On Mon, Aug 28, 2023 at 04:31:21PM +0100, Marc Zyngier wrote:
> Marek reports that his RPi4 spits out a warning at boot time,
> right at the point where the GICv2 virtual CPU interface gets
> mapped.
>
> Upon investigation, it seems that we never return the allocated
> VA and use whatever was on the stack at this point. Yes, this
> is good stuff, and Marek was pretty lucky that he ended-up with
> a VA that intersected with something that was already mapped.
>
> On my setup, this random value is plausible enough for the mapping
> to take place. Who knows what happens...
>
> Cc: Vincent Donnefort <vdonnefort@xxxxxxxxxx>
> Fixes: f156a7d13fc3 ("KVM: arm64: Remove size-order align in the nVHE hyp private VA range")
> Reported-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Tested-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> Link: https://lore.kernel.org/r/79b0ad6e-0c2a-f777-d504-e40e8123d81d@xxxxxxxxxxx
Having a hard time reproducing the issue, but clearly that set is missing from
the original patch!
Sorry about that extra work.
> ---
> arch/arm64/kvm/mmu.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 11c1d786c506..50be51cc40cc 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -652,6 +652,9 @@ int hyp_alloc_private_va_range(size_t size, unsigned long *haddr)
>
> mutex_unlock(&kvm_hyp_pgd_mutex);
>
> + if (!ret)
> + *haddr = base;
> +
> return ret;
> }
>
> --
> 2.34.1
>
