KVM: PPC: Magic Page Book3s support

We need to override EA as well as PA lookups for the magic page. When the guest
tells us to project it, the magic page overrides any guest mappings.

In order to reflect that, we need to hook into all the MMU layers of KVM to
force map the magic page if necessary.

Signed-off-by: Alexander Graf <[e-mail]>
Signed-off-by: Avi Kivity <[e-mail]>

arch/powerpc/include/asm/kvm_book3s.h

···
 			   bool upper, u32 val);
 extern void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr);
 extern int kvmppc_emulate_paired_single(struct kvm_run *run, struct kvm_vcpu *vcpu);
+extern pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
 
 extern u32 kvmppc_trampoline_lowmem;
 extern u32 kvmppc_trampoline_enter;

arch/powerpc/kvm/book3s.c

···
 	}
 }
 
+pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
+{
+	ulong mp_pa = vcpu->arch.magic_page_pa;
+
+	/* Magic page override */
+	if (unlikely(mp_pa) &&
+	    unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
+		     ((mp_pa & PAGE_MASK) & KVM_PAM))) {
+		ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
+		pfn_t pfn;
+
+		pfn = (pfn_t)virt_to_phys((void*)shared_page) >> PAGE_SHIFT;
+		get_page(pfn_to_page(pfn));
+		return pfn;
+	}
+
+	return gfn_to_pfn(vcpu->kvm, gfn);
+}
+
 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
  * emulate 32 bytes dcbz length.
···
 
 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 {
+	ulong mp_pa = vcpu->arch.magic_page_pa;
+
+	if (unlikely(mp_pa) &&
+	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
+		return 1;
+	}
+
 	return kvm_is_visible_gfn(vcpu->kvm, gfn);
 }
 
···
 	struct kvmppc_vcpu_book3s *vcpu_book3s;
 	struct kvm_vcpu *vcpu;
 	int err = -ENOMEM;
+	unsigned long p;
 
 	vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
 	if (!vcpu_book3s)
···
 	if (err)
 		goto free_shadow_vcpu;
 
-	vcpu->arch.shared = (void*)__get_free_page(GFP_KERNEL|__GFP_ZERO);
-	if (!vcpu->arch.shared)
+	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
+	/* the real shared page fills the last 4k of our page */
+	vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
+	if (!p)
 		goto uninit_vcpu;
 
 	vcpu->arch.host_retip = kvm_return_point;
···
 {
 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 
-	free_page((unsigned long)vcpu->arch.shared);
+	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
 	kvm_vcpu_uninit(vcpu);
 	kfree(vcpu_book3s->shadow_vcpu);
 	vfree(vcpu_book3s);

arch/powerpc/kvm/book3s_32_mmu.c

···
 				      struct kvmppc_pte *pte, bool data)
 {
 	int r;
+	ulong mp_ea = vcpu->arch.magic_page_ea;
 
 	pte->eaddr = eaddr;
+
+	/* Magic page override */
+	if (unlikely(mp_ea) &&
+	    unlikely((eaddr & ~0xfffULL) == (mp_ea & ~0xfffULL)) &&
+	    !(vcpu->arch.shared->msr & MSR_PR)) {
+		pte->vpage = kvmppc_mmu_book3s_32_ea_to_vp(vcpu, eaddr, data);
+		pte->raddr = vcpu->arch.magic_page_pa | (pte->raddr & 0xfff);
+		pte->raddr &= KVM_PAM;
+		pte->may_execute = true;
+		pte->may_read = true;
+		pte->may_write = true;
+
+		return 0;
+	}
+
 	r = kvmppc_mmu_book3s_32_xlate_bat(vcpu, eaddr, pte, data);
 	if (r < 0)
 	       r = kvmppc_mmu_book3s_32_xlate_pte(vcpu, eaddr, pte, data, true);

arch/powerpc/kvm/book3s_32_mmu_host.c

···
 	struct hpte_cache *pte;
 
 	/* Get host physical address for gpa */
-	hpaddr = gfn_to_pfn(vcpu->kvm, orig_pte->raddr >> PAGE_SHIFT);
+	hpaddr = kvmppc_gfn_to_pfn(vcpu, orig_pte->raddr >> PAGE_SHIFT);
 	if (kvm_is_error_hva(hpaddr)) {
 		printk(KERN_INFO "Couldn't get guest page for gfn %lx!\n",
 				 orig_pte->eaddr);

arch/powerpc/kvm/book3s_64_mmu.c

···
 	bool found = false;
 	bool perm_err = false;
 	int second = 0;
+	ulong mp_ea = vcpu->arch.magic_page_ea;
+
+	/* Magic page override */
+	if (unlikely(mp_ea) &&
+	    unlikely((eaddr & ~0xfffULL) == (mp_ea & ~0xfffULL)) &&
+	    !(vcpu->arch.shared->msr & MSR_PR)) {
+		gpte->eaddr = eaddr;
+		gpte->vpage = kvmppc_mmu_book3s_64_ea_to_vp(vcpu, eaddr, data);
+		gpte->raddr = vcpu->arch.magic_page_pa | (gpte->raddr & 0xfff);
+		gpte->raddr &= KVM_PAM;
+		gpte->may_execute = true;
+		gpte->may_read = true;
+		gpte->may_write = true;
+
+		return 0;
+	}
 
 	slbe = kvmppc_mmu_book3s_64_find_slbe(vcpu_book3s, eaddr);
 	if (!slbe)
···
 	ulong ea = esid << SID_SHIFT;
 	struct kvmppc_slb *slb;
 	u64 gvsid = esid;
+	ulong mp_ea = vcpu->arch.magic_page_ea;
 
 	if (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 		slb = kvmppc_mmu_book3s_64_find_slbe(to_book3s(vcpu), ea);
···
 		break;
 	case MSR_DR|MSR_IR:
 		if (!slb)
-			return -ENOENT;
+			goto no_slb;
 
 		*vsid = gvsid;
 		break;
···
 		*vsid |= VSID_PR;
 
 	return 0;
+
+no_slb:
+	/* Catch magic page case */
+	if (unlikely(mp_ea) &&
+	    unlikely(esid == (mp_ea >> SID_SHIFT)) &&
+	    !(vcpu->arch.shared->msr & MSR_PR)) {
+		*vsid = VSID_REAL | esid;
+		return 0;
+	}
+
+	return -EINVAL;
 }
 
 static bool kvmppc_mmu_book3s_64_is_dcbz32(struct kvm_vcpu *vcpu)

arch/powerpc/kvm/book3s_64_mmu_host.c

···
 	struct kvmppc_sid_map *map;
 
 	/* Get host physical address for gpa */
-	hpaddr = gfn_to_pfn(vcpu->kvm, orig_pte->raddr >> PAGE_SHIFT);
+	hpaddr = kvmppc_gfn_to_pfn(vcpu, orig_pte->raddr >> PAGE_SHIFT);
 	if (kvm_is_error_hva(hpaddr)) {
 		printk(KERN_INFO "Couldn't get guest page for gfn %lx!\n", orig_pte->eaddr);
 		return -EINVAL;
 	}
 	hpaddr <<= PAGE_SHIFT;
-#if PAGE_SHIFT == 12
-#elif PAGE_SHIFT == 16
-	hpaddr |= orig_pte->raddr & 0xf000;
-#else
-#error Unknown page size
-#endif
+	hpaddr |= orig_pte->raddr & (~0xfffULL & ~PAGE_MASK);
 
 	/* and write the mapping ea -> hpa into the pt */
 	vcpu->arch.mmu.esid_to_vsid(vcpu, orig_pte->eaddr >> SID_SHIFT, &vsid);