The ARM changes are largish, but not too scary. And a simple fix

for x86 (bug introduced in 3.19).
 -----BEGIN PGP SIGNATURE-----
 Version: GnuPG v2.0.22 (GNU/Linux)
 
 iQEcBAABAgAGBQJUy2ulAAoJEL/70l94x66D18kIAJhuh2k5Mt3TfP/zfhi2Y6ER
 IAZqyFODs8txZ3v432PB8yWWvr2XfJ3gwfjvurLygQJ3jCGZqDrmucbUUXzEaPUk
 mPnLpxV0ZEmNweS2HLGPX9HJ6zfsZ1dHRk55Tko9ynAO731q7yPjj6HC0th8wzvE
 BRv5y/18rY2zyar+5Azpj5wpOSllq0ynMgjWXGSlaTLbQoyvgZtzbqNY6nsAGrKw
 e8hSUPogfGUmZkBHHHVDYKpgHvWS1hARyuGFo8LeKXKPo7qhYxZHCDpch8TXnq2y
 21IvQfYddGpcMsaTroA5qyXFigxCX+1j3po6MS3ZH9GGXS5fC3sI8t0EDxKiO6Q=
 =O4X0
 -----END PGP SIGNATURE-----

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "The ARM changes are largish, but not too scary.  And a simple fix for
  x86 (bug introduced in 3.19)"

(Paolo sayus these are the "Final" fixes. We'll see).

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86: check LAPIC presence when building apic_map
  arm/arm64: KVM: Use kernel mapping to perform invalidation on page fault
  arm/arm64: KVM: Invalidate data cache on unmap
  arm/arm64: KVM: Use set/way op trapping to track the state of the caches

arch/arm/include/asm/kvm_emulate.h

@@ -38,6 +38,16 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
 	vcpu->arch.hcr = HCR_GUEST_MASK;
 }
 
+static inline unsigned long vcpu_get_hcr(struct kvm_vcpu *vcpu)
+{
+	return vcpu->arch.hcr;
+}
+
+static inline void vcpu_set_hcr(struct kvm_vcpu *vcpu, unsigned long hcr)
+{
+	vcpu->arch.hcr = hcr;
+}
+
 static inline bool vcpu_mode_is_32bit(struct kvm_vcpu *vcpu)
 {
 	return 1;

arch/arm/include/asm/kvm_host.h

@@ -125,9 +125,6 @@ struct kvm_vcpu_arch {
 	 * Anything that is not used directly from assembly code goes
 	 * here.
 	 */
-	/* dcache set/way operation pending */
-	int last_pcpu;
-	cpumask_t require_dcache_flush;
 
 	/* Don't run the guest on this vcpu */
 	bool pause;

arch/arm/include/asm/kvm_mmu.h

@@ -44,6 +44,7 @@
 
 #ifndef __ASSEMBLY__
 
+#include <linux/highmem.h>
 #include <asm/cacheflush.h>
 #include <asm/pgalloc.h>
 
@@ -161,13 +162,10 @@ static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
 	return (vcpu->arch.cp15[c1_SCTLR] & 0b101) == 0b101;
 }
 
-static inline void coherent_cache_guest_page(struct kvm_vcpu *vcpu, hva_t hva,
+static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
-					     unsigned long size,
+					       unsigned long size,
-					     bool ipa_uncached)
+					       bool ipa_uncached)
 {
-	if (!vcpu_has_cache_enabled(vcpu) || ipa_uncached)
-		kvm_flush_dcache_to_poc((void *)hva, size);
-	
 	/*
 	 * If we are going to insert an instruction page and the icache is
 	 * either VIPT or PIPT, there is a potential problem where the host
@@ -179,18 +177,77 @@ static inline void coherent_cache_guest_page(struct kvm_vcpu *vcpu, hva_t hva,
 	 *
 	 * VIVT caches are tagged using both the ASID and the VMID and doesn't
 	 * need any kind of flushing (DDI 0406C.b - Page B3-1392).
+	 *
+	 * We need to do this through a kernel mapping (using the
+	 * user-space mapping has proved to be the wrong
+	 * solution). For that, we need to kmap one page at a time,
+	 * and iterate over the range.
 	 */
-	if (icache_is_pipt()) {
+
-		__cpuc_coherent_user_range(hva, hva + size);
+	bool need_flush = !vcpu_has_cache_enabled(vcpu) || ipa_uncached;
-	} else if (!icache_is_vivt_asid_tagged()) {
+
+	VM_BUG_ON(size & PAGE_MASK);
+
+	if (!need_flush && !icache_is_pipt())
+		goto vipt_cache;
+
+	while (size) {
+		void *va = kmap_atomic_pfn(pfn);
+
+		if (need_flush)
+			kvm_flush_dcache_to_poc(va, PAGE_SIZE);
+
+		if (icache_is_pipt())
+			__cpuc_coherent_user_range((unsigned long)va,
+						   (unsigned long)va + PAGE_SIZE);
+
+		size -= PAGE_SIZE;
+		pfn++;
+
+		kunmap_atomic(va);
+	}
+
+vipt_cache:
+	if (!icache_is_pipt() && !icache_is_vivt_asid_tagged()) {
 		/* any kind of VIPT cache */
 		__flush_icache_all();
 	}
 }
 
+static inline void __kvm_flush_dcache_pte(pte_t pte)
+{
+	void *va = kmap_atomic(pte_page(pte));
+
+	kvm_flush_dcache_to_poc(va, PAGE_SIZE);
+
+	kunmap_atomic(va);
+}
+
+static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
+{
+	unsigned long size = PMD_SIZE;
+	pfn_t pfn = pmd_pfn(pmd);
+
+	while (size) {
+		void *va = kmap_atomic_pfn(pfn);
+
+		kvm_flush_dcache_to_poc(va, PAGE_SIZE);
+
+		pfn++;
+		size -= PAGE_SIZE;
+
+		kunmap_atomic(va);
+	}
+}
+
+static inline void __kvm_flush_dcache_pud(pud_t pud)
+{
+}
+
 #define kvm_virt_to_phys(x)		virt_to_idmap((unsigned long)(x))
 
-void stage2_flush_vm(struct kvm *kvm);
+void kvm_set_way_flush(struct kvm_vcpu *vcpu);
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
 
 #endif	/* !__ASSEMBLY__ */
 

arch/arm/kvm/arm.c

@@ -281,15 +281,6 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	vcpu->cpu = cpu;
 	vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 
-	/*
-	 * Check whether this vcpu requires the cache to be flushed on
-	 * this physical CPU. This is a consequence of doing dcache
-	 * operations by set/way on this vcpu. We do it here to be in
-	 * a non-preemptible section.
-	 */
-	if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
-		flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
-
 	kvm_arm_set_running_vcpu(vcpu);
 }
 
@@ -541,7 +532,6 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 		ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 
 		vcpu->mode = OUTSIDE_GUEST_MODE;
-		vcpu->arch.last_pcpu = smp_processor_id();
 		kvm_guest_exit();
 		trace_kvm_exit(*vcpu_pc(vcpu));
 		/*

arch/arm/kvm/coproc.c

@@ -189,82 +189,40 @@ static bool access_l2ectlr(struct kvm_vcpu *vcpu,
 	return true;
 }
 
-/* See note at ARM ARM B1.14.4 */
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ */
 static bool access_dcsw(struct kvm_vcpu *vcpu,
 			const struct coproc_params *p,
 			const struct coproc_reg *r)
 {
-	unsigned long val;
-	int cpu;
-
 	if (!p->is_write)
 		return read_from_write_only(vcpu, p);
 
-	cpu = get_cpu();
+	kvm_set_way_flush(vcpu);
-
-	cpumask_setall(&vcpu->arch.require_dcache_flush);
-	cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
-
-	/* If we were already preempted, take the long way around */
-	if (cpu != vcpu->arch.last_pcpu) {
-		flush_cache_all();
-		goto done;
-	}
-
-	val = *vcpu_reg(vcpu, p->Rt1);
-
-	switch (p->CRm) {
-	case 6:			/* Upgrade DCISW to DCCISW, as per HCR.SWIO */
-	case 14:		/* DCCISW */
-		asm volatile("mcr p15, 0, %0, c7, c14, 2" : : "r" (val));
-		break;
-
-	case 10:		/* DCCSW */
-		asm volatile("mcr p15, 0, %0, c7, c10, 2" : : "r" (val));
-		break;
-	}
-
-done:
-	put_cpu();
-
 	return true;
 }
 
 /*
  * Generic accessor for VM registers. Only called as long as HCR_TVM
- * is set.
+ * is set.  If the guest enables the MMU, we stop trapping the VM
+ * sys_regs and leave it in complete control of the caches.
+ *
+ * Used by the cpu-specific code.
  */
-static bool access_vm_reg(struct kvm_vcpu *vcpu,
+bool access_vm_reg(struct kvm_vcpu *vcpu,
-			  const struct coproc_params *p,
+		   const struct coproc_params *p,
-			  const struct coproc_reg *r)
+		   const struct coproc_reg *r)
 {
+	bool was_enabled = vcpu_has_cache_enabled(vcpu);
+
 	BUG_ON(!p->is_write);
 
 	vcpu->arch.cp15[r->reg] = *vcpu_reg(vcpu, p->Rt1);
 	if (p->is_64bit)
 		vcpu->arch.cp15[r->reg + 1] = *vcpu_reg(vcpu, p->Rt2);
 
-	return true;
+	kvm_toggle_cache(vcpu, was_enabled);
-}
-
-/*
- * SCTLR accessor. Only called as long as HCR_TVM is set.  If the
- * guest enables the MMU, we stop trapping the VM sys_regs and leave
- * it in complete control of the caches.
- *
- * Used by the cpu-specific code.
- */
-bool access_sctlr(struct kvm_vcpu *vcpu,
-		  const struct coproc_params *p,
-		  const struct coproc_reg *r)
-{
-	access_vm_reg(vcpu, p, r);
-
-	if (vcpu_has_cache_enabled(vcpu)) {	/* MMU+Caches enabled? */
-		vcpu->arch.hcr &= ~HCR_TVM;
-		stage2_flush_vm(vcpu->kvm);
-	}
-
 	return true;
 }
 

arch/arm/kvm/coproc.h

@@ -153,8 +153,8 @@ static inline int cmp_reg(const struct coproc_reg *i1,
 #define is64		.is_64 = true
 #define is32		.is_64 = false
 
-bool access_sctlr(struct kvm_vcpu *vcpu,
+bool access_vm_reg(struct kvm_vcpu *vcpu,
-		  const struct coproc_params *p,
+		   const struct coproc_params *p,
-		  const struct coproc_reg *r);
+		   const struct coproc_reg *r);
 
 #endif /* __ARM_KVM_COPROC_LOCAL_H__ */

arch/arm/kvm/coproc_a15.c

@@ -34,7 +34,7 @@
 static const struct coproc_reg a15_regs[] = {
 	/* SCTLR: swapped by interrupt.S. */
 	{ CRn( 1), CRm( 0), Op1( 0), Op2( 0), is32,
-			access_sctlr, reset_val, c1_SCTLR, 0x00C50078 },
+			access_vm_reg, reset_val, c1_SCTLR, 0x00C50078 },
 };
 
 static struct kvm_coproc_target_table a15_target_table = {

arch/arm/kvm/coproc_a7.c

@@ -37,7 +37,7 @@
 static const struct coproc_reg a7_regs[] = {
 	/* SCTLR: swapped by interrupt.S. */
 	{ CRn( 1), CRm( 0), Op1( 0), Op2( 0), is32,
-			access_sctlr, reset_val, c1_SCTLR, 0x00C50878 },
+			access_vm_reg, reset_val, c1_SCTLR, 0x00C50878 },
 };
 
 static struct kvm_coproc_target_table a7_target_table = {

arch/arm/kvm/mmu.c

@@ -58,6 +58,26 @@ static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
 		kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
 }
 
+/*
+ * D-Cache management functions. They take the page table entries by
+ * value, as they are flushing the cache using the kernel mapping (or
+ * kmap on 32bit).
+ */
+static void kvm_flush_dcache_pte(pte_t pte)
+{
+	__kvm_flush_dcache_pte(pte);
+}
+
+static void kvm_flush_dcache_pmd(pmd_t pmd)
+{
+	__kvm_flush_dcache_pmd(pmd);
+}
+
+static void kvm_flush_dcache_pud(pud_t pud)
+{
+	__kvm_flush_dcache_pud(pud);
+}
+
 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
 				  int min, int max)
 {
@@ -119,6 +139,26 @@ static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
 	put_page(virt_to_page(pmd));
 }
 
+/*
+ * Unmapping vs dcache management:
+ *
+ * If a guest maps certain memory pages as uncached, all writes will
+ * bypass the data cache and go directly to RAM.  However, the CPUs
+ * can still speculate reads (not writes) and fill cache lines with
+ * data.
+ *
+ * Those cache lines will be *clean* cache lines though, so a
+ * clean+invalidate operation is equivalent to an invalidate
+ * operation, because no cache lines are marked dirty.
+ *
+ * Those clean cache lines could be filled prior to an uncached write
+ * by the guest, and the cache coherent IO subsystem would therefore
+ * end up writing old data to disk.
+ *
+ * This is why right after unmapping a page/section and invalidating
+ * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure
+ * the IO subsystem will never hit in the cache.
+ */
 static void unmap_ptes(struct kvm *kvm, pmd_t *pmd,
 		       phys_addr_t addr, phys_addr_t end)
 {
@@ -128,9 +168,16 @@ static void unmap_ptes(struct kvm *kvm, pmd_t *pmd,
 	start_pte = pte = pte_offset_kernel(pmd, addr);
 	do {
 		if (!pte_none(*pte)) {
+			pte_t old_pte = *pte;
+
 			kvm_set_pte(pte, __pte(0));
-			put_page(virt_to_page(pte));
 			kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+			/* No need to invalidate the cache for device mappings */
+			if ((pte_val(old_pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE)
+				kvm_flush_dcache_pte(old_pte);
+
+			put_page(virt_to_page(pte));
 		}
 	} while (pte++, addr += PAGE_SIZE, addr != end);
 
@@ -149,8 +196,13 @@ static void unmap_pmds(struct kvm *kvm, pud_t *pud,
 		next = kvm_pmd_addr_end(addr, end);
 		if (!pmd_none(*pmd)) {
 			if (kvm_pmd_huge(*pmd)) {
+				pmd_t old_pmd = *pmd;
+
 				pmd_clear(pmd);
 				kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+				kvm_flush_dcache_pmd(old_pmd);
+
 				put_page(virt_to_page(pmd));
 			} else {
 				unmap_ptes(kvm, pmd, addr, next);
@@ -173,8 +225,13 @@ static void unmap_puds(struct kvm *kvm, pgd_t *pgd,
 		next = kvm_pud_addr_end(addr, end);
 		if (!pud_none(*pud)) {
 			if (pud_huge(*pud)) {
+				pud_t old_pud = *pud;
+
 				pud_clear(pud);
 				kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+				kvm_flush_dcache_pud(old_pud);
+
 				put_page(virt_to_page(pud));
 			} else {
 				unmap_pmds(kvm, pud, addr, next);
@@ -209,10 +266,9 @@ static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd,
 
 	pte = pte_offset_kernel(pmd, addr);
 	do {
-		if (!pte_none(*pte)) {
+		if (!pte_none(*pte) &&
-			hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
+		    (pte_val(*pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE)
-			kvm_flush_dcache_to_poc((void*)hva, PAGE_SIZE);
+			kvm_flush_dcache_pte(*pte);
-		}
 	} while (pte++, addr += PAGE_SIZE, addr != end);
 }
 
@@ -226,12 +282,10 @@ static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud,
 	do {
 		next = kvm_pmd_addr_end(addr, end);
 		if (!pmd_none(*pmd)) {
-			if (kvm_pmd_huge(*pmd)) {
+			if (kvm_pmd_huge(*pmd))
-				hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
+				kvm_flush_dcache_pmd(*pmd);
-				kvm_flush_dcache_to_poc((void*)hva, PMD_SIZE);
+			else
-			} else {
 				stage2_flush_ptes(kvm, pmd, addr, next);
-			}
 		}
 	} while (pmd++, addr = next, addr != end);
 }
@@ -246,12 +300,10 @@ static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd,
 	do {
 		next = kvm_pud_addr_end(addr, end);
 		if (!pud_none(*pud)) {
-			if (pud_huge(*pud)) {
+			if (pud_huge(*pud))
-				hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
+				kvm_flush_dcache_pud(*pud);
-				kvm_flush_dcache_to_poc((void*)hva, PUD_SIZE);
+			else
-			} else {
 				stage2_flush_pmds(kvm, pud, addr, next);
-			}
 		}
 	} while (pud++, addr = next, addr != end);
 }
@@ -278,7 +330,7 @@ static void stage2_flush_memslot(struct kvm *kvm,
  * Go through the stage 2 page tables and invalidate any cache lines
  * backing memory already mapped to the VM.
  */
-void stage2_flush_vm(struct kvm *kvm)
+static void stage2_flush_vm(struct kvm *kvm)
 {
 	struct kvm_memslots *slots;
 	struct kvm_memory_slot *memslot;
@@ -905,6 +957,12 @@ static bool kvm_is_device_pfn(unsigned long pfn)
 	return !pfn_valid(pfn);
 }
 
+static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
+				      unsigned long size, bool uncached)
+{
+	__coherent_cache_guest_page(vcpu, pfn, size, uncached);
+}
+
 static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 			  struct kvm_memory_slot *memslot, unsigned long hva,
 			  unsigned long fault_status)
@@ -994,8 +1052,7 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 			kvm_set_s2pmd_writable(&new_pmd);
 			kvm_set_pfn_dirty(pfn);
 		}
-		coherent_cache_guest_page(vcpu, hva & PMD_MASK, PMD_SIZE,
+		coherent_cache_guest_page(vcpu, pfn, PMD_SIZE, fault_ipa_uncached);
-					  fault_ipa_uncached);
 		ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
 	} else {
 		pte_t new_pte = pfn_pte(pfn, mem_type);
@@ -1003,8 +1060,7 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 			kvm_set_s2pte_writable(&new_pte);
 			kvm_set_pfn_dirty(pfn);
 		}
-		coherent_cache_guest_page(vcpu, hva, PAGE_SIZE,
+		coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE, fault_ipa_uncached);
-					  fault_ipa_uncached);
 		ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte,
 			pgprot_val(mem_type) == pgprot_val(PAGE_S2_DEVICE));
 	}
@@ -1411,3 +1467,71 @@ void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 	unmap_stage2_range(kvm, gpa, size);
 	spin_unlock(&kvm->mmu_lock);
 }
+
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ *
+ * Main problems:
+ * - S/W ops are local to a CPU (not broadcast)
+ * - We have line migration behind our back (speculation)
+ * - System caches don't support S/W at all (damn!)
+ *
+ * In the face of the above, the best we can do is to try and convert
+ * S/W ops to VA ops. Because the guest is not allowed to infer the
+ * S/W to PA mapping, it can only use S/W to nuke the whole cache,
+ * which is a rather good thing for us.
+ *
+ * Also, it is only used when turning caches on/off ("The expected
+ * usage of the cache maintenance instructions that operate by set/way
+ * is associated with the cache maintenance instructions associated
+ * with the powerdown and powerup of caches, if this is required by
+ * the implementation.").
+ *
+ * We use the following policy:
+ *
+ * - If we trap a S/W operation, we enable VM trapping to detect
+ *   caches being turned on/off, and do a full clean.
+ *
+ * - We flush the caches on both caches being turned on and off.
+ *
+ * - Once the caches are enabled, we stop trapping VM ops.
+ */
+void kvm_set_way_flush(struct kvm_vcpu *vcpu)
+{
+	unsigned long hcr = vcpu_get_hcr(vcpu);
+
+	/*
+	 * If this is the first time we do a S/W operation
+	 * (i.e. HCR_TVM not set) flush the whole memory, and set the
+	 * VM trapping.
+	 *
+	 * Otherwise, rely on the VM trapping to wait for the MMU +
+	 * Caches to be turned off. At that point, we'll be able to
+	 * clean the caches again.
+	 */
+	if (!(hcr & HCR_TVM)) {
+		trace_kvm_set_way_flush(*vcpu_pc(vcpu),
+					vcpu_has_cache_enabled(vcpu));
+		stage2_flush_vm(vcpu->kvm);
+		vcpu_set_hcr(vcpu, hcr | HCR_TVM);
+	}
+}
+
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
+{
+	bool now_enabled = vcpu_has_cache_enabled(vcpu);
+
+	/*
+	 * If switching the MMU+caches on, need to invalidate the caches.
+	 * If switching it off, need to clean the caches.
+	 * Clean + invalidate does the trick always.
+	 */
+	if (now_enabled != was_enabled)
+		stage2_flush_vm(vcpu->kvm);
+
+	/* Caches are now on, stop trapping VM ops (until a S/W op) */
+	if (now_enabled)
+		vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM);
+
+	trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);
+}

arch/arm/kvm/trace.h

@@ -223,6 +223,45 @@ TRACE_EVENT(kvm_hvc,
 		  __entry->vcpu_pc, __entry->r0, __entry->imm)
 );
 
+TRACE_EVENT(kvm_set_way_flush,
+	    TP_PROTO(unsigned long vcpu_pc, bool cache),
+	    TP_ARGS(vcpu_pc, cache),
+
+	    TP_STRUCT__entry(
+		    __field(	unsigned long,	vcpu_pc		)
+		    __field(	bool,		cache		)
+	    ),
+
+	    TP_fast_assign(
+		    __entry->vcpu_pc		= vcpu_pc;
+		    __entry->cache		= cache;
+	    ),
+
+	    TP_printk("S/W flush at 0x%016lx (cache %s)",
+		      __entry->vcpu_pc, __entry->cache ? "on" : "off")
+);
+
+TRACE_EVENT(kvm_toggle_cache,
+	    TP_PROTO(unsigned long vcpu_pc, bool was, bool now),
+	    TP_ARGS(vcpu_pc, was, now),
+
+	    TP_STRUCT__entry(
+		    __field(	unsigned long,	vcpu_pc		)
+		    __field(	bool,		was		)
+		    __field(	bool,		now		)
+	    ),
+
+	    TP_fast_assign(
+		    __entry->vcpu_pc		= vcpu_pc;
+		    __entry->was		= was;
+		    __entry->now		= now;
+	    ),
+
+	    TP_printk("VM op at 0x%016lx (cache was %s, now %s)",
+		      __entry->vcpu_pc, __entry->was ? "on" : "off",
+		      __entry->now ? "on" : "off")
+);
+
 #endif /* _TRACE_KVM_H */
 
 #undef TRACE_INCLUDE_PATH

arch/arm64/include/asm/kvm_emulate.h

@@ -45,6 +45,16 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
 		vcpu->arch.hcr_el2 &= ~HCR_RW;
 }
 
+static inline unsigned long vcpu_get_hcr(struct kvm_vcpu *vcpu)
+{
+	return vcpu->arch.hcr_el2;
+}
+
+static inline void vcpu_set_hcr(struct kvm_vcpu *vcpu, unsigned long hcr)
+{
+	vcpu->arch.hcr_el2 = hcr;
+}
+
 static inline unsigned long *vcpu_pc(const struct kvm_vcpu *vcpu)
 {
 	return (unsigned long *)&vcpu_gp_regs(vcpu)->regs.pc;

arch/arm64/include/asm/kvm_host.h

@@ -116,9 +116,6 @@ struct kvm_vcpu_arch {
 	 * Anything that is not used directly from assembly code goes
 	 * here.
 	 */
-	/* dcache set/way operation pending */
-	int last_pcpu;
-	cpumask_t require_dcache_flush;
 
 	/* Don't run the guest */
 	bool pause;

arch/arm64/include/asm/kvm_mmu.h

@@ -243,24 +243,46 @@ static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
 	return (vcpu_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
 }
 
-static inline void coherent_cache_guest_page(struct kvm_vcpu *vcpu, hva_t hva,
+static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
-					     unsigned long size,
+					       unsigned long size,
-					     bool ipa_uncached)
+					       bool ipa_uncached)
 {
+	void *va = page_address(pfn_to_page(pfn));
+
 	if (!vcpu_has_cache_enabled(vcpu) || ipa_uncached)
-		kvm_flush_dcache_to_poc((void *)hva, size);
+		kvm_flush_dcache_to_poc(va, size);
 
 	if (!icache_is_aliasing()) {		/* PIPT */
-		flush_icache_range(hva, hva + size);
+		flush_icache_range((unsigned long)va,
+				   (unsigned long)va + size);
 	} else if (!icache_is_aivivt()) {	/* non ASID-tagged VIVT */
 		/* any kind of VIPT cache */
 		__flush_icache_all();
 	}
 }
 
+static inline void __kvm_flush_dcache_pte(pte_t pte)
+{
+	struct page *page = pte_page(pte);
+	kvm_flush_dcache_to_poc(page_address(page), PAGE_SIZE);
+}
+
+static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
+{
+	struct page *page = pmd_page(pmd);
+	kvm_flush_dcache_to_poc(page_address(page), PMD_SIZE);
+}
+
+static inline void __kvm_flush_dcache_pud(pud_t pud)
+{
+	struct page *page = pud_page(pud);
+	kvm_flush_dcache_to_poc(page_address(page), PUD_SIZE);
+}
+
 #define kvm_virt_to_phys(x)		__virt_to_phys((unsigned long)(x))
 
-void stage2_flush_vm(struct kvm *kvm);
+void kvm_set_way_flush(struct kvm_vcpu *vcpu);
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
 
 #endif /* __ASSEMBLY__ */
 #endif /* __ARM64_KVM_MMU_H__ */

arch/arm64/kvm/sys_regs.c

@@ -69,68 +69,31 @@ static u32 get_ccsidr(u32 csselr)
 	return ccsidr;
 }
 
-static void do_dc_cisw(u32 val)
+/*
-{
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
-	asm volatile("dc cisw, %x0" : : "r" (val));
+ */
-	dsb(ish);
-}
-
-static void do_dc_csw(u32 val)
-{
-	asm volatile("dc csw, %x0" : : "r" (val));
-	dsb(ish);
-}
-
-/* See note at ARM ARM B1.14.4 */
 static bool access_dcsw(struct kvm_vcpu *vcpu,
 			const struct sys_reg_params *p,
 			const struct sys_reg_desc *r)
 {
-	unsigned long val;
-	int cpu;
-
 	if (!p->is_write)
 		return read_from_write_only(vcpu, p);
 
-	cpu = get_cpu();
+	kvm_set_way_flush(vcpu);
-
-	cpumask_setall(&vcpu->arch.require_dcache_flush);
-	cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
-
-	/* If we were already preempted, take the long way around */
-	if (cpu != vcpu->arch.last_pcpu) {
-		flush_cache_all();
-		goto done;
-	}
-
-	val = *vcpu_reg(vcpu, p->Rt);
-
-	switch (p->CRm) {
-	case 6:			/* Upgrade DCISW to DCCISW, as per HCR.SWIO */
-	case 14:		/* DCCISW */
-		do_dc_cisw(val);
-		break;
-
-	case 10:		/* DCCSW */
-		do_dc_csw(val);
-		break;
-	}
-
-done:
-	put_cpu();
-
 	return true;
 }
 
 /*
  * Generic accessor for VM registers. Only called as long as HCR_TVM
- * is set.
+ * is set. If the guest enables the MMU, we stop trapping the VM
+ * sys_regs and leave it in complete control of the caches.
  */
 static bool access_vm_reg(struct kvm_vcpu *vcpu,
 			  const struct sys_reg_params *p,
 			  const struct sys_reg_desc *r)
 {
 	unsigned long val;
+	bool was_enabled = vcpu_has_cache_enabled(vcpu);
 
 	BUG_ON(!p->is_write);
 
@@ -143,25 +106,7 @@ static bool access_vm_reg(struct kvm_vcpu *vcpu,
 		vcpu_cp15_64_low(vcpu, r->reg) = val & 0xffffffffUL;
 	}
 
-	return true;
+	kvm_toggle_cache(vcpu, was_enabled);
-}
-
-/*
- * SCTLR_EL1 accessor. Only called as long as HCR_TVM is set.  If the
- * guest enables the MMU, we stop trapping the VM sys_regs and leave
- * it in complete control of the caches.
- */
-static bool access_sctlr(struct kvm_vcpu *vcpu,
-			 const struct sys_reg_params *p,
-			 const struct sys_reg_desc *r)
-{
-	access_vm_reg(vcpu, p, r);
-
-	if (vcpu_has_cache_enabled(vcpu)) {	/* MMU+Caches enabled? */
-		vcpu->arch.hcr_el2 &= ~HCR_TVM;
-		stage2_flush_vm(vcpu->kvm);
-	}
-
 	return true;
 }
 
@@ -377,7 +322,7 @@ static const struct sys_reg_desc sys_reg_descs[] = {
 	  NULL, reset_mpidr, MPIDR_EL1 },
 	/* SCTLR_EL1 */
 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
-	  access_sctlr, reset_val, SCTLR_EL1, 0x00C50078 },
+	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
 	/* CPACR_EL1 */
 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
 	  NULL, reset_val, CPACR_EL1, 0 },
@@ -657,7 +602,7 @@ static const struct sys_reg_desc cp14_64_regs[] = {
  * register).
  */
 static const struct sys_reg_desc cp15_regs[] = {
-	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_sctlr, NULL, c1_SCTLR },
+	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },

arch/x86/kvm/lapic.c

@@ -192,6 +192,9 @@ static void recalculate_apic_map(struct kvm *kvm)
 		u16 cid, lid;
 		u32 ldr, aid;
 
+		if (!kvm_apic_present(vcpu))
+			continue;
+
 		aid = kvm_apic_id(apic);
 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
 		cid = apic_cluster_id(new, ldr);