KVM: PPC: Keep page physical addresses in per-slot arrays

This allocates an array for each memory slot that is added to store
the physical addresses of the pages in the slot.  This array is
vmalloc'd and accessed in kvmppc_h_enter using real_vmalloc_addr().
This allows us to remove the ram_pginfo field from the kvm_arch
struct, and removes the 64GB guest RAM limit that we had.

We use the low-order bits of the array entries to store a flag
indicating that we have done get_page on the corresponding page,
and therefore need to call put_page when we are finished with the
page.  Currently this is set for all pages except those in our
special RMO regions.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>

arch/powerpc/include/asm/kvm_host.h

@@ -38,6 +38,7 @@
 #define KVM_MEMORY_SLOTS 32
 /* memory slots that does not exposed to userspace */
 #define KVM_PRIVATE_MEM_SLOTS 4
+#define KVM_MEM_SLOTS_NUM (KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 
 #ifdef CONFIG_KVM_MMIO
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
@@ -175,25 +176,27 @@ struct revmap_entry {
 	unsigned long guest_rpte;
 };
 
+/* Low-order bits in kvm->arch.slot_phys[][] */
+#define KVMPPC_GOT_PAGE		0x80
+
 struct kvm_arch {
 #ifdef CONFIG_KVM_BOOK3S_64_HV
 	unsigned long hpt_virt;
 	struct revmap_entry *revmap;
-	unsigned long ram_npages;
 	unsigned long ram_psize;
 	unsigned long ram_porder;
-	struct kvmppc_pginfo *ram_pginfo;
 	unsigned int lpid;
 	unsigned int host_lpid;
 	unsigned long host_lpcr;
 	unsigned long sdr1;
 	unsigned long host_sdr1;
 	int tlbie_lock;
-	int n_rma_pages;
 	unsigned long lpcr;
 	unsigned long rmor;
 	struct kvmppc_rma_info *rma;
 	struct list_head spapr_tce_tables;
+	unsigned long *slot_phys[KVM_MEM_SLOTS_NUM];
+	int slot_npages[KVM_MEM_SLOTS_NUM];
 	unsigned short last_vcpu[NR_CPUS];
 	struct kvmppc_vcore *vcores[KVM_MAX_VCORES];
 #endif /* CONFIG_KVM_BOOK3S_64_HV */

arch/powerpc/kvm/book3s_64_mmu_hv.c

@@ -98,16 +98,16 @@ void kvmppc_free_hpt(struct kvm *kvm)
 void kvmppc_map_vrma(struct kvm *kvm, struct kvm_userspace_memory_region *mem)
 {
 	unsigned long i;
-	unsigned long npages = kvm->arch.ram_npages;
-	unsigned long pfn;
+	unsigned long npages;
+	unsigned long pa;
 	unsigned long *hpte;
 	unsigned long hash;
 	unsigned long porder = kvm->arch.ram_porder;
 	struct revmap_entry *rev;
-	struct kvmppc_pginfo *pginfo = kvm->arch.ram_pginfo;
+	unsigned long *physp;
 
-	if (!pginfo)
-		return;
+	physp = kvm->arch.slot_phys[mem->slot];
+	npages = kvm->arch.slot_npages[mem->slot];
 
 	/* VRMA can't be > 1TB */
 	if (npages > 1ul << (40 - porder))
@@ -117,9 +117,10 @@ void kvmppc_map_vrma(struct kvm *kvm, struct kvm_userspace_memory_region *mem)
 		npages = HPT_NPTEG;
 
 	for (i = 0; i < npages; ++i) {
-		pfn = pginfo[i].pfn;
-		if (!pfn)
+		pa = physp[i];
+		if (!pa)
 			break;
+		pa &= PAGE_MASK;
 		/* can't use hpt_hash since va > 64 bits */
 		hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & HPT_HASH_MASK;
 		/*
@@ -131,8 +132,7 @@ void kvmppc_map_vrma(struct kvm *kvm, struct kvm_userspace_memory_region *mem)
 		hash = (hash << 3) + 7;
 		hpte = (unsigned long *) (kvm->arch.hpt_virt + (hash << 4));
 		/* HPTE low word - RPN, protection, etc. */
-		hpte[1] = (pfn << PAGE_SHIFT) | HPTE_R_R | HPTE_R_C |
-			HPTE_R_M | PP_RWXX;
+		hpte[1] = pa | HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
 		smp_wmb();
 		hpte[0] = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
 			(i << (VRMA_PAGE_ORDER - 16)) | HPTE_V_BOLTED |

arch/powerpc/kvm/book3s_hv.c

@@ -50,14 +50,6 @@
 #include <linux/vmalloc.h>
 #include <linux/highmem.h>
 
-/*
- * For now, limit memory to 64GB and require it to be large pages.
- * This value is chosen because it makes the ram_pginfo array be
- * 64kB in size, which is about as large as we want to be trying
- * to allocate with kmalloc.
- */
-#define MAX_MEM_ORDER		36
-
 #define LARGE_PAGE_ORDER	24	/* 16MB pages */
 
 /* #define EXIT_DEBUG */
@@ -147,10 +139,12 @@ static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 				       unsigned long vcpuid, unsigned long vpa)
 {
 	struct kvm *kvm = vcpu->kvm;
-	unsigned long pg_index, ra, len;
+	unsigned long gfn, pg_index, ra, len;
 	unsigned long pg_offset;
 	void *va;
 	struct kvm_vcpu *tvcpu;
+	struct kvm_memory_slot *memslot;
+	unsigned long *physp;
 
 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 	if (!tvcpu)
@@ -164,14 +158,20 @@ static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 		if (vpa & 0x7f)
 			return H_PARAMETER;
 		/* registering new area; convert logical addr to real */
-		pg_index = vpa >> kvm->arch.ram_porder;
-		pg_offset = vpa & (kvm->arch.ram_psize - 1);
-		if (pg_index >= kvm->arch.ram_npages)
+		gfn = vpa >> PAGE_SHIFT;
+		memslot = gfn_to_memslot(kvm, gfn);
+		if (!memslot || !(memslot->flags & KVM_MEMSLOT_INVALID))
+			return H_PARAMETER;
+		physp = kvm->arch.slot_phys[memslot->id];
+		if (!physp)
 			return H_PARAMETER;
-		if (kvm->arch.ram_pginfo[pg_index].pfn == 0)
+		pg_index = (gfn - memslot->base_gfn) >>
+			(kvm->arch.ram_porder - PAGE_SHIFT);
+		pg_offset = vpa & (kvm->arch.ram_psize - 1);
+		ra = physp[pg_index];
+		if (!ra)
 			return H_PARAMETER;
-		ra = kvm->arch.ram_pginfo[pg_index].pfn << PAGE_SHIFT;
-		ra |= pg_offset;
+		ra = (ra & PAGE_MASK) | pg_offset;
 		va = __va(ra);
 		if (flags <= 1)
 			len = *(unsigned short *)(va + 4);
@@ -1075,12 +1075,11 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 				struct kvm_userspace_memory_region *mem)
 {
 	unsigned long psize, porder;
-	unsigned long i, npages, totalpages;
-	unsigned long pg_ix;
-	struct kvmppc_pginfo *pginfo;
+	unsigned long i, npages;
 	unsigned long hva;
 	struct kvmppc_rma_info *ri = NULL;
 	struct page *page;
+	unsigned long *phys;
 
 	/* For now, only allow 16MB pages */
 	porder = LARGE_PAGE_ORDER;
@@ -1092,20 +1091,21 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 		return -EINVAL;
 	}
 
+	/* Allocate a slot_phys array */
 	npages = mem->memory_size >> porder;
-	totalpages = (mem->guest_phys_addr + mem->memory_size) >> porder;
-
-	/* More memory than we have space to track? */
-	if (totalpages > (1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER)))
-		return -EINVAL;
+	phys = kvm->arch.slot_phys[mem->slot];
+	if (!phys) {
+		phys = vzalloc(npages * sizeof(unsigned long));
+		if (!phys)
+			return -ENOMEM;
+		kvm->arch.slot_phys[mem->slot] = phys;
+		kvm->arch.slot_npages[mem->slot] = npages;
+	}
 
 	/* Do we already have an RMA registered? */
 	if (mem->guest_phys_addr == 0 && kvm->arch.rma)
 		return -EINVAL;
 
-	if (totalpages > kvm->arch.ram_npages)
-		kvm->arch.ram_npages = totalpages;
-
 	/* Is this one of our preallocated RMAs? */
 	if (mem->guest_phys_addr == 0) {
 		struct vm_area_struct *vma;
@@ -1138,7 +1138,6 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 		}
 		atomic_inc(&ri->use_count);
 		kvm->arch.rma = ri;
-		kvm->arch.n_rma_pages = rma_size >> porder;
 
 		/* Update LPCR and RMOR */
 		lpcr = kvm->arch.lpcr;
@@ -1162,12 +1161,9 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 			ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
 	}
 
-	pg_ix = mem->guest_phys_addr >> porder;
-	pginfo = kvm->arch.ram_pginfo + pg_ix;
-	for (i = 0; i < npages; ++i, ++pg_ix) {
-		if (ri && pg_ix < kvm->arch.n_rma_pages) {
-			pginfo[i].pfn = ri->base_pfn +
-				(pg_ix << (porder - PAGE_SHIFT));
+	for (i = 0; i < npages; ++i) {
+		if (ri && i < ri->npages) {
+			phys[i] = (ri->base_pfn << PAGE_SHIFT) + (i << porder);
 			continue;
 		}
 		hva = mem->userspace_addr + (i << porder);
@@ -1183,7 +1179,7 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 			       hva, compound_order(page));
 			goto err;
 		}
-		pginfo[i].pfn = page_to_pfn(page);
+		phys[i] = (page_to_pfn(page) << PAGE_SHIFT) | KVMPPC_GOT_PAGE;
 	}
 
 	return 0;
@@ -1192,6 +1188,28 @@ int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 	return -EINVAL;
 }
 
+static void unpin_slot(struct kvm *kvm, int slot_id)
+{
+	unsigned long *physp;
+	unsigned long j, npages, pfn;
+	struct page *page;
+
+	physp = kvm->arch.slot_phys[slot_id];
+	npages = kvm->arch.slot_npages[slot_id];
+	if (physp) {
+		for (j = 0; j < npages; j++) {
+			if (!(physp[j] & KVMPPC_GOT_PAGE))
+				continue;
+			pfn = physp[j] >> PAGE_SHIFT;
+			page = pfn_to_page(pfn);
+			SetPageDirty(page);
+			put_page(page);
+		}
+		vfree(physp);
+		kvm->arch.slot_phys[slot_id] = NULL;
+	}
+}
+
 void kvmppc_core_commit_memory_region(struct kvm *kvm,
 				struct kvm_userspace_memory_region *mem)
 {
@@ -1203,8 +1221,6 @@ void kvmppc_core_commit_memory_region(struct kvm *kvm,
 int kvmppc_core_init_vm(struct kvm *kvm)
 {
 	long r;
-	unsigned long npages = 1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER);
-	long err = -ENOMEM;
 	unsigned long lpcr;
 
 	/* Allocate hashed page table */
@@ -1214,19 +1230,9 @@ int kvmppc_core_init_vm(struct kvm *kvm)
 
 	INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
 
-	kvm->arch.ram_pginfo = kzalloc(npages * sizeof(struct kvmppc_pginfo),
-				       GFP_KERNEL);
-	if (!kvm->arch.ram_pginfo) {
-		pr_err("kvmppc_core_init_vm: couldn't alloc %lu bytes\n",
-		       npages * sizeof(struct kvmppc_pginfo));
-		goto out_free;
-	}
-
-	kvm->arch.ram_npages = 0;
 	kvm->arch.ram_psize = 1ul << LARGE_PAGE_ORDER;
 	kvm->arch.ram_porder = LARGE_PAGE_ORDER;
 	kvm->arch.rma = NULL;
-	kvm->arch.n_rma_pages = 0;
 
 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
 
@@ -1249,25 +1255,15 @@ int kvmppc_core_init_vm(struct kvm *kvm)
 	kvm->arch.lpcr = lpcr;
 
 	return 0;
-
- out_free:
-	kvmppc_free_hpt(kvm);
-	return err;
 }
 
 void kvmppc_core_destroy_vm(struct kvm *kvm)
 {
-	struct kvmppc_pginfo *pginfo;
 	unsigned long i;
 
-	if (kvm->arch.ram_pginfo) {
-		pginfo = kvm->arch.ram_pginfo;
-		kvm->arch.ram_pginfo = NULL;
-		for (i = kvm->arch.n_rma_pages; i < kvm->arch.ram_npages; ++i)
-			if (pginfo[i].pfn)
-				put_page(pfn_to_page(pginfo[i].pfn));
-		kfree(pginfo);
-	}
+	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
+		unpin_slot(kvm, i);
+
 	if (kvm->arch.rma) {
 		kvm_release_rma(kvm->arch.rma);
 		kvm->arch.rma = NULL;

arch/powerpc/kvm/book3s_hv_rm_mmu.c

@@ -20,6 +20,25 @@
 #include <asm/synch.h>
 #include <asm/ppc-opcode.h>
 
+/*
+ * Since this file is built in even if KVM is a module, we need
+ * a local copy of this function for the case where kvm_main.c is
+ * modular.
+ */
+static struct kvm_memory_slot *builtin_gfn_to_memslot(struct kvm *kvm,
+						gfn_t gfn)
+{
+	struct kvm_memslots *slots;
+	struct kvm_memory_slot *memslot;
+
+	slots = kvm_memslots(kvm);
+	kvm_for_each_memslot(memslot, slots)
+		if (gfn >= memslot->base_gfn &&
+		      gfn < memslot->base_gfn + memslot->npages)
+			return memslot;
+	return NULL;
+}
+
 /* Translate address of a vmalloc'd thing to a linear map address */
 static void *real_vmalloc_addr(void *x)
 {
@@ -59,10 +78,12 @@ long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
 {
 	unsigned long porder;
 	struct kvm *kvm = vcpu->kvm;
-	unsigned long i, lpn, pa;
+	unsigned long i, gfn, lpn, pa;
 	unsigned long *hpte;
 	struct revmap_entry *rev;
 	unsigned long g_ptel = ptel;
+	struct kvm_memory_slot *memslot;
+	unsigned long *physp;
 
 	/* only handle 4k, 64k and 16M pages for now */
 	porder = 12;
@@ -80,12 +101,24 @@ long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
 		} else
 			return H_PARAMETER;
 	}
-	lpn = (ptel & HPTE_R_RPN) >> kvm->arch.ram_porder;
-	if (lpn >= kvm->arch.ram_npages || porder > kvm->arch.ram_porder)
+	if (porder > kvm->arch.ram_porder)
 		return H_PARAMETER;
-	pa = kvm->arch.ram_pginfo[lpn].pfn << PAGE_SHIFT;
+
+	gfn = ((ptel & HPTE_R_RPN) & ~((1ul << porder) - 1)) >> PAGE_SHIFT;
+	memslot = builtin_gfn_to_memslot(kvm, gfn);
+	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID)))
+		return H_PARAMETER;
+	physp = kvm->arch.slot_phys[memslot->id];
+	if (!physp)
+		return H_PARAMETER;
+
+	lpn = (gfn - memslot->base_gfn) >> (kvm->arch.ram_porder - PAGE_SHIFT);
+	physp = real_vmalloc_addr(physp + lpn);
+	pa = *physp;
 	if (!pa)
 		return H_PARAMETER;
+	pa &= PAGE_MASK;
+
 	/* Check WIMG */
 	if ((ptel & HPTE_R_WIMG) != HPTE_R_M &&
 	    (ptel & HPTE_R_WIMG) != (HPTE_R_W | HPTE_R_I | HPTE_R_M))