KVM: ppc: support large host pages

KVM on 440 has always been able to handle large guest mappings with 4K host
pages -- we must, since the guest kernel uses 256MB mappings.

This patch makes KVM work when the host has large pages too (tested with 64K).

Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

2 files changed, 62 insertions, 21 deletions

diff --git a/arch/powerpc/kvm/44x_tlb.c b/arch/powerpc/kvm/44x_tlb.c
index ee2461860bcf..d49dc66ab3c3 100644
--- a/arch/powerpc/kvm/44x_tlb.c
+++ b/arch/powerpc/kvm/44x_tlb.c
@@ -28,6 +28,13 @@
 
 #include "44x_tlb.h"
 
+#ifndef PPC44x_TLBE_SIZE
+#define PPC44x_TLBE_SIZE        PPC44x_TLB_4K
+#endif
+
+#define PAGE_SIZE_4K (1<<12)
+#define PAGE_MASK_4K (~(PAGE_SIZE_4K - 1))
+
 #define PPC44x_TLB_UATTR_MASK \
         (PPC44x_TLB_U0|PPC44x_TLB_U1|PPC44x_TLB_U2|PPC44x_TLB_U3)
 #define PPC44x_TLB_USER_PERM_MASK (PPC44x_TLB_UX|PPC44x_TLB_UR|PPC44x_TLB_UW)
@@ -179,15 +186,26 @@ void kvmppc_tlbe_set_modified(struct kvm_vcpu *vcpu, unsigned int i)
         vcpu_44x->shadow_tlb_mod[i] = 1;
 }
 
-/* Caller must ensure that the specified guest TLB entry is safe to insert into
- * the shadow TLB. */
-void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 gvaddr, gfn_t gfn, u64 asid,
-                    u32 flags)
+/**
+ * kvmppc_mmu_map -- create a host mapping for guest memory
+ *
+ * If the guest wanted a larger page than the host supports, only the first
+ * host page is mapped here and the rest are demand faulted.
+ *
+ * If the guest wanted a smaller page than the host page size, we map only the
+ * guest-size page (i.e. not a full host page mapping).
+ *
+ * Caller must ensure that the specified guest TLB entry is safe to insert into
+ * the shadow TLB.
+ */
+void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 gvaddr, gpa_t gpaddr, u64 asid,
+                    u32 flags, u32 max_bytes)
 {
         struct kvmppc_vcpu_44x *vcpu_44x = to_44x(vcpu);
         struct page *new_page;
         struct kvmppc_44x_tlbe *stlbe;
         hpa_t hpaddr;
+        gfn_t gfn;
         unsigned int victim;
 
         /* Future optimization: don't overwrite the TLB entry containing the
@@ -198,6 +216,7 @@ void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 gvaddr, gfn_t gfn, u64 asid,
         stlbe = &vcpu_44x->shadow_tlb[victim];
 
         /* Get reference to new page. */
+        gfn = gpaddr >> PAGE_SHIFT;
         new_page = gfn_to_page(vcpu->kvm, gfn);
         if (is_error_page(new_page)) {
                 printk(KERN_ERR "Couldn't get guest page for gfn %lx!\n", gfn);
@@ -220,10 +239,25 @@ void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 gvaddr, gfn_t gfn, u64 asid,
         stlbe->tid = !(asid & 0xff);
 
         /* Force TS=1 for all guest mappings. */
-        /* For now we hardcode 4KB mappings, but it will be important to
-         * use host large pages in the future. */
-        stlbe->word0 = (gvaddr & PAGE_MASK) | PPC44x_TLB_VALID | PPC44x_TLB_TS
-                       | PPC44x_TLB_4K;
+        stlbe->word0 = PPC44x_TLB_VALID | PPC44x_TLB_TS;
+
+        if (max_bytes >= PAGE_SIZE) {
+                /* Guest mapping is larger than or equal to host page size. We can use
+                 * a "native" host mapping. */
+                stlbe->word0 |= (gvaddr & PAGE_MASK) | PPC44x_TLBE_SIZE;
+        } else {
+                /* Guest mapping is smaller than host page size. We must restrict the
+                 * size of the mapping to be at most the smaller of the two, but for
+                 * simplicity we fall back to a 4K mapping (this is probably what the
+                 * guest is using anyways). */
+                stlbe->word0 |= (gvaddr & PAGE_MASK_4K) | PPC44x_TLB_4K;
+
+                /* 'hpaddr' is a host page, which is larger than the mapping we're
+                 * inserting here. To compensate, we must add the in-page offset to the
+                 * sub-page. */
+                hpaddr |= gpaddr & (PAGE_MASK ^ PAGE_MASK_4K);
+        }
+
         stlbe->word1 = (hpaddr & 0xfffffc00) | ((hpaddr >> 32) & 0xf);
         stlbe->word2 = kvmppc_44x_tlb_shadow_attrib(flags,
                                                     vcpu->arch.msr & MSR_PR);
@@ -322,10 +356,8 @@ static int tlbe_is_host_safe(const struct kvm_vcpu *vcpu,
 int kvmppc_44x_emul_tlbwe(struct kvm_vcpu *vcpu, u8 ra, u8 rs, u8 ws)
 {
         struct kvmppc_vcpu_44x *vcpu_44x = to_44x(vcpu);
-        u64 eaddr;
-        u64 raddr;
+        gva_t eaddr;
         u64 asid;
-        u32 flags;
         struct kvmppc_44x_tlbe *tlbe;
         unsigned int index;
 
@@ -364,15 +396,22 @@ int kvmppc_44x_emul_tlbwe(struct kvm_vcpu *vcpu, u8 ra, u8 rs, u8 ws)
         }
 
         if (tlbe_is_host_safe(vcpu, tlbe)) {
+                gpa_t gpaddr;
+                u32 flags;
+                u32 bytes;
+
                 eaddr = get_tlb_eaddr(tlbe);
-                raddr = get_tlb_raddr(tlbe);
+                gpaddr = get_tlb_raddr(tlbe);
+
+                /* Use the advertised page size to mask effective and real addrs. */
+                bytes = get_tlb_bytes(tlbe);
+                eaddr &= ~(bytes - 1);
+                gpaddr &= ~(bytes - 1);
+
                 asid = (tlbe->word0 & PPC44x_TLB_TS) | tlbe->tid;
                 flags = tlbe->word2 & 0xffff;
 
-                /* Create a 4KB mapping on the host. If the guest wanted a
-                 * large page, only the first 4KB is mapped here and the rest
-                 * are mapped on the fly. */
-                kvmppc_mmu_map(vcpu, eaddr, raddr >> PAGE_SHIFT, asid, flags);
+                kvmppc_mmu_map(vcpu, eaddr, gpaddr, asid, flags, bytes);
         }
 
         KVMTRACE_5D(GTLB_WRITE, vcpu, index,
diff --git a/arch/powerpc/kvm/booke.c b/arch/powerpc/kvm/booke.c
index ec59a6768ec3..924c7b4b1107 100644
--- a/arch/powerpc/kvm/booke.c
+++ b/arch/powerpc/kvm/booke.c
@@ -308,8 +308,8 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
                          * b) the guest used a large mapping which we're faking
                          * Either way, we need to satisfy the fault without
                          * invoking the guest. */
-                        kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
-                                       gtlbe->word2);
+                        kvmppc_mmu_map(vcpu, eaddr, vcpu->arch.paddr_accessed, gtlbe->tid,
+                                       gtlbe->word2, get_tlb_bytes(gtlbe));
                         vcpu->stat.dtlb_virt_miss_exits++;
                         r = RESUME_GUEST;
                 } else {
@@ -325,6 +325,7 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
         case BOOKE_INTERRUPT_ITLB_MISS: {
                 struct kvmppc_44x_tlbe *gtlbe;
                 unsigned long eaddr = vcpu->arch.pc;
+                gpa_t gpaddr;
                 gfn_t gfn;
 
                 r = RESUME_GUEST;
@@ -340,7 +341,8 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 
                 vcpu->stat.itlb_virt_miss_exits++;
 
-                gfn = tlb_xlate(gtlbe, eaddr) >> PAGE_SHIFT;
+                gpaddr = tlb_xlate(gtlbe, eaddr);
+                gfn = gpaddr >> PAGE_SHIFT;
 
                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
                         /* The guest TLB had a mapping, but the shadow TLB
@@ -349,8 +351,8 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
                          * b) the guest used a large mapping which we're faking
                          * Either way, we need to satisfy the fault without
                          * invoking the guest. */
-                        kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
-                                       gtlbe->word2);
+                        kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlbe->tid,
+                                       gtlbe->word2, get_tlb_bytes(gtlbe));
                 } else {
                         /* Guest mapped and leaped at non-RAM! */
                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);