Hi Marc,
On Fri, Jan 28, 2022 at 12:18:39PM +0000, Marc Zyngier wrote:
> From: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
>
> So far we were flushing almost the entire universe whenever a VM would
> load/unload the SCTLR_EL1 and the two versions of that register had
> different MMU enabled settings. This turned out to be so slow that it
> prevented forward progress for a nested VM, because a scheduler timer
> tick interrupt would always be pending when we reached the nested VM.
>
> To avoid this problem, we consider the SCTLR_EL2 when evaluating if
> caches are on or off when entering virtual EL2 (because this is the
> value that we end up shadowing onto the hardware EL1 register).
>
> Signed-off-by: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
> Signed-off-by: Jintack Lim <jintack.lim@xxxxxxxxxx>
> Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> ---
> arch/arm64/include/asm/kvm_mmu.h | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/include/asm/kvm_mmu.h b/arch/arm64/include/asm/kvm_mmu.h
> index 81839e9a8a24..1b314b2a69bc 100644
> --- a/arch/arm64/include/asm/kvm_mmu.h
> +++ b/arch/arm64/include/asm/kvm_mmu.h
> @@ -115,6 +115,7 @@ alternative_cb_end
> #include <asm/cache.h>
> #include <asm/cacheflush.h>
> #include <asm/mmu_context.h>
> +#include <asm/kvm_emulate.h>
>
> void kvm_update_va_mask(struct alt_instr *alt,
> __le32 *origptr, __le32 *updptr, int nr_inst);
> @@ -187,7 +188,10 @@ struct kvm;
>
> static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
> {
> - return (vcpu_read_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
> + if (vcpu_is_el2(vcpu))
> + return (__vcpu_sys_reg(vcpu, SCTLR_EL2) & 0b101) == 0b101;
> + else
> + return (vcpu_read_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
Might be more readable if instead of 0b101 KVM would use defines, something
like:
static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
{
- return (vcpu_read_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
+ u64 cache_bits = SCTLR_ELx_M | SCTLR_ELx_C;
+
+ if (vcpu_is_el2(vcpu))
+ return (__vcpu_sys_reg(vcpu, SCTLR_EL2) & cache_bits) == cache_bits;
+ else
+ return (vcpu_read_sys_reg(vcpu, SCTLR_EL1) & cache_bits) == cache_bits;
}
Regardless, it is correct to use vcpu_read_sys_reg() for the SCTLR_EL1 case, as
the most recent register value could be on the CPU in the VHE case, instead of
being in memory, like it's always the case with the SCTLR_EL2 register:
Reviewed-by: Alexandru Elisei <alexandru.elisei@xxxxxxx>
Thanks,
Alex
> }
>
> static inline void __clean_dcache_guest_page(void *va, size_t size)
> --
> 2.30.2
>
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
