the discussion started on how we can resolve this
> before the Linux 5.19 release.
>

Agreed, I just sent by pull request for this literally few hours ago.

> Sudeep Holla has already told me they'll gladly add a workaround before
> the 5.19 release, but would rather see this fixed in the vendor firmware
> first. Would rockchip be able and willing to fix it and publish a new
> bl31 for rk3568?
>

Indeed and as mentioned above details on how other such f/w bugs are dealt
in general esp that the firmware is blob release and one can't fix it easily.
Do we have a bugzilla kind of setup to report and get the bugs fixed ?

-- 
Regards,
Sudeep