On Wed, Feb 13, 2013 at 03:47:04PM +0000, Marc Zyngier wrote:
> At the moment, we're killing the i-cache each time we map a page
> in stage-2, as we have no idea if this page is used as a data or
> code page.
>
> In order to preserve the cache a bit more, mark all pages with the
> XN flag. If we get an exec permission fault, invalidate the i-cache
> and clear the XN bit.
>
> We end up getting more faults, but we preserve the i-cache.
>
> Signed-off-by: Marc Zyngier <marc.zyngier@xxxxxxx>
> ---
> arch/arm/include/asm/kvm_mmu.h | 5 +++++
> arch/arm/include/asm/pgtable.h | 2 +-
> arch/arm/kvm/mmu.c | 7 +++++--
> 3 files changed, 11 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/include/asm/kvm_mmu.h b/arch/arm/include/asm/kvm_mmu.h
> index 257e2bc..dbfeea1 100644
> --- a/arch/arm/include/asm/kvm_mmu.h
> +++ b/arch/arm/include/asm/kvm_mmu.h
> @@ -95,6 +95,11 @@ static inline void kvm_set_s2pte_writable(pte_t *pte)
> pte_val(*pte) |= L_PTE_S2_RDWR;
> }
>
> +static inline void kvm_set_s2pte_exec(pte_t *pte)
> +{
> + pte_val(*pte) &= ~L_PTE_XN;
> +}
> +
> struct kvm;
>
> static inline void coherent_icache_guest_page(struct kvm *kvm, gfn_t gfn)
> diff --git a/arch/arm/include/asm/pgtable.h b/arch/arm/include/asm/pgtable.h
> index f30ac3b..0bb73cc 100644
> --- a/arch/arm/include/asm/pgtable.h
> +++ b/arch/arm/include/asm/pgtable.h
> @@ -87,7 +87,7 @@ extern pgprot_t pgprot_s2_device;
> #define PAGE_KERNEL_EXEC pgprot_kernel
> #define PAGE_HYP _MOD_PROT(pgprot_kernel, L_PTE_HYP)
> #define PAGE_HYP_DEVICE _MOD_PROT(pgprot_hyp_device, L_PTE_HYP)
> -#define PAGE_S2 _MOD_PROT(pgprot_s2, L_PTE_S2_RDONLY)
> +#define PAGE_S2 _MOD_PROT(pgprot_s2, L_PTE_S2_RDONLY | L_PTE_XN)
> #define PAGE_S2_DEVICE _MOD_PROT(pgprot_s2_device, L_PTE_USER | L_PTE_S2_RDONLY)
>
> #define __PAGE_NONE __pgprot(_L_PTE_DEFAULT | L_PTE_RDONLY | L_PTE_XN | L_PTE_NONE)
> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> index 58a45d1..d727a4f 100644
> --- a/arch/arm/kvm/mmu.c
> +++ b/arch/arm/kvm/mmu.c
> @@ -543,8 +543,6 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
> if (is_error_pfn(pfn))
> return -EFAULT;
>
> - coherent_icache_guest_page(vcpu->kvm, gfn);
> -
> spin_lock(&vcpu->kvm->mmu_lock);
> if (mmu_notifier_retry(vcpu->kvm, mmu_seq))
> goto out_unlock;
> @@ -560,6 +558,11 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
> kvm_set_pfn_dirty(pfn);
> }
>
> + if (fault_type == KVM_EXEC_FAULT) {
> + kvm_set_s2pte_exec(&new_pte);
> + coherent_icache_guest_page(vcpu->kvm, gfn);
> + }
> +
> stage2_set_pte_at(vcpu->kvm, fault_ipa, ptep, new_pte);
>
> out_unlock:
> --
> 1.8.1.2
>
This one looks fine, but I'd like to see some more results using real
workloads on both guest/host before we apply this, as we have discussed.
Thanks,
-Christoffer
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
