[PATCH v5 33/36] KVM: arm64: Wrap the host with a stage 2

[Quentin Perret]

On Tuesday 16 Mar 2021 at 16:16:18 (+0100), Mate Toth-Pal wrote:
> On 2021-03-16 15:29, Quentin Perret wrote:
> > On Tuesday 16 Mar 2021 at 12:53:53 (+0000), Quentin Perret wrote:
> > > On Tuesday 16 Mar 2021 at 13:28:42 (+0100), Mate Toth-Pal wrote:
> > > > Changing the value of MT_S2_FWB_NORMAL to 7 would change this behavior, and
> > > > the resulting memory type would be device.
> > > 
> > > Sounds like the correct fix here -- see below.
> > 
> > Just to clarify this, I meant this should be the configuration for the
> > host stage-2. We'll want to keep the existing behaviour for guests I
> > believe.
> 
> I Agree.

OK, so the below seems to boot on my non-FWB-capable hardware and should
fix the issue. Could you by any chance give it a spin?

diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
index b93a2a3526ab..b2066bd03ca2 100644
--- a/arch/arm64/include/asm/kvm_pgtable.h
+++ b/arch/arm64/include/asm/kvm_pgtable.h
@@ -76,10 +76,11 @@ struct kvm_pgtable {

 /**
  * enum kvm_pgtable_prot - Page-table permissions and attributes.
- * @KVM_PGTABLE_PROT_X:                Execute permission.
- * @KVM_PGTABLE_PROT_W:                Write permission.
- * @KVM_PGTABLE_PROT_R:                Read permission.
- * @KVM_PGTABLE_PROT_DEVICE:   Device attributes.
+ * @KVM_PGTABLE_PROT_X:                        Execute permission.
+ * @KVM_PGTABLE_PROT_W:                        Write permission.
+ * @KVM_PGTABLE_PROT_R:                        Read permission.
+ * @KVM_PGTABLE_PROT_DEVICE:           Device attributes.
+ * @KVM_PGTABLE_PROT_S2_NOFWB:         Don't enforce Normal-WB with FWB.
  */
 enum kvm_pgtable_prot {
        KVM_PGTABLE_PROT_X                      = BIT(0),
@@ -87,6 +88,8 @@ enum kvm_pgtable_prot {
        KVM_PGTABLE_PROT_R                      = BIT(2),

        KVM_PGTABLE_PROT_DEVICE                 = BIT(3),
+
+       KVM_PGTABLE_PROT_S2_NOFWB               = BIT(4),
 };

 #define PAGE_HYP               (KVM_PGTABLE_PROT_R | KVM_PGTABLE_PROT_W)
diff --git a/arch/arm64/include/asm/memory.h b/arch/arm64/include/asm/memory.h
index c759faf7a1ff..e695d2e1839d 100644
--- a/arch/arm64/include/asm/memory.h
+++ b/arch/arm64/include/asm/memory.h
@@ -144,13 +144,16 @@
  * Memory types for Stage-2 translation
  */
 #define MT_S2_NORMAL           0xf
+#define MT_S2_WEAK             MT_S2_NORMAL
 #define MT_S2_DEVICE_nGnRE     0x1

 /*
  * Memory types for Stage-2 translation when ID_AA64MMFR2_EL1.FWB is 0001
- * Stage-2 enforces Normal-WB and Device-nGnRE
+ * Stage-2 enforces Normal-WB and Device-nGnRE by default. The 'weak' mode
+ * honors Stage-1 attributes.
  */
 #define MT_S2_FWB_NORMAL       6
+#define MT_S2_FWB_WEAK         7
 #define MT_S2_FWB_DEVICE_nGnRE 1

 #ifdef CONFIG_ARM64_4K_PAGES
diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
index dd03252b9574..1ff72babe565 100644
--- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
+++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
@@ -214,6 +214,8 @@ static int host_stage2_idmap(u64 addr)

        if (is_memory)
                prot |= KVM_PGTABLE_PROT_X;
+       else
+               prot |= KVM_PGTABLE_PROT_S2_NOFWB;

        hyp_spin_lock(&host_kvm.lock);
        ret = kvm_pgtable_stage2_find_range(&host_kvm.pgt, addr, prot, &range);
diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
index 3a971df278bd..bd1b8464a537 100644
--- a/arch/arm64/kvm/hyp/pgtable.c
+++ b/arch/arm64/kvm/hyp/pgtable.c
@@ -343,6 +343,9 @@ static int hyp_set_prot_attr(enum kvm_pgtable_prot prot, kvm_pte_t *ptep)
        if (!(prot & KVM_PGTABLE_PROT_R))
                return -EINVAL;

+       if (prot & KVM_PGTABLE_PROT_S2_NOFWB)
+               return -EINVAL;
+
        if (prot & KVM_PGTABLE_PROT_X) {
                if (prot & KVM_PGTABLE_PROT_W)
                        return -EINVAL;
@@ -510,9 +513,18 @@ u64 kvm_get_vtcr(u64 mmfr0, u64 mmfr1, u32 phys_shift)
 static int stage2_set_prot_attr(enum kvm_pgtable_prot prot, kvm_pte_t *ptep)
 {
        bool device = prot & KVM_PGTABLE_PROT_DEVICE;
-       kvm_pte_t attr = device ? PAGE_S2_MEMATTR(DEVICE_nGnRE) :
-                           PAGE_S2_MEMATTR(NORMAL);
+       bool nofwb = prot & KVM_PGTABLE_PROT_S2_NOFWB;
        u32 sh = KVM_PTE_LEAF_ATTR_LO_S2_SH_IS;
+       kvm_pte_t attr;
+
+       WARN_ON(nofwb && device);
+
+       if (device)
+               attr = PAGE_S2_MEMATTR(DEVICE_nGnRE);
+       else if (nofwb)
+               attr = PAGE_S2_MEMATTR(WEAK);
+       else
+               attr = PAGE_S2_MEMATTR(NORMAL);

        if (!(prot & KVM_PGTABLE_PROT_X))
                attr |= KVM_PTE_LEAF_ATTR_HI_S2_XN;