[PATCH] ARM64: KVM: Nuke EL2 mode tlbs before enabling MMU for EL2

[Marc Zyngier]

On Wed, Jul 30 2014 at 12:13:19 pm BST, Pranavkumar Sawargaonkar <pranavkumar at linaro.org> wrote:
> Hi Marc,
>
> On 30 July 2014 16:30, Marc Zyngier <marc.zyngier at arm.com> wrote:
>
>     Hi Pranav,
>    
>     On Wed, Jul 30 2014 at 10:46:47 am BST, Pranavkumar Sawargaonkar <
>     pranavkumar at linaro.org> wrote:
>     > X-Gene u-boot runs in EL2 mode with MMU enabled hence we might
>     > have stale EL2 tlb enteris when we enable EL2 MMU on each host CPU.
>     >
>     > This can happen on any ARM64 board running bootloader in EL2
>     > with MMU enabled.
>     >
>     > This patch ensures that we flush all EL2 TLBs on each host CPU
>     > before enabling EL2 MMU. This is very similar to what we do
>     > for EL1 TLBs before enabling EL1 MMU.
>     > (Refer __cpu_setup() of arch/arm64/mm/proc.S)
>     >
>     > Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar at linaro.org>
>     > Signed-off-by: Anup Patel <anup.patel at linaro.org>
>     > ---
>     >  arch/arm64/kvm/hyp-init.S |    4 ++++
>     >  1 file changed, 4 insertions(+)
>     >
>     > diff --git a/arch/arm64/kvm/hyp-init.S b/arch/arm64/kvm/hyp-init.S
>     > index d968796..c319116 100644
>     > --- a/arch/arm64/kvm/hyp-init.S
>     > +++ b/arch/arm64/kvm/hyp-init.S
>     > @@ -80,6 +80,10 @@ __do_hyp_init:
>     >       msr     mair_el2, x4
>     >       isb
>     >
>     > +     /* Invalidate the stale TLBs from Bootloader */
>     > +     tlbi    alle2
>     > +     dsb     sy
>     > +
>     >       mrs     x4, sctlr_el2
>     >       and     x4, x4, #SCTLR_EL2_EE   // preserve endianness of EL2
>     >       ldr     x5, =SCTLR_EL2_FLAGS
>    
>     This looks like a valid fix. But I wonder if it wouldn't be better
>     located in kernel/head.S::el2_setup?
>
> Sure, I can try this on one of my flaky setup where I see this issue :)
> One question is , if someone tries to do load/unload KVM module frequently
> then do "__do_hyp_init" will require the flushing again ?

Well, we can't really "unload" KVM, it is not a module. But CPU hotplug
could trigger something similar (though I'd expect it to come back
through el2_setup anyway).

Anyway, we can probably keep it in KVM, and revisit that later if that
makes sense.

>  
>
>     Also, the 32bit code seems to suffer from the same issue. Care to fix it
>     as well?
>
> I am fine with this but as I do not have arm32 board to test it
> somebody will need to add tested-by.

Sure. Fire away!

	M.
-- 
Jazz is not dead. It just smells funny.