Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Kernel-based Virtual Machine driver for Linux |
| 3 | * |
| 4 | * This module enables machines with Intel VT-x extensions to run virtual |
| 5 | * machines without emulation or binary translation. |
| 6 | * |
| 7 | * MMU support |
| 8 | * |
| 9 | * Copyright (C) 2006 Qumranet, Inc. |
| 10 | * |
| 11 | * Authors: |
| 12 | * Yaniv Kamay <yaniv@qumranet.com> |
| 13 | * Avi Kivity <avi@qumranet.com> |
| 14 | * |
| 15 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 16 | * the COPYING file in the top-level directory. |
| 17 | * |
| 18 | */ |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 19 | |
| 20 | #include "vmx.h" |
Zhang Xiantao | 1d737c8 | 2007-12-14 09:35:10 +0800 | [diff] [blame] | 21 | #include "mmu.h" |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 22 | |
Avi Kivity | edf8841 | 2007-12-16 11:02:48 +0200 | [diff] [blame] | 23 | #include <linux/kvm_host.h> |
Avi Kivity | e495606 | 2007-06-28 14:15:57 -0400 | [diff] [blame] | 24 | #include <linux/types.h> |
| 25 | #include <linux/string.h> |
| 26 | #include <linux/mm.h> |
| 27 | #include <linux/highmem.h> |
| 28 | #include <linux/module.h> |
Izik Eidus | 448353c | 2007-11-26 14:08:14 +0200 | [diff] [blame] | 29 | #include <linux/swap.h> |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 30 | #include <linux/hugetlb.h> |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 31 | #include <linux/compiler.h> |
Avi Kivity | e495606 | 2007-06-28 14:15:57 -0400 | [diff] [blame] | 32 | |
| 33 | #include <asm/page.h> |
| 34 | #include <asm/cmpxchg.h> |
Avi Kivity | 4e54237 | 2007-11-21 14:08:40 +0200 | [diff] [blame] | 35 | #include <asm/io.h> |
Avi Kivity | e495606 | 2007-06-28 14:15:57 -0400 | [diff] [blame] | 36 | |
Joerg Roedel | 1855267 | 2008-02-07 13:47:41 +0100 | [diff] [blame] | 37 | /* |
| 38 | * When setting this variable to true it enables Two-Dimensional-Paging |
| 39 | * where the hardware walks 2 page tables: |
| 40 | * 1. the guest-virtual to guest-physical |
| 41 | * 2. while doing 1. it walks guest-physical to host-physical |
| 42 | * If the hardware supports that we don't need to do shadow paging. |
| 43 | */ |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 44 | bool tdp_enabled = false; |
Joerg Roedel | 1855267 | 2008-02-07 13:47:41 +0100 | [diff] [blame] | 45 | |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 46 | #undef MMU_DEBUG |
| 47 | |
| 48 | #undef AUDIT |
| 49 | |
| 50 | #ifdef AUDIT |
| 51 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg); |
| 52 | #else |
| 53 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {} |
| 54 | #endif |
| 55 | |
| 56 | #ifdef MMU_DEBUG |
| 57 | |
| 58 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) |
| 59 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) |
| 60 | |
| 61 | #else |
| 62 | |
| 63 | #define pgprintk(x...) do { } while (0) |
| 64 | #define rmap_printk(x...) do { } while (0) |
| 65 | |
| 66 | #endif |
| 67 | |
| 68 | #if defined(MMU_DEBUG) || defined(AUDIT) |
| 69 | static int dbg = 1; |
| 70 | #endif |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 71 | |
Yaozu Dong | d6c69ee | 2007-04-25 14:17:25 +0800 | [diff] [blame] | 72 | #ifndef MMU_DEBUG |
| 73 | #define ASSERT(x) do { } while (0) |
| 74 | #else |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 75 | #define ASSERT(x) \ |
| 76 | if (!(x)) { \ |
| 77 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ |
| 78 | __FILE__, __LINE__, #x); \ |
| 79 | } |
Yaozu Dong | d6c69ee | 2007-04-25 14:17:25 +0800 | [diff] [blame] | 80 | #endif |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 81 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 82 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
| 83 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 |
| 84 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 85 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
| 86 | |
| 87 | #define PT64_LEVEL_BITS 9 |
| 88 | |
| 89 | #define PT64_LEVEL_SHIFT(level) \ |
Mike Day | d77c26f | 2007-10-08 09:02:08 -0400 | [diff] [blame] | 90 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 91 | |
| 92 | #define PT64_LEVEL_MASK(level) \ |
| 93 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) |
| 94 | |
| 95 | #define PT64_INDEX(address, level)\ |
| 96 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) |
| 97 | |
| 98 | |
| 99 | #define PT32_LEVEL_BITS 10 |
| 100 | |
| 101 | #define PT32_LEVEL_SHIFT(level) \ |
Mike Day | d77c26f | 2007-10-08 09:02:08 -0400 | [diff] [blame] | 102 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 103 | |
| 104 | #define PT32_LEVEL_MASK(level) \ |
| 105 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) |
| 106 | |
| 107 | #define PT32_INDEX(address, level)\ |
| 108 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) |
| 109 | |
| 110 | |
Avi Kivity | 27aba76 | 2007-03-09 13:04:31 +0200 | [diff] [blame] | 111 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 112 | #define PT64_DIR_BASE_ADDR_MASK \ |
| 113 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) |
| 114 | |
| 115 | #define PT32_BASE_ADDR_MASK PAGE_MASK |
| 116 | #define PT32_DIR_BASE_ADDR_MASK \ |
| 117 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) |
| 118 | |
Avi Kivity | 79539ce | 2007-11-21 02:06:21 +0200 | [diff] [blame] | 119 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \ |
| 120 | | PT64_NX_MASK) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 121 | |
| 122 | #define PFERR_PRESENT_MASK (1U << 0) |
| 123 | #define PFERR_WRITE_MASK (1U << 1) |
| 124 | #define PFERR_USER_MASK (1U << 2) |
Avi Kivity | 73b1087 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 125 | #define PFERR_FETCH_MASK (1U << 4) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 126 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 127 | #define PT_DIRECTORY_LEVEL 2 |
| 128 | #define PT_PAGE_TABLE_LEVEL 1 |
| 129 | |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 130 | #define RMAP_EXT 4 |
| 131 | |
Avi Kivity | fe135d2 | 2007-12-09 16:15:46 +0200 | [diff] [blame] | 132 | #define ACC_EXEC_MASK 1 |
| 133 | #define ACC_WRITE_MASK PT_WRITABLE_MASK |
| 134 | #define ACC_USER_MASK PT_USER_MASK |
| 135 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) |
| 136 | |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 137 | struct kvm_pv_mmu_op_buffer { |
| 138 | void *ptr; |
| 139 | unsigned len; |
| 140 | unsigned processed; |
| 141 | char buf[512] __aligned(sizeof(long)); |
| 142 | }; |
| 143 | |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 144 | struct kvm_rmap_desc { |
| 145 | u64 *shadow_ptes[RMAP_EXT]; |
| 146 | struct kvm_rmap_desc *more; |
| 147 | }; |
| 148 | |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 149 | static struct kmem_cache *pte_chain_cache; |
| 150 | static struct kmem_cache *rmap_desc_cache; |
Avi Kivity | d3d25b0 | 2007-05-30 12:34:53 +0300 | [diff] [blame] | 151 | static struct kmem_cache *mmu_page_header_cache; |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 152 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 153 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
| 154 | static u64 __read_mostly shadow_notrap_nonpresent_pte; |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 155 | static u64 __read_mostly shadow_base_present_pte; |
| 156 | static u64 __read_mostly shadow_nx_mask; |
| 157 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ |
| 158 | static u64 __read_mostly shadow_user_mask; |
| 159 | static u64 __read_mostly shadow_accessed_mask; |
| 160 | static u64 __read_mostly shadow_dirty_mask; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 161 | |
| 162 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) |
| 163 | { |
| 164 | shadow_trap_nonpresent_pte = trap_pte; |
| 165 | shadow_notrap_nonpresent_pte = notrap_pte; |
| 166 | } |
| 167 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); |
| 168 | |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 169 | void kvm_mmu_set_base_ptes(u64 base_pte) |
| 170 | { |
| 171 | shadow_base_present_pte = base_pte; |
| 172 | } |
| 173 | EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes); |
| 174 | |
| 175 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, |
| 176 | u64 dirty_mask, u64 nx_mask, u64 x_mask) |
| 177 | { |
| 178 | shadow_user_mask = user_mask; |
| 179 | shadow_accessed_mask = accessed_mask; |
| 180 | shadow_dirty_mask = dirty_mask; |
| 181 | shadow_nx_mask = nx_mask; |
| 182 | shadow_x_mask = x_mask; |
| 183 | } |
| 184 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); |
| 185 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 186 | static int is_write_protection(struct kvm_vcpu *vcpu) |
| 187 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 188 | return vcpu->arch.cr0 & X86_CR0_WP; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 189 | } |
| 190 | |
| 191 | static int is_cpuid_PSE36(void) |
| 192 | { |
| 193 | return 1; |
| 194 | } |
| 195 | |
Avi Kivity | 73b1087 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 196 | static int is_nx(struct kvm_vcpu *vcpu) |
| 197 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 198 | return vcpu->arch.shadow_efer & EFER_NX; |
Avi Kivity | 73b1087 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 199 | } |
| 200 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 201 | static int is_present_pte(unsigned long pte) |
| 202 | { |
| 203 | return pte & PT_PRESENT_MASK; |
| 204 | } |
| 205 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 206 | static int is_shadow_present_pte(u64 pte) |
| 207 | { |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 208 | return pte != shadow_trap_nonpresent_pte |
| 209 | && pte != shadow_notrap_nonpresent_pte; |
| 210 | } |
| 211 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 212 | static int is_large_pte(u64 pte) |
| 213 | { |
| 214 | return pte & PT_PAGE_SIZE_MASK; |
| 215 | } |
| 216 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 217 | static int is_writeble_pte(unsigned long pte) |
| 218 | { |
| 219 | return pte & PT_WRITABLE_MASK; |
| 220 | } |
| 221 | |
Avi Kivity | e3c5e7ec | 2007-10-11 12:32:30 +0200 | [diff] [blame] | 222 | static int is_dirty_pte(unsigned long pte) |
| 223 | { |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 224 | return pte & shadow_dirty_mask; |
Avi Kivity | e3c5e7ec | 2007-10-11 12:32:30 +0200 | [diff] [blame] | 225 | } |
| 226 | |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 227 | static int is_rmap_pte(u64 pte) |
| 228 | { |
Avi Kivity | 4b1a80f | 2008-03-23 12:18:19 +0200 | [diff] [blame] | 229 | return is_shadow_present_pte(pte); |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 230 | } |
| 231 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 232 | static pfn_t spte_to_pfn(u64 pte) |
Avi Kivity | 0b49ea8 | 2008-03-23 15:06:23 +0200 | [diff] [blame] | 233 | { |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 234 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
Avi Kivity | 0b49ea8 | 2008-03-23 15:06:23 +0200 | [diff] [blame] | 235 | } |
| 236 | |
Avi Kivity | da92852 | 2007-11-21 13:54:47 +0200 | [diff] [blame] | 237 | static gfn_t pse36_gfn_delta(u32 gpte) |
| 238 | { |
| 239 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; |
| 240 | |
| 241 | return (gpte & PT32_DIR_PSE36_MASK) << shift; |
| 242 | } |
| 243 | |
Avi Kivity | e663ee6 | 2007-05-31 15:46:04 +0300 | [diff] [blame] | 244 | static void set_shadow_pte(u64 *sptep, u64 spte) |
| 245 | { |
| 246 | #ifdef CONFIG_X86_64 |
| 247 | set_64bit((unsigned long *)sptep, spte); |
| 248 | #else |
| 249 | set_64bit((unsigned long long *)sptep, spte); |
| 250 | #endif |
| 251 | } |
| 252 | |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 253 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 254 | struct kmem_cache *base_cache, int min) |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 255 | { |
| 256 | void *obj; |
| 257 | |
| 258 | if (cache->nobjs >= min) |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 259 | return 0; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 260 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 261 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 262 | if (!obj) |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 263 | return -ENOMEM; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 264 | cache->objects[cache->nobjs++] = obj; |
| 265 | } |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 266 | return 0; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 267 | } |
| 268 | |
| 269 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) |
| 270 | { |
| 271 | while (mc->nobjs) |
| 272 | kfree(mc->objects[--mc->nobjs]); |
| 273 | } |
| 274 | |
Avi Kivity | c1158e6 | 2007-07-20 08:18:27 +0300 | [diff] [blame] | 275 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 276 | int min) |
Avi Kivity | c1158e6 | 2007-07-20 08:18:27 +0300 | [diff] [blame] | 277 | { |
| 278 | struct page *page; |
| 279 | |
| 280 | if (cache->nobjs >= min) |
| 281 | return 0; |
| 282 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 283 | page = alloc_page(GFP_KERNEL); |
Avi Kivity | c1158e6 | 2007-07-20 08:18:27 +0300 | [diff] [blame] | 284 | if (!page) |
| 285 | return -ENOMEM; |
| 286 | set_page_private(page, 0); |
| 287 | cache->objects[cache->nobjs++] = page_address(page); |
| 288 | } |
| 289 | return 0; |
| 290 | } |
| 291 | |
| 292 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) |
| 293 | { |
| 294 | while (mc->nobjs) |
Avi Kivity | c4d198d | 2007-07-21 09:06:46 +0300 | [diff] [blame] | 295 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
Avi Kivity | c1158e6 | 2007-07-20 08:18:27 +0300 | [diff] [blame] | 296 | } |
| 297 | |
Avi Kivity | 8c43850 | 2007-04-16 11:53:17 +0300 | [diff] [blame] | 298 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
| 299 | { |
| 300 | int r; |
| 301 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 302 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache, |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 303 | pte_chain_cache, 4); |
| 304 | if (r) |
| 305 | goto out; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 306 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache, |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 307 | rmap_desc_cache, 1); |
| 308 | if (r) |
| 309 | goto out; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 310 | r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8); |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 311 | if (r) |
| 312 | goto out; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 313 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
Avi Kivity | 2e3e588 | 2007-09-10 11:28:17 +0300 | [diff] [blame] | 314 | mmu_page_header_cache, 4); |
| 315 | out: |
Avi Kivity | 8c43850 | 2007-04-16 11:53:17 +0300 | [diff] [blame] | 316 | return r; |
| 317 | } |
| 318 | |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 319 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) |
| 320 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 321 | mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache); |
| 322 | mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache); |
| 323 | mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache); |
| 324 | mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 325 | } |
| 326 | |
| 327 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, |
| 328 | size_t size) |
| 329 | { |
| 330 | void *p; |
| 331 | |
| 332 | BUG_ON(!mc->nobjs); |
| 333 | p = mc->objects[--mc->nobjs]; |
| 334 | memset(p, 0, size); |
| 335 | return p; |
| 336 | } |
| 337 | |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 338 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
| 339 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 340 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache, |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 341 | sizeof(struct kvm_pte_chain)); |
| 342 | } |
| 343 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 344 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 345 | { |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 346 | kfree(pc); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 347 | } |
| 348 | |
| 349 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) |
| 350 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 351 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache, |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 352 | sizeof(struct kvm_rmap_desc)); |
| 353 | } |
| 354 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 355 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 356 | { |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 357 | kfree(rd); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 358 | } |
| 359 | |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 360 | /* |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 361 | * Return the pointer to the largepage write count for a given |
| 362 | * gfn, handling slots that are not large page aligned. |
| 363 | */ |
| 364 | static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot) |
| 365 | { |
| 366 | unsigned long idx; |
| 367 | |
| 368 | idx = (gfn / KVM_PAGES_PER_HPAGE) - |
| 369 | (slot->base_gfn / KVM_PAGES_PER_HPAGE); |
| 370 | return &slot->lpage_info[idx].write_count; |
| 371 | } |
| 372 | |
| 373 | static void account_shadowed(struct kvm *kvm, gfn_t gfn) |
| 374 | { |
| 375 | int *write_count; |
| 376 | |
| 377 | write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn)); |
| 378 | *write_count += 1; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 379 | } |
| 380 | |
| 381 | static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn) |
| 382 | { |
| 383 | int *write_count; |
| 384 | |
| 385 | write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn)); |
| 386 | *write_count -= 1; |
| 387 | WARN_ON(*write_count < 0); |
| 388 | } |
| 389 | |
| 390 | static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn) |
| 391 | { |
| 392 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 393 | int *largepage_idx; |
| 394 | |
| 395 | if (slot) { |
| 396 | largepage_idx = slot_largepage_idx(gfn, slot); |
| 397 | return *largepage_idx; |
| 398 | } |
| 399 | |
| 400 | return 1; |
| 401 | } |
| 402 | |
| 403 | static int host_largepage_backed(struct kvm *kvm, gfn_t gfn) |
| 404 | { |
| 405 | struct vm_area_struct *vma; |
| 406 | unsigned long addr; |
| 407 | |
| 408 | addr = gfn_to_hva(kvm, gfn); |
| 409 | if (kvm_is_error_hva(addr)) |
| 410 | return 0; |
| 411 | |
| 412 | vma = find_vma(current->mm, addr); |
| 413 | if (vma && is_vm_hugetlb_page(vma)) |
| 414 | return 1; |
| 415 | |
| 416 | return 0; |
| 417 | } |
| 418 | |
| 419 | static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn) |
| 420 | { |
| 421 | struct kvm_memory_slot *slot; |
| 422 | |
| 423 | if (has_wrprotected_page(vcpu->kvm, large_gfn)) |
| 424 | return 0; |
| 425 | |
| 426 | if (!host_largepage_backed(vcpu->kvm, large_gfn)) |
| 427 | return 0; |
| 428 | |
| 429 | slot = gfn_to_memslot(vcpu->kvm, large_gfn); |
| 430 | if (slot && slot->dirty_bitmap) |
| 431 | return 0; |
| 432 | |
| 433 | return 1; |
| 434 | } |
| 435 | |
| 436 | /* |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 437 | * Take gfn and return the reverse mapping to it. |
| 438 | * Note: gfn must be unaliased before this function get called |
| 439 | */ |
| 440 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 441 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage) |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 442 | { |
| 443 | struct kvm_memory_slot *slot; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 444 | unsigned long idx; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 445 | |
| 446 | slot = gfn_to_memslot(kvm, gfn); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 447 | if (!lpage) |
| 448 | return &slot->rmap[gfn - slot->base_gfn]; |
| 449 | |
| 450 | idx = (gfn / KVM_PAGES_PER_HPAGE) - |
| 451 | (slot->base_gfn / KVM_PAGES_PER_HPAGE); |
| 452 | |
| 453 | return &slot->lpage_info[idx].rmap_pde; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 454 | } |
| 455 | |
| 456 | /* |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 457 | * Reverse mapping data structures: |
| 458 | * |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 459 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
| 460 | * that points to page_address(page). |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 461 | * |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 462 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
| 463 | * containing more mappings. |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 464 | */ |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 465 | static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage) |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 466 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 467 | struct kvm_mmu_page *sp; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 468 | struct kvm_rmap_desc *desc; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 469 | unsigned long *rmapp; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 470 | int i; |
| 471 | |
| 472 | if (!is_rmap_pte(*spte)) |
| 473 | return; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 474 | gfn = unalias_gfn(vcpu->kvm, gfn); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 475 | sp = page_header(__pa(spte)); |
| 476 | sp->gfns[spte - sp->spt] = gfn; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 477 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 478 | if (!*rmapp) { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 479 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 480 | *rmapp = (unsigned long)spte; |
| 481 | } else if (!(*rmapp & 1)) { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 482 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 483 | desc = mmu_alloc_rmap_desc(vcpu); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 484 | desc->shadow_ptes[0] = (u64 *)*rmapp; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 485 | desc->shadow_ptes[1] = spte; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 486 | *rmapp = (unsigned long)desc | 1; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 487 | } else { |
| 488 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 489 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 490 | while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) |
| 491 | desc = desc->more; |
| 492 | if (desc->shadow_ptes[RMAP_EXT-1]) { |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 493 | desc->more = mmu_alloc_rmap_desc(vcpu); |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 494 | desc = desc->more; |
| 495 | } |
| 496 | for (i = 0; desc->shadow_ptes[i]; ++i) |
| 497 | ; |
| 498 | desc->shadow_ptes[i] = spte; |
| 499 | } |
| 500 | } |
| 501 | |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 502 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 503 | struct kvm_rmap_desc *desc, |
| 504 | int i, |
| 505 | struct kvm_rmap_desc *prev_desc) |
| 506 | { |
| 507 | int j; |
| 508 | |
| 509 | for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j) |
| 510 | ; |
| 511 | desc->shadow_ptes[i] = desc->shadow_ptes[j]; |
Al Viro | 11718b4d | 2007-02-09 16:39:20 +0000 | [diff] [blame] | 512 | desc->shadow_ptes[j] = NULL; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 513 | if (j != 0) |
| 514 | return; |
| 515 | if (!prev_desc && !desc->more) |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 516 | *rmapp = (unsigned long)desc->shadow_ptes[0]; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 517 | else |
| 518 | if (prev_desc) |
| 519 | prev_desc->more = desc->more; |
| 520 | else |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 521 | *rmapp = (unsigned long)desc->more | 1; |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 522 | mmu_free_rmap_desc(desc); |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 523 | } |
| 524 | |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 525 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 526 | { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 527 | struct kvm_rmap_desc *desc; |
| 528 | struct kvm_rmap_desc *prev_desc; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 529 | struct kvm_mmu_page *sp; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 530 | pfn_t pfn; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 531 | unsigned long *rmapp; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 532 | int i; |
| 533 | |
| 534 | if (!is_rmap_pte(*spte)) |
| 535 | return; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 536 | sp = page_header(__pa(spte)); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 537 | pfn = spte_to_pfn(*spte); |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 538 | if (*spte & shadow_accessed_mask) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 539 | kvm_set_pfn_accessed(pfn); |
Izik Eidus | b4231d6 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 540 | if (is_writeble_pte(*spte)) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 541 | kvm_release_pfn_dirty(pfn); |
Izik Eidus | b4231d6 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 542 | else |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 543 | kvm_release_pfn_clean(pfn); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 544 | rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte)); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 545 | if (!*rmapp) { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 546 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
| 547 | BUG(); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 548 | } else if (!(*rmapp & 1)) { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 549 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 550 | if ((u64 *)*rmapp != spte) { |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 551 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
| 552 | spte, *spte); |
| 553 | BUG(); |
| 554 | } |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 555 | *rmapp = 0; |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 556 | } else { |
| 557 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 558 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 559 | prev_desc = NULL; |
| 560 | while (desc) { |
| 561 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) |
| 562 | if (desc->shadow_ptes[i] == spte) { |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 563 | rmap_desc_remove_entry(rmapp, |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 564 | desc, i, |
Avi Kivity | cd4a4e5 | 2007-01-05 16:36:38 -0800 | [diff] [blame] | 565 | prev_desc); |
| 566 | return; |
| 567 | } |
| 568 | prev_desc = desc; |
| 569 | desc = desc->more; |
| 570 | } |
| 571 | BUG(); |
| 572 | } |
| 573 | } |
| 574 | |
Izik Eidus | 98348e9 | 2007-10-16 14:42:30 +0200 | [diff] [blame] | 575 | static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte) |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 576 | { |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 577 | struct kvm_rmap_desc *desc; |
Izik Eidus | 98348e9 | 2007-10-16 14:42:30 +0200 | [diff] [blame] | 578 | struct kvm_rmap_desc *prev_desc; |
| 579 | u64 *prev_spte; |
| 580 | int i; |
| 581 | |
| 582 | if (!*rmapp) |
| 583 | return NULL; |
| 584 | else if (!(*rmapp & 1)) { |
| 585 | if (!spte) |
| 586 | return (u64 *)*rmapp; |
| 587 | return NULL; |
| 588 | } |
| 589 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
| 590 | prev_desc = NULL; |
| 591 | prev_spte = NULL; |
| 592 | while (desc) { |
| 593 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) { |
| 594 | if (prev_spte == spte) |
| 595 | return desc->shadow_ptes[i]; |
| 596 | prev_spte = desc->shadow_ptes[i]; |
| 597 | } |
| 598 | desc = desc->more; |
| 599 | } |
| 600 | return NULL; |
| 601 | } |
| 602 | |
| 603 | static void rmap_write_protect(struct kvm *kvm, u64 gfn) |
| 604 | { |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 605 | unsigned long *rmapp; |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 606 | u64 *spte; |
Eddie Dong | caa5b8a | 2007-12-18 06:08:27 +0800 | [diff] [blame] | 607 | int write_protected = 0; |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 608 | |
Anthony Liguori | 4a4c992 | 2007-10-10 20:08:41 -0500 | [diff] [blame] | 609 | gfn = unalias_gfn(kvm, gfn); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 610 | rmapp = gfn_to_rmap(kvm, gfn, 0); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 611 | |
Izik Eidus | 98348e9 | 2007-10-16 14:42:30 +0200 | [diff] [blame] | 612 | spte = rmap_next(kvm, rmapp, NULL); |
| 613 | while (spte) { |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 614 | BUG_ON(!spte); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 615 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 616 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); |
Eddie Dong | caa5b8a | 2007-12-18 06:08:27 +0800 | [diff] [blame] | 617 | if (is_writeble_pte(*spte)) { |
Izik Eidus | 9647c14 | 2007-10-16 14:43:46 +0200 | [diff] [blame] | 618 | set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK); |
Eddie Dong | caa5b8a | 2007-12-18 06:08:27 +0800 | [diff] [blame] | 619 | write_protected = 1; |
| 620 | } |
Izik Eidus | 9647c14 | 2007-10-16 14:43:46 +0200 | [diff] [blame] | 621 | spte = rmap_next(kvm, rmapp, spte); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 622 | } |
Izik Eidus | 855149a | 2008-03-20 18:17:24 +0200 | [diff] [blame] | 623 | if (write_protected) { |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 624 | pfn_t pfn; |
Izik Eidus | 855149a | 2008-03-20 18:17:24 +0200 | [diff] [blame] | 625 | |
| 626 | spte = rmap_next(kvm, rmapp, NULL); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 627 | pfn = spte_to_pfn(*spte); |
| 628 | kvm_set_pfn_dirty(pfn); |
Izik Eidus | 855149a | 2008-03-20 18:17:24 +0200 | [diff] [blame] | 629 | } |
| 630 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 631 | /* check for huge page mappings */ |
| 632 | rmapp = gfn_to_rmap(kvm, gfn, 1); |
| 633 | spte = rmap_next(kvm, rmapp, NULL); |
| 634 | while (spte) { |
| 635 | BUG_ON(!spte); |
| 636 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
| 637 | BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)); |
| 638 | pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn); |
| 639 | if (is_writeble_pte(*spte)) { |
| 640 | rmap_remove(kvm, spte); |
| 641 | --kvm->stat.lpages; |
| 642 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
| 643 | write_protected = 1; |
| 644 | } |
| 645 | spte = rmap_next(kvm, rmapp, spte); |
| 646 | } |
| 647 | |
Eddie Dong | caa5b8a | 2007-12-18 06:08:27 +0800 | [diff] [blame] | 648 | if (write_protected) |
| 649 | kvm_flush_remote_tlbs(kvm); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 650 | |
| 651 | account_shadowed(kvm, gfn); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 652 | } |
| 653 | |
Yaozu Dong | d6c69ee | 2007-04-25 14:17:25 +0800 | [diff] [blame] | 654 | #ifdef MMU_DEBUG |
Avi Kivity | 47ad8e6 | 2007-05-06 15:50:58 +0300 | [diff] [blame] | 655 | static int is_empty_shadow_page(u64 *spt) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 656 | { |
Avi Kivity | 139bdb2 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 657 | u64 *pos; |
| 658 | u64 *end; |
| 659 | |
Avi Kivity | 47ad8e6 | 2007-05-06 15:50:58 +0300 | [diff] [blame] | 660 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
Avi Kivity | d196e34 | 2008-01-24 11:44:11 +0200 | [diff] [blame] | 661 | if (*pos != shadow_trap_nonpresent_pte) { |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 662 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
Avi Kivity | 139bdb2 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 663 | pos, *pos); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 664 | return 0; |
Avi Kivity | 139bdb2 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 665 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 666 | return 1; |
| 667 | } |
Yaozu Dong | d6c69ee | 2007-04-25 14:17:25 +0800 | [diff] [blame] | 668 | #endif |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 669 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 670 | static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
Avi Kivity | 260746c | 2007-01-05 16:36:49 -0800 | [diff] [blame] | 671 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 672 | ASSERT(is_empty_shadow_page(sp->spt)); |
| 673 | list_del(&sp->link); |
| 674 | __free_page(virt_to_page(sp->spt)); |
| 675 | __free_page(virt_to_page(sp->gfns)); |
| 676 | kfree(sp); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 677 | ++kvm->arch.n_free_mmu_pages; |
Avi Kivity | 260746c | 2007-01-05 16:36:49 -0800 | [diff] [blame] | 678 | } |
| 679 | |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 680 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
| 681 | { |
Dong, Eddie | 1ae0a13 | 2008-01-07 13:20:25 +0200 | [diff] [blame] | 682 | return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 683 | } |
| 684 | |
Avi Kivity | 25c0de2 | 2007-01-05 16:36:42 -0800 | [diff] [blame] | 685 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
| 686 | u64 *parent_pte) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 687 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 688 | struct kvm_mmu_page *sp; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 689 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 690 | sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp); |
| 691 | sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); |
| 692 | sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 693 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 694 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 695 | ASSERT(is_empty_shadow_page(sp->spt)); |
| 696 | sp->slot_bitmap = 0; |
| 697 | sp->multimapped = 0; |
| 698 | sp->parent_pte = parent_pte; |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 699 | --vcpu->kvm->arch.n_free_mmu_pages; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 700 | return sp; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 701 | } |
| 702 | |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 703 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 704 | struct kvm_mmu_page *sp, u64 *parent_pte) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 705 | { |
| 706 | struct kvm_pte_chain *pte_chain; |
| 707 | struct hlist_node *node; |
| 708 | int i; |
| 709 | |
| 710 | if (!parent_pte) |
| 711 | return; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 712 | if (!sp->multimapped) { |
| 713 | u64 *old = sp->parent_pte; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 714 | |
| 715 | if (!old) { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 716 | sp->parent_pte = parent_pte; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 717 | return; |
| 718 | } |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 719 | sp->multimapped = 1; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 720 | pte_chain = mmu_alloc_pte_chain(vcpu); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 721 | INIT_HLIST_HEAD(&sp->parent_ptes); |
| 722 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 723 | pte_chain->parent_ptes[0] = old; |
| 724 | } |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 725 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) { |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 726 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) |
| 727 | continue; |
| 728 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) |
| 729 | if (!pte_chain->parent_ptes[i]) { |
| 730 | pte_chain->parent_ptes[i] = parent_pte; |
| 731 | return; |
| 732 | } |
| 733 | } |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 734 | pte_chain = mmu_alloc_pte_chain(vcpu); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 735 | BUG_ON(!pte_chain); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 736 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 737 | pte_chain->parent_ptes[0] = parent_pte; |
| 738 | } |
| 739 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 740 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 741 | u64 *parent_pte) |
| 742 | { |
| 743 | struct kvm_pte_chain *pte_chain; |
| 744 | struct hlist_node *node; |
| 745 | int i; |
| 746 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 747 | if (!sp->multimapped) { |
| 748 | BUG_ON(sp->parent_pte != parent_pte); |
| 749 | sp->parent_pte = NULL; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 750 | return; |
| 751 | } |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 752 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 753 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { |
| 754 | if (!pte_chain->parent_ptes[i]) |
| 755 | break; |
| 756 | if (pte_chain->parent_ptes[i] != parent_pte) |
| 757 | continue; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 758 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
| 759 | && pte_chain->parent_ptes[i + 1]) { |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 760 | pte_chain->parent_ptes[i] |
| 761 | = pte_chain->parent_ptes[i + 1]; |
| 762 | ++i; |
| 763 | } |
| 764 | pte_chain->parent_ptes[i] = NULL; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 765 | if (i == 0) { |
| 766 | hlist_del(&pte_chain->link); |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 767 | mmu_free_pte_chain(pte_chain); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 768 | if (hlist_empty(&sp->parent_ptes)) { |
| 769 | sp->multimapped = 0; |
| 770 | sp->parent_pte = NULL; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 771 | } |
| 772 | } |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 773 | return; |
| 774 | } |
| 775 | BUG(); |
| 776 | } |
| 777 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 778 | static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 779 | { |
| 780 | unsigned index; |
| 781 | struct hlist_head *bucket; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 782 | struct kvm_mmu_page *sp; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 783 | struct hlist_node *node; |
| 784 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 785 | pgprintk("%s: looking for gfn %lx\n", __func__, gfn); |
Dong, Eddie | 1ae0a13 | 2008-01-07 13:20:25 +0200 | [diff] [blame] | 786 | index = kvm_page_table_hashfn(gfn); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 787 | bucket = &kvm->arch.mmu_page_hash[index]; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 788 | hlist_for_each_entry(sp, node, bucket, hash_link) |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 789 | if (sp->gfn == gfn && !sp->role.metaphysical |
| 790 | && !sp->role.invalid) { |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 791 | pgprintk("%s: found role %x\n", |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 792 | __func__, sp->role.word); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 793 | return sp; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 794 | } |
| 795 | return NULL; |
| 796 | } |
| 797 | |
| 798 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
| 799 | gfn_t gfn, |
| 800 | gva_t gaddr, |
| 801 | unsigned level, |
| 802 | int metaphysical, |
Avi Kivity | 41074d0 | 2007-12-09 17:00:02 +0200 | [diff] [blame] | 803 | unsigned access, |
Avi Kivity | f7d9c7b | 2008-02-26 22:12:10 +0200 | [diff] [blame] | 804 | u64 *parent_pte) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 805 | { |
| 806 | union kvm_mmu_page_role role; |
| 807 | unsigned index; |
| 808 | unsigned quadrant; |
| 809 | struct hlist_head *bucket; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 810 | struct kvm_mmu_page *sp; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 811 | struct hlist_node *node; |
| 812 | |
| 813 | role.word = 0; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 814 | role.glevels = vcpu->arch.mmu.root_level; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 815 | role.level = level; |
| 816 | role.metaphysical = metaphysical; |
Avi Kivity | 41074d0 | 2007-12-09 17:00:02 +0200 | [diff] [blame] | 817 | role.access = access; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 818 | if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) { |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 819 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
| 820 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; |
| 821 | role.quadrant = quadrant; |
| 822 | } |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 823 | pgprintk("%s: looking gfn %lx role %x\n", __func__, |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 824 | gfn, role.word); |
Dong, Eddie | 1ae0a13 | 2008-01-07 13:20:25 +0200 | [diff] [blame] | 825 | index = kvm_page_table_hashfn(gfn); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 826 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 827 | hlist_for_each_entry(sp, node, bucket, hash_link) |
| 828 | if (sp->gfn == gfn && sp->role.word == role.word) { |
| 829 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 830 | pgprintk("%s: found\n", __func__); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 831 | return sp; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 832 | } |
Avi Kivity | dfc5aa0 | 2007-12-18 19:47:18 +0200 | [diff] [blame] | 833 | ++vcpu->kvm->stat.mmu_cache_miss; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 834 | sp = kvm_mmu_alloc_page(vcpu, parent_pte); |
| 835 | if (!sp) |
| 836 | return sp; |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 837 | pgprintk("%s: adding gfn %lx role %x\n", __func__, gfn, role.word); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 838 | sp->gfn = gfn; |
| 839 | sp->role = role; |
| 840 | hlist_add_head(&sp->hash_link, bucket); |
Avi Kivity | 374cbac | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 841 | if (!metaphysical) |
Anthony Liguori | 4a4c992 | 2007-10-10 20:08:41 -0500 | [diff] [blame] | 842 | rmap_write_protect(vcpu->kvm, gfn); |
Marcelo Tosatti | bed1d1d | 2008-04-04 14:56:44 -0300 | [diff] [blame] | 843 | vcpu->arch.mmu.prefetch_page(vcpu, sp); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 844 | return sp; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 845 | } |
| 846 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 847 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 848 | struct kvm_mmu_page *sp) |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 849 | { |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 850 | unsigned i; |
| 851 | u64 *pt; |
| 852 | u64 ent; |
| 853 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 854 | pt = sp->spt; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 855 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 856 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 857 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 858 | if (is_shadow_present_pte(pt[i])) |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 859 | rmap_remove(kvm, &pt[i]); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 860 | pt[i] = shadow_trap_nonpresent_pte; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 861 | } |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 862 | kvm_flush_remote_tlbs(kvm); |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 863 | return; |
| 864 | } |
| 865 | |
| 866 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
| 867 | ent = pt[i]; |
| 868 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 869 | if (is_shadow_present_pte(ent)) { |
| 870 | if (!is_large_pte(ent)) { |
| 871 | ent &= PT64_BASE_ADDR_MASK; |
| 872 | mmu_page_remove_parent_pte(page_header(ent), |
| 873 | &pt[i]); |
| 874 | } else { |
| 875 | --kvm->stat.lpages; |
| 876 | rmap_remove(kvm, &pt[i]); |
| 877 | } |
| 878 | } |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 879 | pt[i] = shadow_trap_nonpresent_pte; |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 880 | } |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 881 | kvm_flush_remote_tlbs(kvm); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 882 | } |
| 883 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 884 | static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 885 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 886 | mmu_page_remove_parent_pte(sp, parent_pte); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 887 | } |
| 888 | |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 889 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
| 890 | { |
| 891 | int i; |
| 892 | |
| 893 | for (i = 0; i < KVM_MAX_VCPUS; ++i) |
| 894 | if (kvm->vcpus[i]) |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 895 | kvm->vcpus[i]->arch.last_pte_updated = NULL; |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 896 | } |
| 897 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 898 | static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 899 | { |
| 900 | u64 *parent_pte; |
| 901 | |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 902 | ++kvm->stat.mmu_shadow_zapped; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 903 | while (sp->multimapped || sp->parent_pte) { |
| 904 | if (!sp->multimapped) |
| 905 | parent_pte = sp->parent_pte; |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 906 | else { |
| 907 | struct kvm_pte_chain *chain; |
| 908 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 909 | chain = container_of(sp->parent_ptes.first, |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 910 | struct kvm_pte_chain, link); |
| 911 | parent_pte = chain->parent_ptes[0]; |
| 912 | } |
Avi Kivity | 697fe2e | 2007-01-05 16:36:46 -0800 | [diff] [blame] | 913 | BUG_ON(!parent_pte); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 914 | kvm_mmu_put_page(sp, parent_pte); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 915 | set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 916 | } |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 917 | kvm_mmu_page_unlink_children(kvm, sp); |
| 918 | if (!sp->root_count) { |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 919 | if (!sp->role.metaphysical) |
| 920 | unaccount_shadowed(kvm, sp->gfn); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 921 | hlist_del(&sp->hash_link); |
| 922 | kvm_mmu_free_page(kvm, sp); |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 923 | } else { |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 924 | list_move(&sp->link, &kvm->arch.active_mmu_pages); |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 925 | sp->role.invalid = 1; |
| 926 | kvm_reload_remote_mmus(kvm); |
| 927 | } |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 928 | kvm_mmu_reset_last_pte_updated(kvm); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 929 | } |
| 930 | |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 931 | /* |
| 932 | * Changing the number of mmu pages allocated to the vm |
| 933 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock |
| 934 | */ |
| 935 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) |
| 936 | { |
| 937 | /* |
| 938 | * If we set the number of mmu pages to be smaller be than the |
| 939 | * number of actived pages , we must to free some mmu pages before we |
| 940 | * change the value |
| 941 | */ |
| 942 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 943 | if ((kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages) > |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 944 | kvm_nr_mmu_pages) { |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 945 | int n_used_mmu_pages = kvm->arch.n_alloc_mmu_pages |
| 946 | - kvm->arch.n_free_mmu_pages; |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 947 | |
| 948 | while (n_used_mmu_pages > kvm_nr_mmu_pages) { |
| 949 | struct kvm_mmu_page *page; |
| 950 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 951 | page = container_of(kvm->arch.active_mmu_pages.prev, |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 952 | struct kvm_mmu_page, link); |
| 953 | kvm_mmu_zap_page(kvm, page); |
| 954 | n_used_mmu_pages--; |
| 955 | } |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 956 | kvm->arch.n_free_mmu_pages = 0; |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 957 | } |
| 958 | else |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 959 | kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages |
| 960 | - kvm->arch.n_alloc_mmu_pages; |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 961 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 962 | kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages; |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 963 | } |
| 964 | |
Anthony Liguori | f67a46f | 2007-10-10 19:25:50 -0500 | [diff] [blame] | 965 | static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 966 | { |
| 967 | unsigned index; |
| 968 | struct hlist_head *bucket; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 969 | struct kvm_mmu_page *sp; |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 970 | struct hlist_node *node, *n; |
| 971 | int r; |
| 972 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 973 | pgprintk("%s: looking for gfn %lx\n", __func__, gfn); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 974 | r = 0; |
Dong, Eddie | 1ae0a13 | 2008-01-07 13:20:25 +0200 | [diff] [blame] | 975 | index = kvm_page_table_hashfn(gfn); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 976 | bucket = &kvm->arch.mmu_page_hash[index]; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 977 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) |
| 978 | if (sp->gfn == gfn && !sp->role.metaphysical) { |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 979 | pgprintk("%s: gfn %lx role %x\n", __func__, gfn, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 980 | sp->role.word); |
| 981 | kvm_mmu_zap_page(kvm, sp); |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 982 | r = 1; |
| 983 | } |
| 984 | return r; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 985 | } |
| 986 | |
Anthony Liguori | f67a46f | 2007-10-10 19:25:50 -0500 | [diff] [blame] | 987 | static void mmu_unshadow(struct kvm *kvm, gfn_t gfn) |
Avi Kivity | 97a0a01 | 2007-05-31 15:08:29 +0300 | [diff] [blame] | 988 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 989 | struct kvm_mmu_page *sp; |
Avi Kivity | 97a0a01 | 2007-05-31 15:08:29 +0300 | [diff] [blame] | 990 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 991 | while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) { |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 992 | pgprintk("%s: zap %lx %x\n", __func__, gfn, sp->role.word); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 993 | kvm_mmu_zap_page(kvm, sp); |
Avi Kivity | 97a0a01 | 2007-05-31 15:08:29 +0300 | [diff] [blame] | 994 | } |
| 995 | } |
| 996 | |
Avi Kivity | 38c335f | 2007-11-21 14:20:22 +0200 | [diff] [blame] | 997 | static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 998 | { |
Avi Kivity | 38c335f | 2007-11-21 14:20:22 +0200 | [diff] [blame] | 999 | int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn)); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1000 | struct kvm_mmu_page *sp = page_header(__pa(pte)); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1001 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1002 | __set_bit(slot, &sp->slot_bitmap); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1003 | } |
| 1004 | |
Avi Kivity | 039576c | 2007-03-20 12:46:50 +0200 | [diff] [blame] | 1005 | struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) |
| 1006 | { |
Izik Eidus | 72dc67a | 2008-02-10 18:04:15 +0200 | [diff] [blame] | 1007 | struct page *page; |
| 1008 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1009 | gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva); |
Avi Kivity | 039576c | 2007-03-20 12:46:50 +0200 | [diff] [blame] | 1010 | |
| 1011 | if (gpa == UNMAPPED_GVA) |
| 1012 | return NULL; |
Izik Eidus | 72dc67a | 2008-02-10 18:04:15 +0200 | [diff] [blame] | 1013 | |
| 1014 | down_read(¤t->mm->mmap_sem); |
| 1015 | page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
| 1016 | up_read(¤t->mm->mmap_sem); |
| 1017 | |
| 1018 | return page; |
Avi Kivity | 039576c | 2007-03-20 12:46:50 +0200 | [diff] [blame] | 1019 | } |
| 1020 | |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1021 | static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, |
| 1022 | unsigned pt_access, unsigned pte_access, |
| 1023 | int user_fault, int write_fault, int dirty, |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1024 | int *ptwrite, int largepage, gfn_t gfn, |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1025 | pfn_t pfn, bool speculative) |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1026 | { |
| 1027 | u64 spte; |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1028 | int was_rmapped = 0; |
Izik Eidus | 75e68e6 | 2008-01-12 23:49:09 +0200 | [diff] [blame] | 1029 | int was_writeble = is_writeble_pte(*shadow_pte); |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1030 | |
Avi Kivity | bc750ba | 2007-12-09 18:39:41 +0200 | [diff] [blame] | 1031 | pgprintk("%s: spte %llx access %x write_fault %d" |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1032 | " user_fault %d gfn %lx\n", |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1033 | __func__, *shadow_pte, pt_access, |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1034 | write_fault, user_fault, gfn); |
| 1035 | |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1036 | if (is_rmap_pte(*shadow_pte)) { |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1037 | /* |
| 1038 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink |
| 1039 | * the parent of the now unreachable PTE. |
| 1040 | */ |
| 1041 | if (largepage && !is_large_pte(*shadow_pte)) { |
| 1042 | struct kvm_mmu_page *child; |
| 1043 | u64 pte = *shadow_pte; |
| 1044 | |
| 1045 | child = page_header(pte & PT64_BASE_ADDR_MASK); |
| 1046 | mmu_page_remove_parent_pte(child, shadow_pte); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1047 | } else if (pfn != spte_to_pfn(*shadow_pte)) { |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1048 | pgprintk("hfn old %lx new %lx\n", |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1049 | spte_to_pfn(*shadow_pte), pfn); |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1050 | rmap_remove(vcpu->kvm, shadow_pte); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1051 | } else { |
| 1052 | if (largepage) |
| 1053 | was_rmapped = is_large_pte(*shadow_pte); |
| 1054 | else |
| 1055 | was_rmapped = 1; |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1056 | } |
Marcelo Tosatti | 15aaa81 | 2008-03-17 10:08:18 -0300 | [diff] [blame] | 1057 | } |
| 1058 | |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1059 | /* |
| 1060 | * We don't set the accessed bit, since we sometimes want to see |
| 1061 | * whether the guest actually used the pte (in order to detect |
| 1062 | * demand paging). |
| 1063 | */ |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 1064 | spte = shadow_base_present_pte | shadow_dirty_mask; |
Avi Kivity | 947da53 | 2008-03-18 11:05:52 +0200 | [diff] [blame] | 1065 | if (!speculative) |
| 1066 | pte_access |= PT_ACCESSED_MASK; |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1067 | if (!dirty) |
| 1068 | pte_access &= ~ACC_WRITE_MASK; |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 1069 | if (pte_access & ACC_EXEC_MASK) |
| 1070 | spte |= shadow_x_mask; |
| 1071 | else |
| 1072 | spte |= shadow_nx_mask; |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1073 | if (pte_access & ACC_USER_MASK) |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 1074 | spte |= shadow_user_mask; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1075 | if (largepage) |
| 1076 | spte |= PT_PAGE_SIZE_MASK; |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1077 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1078 | spte |= (u64)pfn << PAGE_SHIFT; |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1079 | |
| 1080 | if ((pte_access & ACC_WRITE_MASK) |
| 1081 | || (write_fault && !is_write_protection(vcpu) && !user_fault)) { |
| 1082 | struct kvm_mmu_page *shadow; |
| 1083 | |
| 1084 | spte |= PT_WRITABLE_MASK; |
| 1085 | if (user_fault) { |
| 1086 | mmu_unshadow(vcpu->kvm, gfn); |
| 1087 | goto unshadowed; |
| 1088 | } |
| 1089 | |
| 1090 | shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1091 | if (shadow || |
| 1092 | (largepage && has_wrprotected_page(vcpu->kvm, gfn))) { |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1093 | pgprintk("%s: found shadow page for %lx, marking ro\n", |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1094 | __func__, gfn); |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1095 | pte_access &= ~ACC_WRITE_MASK; |
| 1096 | if (is_writeble_pte(spte)) { |
| 1097 | spte &= ~PT_WRITABLE_MASK; |
| 1098 | kvm_x86_ops->tlb_flush(vcpu); |
| 1099 | } |
| 1100 | if (write_fault) |
| 1101 | *ptwrite = 1; |
| 1102 | } |
| 1103 | } |
| 1104 | |
| 1105 | unshadowed: |
| 1106 | |
| 1107 | if (pte_access & ACC_WRITE_MASK) |
| 1108 | mark_page_dirty(vcpu->kvm, gfn); |
| 1109 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1110 | pgprintk("%s: setting spte %llx\n", __func__, spte); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1111 | pgprintk("instantiating %s PTE (%s) at %d (%llx) addr %llx\n", |
| 1112 | (spte&PT_PAGE_SIZE_MASK)? "2MB" : "4kB", |
| 1113 | (spte&PT_WRITABLE_MASK)?"RW":"R", gfn, spte, shadow_pte); |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1114 | set_shadow_pte(shadow_pte, spte); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1115 | if (!was_rmapped && (spte & PT_PAGE_SIZE_MASK) |
| 1116 | && (spte & PT_PRESENT_MASK)) |
| 1117 | ++vcpu->kvm->stat.lpages; |
| 1118 | |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1119 | page_header_update_slot(vcpu->kvm, shadow_pte, gfn); |
| 1120 | if (!was_rmapped) { |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1121 | rmap_add(vcpu, shadow_pte, gfn, largepage); |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1122 | if (!is_rmap_pte(*shadow_pte)) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1123 | kvm_release_pfn_clean(pfn); |
Izik Eidus | 75e68e6 | 2008-01-12 23:49:09 +0200 | [diff] [blame] | 1124 | } else { |
| 1125 | if (was_writeble) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1126 | kvm_release_pfn_dirty(pfn); |
Izik Eidus | 75e68e6 | 2008-01-12 23:49:09 +0200 | [diff] [blame] | 1127 | else |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1128 | kvm_release_pfn_clean(pfn); |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1129 | } |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1130 | if (!ptwrite || !*ptwrite) |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1131 | vcpu->arch.last_pte_updated = shadow_pte; |
Avi Kivity | 1c4f1fd | 2007-12-09 17:40:31 +0200 | [diff] [blame] | 1132 | } |
| 1133 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1134 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
| 1135 | { |
| 1136 | } |
| 1137 | |
Joerg Roedel | 4d9976b | 2008-02-07 13:47:42 +0100 | [diff] [blame] | 1138 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1139 | int largepage, gfn_t gfn, pfn_t pfn, |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1140 | int level) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1141 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1142 | hpa_t table_addr = vcpu->arch.mmu.root_hpa; |
Avi Kivity | e833240 | 2007-12-09 18:43:00 +0200 | [diff] [blame] | 1143 | int pt_write = 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1144 | |
| 1145 | for (; ; level--) { |
| 1146 | u32 index = PT64_INDEX(v, level); |
| 1147 | u64 *table; |
| 1148 | |
| 1149 | ASSERT(VALID_PAGE(table_addr)); |
| 1150 | table = __va(table_addr); |
| 1151 | |
| 1152 | if (level == 1) { |
Avi Kivity | e833240 | 2007-12-09 18:43:00 +0200 | [diff] [blame] | 1153 | mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL, |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1154 | 0, write, 1, &pt_write, 0, gfn, pfn, false); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1155 | return pt_write; |
| 1156 | } |
| 1157 | |
| 1158 | if (largepage && level == 2) { |
| 1159 | mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL, |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1160 | 0, write, 1, &pt_write, 1, gfn, pfn, false); |
Avi Kivity | d196e34 | 2008-01-24 11:44:11 +0200 | [diff] [blame] | 1161 | return pt_write; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1162 | } |
| 1163 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1164 | if (table[index] == shadow_trap_nonpresent_pte) { |
Avi Kivity | 25c0de2 | 2007-01-05 16:36:42 -0800 | [diff] [blame] | 1165 | struct kvm_mmu_page *new_table; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1166 | gfn_t pseudo_gfn; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1167 | |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1168 | pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK) |
| 1169 | >> PAGE_SHIFT; |
| 1170 | new_table = kvm_mmu_get_page(vcpu, pseudo_gfn, |
| 1171 | v, level - 1, |
Avi Kivity | f7d9c7b | 2008-02-26 22:12:10 +0200 | [diff] [blame] | 1172 | 1, ACC_ALL, &table[index]); |
Avi Kivity | 25c0de2 | 2007-01-05 16:36:42 -0800 | [diff] [blame] | 1173 | if (!new_table) { |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1174 | pgprintk("nonpaging_map: ENOMEM\n"); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1175 | kvm_release_pfn_clean(pfn); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1176 | return -ENOMEM; |
| 1177 | } |
| 1178 | |
Sheng Yang | 1439442 | 2008-04-28 12:24:45 +0800 | [diff] [blame] | 1179 | table[index] = __pa(new_table->spt) |
| 1180 | | PT_PRESENT_MASK | PT_WRITABLE_MASK |
| 1181 | | shadow_user_mask | shadow_x_mask; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1182 | } |
| 1183 | table_addr = table[index] & PT64_BASE_ADDR_MASK; |
| 1184 | } |
| 1185 | } |
| 1186 | |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1187 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn) |
| 1188 | { |
| 1189 | int r; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1190 | int largepage = 0; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1191 | pfn_t pfn; |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1192 | |
| 1193 | down_read(¤t->mm->mmap_sem); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1194 | if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { |
| 1195 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); |
| 1196 | largepage = 1; |
| 1197 | } |
| 1198 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1199 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
Izik Eidus | 72dc67a | 2008-02-10 18:04:15 +0200 | [diff] [blame] | 1200 | up_read(¤t->mm->mmap_sem); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1201 | |
Avi Kivity | d196e34 | 2008-01-24 11:44:11 +0200 | [diff] [blame] | 1202 | /* mmio */ |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1203 | if (is_error_pfn(pfn)) { |
| 1204 | kvm_release_pfn_clean(pfn); |
Avi Kivity | d196e34 | 2008-01-24 11:44:11 +0200 | [diff] [blame] | 1205 | return 1; |
| 1206 | } |
| 1207 | |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1208 | spin_lock(&vcpu->kvm->mmu_lock); |
Avi Kivity | eb787d1 | 2007-12-31 15:27:49 +0200 | [diff] [blame] | 1209 | kvm_mmu_free_some_pages(vcpu); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1210 | r = __direct_map(vcpu, v, write, largepage, gfn, pfn, |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1211 | PT32E_ROOT_LEVEL); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1212 | spin_unlock(&vcpu->kvm->mmu_lock); |
| 1213 | |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1214 | |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1215 | return r; |
| 1216 | } |
| 1217 | |
| 1218 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1219 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
| 1220 | struct kvm_mmu_page *sp) |
| 1221 | { |
| 1222 | int i; |
| 1223 | |
| 1224 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
| 1225 | sp->spt[i] = shadow_trap_nonpresent_pte; |
| 1226 | } |
| 1227 | |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1228 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
| 1229 | { |
| 1230 | int i; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1231 | struct kvm_mmu_page *sp; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1232 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1233 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
Avi Kivity | 7b53aa5 | 2007-06-05 12:17:03 +0300 | [diff] [blame] | 1234 | return; |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1235 | spin_lock(&vcpu->kvm->mmu_lock); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1236 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
| 1237 | hpa_t root = vcpu->arch.mmu.root_hpa; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1238 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1239 | sp = page_header(root); |
| 1240 | --sp->root_count; |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 1241 | if (!sp->root_count && sp->role.invalid) |
| 1242 | kvm_mmu_zap_page(vcpu->kvm, sp); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1243 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1244 | spin_unlock(&vcpu->kvm->mmu_lock); |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1245 | return; |
| 1246 | } |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1247 | for (i = 0; i < 4; ++i) { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1248 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1249 | |
Avi Kivity | 417726a | 2007-04-12 17:35:58 +0300 | [diff] [blame] | 1250 | if (root) { |
Avi Kivity | 417726a | 2007-04-12 17:35:58 +0300 | [diff] [blame] | 1251 | root &= PT64_BASE_ADDR_MASK; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1252 | sp = page_header(root); |
| 1253 | --sp->root_count; |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 1254 | if (!sp->root_count && sp->role.invalid) |
| 1255 | kvm_mmu_zap_page(vcpu->kvm, sp); |
Avi Kivity | 417726a | 2007-04-12 17:35:58 +0300 | [diff] [blame] | 1256 | } |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1257 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1258 | } |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1259 | spin_unlock(&vcpu->kvm->mmu_lock); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1260 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1261 | } |
| 1262 | |
| 1263 | static void mmu_alloc_roots(struct kvm_vcpu *vcpu) |
| 1264 | { |
| 1265 | int i; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1266 | gfn_t root_gfn; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1267 | struct kvm_mmu_page *sp; |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1268 | int metaphysical = 0; |
Avi Kivity | 3bb65a2 | 2007-01-05 16:36:51 -0800 | [diff] [blame] | 1269 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1270 | root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1271 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1272 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
| 1273 | hpa_t root = vcpu->arch.mmu.root_hpa; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1274 | |
| 1275 | ASSERT(!VALID_PAGE(root)); |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1276 | if (tdp_enabled) |
| 1277 | metaphysical = 1; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1278 | sp = kvm_mmu_get_page(vcpu, root_gfn, 0, |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1279 | PT64_ROOT_LEVEL, metaphysical, |
| 1280 | ACC_ALL, NULL); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1281 | root = __pa(sp->spt); |
| 1282 | ++sp->root_count; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1283 | vcpu->arch.mmu.root_hpa = root; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1284 | return; |
| 1285 | } |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1286 | metaphysical = !is_paging(vcpu); |
| 1287 | if (tdp_enabled) |
| 1288 | metaphysical = 1; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1289 | for (i = 0; i < 4; ++i) { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1290 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1291 | |
| 1292 | ASSERT(!VALID_PAGE(root)); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1293 | if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) { |
| 1294 | if (!is_present_pte(vcpu->arch.pdptrs[i])) { |
| 1295 | vcpu->arch.mmu.pae_root[i] = 0; |
Avi Kivity | 417726a | 2007-04-12 17:35:58 +0300 | [diff] [blame] | 1296 | continue; |
| 1297 | } |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1298 | root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT; |
| 1299 | } else if (vcpu->arch.mmu.root_level == 0) |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1300 | root_gfn = 0; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1301 | sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1302 | PT32_ROOT_LEVEL, metaphysical, |
Avi Kivity | f7d9c7b | 2008-02-26 22:12:10 +0200 | [diff] [blame] | 1303 | ACC_ALL, NULL); |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1304 | root = __pa(sp->spt); |
| 1305 | ++sp->root_count; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1306 | vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1307 | } |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1308 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1309 | } |
| 1310 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1311 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) |
| 1312 | { |
| 1313 | return vaddr; |
| 1314 | } |
| 1315 | |
| 1316 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, |
Avi Kivity | 3f3e712 | 2007-11-21 14:54:16 +0200 | [diff] [blame] | 1317 | u32 error_code) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1318 | { |
Avi Kivity | e833240 | 2007-12-09 18:43:00 +0200 | [diff] [blame] | 1319 | gfn_t gfn; |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 1320 | int r; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1321 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1322 | pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code); |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 1323 | r = mmu_topup_memory_caches(vcpu); |
| 1324 | if (r) |
| 1325 | return r; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 1326 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1327 | ASSERT(vcpu); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1328 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1329 | |
Avi Kivity | e833240 | 2007-12-09 18:43:00 +0200 | [diff] [blame] | 1330 | gfn = gva >> PAGE_SHIFT; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1331 | |
Avi Kivity | e833240 | 2007-12-09 18:43:00 +0200 | [diff] [blame] | 1332 | return nonpaging_map(vcpu, gva & PAGE_MASK, |
| 1333 | error_code & PFERR_WRITE_MASK, gfn); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1334 | } |
| 1335 | |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1336 | static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, |
| 1337 | u32 error_code) |
| 1338 | { |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1339 | pfn_t pfn; |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1340 | int r; |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1341 | int largepage = 0; |
| 1342 | gfn_t gfn = gpa >> PAGE_SHIFT; |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1343 | |
| 1344 | ASSERT(vcpu); |
| 1345 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
| 1346 | |
| 1347 | r = mmu_topup_memory_caches(vcpu); |
| 1348 | if (r) |
| 1349 | return r; |
| 1350 | |
| 1351 | down_read(¤t->mm->mmap_sem); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1352 | if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { |
| 1353 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); |
| 1354 | largepage = 1; |
| 1355 | } |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1356 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
Marcelo Tosatti | 3200f40 | 2008-03-29 20:17:59 -0300 | [diff] [blame] | 1357 | up_read(¤t->mm->mmap_sem); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1358 | if (is_error_pfn(pfn)) { |
| 1359 | kvm_release_pfn_clean(pfn); |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1360 | return 1; |
| 1361 | } |
| 1362 | spin_lock(&vcpu->kvm->mmu_lock); |
| 1363 | kvm_mmu_free_some_pages(vcpu); |
| 1364 | r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK, |
Sheng Yang | 67253af | 2008-04-25 10:20:22 +0800 | [diff] [blame] | 1365 | largepage, gfn, pfn, kvm_x86_ops->get_tdp_level()); |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1366 | spin_unlock(&vcpu->kvm->mmu_lock); |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1367 | |
| 1368 | return r; |
| 1369 | } |
| 1370 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1371 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
| 1372 | { |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1373 | mmu_free_roots(vcpu); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1374 | } |
| 1375 | |
| 1376 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) |
| 1377 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1378 | struct kvm_mmu *context = &vcpu->arch.mmu; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1379 | |
| 1380 | context->new_cr3 = nonpaging_new_cr3; |
| 1381 | context->page_fault = nonpaging_page_fault; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1382 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
| 1383 | context->free = nonpaging_free; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1384 | context->prefetch_page = nonpaging_prefetch_page; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1385 | context->root_level = 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1386 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1387 | context->root_hpa = INVALID_PAGE; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1388 | return 0; |
| 1389 | } |
| 1390 | |
Avi Kivity | d835dfe | 2007-11-21 02:57:59 +0200 | [diff] [blame] | 1391 | void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1392 | { |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 1393 | ++vcpu->stat.tlb_flush; |
Christian Ehrhardt | cbdd1be | 2007-09-09 15:41:59 +0300 | [diff] [blame] | 1394 | kvm_x86_ops->tlb_flush(vcpu); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1395 | } |
| 1396 | |
| 1397 | static void paging_new_cr3(struct kvm_vcpu *vcpu) |
| 1398 | { |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1399 | pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 1400 | mmu_free_roots(vcpu); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1401 | } |
| 1402 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1403 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
| 1404 | u64 addr, |
| 1405 | u32 err_code) |
| 1406 | { |
Avi Kivity | c3c91fe | 2007-11-25 14:04:58 +0200 | [diff] [blame] | 1407 | kvm_inject_page_fault(vcpu, addr, err_code); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1408 | } |
| 1409 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1410 | static void paging_free(struct kvm_vcpu *vcpu) |
| 1411 | { |
| 1412 | nonpaging_free(vcpu); |
| 1413 | } |
| 1414 | |
| 1415 | #define PTTYPE 64 |
| 1416 | #include "paging_tmpl.h" |
| 1417 | #undef PTTYPE |
| 1418 | |
| 1419 | #define PTTYPE 32 |
| 1420 | #include "paging_tmpl.h" |
| 1421 | #undef PTTYPE |
| 1422 | |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1423 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1424 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1425 | struct kvm_mmu *context = &vcpu->arch.mmu; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1426 | |
| 1427 | ASSERT(is_pae(vcpu)); |
| 1428 | context->new_cr3 = paging_new_cr3; |
| 1429 | context->page_fault = paging64_page_fault; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1430 | context->gva_to_gpa = paging64_gva_to_gpa; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1431 | context->prefetch_page = paging64_prefetch_page; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1432 | context->free = paging_free; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1433 | context->root_level = level; |
| 1434 | context->shadow_root_level = level; |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1435 | context->root_hpa = INVALID_PAGE; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1436 | return 0; |
| 1437 | } |
| 1438 | |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1439 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
| 1440 | { |
| 1441 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); |
| 1442 | } |
| 1443 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1444 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
| 1445 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1446 | struct kvm_mmu *context = &vcpu->arch.mmu; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1447 | |
| 1448 | context->new_cr3 = paging_new_cr3; |
| 1449 | context->page_fault = paging32_page_fault; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1450 | context->gva_to_gpa = paging32_gva_to_gpa; |
| 1451 | context->free = paging_free; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1452 | context->prefetch_page = paging32_prefetch_page; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1453 | context->root_level = PT32_ROOT_LEVEL; |
| 1454 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1455 | context->root_hpa = INVALID_PAGE; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1456 | return 0; |
| 1457 | } |
| 1458 | |
| 1459 | static int paging32E_init_context(struct kvm_vcpu *vcpu) |
| 1460 | { |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1461 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1462 | } |
| 1463 | |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1464 | static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
| 1465 | { |
| 1466 | struct kvm_mmu *context = &vcpu->arch.mmu; |
| 1467 | |
| 1468 | context->new_cr3 = nonpaging_new_cr3; |
| 1469 | context->page_fault = tdp_page_fault; |
| 1470 | context->free = nonpaging_free; |
| 1471 | context->prefetch_page = nonpaging_prefetch_page; |
Sheng Yang | 67253af | 2008-04-25 10:20:22 +0800 | [diff] [blame] | 1472 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1473 | context->root_hpa = INVALID_PAGE; |
| 1474 | |
| 1475 | if (!is_paging(vcpu)) { |
| 1476 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
| 1477 | context->root_level = 0; |
| 1478 | } else if (is_long_mode(vcpu)) { |
| 1479 | context->gva_to_gpa = paging64_gva_to_gpa; |
| 1480 | context->root_level = PT64_ROOT_LEVEL; |
| 1481 | } else if (is_pae(vcpu)) { |
| 1482 | context->gva_to_gpa = paging64_gva_to_gpa; |
| 1483 | context->root_level = PT32E_ROOT_LEVEL; |
| 1484 | } else { |
| 1485 | context->gva_to_gpa = paging32_gva_to_gpa; |
| 1486 | context->root_level = PT32_ROOT_LEVEL; |
| 1487 | } |
| 1488 | |
| 1489 | return 0; |
| 1490 | } |
| 1491 | |
| 1492 | static int init_kvm_softmmu(struct kvm_vcpu *vcpu) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1493 | { |
| 1494 | ASSERT(vcpu); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1495 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1496 | |
| 1497 | if (!is_paging(vcpu)) |
| 1498 | return nonpaging_init_context(vcpu); |
Avi Kivity | a9058ec | 2006-12-29 16:49:37 -0800 | [diff] [blame] | 1499 | else if (is_long_mode(vcpu)) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1500 | return paging64_init_context(vcpu); |
| 1501 | else if (is_pae(vcpu)) |
| 1502 | return paging32E_init_context(vcpu); |
| 1503 | else |
| 1504 | return paging32_init_context(vcpu); |
| 1505 | } |
| 1506 | |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1507 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) |
| 1508 | { |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1509 | vcpu->arch.update_pte.pfn = bad_pfn; |
| 1510 | |
Joerg Roedel | fb72d16 | 2008-02-07 13:47:44 +0100 | [diff] [blame] | 1511 | if (tdp_enabled) |
| 1512 | return init_kvm_tdp_mmu(vcpu); |
| 1513 | else |
| 1514 | return init_kvm_softmmu(vcpu); |
| 1515 | } |
| 1516 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1517 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) |
| 1518 | { |
| 1519 | ASSERT(vcpu); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1520 | if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) { |
| 1521 | vcpu->arch.mmu.free(vcpu); |
| 1522 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1523 | } |
| 1524 | } |
| 1525 | |
| 1526 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) |
| 1527 | { |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1528 | destroy_kvm_mmu(vcpu); |
| 1529 | return init_kvm_mmu(vcpu); |
| 1530 | } |
Eddie Dong | 8668a3c | 2007-10-10 14:26:45 +0800 | [diff] [blame] | 1531 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1532 | |
| 1533 | int kvm_mmu_load(struct kvm_vcpu *vcpu) |
| 1534 | { |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 1535 | int r; |
| 1536 | |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 1537 | r = mmu_topup_memory_caches(vcpu); |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1538 | if (r) |
| 1539 | goto out; |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1540 | spin_lock(&vcpu->kvm->mmu_lock); |
Avi Kivity | eb787d1 | 2007-12-31 15:27:49 +0200 | [diff] [blame] | 1541 | kvm_mmu_free_some_pages(vcpu); |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1542 | mmu_alloc_roots(vcpu); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1543 | spin_unlock(&vcpu->kvm->mmu_lock); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1544 | kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa); |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1545 | kvm_mmu_flush_tlb(vcpu); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 1546 | out: |
| 1547 | return r; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1548 | } |
Avi Kivity | 17c3ba9 | 2007-06-04 15:58:30 +0300 | [diff] [blame] | 1549 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
| 1550 | |
| 1551 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) |
| 1552 | { |
| 1553 | mmu_free_roots(vcpu); |
| 1554 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1555 | |
Avi Kivity | 09072da | 2007-05-01 14:16:52 +0300 | [diff] [blame] | 1556 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1557 | struct kvm_mmu_page *sp, |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1558 | u64 *spte) |
| 1559 | { |
| 1560 | u64 pte; |
| 1561 | struct kvm_mmu_page *child; |
| 1562 | |
| 1563 | pte = *spte; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1564 | if (is_shadow_present_pte(pte)) { |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1565 | if (sp->role.level == PT_PAGE_TABLE_LEVEL || |
| 1566 | is_large_pte(pte)) |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 1567 | rmap_remove(vcpu->kvm, spte); |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1568 | else { |
| 1569 | child = page_header(pte & PT64_BASE_ADDR_MASK); |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 1570 | mmu_page_remove_parent_pte(child, spte); |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1571 | } |
| 1572 | } |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1573 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1574 | if (is_large_pte(pte)) |
| 1575 | --vcpu->kvm->stat.lpages; |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1576 | } |
| 1577 | |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1578 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1579 | struct kvm_mmu_page *sp, |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1580 | u64 *spte, |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1581 | const void *new) |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1582 | { |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1583 | if ((sp->role.level != PT_PAGE_TABLE_LEVEL) |
| 1584 | && !vcpu->arch.update_pte.largepage) { |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1585 | ++vcpu->kvm->stat.mmu_pde_zapped; |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1586 | return; |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1587 | } |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1588 | |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1589 | ++vcpu->kvm->stat.mmu_pte_updated; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1590 | if (sp->role.glevels == PT32_ROOT_LEVEL) |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1591 | paging32_update_pte(vcpu, sp, spte, new); |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1592 | else |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1593 | paging64_update_pte(vcpu, sp, spte, new); |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 1594 | } |
| 1595 | |
Avi Kivity | 79539ce | 2007-11-21 02:06:21 +0200 | [diff] [blame] | 1596 | static bool need_remote_flush(u64 old, u64 new) |
| 1597 | { |
| 1598 | if (!is_shadow_present_pte(old)) |
| 1599 | return false; |
| 1600 | if (!is_shadow_present_pte(new)) |
| 1601 | return true; |
| 1602 | if ((old ^ new) & PT64_BASE_ADDR_MASK) |
| 1603 | return true; |
| 1604 | old ^= PT64_NX_MASK; |
| 1605 | new ^= PT64_NX_MASK; |
| 1606 | return (old & ~new & PT64_PERM_MASK) != 0; |
| 1607 | } |
| 1608 | |
| 1609 | static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new) |
| 1610 | { |
| 1611 | if (need_remote_flush(old, new)) |
| 1612 | kvm_flush_remote_tlbs(vcpu->kvm); |
| 1613 | else |
| 1614 | kvm_mmu_flush_tlb(vcpu); |
| 1615 | } |
| 1616 | |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 1617 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
| 1618 | { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1619 | u64 *spte = vcpu->arch.last_pte_updated; |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 1620 | |
Sheng Yang | 7b52345 | 2008-04-25 21:13:50 +0800 | [diff] [blame] | 1621 | return !!(spte && (*spte & shadow_accessed_mask)); |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 1622 | } |
| 1623 | |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1624 | static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
| 1625 | const u8 *new, int bytes) |
| 1626 | { |
| 1627 | gfn_t gfn; |
| 1628 | int r; |
| 1629 | u64 gpte = 0; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1630 | pfn_t pfn; |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1631 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1632 | vcpu->arch.update_pte.largepage = 0; |
| 1633 | |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1634 | if (bytes != 4 && bytes != 8) |
| 1635 | return; |
| 1636 | |
| 1637 | /* |
| 1638 | * Assume that the pte write on a page table of the same type |
| 1639 | * as the current vcpu paging mode. This is nearly always true |
| 1640 | * (might be false while changing modes). Note it is verified later |
| 1641 | * by update_pte(). |
| 1642 | */ |
| 1643 | if (is_pae(vcpu)) { |
| 1644 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
| 1645 | if ((bytes == 4) && (gpa % 4 == 0)) { |
| 1646 | r = kvm_read_guest(vcpu->kvm, gpa & ~(u64)7, &gpte, 8); |
| 1647 | if (r) |
| 1648 | return; |
| 1649 | memcpy((void *)&gpte + (gpa % 8), new, 4); |
| 1650 | } else if ((bytes == 8) && (gpa % 8 == 0)) { |
| 1651 | memcpy((void *)&gpte, new, 8); |
| 1652 | } |
| 1653 | } else { |
| 1654 | if ((bytes == 4) && (gpa % 4 == 0)) |
| 1655 | memcpy((void *)&gpte, new, 4); |
| 1656 | } |
| 1657 | if (!is_present_pte(gpte)) |
| 1658 | return; |
| 1659 | gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
Izik Eidus | 72dc67a | 2008-02-10 18:04:15 +0200 | [diff] [blame] | 1660 | |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1661 | down_read(¤t->mm->mmap_sem); |
| 1662 | if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) { |
| 1663 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); |
| 1664 | vcpu->arch.update_pte.largepage = 1; |
| 1665 | } |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1666 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 1667 | up_read(¤t->mm->mmap_sem); |
Izik Eidus | 72dc67a | 2008-02-10 18:04:15 +0200 | [diff] [blame] | 1668 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1669 | if (is_error_pfn(pfn)) { |
| 1670 | kvm_release_pfn_clean(pfn); |
Avi Kivity | d196e34 | 2008-01-24 11:44:11 +0200 | [diff] [blame] | 1671 | return; |
| 1672 | } |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1673 | vcpu->arch.update_pte.gfn = gfn; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1674 | vcpu->arch.update_pte.pfn = pfn; |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1675 | } |
| 1676 | |
Avi Kivity | 09072da | 2007-05-01 14:16:52 +0300 | [diff] [blame] | 1677 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
Shaohua Li | fe55188 | 2007-07-23 14:51:39 +0800 | [diff] [blame] | 1678 | const u8 *new, int bytes) |
Avi Kivity | da4a00f | 2007-01-05 16:36:44 -0800 | [diff] [blame] | 1679 | { |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1680 | gfn_t gfn = gpa >> PAGE_SHIFT; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1681 | struct kvm_mmu_page *sp; |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1682 | struct hlist_node *node, *n; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1683 | struct hlist_head *bucket; |
| 1684 | unsigned index; |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1685 | u64 entry, gentry; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1686 | u64 *spte; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1687 | unsigned offset = offset_in_page(gpa); |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1688 | unsigned pte_size; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1689 | unsigned page_offset; |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1690 | unsigned misaligned; |
Avi Kivity | fce0657 | 2007-05-01 16:44:05 +0300 | [diff] [blame] | 1691 | unsigned quadrant; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1692 | int level; |
Avi Kivity | 86a5ba0 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 1693 | int flooded = 0; |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1694 | int npte; |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1695 | int r; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1696 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 1697 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1698 | mmu_guess_page_from_pte_write(vcpu, gpa, new, bytes); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1699 | spin_lock(&vcpu->kvm->mmu_lock); |
Avi Kivity | eb787d1 | 2007-12-31 15:27:49 +0200 | [diff] [blame] | 1700 | kvm_mmu_free_some_pages(vcpu); |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1701 | ++vcpu->kvm->stat.mmu_pte_write; |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1702 | kvm_mmu_audit(vcpu, "pre pte write"); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1703 | if (gfn == vcpu->arch.last_pt_write_gfn |
Avi Kivity | 12b7d28 | 2007-09-23 14:10:49 +0200 | [diff] [blame] | 1704 | && !last_updated_pte_accessed(vcpu)) { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1705 | ++vcpu->arch.last_pt_write_count; |
| 1706 | if (vcpu->arch.last_pt_write_count >= 3) |
Avi Kivity | 86a5ba0 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 1707 | flooded = 1; |
| 1708 | } else { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1709 | vcpu->arch.last_pt_write_gfn = gfn; |
| 1710 | vcpu->arch.last_pt_write_count = 1; |
| 1711 | vcpu->arch.last_pte_updated = NULL; |
Avi Kivity | 86a5ba0 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 1712 | } |
Dong, Eddie | 1ae0a13 | 2008-01-07 13:20:25 +0200 | [diff] [blame] | 1713 | index = kvm_page_table_hashfn(gfn); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1714 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1715 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) { |
| 1716 | if (sp->gfn != gfn || sp->role.metaphysical) |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1717 | continue; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1718 | pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1719 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
Avi Kivity | e925c5b | 2007-04-30 14:47:02 +0300 | [diff] [blame] | 1720 | misaligned |= bytes < 4; |
Avi Kivity | 86a5ba0 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 1721 | if (misaligned || flooded) { |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1722 | /* |
| 1723 | * Misaligned accesses are too much trouble to fix |
| 1724 | * up; also, they usually indicate a page is not used |
| 1725 | * as a page table. |
Avi Kivity | 86a5ba0 | 2007-01-05 16:36:50 -0800 | [diff] [blame] | 1726 | * |
| 1727 | * If we're seeing too many writes to a page, |
| 1728 | * it may no longer be a page table, or we may be |
| 1729 | * forking, in which case it is better to unmap the |
| 1730 | * page. |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1731 | */ |
| 1732 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1733 | gpa, bytes, sp->role.word); |
| 1734 | kvm_mmu_zap_page(vcpu->kvm, sp); |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1735 | ++vcpu->kvm->stat.mmu_flooded; |
Avi Kivity | 0e7bc4b | 2007-01-05 16:36:48 -0800 | [diff] [blame] | 1736 | continue; |
| 1737 | } |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1738 | page_offset = offset; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1739 | level = sp->role.level; |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1740 | npte = 1; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1741 | if (sp->role.glevels == PT32_ROOT_LEVEL) { |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1742 | page_offset <<= 1; /* 32->64 */ |
| 1743 | /* |
| 1744 | * A 32-bit pde maps 4MB while the shadow pdes map |
| 1745 | * only 2MB. So we need to double the offset again |
| 1746 | * and zap two pdes instead of one. |
| 1747 | */ |
| 1748 | if (level == PT32_ROOT_LEVEL) { |
Avi Kivity | 6b8d0f9 | 2007-04-18 11:18:18 +0300 | [diff] [blame] | 1749 | page_offset &= ~7; /* kill rounding error */ |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1750 | page_offset <<= 1; |
| 1751 | npte = 2; |
| 1752 | } |
Avi Kivity | fce0657 | 2007-05-01 16:44:05 +0300 | [diff] [blame] | 1753 | quadrant = page_offset >> PAGE_SHIFT; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1754 | page_offset &= ~PAGE_MASK; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1755 | if (quadrant != sp->role.quadrant) |
Avi Kivity | fce0657 | 2007-05-01 16:44:05 +0300 | [diff] [blame] | 1756 | continue; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1757 | } |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1758 | spte = &sp->spt[page_offset / sizeof(*spte)]; |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1759 | if ((gpa & (pte_size - 1)) || (bytes < pte_size)) { |
| 1760 | gentry = 0; |
| 1761 | r = kvm_read_guest_atomic(vcpu->kvm, |
| 1762 | gpa & ~(u64)(pte_size - 1), |
| 1763 | &gentry, pte_size); |
| 1764 | new = (const void *)&gentry; |
| 1765 | if (r < 0) |
| 1766 | new = NULL; |
| 1767 | } |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1768 | while (npte--) { |
Avi Kivity | 79539ce | 2007-11-21 02:06:21 +0200 | [diff] [blame] | 1769 | entry = *spte; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1770 | mmu_pte_write_zap_pte(vcpu, sp, spte); |
Dong, Eddie | 489f1d6 | 2008-01-07 11:14:20 +0200 | [diff] [blame] | 1771 | if (new) |
| 1772 | mmu_pte_write_new_pte(vcpu, sp, spte, new); |
Avi Kivity | 79539ce | 2007-11-21 02:06:21 +0200 | [diff] [blame] | 1773 | mmu_pte_write_flush_tlb(vcpu, entry, *spte); |
Avi Kivity | ac1b714 | 2007-03-08 17:13:32 +0200 | [diff] [blame] | 1774 | ++spte; |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1775 | } |
Avi Kivity | 9b7a032 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1776 | } |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 1777 | kvm_mmu_audit(vcpu, "post pte write"); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1778 | spin_unlock(&vcpu->kvm->mmu_lock); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1779 | if (!is_error_pfn(vcpu->arch.update_pte.pfn)) { |
| 1780 | kvm_release_pfn_clean(vcpu->arch.update_pte.pfn); |
| 1781 | vcpu->arch.update_pte.pfn = bad_pfn; |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 1782 | } |
Avi Kivity | da4a00f | 2007-01-05 16:36:44 -0800 | [diff] [blame] | 1783 | } |
| 1784 | |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1785 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
| 1786 | { |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1787 | gpa_t gpa; |
| 1788 | int r; |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1789 | |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1790 | gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva); |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1791 | |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1792 | spin_lock(&vcpu->kvm->mmu_lock); |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1793 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1794 | spin_unlock(&vcpu->kvm->mmu_lock); |
Marcelo Tosatti | 10589a4 | 2007-12-20 19:18:22 -0500 | [diff] [blame] | 1795 | return r; |
Avi Kivity | a436036 | 2007-01-05 16:36:45 -0800 | [diff] [blame] | 1796 | } |
| 1797 | |
Avi Kivity | 22d95b1 | 2007-09-14 20:26:06 +0300 | [diff] [blame] | 1798 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
Avi Kivity | ebeace8 | 2007-01-05 16:36:47 -0800 | [diff] [blame] | 1799 | { |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1800 | while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES) { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1801 | struct kvm_mmu_page *sp; |
Avi Kivity | ebeace8 | 2007-01-05 16:36:47 -0800 | [diff] [blame] | 1802 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1803 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1804 | struct kvm_mmu_page, link); |
| 1805 | kvm_mmu_zap_page(vcpu->kvm, sp); |
Avi Kivity | 4cee576 | 2007-11-18 16:37:07 +0200 | [diff] [blame] | 1806 | ++vcpu->kvm->stat.mmu_recycled; |
Avi Kivity | ebeace8 | 2007-01-05 16:36:47 -0800 | [diff] [blame] | 1807 | } |
| 1808 | } |
Avi Kivity | ebeace8 | 2007-01-05 16:36:47 -0800 | [diff] [blame] | 1809 | |
Avi Kivity | 3067714 | 2007-10-28 18:48:59 +0200 | [diff] [blame] | 1810 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code) |
| 1811 | { |
| 1812 | int r; |
| 1813 | enum emulation_result er; |
| 1814 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1815 | r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code); |
Avi Kivity | 3067714 | 2007-10-28 18:48:59 +0200 | [diff] [blame] | 1816 | if (r < 0) |
| 1817 | goto out; |
| 1818 | |
| 1819 | if (!r) { |
| 1820 | r = 1; |
| 1821 | goto out; |
| 1822 | } |
| 1823 | |
Avi Kivity | b733bfb | 2007-10-28 18:52:05 +0200 | [diff] [blame] | 1824 | r = mmu_topup_memory_caches(vcpu); |
| 1825 | if (r) |
| 1826 | goto out; |
| 1827 | |
Avi Kivity | 3067714 | 2007-10-28 18:48:59 +0200 | [diff] [blame] | 1828 | er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0); |
Avi Kivity | 3067714 | 2007-10-28 18:48:59 +0200 | [diff] [blame] | 1829 | |
| 1830 | switch (er) { |
| 1831 | case EMULATE_DONE: |
| 1832 | return 1; |
| 1833 | case EMULATE_DO_MMIO: |
| 1834 | ++vcpu->stat.mmio_exits; |
| 1835 | return 0; |
| 1836 | case EMULATE_FAIL: |
| 1837 | kvm_report_emulation_failure(vcpu, "pagetable"); |
| 1838 | return 1; |
| 1839 | default: |
| 1840 | BUG(); |
| 1841 | } |
| 1842 | out: |
Avi Kivity | 3067714 | 2007-10-28 18:48:59 +0200 | [diff] [blame] | 1843 | return r; |
| 1844 | } |
| 1845 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); |
| 1846 | |
Joerg Roedel | 1855267 | 2008-02-07 13:47:41 +0100 | [diff] [blame] | 1847 | void kvm_enable_tdp(void) |
| 1848 | { |
| 1849 | tdp_enabled = true; |
| 1850 | } |
| 1851 | EXPORT_SYMBOL_GPL(kvm_enable_tdp); |
| 1852 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1853 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
| 1854 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1855 | struct kvm_mmu_page *sp; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1856 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1857 | while (!list_empty(&vcpu->kvm->arch.active_mmu_pages)) { |
| 1858 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.next, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1859 | struct kvm_mmu_page, link); |
| 1860 | kvm_mmu_zap_page(vcpu->kvm, sp); |
Avi Kivity | f51234c | 2007-01-05 16:36:52 -0800 | [diff] [blame] | 1861 | } |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1862 | free_page((unsigned long)vcpu->arch.mmu.pae_root); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1863 | } |
| 1864 | |
| 1865 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) |
| 1866 | { |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1867 | struct page *page; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1868 | int i; |
| 1869 | |
| 1870 | ASSERT(vcpu); |
| 1871 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1872 | if (vcpu->kvm->arch.n_requested_mmu_pages) |
| 1873 | vcpu->kvm->arch.n_free_mmu_pages = |
| 1874 | vcpu->kvm->arch.n_requested_mmu_pages; |
Izik Eidus | 82ce2c9 | 2007-10-02 18:52:55 +0200 | [diff] [blame] | 1875 | else |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1876 | vcpu->kvm->arch.n_free_mmu_pages = |
| 1877 | vcpu->kvm->arch.n_alloc_mmu_pages; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1878 | /* |
| 1879 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. |
| 1880 | * Therefore we need to allocate shadow page tables in the first |
| 1881 | * 4GB of memory, which happens to fit the DMA32 zone. |
| 1882 | */ |
| 1883 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); |
| 1884 | if (!page) |
| 1885 | goto error_1; |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1886 | vcpu->arch.mmu.pae_root = page_address(page); |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1887 | for (i = 0; i < 4; ++i) |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1888 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
Avi Kivity | 17ac10a | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 1889 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1890 | return 0; |
| 1891 | |
| 1892 | error_1: |
| 1893 | free_mmu_pages(vcpu); |
| 1894 | return -ENOMEM; |
| 1895 | } |
| 1896 | |
Ingo Molnar | 8018c27 | 2006-12-29 16:50:01 -0800 | [diff] [blame] | 1897 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1898 | { |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1899 | ASSERT(vcpu); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1900 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1901 | |
Ingo Molnar | 8018c27 | 2006-12-29 16:50:01 -0800 | [diff] [blame] | 1902 | return alloc_mmu_pages(vcpu); |
| 1903 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1904 | |
Ingo Molnar | 8018c27 | 2006-12-29 16:50:01 -0800 | [diff] [blame] | 1905 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
| 1906 | { |
| 1907 | ASSERT(vcpu); |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 1908 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
Avi Kivity | 2c26495 | 2006-12-22 01:05:28 -0800 | [diff] [blame] | 1909 | |
Ingo Molnar | 8018c27 | 2006-12-29 16:50:01 -0800 | [diff] [blame] | 1910 | return init_kvm_mmu(vcpu); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1911 | } |
| 1912 | |
| 1913 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) |
| 1914 | { |
| 1915 | ASSERT(vcpu); |
| 1916 | |
| 1917 | destroy_kvm_mmu(vcpu); |
| 1918 | free_mmu_pages(vcpu); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 1919 | mmu_free_memory_caches(vcpu); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1920 | } |
| 1921 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 1922 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1923 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1924 | struct kvm_mmu_page *sp; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1925 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1926 | list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) { |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1927 | int i; |
| 1928 | u64 *pt; |
| 1929 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1930 | if (!test_bit(slot, &sp->slot_bitmap)) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1931 | continue; |
| 1932 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1933 | pt = sp->spt; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1934 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
| 1935 | /* avoid RMW */ |
Izik Eidus | 9647c14 | 2007-10-16 14:43:46 +0200 | [diff] [blame] | 1936 | if (pt[i] & PT_WRITABLE_MASK) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1937 | pt[i] &= ~PT_WRITABLE_MASK; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1938 | } |
| 1939 | } |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 1940 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 1941 | void kvm_mmu_zap_all(struct kvm *kvm) |
Dor Laor | e0fa826 | 2007-03-30 13:06:33 +0300 | [diff] [blame] | 1942 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1943 | struct kvm_mmu_page *sp, *node; |
Dor Laor | e0fa826 | 2007-03-30 13:06:33 +0300 | [diff] [blame] | 1944 | |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1945 | spin_lock(&kvm->mmu_lock); |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 1946 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 1947 | kvm_mmu_zap_page(kvm, sp); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 1948 | spin_unlock(&kvm->mmu_lock); |
Dor Laor | e0fa826 | 2007-03-30 13:06:33 +0300 | [diff] [blame] | 1949 | |
Avi Kivity | 90cb052 | 2007-07-17 13:04:56 +0300 | [diff] [blame] | 1950 | kvm_flush_remote_tlbs(kvm); |
Dor Laor | e0fa826 | 2007-03-30 13:06:33 +0300 | [diff] [blame] | 1951 | } |
| 1952 | |
Izik Eidus | 3ee16c8 | 2008-03-30 15:17:21 +0300 | [diff] [blame] | 1953 | void kvm_mmu_remove_one_alloc_mmu_page(struct kvm *kvm) |
| 1954 | { |
| 1955 | struct kvm_mmu_page *page; |
| 1956 | |
| 1957 | page = container_of(kvm->arch.active_mmu_pages.prev, |
| 1958 | struct kvm_mmu_page, link); |
| 1959 | kvm_mmu_zap_page(kvm, page); |
| 1960 | } |
| 1961 | |
| 1962 | static int mmu_shrink(int nr_to_scan, gfp_t gfp_mask) |
| 1963 | { |
| 1964 | struct kvm *kvm; |
| 1965 | struct kvm *kvm_freed = NULL; |
| 1966 | int cache_count = 0; |
| 1967 | |
| 1968 | spin_lock(&kvm_lock); |
| 1969 | |
| 1970 | list_for_each_entry(kvm, &vm_list, vm_list) { |
| 1971 | int npages; |
| 1972 | |
| 1973 | spin_lock(&kvm->mmu_lock); |
| 1974 | npages = kvm->arch.n_alloc_mmu_pages - |
| 1975 | kvm->arch.n_free_mmu_pages; |
| 1976 | cache_count += npages; |
| 1977 | if (!kvm_freed && nr_to_scan > 0 && npages > 0) { |
| 1978 | kvm_mmu_remove_one_alloc_mmu_page(kvm); |
| 1979 | cache_count--; |
| 1980 | kvm_freed = kvm; |
| 1981 | } |
| 1982 | nr_to_scan--; |
| 1983 | |
| 1984 | spin_unlock(&kvm->mmu_lock); |
| 1985 | } |
| 1986 | if (kvm_freed) |
| 1987 | list_move_tail(&kvm_freed->vm_list, &vm_list); |
| 1988 | |
| 1989 | spin_unlock(&kvm_lock); |
| 1990 | |
| 1991 | return cache_count; |
| 1992 | } |
| 1993 | |
| 1994 | static struct shrinker mmu_shrinker = { |
| 1995 | .shrink = mmu_shrink, |
| 1996 | .seeks = DEFAULT_SEEKS * 10, |
| 1997 | }; |
| 1998 | |
Ingo Molnar | 2ddfd20 | 2008-05-22 10:37:48 +0200 | [diff] [blame^] | 1999 | static void mmu_destroy_caches(void) |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2000 | { |
| 2001 | if (pte_chain_cache) |
| 2002 | kmem_cache_destroy(pte_chain_cache); |
| 2003 | if (rmap_desc_cache) |
| 2004 | kmem_cache_destroy(rmap_desc_cache); |
Avi Kivity | d3d25b0 | 2007-05-30 12:34:53 +0300 | [diff] [blame] | 2005 | if (mmu_page_header_cache) |
| 2006 | kmem_cache_destroy(mmu_page_header_cache); |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2007 | } |
| 2008 | |
Izik Eidus | 3ee16c8 | 2008-03-30 15:17:21 +0300 | [diff] [blame] | 2009 | void kvm_mmu_module_exit(void) |
| 2010 | { |
| 2011 | mmu_destroy_caches(); |
| 2012 | unregister_shrinker(&mmu_shrinker); |
| 2013 | } |
| 2014 | |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2015 | int kvm_mmu_module_init(void) |
| 2016 | { |
| 2017 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", |
| 2018 | sizeof(struct kvm_pte_chain), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2019 | 0, 0, NULL); |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2020 | if (!pte_chain_cache) |
| 2021 | goto nomem; |
| 2022 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", |
| 2023 | sizeof(struct kvm_rmap_desc), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2024 | 0, 0, NULL); |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2025 | if (!rmap_desc_cache) |
| 2026 | goto nomem; |
| 2027 | |
Avi Kivity | d3d25b0 | 2007-05-30 12:34:53 +0300 | [diff] [blame] | 2028 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
| 2029 | sizeof(struct kvm_mmu_page), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2030 | 0, 0, NULL); |
Avi Kivity | d3d25b0 | 2007-05-30 12:34:53 +0300 | [diff] [blame] | 2031 | if (!mmu_page_header_cache) |
| 2032 | goto nomem; |
| 2033 | |
Izik Eidus | 3ee16c8 | 2008-03-30 15:17:21 +0300 | [diff] [blame] | 2034 | register_shrinker(&mmu_shrinker); |
| 2035 | |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2036 | return 0; |
| 2037 | |
| 2038 | nomem: |
Izik Eidus | 3ee16c8 | 2008-03-30 15:17:21 +0300 | [diff] [blame] | 2039 | mmu_destroy_caches(); |
Avi Kivity | b5a33a7 | 2007-04-15 16:31:09 +0300 | [diff] [blame] | 2040 | return -ENOMEM; |
| 2041 | } |
| 2042 | |
Zhang Xiantao | 3ad82a7 | 2007-11-20 13:11:38 +0800 | [diff] [blame] | 2043 | /* |
| 2044 | * Caculate mmu pages needed for kvm. |
| 2045 | */ |
| 2046 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) |
| 2047 | { |
| 2048 | int i; |
| 2049 | unsigned int nr_mmu_pages; |
| 2050 | unsigned int nr_pages = 0; |
| 2051 | |
| 2052 | for (i = 0; i < kvm->nmemslots; i++) |
| 2053 | nr_pages += kvm->memslots[i].npages; |
| 2054 | |
| 2055 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; |
| 2056 | nr_mmu_pages = max(nr_mmu_pages, |
| 2057 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); |
| 2058 | |
| 2059 | return nr_mmu_pages; |
| 2060 | } |
| 2061 | |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 2062 | static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer, |
| 2063 | unsigned len) |
| 2064 | { |
| 2065 | if (len > buffer->len) |
| 2066 | return NULL; |
| 2067 | return buffer->ptr; |
| 2068 | } |
| 2069 | |
| 2070 | static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer, |
| 2071 | unsigned len) |
| 2072 | { |
| 2073 | void *ret; |
| 2074 | |
| 2075 | ret = pv_mmu_peek_buffer(buffer, len); |
| 2076 | if (!ret) |
| 2077 | return ret; |
| 2078 | buffer->ptr += len; |
| 2079 | buffer->len -= len; |
| 2080 | buffer->processed += len; |
| 2081 | return ret; |
| 2082 | } |
| 2083 | |
| 2084 | static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu, |
| 2085 | gpa_t addr, gpa_t value) |
| 2086 | { |
| 2087 | int bytes = 8; |
| 2088 | int r; |
| 2089 | |
| 2090 | if (!is_long_mode(vcpu) && !is_pae(vcpu)) |
| 2091 | bytes = 4; |
| 2092 | |
| 2093 | r = mmu_topup_memory_caches(vcpu); |
| 2094 | if (r) |
| 2095 | return r; |
| 2096 | |
Marcelo Tosatti | 3200f40 | 2008-03-29 20:17:59 -0300 | [diff] [blame] | 2097 | if (!emulator_write_phys(vcpu, addr, &value, bytes)) |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 2098 | return -EFAULT; |
| 2099 | |
| 2100 | return 1; |
| 2101 | } |
| 2102 | |
| 2103 | static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
| 2104 | { |
| 2105 | kvm_x86_ops->tlb_flush(vcpu); |
| 2106 | return 1; |
| 2107 | } |
| 2108 | |
| 2109 | static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr) |
| 2110 | { |
| 2111 | spin_lock(&vcpu->kvm->mmu_lock); |
| 2112 | mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT); |
| 2113 | spin_unlock(&vcpu->kvm->mmu_lock); |
| 2114 | return 1; |
| 2115 | } |
| 2116 | |
| 2117 | static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu, |
| 2118 | struct kvm_pv_mmu_op_buffer *buffer) |
| 2119 | { |
| 2120 | struct kvm_mmu_op_header *header; |
| 2121 | |
| 2122 | header = pv_mmu_peek_buffer(buffer, sizeof *header); |
| 2123 | if (!header) |
| 2124 | return 0; |
| 2125 | switch (header->op) { |
| 2126 | case KVM_MMU_OP_WRITE_PTE: { |
| 2127 | struct kvm_mmu_op_write_pte *wpte; |
| 2128 | |
| 2129 | wpte = pv_mmu_read_buffer(buffer, sizeof *wpte); |
| 2130 | if (!wpte) |
| 2131 | return 0; |
| 2132 | return kvm_pv_mmu_write(vcpu, wpte->pte_phys, |
| 2133 | wpte->pte_val); |
| 2134 | } |
| 2135 | case KVM_MMU_OP_FLUSH_TLB: { |
| 2136 | struct kvm_mmu_op_flush_tlb *ftlb; |
| 2137 | |
| 2138 | ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb); |
| 2139 | if (!ftlb) |
| 2140 | return 0; |
| 2141 | return kvm_pv_mmu_flush_tlb(vcpu); |
| 2142 | } |
| 2143 | case KVM_MMU_OP_RELEASE_PT: { |
| 2144 | struct kvm_mmu_op_release_pt *rpt; |
| 2145 | |
| 2146 | rpt = pv_mmu_read_buffer(buffer, sizeof *rpt); |
| 2147 | if (!rpt) |
| 2148 | return 0; |
| 2149 | return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys); |
| 2150 | } |
| 2151 | default: return 0; |
| 2152 | } |
| 2153 | } |
| 2154 | |
| 2155 | int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes, |
| 2156 | gpa_t addr, unsigned long *ret) |
| 2157 | { |
| 2158 | int r; |
| 2159 | struct kvm_pv_mmu_op_buffer buffer; |
| 2160 | |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 2161 | buffer.ptr = buffer.buf; |
| 2162 | buffer.len = min_t(unsigned long, bytes, sizeof buffer.buf); |
| 2163 | buffer.processed = 0; |
| 2164 | |
| 2165 | r = kvm_read_guest(vcpu->kvm, addr, buffer.buf, buffer.len); |
| 2166 | if (r) |
| 2167 | goto out; |
| 2168 | |
| 2169 | while (buffer.len) { |
| 2170 | r = kvm_pv_mmu_op_one(vcpu, &buffer); |
| 2171 | if (r < 0) |
| 2172 | goto out; |
| 2173 | if (r == 0) |
| 2174 | break; |
| 2175 | } |
| 2176 | |
| 2177 | r = 1; |
| 2178 | out: |
| 2179 | *ret = buffer.processed; |
Marcelo Tosatti | 2f333bc | 2008-02-22 12:21:37 -0500 | [diff] [blame] | 2180 | return r; |
| 2181 | } |
| 2182 | |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2183 | #ifdef AUDIT |
| 2184 | |
| 2185 | static const char *audit_msg; |
| 2186 | |
| 2187 | static gva_t canonicalize(gva_t gva) |
| 2188 | { |
| 2189 | #ifdef CONFIG_X86_64 |
| 2190 | gva = (long long)(gva << 16) >> 16; |
| 2191 | #endif |
| 2192 | return gva; |
| 2193 | } |
| 2194 | |
| 2195 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, |
| 2196 | gva_t va, int level) |
| 2197 | { |
| 2198 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); |
| 2199 | int i; |
| 2200 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); |
| 2201 | |
| 2202 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { |
| 2203 | u64 ent = pt[i]; |
| 2204 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2205 | if (ent == shadow_trap_nonpresent_pte) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2206 | continue; |
| 2207 | |
| 2208 | va = canonicalize(va); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2209 | if (level > 1) { |
| 2210 | if (ent == shadow_notrap_nonpresent_pte) |
| 2211 | printk(KERN_ERR "audit: (%s) nontrapping pte" |
| 2212 | " in nonleaf level: levels %d gva %lx" |
| 2213 | " level %d pte %llx\n", audit_msg, |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2214 | vcpu->arch.mmu.root_level, va, level, ent); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2215 | |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2216 | audit_mappings_page(vcpu, ent, va, level - 1); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2217 | } else { |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2218 | gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, va); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 2219 | hpa_t hpa = (hpa_t)gpa_to_pfn(vcpu, gpa) << PAGE_SHIFT; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2220 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2221 | if (is_shadow_present_pte(ent) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2222 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2223 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
| 2224 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2225 | audit_msg, vcpu->arch.mmu.root_level, |
Mike Day | d77c26f | 2007-10-08 09:02:08 -0400 | [diff] [blame] | 2226 | va, gpa, hpa, ent, |
| 2227 | is_shadow_present_pte(ent)); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2228 | else if (ent == shadow_notrap_nonpresent_pte |
| 2229 | && !is_error_hpa(hpa)) |
| 2230 | printk(KERN_ERR "audit: (%s) notrap shadow," |
| 2231 | " valid guest gva %lx\n", audit_msg, va); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 2232 | kvm_release_pfn_clean(pfn); |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 2233 | |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2234 | } |
| 2235 | } |
| 2236 | } |
| 2237 | |
| 2238 | static void audit_mappings(struct kvm_vcpu *vcpu) |
| 2239 | { |
Avi Kivity | 1ea252a | 2007-03-08 11:48:09 +0200 | [diff] [blame] | 2240 | unsigned i; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2241 | |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2242 | if (vcpu->arch.mmu.root_level == 4) |
| 2243 | audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4); |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2244 | else |
| 2245 | for (i = 0; i < 4; ++i) |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2246 | if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2247 | audit_mappings_page(vcpu, |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2248 | vcpu->arch.mmu.pae_root[i], |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2249 | i << 30, |
| 2250 | 2); |
| 2251 | } |
| 2252 | |
| 2253 | static int count_rmaps(struct kvm_vcpu *vcpu) |
| 2254 | { |
| 2255 | int nmaps = 0; |
| 2256 | int i, j, k; |
| 2257 | |
| 2258 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { |
| 2259 | struct kvm_memory_slot *m = &vcpu->kvm->memslots[i]; |
| 2260 | struct kvm_rmap_desc *d; |
| 2261 | |
| 2262 | for (j = 0; j < m->npages; ++j) { |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2263 | unsigned long *rmapp = &m->rmap[j]; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2264 | |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2265 | if (!*rmapp) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2266 | continue; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2267 | if (!(*rmapp & 1)) { |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2268 | ++nmaps; |
| 2269 | continue; |
| 2270 | } |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2271 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2272 | while (d) { |
| 2273 | for (k = 0; k < RMAP_EXT; ++k) |
| 2274 | if (d->shadow_ptes[k]) |
| 2275 | ++nmaps; |
| 2276 | else |
| 2277 | break; |
| 2278 | d = d->more; |
| 2279 | } |
| 2280 | } |
| 2281 | } |
| 2282 | return nmaps; |
| 2283 | } |
| 2284 | |
| 2285 | static int count_writable_mappings(struct kvm_vcpu *vcpu) |
| 2286 | { |
| 2287 | int nmaps = 0; |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2288 | struct kvm_mmu_page *sp; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2289 | int i; |
| 2290 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 2291 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2292 | u64 *pt = sp->spt; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2293 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2294 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2295 | continue; |
| 2296 | |
| 2297 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
| 2298 | u64 ent = pt[i]; |
| 2299 | |
| 2300 | if (!(ent & PT_PRESENT_MASK)) |
| 2301 | continue; |
| 2302 | if (!(ent & PT_WRITABLE_MASK)) |
| 2303 | continue; |
| 2304 | ++nmaps; |
| 2305 | } |
| 2306 | } |
| 2307 | return nmaps; |
| 2308 | } |
| 2309 | |
| 2310 | static void audit_rmap(struct kvm_vcpu *vcpu) |
| 2311 | { |
| 2312 | int n_rmap = count_rmaps(vcpu); |
| 2313 | int n_actual = count_writable_mappings(vcpu); |
| 2314 | |
| 2315 | if (n_rmap != n_actual) |
| 2316 | printk(KERN_ERR "%s: (%s) rmap %d actual %d\n", |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 2317 | __func__, audit_msg, n_rmap, n_actual); |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2318 | } |
| 2319 | |
| 2320 | static void audit_write_protection(struct kvm_vcpu *vcpu) |
| 2321 | { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2322 | struct kvm_mmu_page *sp; |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2323 | struct kvm_memory_slot *slot; |
| 2324 | unsigned long *rmapp; |
| 2325 | gfn_t gfn; |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2326 | |
Zhang Xiantao | f05e70a | 2007-12-14 10:01:48 +0800 | [diff] [blame] | 2327 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2328 | if (sp->role.metaphysical) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2329 | continue; |
| 2330 | |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2331 | slot = gfn_to_memslot(vcpu->kvm, sp->gfn); |
| 2332 | gfn = unalias_gfn(vcpu->kvm, sp->gfn); |
Izik Eidus | 290fc38 | 2007-09-27 14:11:22 +0200 | [diff] [blame] | 2333 | rmapp = &slot->rmap[gfn - slot->base_gfn]; |
| 2334 | if (*rmapp) |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2335 | printk(KERN_ERR "%s: (%s) shadow page has writable" |
| 2336 | " mappings: gfn %lx role %x\n", |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 2337 | __func__, audit_msg, sp->gfn, |
Avi Kivity | 4db3531 | 2007-11-21 15:28:32 +0200 | [diff] [blame] | 2338 | sp->role.word); |
Avi Kivity | 37a7d8b | 2007-01-05 16:36:56 -0800 | [diff] [blame] | 2339 | } |
| 2340 | } |
| 2341 | |
| 2342 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) |
| 2343 | { |
| 2344 | int olddbg = dbg; |
| 2345 | |
| 2346 | dbg = 0; |
| 2347 | audit_msg = msg; |
| 2348 | audit_rmap(vcpu); |
| 2349 | audit_write_protection(vcpu); |
| 2350 | audit_mappings(vcpu); |
| 2351 | dbg = olddbg; |
| 2352 | } |
| 2353 | |
| 2354 | #endif |