Re: [PATCH v3 4/5] arm64: initialise SCTLR2_EL1 at cpu_soft_restart()

[Dave Martin <Dave.Martin@arm.com>]

Hi,

On Wed, Aug 20, 2025 at 06:32:34PM +0100, Yeoreum Yun wrote:
> Hi Dave,
> 
> > On Wed, Aug 13, 2025 at 01:01:17PM +0100, Yeoreum Yun wrote:
> > > SCTLR2_EL1 register is optional starting from ARMv8.8/ARMv9.3,
> > > and becomes mandatory from ARMv8.9/ARMv9.4
> > > and serveral architectural feature are controled by bits in
> > > these registers.
> > >
> > > Before, launching new kernel via kexec, initialise SCTLR2_EL1 explicitly.
> > >
> > > Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
> > > ---
> > >  arch/arm64/kernel/cpu-reset.S | 6 ++++++
> > >  1 file changed, 6 insertions(+)
> > >
> > > diff --git a/arch/arm64/kernel/cpu-reset.S b/arch/arm64/kernel/cpu-reset.S
> > > index c87445dde674..123564af345b 100644
> > > --- a/arch/arm64/kernel/cpu-reset.S
> > > +++ b/arch/arm64/kernel/cpu-reset.S
> > > @@ -37,6 +37,12 @@ SYM_TYPED_FUNC_START(cpu_soft_restart)
> > >  	 * regime if HCR_EL2.E2H == 1
> > >  	 */
> > >  	msr	sctlr_el1, x12
> > > +
> > > +alternative_if ARM64_HAS_SCTLR2
> > > +	mov_q	x12, INIT_SCTLR2_EL1
> > > +	msr_s	SYS_SCTLR2_EL1, x12
> > > +alternative_else_nop_endif
> > > +
> >
> > It would be better to do this based on the ID regs.
> >
> > Although the previous kernel _shouldn't_ have used SCTLR2 if the
> > capability ARM64_HAS_SCTLR2 did not get enabled, it would be better to
> > enforce a clean state here for the new kernel.
> >
> > If so, maybe one of the macros that you already defined can be used
> > here?  (But it's also fine to open-code it.)
> 
> But cpu_soft_restart() can be called before capability is enabled?
> I think this function is called after "capability" setup,
> Was it good to use alternative than check the ID register?

What I meant is that we should reset SCTLR2_EL1 here even if the
ARM64_HAS_SCTLR2 capability is not set.

The cpu_soft_restart() code has responsibilities similar to those of a
bootloader.  We want to put the CPU into a known state, irrespective of
how the current kernel has been using the CPU.

For one thing, we come through this path when booting a crash kernel if
the current kernel panicked.  So we should avoid making too many
assumptions about anything being in a sensible state here.

(Your rewrite of this in v4 looks fine.)

> > >  	isb
> > >
> > >  	cbz	x0, 1f				// el2_switch?
> >
> > [...]
> >
> > In the case where the el2_switch argument in non-zero, don't we also
> > need to do something to reinitialise SCTLR2_EL2 after switching back
> > to EL2, in the HVC_SOFT_RESTART handler?
> >
> > Maybe I missed something.
> 
> No. I'm missing to init in NVHE's HVC_SOFT_RESTART handler to clear SCTLR2_EL2.
> 
> Thanks!

I'll take a look at this in v4.

Cheers
---Dave