Re: [BUG] ARM64 regression: NULL pointer dereference in arm_smccc_version_init+0x90/0x1ac

[Will Deacon <will@kernel.org>]

On Fri, Sep 26, 2025 at 01:36:34PM +0300, Dan Carpenter wrote:
> On Thu, Mar 13, 2025 at 10:08:36PM +0000, Will Deacon wrote:
> > On Thu, Feb 06, 2025 at 05:57:41PM +0100, Emanuele Rocca wrote:
> > > On 2025-02-06 01:11, Will Deacon wrote:
> > > > In your screenshot, x6 looks like it's retained its poison value, but
> > > > that was the register being corrupted in the initial report. Maybe you
> > > > could share the diff you made?
> > > 
> > > Sure, please see [1]. A kernel built with that patch crashes and all
> > > registers retain their poison values.
> > 
> > Okey doke, at least it sounds like the firmware isn't taking advantage
> > of the broken spec, then.
> > 
> > In any case, I think the right way forward here is to use the
> > out-of-line 1.2 helpers whenever we can (the spectre mitigations can
> > stay as they are).
> > 
> > Will
> > 
> 
> Any updates on this?  I guess it only affects gcc13?  Clang seems to boot
> okay.

I don't recall seeing a patch but we should just move to the out-of-line
SMCCC helpers wherever we can. The inline macros are fiddly to use,
especially when you have to pad things out with zeroes at the caller to
prevent registers from getting corrupted.

Will