[PATCH v7 55/68] KVM: arm64: nv: Invalidate TLBs based on shadow S2 TTL-like information

[Marc Zyngier]

In order to be able to make S2 TLB invalidations more performant on NV,
let's use a scheme derived from the ARMv8.4 TTL extension.

If bits [56:55] in the descriptor are non-zero, they indicate a level
which can be used as an invalidation range.

Signed-off-by: Marc Zyngier <maz at kernel.org>
---
 arch/arm64/include/asm/kvm_nested.h |  4 ++
 arch/arm64/kvm/nested.c             | 99 +++++++++++++++++++++++++++++
 arch/arm64/kvm/sys_regs.c           |  9 ++-
 3 files changed, 109 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_nested.h b/arch/arm64/include/asm/kvm_nested.h
index c5b661517cb6..63bb4898e2d9 100644
--- a/arch/arm64/include/asm/kvm_nested.h
+++ b/arch/arm64/include/asm/kvm_nested.h
@@ -122,6 +122,8 @@ extern bool __forward_traps(struct kvm_vcpu *vcpu, unsigned int reg,
 extern bool forward_traps(struct kvm_vcpu *vcpu, u64 control_bit);
 extern bool forward_nv_traps(struct kvm_vcpu *vcpu);
 extern bool forward_nv1_traps(struct kvm_vcpu *vcpu);
+u8 get_guest_mapping_ttl(struct kvm_vcpu *vcpu, struct kvm_s2_mmu *mmu,
+			 u64 addr);
 unsigned int ttl_to_size(u8 ttl);
 
 struct sys_reg_params;
@@ -130,4 +132,6 @@ struct sys_reg_desc;
 void access_nested_id_reg(struct kvm_vcpu *v, struct sys_reg_params *p,
 			  const struct sys_reg_desc *r);
 
+#define KVM_NV_GUEST_MAP_SZ	(KVM_PGTABLE_PROT_SW1 | KVM_PGTABLE_PROT_SW0)
+
 #endif /* __ARM64_KVM_NESTED_H */
diff --git a/arch/arm64/kvm/nested.c b/arch/arm64/kvm/nested.c
index 8b22984a3112..f93f4248c6f3 100644
--- a/arch/arm64/kvm/nested.c
+++ b/arch/arm64/kvm/nested.c
@@ -4,6 +4,7 @@
  * Author: Jintack Lim <jintack.lim at linaro.org>
  */
 
+#include <linux/bitfield.h>
 #include <linux/kvm.h>
 #include <linux/kvm_host.h>
 
@@ -357,6 +358,30 @@ int kvm_walk_nested_s2(struct kvm_vcpu *vcpu, phys_addr_t gipa,
 	return ret;
 }
 
+static int read_host_s2_desc(phys_addr_t pa, u64 *desc, void *data)
+{
+	u64 *va = phys_to_virt(pa);
+
+	*desc = *va;
+
+	return 0;
+}
+
+static int kvm_walk_shadow_s2(struct kvm_s2_mmu *mmu, phys_addr_t gipa,
+			      struct kvm_s2_trans *result)
+{
+	struct s2_walk_info wi = { };
+
+	wi.read_desc = read_host_s2_desc;
+	wi.baddr = mmu->pgd_phys;
+
+	vtcr_to_walk_info(mmu->arch->vtcr, &wi);
+
+	wi.be = IS_ENABLED(CONFIG_CPU_BIG_ENDIAN);
+
+	return walk_nested_s2_pgd(gipa, &wi, result);
+}
+
 unsigned int ttl_to_size(u8 ttl)
 {
 	int level = ttl & 3;
@@ -413,6 +438,80 @@ unsigned int ttl_to_size(u8 ttl)
 	return max_size;
 }
 
+/*
+ * Compute the equivalent of the TTL field by parsing the shadow PT.
+ * The granule size is extracted from VTCR_EL2.TG0 while the level is
+ * retrieved from first entry carrying the level as a tag.
+ */
+u8 get_guest_mapping_ttl(struct kvm_vcpu *vcpu, struct kvm_s2_mmu *mmu,
+			 u64 addr)
+{
+	u64 tmp, sz = 0, vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
+	struct kvm_s2_trans out;
+	u8 ttl, level;
+
+	switch (vtcr & VTCR_EL2_TG0_MASK) {
+	case VTCR_EL2_TG0_4K:
+		ttl = (1 << 2);
+		break;
+	case VTCR_EL2_TG0_16K:
+		ttl = (2 << 2);
+		break;
+	case VTCR_EL2_TG0_64K:
+		ttl = (3 << 2);
+		break;
+	default:
+		BUG();
+	}
+
+	tmp = addr;
+
+again:
+	/* Iteratively compute the block sizes for a particular granule size */
+	switch (vtcr & VTCR_EL2_TG0_MASK) {
+	case VTCR_EL2_TG0_4K:
+		if	(sz < SZ_4K)	sz = SZ_4K;
+		else if (sz < SZ_2M)	sz = SZ_2M;
+		else if (sz < SZ_1G)	sz = SZ_1G;
+		else			sz = 0;
+		break;
+	case VTCR_EL2_TG0_16K:
+		if	(sz < SZ_16K)	sz = SZ_16K;
+		else if (sz < SZ_32M)	sz = SZ_32M;
+		else			sz = 0;
+		break;
+	case VTCR_EL2_TG0_64K:
+		if	(sz < SZ_64K)	sz = SZ_64K;
+		else if (sz < SZ_512M)	sz = SZ_512M;
+		else			sz = 0;
+		break;
+	default:
+		BUG();
+	}
+
+	if (sz == 0)
+		return 0;
+
+	tmp &= ~(sz - 1);
+	out = (struct kvm_s2_trans) { };
+	kvm_walk_shadow_s2(mmu, tmp, &out);
+	level = FIELD_GET(KVM_NV_GUEST_MAP_SZ, out.upper_attr);
+	if (!level)
+		goto again;
+
+	ttl |= level;
+
+	/*
+	 * We now have found some level information in the shadow S2. Check
+	 * that the resulting range is actually including the original IPA.
+	 */
+	sz = ttl_to_size(ttl);
+	if (addr < (tmp + sz))
+		return ttl;
+
+	return 0;
+}
+
 /* Must be called with kvm->mmu_lock held */
 struct kvm_s2_mmu *lookup_s2_mmu(struct kvm *kvm, u64 vttbr, u64 hcr)
 {
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
index aa15cf3b29eb..89fc11024786 100644
--- a/arch/arm64/kvm/sys_regs.c
+++ b/arch/arm64/kvm/sys_regs.c
@@ -2870,10 +2870,13 @@ static unsigned long compute_tlb_inval_range(struct kvm_vcpu *vcpu,
 					     u64 val)
 {
 	unsigned long max_size;
-	u8 ttl = 0;
+	u8 ttl;
 
-	if (cpus_have_const_cap(ARM64_HAS_ARMv8_4_TTL)) {
-		ttl = FIELD_GET(GENMASK_ULL(47, 44), val);
+	ttl = FIELD_GET(GENMASK_ULL(47, 44), val);
+
+	if (!(cpus_have_const_cap(ARM64_HAS_ARMv8_4_TTL) && ttl)) {
+		u64 addr = (val & GENMASK_ULL(35, 0)) << 12;
+		ttl = get_guest_mapping_ttl(vcpu, mmu, addr);
 	}
 
 	max_size = ttl_to_size(ttl);
-- 
2.34.1