blob: 323c3f5f5c84968eb35561978392286405323afe [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
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.
Nicolas Kaiser9611c182010-10-06 14:23:22 +020010 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -080011 *
12 * Authors:
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Avi Kivity <avi@qumranet.com>
15 *
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
18 *
19 */
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Gleb Natapovaf585b92010-10-14 11:22:46 +020021#include "irq.h"
Zhang Xiantao1d737c82007-12-14 09:35:10 +080022#include "mmu.h"
Avi Kivity836a1b32010-01-21 15:31:49 +020023#include "x86.h"
Avi Kivity6de4f3a2009-05-31 22:58:47 +030024#include "kvm_cache_regs.h"
Nadav Amit5f7dde72014-05-07 15:32:50 +030025#include "cpuid.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080026
Avi Kivityedf88412007-12-16 11:02:48 +020027#include <linux/kvm_host.h>
Avi Kivitye4956062007-06-28 14:15:57 -040028#include <linux/types.h>
29#include <linux/string.h>
30#include <linux/mm.h>
31#include <linux/highmem.h>
32#include <linux/module.h>
Izik Eidus448353c2007-11-26 14:08:14 +020033#include <linux/swap.h>
Marcelo Tosatti05da4552008-02-23 11:44:30 -030034#include <linux/hugetlb.h>
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -050035#include <linux/compiler.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020036#include <linux/srcu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Huang Yingbf998152010-05-31 14:28:19 +080038#include <linux/uaccess.h>
Avi Kivitye4956062007-06-28 14:15:57 -040039
40#include <asm/page.h>
41#include <asm/cmpxchg.h>
Avi Kivity4e542372007-11-21 14:08:40 +020042#include <asm/io.h>
Eduardo Habkost13673a92008-11-17 19:03:13 -020043#include <asm/vmx.h>
Avi Kivitye4956062007-06-28 14:15:57 -040044
Joerg Roedel18552672008-02-07 13:47:41 +010045/*
46 * When setting this variable to true it enables Two-Dimensional-Paging
47 * where the hardware walks 2 page tables:
48 * 1. the guest-virtual to guest-physical
49 * 2. while doing 1. it walks guest-physical to host-physical
50 * If the hardware supports that we don't need to do shadow paging.
51 */
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -050052bool tdp_enabled = false;
Joerg Roedel18552672008-02-07 13:47:41 +010053
Xiao Guangrong8b1fe172010-08-30 18:22:53 +080054enum {
55 AUDIT_PRE_PAGE_FAULT,
56 AUDIT_POST_PAGE_FAULT,
57 AUDIT_PRE_PTE_WRITE,
Xiao Guangrong69030742010-09-27 18:09:29 +080058 AUDIT_POST_PTE_WRITE,
59 AUDIT_PRE_SYNC,
60 AUDIT_POST_SYNC
Xiao Guangrong8b1fe172010-08-30 18:22:53 +080061};
62
Avi Kivity37a7d8b2007-01-05 16:36:56 -080063#undef MMU_DEBUG
64
Avi Kivity37a7d8b2007-01-05 16:36:56 -080065#ifdef MMU_DEBUG
66
67#define pgprintk(x...) do { if (dbg) printk(x); } while (0)
68#define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
69
70#else
71
72#define pgprintk(x...) do { } while (0)
73#define rmap_printk(x...) do { } while (0)
74
75#endif
76
Xiao Guangrong8b1fe172010-08-30 18:22:53 +080077#ifdef MMU_DEBUG
Rusty Russell476bc002012-01-13 09:32:18 +103078static bool dbg = 0;
Avi Kivity6ada8cc2008-06-22 16:45:24 +030079module_param(dbg, bool, 0644);
Avi Kivity37a7d8b2007-01-05 16:36:56 -080080#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080081
Yaozu Dongd6c69ee2007-04-25 14:17:25 +080082#ifndef MMU_DEBUG
83#define ASSERT(x) do { } while (0)
84#else
Avi Kivity6aa8b732006-12-10 02:21:36 -080085#define ASSERT(x) \
86 if (!(x)) { \
87 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
88 __FILE__, __LINE__, #x); \
89 }
Yaozu Dongd6c69ee2007-04-25 14:17:25 +080090#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080091
Xiao Guangrong957ed9e2010-08-22 19:12:48 +080092#define PTE_PREFETCH_NUM 8
93
Xudong Hao00763e42012-06-07 18:26:07 +080094#define PT_FIRST_AVAIL_BITS_SHIFT 10
Avi Kivity6aa8b732006-12-10 02:21:36 -080095#define PT64_SECOND_AVAIL_BITS_SHIFT 52
96
Avi Kivity6aa8b732006-12-10 02:21:36 -080097#define PT64_LEVEL_BITS 9
98
99#define PT64_LEVEL_SHIFT(level) \
Mike Dayd77c26f2007-10-08 09:02:08 -0400100 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800101
Avi Kivity6aa8b732006-12-10 02:21:36 -0800102#define PT64_INDEX(address, level)\
103 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
104
105
106#define PT32_LEVEL_BITS 10
107
108#define PT32_LEVEL_SHIFT(level) \
Mike Dayd77c26f2007-10-08 09:02:08 -0400109 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800110
Joerg Roedele04da982009-07-27 16:30:45 +0200111#define PT32_LVL_OFFSET_MASK(level) \
112 (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
113 * PT32_LEVEL_BITS))) - 1))
Avi Kivity6aa8b732006-12-10 02:21:36 -0800114
115#define PT32_INDEX(address, level)\
116 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
117
118
Avi Kivity27aba762007-03-09 13:04:31 +0200119#define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
Avi Kivity6aa8b732006-12-10 02:21:36 -0800120#define PT64_DIR_BASE_ADDR_MASK \
121 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
Joerg Roedele04da982009-07-27 16:30:45 +0200122#define PT64_LVL_ADDR_MASK(level) \
123 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
124 * PT64_LEVEL_BITS))) - 1))
125#define PT64_LVL_OFFSET_MASK(level) \
126 (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
127 * PT64_LEVEL_BITS))) - 1))
Avi Kivity6aa8b732006-12-10 02:21:36 -0800128
129#define PT32_BASE_ADDR_MASK PAGE_MASK
130#define PT32_DIR_BASE_ADDR_MASK \
131 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
Joerg Roedele04da982009-07-27 16:30:45 +0200132#define PT32_LVL_ADDR_MASK(level) \
133 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
134 * PT32_LEVEL_BITS))) - 1))
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135
Gleb Natapov53166222013-08-05 11:07:14 +0300136#define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | shadow_user_mask \
137 | shadow_x_mask | shadow_nx_mask)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800138
Avi Kivityfe135d22007-12-09 16:15:46 +0200139#define ACC_EXEC_MASK 1
140#define ACC_WRITE_MASK PT_WRITABLE_MASK
141#define ACC_USER_MASK PT_USER_MASK
142#define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
143
Avi Kivity90bb6fc2009-12-31 12:10:16 +0200144#include <trace/events/kvm.h>
145
Avi Kivity07420172009-07-06 12:21:32 +0300146#define CREATE_TRACE_POINTS
147#include "mmutrace.h"
148
Xiao Guangrong49fde342012-06-20 15:58:58 +0800149#define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
150#define SPTE_MMU_WRITEABLE (1ULL << (PT_FIRST_AVAIL_BITS_SHIFT + 1))
Izik Eidus14032832009-09-23 21:47:17 +0300151
Avi Kivity135f8c22008-08-21 17:49:56 +0300152#define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
153
Takuya Yoshikawa220f7732012-03-21 23:49:39 +0900154/* make pte_list_desc fit well in cache line */
155#define PTE_LIST_EXT 3
156
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800157struct pte_list_desc {
158 u64 *sptes[PTE_LIST_EXT];
159 struct pte_list_desc *more;
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800160};
161
Avi Kivity2d111232008-12-25 14:39:47 +0200162struct kvm_shadow_walk_iterator {
163 u64 addr;
164 hpa_t shadow_addr;
Avi Kivity2d111232008-12-25 14:39:47 +0200165 u64 *sptep;
Xiao Guangrongdd3bfd52011-07-12 03:32:54 +0800166 int level;
Avi Kivity2d111232008-12-25 14:39:47 +0200167 unsigned index;
168};
169
170#define for_each_shadow_entry(_vcpu, _addr, _walker) \
171 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
172 shadow_walk_okay(&(_walker)); \
173 shadow_walk_next(&(_walker)))
174
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800175#define for_each_shadow_entry_lockless(_vcpu, _addr, _walker, spte) \
176 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
177 shadow_walk_okay(&(_walker)) && \
178 ({ spte = mmu_spte_get_lockless(_walker.sptep); 1; }); \
179 __shadow_walk_next(&(_walker), spte))
180
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800181static struct kmem_cache *pte_list_desc_cache;
Avi Kivityd3d25b02007-05-30 12:34:53 +0300182static struct kmem_cache *mmu_page_header_cache;
Dave Hansen45221ab2010-08-19 18:11:37 -0700183static struct percpu_counter kvm_total_used_mmu_pages;
Avi Kivityb5a33a72007-04-15 16:31:09 +0300184
Sheng Yang7b523452008-04-25 21:13:50 +0800185static u64 __read_mostly shadow_nx_mask;
186static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
187static u64 __read_mostly shadow_user_mask;
188static u64 __read_mostly shadow_accessed_mask;
189static u64 __read_mostly shadow_dirty_mask;
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800190static u64 __read_mostly shadow_mmio_mask;
191
192static void mmu_spte_set(u64 *sptep, u64 spte);
Avi Kivitye6765052012-07-08 17:16:30 +0300193static void mmu_free_roots(struct kvm_vcpu *vcpu);
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800194
195void kvm_mmu_set_mmio_spte_mask(u64 mmio_mask)
196{
197 shadow_mmio_mask = mmio_mask;
198}
199EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_mask);
200
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800201/*
202 * spte bits of bit 3 ~ bit 11 are used as low 9 bits of generation number,
203 * the bits of bits 52 ~ bit 61 are used as high 10 bits of generation
204 * number.
205 */
206#define MMIO_SPTE_GEN_LOW_SHIFT 3
207#define MMIO_SPTE_GEN_HIGH_SHIFT 52
208
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800209#define MMIO_GEN_SHIFT 19
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800210#define MMIO_GEN_LOW_SHIFT 9
211#define MMIO_GEN_LOW_MASK ((1 << MMIO_GEN_LOW_SHIFT) - 1)
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800212#define MMIO_GEN_MASK ((1 << MMIO_GEN_SHIFT) - 1)
213#define MMIO_MAX_GEN ((1 << MMIO_GEN_SHIFT) - 1)
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800214
215static u64 generation_mmio_spte_mask(unsigned int gen)
216{
217 u64 mask;
218
219 WARN_ON(gen > MMIO_MAX_GEN);
220
221 mask = (gen & MMIO_GEN_LOW_MASK) << MMIO_SPTE_GEN_LOW_SHIFT;
222 mask |= ((u64)gen >> MMIO_GEN_LOW_SHIFT) << MMIO_SPTE_GEN_HIGH_SHIFT;
223 return mask;
224}
225
226static unsigned int get_mmio_spte_generation(u64 spte)
227{
228 unsigned int gen;
229
230 spte &= ~shadow_mmio_mask;
231
232 gen = (spte >> MMIO_SPTE_GEN_LOW_SHIFT) & MMIO_GEN_LOW_MASK;
233 gen |= (spte >> MMIO_SPTE_GEN_HIGH_SHIFT) << MMIO_GEN_LOW_SHIFT;
234 return gen;
235}
236
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800237static unsigned int kvm_current_mmio_generation(struct kvm *kvm)
238{
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200239 return kvm_memslots(kvm)->generation & MMIO_GEN_MASK;
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800240}
241
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800242static void mark_mmio_spte(struct kvm *kvm, u64 *sptep, u64 gfn,
243 unsigned access)
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800244{
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800245 unsigned int gen = kvm_current_mmio_generation(kvm);
246 u64 mask = generation_mmio_spte_mask(gen);
Takuya Yoshikawa95b04302013-03-12 17:44:40 +0900247
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800248 access &= ACC_WRITE_MASK | ACC_USER_MASK;
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800249 mask |= shadow_mmio_mask | access | gfn << PAGE_SHIFT;
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800250
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800251 trace_mark_mmio_spte(sptep, gfn, access, gen);
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800252 mmu_spte_set(sptep, mask);
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800253}
254
255static bool is_mmio_spte(u64 spte)
256{
257 return (spte & shadow_mmio_mask) == shadow_mmio_mask;
258}
259
260static gfn_t get_mmio_spte_gfn(u64 spte)
261{
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800262 u64 mask = generation_mmio_spte_mask(MMIO_MAX_GEN) | shadow_mmio_mask;
263 return (spte & ~mask) >> PAGE_SHIFT;
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800264}
265
266static unsigned get_mmio_spte_access(u64 spte)
267{
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800268 u64 mask = generation_mmio_spte_mask(MMIO_MAX_GEN) | shadow_mmio_mask;
269 return (spte & ~mask) & ~PAGE_MASK;
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800270}
271
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800272static bool set_mmio_spte(struct kvm *kvm, u64 *sptep, gfn_t gfn,
273 pfn_t pfn, unsigned access)
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800274{
275 if (unlikely(is_noslot_pfn(pfn))) {
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800276 mark_mmio_spte(kvm, sptep, gfn, access);
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800277 return true;
278 }
279
280 return false;
281}
Avi Kivityc7addb92007-09-16 18:58:32 +0200282
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800283static bool check_mmio_spte(struct kvm *kvm, u64 spte)
284{
Xiao Guangrong089504c2013-06-07 16:51:27 +0800285 unsigned int kvm_gen, spte_gen;
286
287 kvm_gen = kvm_current_mmio_generation(kvm);
288 spte_gen = get_mmio_spte_generation(spte);
289
290 trace_check_mmio_spte(spte, kvm_gen, spte_gen);
291 return likely(kvm_gen == spte_gen);
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800292}
293
Dong, Eddie82725b22009-03-30 16:21:08 +0800294static inline u64 rsvd_bits(int s, int e)
295{
296 return ((1ULL << (e - s + 1)) - 1) << s;
297}
298
Sheng Yang7b523452008-04-25 21:13:50 +0800299void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
Sheng Yang4b12f0d2009-04-27 20:35:42 +0800300 u64 dirty_mask, u64 nx_mask, u64 x_mask)
Sheng Yang7b523452008-04-25 21:13:50 +0800301{
302 shadow_user_mask = user_mask;
303 shadow_accessed_mask = accessed_mask;
304 shadow_dirty_mask = dirty_mask;
305 shadow_nx_mask = nx_mask;
306 shadow_x_mask = x_mask;
307}
308EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes);
309
Avi Kivity6aa8b732006-12-10 02:21:36 -0800310static int is_cpuid_PSE36(void)
311{
312 return 1;
313}
314
Avi Kivity73b10872007-01-26 00:56:41 -0800315static int is_nx(struct kvm_vcpu *vcpu)
316{
Avi Kivityf6801df2010-01-21 15:31:50 +0200317 return vcpu->arch.efer & EFER_NX;
Avi Kivity73b10872007-01-26 00:56:41 -0800318}
319
Avi Kivityc7addb92007-09-16 18:58:32 +0200320static int is_shadow_present_pte(u64 pte)
321{
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800322 return pte & PT_PRESENT_MASK && !is_mmio_spte(pte);
Avi Kivityc7addb92007-09-16 18:58:32 +0200323}
324
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300325static int is_large_pte(u64 pte)
326{
327 return pte & PT_PAGE_SIZE_MASK;
328}
329
Avi Kivity43a37952009-06-10 14:12:05 +0300330static int is_rmap_spte(u64 pte)
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800331{
Avi Kivity4b1a80f2008-03-23 12:18:19 +0200332 return is_shadow_present_pte(pte);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800333}
334
Marcelo Tosatti776e6632009-06-10 12:27:03 -0300335static int is_last_spte(u64 pte, int level)
336{
337 if (level == PT_PAGE_TABLE_LEVEL)
338 return 1;
Joerg Roedel852e3c12009-07-27 16:30:44 +0200339 if (is_large_pte(pte))
Marcelo Tosatti776e6632009-06-10 12:27:03 -0300340 return 1;
341 return 0;
342}
343
Anthony Liguori35149e22008-04-02 14:46:56 -0500344static pfn_t spte_to_pfn(u64 pte)
Avi Kivity0b49ea82008-03-23 15:06:23 +0200345{
Anthony Liguori35149e22008-04-02 14:46:56 -0500346 return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
Avi Kivity0b49ea82008-03-23 15:06:23 +0200347}
348
Avi Kivityda928522007-11-21 13:54:47 +0200349static gfn_t pse36_gfn_delta(u32 gpte)
350{
351 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
352
353 return (gpte & PT32_DIR_PSE36_MASK) << shift;
354}
355
Xiao Guangrong603e0652011-07-12 03:31:28 +0800356#ifdef CONFIG_X86_64
Avi Kivityd555c332009-06-10 14:24:23 +0300357static void __set_spte(u64 *sptep, u64 spte)
Avi Kivitye663ee62007-05-31 15:46:04 +0300358{
Xiao Guangrong603e0652011-07-12 03:31:28 +0800359 *sptep = spte;
Avi Kivitye663ee62007-05-31 15:46:04 +0300360}
361
Xiao Guangrong603e0652011-07-12 03:31:28 +0800362static void __update_clear_spte_fast(u64 *sptep, u64 spte)
Avi Kivitya9221dd2010-06-06 14:48:06 +0300363{
Xiao Guangrong603e0652011-07-12 03:31:28 +0800364 *sptep = spte;
Avi Kivitya9221dd2010-06-06 14:48:06 +0300365}
366
Xiao Guangrong603e0652011-07-12 03:31:28 +0800367static u64 __update_clear_spte_slow(u64 *sptep, u64 spte)
368{
369 return xchg(sptep, spte);
370}
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800371
372static u64 __get_spte_lockless(u64 *sptep)
373{
374 return ACCESS_ONCE(*sptep);
375}
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800376
377static bool __check_direct_spte_mmio_pf(u64 spte)
378{
379 /* It is valid if the spte is zapped. */
380 return spte == 0ull;
381}
Xiao Guangrong603e0652011-07-12 03:31:28 +0800382#else
383union split_spte {
384 struct {
385 u32 spte_low;
386 u32 spte_high;
387 };
388 u64 spte;
389};
390
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800391static void count_spte_clear(u64 *sptep, u64 spte)
392{
393 struct kvm_mmu_page *sp = page_header(__pa(sptep));
394
395 if (is_shadow_present_pte(spte))
396 return;
397
398 /* Ensure the spte is completely set before we increase the count */
399 smp_wmb();
400 sp->clear_spte_count++;
401}
402
Xiao Guangrong603e0652011-07-12 03:31:28 +0800403static void __set_spte(u64 *sptep, u64 spte)
404{
405 union split_spte *ssptep, sspte;
406
407 ssptep = (union split_spte *)sptep;
408 sspte = (union split_spte)spte;
409
410 ssptep->spte_high = sspte.spte_high;
411
412 /*
413 * If we map the spte from nonpresent to present, We should store
414 * the high bits firstly, then set present bit, so cpu can not
415 * fetch this spte while we are setting the spte.
416 */
417 smp_wmb();
418
419 ssptep->spte_low = sspte.spte_low;
420}
421
422static void __update_clear_spte_fast(u64 *sptep, u64 spte)
423{
424 union split_spte *ssptep, sspte;
425
426 ssptep = (union split_spte *)sptep;
427 sspte = (union split_spte)spte;
428
429 ssptep->spte_low = sspte.spte_low;
430
431 /*
432 * If we map the spte from present to nonpresent, we should clear
433 * present bit firstly to avoid vcpu fetch the old high bits.
434 */
435 smp_wmb();
436
437 ssptep->spte_high = sspte.spte_high;
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800438 count_spte_clear(sptep, spte);
Xiao Guangrong603e0652011-07-12 03:31:28 +0800439}
440
441static u64 __update_clear_spte_slow(u64 *sptep, u64 spte)
442{
443 union split_spte *ssptep, sspte, orig;
444
445 ssptep = (union split_spte *)sptep;
446 sspte = (union split_spte)spte;
447
448 /* xchg acts as a barrier before the setting of the high bits */
449 orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low);
Zhao Jin41bc3182011-09-19 12:19:51 +0800450 orig.spte_high = ssptep->spte_high;
451 ssptep->spte_high = sspte.spte_high;
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800452 count_spte_clear(sptep, spte);
Xiao Guangrong603e0652011-07-12 03:31:28 +0800453
454 return orig.spte;
455}
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800456
457/*
458 * The idea using the light way get the spte on x86_32 guest is from
459 * gup_get_pte(arch/x86/mm/gup.c).
Xiao Guangrongaccaefe2013-06-19 17:09:20 +0800460 *
461 * An spte tlb flush may be pending, because kvm_set_pte_rmapp
462 * coalesces them and we are running out of the MMU lock. Therefore
463 * we need to protect against in-progress updates of the spte.
464 *
465 * Reading the spte while an update is in progress may get the old value
466 * for the high part of the spte. The race is fine for a present->non-present
467 * change (because the high part of the spte is ignored for non-present spte),
468 * but for a present->present change we must reread the spte.
469 *
470 * All such changes are done in two steps (present->non-present and
471 * non-present->present), hence it is enough to count the number of
472 * present->non-present updates: if it changed while reading the spte,
473 * we might have hit the race. This is done using clear_spte_count.
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800474 */
475static u64 __get_spte_lockless(u64 *sptep)
476{
477 struct kvm_mmu_page *sp = page_header(__pa(sptep));
478 union split_spte spte, *orig = (union split_spte *)sptep;
479 int count;
480
481retry:
482 count = sp->clear_spte_count;
483 smp_rmb();
484
485 spte.spte_low = orig->spte_low;
486 smp_rmb();
487
488 spte.spte_high = orig->spte_high;
489 smp_rmb();
490
491 if (unlikely(spte.spte_low != orig->spte_low ||
492 count != sp->clear_spte_count))
493 goto retry;
494
495 return spte.spte;
496}
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800497
498static bool __check_direct_spte_mmio_pf(u64 spte)
499{
500 union split_spte sspte = (union split_spte)spte;
501 u32 high_mmio_mask = shadow_mmio_mask >> 32;
502
503 /* It is valid if the spte is zapped. */
504 if (spte == 0ull)
505 return true;
506
507 /* It is valid if the spte is being zapped. */
508 if (sspte.spte_low == 0ull &&
509 (sspte.spte_high & high_mmio_mask) == high_mmio_mask)
510 return true;
511
512 return false;
513}
Xiao Guangrong603e0652011-07-12 03:31:28 +0800514#endif
515
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800516static bool spte_is_locklessly_modifiable(u64 spte)
517{
Gleb Natapovfeb3eb72013-01-30 16:45:00 +0200518 return (spte & (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE)) ==
519 (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE);
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800520}
521
Xiao Guangrong8672b722010-08-02 16:14:04 +0800522static bool spte_has_volatile_bits(u64 spte)
523{
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800524 /*
525 * Always atomicly update spte if it can be updated
526 * out of mmu-lock, it can ensure dirty bit is not lost,
527 * also, it can help us to get a stable is_writable_pte()
528 * to ensure tlb flush is not missed.
529 */
530 if (spte_is_locklessly_modifiable(spte))
531 return true;
532
Xiao Guangrong8672b722010-08-02 16:14:04 +0800533 if (!shadow_accessed_mask)
534 return false;
535
536 if (!is_shadow_present_pte(spte))
537 return false;
538
Xiao Guangrong41327792010-08-02 16:15:08 +0800539 if ((spte & shadow_accessed_mask) &&
540 (!is_writable_pte(spte) || (spte & shadow_dirty_mask)))
Xiao Guangrong8672b722010-08-02 16:14:04 +0800541 return false;
542
543 return true;
544}
545
Xiao Guangrong41327792010-08-02 16:15:08 +0800546static bool spte_is_bit_cleared(u64 old_spte, u64 new_spte, u64 bit_mask)
547{
548 return (old_spte & bit_mask) && !(new_spte & bit_mask);
549}
550
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800551/* Rules for using mmu_spte_set:
552 * Set the sptep from nonpresent to present.
553 * Note: the sptep being assigned *must* be either not present
554 * or in a state where the hardware will not attempt to update
555 * the spte.
556 */
557static void mmu_spte_set(u64 *sptep, u64 new_spte)
558{
559 WARN_ON(is_shadow_present_pte(*sptep));
560 __set_spte(sptep, new_spte);
561}
562
563/* Rules for using mmu_spte_update:
564 * Update the state bits, it means the mapped pfn is not changged.
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800565 *
566 * Whenever we overwrite a writable spte with a read-only one we
567 * should flush remote TLBs. Otherwise rmap_write_protect
568 * will find a read-only spte, even though the writable spte
569 * might be cached on a CPU's TLB, the return value indicates this
570 * case.
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800571 */
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800572static bool mmu_spte_update(u64 *sptep, u64 new_spte)
Avi Kivityb79b93f2010-06-06 15:46:44 +0300573{
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800574 u64 old_spte = *sptep;
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800575 bool ret = false;
Avi Kivityb79b93f2010-06-06 15:46:44 +0300576
Xiao Guangrong41327792010-08-02 16:15:08 +0800577 WARN_ON(!is_rmap_spte(new_spte));
578
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800579 if (!is_shadow_present_pte(old_spte)) {
580 mmu_spte_set(sptep, new_spte);
581 return ret;
582 }
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800583
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800584 if (!spte_has_volatile_bits(old_spte))
Xiao Guangrong603e0652011-07-12 03:31:28 +0800585 __update_clear_spte_fast(sptep, new_spte);
Xiao Guangrong41327792010-08-02 16:15:08 +0800586 else
Xiao Guangrong603e0652011-07-12 03:31:28 +0800587 old_spte = __update_clear_spte_slow(sptep, new_spte);
Xiao Guangrong41327792010-08-02 16:15:08 +0800588
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +0800589 /*
590 * For the spte updated out of mmu-lock is safe, since
591 * we always atomicly update it, see the comments in
592 * spte_has_volatile_bits().
593 */
Xiao Guangrong7f31c952014-04-17 17:06:15 +0800594 if (spte_is_locklessly_modifiable(old_spte) &&
595 !is_writable_pte(new_spte))
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800596 ret = true;
597
Xiao Guangrong41327792010-08-02 16:15:08 +0800598 if (!shadow_accessed_mask)
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800599 return ret;
Xiao Guangrong41327792010-08-02 16:15:08 +0800600
601 if (spte_is_bit_cleared(old_spte, new_spte, shadow_accessed_mask))
602 kvm_set_pfn_accessed(spte_to_pfn(old_spte));
603 if (spte_is_bit_cleared(old_spte, new_spte, shadow_dirty_mask))
604 kvm_set_pfn_dirty(spte_to_pfn(old_spte));
Xiao Guangrong6e7d0352012-06-20 15:58:33 +0800605
606 return ret;
Avi Kivityb79b93f2010-06-06 15:46:44 +0300607}
608
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800609/*
610 * Rules for using mmu_spte_clear_track_bits:
611 * It sets the sptep from present to nonpresent, and track the
612 * state bits, it is used to clear the last level sptep.
613 */
614static int mmu_spte_clear_track_bits(u64 *sptep)
615{
616 pfn_t pfn;
617 u64 old_spte = *sptep;
618
619 if (!spte_has_volatile_bits(old_spte))
Xiao Guangrong603e0652011-07-12 03:31:28 +0800620 __update_clear_spte_fast(sptep, 0ull);
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800621 else
Xiao Guangrong603e0652011-07-12 03:31:28 +0800622 old_spte = __update_clear_spte_slow(sptep, 0ull);
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800623
624 if (!is_rmap_spte(old_spte))
625 return 0;
626
627 pfn = spte_to_pfn(old_spte);
Xiao Guangrong86fde742012-07-17 21:52:52 +0800628
629 /*
630 * KVM does not hold the refcount of the page used by
631 * kvm mmu, before reclaiming the page, we should
632 * unmap it from mmu first.
633 */
634 WARN_ON(!kvm_is_mmio_pfn(pfn) && !page_count(pfn_to_page(pfn)));
635
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800636 if (!shadow_accessed_mask || old_spte & shadow_accessed_mask)
637 kvm_set_pfn_accessed(pfn);
638 if (!shadow_dirty_mask || (old_spte & shadow_dirty_mask))
639 kvm_set_pfn_dirty(pfn);
640 return 1;
641}
642
643/*
644 * Rules for using mmu_spte_clear_no_track:
645 * Directly clear spte without caring the state bits of sptep,
646 * it is used to set the upper level spte.
647 */
648static void mmu_spte_clear_no_track(u64 *sptep)
649{
Xiao Guangrong603e0652011-07-12 03:31:28 +0800650 __update_clear_spte_fast(sptep, 0ull);
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +0800651}
652
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800653static u64 mmu_spte_get_lockless(u64 *sptep)
654{
655 return __get_spte_lockless(sptep);
656}
657
658static void walk_shadow_page_lockless_begin(struct kvm_vcpu *vcpu)
659{
Avi Kivityc1427862012-05-14 15:44:06 +0300660 /*
661 * Prevent page table teardown by making any free-er wait during
662 * kvm_flush_remote_tlbs() IPI to all active vcpus.
663 */
664 local_irq_disable();
665 vcpu->mode = READING_SHADOW_PAGE_TABLES;
666 /*
667 * Make sure a following spte read is not reordered ahead of the write
668 * to vcpu->mode.
669 */
670 smp_mb();
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800671}
672
673static void walk_shadow_page_lockless_end(struct kvm_vcpu *vcpu)
674{
Avi Kivityc1427862012-05-14 15:44:06 +0300675 /*
676 * Make sure the write to vcpu->mode is not reordered in front of
677 * reads to sptes. If it does, kvm_commit_zap_page() can see us
678 * OUTSIDE_GUEST_MODE and proceed to free the shadow page table.
679 */
680 smp_mb();
681 vcpu->mode = OUTSIDE_GUEST_MODE;
682 local_irq_enable();
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +0800683}
684
Avi Kivitye2dec932007-01-05 16:36:54 -0800685static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
Avi Kivity2e3e5882007-09-10 11:28:17 +0300686 struct kmem_cache *base_cache, int min)
Avi Kivity714b93d2007-01-05 16:36:53 -0800687{
688 void *obj;
689
690 if (cache->nobjs >= min)
Avi Kivitye2dec932007-01-05 16:36:54 -0800691 return 0;
Avi Kivity714b93d2007-01-05 16:36:53 -0800692 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
Avi Kivity2e3e5882007-09-10 11:28:17 +0300693 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
Avi Kivity714b93d2007-01-05 16:36:53 -0800694 if (!obj)
Avi Kivitye2dec932007-01-05 16:36:54 -0800695 return -ENOMEM;
Avi Kivity714b93d2007-01-05 16:36:53 -0800696 cache->objects[cache->nobjs++] = obj;
697 }
Avi Kivitye2dec932007-01-05 16:36:54 -0800698 return 0;
Avi Kivity714b93d2007-01-05 16:36:53 -0800699}
700
Xiao Guangrongf759e2b2011-09-22 16:53:17 +0800701static int mmu_memory_cache_free_objects(struct kvm_mmu_memory_cache *cache)
702{
703 return cache->nobjs;
704}
705
Xiao Guangronge8ad9a72010-05-13 10:06:02 +0800706static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc,
707 struct kmem_cache *cache)
Avi Kivity714b93d2007-01-05 16:36:53 -0800708{
709 while (mc->nobjs)
Xiao Guangronge8ad9a72010-05-13 10:06:02 +0800710 kmem_cache_free(cache, mc->objects[--mc->nobjs]);
Avi Kivity714b93d2007-01-05 16:36:53 -0800711}
712
Avi Kivityc1158e62007-07-20 08:18:27 +0300713static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
Avi Kivity2e3e5882007-09-10 11:28:17 +0300714 int min)
Avi Kivityc1158e62007-07-20 08:18:27 +0300715{
Xiao Guangrong842f22e2011-03-04 19:01:10 +0800716 void *page;
Avi Kivityc1158e62007-07-20 08:18:27 +0300717
718 if (cache->nobjs >= min)
719 return 0;
720 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
Xiao Guangrong842f22e2011-03-04 19:01:10 +0800721 page = (void *)__get_free_page(GFP_KERNEL);
Avi Kivityc1158e62007-07-20 08:18:27 +0300722 if (!page)
723 return -ENOMEM;
Xiao Guangrong842f22e2011-03-04 19:01:10 +0800724 cache->objects[cache->nobjs++] = page;
Avi Kivityc1158e62007-07-20 08:18:27 +0300725 }
726 return 0;
727}
728
729static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
730{
731 while (mc->nobjs)
Avi Kivityc4d198d2007-07-21 09:06:46 +0300732 free_page((unsigned long)mc->objects[--mc->nobjs]);
Avi Kivityc1158e62007-07-20 08:18:27 +0300733}
734
Avi Kivity8c438502007-04-16 11:53:17 +0300735static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
736{
737 int r;
738
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800739 r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache,
Xiao Guangrong67052b32011-05-15 23:27:08 +0800740 pte_list_desc_cache, 8 + PTE_PREFETCH_NUM);
Avi Kivity2e3e5882007-09-10 11:28:17 +0300741 if (r)
742 goto out;
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800743 r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8);
Avi Kivity2e3e5882007-09-10 11:28:17 +0300744 if (r)
745 goto out;
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800746 r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache,
Avi Kivity2e3e5882007-09-10 11:28:17 +0300747 mmu_page_header_cache, 4);
748out:
Avi Kivity8c438502007-04-16 11:53:17 +0300749 return r;
750}
751
Avi Kivity714b93d2007-01-05 16:36:53 -0800752static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
753{
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800754 mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache,
755 pte_list_desc_cache);
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800756 mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache);
Xiao Guangronge8ad9a72010-05-13 10:06:02 +0800757 mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache,
758 mmu_page_header_cache);
Avi Kivity714b93d2007-01-05 16:36:53 -0800759}
760
Takuya Yoshikawa80feb892012-05-29 23:54:26 +0900761static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
Avi Kivity714b93d2007-01-05 16:36:53 -0800762{
763 void *p;
764
765 BUG_ON(!mc->nobjs);
766 p = mc->objects[--mc->nobjs];
Avi Kivity714b93d2007-01-05 16:36:53 -0800767 return p;
768}
769
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800770static struct pte_list_desc *mmu_alloc_pte_list_desc(struct kvm_vcpu *vcpu)
Avi Kivity714b93d2007-01-05 16:36:53 -0800771{
Takuya Yoshikawa80feb892012-05-29 23:54:26 +0900772 return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_list_desc_cache);
Avi Kivity714b93d2007-01-05 16:36:53 -0800773}
774
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800775static void mmu_free_pte_list_desc(struct pte_list_desc *pte_list_desc)
Avi Kivity714b93d2007-01-05 16:36:53 -0800776{
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800777 kmem_cache_free(pte_list_desc_cache, pte_list_desc);
Avi Kivity714b93d2007-01-05 16:36:53 -0800778}
779
Lai Jiangshan2032a932010-05-26 16:49:59 +0800780static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index)
781{
782 if (!sp->role.direct)
783 return sp->gfns[index];
784
785 return sp->gfn + (index << ((sp->role.level - 1) * PT64_LEVEL_BITS));
786}
787
788static void kvm_mmu_page_set_gfn(struct kvm_mmu_page *sp, int index, gfn_t gfn)
789{
790 if (sp->role.direct)
791 BUG_ON(gfn != kvm_mmu_page_get_gfn(sp, index));
792 else
793 sp->gfns[index] = gfn;
794}
795
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800796/*
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900797 * Return the pointer to the large page information for a given gfn,
798 * handling slots that are not large page aligned.
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300799 */
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900800static struct kvm_lpage_info *lpage_info_slot(gfn_t gfn,
801 struct kvm_memory_slot *slot,
802 int level)
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300803{
804 unsigned long idx;
805
Takuya Yoshikawafb03cb62012-02-08 12:59:10 +0900806 idx = gfn_to_index(gfn, slot->base_gfn, level);
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900807 return &slot->arch.lpage_info[level - 2][idx];
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300808}
809
810static void account_shadowed(struct kvm *kvm, gfn_t gfn)
811{
Joerg Roedeld25797b2009-07-27 16:30:43 +0200812 struct kvm_memory_slot *slot;
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900813 struct kvm_lpage_info *linfo;
Joerg Roedeld25797b2009-07-27 16:30:43 +0200814 int i;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300815
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300816 slot = gfn_to_memslot(kvm, gfn);
Joerg Roedeld25797b2009-07-27 16:30:43 +0200817 for (i = PT_DIRECTORY_LEVEL;
818 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900819 linfo = lpage_info_slot(gfn, slot, i);
820 linfo->write_count += 1;
Joerg Roedeld25797b2009-07-27 16:30:43 +0200821 }
Xiao Guangrong332b2072011-05-15 23:20:27 +0800822 kvm->arch.indirect_shadow_pages++;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300823}
824
825static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
826{
Joerg Roedeld25797b2009-07-27 16:30:43 +0200827 struct kvm_memory_slot *slot;
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900828 struct kvm_lpage_info *linfo;
Joerg Roedeld25797b2009-07-27 16:30:43 +0200829 int i;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300830
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300831 slot = gfn_to_memslot(kvm, gfn);
Joerg Roedeld25797b2009-07-27 16:30:43 +0200832 for (i = PT_DIRECTORY_LEVEL;
833 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900834 linfo = lpage_info_slot(gfn, slot, i);
835 linfo->write_count -= 1;
836 WARN_ON(linfo->write_count < 0);
Joerg Roedeld25797b2009-07-27 16:30:43 +0200837 }
Xiao Guangrong332b2072011-05-15 23:20:27 +0800838 kvm->arch.indirect_shadow_pages--;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300839}
840
Joerg Roedeld25797b2009-07-27 16:30:43 +0200841static int has_wrprotected_page(struct kvm *kvm,
842 gfn_t gfn,
843 int level)
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300844{
Izik Eidus28430992008-10-03 17:40:32 +0300845 struct kvm_memory_slot *slot;
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900846 struct kvm_lpage_info *linfo;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300847
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300848 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300849 if (slot) {
Takuya Yoshikawad4dbf472010-12-07 12:59:07 +0900850 linfo = lpage_info_slot(gfn, slot, level);
851 return linfo->write_count;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300852 }
853
854 return 1;
855}
856
Joerg Roedeld25797b2009-07-27 16:30:43 +0200857static int host_mapping_level(struct kvm *kvm, gfn_t gfn)
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300858{
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100859 unsigned long page_size;
Joerg Roedeld25797b2009-07-27 16:30:43 +0200860 int i, ret = 0;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300861
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100862 page_size = kvm_host_page_size(kvm, gfn);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300863
Joerg Roedeld25797b2009-07-27 16:30:43 +0200864 for (i = PT_PAGE_TABLE_LEVEL;
865 i < (PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES); ++i) {
866 if (page_size >= KVM_HPAGE_SIZE(i))
867 ret = i;
868 else
869 break;
870 }
871
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -0300872 return ret;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300873}
874
Xiao Guangrong5d163b12011-03-09 15:43:00 +0800875static struct kvm_memory_slot *
876gfn_to_memslot_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t gfn,
877 bool no_dirty_log)
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300878{
879 struct kvm_memory_slot *slot;
Xiao Guangrong5d163b12011-03-09 15:43:00 +0800880
881 slot = gfn_to_memslot(vcpu->kvm, gfn);
882 if (!slot || slot->flags & KVM_MEMSLOT_INVALID ||
883 (no_dirty_log && slot->dirty_bitmap))
884 slot = NULL;
885
886 return slot;
887}
888
889static bool mapping_level_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t large_gfn)
890{
Stevea0a8eab2011-06-17 10:25:39 +0800891 return !gfn_to_memslot_dirty_bitmap(vcpu, large_gfn, true);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800892}
893
894static int mapping_level(struct kvm_vcpu *vcpu, gfn_t large_gfn)
895{
896 int host_level, level, max_level;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300897
Joerg Roedeld25797b2009-07-27 16:30:43 +0200898 host_level = host_mapping_level(vcpu->kvm, large_gfn);
899
900 if (host_level == PT_PAGE_TABLE_LEVEL)
901 return host_level;
902
Xiao Guangrong55dd98c2013-02-05 15:26:54 +0800903 max_level = min(kvm_x86_ops->get_lpage_level(), host_level);
Sheng Yang878403b2010-01-05 19:02:29 +0800904
905 for (level = PT_DIRECTORY_LEVEL; level <= max_level; ++level)
Joerg Roedeld25797b2009-07-27 16:30:43 +0200906 if (has_wrprotected_page(vcpu->kvm, large_gfn, level))
907 break;
Joerg Roedeld25797b2009-07-27 16:30:43 +0200908
909 return level - 1;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300910}
911
912/*
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800913 * Pte mapping structures:
914 *
915 * If pte_list bit zero is zero, then pte_list point to the spte.
916 *
917 * If pte_list bit zero is one, (then pte_list & ~1) points to a struct
918 * pte_list_desc containing more mappings.
919 *
920 * Returns the number of pte entries before the spte was added or zero if
921 * the spte was not added.
922 *
923 */
924static int pte_list_add(struct kvm_vcpu *vcpu, u64 *spte,
925 unsigned long *pte_list)
926{
927 struct pte_list_desc *desc;
928 int i, count = 0;
929
930 if (!*pte_list) {
931 rmap_printk("pte_list_add: %p %llx 0->1\n", spte, *spte);
932 *pte_list = (unsigned long)spte;
933 } else if (!(*pte_list & 1)) {
934 rmap_printk("pte_list_add: %p %llx 1->many\n", spte, *spte);
935 desc = mmu_alloc_pte_list_desc(vcpu);
936 desc->sptes[0] = (u64 *)*pte_list;
937 desc->sptes[1] = spte;
938 *pte_list = (unsigned long)desc | 1;
939 ++count;
940 } else {
941 rmap_printk("pte_list_add: %p %llx many->many\n", spte, *spte);
942 desc = (struct pte_list_desc *)(*pte_list & ~1ul);
943 while (desc->sptes[PTE_LIST_EXT-1] && desc->more) {
944 desc = desc->more;
945 count += PTE_LIST_EXT;
946 }
947 if (desc->sptes[PTE_LIST_EXT-1]) {
948 desc->more = mmu_alloc_pte_list_desc(vcpu);
949 desc = desc->more;
950 }
951 for (i = 0; desc->sptes[i]; ++i)
952 ++count;
953 desc->sptes[i] = spte;
954 }
955 return count;
956}
957
Xiao Guangrong53c07b12011-05-15 23:26:20 +0800958static void
959pte_list_desc_remove_entry(unsigned long *pte_list, struct pte_list_desc *desc,
960 int i, struct pte_list_desc *prev_desc)
961{
962 int j;
963
964 for (j = PTE_LIST_EXT - 1; !desc->sptes[j] && j > i; --j)
965 ;
966 desc->sptes[i] = desc->sptes[j];
967 desc->sptes[j] = NULL;
968 if (j != 0)
969 return;
970 if (!prev_desc && !desc->more)
971 *pte_list = (unsigned long)desc->sptes[0];
972 else
973 if (prev_desc)
974 prev_desc->more = desc->more;
975 else
976 *pte_list = (unsigned long)desc->more | 1;
977 mmu_free_pte_list_desc(desc);
978}
979
980static void pte_list_remove(u64 *spte, unsigned long *pte_list)
981{
982 struct pte_list_desc *desc;
983 struct pte_list_desc *prev_desc;
984 int i;
985
986 if (!*pte_list) {
987 printk(KERN_ERR "pte_list_remove: %p 0->BUG\n", spte);
988 BUG();
989 } else if (!(*pte_list & 1)) {
990 rmap_printk("pte_list_remove: %p 1->0\n", spte);
991 if ((u64 *)*pte_list != spte) {
992 printk(KERN_ERR "pte_list_remove: %p 1->BUG\n", spte);
993 BUG();
994 }
995 *pte_list = 0;
996 } else {
997 rmap_printk("pte_list_remove: %p many->many\n", spte);
998 desc = (struct pte_list_desc *)(*pte_list & ~1ul);
999 prev_desc = NULL;
1000 while (desc) {
1001 for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i)
1002 if (desc->sptes[i] == spte) {
1003 pte_list_desc_remove_entry(pte_list,
1004 desc, i,
1005 prev_desc);
1006 return;
1007 }
1008 prev_desc = desc;
1009 desc = desc->more;
1010 }
1011 pr_err("pte_list_remove: %p many->many\n", spte);
1012 BUG();
1013 }
1014}
1015
Xiao Guangrong67052b32011-05-15 23:27:08 +08001016typedef void (*pte_list_walk_fn) (u64 *spte);
1017static void pte_list_walk(unsigned long *pte_list, pte_list_walk_fn fn)
1018{
1019 struct pte_list_desc *desc;
1020 int i;
1021
1022 if (!*pte_list)
1023 return;
1024
1025 if (!(*pte_list & 1))
1026 return fn((u64 *)*pte_list);
1027
1028 desc = (struct pte_list_desc *)(*pte_list & ~1ul);
1029 while (desc) {
1030 for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i)
1031 fn(desc->sptes[i]);
1032 desc = desc->more;
1033 }
1034}
1035
Takuya Yoshikawa9373e2c02012-01-17 19:51:20 +09001036static unsigned long *__gfn_to_rmap(gfn_t gfn, int level,
Takuya Yoshikawa9b9b1492011-11-14 18:22:28 +09001037 struct kvm_memory_slot *slot)
1038{
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09001039 unsigned long idx;
Takuya Yoshikawa9b9b1492011-11-14 18:22:28 +09001040
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09001041 idx = gfn_to_index(gfn, slot->base_gfn, level);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09001042 return &slot->arch.rmap[level - PT_PAGE_TABLE_LEVEL][idx];
Takuya Yoshikawa9b9b1492011-11-14 18:22:28 +09001043}
1044
Xiao Guangrong53c07b12011-05-15 23:26:20 +08001045/*
Izik Eidus290fc382007-09-27 14:11:22 +02001046 * Take gfn and return the reverse mapping to it.
Izik Eidus290fc382007-09-27 14:11:22 +02001047 */
Joerg Roedel44ad9942009-07-27 16:30:42 +02001048static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level)
Izik Eidus290fc382007-09-27 14:11:22 +02001049{
1050 struct kvm_memory_slot *slot;
1051
1052 slot = gfn_to_memslot(kvm, gfn);
Takuya Yoshikawa9373e2c02012-01-17 19:51:20 +09001053 return __gfn_to_rmap(gfn, level, slot);
Izik Eidus290fc382007-09-27 14:11:22 +02001054}
1055
Xiao Guangrongf759e2b2011-09-22 16:53:17 +08001056static bool rmap_can_add(struct kvm_vcpu *vcpu)
1057{
1058 struct kvm_mmu_memory_cache *cache;
1059
1060 cache = &vcpu->arch.mmu_pte_list_desc_cache;
1061 return mmu_memory_cache_free_objects(cache);
1062}
1063
Joerg Roedel44ad9942009-07-27 16:30:42 +02001064static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001065{
Avi Kivity4db35312007-11-21 15:28:32 +02001066 struct kvm_mmu_page *sp;
Izik Eidus290fc382007-09-27 14:11:22 +02001067 unsigned long *rmapp;
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001068
Avi Kivity4db35312007-11-21 15:28:32 +02001069 sp = page_header(__pa(spte));
Lai Jiangshan2032a932010-05-26 16:49:59 +08001070 kvm_mmu_page_set_gfn(sp, spte - sp->spt, gfn);
Joerg Roedel44ad9942009-07-27 16:30:42 +02001071 rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level);
Xiao Guangrong53c07b12011-05-15 23:26:20 +08001072 return pte_list_add(vcpu, spte, rmapp);
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001073}
1074
Izik Eidus290fc382007-09-27 14:11:22 +02001075static void rmap_remove(struct kvm *kvm, u64 *spte)
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001076{
Avi Kivity4db35312007-11-21 15:28:32 +02001077 struct kvm_mmu_page *sp;
Lai Jiangshan2032a932010-05-26 16:49:59 +08001078 gfn_t gfn;
Izik Eidus290fc382007-09-27 14:11:22 +02001079 unsigned long *rmapp;
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001080
Avi Kivity4db35312007-11-21 15:28:32 +02001081 sp = page_header(__pa(spte));
Lai Jiangshan2032a932010-05-26 16:49:59 +08001082 gfn = kvm_mmu_page_get_gfn(sp, spte - sp->spt);
1083 rmapp = gfn_to_rmap(kvm, gfn, sp->role.level);
Xiao Guangrong53c07b12011-05-15 23:26:20 +08001084 pte_list_remove(spte, rmapp);
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001085}
1086
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001087/*
1088 * Used by the following functions to iterate through the sptes linked by a
1089 * rmap. All fields are private and not assumed to be used outside.
1090 */
1091struct rmap_iterator {
1092 /* private fields */
1093 struct pte_list_desc *desc; /* holds the sptep if not NULL */
1094 int pos; /* index of the sptep */
1095};
1096
1097/*
1098 * Iteration must be started by this function. This should also be used after
1099 * removing/dropping sptes from the rmap link because in such cases the
1100 * information in the itererator may not be valid.
1101 *
1102 * Returns sptep if found, NULL otherwise.
1103 */
1104static u64 *rmap_get_first(unsigned long rmap, struct rmap_iterator *iter)
1105{
1106 if (!rmap)
1107 return NULL;
1108
1109 if (!(rmap & 1)) {
1110 iter->desc = NULL;
1111 return (u64 *)rmap;
1112 }
1113
1114 iter->desc = (struct pte_list_desc *)(rmap & ~1ul);
1115 iter->pos = 0;
1116 return iter->desc->sptes[iter->pos];
1117}
1118
1119/*
1120 * Must be used with a valid iterator: e.g. after rmap_get_first().
1121 *
1122 * Returns sptep if found, NULL otherwise.
1123 */
1124static u64 *rmap_get_next(struct rmap_iterator *iter)
1125{
1126 if (iter->desc) {
1127 if (iter->pos < PTE_LIST_EXT - 1) {
1128 u64 *sptep;
1129
1130 ++iter->pos;
1131 sptep = iter->desc->sptes[iter->pos];
1132 if (sptep)
1133 return sptep;
1134 }
1135
1136 iter->desc = iter->desc->more;
1137
1138 if (iter->desc) {
1139 iter->pos = 0;
1140 /* desc->sptes[0] cannot be NULL */
1141 return iter->desc->sptes[iter->pos];
1142 }
1143 }
1144
1145 return NULL;
1146}
1147
Xiao Guangrongc3707952011-07-12 03:28:04 +08001148static void drop_spte(struct kvm *kvm, u64 *sptep)
Xiao Guangronge4b502e2010-07-16 11:28:09 +08001149{
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +08001150 if (mmu_spte_clear_track_bits(sptep))
Marcelo Tosattieb45fda2010-10-25 11:58:22 -02001151 rmap_remove(kvm, sptep);
Avi Kivitybe38d272010-06-06 14:31:27 +03001152}
1153
Xiao Guangrong8e22f952012-06-20 15:57:39 +08001154
1155static bool __drop_large_spte(struct kvm *kvm, u64 *sptep)
1156{
1157 if (is_large_pte(*sptep)) {
1158 WARN_ON(page_header(__pa(sptep))->role.level ==
1159 PT_PAGE_TABLE_LEVEL);
1160 drop_spte(kvm, sptep);
1161 --kvm->stat.lpages;
1162 return true;
1163 }
1164
1165 return false;
1166}
1167
1168static void drop_large_spte(struct kvm_vcpu *vcpu, u64 *sptep)
1169{
1170 if (__drop_large_spte(vcpu->kvm, sptep))
1171 kvm_flush_remote_tlbs(vcpu->kvm);
1172}
1173
1174/*
Xiao Guangrong49fde342012-06-20 15:58:58 +08001175 * Write-protect on the specified @sptep, @pt_protect indicates whether
Xiao Guangrongc126d942014-04-17 17:06:14 +08001176 * spte write-protection is caused by protecting shadow page table.
Xiao Guangrong49fde342012-06-20 15:58:58 +08001177 *
1178 * Note: write protection is difference between drity logging and spte
1179 * protection:
1180 * - for dirty logging, the spte can be set to writable at anytime if
1181 * its dirty bitmap is properly set.
1182 * - for spte protection, the spte can be writable only after unsync-ing
1183 * shadow page.
Xiao Guangrong8e22f952012-06-20 15:57:39 +08001184 *
Xiao Guangrongc126d942014-04-17 17:06:14 +08001185 * Return true if tlb need be flushed.
Xiao Guangrong8e22f952012-06-20 15:57:39 +08001186 */
Xiao Guangrongc126d942014-04-17 17:06:14 +08001187static bool spte_write_protect(struct kvm *kvm, u64 *sptep, bool pt_protect)
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001188{
1189 u64 spte = *sptep;
1190
Xiao Guangrong49fde342012-06-20 15:58:58 +08001191 if (!is_writable_pte(spte) &&
1192 !(pt_protect && spte_is_locklessly_modifiable(spte)))
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001193 return false;
1194
1195 rmap_printk("rmap_write_protect: spte %p %llx\n", sptep, *sptep);
1196
Xiao Guangrong49fde342012-06-20 15:58:58 +08001197 if (pt_protect)
1198 spte &= ~SPTE_MMU_WRITEABLE;
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001199 spte = spte & ~PT_WRITABLE_MASK;
Xiao Guangrong49fde342012-06-20 15:58:58 +08001200
Xiao Guangrongc126d942014-04-17 17:06:14 +08001201 return mmu_spte_update(sptep, spte);
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001202}
1203
Xiao Guangrong49fde342012-06-20 15:58:58 +08001204static bool __rmap_write_protect(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa245c3912013-01-08 19:44:09 +09001205 bool pt_protect)
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001206{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001207 u64 *sptep;
1208 struct rmap_iterator iter;
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001209 bool flush = false;
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001210
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001211 for (sptep = rmap_get_first(*rmapp, &iter); sptep;) {
1212 BUG_ON(!(*sptep & PT_PRESENT_MASK));
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001213
Xiao Guangrongc126d942014-04-17 17:06:14 +08001214 flush |= spte_write_protect(kvm, sptep, pt_protect);
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001215 sptep = rmap_get_next(&iter);
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001216 }
1217
Xiao Guangrongd13bc5b2012-06-20 15:57:15 +08001218 return flush;
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001219}
1220
Takuya Yoshikawa5dc99b232012-03-01 19:32:16 +09001221/**
1222 * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages
1223 * @kvm: kvm instance
1224 * @slot: slot to protect
1225 * @gfn_offset: start of the BITS_PER_LONG pages we care about
1226 * @mask: indicates which pages we should protect
1227 *
1228 * Used when we do not need to care about huge page mappings: e.g. during dirty
1229 * logging we do not have any such mappings.
1230 */
1231void kvm_mmu_write_protect_pt_masked(struct kvm *kvm,
1232 struct kvm_memory_slot *slot,
1233 gfn_t gfn_offset, unsigned long mask)
Izik Eidus98348e92007-10-16 14:42:30 +02001234{
Izik Eidus290fc382007-09-27 14:11:22 +02001235 unsigned long *rmapp;
Takuya Yoshikawa5dc99b232012-03-01 19:32:16 +09001236
1237 while (mask) {
Takuya Yoshikawa65fbe372012-08-01 18:02:01 +09001238 rmapp = __gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask),
1239 PT_PAGE_TABLE_LEVEL, slot);
Takuya Yoshikawa245c3912013-01-08 19:44:09 +09001240 __rmap_write_protect(kvm, rmapp, false);
Takuya Yoshikawa5dc99b232012-03-01 19:32:16 +09001241
1242 /* clear the first set bit */
1243 mask &= mask - 1;
1244 }
1245}
1246
Xiao Guangrong2f845692012-06-20 15:56:53 +08001247static bool rmap_write_protect(struct kvm *kvm, u64 gfn)
Takuya Yoshikawa5dc99b232012-03-01 19:32:16 +09001248{
1249 struct kvm_memory_slot *slot;
1250 unsigned long *rmapp;
1251 int i;
Xiao Guangrong2f845692012-06-20 15:56:53 +08001252 bool write_protected = false;
Takuya Yoshikawa5dc99b232012-03-01 19:32:16 +09001253
1254 slot = gfn_to_memslot(kvm, gfn);
Avi Kivity374cbac2007-01-05 16:36:43 -08001255
Takuya Yoshikawaa0ed4602012-03-01 19:31:22 +09001256 for (i = PT_PAGE_TABLE_LEVEL;
Joerg Roedel44ad9942009-07-27 16:30:42 +02001257 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
Takuya Yoshikawa9373e2c02012-01-17 19:51:20 +09001258 rmapp = __gfn_to_rmap(gfn, i, slot);
Takuya Yoshikawa245c3912013-01-08 19:44:09 +09001259 write_protected |= __rmap_write_protect(kvm, rmapp, true);
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001260 }
1261
Marcelo Tosattib1a36822008-12-01 22:32:03 -02001262 return write_protected;
Avi Kivity374cbac2007-01-05 16:36:43 -08001263}
1264
Frederik Deweerdt8a8365c2009-10-09 11:42:56 +00001265static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001266 struct kvm_memory_slot *slot, unsigned long data)
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001267{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001268 u64 *sptep;
1269 struct rmap_iterator iter;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001270 int need_tlb_flush = 0;
1271
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001272 while ((sptep = rmap_get_first(*rmapp, &iter))) {
1273 BUG_ON(!(*sptep & PT_PRESENT_MASK));
1274 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", sptep, *sptep);
1275
1276 drop_spte(kvm, sptep);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001277 need_tlb_flush = 1;
1278 }
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001279
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001280 return need_tlb_flush;
1281}
1282
Frederik Deweerdt8a8365c2009-10-09 11:42:56 +00001283static int kvm_set_pte_rmapp(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001284 struct kvm_memory_slot *slot, unsigned long data)
Izik Eidus3da0dd42009-09-23 21:47:18 +03001285{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001286 u64 *sptep;
1287 struct rmap_iterator iter;
Izik Eidus3da0dd42009-09-23 21:47:18 +03001288 int need_flush = 0;
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001289 u64 new_spte;
Izik Eidus3da0dd42009-09-23 21:47:18 +03001290 pte_t *ptep = (pte_t *)data;
1291 pfn_t new_pfn;
1292
1293 WARN_ON(pte_huge(*ptep));
1294 new_pfn = pte_pfn(*ptep);
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001295
1296 for (sptep = rmap_get_first(*rmapp, &iter); sptep;) {
1297 BUG_ON(!is_shadow_present_pte(*sptep));
1298 rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", sptep, *sptep);
1299
Izik Eidus3da0dd42009-09-23 21:47:18 +03001300 need_flush = 1;
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001301
Izik Eidus3da0dd42009-09-23 21:47:18 +03001302 if (pte_write(*ptep)) {
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001303 drop_spte(kvm, sptep);
1304 sptep = rmap_get_first(*rmapp, &iter);
Izik Eidus3da0dd42009-09-23 21:47:18 +03001305 } else {
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001306 new_spte = *sptep & ~PT64_BASE_ADDR_MASK;
Izik Eidus3da0dd42009-09-23 21:47:18 +03001307 new_spte |= (u64)new_pfn << PAGE_SHIFT;
1308
1309 new_spte &= ~PT_WRITABLE_MASK;
1310 new_spte &= ~SPTE_HOST_WRITEABLE;
Avi Kivityb79b93f2010-06-06 15:46:44 +03001311 new_spte &= ~shadow_accessed_mask;
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001312
1313 mmu_spte_clear_track_bits(sptep);
1314 mmu_spte_set(sptep, new_spte);
1315 sptep = rmap_get_next(&iter);
Izik Eidus3da0dd42009-09-23 21:47:18 +03001316 }
1317 }
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001318
Izik Eidus3da0dd42009-09-23 21:47:18 +03001319 if (need_flush)
1320 kvm_flush_remote_tlbs(kvm);
1321
1322 return 0;
1323}
1324
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001325static int kvm_handle_hva_range(struct kvm *kvm,
1326 unsigned long start,
1327 unsigned long end,
1328 unsigned long data,
1329 int (*handler)(struct kvm *kvm,
1330 unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001331 struct kvm_memory_slot *slot,
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001332 unsigned long data))
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001333{
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +08001334 int j;
Takuya Yoshikawaf3953022012-07-02 17:58:48 +09001335 int ret = 0;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001336 struct kvm_memslots *slots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +08001337 struct kvm_memory_slot *memslot;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001338
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08001339 slots = kvm_memslots(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001340
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +08001341 kvm_for_each_memslot(memslot, slots) {
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001342 unsigned long hva_start, hva_end;
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001343 gfn_t gfn_start, gfn_end;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001344
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001345 hva_start = max(start, memslot->userspace_addr);
1346 hva_end = min(end, memslot->userspace_addr +
1347 (memslot->npages << PAGE_SHIFT));
1348 if (hva_start >= hva_end)
1349 continue;
1350 /*
1351 * {gfn(page) | page intersects with [hva_start, hva_end)} =
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001352 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001353 */
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001354 gfn_start = hva_to_gfn_memslot(hva_start, memslot);
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001355 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
Joerg Roedel852e3c12009-07-27 16:30:44 +02001356
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001357 for (j = PT_PAGE_TABLE_LEVEL;
1358 j < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++j) {
1359 unsigned long idx, idx_end;
1360 unsigned long *rmapp;
Joerg Roedel852e3c12009-07-27 16:30:44 +02001361
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001362 /*
1363 * {idx(page_j) | page_j intersects with
1364 * [hva_start, hva_end)} = {idx, idx+1, ..., idx_end}.
1365 */
1366 idx = gfn_to_index(gfn_start, memslot->base_gfn, j);
1367 idx_end = gfn_to_index(gfn_end - 1, memslot->base_gfn, j);
Andrea Arcangeli6e3e2432010-07-16 11:52:55 +02001368
Takuya Yoshikawabcd3ef52012-07-02 17:59:33 +09001369 rmapp = __gfn_to_rmap(gfn_start, j, memslot);
1370
1371 for (; idx <= idx_end; ++idx)
1372 ret |= handler(kvm, rmapp++, memslot, data);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001373 }
1374 }
1375
Takuya Yoshikawaf3953022012-07-02 17:58:48 +09001376 return ret;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001377}
1378
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001379static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
1380 unsigned long data,
1381 int (*handler)(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001382 struct kvm_memory_slot *slot,
Takuya Yoshikawa84504ef2012-07-02 17:55:48 +09001383 unsigned long data))
1384{
1385 return kvm_handle_hva_range(kvm, hva, hva + 1, data, handler);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001386}
1387
1388int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
1389{
Izik Eidus3da0dd42009-09-23 21:47:18 +03001390 return kvm_handle_hva(kvm, hva, 0, kvm_unmap_rmapp);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001391}
1392
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +09001393int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
1394{
1395 return kvm_handle_hva_range(kvm, start, end, 0, kvm_unmap_rmapp);
1396}
1397
Izik Eidus3da0dd42009-09-23 21:47:18 +03001398void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
1399{
Frederik Deweerdt8a8365c2009-10-09 11:42:56 +00001400 kvm_handle_hva(kvm, hva, (unsigned long)&pte, kvm_set_pte_rmapp);
Izik Eidus3da0dd42009-09-23 21:47:18 +03001401}
1402
Frederik Deweerdt8a8365c2009-10-09 11:42:56 +00001403static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001404 struct kvm_memory_slot *slot, unsigned long data)
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001405{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001406 u64 *sptep;
Michael S. Tsirkin79f702a2012-06-03 11:34:08 +03001407 struct rmap_iterator uninitialized_var(iter);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001408 int young = 0;
1409
Rik van Riel6316e1c2010-02-03 16:11:03 -05001410 /*
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001411 * In case of absence of EPT Access and Dirty Bits supports,
1412 * emulate the accessed bit for EPT, by checking if this page has
Rik van Riel6316e1c2010-02-03 16:11:03 -05001413 * an EPT mapping, and clearing it if it does. On the next access,
1414 * a new EPT mapping will be established.
1415 * This has some overhead, but not as much as the cost of swapping
1416 * out actively used pages or breaking up actively used hugepages.
1417 */
Takuya Yoshikawaf3953022012-07-02 17:58:48 +09001418 if (!shadow_accessed_mask) {
1419 young = kvm_unmap_rmapp(kvm, rmapp, slot, data);
1420 goto out;
1421 }
Sheng Yang534e38b2008-09-08 15:12:30 +08001422
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001423 for (sptep = rmap_get_first(*rmapp, &iter); sptep;
1424 sptep = rmap_get_next(&iter)) {
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001425 BUG_ON(!is_shadow_present_pte(*sptep));
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001426
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001427 if (*sptep & shadow_accessed_mask) {
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001428 young = 1;
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001429 clear_bit((ffs(shadow_accessed_mask) - 1),
1430 (unsigned long *)sptep);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001431 }
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001432 }
Takuya Yoshikawaf3953022012-07-02 17:58:48 +09001433out:
1434 /* @data has hva passed to kvm_age_hva(). */
1435 trace_kvm_age_page(data, slot, young);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001436 return young;
1437}
1438
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001439static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001440 struct kvm_memory_slot *slot, unsigned long data)
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001441{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001442 u64 *sptep;
1443 struct rmap_iterator iter;
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001444 int young = 0;
1445
1446 /*
1447 * If there's no access bit in the secondary pte set by the
1448 * hardware it's up to gup-fast/gup to set the access bit in
1449 * the primary pte or in the page structure.
1450 */
1451 if (!shadow_accessed_mask)
1452 goto out;
1453
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001454 for (sptep = rmap_get_first(*rmapp, &iter); sptep;
1455 sptep = rmap_get_next(&iter)) {
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001456 BUG_ON(!is_shadow_present_pte(*sptep));
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09001457
Xudong Hao3f6d8c82012-05-22 11:23:15 +08001458 if (*sptep & shadow_accessed_mask) {
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001459 young = 1;
1460 break;
1461 }
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001462 }
1463out:
1464 return young;
1465}
1466
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03001467#define RMAP_RECYCLE_THRESHOLD 1000
1468
Joerg Roedel852e3c12009-07-27 16:30:44 +02001469static void rmap_recycle(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03001470{
1471 unsigned long *rmapp;
Joerg Roedel852e3c12009-07-27 16:30:44 +02001472 struct kvm_mmu_page *sp;
1473
1474 sp = page_header(__pa(spte));
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03001475
Joerg Roedel852e3c12009-07-27 16:30:44 +02001476 rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level);
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03001477
Takuya Yoshikawa048212d2012-07-02 17:57:59 +09001478 kvm_unmap_rmapp(vcpu->kvm, rmapp, NULL, 0);
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03001479 kvm_flush_remote_tlbs(vcpu->kvm);
1480}
1481
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001482int kvm_age_hva(struct kvm *kvm, unsigned long hva)
1483{
Takuya Yoshikawaf3953022012-07-02 17:58:48 +09001484 return kvm_handle_hva(kvm, hva, hva, kvm_age_rmapp);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001485}
1486
Andrea Arcangeli8ee53822011-01-13 15:47:10 -08001487int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
1488{
1489 return kvm_handle_hva(kvm, hva, 0, kvm_test_age_rmapp);
1490}
1491
Yaozu Dongd6c69ee2007-04-25 14:17:25 +08001492#ifdef MMU_DEBUG
Avi Kivity47ad8e62007-05-06 15:50:58 +03001493static int is_empty_shadow_page(u64 *spt)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001494{
Avi Kivity139bdb22007-01-05 16:36:50 -08001495 u64 *pos;
1496 u64 *end;
1497
Avi Kivity47ad8e62007-05-06 15:50:58 +03001498 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
Avi Kivity3c915512008-05-20 16:21:13 +03001499 if (is_shadow_present_pte(*pos)) {
Harvey Harrisonb8688d52008-03-03 12:59:56 -08001500 printk(KERN_ERR "%s: %p %llx\n", __func__,
Avi Kivity139bdb22007-01-05 16:36:50 -08001501 pos, *pos);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001502 return 0;
Avi Kivity139bdb22007-01-05 16:36:50 -08001503 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001504 return 1;
1505}
Yaozu Dongd6c69ee2007-04-25 14:17:25 +08001506#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -08001507
Dave Hansen45221ab2010-08-19 18:11:37 -07001508/*
1509 * This value is the sum of all of the kvm instances's
1510 * kvm->arch.n_used_mmu_pages values. We need a global,
1511 * aggregate version in order to make the slab shrinker
1512 * faster
1513 */
1514static inline void kvm_mod_used_mmu_pages(struct kvm *kvm, int nr)
1515{
1516 kvm->arch.n_used_mmu_pages += nr;
1517 percpu_counter_add(&kvm_total_used_mmu_pages, nr);
1518}
1519
Gleb Natapov834be0d2013-01-30 16:45:05 +02001520static void kvm_mmu_free_page(struct kvm_mmu_page *sp)
Avi Kivity260746c2007-01-05 16:36:49 -08001521{
Avi Kivity4db35312007-11-21 15:28:32 +02001522 ASSERT(is_empty_shadow_page(sp->spt));
Xiao Guangrong77758342010-06-04 21:53:54 +08001523 hlist_del(&sp->hash_link);
Xiao Guangrongbd4c86e2011-07-12 03:27:14 +08001524 list_del(&sp->link);
1525 free_page((unsigned long)sp->spt);
Gleb Natapov834be0d2013-01-30 16:45:05 +02001526 if (!sp->role.direct)
1527 free_page((unsigned long)sp->gfns);
Xiao Guangronge8ad9a72010-05-13 10:06:02 +08001528 kmem_cache_free(mmu_page_header_cache, sp);
Avi Kivity260746c2007-01-05 16:36:49 -08001529}
1530
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001531static unsigned kvm_page_table_hashfn(gfn_t gfn)
1532{
Dong, Eddie1ae0a132008-01-07 13:20:25 +02001533 return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1);
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001534}
1535
Xiao Guangrong67052b32011-05-15 23:27:08 +08001536static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
1537 struct kvm_mmu_page *sp, u64 *parent_pte)
1538{
1539 if (!parent_pte)
1540 return;
1541
1542 pte_list_add(vcpu, parent_pte, &sp->parent_ptes);
1543}
1544
1545static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
1546 u64 *parent_pte)
1547{
1548 pte_list_remove(parent_pte, &sp->parent_ptes);
1549}
1550
Xiao Guangrongbcdd9a92011-05-15 23:28:29 +08001551static void drop_parent_pte(struct kvm_mmu_page *sp,
1552 u64 *parent_pte)
1553{
1554 mmu_page_remove_parent_pte(sp, parent_pte);
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +08001555 mmu_spte_clear_no_track(parent_pte);
Xiao Guangrongbcdd9a92011-05-15 23:28:29 +08001556}
1557
Avi Kivity25c0de22007-01-05 16:36:42 -08001558static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
Lai Jiangshan2032a932010-05-26 16:49:59 +08001559 u64 *parent_pte, int direct)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001560{
Avi Kivity4db35312007-11-21 15:28:32 +02001561 struct kvm_mmu_page *sp;
Takuya Yoshikawa7ddca7e2013-03-21 19:33:43 +09001562
Takuya Yoshikawa80feb892012-05-29 23:54:26 +09001563 sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache);
1564 sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache);
Lai Jiangshan2032a932010-05-26 16:49:59 +08001565 if (!direct)
Takuya Yoshikawa80feb892012-05-29 23:54:26 +09001566 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache);
Avi Kivity4db35312007-11-21 15:28:32 +02001567 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08001568
1569 /*
1570 * The active_mmu_pages list is the FIFO list, do not move the
1571 * page until it is zapped. kvm_zap_obsolete_pages depends on
1572 * this feature. See the comments in kvm_zap_obsolete_pages().
1573 */
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08001574 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
Xiao Guangrong67052b32011-05-15 23:27:08 +08001575 sp->parent_ptes = 0;
1576 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
Dave Hansen45221ab2010-08-19 18:11:37 -07001577 kvm_mod_used_mmu_pages(vcpu->kvm, +1);
Avi Kivity4db35312007-11-21 15:28:32 +02001578 return sp;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001579}
1580
Xiao Guangrong67052b32011-05-15 23:27:08 +08001581static void mark_unsync(u64 *spte);
Xiao Guangrong6b184932010-04-16 21:29:17 +08001582static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp)
Marcelo Tosatti0074ff62008-09-23 13:18:40 -03001583{
Xiao Guangrong67052b32011-05-15 23:27:08 +08001584 pte_list_walk(&sp->parent_ptes, mark_unsync);
Xiao Guangrong1047df12010-06-11 21:35:15 +08001585}
1586
Xiao Guangrong67052b32011-05-15 23:27:08 +08001587static void mark_unsync(u64 *spte)
Xiao Guangrong1047df12010-06-11 21:35:15 +08001588{
Xiao Guangrong67052b32011-05-15 23:27:08 +08001589 struct kvm_mmu_page *sp;
Xiao Guangrong1047df12010-06-11 21:35:15 +08001590 unsigned int index;
1591
Xiao Guangrong67052b32011-05-15 23:27:08 +08001592 sp = page_header(__pa(spte));
Xiao Guangrong1047df12010-06-11 21:35:15 +08001593 index = spte - sp->spt;
1594 if (__test_and_set_bit(index, sp->unsync_child_bitmap))
1595 return;
1596 if (sp->unsync_children++)
1597 return;
1598 kvm_mmu_mark_parents_unsync(sp);
Marcelo Tosatti0074ff62008-09-23 13:18:40 -03001599}
1600
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03001601static int nonpaging_sync_page(struct kvm_vcpu *vcpu,
Xiao Guangronga4a8e6f2010-11-19 17:04:03 +08001602 struct kvm_mmu_page *sp)
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03001603{
1604 return 1;
1605}
1606
Marcelo Tosattia7052892008-09-23 13:18:35 -03001607static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
1608{
1609}
1610
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08001611static void nonpaging_update_pte(struct kvm_vcpu *vcpu,
1612 struct kvm_mmu_page *sp, u64 *spte,
Xiao Guangrong7c562522011-03-28 10:29:27 +08001613 const void *pte)
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08001614{
1615 WARN_ON(1);
1616}
1617
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001618#define KVM_PAGE_ARRAY_NR 16
1619
1620struct kvm_mmu_pages {
1621 struct mmu_page_and_offset {
1622 struct kvm_mmu_page *sp;
1623 unsigned int idx;
1624 } page[KVM_PAGE_ARRAY_NR];
1625 unsigned int nr;
1626};
1627
Hannes Edercded19f2009-02-21 02:19:13 +01001628static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp,
1629 int idx)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001630{
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001631 int i;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001632
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001633 if (sp->unsync)
1634 for (i=0; i < pvec->nr; i++)
1635 if (pvec->page[i].sp == sp)
1636 return 0;
1637
1638 pvec->page[pvec->nr].sp = sp;
1639 pvec->page[pvec->nr].idx = idx;
1640 pvec->nr++;
1641 return (pvec->nr == KVM_PAGE_ARRAY_NR);
1642}
1643
1644static int __mmu_unsync_walk(struct kvm_mmu_page *sp,
1645 struct kvm_mmu_pages *pvec)
1646{
1647 int i, ret, nr_unsync_leaf = 0;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001648
Takuya Yoshikawa37178b82011-11-29 14:02:45 +09001649 for_each_set_bit(i, sp->unsync_child_bitmap, 512) {
Xiao Guangrong7a8f1a72010-06-11 21:34:04 +08001650 struct kvm_mmu_page *child;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001651 u64 ent = sp->spt[i];
1652
Xiao Guangrong7a8f1a72010-06-11 21:34:04 +08001653 if (!is_shadow_present_pte(ent) || is_large_pte(ent))
1654 goto clear_child_bitmap;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001655
Xiao Guangrong7a8f1a72010-06-11 21:34:04 +08001656 child = page_header(ent & PT64_BASE_ADDR_MASK);
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001657
Xiao Guangrong7a8f1a72010-06-11 21:34:04 +08001658 if (child->unsync_children) {
1659 if (mmu_pages_add(pvec, child, i))
1660 return -ENOSPC;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001661
Xiao Guangrong7a8f1a72010-06-11 21:34:04 +08001662 ret = __mmu_unsync_walk(child, pvec);
1663 if (!ret)
1664 goto clear_child_bitmap;
1665 else if (ret > 0)
1666 nr_unsync_leaf += ret;
1667 else
1668 return ret;
1669 } else if (child->unsync) {
1670 nr_unsync_leaf++;
1671 if (mmu_pages_add(pvec, child, i))
1672 return -ENOSPC;
1673 } else
1674 goto clear_child_bitmap;
1675
1676 continue;
1677
1678clear_child_bitmap:
1679 __clear_bit(i, sp->unsync_child_bitmap);
1680 sp->unsync_children--;
1681 WARN_ON((int)sp->unsync_children < 0);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001682 }
1683
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001684
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001685 return nr_unsync_leaf;
1686}
1687
1688static int mmu_unsync_walk(struct kvm_mmu_page *sp,
1689 struct kvm_mmu_pages *pvec)
1690{
1691 if (!sp->unsync_children)
1692 return 0;
1693
1694 mmu_pages_add(pvec, sp, 0);
1695 return __mmu_unsync_walk(sp, pvec);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001696}
1697
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001698static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1699{
1700 WARN_ON(!sp->unsync);
Xiao Guangrong5e1b3dd2010-04-28 11:55:06 +08001701 trace_kvm_mmu_sync_page(sp);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001702 sp->unsync = 0;
1703 --kvm->stat.mmu_unsync;
1704}
1705
Xiao Guangrong77758342010-06-04 21:53:54 +08001706static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp,
1707 struct list_head *invalid_list);
1708static void kvm_mmu_commit_zap_page(struct kvm *kvm,
1709 struct list_head *invalid_list);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001710
Xiao Guangrongf34d2512013-05-31 08:36:28 +08001711/*
1712 * NOTE: we should pay more attention on the zapped-obsolete page
1713 * (is_obsolete_sp(sp) && sp->role.invalid) when you do hash list walk
1714 * since it has been deleted from active_mmu_pages but still can be found
1715 * at hast list.
1716 *
1717 * for_each_gfn_indirect_valid_sp has skipped that kind of page and
1718 * kvm_mmu_get_page(), the only user of for_each_gfn_sp(), has skipped
1719 * all the obsolete pages.
1720 */
Takuya Yoshikawa1044b032013-03-06 16:05:07 +09001721#define for_each_gfn_sp(_kvm, _sp, _gfn) \
1722 hlist_for_each_entry(_sp, \
1723 &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)], hash_link) \
1724 if ((_sp)->gfn != (_gfn)) {} else
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001725
Takuya Yoshikawa1044b032013-03-06 16:05:07 +09001726#define for_each_gfn_indirect_valid_sp(_kvm, _sp, _gfn) \
1727 for_each_gfn_sp(_kvm, _sp, _gfn) \
1728 if ((_sp)->role.direct || (_sp)->role.invalid) {} else
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001729
Xiao Guangrongf918b442010-06-11 21:30:36 +08001730/* @sp->gfn should be write-protected at the call site */
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001731static int __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001732 struct list_head *invalid_list, bool clear_unsync)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001733{
Avi Kivity5b7e0102010-04-14 19:20:03 +03001734 if (sp->role.cr4_pae != !!is_pae(vcpu)) {
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001735 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001736 return 1;
1737 }
1738
Xiao Guangrongf918b442010-06-11 21:30:36 +08001739 if (clear_unsync)
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001740 kvm_unlink_unsync_page(vcpu->kvm, sp);
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001741
Xiao Guangronga4a8e6f2010-11-19 17:04:03 +08001742 if (vcpu->arch.mmu.sync_page(vcpu, sp)) {
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001743 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001744 return 1;
1745 }
1746
1747 kvm_mmu_flush_tlb(vcpu);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001748 return 0;
1749}
1750
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001751static int kvm_sync_page_transient(struct kvm_vcpu *vcpu,
1752 struct kvm_mmu_page *sp)
1753{
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001754 LIST_HEAD(invalid_list);
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001755 int ret;
1756
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001757 ret = __kvm_sync_page(vcpu, sp, &invalid_list, false);
Xiao Guangrongbe71e062010-06-11 21:31:38 +08001758 if (ret)
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001759 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
1760
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001761 return ret;
1762}
1763
Xiao Guangronge37fa782011-11-30 17:43:24 +08001764#ifdef CONFIG_KVM_MMU_AUDIT
1765#include "mmu_audit.c"
1766#else
1767static void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point) { }
1768static void mmu_audit_disable(void) { }
1769#endif
1770
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001771static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
1772 struct list_head *invalid_list)
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001773{
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001774 return __kvm_sync_page(vcpu, sp, invalid_list, true);
Xiao Guangrong1d9dc7e2010-05-15 18:51:24 +08001775}
1776
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001777/* @gfn should be write-protected at the call site */
1778static void kvm_sync_pages(struct kvm_vcpu *vcpu, gfn_t gfn)
1779{
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001780 struct kvm_mmu_page *s;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001781 LIST_HEAD(invalid_list);
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001782 bool flush = false;
1783
Sasha Levinb67bfe02013-02-27 17:06:00 -08001784 for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) {
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001785 if (!s->unsync)
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001786 continue;
1787
1788 WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL);
Xiao Guangronga4a8e6f2010-11-19 17:04:03 +08001789 kvm_unlink_unsync_page(vcpu->kvm, s);
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001790 if ((s->role.cr4_pae != !!is_pae(vcpu)) ||
Xiao Guangronga4a8e6f2010-11-19 17:04:03 +08001791 (vcpu->arch.mmu.sync_page(vcpu, s))) {
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001792 kvm_mmu_prepare_zap_page(vcpu->kvm, s, &invalid_list);
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001793 continue;
1794 }
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001795 flush = true;
1796 }
1797
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001798 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001799 if (flush)
1800 kvm_mmu_flush_tlb(vcpu);
1801}
1802
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001803struct mmu_page_path {
1804 struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1];
1805 unsigned int idx[PT64_ROOT_LEVEL-1];
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001806};
1807
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001808#define for_each_sp(pvec, sp, parents, i) \
1809 for (i = mmu_pages_next(&pvec, &parents, -1), \
1810 sp = pvec.page[i].sp; \
1811 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1812 i = mmu_pages_next(&pvec, &parents, i))
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001813
Hannes Edercded19f2009-02-21 02:19:13 +01001814static int mmu_pages_next(struct kvm_mmu_pages *pvec,
1815 struct mmu_page_path *parents,
1816 int i)
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001817{
1818 int n;
1819
1820 for (n = i+1; n < pvec->nr; n++) {
1821 struct kvm_mmu_page *sp = pvec->page[n].sp;
1822
1823 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
1824 parents->idx[0] = pvec->page[n].idx;
1825 return n;
1826 }
1827
1828 parents->parent[sp->role.level-2] = sp;
1829 parents->idx[sp->role.level-1] = pvec->page[n].idx;
1830 }
1831
1832 return n;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001833}
1834
Hannes Edercded19f2009-02-21 02:19:13 +01001835static void mmu_pages_clear_parents(struct mmu_page_path *parents)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001836{
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001837 struct kvm_mmu_page *sp;
1838 unsigned int level = 0;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001839
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001840 do {
1841 unsigned int idx = parents->idx[level];
1842
1843 sp = parents->parent[level];
1844 if (!sp)
1845 return;
1846
1847 --sp->unsync_children;
1848 WARN_ON((int)sp->unsync_children < 0);
1849 __clear_bit(idx, sp->unsync_child_bitmap);
1850 level++;
1851 } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children);
1852}
1853
1854static void kvm_mmu_pages_init(struct kvm_mmu_page *parent,
1855 struct mmu_page_path *parents,
1856 struct kvm_mmu_pages *pvec)
1857{
1858 parents->parent[parent->role.level-1] = NULL;
1859 pvec->nr = 0;
1860}
1861
1862static void mmu_sync_children(struct kvm_vcpu *vcpu,
1863 struct kvm_mmu_page *parent)
1864{
1865 int i;
1866 struct kvm_mmu_page *sp;
1867 struct mmu_page_path parents;
1868 struct kvm_mmu_pages pages;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001869 LIST_HEAD(invalid_list);
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001870
1871 kvm_mmu_pages_init(parent, &parents, &pages);
1872 while (mmu_unsync_walk(parent, &pages)) {
Xiao Guangrong2f845692012-06-20 15:56:53 +08001873 bool protected = false;
Marcelo Tosattib1a36822008-12-01 22:32:03 -02001874
1875 for_each_sp(pages, sp, parents, i)
1876 protected |= rmap_write_protect(vcpu->kvm, sp->gfn);
1877
1878 if (protected)
1879 kvm_flush_remote_tlbs(vcpu->kvm);
1880
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001881 for_each_sp(pages, sp, parents, i) {
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001882 kvm_sync_page(vcpu, sp, &invalid_list);
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001883 mmu_pages_clear_parents(&parents);
1884 }
Xiao Guangrongd98ba052010-06-04 21:55:29 +08001885 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001886 cond_resched_lock(&vcpu->kvm->mmu_lock);
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02001887 kvm_mmu_pages_init(parent, &parents, &pages);
1888 }
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001889}
1890
Xiao Guangrongc3707952011-07-12 03:28:04 +08001891static void init_shadow_page_table(struct kvm_mmu_page *sp)
1892{
1893 int i;
1894
1895 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
1896 sp->spt[i] = 0ull;
1897}
1898
Xiao Guangronga30f47c2011-09-22 16:58:36 +08001899static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp)
1900{
1901 sp->write_flooding_count = 0;
1902}
1903
1904static void clear_sp_write_flooding_count(u64 *spte)
1905{
1906 struct kvm_mmu_page *sp = page_header(__pa(spte));
1907
1908 __clear_sp_write_flooding_count(sp);
1909}
1910
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08001911static bool is_obsolete_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
1912{
1913 return unlikely(sp->mmu_valid_gen != kvm->arch.mmu_valid_gen);
1914}
1915
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001916static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
1917 gfn_t gfn,
1918 gva_t gaddr,
1919 unsigned level,
Avi Kivityf6e2c022009-01-11 13:02:10 +02001920 int direct,
Avi Kivity41074d02007-12-09 17:00:02 +02001921 unsigned access,
Avi Kivityf7d9c7b2008-02-26 22:12:10 +02001922 u64 *parent_pte)
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001923{
1924 union kvm_mmu_page_role role;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001925 unsigned quadrant;
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001926 struct kvm_mmu_page *sp;
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001927 bool need_sync = false;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001928
Avi Kivitya770f6f2008-12-21 19:20:09 +02001929 role = vcpu->arch.mmu.base_role;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001930 role.level = level;
Avi Kivityf6e2c022009-01-11 13:02:10 +02001931 role.direct = direct;
Avi Kivity84b0c8c2010-03-14 10:16:40 +02001932 if (role.direct)
Avi Kivity5b7e0102010-04-14 19:20:03 +03001933 role.cr4_pae = 0;
Avi Kivity41074d02007-12-09 17:00:02 +02001934 role.access = access;
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02001935 if (!vcpu->arch.mmu.direct_map
1936 && vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) {
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001937 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
1938 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
1939 role.quadrant = quadrant;
1940 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001941 for_each_gfn_sp(vcpu->kvm, sp, gfn) {
Xiao Guangrong7f52af72013-05-31 08:36:26 +08001942 if (is_obsolete_sp(vcpu->kvm, sp))
1943 continue;
1944
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001945 if (!need_sync && sp->unsync)
1946 need_sync = true;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001947
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001948 if (sp->role.word != role.word)
1949 continue;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001950
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001951 if (sp->unsync && kvm_sync_page_transient(vcpu, sp))
1952 break;
Xiao Guangronge02aa902010-05-15 18:52:34 +08001953
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001954 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
1955 if (sp->unsync_children) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001956 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001957 kvm_mmu_mark_parents_unsync(sp);
1958 } else if (sp->unsync)
1959 kvm_mmu_mark_parents_unsync(sp);
Xiao Guangronge02aa902010-05-15 18:52:34 +08001960
Xiao Guangronga30f47c2011-09-22 16:58:36 +08001961 __clear_sp_write_flooding_count(sp);
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001962 trace_kvm_mmu_get_page(sp, false);
1963 return sp;
1964 }
Avi Kivitydfc5aa02007-12-18 19:47:18 +02001965 ++vcpu->kvm->stat.mmu_cache_miss;
Lai Jiangshan2032a932010-05-26 16:49:59 +08001966 sp = kvm_mmu_alloc_page(vcpu, parent_pte, direct);
Avi Kivity4db35312007-11-21 15:28:32 +02001967 if (!sp)
1968 return sp;
Avi Kivity4db35312007-11-21 15:28:32 +02001969 sp->gfn = gfn;
1970 sp->role = role;
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08001971 hlist_add_head(&sp->hash_link,
1972 &vcpu->kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]);
Avi Kivityf6e2c022009-01-11 13:02:10 +02001973 if (!direct) {
Marcelo Tosattib1a36822008-12-01 22:32:03 -02001974 if (rmap_write_protect(vcpu->kvm, gfn))
1975 kvm_flush_remote_tlbs(vcpu->kvm);
Xiao Guangrong9f1a1222010-05-24 15:41:33 +08001976 if (level > PT_PAGE_TABLE_LEVEL && need_sync)
1977 kvm_sync_pages(vcpu, gfn);
1978
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03001979 account_shadowed(vcpu->kvm, gfn);
1980 }
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08001981 sp->mmu_valid_gen = vcpu->kvm->arch.mmu_valid_gen;
Xiao Guangrongc3707952011-07-12 03:28:04 +08001982 init_shadow_page_table(sp);
Avi Kivityf691fe12009-07-06 15:58:14 +03001983 trace_kvm_mmu_get_page(sp, true);
Avi Kivity4db35312007-11-21 15:28:32 +02001984 return sp;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08001985}
1986
Avi Kivity2d111232008-12-25 14:39:47 +02001987static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator,
1988 struct kvm_vcpu *vcpu, u64 addr)
1989{
1990 iterator->addr = addr;
1991 iterator->shadow_addr = vcpu->arch.mmu.root_hpa;
1992 iterator->level = vcpu->arch.mmu.shadow_root_level;
Joerg Roedel81407ca2010-09-10 17:31:00 +02001993
1994 if (iterator->level == PT64_ROOT_LEVEL &&
1995 vcpu->arch.mmu.root_level < PT64_ROOT_LEVEL &&
1996 !vcpu->arch.mmu.direct_map)
1997 --iterator->level;
1998
Avi Kivity2d111232008-12-25 14:39:47 +02001999 if (iterator->level == PT32E_ROOT_LEVEL) {
2000 iterator->shadow_addr
2001 = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
2002 iterator->shadow_addr &= PT64_BASE_ADDR_MASK;
2003 --iterator->level;
2004 if (!iterator->shadow_addr)
2005 iterator->level = 0;
2006 }
2007}
2008
2009static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator)
2010{
2011 if (iterator->level < PT_PAGE_TABLE_LEVEL)
2012 return false;
Marcelo Tosatti4d889542009-06-11 12:07:41 -03002013
Avi Kivity2d111232008-12-25 14:39:47 +02002014 iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level);
2015 iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index;
2016 return true;
2017}
2018
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08002019static void __shadow_walk_next(struct kvm_shadow_walk_iterator *iterator,
2020 u64 spte)
Avi Kivity2d111232008-12-25 14:39:47 +02002021{
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08002022 if (is_last_spte(spte, iterator->level)) {
Xiao Guangrong052331b2011-07-12 03:21:17 +08002023 iterator->level = 0;
2024 return;
2025 }
2026
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08002027 iterator->shadow_addr = spte & PT64_BASE_ADDR_MASK;
Avi Kivity2d111232008-12-25 14:39:47 +02002028 --iterator->level;
2029}
2030
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08002031static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator)
2032{
2033 return __shadow_walk_next(iterator, *iterator->sptep);
2034}
2035
Yang Zhang7a1638c2013-08-05 11:07:13 +03002036static void link_shadow_page(u64 *sptep, struct kvm_mmu_page *sp, bool accessed)
Avi Kivity32ef26a2010-07-13 14:27:04 +03002037{
2038 u64 spte;
2039
Yang Zhang7a1638c2013-08-05 11:07:13 +03002040 BUILD_BUG_ON(VMX_EPT_READABLE_MASK != PT_PRESENT_MASK ||
2041 VMX_EPT_WRITABLE_MASK != PT_WRITABLE_MASK);
2042
Xiao Guangrong24db2732013-02-05 15:28:02 +08002043 spte = __pa(sp->spt) | PT_PRESENT_MASK | PT_WRITABLE_MASK |
Yang Zhang7a1638c2013-08-05 11:07:13 +03002044 shadow_user_mask | shadow_x_mask;
2045
2046 if (accessed)
2047 spte |= shadow_accessed_mask;
Xiao Guangrong24db2732013-02-05 15:28:02 +08002048
Xiao Guangrong1df9f2d2011-07-12 03:30:35 +08002049 mmu_spte_set(sptep, spte);
Avi Kivity32ef26a2010-07-13 14:27:04 +03002050}
2051
Avi Kivitya357bd22010-07-13 14:27:07 +03002052static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep,
2053 unsigned direct_access)
2054{
2055 if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) {
2056 struct kvm_mmu_page *child;
2057
2058 /*
2059 * For the direct sp, if the guest pte's dirty bit
2060 * changed form clean to dirty, it will corrupt the
2061 * sp's access: allow writable in the read-only sp,
2062 * so we should update the spte at this point to get
2063 * a new sp with the correct access.
2064 */
2065 child = page_header(*sptep & PT64_BASE_ADDR_MASK);
2066 if (child->role.access == direct_access)
2067 return;
2068
Xiao Guangrongbcdd9a92011-05-15 23:28:29 +08002069 drop_parent_pte(child, sptep);
Avi Kivitya357bd22010-07-13 14:27:07 +03002070 kvm_flush_remote_tlbs(vcpu->kvm);
2071 }
2072}
2073
Xiao Guangrong505aef82011-09-22 16:56:06 +08002074static bool mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp,
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08002075 u64 *spte)
2076{
2077 u64 pte;
2078 struct kvm_mmu_page *child;
2079
2080 pte = *spte;
2081 if (is_shadow_present_pte(pte)) {
Xiao Guangrong505aef82011-09-22 16:56:06 +08002082 if (is_last_spte(pte, sp->role.level)) {
Xiao Guangrongc3707952011-07-12 03:28:04 +08002083 drop_spte(kvm, spte);
Xiao Guangrong505aef82011-09-22 16:56:06 +08002084 if (is_large_pte(pte))
2085 --kvm->stat.lpages;
2086 } else {
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08002087 child = page_header(pte & PT64_BASE_ADDR_MASK);
Xiao Guangrongbcdd9a92011-05-15 23:28:29 +08002088 drop_parent_pte(child, spte);
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08002089 }
Xiao Guangrong505aef82011-09-22 16:56:06 +08002090 return true;
2091 }
2092
2093 if (is_mmio_spte(pte))
Xiao Guangrongce88dec2011-07-12 03:33:44 +08002094 mmu_spte_clear_no_track(spte);
Xiao Guangrongc3707952011-07-12 03:28:04 +08002095
Xiao Guangrong505aef82011-09-22 16:56:06 +08002096 return false;
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08002097}
2098
Avi Kivity90cb0522007-07-17 13:04:56 +03002099static void kvm_mmu_page_unlink_children(struct kvm *kvm,
Avi Kivity4db35312007-11-21 15:28:32 +02002100 struct kvm_mmu_page *sp)
Avi Kivitya4360362007-01-05 16:36:45 -08002101{
Avi Kivity697fe2e2007-01-05 16:36:46 -08002102 unsigned i;
Avi Kivity697fe2e2007-01-05 16:36:46 -08002103
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08002104 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
2105 mmu_page_zap_pte(kvm, sp, sp->spt + i);
Avi Kivitya4360362007-01-05 16:36:45 -08002106}
2107
Avi Kivity4db35312007-11-21 15:28:32 +02002108static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
Avi Kivitycea0f0e2007-01-05 16:36:43 -08002109{
Avi Kivity4db35312007-11-21 15:28:32 +02002110 mmu_page_remove_parent_pte(sp, parent_pte);
Avi Kivitya4360362007-01-05 16:36:45 -08002111}
2112
Avi Kivity31aa2b42008-07-11 17:59:46 +03002113static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
Avi Kivitya4360362007-01-05 16:36:45 -08002114{
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09002115 u64 *sptep;
2116 struct rmap_iterator iter;
Avi Kivitya4360362007-01-05 16:36:45 -08002117
Takuya Yoshikawa1e3f42f2012-03-21 23:50:34 +09002118 while ((sptep = rmap_get_first(sp->parent_ptes, &iter)))
2119 drop_parent_pte(sp, sptep);
Avi Kivity31aa2b42008-07-11 17:59:46 +03002120}
2121
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002122static int mmu_zap_unsync_children(struct kvm *kvm,
Xiao Guangrong77758342010-06-04 21:53:54 +08002123 struct kvm_mmu_page *parent,
2124 struct list_head *invalid_list)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002125{
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002126 int i, zapped = 0;
2127 struct mmu_page_path parents;
2128 struct kvm_mmu_pages pages;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002129
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002130 if (parent->role.level == PT_PAGE_TABLE_LEVEL)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002131 return 0;
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002132
2133 kvm_mmu_pages_init(parent, &parents, &pages);
2134 while (mmu_unsync_walk(parent, &pages)) {
2135 struct kvm_mmu_page *sp;
2136
2137 for_each_sp(pages, sp, parents, i) {
Xiao Guangrong77758342010-06-04 21:53:54 +08002138 kvm_mmu_prepare_zap_page(kvm, sp, invalid_list);
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002139 mmu_pages_clear_parents(&parents);
Xiao Guangrong77662e02010-04-16 16:34:42 +08002140 zapped++;
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002141 }
Marcelo Tosatti60c8aec2008-12-01 22:32:02 -02002142 kvm_mmu_pages_init(parent, &parents, &pages);
2143 }
2144
2145 return zapped;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002146}
2147
Xiao Guangrong77758342010-06-04 21:53:54 +08002148static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp,
2149 struct list_head *invalid_list)
Avi Kivity31aa2b42008-07-11 17:59:46 +03002150{
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002151 int ret;
Avi Kivityf691fe12009-07-06 15:58:14 +03002152
Xiao Guangrong77758342010-06-04 21:53:54 +08002153 trace_kvm_mmu_prepare_zap_page(sp);
Avi Kivity31aa2b42008-07-11 17:59:46 +03002154 ++kvm->stat.mmu_shadow_zapped;
Xiao Guangrong77758342010-06-04 21:53:54 +08002155 ret = mmu_zap_unsync_children(kvm, sp, invalid_list);
Avi Kivity4db35312007-11-21 15:28:32 +02002156 kvm_mmu_page_unlink_children(kvm, sp);
Avi Kivity31aa2b42008-07-11 17:59:46 +03002157 kvm_mmu_unlink_parents(kvm, sp);
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08002158
Avi Kivityf6e2c022009-01-11 13:02:10 +02002159 if (!sp->role.invalid && !sp->role.direct)
Avi Kivity5b5c6a52008-07-11 18:07:26 +03002160 unaccount_shadowed(kvm, sp->gfn);
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08002161
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002162 if (sp->unsync)
2163 kvm_unlink_unsync_page(kvm, sp);
Avi Kivity4db35312007-11-21 15:28:32 +02002164 if (!sp->root_count) {
Gui Jianfeng54a4f022010-05-05 09:03:49 +08002165 /* Count self */
2166 ret++;
Xiao Guangrong77758342010-06-04 21:53:54 +08002167 list_move(&sp->link, invalid_list);
Xiao Guangrongaa6bd182011-07-12 03:26:40 +08002168 kvm_mod_used_mmu_pages(kvm, -1);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05002169 } else {
Avi Kivity5b5c6a52008-07-11 18:07:26 +03002170 list_move(&sp->link, &kvm->arch.active_mmu_pages);
Gleb Natapov05988d72013-05-31 08:36:30 +08002171
2172 /*
2173 * The obsolete pages can not be used on any vcpus.
2174 * See the comments in kvm_mmu_invalidate_zap_all_pages().
2175 */
2176 if (!sp->role.invalid && !is_obsolete_sp(kvm, sp))
2177 kvm_reload_remote_mmus(kvm);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05002178 }
Xiao Guangrong77758342010-06-04 21:53:54 +08002179
2180 sp->role.invalid = 1;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002181 return ret;
Avi Kivitya4360362007-01-05 16:36:45 -08002182}
2183
Xiao Guangrong77758342010-06-04 21:53:54 +08002184static void kvm_mmu_commit_zap_page(struct kvm *kvm,
2185 struct list_head *invalid_list)
2186{
Takuya Yoshikawa945315b2013-03-06 16:05:52 +09002187 struct kvm_mmu_page *sp, *nsp;
Xiao Guangrong77758342010-06-04 21:53:54 +08002188
2189 if (list_empty(invalid_list))
2190 return;
2191
Avi Kivityc1427862012-05-14 15:44:06 +03002192 /*
2193 * wmb: make sure everyone sees our modifications to the page tables
2194 * rmb: make sure we see changes to vcpu->mode
2195 */
2196 smp_mb();
2197
2198 /*
2199 * Wait for all vcpus to exit guest mode and/or lockless shadow
2200 * page table walks.
2201 */
Xiao Guangrong77758342010-06-04 21:53:54 +08002202 kvm_flush_remote_tlbs(kvm);
2203
Takuya Yoshikawa945315b2013-03-06 16:05:52 +09002204 list_for_each_entry_safe(sp, nsp, invalid_list, link) {
Xiao Guangrong77758342010-06-04 21:53:54 +08002205 WARN_ON(!sp->role.invalid || sp->root_count);
Xiao Guangrongaa6bd182011-07-12 03:26:40 +08002206 kvm_mmu_free_page(sp);
Takuya Yoshikawa945315b2013-03-06 16:05:52 +09002207 }
Xiao Guangrong77758342010-06-04 21:53:54 +08002208}
2209
Takuya Yoshikawa5da59602013-03-06 16:06:58 +09002210static bool prepare_zap_oldest_mmu_page(struct kvm *kvm,
2211 struct list_head *invalid_list)
2212{
2213 struct kvm_mmu_page *sp;
2214
2215 if (list_empty(&kvm->arch.active_mmu_pages))
2216 return false;
2217
2218 sp = list_entry(kvm->arch.active_mmu_pages.prev,
2219 struct kvm_mmu_page, link);
2220 kvm_mmu_prepare_zap_page(kvm, sp, invalid_list);
2221
2222 return true;
2223}
2224
Izik Eidus82ce2c92007-10-02 18:52:55 +02002225/*
2226 * Changing the number of mmu pages allocated to the vm
Dave Hansen49d5ca22010-08-19 18:11:28 -07002227 * Note: if goal_nr_mmu_pages is too small, you will get dead lock
Izik Eidus82ce2c92007-10-02 18:52:55 +02002228 */
Dave Hansen49d5ca22010-08-19 18:11:28 -07002229void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int goal_nr_mmu_pages)
Izik Eidus82ce2c92007-10-02 18:52:55 +02002230{
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002231 LIST_HEAD(invalid_list);
Izik Eidus82ce2c92007-10-02 18:52:55 +02002232
Takuya Yoshikawab34cb592013-01-08 19:46:07 +09002233 spin_lock(&kvm->mmu_lock);
2234
Dave Hansen49d5ca22010-08-19 18:11:28 -07002235 if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) {
Takuya Yoshikawa5da59602013-03-06 16:06:58 +09002236 /* Need to free some mmu pages to achieve the goal. */
2237 while (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages)
2238 if (!prepare_zap_oldest_mmu_page(kvm, &invalid_list))
2239 break;
Izik Eidus82ce2c92007-10-02 18:52:55 +02002240
Xiao Guangrongaa6bd182011-07-12 03:26:40 +08002241 kvm_mmu_commit_zap_page(kvm, &invalid_list);
Dave Hansen49d5ca22010-08-19 18:11:28 -07002242 goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages;
Izik Eidus82ce2c92007-10-02 18:52:55 +02002243 }
Izik Eidus82ce2c92007-10-02 18:52:55 +02002244
Dave Hansen49d5ca22010-08-19 18:11:28 -07002245 kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages;
Takuya Yoshikawab34cb592013-01-08 19:46:07 +09002246
2247 spin_unlock(&kvm->mmu_lock);
Izik Eidus82ce2c92007-10-02 18:52:55 +02002248}
2249
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08002250int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
Avi Kivitya4360362007-01-05 16:36:45 -08002251{
Avi Kivity4db35312007-11-21 15:28:32 +02002252 struct kvm_mmu_page *sp;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002253 LIST_HEAD(invalid_list);
Avi Kivitya4360362007-01-05 16:36:45 -08002254 int r;
2255
Xiao Guangrong9ad17b102010-08-28 19:19:42 +08002256 pgprintk("%s: looking for gfn %llx\n", __func__, gfn);
Avi Kivitya4360362007-01-05 16:36:45 -08002257 r = 0;
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08002258 spin_lock(&kvm->mmu_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08002259 for_each_gfn_indirect_valid_sp(kvm, sp, gfn) {
Xiao Guangrong9ad17b102010-08-28 19:19:42 +08002260 pgprintk("%s: gfn %llx role %x\n", __func__, gfn,
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08002261 sp->role.word);
2262 r = 1;
Xiao Guangrongf41d3352010-06-04 21:56:11 +08002263 kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list);
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08002264 }
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002265 kvm_mmu_commit_zap_page(kvm, &invalid_list);
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08002266 spin_unlock(&kvm->mmu_lock);
2267
Avi Kivitya4360362007-01-05 16:36:45 -08002268 return r;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08002269}
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08002270EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page);
Avi Kivitycea0f0e2007-01-05 16:36:43 -08002271
Sheng Yang74be52e2008-10-09 16:01:56 +08002272/*
2273 * The function is based on mtrr_type_lookup() in
2274 * arch/x86/kernel/cpu/mtrr/generic.c
2275 */
2276static int get_mtrr_type(struct mtrr_state_type *mtrr_state,
2277 u64 start, u64 end)
2278{
2279 int i;
2280 u64 base, mask;
2281 u8 prev_match, curr_match;
2282 int num_var_ranges = KVM_NR_VAR_MTRR;
2283
2284 if (!mtrr_state->enabled)
2285 return 0xFF;
2286
2287 /* Make end inclusive end, instead of exclusive */
2288 end--;
2289
2290 /* Look in fixed ranges. Just return the type as per start */
2291 if (mtrr_state->have_fixed && (start < 0x100000)) {
2292 int idx;
2293
2294 if (start < 0x80000) {
2295 idx = 0;
2296 idx += (start >> 16);
2297 return mtrr_state->fixed_ranges[idx];
2298 } else if (start < 0xC0000) {
2299 idx = 1 * 8;
2300 idx += ((start - 0x80000) >> 14);
2301 return mtrr_state->fixed_ranges[idx];
2302 } else if (start < 0x1000000) {
2303 idx = 3 * 8;
2304 idx += ((start - 0xC0000) >> 12);
2305 return mtrr_state->fixed_ranges[idx];
2306 }
2307 }
2308
2309 /*
2310 * Look in variable ranges
2311 * Look of multiple ranges matching this address and pick type
2312 * as per MTRR precedence
2313 */
2314 if (!(mtrr_state->enabled & 2))
2315 return mtrr_state->def_type;
2316
2317 prev_match = 0xFF;
2318 for (i = 0; i < num_var_ranges; ++i) {
2319 unsigned short start_state, end_state;
2320
2321 if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11)))
2322 continue;
2323
2324 base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) +
2325 (mtrr_state->var_ranges[i].base_lo & PAGE_MASK);
2326 mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) +
2327 (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK);
2328
2329 start_state = ((start & mask) == (base & mask));
2330 end_state = ((end & mask) == (base & mask));
2331 if (start_state != end_state)
2332 return 0xFE;
2333
2334 if ((start & mask) != (base & mask))
2335 continue;
2336
2337 curr_match = mtrr_state->var_ranges[i].base_lo & 0xff;
2338 if (prev_match == 0xFF) {
2339 prev_match = curr_match;
2340 continue;
2341 }
2342
2343 if (prev_match == MTRR_TYPE_UNCACHABLE ||
2344 curr_match == MTRR_TYPE_UNCACHABLE)
2345 return MTRR_TYPE_UNCACHABLE;
2346
2347 if ((prev_match == MTRR_TYPE_WRBACK &&
2348 curr_match == MTRR_TYPE_WRTHROUGH) ||
2349 (prev_match == MTRR_TYPE_WRTHROUGH &&
2350 curr_match == MTRR_TYPE_WRBACK)) {
2351 prev_match = MTRR_TYPE_WRTHROUGH;
2352 curr_match = MTRR_TYPE_WRTHROUGH;
2353 }
2354
2355 if (prev_match != curr_match)
2356 return MTRR_TYPE_UNCACHABLE;
2357 }
2358
2359 if (prev_match != 0xFF)
2360 return prev_match;
2361
2362 return mtrr_state->def_type;
2363}
2364
Sheng Yang4b12f0d2009-04-27 20:35:42 +08002365u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
Sheng Yang74be52e2008-10-09 16:01:56 +08002366{
2367 u8 mtrr;
2368
2369 mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT,
2370 (gfn << PAGE_SHIFT) + PAGE_SIZE);
2371 if (mtrr == 0xfe || mtrr == 0xff)
2372 mtrr = MTRR_TYPE_WRBACK;
2373 return mtrr;
2374}
Sheng Yang4b12f0d2009-04-27 20:35:42 +08002375EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type);
Sheng Yang74be52e2008-10-09 16:01:56 +08002376
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002377static void __kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002378{
Xiao Guangrong5e1b3dd2010-04-28 11:55:06 +08002379 trace_kvm_mmu_unsync_page(sp);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002380 ++vcpu->kvm->stat.mmu_unsync;
2381 sp->unsync = 1;
Marcelo Tosatti6cffe8c2008-12-01 22:32:04 -02002382
Xiao Guangrong6b184932010-04-16 21:29:17 +08002383 kvm_mmu_mark_parents_unsync(sp);
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002384}
2385
2386static void kvm_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn)
2387{
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002388 struct kvm_mmu_page *s;
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002389
Sasha Levinb67bfe02013-02-27 17:06:00 -08002390 for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) {
Xiao Guangrong7ae680e2010-06-04 21:53:07 +08002391 if (s->unsync)
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002392 continue;
2393 WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL);
2394 __kvm_unsync_page(vcpu, s);
2395 }
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002396}
2397
2398static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
2399 bool can_unsync)
2400{
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002401 struct kvm_mmu_page *s;
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002402 bool need_unsync = false;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002403
Sasha Levinb67bfe02013-02-27 17:06:00 -08002404 for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) {
Xiao Guangrong36a2e672010-06-30 16:02:02 +08002405 if (!can_unsync)
2406 return 1;
2407
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002408 if (s->role.level != PT_PAGE_TABLE_LEVEL)
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002409 return 1;
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002410
Gleb Natapov9bb4f6b2013-01-30 16:45:01 +02002411 if (!s->unsync)
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002412 need_unsync = true;
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002413 }
Xiao Guangrong9cf5cf52010-05-24 15:40:07 +08002414 if (need_unsync)
2415 kvm_unsync_pages(vcpu, gfn);
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002416 return 0;
2417}
2418
Avi Kivityd555c332009-06-10 14:24:23 +03002419static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
Xiao Guangrongc22885052013-01-08 14:36:04 +08002420 unsigned pte_access, int level,
Marcelo Tosattic2d0ee42009-04-05 14:54:47 -03002421 gfn_t gfn, pfn_t pfn, bool speculative,
Lai Jiangshan9bdbba12010-11-19 17:03:22 +08002422 bool can_unsync, bool host_writable)
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002423{
Xiao Guangrong6e7d0352012-06-20 15:58:33 +08002424 u64 spte;
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002425 int ret = 0;
Sheng Yang64d4d522008-10-09 16:01:57 +08002426
Xiao Guangrongf2fd1252013-06-07 16:51:24 +08002427 if (set_mmio_spte(vcpu->kvm, sptep, gfn, pfn, pte_access))
Xiao Guangrongce88dec2011-07-12 03:33:44 +08002428 return 0;
2429
Marcelo Tosatti982c2562010-10-22 14:18:16 -02002430 spte = PT_PRESENT_MASK;
Avi Kivity947da532008-03-18 11:05:52 +02002431 if (!speculative)
Avi Kivity3201b5d2008-08-27 20:01:04 +03002432 spte |= shadow_accessed_mask;
Xiao Guangrong640d9b02011-07-12 03:24:39 +08002433
Sheng Yang7b523452008-04-25 21:13:50 +08002434 if (pte_access & ACC_EXEC_MASK)
2435 spte |= shadow_x_mask;
2436 else
2437 spte |= shadow_nx_mask;
Xiao Guangrong49fde342012-06-20 15:58:58 +08002438
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002439 if (pte_access & ACC_USER_MASK)
Sheng Yang7b523452008-04-25 21:13:50 +08002440 spte |= shadow_user_mask;
Xiao Guangrong49fde342012-06-20 15:58:58 +08002441
Joerg Roedel852e3c12009-07-27 16:30:44 +02002442 if (level > PT_PAGE_TABLE_LEVEL)
Marcelo Tosatti05da4552008-02-23 11:44:30 -03002443 spte |= PT_PAGE_SIZE_MASK;
Avi Kivityb0bc3ee2010-09-13 16:45:28 +02002444 if (tdp_enabled)
Sheng Yang4b12f0d2009-04-27 20:35:42 +08002445 spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn,
2446 kvm_is_mmio_pfn(pfn));
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002447
Lai Jiangshan9bdbba12010-11-19 17:03:22 +08002448 if (host_writable)
Izik Eidus14032832009-09-23 21:47:17 +03002449 spte |= SPTE_HOST_WRITEABLE;
Xiao Guangrongf8e453b2010-12-23 16:09:29 +08002450 else
2451 pte_access &= ~ACC_WRITE_MASK;
Izik Eidus14032832009-09-23 21:47:17 +03002452
Anthony Liguori35149e22008-04-02 14:46:56 -05002453 spte |= (u64)pfn << PAGE_SHIFT;
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002454
Xiao Guangrongc22885052013-01-08 14:36:04 +08002455 if (pte_access & ACC_WRITE_MASK) {
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002456
Xiao Guangrongc2193462012-12-04 07:17:11 +08002457 /*
Xiao Guangrong7751bab2013-01-08 14:36:51 +08002458 * Other vcpu creates new sp in the window between
2459 * mapping_level() and acquiring mmu-lock. We can
2460 * allow guest to retry the access, the mapping can
2461 * be fixed if guest refault.
Xiao Guangrongc2193462012-12-04 07:17:11 +08002462 */
Joerg Roedel852e3c12009-07-27 16:30:44 +02002463 if (level > PT_PAGE_TABLE_LEVEL &&
Xiao Guangrongc2193462012-12-04 07:17:11 +08002464 has_wrprotected_page(vcpu->kvm, gfn, level))
Avi Kivitybe38d272010-06-06 14:31:27 +03002465 goto done;
Marcelo Tosatti38187c82008-09-23 13:18:32 -03002466
Xiao Guangrong49fde342012-06-20 15:58:58 +08002467 spte |= PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE;
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002468
Marcelo Tosattiecc55892008-11-25 15:58:07 +01002469 /*
2470 * Optimization: for pte sync, if spte was writable the hash
2471 * lookup is unnecessary (and expensive). Write protection
2472 * is responsibility of mmu_get_page / kvm_sync_page.
2473 * Same reasoning can be applied to dirty page accounting.
2474 */
Takuya Yoshikawa8dae4442010-01-18 18:45:10 +09002475 if (!can_unsync && is_writable_pte(*sptep))
Marcelo Tosattiecc55892008-11-25 15:58:07 +01002476 goto set_pte;
2477
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03002478 if (mmu_need_write_protect(vcpu, gfn, can_unsync)) {
Xiao Guangrong9ad17b102010-08-28 19:19:42 +08002479 pgprintk("%s: found shadow page for %llx, marking ro\n",
Harvey Harrisonb8688d52008-03-03 12:59:56 -08002480 __func__, gfn);
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002481 ret = 1;
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002482 pte_access &= ~ACC_WRITE_MASK;
Xiao Guangrong49fde342012-06-20 15:58:58 +08002483 spte &= ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE);
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002484 }
2485 }
2486
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002487 if (pte_access & ACC_WRITE_MASK)
2488 mark_page_dirty(vcpu->kvm, gfn);
2489
Marcelo Tosatti38187c82008-09-23 13:18:32 -03002490set_pte:
Xiao Guangrong6e7d0352012-06-20 15:58:33 +08002491 if (mmu_spte_update(sptep, spte))
Xiao Guangrongb330aa02010-11-19 17:02:35 +08002492 kvm_flush_remote_tlbs(vcpu->kvm);
Avi Kivitybe38d272010-06-06 14:31:27 +03002493done:
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002494 return ret;
2495}
2496
Avi Kivityd555c332009-06-10 14:24:23 +03002497static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
Xiao Guangrongf7616202013-02-05 15:27:27 +08002498 unsigned pte_access, int write_fault, int *emulate,
2499 int level, gfn_t gfn, pfn_t pfn, bool speculative,
2500 bool host_writable)
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002501{
2502 int was_rmapped = 0;
Marcelo Tosatti53a27b32009-08-05 15:43:58 -03002503 int rmap_count;
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002504
Xiao Guangrongf7616202013-02-05 15:27:27 +08002505 pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__,
2506 *sptep, write_fault, gfn);
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002507
Avi Kivityd555c332009-06-10 14:24:23 +03002508 if (is_rmap_spte(*sptep)) {
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002509 /*
2510 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
2511 * the parent of the now unreachable PTE.
2512 */
Joerg Roedel852e3c12009-07-27 16:30:44 +02002513 if (level > PT_PAGE_TABLE_LEVEL &&
2514 !is_large_pte(*sptep)) {
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002515 struct kvm_mmu_page *child;
Avi Kivityd555c332009-06-10 14:24:23 +03002516 u64 pte = *sptep;
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002517
2518 child = page_header(pte & PT64_BASE_ADDR_MASK);
Xiao Guangrongbcdd9a92011-05-15 23:28:29 +08002519 drop_parent_pte(child, sptep);
Marcelo Tosatti3be22642010-05-28 09:44:59 -03002520 kvm_flush_remote_tlbs(vcpu->kvm);
Avi Kivityd555c332009-06-10 14:24:23 +03002521 } else if (pfn != spte_to_pfn(*sptep)) {
Xiao Guangrong9ad17b102010-08-28 19:19:42 +08002522 pgprintk("hfn old %llx new %llx\n",
Avi Kivityd555c332009-06-10 14:24:23 +03002523 spte_to_pfn(*sptep), pfn);
Xiao Guangrongc3707952011-07-12 03:28:04 +08002524 drop_spte(vcpu->kvm, sptep);
Xiao Guangrong91546352010-06-30 16:04:06 +08002525 kvm_flush_remote_tlbs(vcpu->kvm);
Joerg Roedel6bed6b92009-02-18 14:08:59 +01002526 } else
2527 was_rmapped = 1;
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002528 }
Joerg Roedel852e3c12009-07-27 16:30:44 +02002529
Xiao Guangrongc22885052013-01-08 14:36:04 +08002530 if (set_spte(vcpu, sptep, pte_access, level, gfn, pfn, speculative,
2531 true, host_writable)) {
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002532 if (write_fault)
Xiao Guangrongb90a0e62011-07-12 03:25:56 +08002533 *emulate = 1;
Xiao Guangrong5304efd2010-06-08 20:05:57 +08002534 kvm_mmu_flush_tlb(vcpu);
Marcelo Tosattia378b4e2008-09-23 13:18:31 -03002535 }
Marcelo Tosatti1e73f9d2008-09-23 13:18:30 -03002536
Xiao Guangrongce88dec2011-07-12 03:33:44 +08002537 if (unlikely(is_mmio_spte(*sptep) && emulate))
2538 *emulate = 1;
2539
Avi Kivityd555c332009-06-10 14:24:23 +03002540 pgprintk("%s: setting spte %llx\n", __func__, *sptep);
Xiao Guangrong9ad17b102010-08-28 19:19:42 +08002541 pgprintk("instantiating %s PTE (%s) at %llx (%llx) addr %p\n",
Avi Kivityd555c332009-06-10 14:24:23 +03002542 is_large_pte(*sptep)? "2MB" : "4kB",
Joerg Roedela205bc12009-07-09 16:36:01 +02002543 *sptep & PT_PRESENT_MASK ?"RW":"R", gfn,
2544 *sptep, sptep);
Avi Kivityd555c332009-06-10 14:24:23 +03002545 if (!was_rmapped && is_large_pte(*sptep))
Marcelo Tosatti05da4552008-02-23 11:44:30 -03002546 ++vcpu->kvm->stat.lpages;
2547
Xiao Guangrongffb61bb2011-07-12 03:22:01 +08002548 if (is_shadow_present_pte(*sptep)) {
Xiao Guangrongffb61bb2011-07-12 03:22:01 +08002549 if (!was_rmapped) {
2550 rmap_count = rmap_add(vcpu, sptep, gfn);
2551 if (rmap_count > RMAP_RECYCLE_THRESHOLD)
2552 rmap_recycle(vcpu, sptep, gfn);
2553 }
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002554 }
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08002555
Xiao Guangrongf3ac1a42012-10-16 20:07:03 +08002556 kvm_release_pfn_clean(pfn);
Avi Kivity1c4f1fd2007-12-09 17:40:31 +02002557}
2558
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002559static pfn_t pte_prefetch_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn,
2560 bool no_dirty_log)
2561{
2562 struct kvm_memory_slot *slot;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002563
Xiao Guangrong5d163b12011-03-09 15:43:00 +08002564 slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, no_dirty_log);
Xiao Guangrong903816f2012-07-17 21:54:11 +08002565 if (!slot)
Xiao Guangrong6c8ee572012-08-03 15:37:54 +08002566 return KVM_PFN_ERR_FAULT;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002567
Xiao Guangrong037d92d2012-08-21 10:59:12 +08002568 return gfn_to_pfn_memslot_atomic(slot, gfn);
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002569}
2570
2571static int direct_pte_prefetch_many(struct kvm_vcpu *vcpu,
2572 struct kvm_mmu_page *sp,
2573 u64 *start, u64 *end)
2574{
2575 struct page *pages[PTE_PREFETCH_NUM];
2576 unsigned access = sp->role.access;
2577 int i, ret;
2578 gfn_t gfn;
2579
2580 gfn = kvm_mmu_page_get_gfn(sp, start - sp->spt);
Xiao Guangrong5d163b12011-03-09 15:43:00 +08002581 if (!gfn_to_memslot_dirty_bitmap(vcpu, gfn, access & ACC_WRITE_MASK))
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002582 return -1;
2583
2584 ret = gfn_to_page_many_atomic(vcpu->kvm, gfn, pages, end - start);
2585 if (ret <= 0)
2586 return -1;
2587
2588 for (i = 0; i < ret; i++, gfn++, start++)
Xiao Guangrongf7616202013-02-05 15:27:27 +08002589 mmu_set_spte(vcpu, start, access, 0, NULL,
Xiao Guangrongc22885052013-01-08 14:36:04 +08002590 sp->role.level, gfn, page_to_pfn(pages[i]),
2591 true, true);
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002592
2593 return 0;
2594}
2595
2596static void __direct_pte_prefetch(struct kvm_vcpu *vcpu,
2597 struct kvm_mmu_page *sp, u64 *sptep)
2598{
2599 u64 *spte, *start = NULL;
2600 int i;
2601
2602 WARN_ON(!sp->role.direct);
2603
2604 i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
2605 spte = sp->spt + i;
2606
2607 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
Xiao Guangrongc3707952011-07-12 03:28:04 +08002608 if (is_shadow_present_pte(*spte) || spte == sptep) {
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002609 if (!start)
2610 continue;
2611 if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0)
2612 break;
2613 start = NULL;
2614 } else if (!start)
2615 start = spte;
2616 }
2617}
2618
2619static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep)
2620{
2621 struct kvm_mmu_page *sp;
2622
2623 /*
2624 * Since it's no accessed bit on EPT, it's no way to
2625 * distinguish between actually accessed translations
2626 * and prefetched, so disable pte prefetch if EPT is
2627 * enabled.
2628 */
2629 if (!shadow_accessed_mask)
2630 return;
2631
2632 sp = page_header(__pa(sptep));
2633 if (sp->role.level > PT_PAGE_TABLE_LEVEL)
2634 return;
2635
2636 __direct_pte_prefetch(vcpu, sp, sptep);
2637}
2638
Joerg Roedel4d9976b2008-02-07 13:47:42 +01002639static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
Xiao Guangrong2ec47392010-12-07 10:34:42 +08002640 int map_writable, int level, gfn_t gfn, pfn_t pfn,
2641 bool prefault)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002642{
Avi Kivity9f652d22008-12-25 14:54:25 +02002643 struct kvm_shadow_walk_iterator iterator;
2644 struct kvm_mmu_page *sp;
Xiao Guangrongb90a0e62011-07-12 03:25:56 +08002645 int emulate = 0;
Avi Kivity9f652d22008-12-25 14:54:25 +02002646 gfn_t pseudo_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002647
Marcelo Tosatti989c6b32013-12-19 15:28:51 -02002648 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2649 return 0;
2650
Avi Kivity9f652d22008-12-25 14:54:25 +02002651 for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) {
Joerg Roedel852e3c12009-07-27 16:30:44 +02002652 if (iterator.level == level) {
Xiao Guangrongf7616202013-02-05 15:27:27 +08002653 mmu_set_spte(vcpu, iterator.sptep, ACC_ALL,
Xiao Guangrongc22885052013-01-08 14:36:04 +08002654 write, &emulate, level, gfn, pfn,
2655 prefault, map_writable);
Xiao Guangrong957ed9e2010-08-22 19:12:48 +08002656 direct_pte_prefetch(vcpu, iterator.sptep);
Avi Kivity9f652d22008-12-25 14:54:25 +02002657 ++vcpu->stat.pf_fixed;
2658 break;
2659 }
2660
Marcelo Tosatti404381c2014-02-24 13:59:32 -03002661 drop_large_spte(vcpu, iterator.sptep);
Xiao Guangrongc3707952011-07-12 03:28:04 +08002662 if (!is_shadow_present_pte(*iterator.sptep)) {
Lai Jiangshanc9fa0b32010-05-26 16:48:25 +08002663 u64 base_addr = iterator.addr;
2664
2665 base_addr &= PT64_LVL_ADDR_MASK(iterator.level);
2666 pseudo_gfn = base_addr >> PAGE_SHIFT;
Avi Kivity9f652d22008-12-25 14:54:25 +02002667 sp = kvm_mmu_get_page(vcpu, pseudo_gfn, iterator.addr,
2668 iterator.level - 1,
2669 1, ACC_ALL, iterator.sptep);
Avi Kivity9f652d22008-12-25 14:54:25 +02002670
Yang Zhang7a1638c2013-08-05 11:07:13 +03002671 link_shadow_page(iterator.sptep, sp, true);
Avi Kivity9f652d22008-12-25 14:54:25 +02002672 }
2673 }
Xiao Guangrongb90a0e62011-07-12 03:25:56 +08002674 return emulate;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002675}
2676
Huang Ying77db5cb2010-10-08 16:24:15 +08002677static void kvm_send_hwpoison_signal(unsigned long address, struct task_struct *tsk)
Huang Yingbf998152010-05-31 14:28:19 +08002678{
Huang Ying77db5cb2010-10-08 16:24:15 +08002679 siginfo_t info;
Huang Yingbf998152010-05-31 14:28:19 +08002680
Huang Ying77db5cb2010-10-08 16:24:15 +08002681 info.si_signo = SIGBUS;
2682 info.si_errno = 0;
2683 info.si_code = BUS_MCEERR_AR;
2684 info.si_addr = (void __user *)address;
2685 info.si_addr_lsb = PAGE_SHIFT;
2686
2687 send_sig_info(SIGBUS, &info, tsk);
Huang Yingbf998152010-05-31 14:28:19 +08002688}
2689
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002690static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, pfn_t pfn)
Huang Yingbf998152010-05-31 14:28:19 +08002691{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08002692 /*
2693 * Do not cache the mmio info caused by writing the readonly gfn
2694 * into the spte otherwise read access on readonly gfn also can
2695 * caused mmio page fault and treat it as mmio access.
2696 * Return 1 to tell kvm to emulate it.
2697 */
2698 if (pfn == KVM_PFN_ERR_RO_FAULT)
2699 return 1;
2700
Xiao Guangronge6c15022012-08-03 15:38:36 +08002701 if (pfn == KVM_PFN_ERR_HWPOISON) {
Xiao Guangrongbebb1062011-07-12 03:23:20 +08002702 kvm_send_hwpoison_signal(gfn_to_hva(vcpu->kvm, gfn), current);
Huang Yingbf998152010-05-31 14:28:19 +08002703 return 0;
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002704 }
Gleb Natapovedba23e2010-07-07 20:16:45 +03002705
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002706 return -EFAULT;
Huang Yingbf998152010-05-31 14:28:19 +08002707}
2708
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002709static void transparent_hugepage_adjust(struct kvm_vcpu *vcpu,
2710 gfn_t *gfnp, pfn_t *pfnp, int *levelp)
2711{
2712 pfn_t pfn = *pfnp;
2713 gfn_t gfn = *gfnp;
2714 int level = *levelp;
2715
2716 /*
2717 * Check if it's a transparent hugepage. If this would be an
2718 * hugetlbfs page, level wouldn't be set to
2719 * PT_PAGE_TABLE_LEVEL and there would be no adjustment done
2720 * here.
2721 */
Xiao Guangrong81c52c52012-10-16 20:10:59 +08002722 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn) &&
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002723 level == PT_PAGE_TABLE_LEVEL &&
2724 PageTransCompound(pfn_to_page(pfn)) &&
2725 !has_wrprotected_page(vcpu->kvm, gfn, PT_DIRECTORY_LEVEL)) {
2726 unsigned long mask;
2727 /*
2728 * mmu_notifier_retry was successful and we hold the
2729 * mmu_lock here, so the pmd can't become splitting
2730 * from under us, and in turn
2731 * __split_huge_page_refcount() can't run from under
2732 * us and we can safely transfer the refcount from
2733 * PG_tail to PG_head as we switch the pfn to tail to
2734 * head.
2735 */
2736 *levelp = level = PT_DIRECTORY_LEVEL;
2737 mask = KVM_PAGES_PER_HPAGE(level) - 1;
2738 VM_BUG_ON((gfn & mask) != (pfn & mask));
2739 if (pfn & mask) {
2740 gfn &= ~mask;
2741 *gfnp = gfn;
2742 kvm_release_pfn_clean(pfn);
2743 pfn &= ~mask;
Xiao Guangrongc3586662012-05-28 14:10:43 +08002744 kvm_get_pfn(pfn);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002745 *pfnp = pfn;
2746 }
2747 }
2748}
2749
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002750static bool handle_abnormal_pfn(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
2751 pfn_t pfn, unsigned access, int *ret_val)
2752{
2753 bool ret = true;
2754
2755 /* The pfn is invalid, report the error! */
Xiao Guangrong81c52c52012-10-16 20:10:59 +08002756 if (unlikely(is_error_pfn(pfn))) {
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002757 *ret_val = kvm_handle_bad_page(vcpu, gfn, pfn);
2758 goto exit;
2759 }
2760
Xiao Guangrongce88dec2011-07-12 03:33:44 +08002761 if (unlikely(is_noslot_pfn(pfn)))
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002762 vcpu_cache_mmio_info(vcpu, gva, gfn, access);
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002763
2764 ret = false;
2765exit:
2766 return ret;
2767}
2768
Xiao Guangronge5552fd2013-07-30 21:01:59 +08002769static bool page_fault_can_be_fast(u32 error_code)
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002770{
2771 /*
Xiao Guangrong1c118b82013-07-18 12:52:37 +08002772 * Do not fix the mmio spte with invalid generation number which
2773 * need to be updated by slow page fault path.
2774 */
2775 if (unlikely(error_code & PFERR_RSVD_MASK))
2776 return false;
2777
2778 /*
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002779 * #PF can be fast only if the shadow page table is present and it
2780 * is caused by write-protect, that means we just need change the
2781 * W bit of the spte which can be done out of mmu-lock.
2782 */
2783 if (!(error_code & PFERR_PRESENT_MASK) ||
2784 !(error_code & PFERR_WRITE_MASK))
2785 return false;
2786
2787 return true;
2788}
2789
2790static bool
Xiao Guangrong92a476c2014-04-17 17:06:13 +08002791fast_pf_fix_direct_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
2792 u64 *sptep, u64 spte)
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002793{
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002794 gfn_t gfn;
2795
2796 WARN_ON(!sp->role.direct);
2797
2798 /*
2799 * The gfn of direct spte is stable since it is calculated
2800 * by sp->gfn.
2801 */
2802 gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt);
2803
2804 if (cmpxchg64(sptep, spte, spte | PT_WRITABLE_MASK) == spte)
2805 mark_page_dirty(vcpu->kvm, gfn);
2806
2807 return true;
2808}
2809
2810/*
2811 * Return value:
2812 * - true: let the vcpu to access on the same address again.
2813 * - false: let the real page fault path to fix it.
2814 */
2815static bool fast_page_fault(struct kvm_vcpu *vcpu, gva_t gva, int level,
2816 u32 error_code)
2817{
2818 struct kvm_shadow_walk_iterator iterator;
Xiao Guangrong92a476c2014-04-17 17:06:13 +08002819 struct kvm_mmu_page *sp;
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002820 bool ret = false;
2821 u64 spte = 0ull;
2822
Marcelo Tosatti37f6a4e2014-01-03 17:09:32 -02002823 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2824 return false;
2825
Xiao Guangronge5552fd2013-07-30 21:01:59 +08002826 if (!page_fault_can_be_fast(error_code))
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002827 return false;
2828
2829 walk_shadow_page_lockless_begin(vcpu);
2830 for_each_shadow_entry_lockless(vcpu, gva, iterator, spte)
2831 if (!is_shadow_present_pte(spte) || iterator.level < level)
2832 break;
2833
2834 /*
2835 * If the mapping has been changed, let the vcpu fault on the
2836 * same address again.
2837 */
2838 if (!is_rmap_spte(spte)) {
2839 ret = true;
2840 goto exit;
2841 }
2842
Xiao Guangrong92a476c2014-04-17 17:06:13 +08002843 sp = page_header(__pa(iterator.sptep));
2844 if (!is_last_spte(spte, sp->role.level))
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002845 goto exit;
2846
2847 /*
2848 * Check if it is a spurious fault caused by TLB lazily flushed.
2849 *
2850 * Need not check the access of upper level table entries since
2851 * they are always ACC_ALL.
2852 */
2853 if (is_writable_pte(spte)) {
2854 ret = true;
2855 goto exit;
2856 }
2857
2858 /*
2859 * Currently, to simplify the code, only the spte write-protected
2860 * by dirty-log can be fast fixed.
2861 */
2862 if (!spte_is_locklessly_modifiable(spte))
2863 goto exit;
2864
2865 /*
Xiao Guangrongc126d942014-04-17 17:06:14 +08002866 * Do not fix write-permission on the large spte since we only dirty
2867 * the first page into the dirty-bitmap in fast_pf_fix_direct_spte()
2868 * that means other pages are missed if its slot is dirty-logged.
2869 *
2870 * Instead, we let the slow page fault path create a normal spte to
2871 * fix the access.
2872 *
2873 * See the comments in kvm_arch_commit_memory_region().
2874 */
2875 if (sp->role.level > PT_PAGE_TABLE_LEVEL)
2876 goto exit;
2877
2878 /*
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002879 * Currently, fast page fault only works for direct mapping since
2880 * the gfn is not stable for indirect shadow page.
2881 * See Documentation/virtual/kvm/locking.txt to get more detail.
2882 */
Xiao Guangrong92a476c2014-04-17 17:06:13 +08002883 ret = fast_pf_fix_direct_spte(vcpu, sp, iterator.sptep, spte);
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002884exit:
Xiao Guangronga72faf22012-06-20 15:59:41 +08002885 trace_fast_page_fault(vcpu, gva, error_code, iterator.sptep,
2886 spte, ret);
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002887 walk_shadow_page_lockless_end(vcpu);
2888
2889 return ret;
2890}
2891
Xiao Guangrong78b2c542010-12-07 10:48:06 +08002892static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn,
Xiao Guangrong060c2ab2010-11-12 14:49:11 +08002893 gva_t gva, pfn_t *pfn, bool write, bool *writable);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09002894static void make_mmu_pages_available(struct kvm_vcpu *vcpu);
Xiao Guangrong060c2ab2010-11-12 14:49:11 +08002895
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002896static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, u32 error_code,
2897 gfn_t gfn, bool prefault)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002898{
2899 int r;
Joerg Roedel852e3c12009-07-27 16:30:44 +02002900 int level;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002901 int force_pt_level;
Anthony Liguori35149e22008-04-02 14:46:56 -05002902 pfn_t pfn;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02002903 unsigned long mmu_seq;
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002904 bool map_writable, write = error_code & PFERR_WRITE_MASK;
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002905
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002906 force_pt_level = mapping_level_dirty_bitmap(vcpu, gfn);
2907 if (likely(!force_pt_level)) {
2908 level = mapping_level(vcpu, gfn);
2909 /*
2910 * This path builds a PAE pagetable - so we can map
2911 * 2mb pages at maximum. Therefore check if the level
2912 * is larger than that.
2913 */
2914 if (level > PT_DIRECTORY_LEVEL)
2915 level = PT_DIRECTORY_LEVEL;
Joerg Roedel852e3c12009-07-27 16:30:44 +02002916
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002917 gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1);
2918 } else
2919 level = PT_PAGE_TABLE_LEVEL;
Marcelo Tosatti05da4552008-02-23 11:44:30 -03002920
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08002921 if (fast_page_fault(vcpu, v, level, error_code))
2922 return 0;
2923
Andrea Arcangelie930bff2008-07-25 16:24:52 +02002924 mmu_seq = vcpu->kvm->mmu_notifier_seq;
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03002925 smp_rmb();
Xiao Guangrong060c2ab2010-11-12 14:49:11 +08002926
Xiao Guangrong78b2c542010-12-07 10:48:06 +08002927 if (try_async_pf(vcpu, prefault, gfn, v, &pfn, write, &map_writable))
Xiao Guangrong060c2ab2010-11-12 14:49:11 +08002928 return 0;
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002929
Xiao Guangrongd7c55202011-07-12 03:29:38 +08002930 if (handle_abnormal_pfn(vcpu, v, gfn, pfn, ACC_ALL, &r))
2931 return r;
Avi Kivityd196e342008-01-24 11:44:11 +02002932
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002933 spin_lock(&vcpu->kvm->mmu_lock);
Christoffer Dall8ca40a72012-10-14 23:10:18 -04002934 if (mmu_notifier_retry(vcpu->kvm, mmu_seq))
Andrea Arcangelie930bff2008-07-25 16:24:52 +02002935 goto out_unlock;
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09002936 make_mmu_pages_available(vcpu);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08002937 if (likely(!force_pt_level))
2938 transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level);
Xiao Guangrong2ec47392010-12-07 10:34:42 +08002939 r = __direct_map(vcpu, v, write, map_writable, level, gfn, pfn,
2940 prefault);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002941 spin_unlock(&vcpu->kvm->mmu_lock);
2942
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002943
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002944 return r;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02002945
2946out_unlock:
2947 spin_unlock(&vcpu->kvm->mmu_lock);
2948 kvm_release_pfn_clean(pfn);
2949 return 0;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002950}
2951
2952
Avi Kivity17ac10a2007-01-05 16:36:40 -08002953static void mmu_free_roots(struct kvm_vcpu *vcpu)
2954{
2955 int i;
Avi Kivity4db35312007-11-21 15:28:32 +02002956 struct kvm_mmu_page *sp;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002957 LIST_HEAD(invalid_list);
Avi Kivity17ac10a2007-01-05 16:36:40 -08002958
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002959 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
Avi Kivity7b53aa52007-06-05 12:17:03 +03002960 return;
Gleb Natapov35af5772013-05-16 11:55:51 +03002961
Joerg Roedel81407ca2010-09-10 17:31:00 +02002962 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL &&
2963 (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL ||
2964 vcpu->arch.mmu.direct_map)) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002965 hpa_t root = vcpu->arch.mmu.root_hpa;
Avi Kivity17ac10a2007-01-05 16:36:40 -08002966
Gleb Natapov35af5772013-05-16 11:55:51 +03002967 spin_lock(&vcpu->kvm->mmu_lock);
Avi Kivity4db35312007-11-21 15:28:32 +02002968 sp = page_header(root);
2969 --sp->root_count;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002970 if (!sp->root_count && sp->role.invalid) {
2971 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list);
2972 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
2973 }
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002974 spin_unlock(&vcpu->kvm->mmu_lock);
Gleb Natapov35af5772013-05-16 11:55:51 +03002975 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
Avi Kivity17ac10a2007-01-05 16:36:40 -08002976 return;
2977 }
Gleb Natapov35af5772013-05-16 11:55:51 +03002978
2979 spin_lock(&vcpu->kvm->mmu_lock);
Avi Kivity17ac10a2007-01-05 16:36:40 -08002980 for (i = 0; i < 4; ++i) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002981 hpa_t root = vcpu->arch.mmu.pae_root[i];
Avi Kivity17ac10a2007-01-05 16:36:40 -08002982
Avi Kivity417726a2007-04-12 17:35:58 +03002983 if (root) {
Avi Kivity417726a2007-04-12 17:35:58 +03002984 root &= PT64_BASE_ADDR_MASK;
Avi Kivity4db35312007-11-21 15:28:32 +02002985 sp = page_header(root);
2986 --sp->root_count;
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05002987 if (!sp->root_count && sp->role.invalid)
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002988 kvm_mmu_prepare_zap_page(vcpu->kvm, sp,
2989 &invalid_list);
Avi Kivity417726a2007-04-12 17:35:58 +03002990 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002991 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
Avi Kivity17ac10a2007-01-05 16:36:40 -08002992 }
Xiao Guangrongd98ba052010-06-04 21:55:29 +08002993 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05002994 spin_unlock(&vcpu->kvm->mmu_lock);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002995 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
Avi Kivity17ac10a2007-01-05 16:36:40 -08002996}
2997
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03002998static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn)
2999{
3000 int ret = 0;
3001
3002 if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03003003 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003004 ret = 1;
3005 }
3006
3007 return ret;
3008}
3009
Joerg Roedel651dd372010-09-10 17:30:59 +02003010static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu)
3011{
3012 struct kvm_mmu_page *sp;
Avi Kivity7ebaf152010-10-03 18:51:39 +02003013 unsigned i;
Joerg Roedel651dd372010-09-10 17:30:59 +02003014
3015 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
3016 spin_lock(&vcpu->kvm->mmu_lock);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09003017 make_mmu_pages_available(vcpu);
Joerg Roedel651dd372010-09-10 17:30:59 +02003018 sp = kvm_mmu_get_page(vcpu, 0, 0, PT64_ROOT_LEVEL,
3019 1, ACC_ALL, NULL);
3020 ++sp->root_count;
3021 spin_unlock(&vcpu->kvm->mmu_lock);
3022 vcpu->arch.mmu.root_hpa = __pa(sp->spt);
3023 } else if (vcpu->arch.mmu.shadow_root_level == PT32E_ROOT_LEVEL) {
3024 for (i = 0; i < 4; ++i) {
3025 hpa_t root = vcpu->arch.mmu.pae_root[i];
3026
3027 ASSERT(!VALID_PAGE(root));
3028 spin_lock(&vcpu->kvm->mmu_lock);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09003029 make_mmu_pages_available(vcpu);
Avi Kivity649497d2010-12-28 12:09:07 +02003030 sp = kvm_mmu_get_page(vcpu, i << (30 - PAGE_SHIFT),
3031 i << 30,
Joerg Roedel651dd372010-09-10 17:30:59 +02003032 PT32_ROOT_LEVEL, 1, ACC_ALL,
3033 NULL);
3034 root = __pa(sp->spt);
3035 ++sp->root_count;
3036 spin_unlock(&vcpu->kvm->mmu_lock);
3037 vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK;
Joerg Roedel651dd372010-09-10 17:30:59 +02003038 }
Xiao Guangrong62927572010-09-27 18:02:12 +08003039 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
Joerg Roedel651dd372010-09-10 17:30:59 +02003040 } else
3041 BUG();
3042
3043 return 0;
3044}
3045
3046static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
Avi Kivity17ac10a2007-01-05 16:36:40 -08003047{
Avi Kivity4db35312007-11-21 15:28:32 +02003048 struct kvm_mmu_page *sp;
Joerg Roedel81407ca2010-09-10 17:31:00 +02003049 u64 pdptr, pm_mask;
3050 gfn_t root_gfn;
3051 int i;
Avi Kivity3bb65a22007-01-05 16:36:51 -08003052
Joerg Roedel5777ed32010-09-10 17:30:42 +02003053 root_gfn = vcpu->arch.mmu.get_cr3(vcpu) >> PAGE_SHIFT;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003054
Joerg Roedel651dd372010-09-10 17:30:59 +02003055 if (mmu_check_root(vcpu, root_gfn))
3056 return 1;
3057
3058 /*
3059 * Do we shadow a long mode page table? If so we need to
3060 * write-protect the guests page table root.
3061 */
3062 if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003063 hpa_t root = vcpu->arch.mmu.root_hpa;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003064
3065 ASSERT(!VALID_PAGE(root));
Joerg Roedel651dd372010-09-10 17:30:59 +02003066
Avi Kivity8facbbf2010-05-04 12:58:32 +03003067 spin_lock(&vcpu->kvm->mmu_lock);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09003068 make_mmu_pages_available(vcpu);
Joerg Roedel651dd372010-09-10 17:30:59 +02003069 sp = kvm_mmu_get_page(vcpu, root_gfn, 0, PT64_ROOT_LEVEL,
3070 0, ACC_ALL, NULL);
Avi Kivity4db35312007-11-21 15:28:32 +02003071 root = __pa(sp->spt);
3072 ++sp->root_count;
Avi Kivity8facbbf2010-05-04 12:58:32 +03003073 spin_unlock(&vcpu->kvm->mmu_lock);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003074 vcpu->arch.mmu.root_hpa = root;
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003075 return 0;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003076 }
Joerg Roedelf87f9282010-09-02 17:29:45 +02003077
Joerg Roedel651dd372010-09-10 17:30:59 +02003078 /*
3079 * We shadow a 32 bit page table. This may be a legacy 2-level
Joerg Roedel81407ca2010-09-10 17:31:00 +02003080 * or a PAE 3-level page table. In either case we need to be aware that
3081 * the shadow page table may be a PAE or a long mode page table.
Joerg Roedel651dd372010-09-10 17:30:59 +02003082 */
Joerg Roedel81407ca2010-09-10 17:31:00 +02003083 pm_mask = PT_PRESENT_MASK;
3084 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL)
3085 pm_mask |= PT_ACCESSED_MASK | PT_WRITABLE_MASK | PT_USER_MASK;
3086
Avi Kivity17ac10a2007-01-05 16:36:40 -08003087 for (i = 0; i < 4; ++i) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003088 hpa_t root = vcpu->arch.mmu.pae_root[i];
Avi Kivity17ac10a2007-01-05 16:36:40 -08003089
3090 ASSERT(!VALID_PAGE(root));
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003091 if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) {
Avi Kivitye4e517b2011-07-28 11:36:17 +03003092 pdptr = vcpu->arch.mmu.get_pdptr(vcpu, i);
Avi Kivity43a37952009-06-10 14:12:05 +03003093 if (!is_present_gpte(pdptr)) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003094 vcpu->arch.mmu.pae_root[i] = 0;
Avi Kivity417726a2007-04-12 17:35:58 +03003095 continue;
3096 }
Avi Kivity6de4f3a2009-05-31 22:58:47 +03003097 root_gfn = pdptr >> PAGE_SHIFT;
Joerg Roedelf87f9282010-09-02 17:29:45 +02003098 if (mmu_check_root(vcpu, root_gfn))
3099 return 1;
Eric Northup5a7388c2010-04-26 17:00:05 -07003100 }
Avi Kivity8facbbf2010-05-04 12:58:32 +03003101 spin_lock(&vcpu->kvm->mmu_lock);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09003102 make_mmu_pages_available(vcpu);
Avi Kivity4db35312007-11-21 15:28:32 +02003103 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
Joerg Roedel651dd372010-09-10 17:30:59 +02003104 PT32_ROOT_LEVEL, 0,
Avi Kivityf7d9c7b2008-02-26 22:12:10 +02003105 ACC_ALL, NULL);
Avi Kivity4db35312007-11-21 15:28:32 +02003106 root = __pa(sp->spt);
3107 ++sp->root_count;
Avi Kivity8facbbf2010-05-04 12:58:32 +03003108 spin_unlock(&vcpu->kvm->mmu_lock);
3109
Joerg Roedel81407ca2010-09-10 17:31:00 +02003110 vcpu->arch.mmu.pae_root[i] = root | pm_mask;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003111 }
Xiao Guangrong62927572010-09-27 18:02:12 +08003112 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
Joerg Roedel81407ca2010-09-10 17:31:00 +02003113
3114 /*
3115 * If we shadow a 32 bit page table with a long mode page
3116 * table we enter this path.
3117 */
3118 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
3119 if (vcpu->arch.mmu.lm_root == NULL) {
3120 /*
3121 * The additional page necessary for this is only
3122 * allocated on demand.
3123 */
3124
3125 u64 *lm_root;
3126
3127 lm_root = (void*)get_zeroed_page(GFP_KERNEL);
3128 if (lm_root == NULL)
3129 return 1;
3130
3131 lm_root[0] = __pa(vcpu->arch.mmu.pae_root) | pm_mask;
3132
3133 vcpu->arch.mmu.lm_root = lm_root;
3134 }
3135
3136 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.lm_root);
3137 }
3138
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003139 return 0;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003140}
3141
Joerg Roedel651dd372010-09-10 17:30:59 +02003142static int mmu_alloc_roots(struct kvm_vcpu *vcpu)
3143{
3144 if (vcpu->arch.mmu.direct_map)
3145 return mmu_alloc_direct_roots(vcpu);
3146 else
3147 return mmu_alloc_shadow_roots(vcpu);
3148}
3149
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003150static void mmu_sync_roots(struct kvm_vcpu *vcpu)
3151{
3152 int i;
3153 struct kvm_mmu_page *sp;
3154
Joerg Roedel81407ca2010-09-10 17:31:00 +02003155 if (vcpu->arch.mmu.direct_map)
3156 return;
3157
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003158 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3159 return;
Xiao Guangrong69030742010-09-27 18:09:29 +08003160
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003161 vcpu_clear_mmio_info(vcpu, ~0ul);
Xiao Guangrong0375f7f2011-11-28 20:41:00 +08003162 kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC);
Joerg Roedel81407ca2010-09-10 17:31:00 +02003163 if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) {
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003164 hpa_t root = vcpu->arch.mmu.root_hpa;
3165 sp = page_header(root);
3166 mmu_sync_children(vcpu, sp);
Xiao Guangrong0375f7f2011-11-28 20:41:00 +08003167 kvm_mmu_audit(vcpu, AUDIT_POST_SYNC);
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003168 return;
3169 }
3170 for (i = 0; i < 4; ++i) {
3171 hpa_t root = vcpu->arch.mmu.pae_root[i];
3172
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003173 if (root && VALID_PAGE(root)) {
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003174 root &= PT64_BASE_ADDR_MASK;
3175 sp = page_header(root);
3176 mmu_sync_children(vcpu, sp);
3177 }
3178 }
Xiao Guangrong0375f7f2011-11-28 20:41:00 +08003179 kvm_mmu_audit(vcpu, AUDIT_POST_SYNC);
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003180}
3181
3182void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
3183{
3184 spin_lock(&vcpu->kvm->mmu_lock);
3185 mmu_sync_roots(vcpu);
3186 spin_unlock(&vcpu->kvm->mmu_lock);
3187}
Nadav Har'Elbfd0a562013-08-05 11:07:17 +03003188EXPORT_SYMBOL_GPL(kvm_mmu_sync_roots);
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -03003189
Gleb Natapov1871c602010-02-10 14:21:32 +02003190static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr,
Avi Kivityab9ae312010-11-22 17:53:26 +02003191 u32 access, struct x86_exception *exception)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003192{
Avi Kivityab9ae312010-11-22 17:53:26 +02003193 if (exception)
3194 exception->error_code = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003195 return vaddr;
3196}
3197
Joerg Roedel6539e732010-09-10 17:30:50 +02003198static gpa_t nonpaging_gva_to_gpa_nested(struct kvm_vcpu *vcpu, gva_t vaddr,
Avi Kivityab9ae312010-11-22 17:53:26 +02003199 u32 access,
3200 struct x86_exception *exception)
Joerg Roedel6539e732010-09-10 17:30:50 +02003201{
Avi Kivityab9ae312010-11-22 17:53:26 +02003202 if (exception)
3203 exception->error_code = 0;
Joerg Roedel6539e732010-09-10 17:30:50 +02003204 return vcpu->arch.nested_mmu.translate_gpa(vcpu, vaddr, access);
3205}
3206
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003207static bool quickly_check_mmio_pf(struct kvm_vcpu *vcpu, u64 addr, bool direct)
3208{
3209 if (direct)
3210 return vcpu_match_mmio_gpa(vcpu, addr);
3211
3212 return vcpu_match_mmio_gva(vcpu, addr);
3213}
3214
3215
3216/*
3217 * On direct hosts, the last spte is only allows two states
3218 * for mmio page fault:
3219 * - It is the mmio spte
3220 * - It is zapped or it is being zapped.
3221 *
3222 * This function completely checks the spte when the last spte
3223 * is not the mmio spte.
3224 */
3225static bool check_direct_spte_mmio_pf(u64 spte)
3226{
3227 return __check_direct_spte_mmio_pf(spte);
3228}
3229
3230static u64 walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr)
3231{
3232 struct kvm_shadow_walk_iterator iterator;
3233 u64 spte = 0ull;
3234
Marcelo Tosatti37f6a4e2014-01-03 17:09:32 -02003235 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3236 return spte;
3237
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003238 walk_shadow_page_lockless_begin(vcpu);
3239 for_each_shadow_entry_lockless(vcpu, addr, iterator, spte)
3240 if (!is_shadow_present_pte(spte))
3241 break;
3242 walk_shadow_page_lockless_end(vcpu);
3243
3244 return spte;
3245}
3246
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003247int handle_mmio_page_fault_common(struct kvm_vcpu *vcpu, u64 addr, bool direct)
3248{
3249 u64 spte;
3250
3251 if (quickly_check_mmio_pf(vcpu, addr, direct))
Xiao Guangrongb37fbea2013-06-07 16:51:25 +08003252 return RET_MMIO_PF_EMULATE;
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003253
3254 spte = walk_shadow_page_get_mmio_spte(vcpu, addr);
3255
3256 if (is_mmio_spte(spte)) {
3257 gfn_t gfn = get_mmio_spte_gfn(spte);
3258 unsigned access = get_mmio_spte_access(spte);
3259
Xiao Guangrongf8f55942013-06-07 16:51:26 +08003260 if (!check_mmio_spte(vcpu->kvm, spte))
3261 return RET_MMIO_PF_INVALID;
3262
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003263 if (direct)
3264 addr = 0;
Xiao Guangrong4f022642011-07-12 03:34:24 +08003265
3266 trace_handle_mmio_page_fault(addr, gfn, access);
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003267 vcpu_cache_mmio_info(vcpu, addr, gfn, access);
Xiao Guangrongb37fbea2013-06-07 16:51:25 +08003268 return RET_MMIO_PF_EMULATE;
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003269 }
3270
3271 /*
3272 * It's ok if the gva is remapped by other cpus on shadow guest,
3273 * it's a BUG if the gfn is not a mmio page.
3274 */
3275 if (direct && !check_direct_spte_mmio_pf(spte))
Xiao Guangrongb37fbea2013-06-07 16:51:25 +08003276 return RET_MMIO_PF_BUG;
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003277
3278 /*
3279 * If the page table is zapped by other cpus, let CPU fault again on
3280 * the address.
3281 */
Xiao Guangrongb37fbea2013-06-07 16:51:25 +08003282 return RET_MMIO_PF_RETRY;
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003283}
3284EXPORT_SYMBOL_GPL(handle_mmio_page_fault_common);
3285
3286static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr,
3287 u32 error_code, bool direct)
3288{
3289 int ret;
3290
3291 ret = handle_mmio_page_fault_common(vcpu, addr, direct);
Xiao Guangrongb37fbea2013-06-07 16:51:25 +08003292 WARN_ON(ret == RET_MMIO_PF_BUG);
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003293 return ret;
3294}
3295
Avi Kivity6aa8b732006-12-10 02:21:36 -08003296static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
Xiao Guangrong78b2c542010-12-07 10:48:06 +08003297 u32 error_code, bool prefault)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003298{
Avi Kivitye8332402007-12-09 18:43:00 +02003299 gfn_t gfn;
Avi Kivitye2dec932007-01-05 16:36:54 -08003300 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003301
Harvey Harrisonb8688d52008-03-03 12:59:56 -08003302 pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code);
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003303
Xiao Guangrongf8f55942013-06-07 16:51:26 +08003304 if (unlikely(error_code & PFERR_RSVD_MASK)) {
3305 r = handle_mmio_page_fault(vcpu, gva, error_code, true);
3306
3307 if (likely(r != RET_MMIO_PF_INVALID))
3308 return r;
3309 }
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003310
Avi Kivitye2dec932007-01-05 16:36:54 -08003311 r = mmu_topup_memory_caches(vcpu);
3312 if (r)
3313 return r;
Avi Kivity714b93d2007-01-05 16:36:53 -08003314
Avi Kivity6aa8b732006-12-10 02:21:36 -08003315 ASSERT(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003316 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
Avi Kivity6aa8b732006-12-10 02:21:36 -08003317
Avi Kivitye8332402007-12-09 18:43:00 +02003318 gfn = gva >> PAGE_SHIFT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003319
Avi Kivitye8332402007-12-09 18:43:00 +02003320 return nonpaging_map(vcpu, gva & PAGE_MASK,
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08003321 error_code, gfn, prefault);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003322}
3323
Jan Kiszka7e1fbea2010-10-20 15:18:02 +02003324static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn)
Gleb Natapovaf585b92010-10-14 11:22:46 +02003325{
3326 struct kvm_arch_async_pf arch;
Xiao Guangrongfb67e142010-12-07 10:35:25 +08003327
Gleb Natapov7c907052010-10-14 11:22:53 +02003328 arch.token = (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id;
Gleb Natapovaf585b92010-10-14 11:22:46 +02003329 arch.gfn = gfn;
Xiao Guangrongc4806ac2010-11-12 14:49:55 +08003330 arch.direct_map = vcpu->arch.mmu.direct_map;
Xiao Guangrongfb67e142010-12-07 10:35:25 +08003331 arch.cr3 = vcpu->arch.mmu.get_cr3(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003332
Dominik Dingele0ead412013-06-06 15:32:37 +02003333 return kvm_setup_async_pf(vcpu, gva, gfn_to_hva(vcpu->kvm, gfn), &arch);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003334}
3335
3336static bool can_do_async_pf(struct kvm_vcpu *vcpu)
3337{
3338 if (unlikely(!irqchip_in_kernel(vcpu->kvm) ||
3339 kvm_event_needs_reinjection(vcpu)))
3340 return false;
3341
3342 return kvm_x86_ops->interrupt_allowed(vcpu);
3343}
3344
Xiao Guangrong78b2c542010-12-07 10:48:06 +08003345static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn,
Marcelo Tosatti612819c2010-10-22 14:18:18 -02003346 gva_t gva, pfn_t *pfn, bool write, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02003347{
3348 bool async;
3349
Marcelo Tosatti612819c2010-10-22 14:18:18 -02003350 *pfn = gfn_to_pfn_async(vcpu->kvm, gfn, &async, write, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003351
3352 if (!async)
3353 return false; /* *pfn has correct page already */
3354
Xiao Guangrong78b2c542010-12-07 10:48:06 +08003355 if (!prefault && can_do_async_pf(vcpu)) {
Xiao Guangrongc9b263d2010-11-01 16:58:43 +08003356 trace_kvm_try_async_get_page(gva, gfn);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003357 if (kvm_find_async_pf_gfn(vcpu, gfn)) {
3358 trace_kvm_async_pf_doublefault(gva, gfn);
3359 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
3360 return true;
3361 } else if (kvm_arch_setup_async_pf(vcpu, gva, gfn))
3362 return true;
3363 }
3364
Marcelo Tosatti612819c2010-10-22 14:18:18 -02003365 *pfn = gfn_to_pfn_prot(vcpu->kvm, gfn, write, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003366
3367 return false;
3368}
3369
Gleb Natapov56028d02010-10-17 18:13:42 +02003370static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code,
Xiao Guangrong78b2c542010-12-07 10:48:06 +08003371 bool prefault)
Joerg Roedelfb72d162008-02-07 13:47:44 +01003372{
Anthony Liguori35149e22008-04-02 14:46:56 -05003373 pfn_t pfn;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003374 int r;
Joerg Roedel852e3c12009-07-27 16:30:44 +02003375 int level;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08003376 int force_pt_level;
Marcelo Tosatti05da4552008-02-23 11:44:30 -03003377 gfn_t gfn = gpa >> PAGE_SHIFT;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02003378 unsigned long mmu_seq;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02003379 int write = error_code & PFERR_WRITE_MASK;
3380 bool map_writable;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003381
3382 ASSERT(vcpu);
3383 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
3384
Xiao Guangrongf8f55942013-06-07 16:51:26 +08003385 if (unlikely(error_code & PFERR_RSVD_MASK)) {
3386 r = handle_mmio_page_fault(vcpu, gpa, error_code, true);
3387
3388 if (likely(r != RET_MMIO_PF_INVALID))
3389 return r;
3390 }
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003391
Joerg Roedelfb72d162008-02-07 13:47:44 +01003392 r = mmu_topup_memory_caches(vcpu);
3393 if (r)
3394 return r;
3395
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08003396 force_pt_level = mapping_level_dirty_bitmap(vcpu, gfn);
3397 if (likely(!force_pt_level)) {
3398 level = mapping_level(vcpu, gfn);
3399 gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1);
3400 } else
3401 level = PT_PAGE_TABLE_LEVEL;
Joerg Roedel852e3c12009-07-27 16:30:44 +02003402
Xiao Guangrongc7ba5b42012-06-20 15:59:18 +08003403 if (fast_page_fault(vcpu, gpa, level, error_code))
3404 return 0;
3405
Andrea Arcangelie930bff2008-07-25 16:24:52 +02003406 mmu_seq = vcpu->kvm->mmu_notifier_seq;
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03003407 smp_rmb();
Gleb Natapovaf585b92010-10-14 11:22:46 +02003408
Xiao Guangrong78b2c542010-12-07 10:48:06 +08003409 if (try_async_pf(vcpu, prefault, gfn, gpa, &pfn, write, &map_writable))
Gleb Natapovaf585b92010-10-14 11:22:46 +02003410 return 0;
3411
Xiao Guangrongd7c55202011-07-12 03:29:38 +08003412 if (handle_abnormal_pfn(vcpu, 0, gfn, pfn, ACC_ALL, &r))
3413 return r;
3414
Joerg Roedelfb72d162008-02-07 13:47:44 +01003415 spin_lock(&vcpu->kvm->mmu_lock);
Christoffer Dall8ca40a72012-10-14 23:10:18 -04003416 if (mmu_notifier_retry(vcpu->kvm, mmu_seq))
Andrea Arcangelie930bff2008-07-25 16:24:52 +02003417 goto out_unlock;
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +09003418 make_mmu_pages_available(vcpu);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -08003419 if (likely(!force_pt_level))
3420 transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level);
Marcelo Tosatti612819c2010-10-22 14:18:18 -02003421 r = __direct_map(vcpu, gpa, write, map_writable,
Xiao Guangrong2ec47392010-12-07 10:34:42 +08003422 level, gfn, pfn, prefault);
Joerg Roedelfb72d162008-02-07 13:47:44 +01003423 spin_unlock(&vcpu->kvm->mmu_lock);
Joerg Roedelfb72d162008-02-07 13:47:44 +01003424
3425 return r;
Andrea Arcangelie930bff2008-07-25 16:24:52 +02003426
3427out_unlock:
3428 spin_unlock(&vcpu->kvm->mmu_lock);
3429 kvm_release_pfn_clean(pfn);
3430 return 0;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003431}
3432
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003433static void nonpaging_init_context(struct kvm_vcpu *vcpu,
3434 struct kvm_mmu *context)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003435{
Avi Kivity6aa8b732006-12-10 02:21:36 -08003436 context->page_fault = nonpaging_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003437 context->gva_to_gpa = nonpaging_gva_to_gpa;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03003438 context->sync_page = nonpaging_sync_page;
Marcelo Tosattia7052892008-09-23 13:18:35 -03003439 context->invlpg = nonpaging_invlpg;
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08003440 context->update_pte = nonpaging_update_pte;
Avi Kivitycea0f0e2007-01-05 16:36:43 -08003441 context->root_level = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003442 context->shadow_root_level = PT32E_ROOT_LEVEL;
Avi Kivity17c3ba92007-06-04 15:58:30 +03003443 context->root_hpa = INVALID_PAGE;
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02003444 context->direct_map = true;
Joerg Roedel2d48a982010-09-10 17:31:01 +02003445 context->nx = false;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003446}
3447
Avi Kivityd835dfe2007-11-21 02:57:59 +02003448void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003449{
Avi Kivity1165f5f2007-04-19 17:27:43 +03003450 ++vcpu->stat.tlb_flush;
Avi Kivitya8eeb042010-05-10 12:34:53 +03003451 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003452}
Nadav Har'Elbfd0a562013-08-05 11:07:17 +03003453EXPORT_SYMBOL_GPL(kvm_mmu_flush_tlb);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003454
Paolo Bonzinid8d173d2013-10-02 16:56:11 +02003455void kvm_mmu_new_cr3(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003456{
Avi Kivitycea0f0e2007-01-05 16:36:43 -08003457 mmu_free_roots(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003458}
3459
Joerg Roedel5777ed32010-09-10 17:30:42 +02003460static unsigned long get_cr3(struct kvm_vcpu *vcpu)
3461{
Avi Kivity9f8fe502010-12-05 17:30:00 +02003462 return kvm_read_cr3(vcpu);
Joerg Roedel5777ed32010-09-10 17:30:42 +02003463}
3464
Avi Kivity6389ee92010-11-29 16:12:30 +02003465static void inject_page_fault(struct kvm_vcpu *vcpu,
3466 struct x86_exception *fault)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003467{
Avi Kivity6389ee92010-11-29 16:12:30 +02003468 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003469}
3470
Xiao Guangrongf2fd1252013-06-07 16:51:24 +08003471static bool sync_mmio_spte(struct kvm *kvm, u64 *sptep, gfn_t gfn,
3472 unsigned access, int *nr_present)
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003473{
3474 if (unlikely(is_mmio_spte(*sptep))) {
3475 if (gfn != get_mmio_spte_gfn(*sptep)) {
3476 mmu_spte_clear_no_track(sptep);
3477 return true;
3478 }
3479
3480 (*nr_present)++;
Xiao Guangrongf2fd1252013-06-07 16:51:24 +08003481 mark_mmio_spte(kvm, sptep, gfn, access);
Xiao Guangrongce88dec2011-07-12 03:33:44 +08003482 return true;
3483 }
3484
3485 return false;
3486}
3487
Avi Kivity6fd01b72012-09-12 20:46:56 +03003488static inline bool is_last_gpte(struct kvm_mmu *mmu, unsigned level, unsigned gpte)
3489{
3490 unsigned index;
3491
3492 index = level - 1;
3493 index |= (gpte & PT_PAGE_SIZE_MASK) >> (PT_PAGE_SIZE_SHIFT - 2);
3494 return mmu->last_pte_bitmap & (1 << index);
3495}
3496
Nadav Har'El37406aa2013-08-05 11:07:12 +03003497#define PTTYPE_EPT 18 /* arbitrary */
3498#define PTTYPE PTTYPE_EPT
3499#include "paging_tmpl.h"
3500#undef PTTYPE
3501
Avi Kivity6aa8b732006-12-10 02:21:36 -08003502#define PTTYPE 64
3503#include "paging_tmpl.h"
3504#undef PTTYPE
3505
3506#define PTTYPE 32
3507#include "paging_tmpl.h"
3508#undef PTTYPE
3509
Joerg Roedel52fde8d2010-09-10 17:30:44 +02003510static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu,
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003511 struct kvm_mmu *context)
Dong, Eddie82725b22009-03-30 16:21:08 +08003512{
Dong, Eddie82725b22009-03-30 16:21:08 +08003513 int maxphyaddr = cpuid_maxphyaddr(vcpu);
3514 u64 exb_bit_rsvd = 0;
Nadav Amit5f7dde72014-05-07 15:32:50 +03003515 u64 gbpages_bit_rsvd = 0;
Dong, Eddie82725b22009-03-30 16:21:08 +08003516
Yang Zhang25d92082013-08-06 12:00:32 +03003517 context->bad_mt_xwr = 0;
3518
Joerg Roedel2d48a982010-09-10 17:31:01 +02003519 if (!context->nx)
Dong, Eddie82725b22009-03-30 16:21:08 +08003520 exb_bit_rsvd = rsvd_bits(63, 63);
Nadav Amit5f7dde72014-05-07 15:32:50 +03003521 if (!guest_cpuid_has_gbpages(vcpu))
3522 gbpages_bit_rsvd = rsvd_bits(7, 7);
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003523 switch (context->root_level) {
Dong, Eddie82725b22009-03-30 16:21:08 +08003524 case PT32_ROOT_LEVEL:
3525 /* no rsvd bits for 2 level 4K page table entries */
3526 context->rsvd_bits_mask[0][1] = 0;
3527 context->rsvd_bits_mask[0][0] = 0;
Xiao Guangrongf815bce2010-03-19 17:58:53 +08003528 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0];
3529
3530 if (!is_pse(vcpu)) {
3531 context->rsvd_bits_mask[1][1] = 0;
3532 break;
3533 }
3534
Dong, Eddie82725b22009-03-30 16:21:08 +08003535 if (is_cpuid_PSE36())
3536 /* 36bits PSE 4MB page */
3537 context->rsvd_bits_mask[1][1] = rsvd_bits(17, 21);
3538 else
3539 /* 32 bits PSE 4MB page */
3540 context->rsvd_bits_mask[1][1] = rsvd_bits(13, 21);
Dong, Eddie82725b22009-03-30 16:21:08 +08003541 break;
3542 case PT32E_ROOT_LEVEL:
Dong, Eddie20c466b2009-03-31 23:03:45 +08003543 context->rsvd_bits_mask[0][2] =
3544 rsvd_bits(maxphyaddr, 63) |
Nadav Amitcd9ae5f2014-04-04 06:31:04 +03003545 rsvd_bits(5, 8) | rsvd_bits(1, 2); /* PDPTE */
Dong, Eddie82725b22009-03-30 16:21:08 +08003546 context->rsvd_bits_mask[0][1] = exb_bit_rsvd |
Sheng Yang4c26b4c2009-04-02 10:28:37 +08003547 rsvd_bits(maxphyaddr, 62); /* PDE */
Dong, Eddie82725b22009-03-30 16:21:08 +08003548 context->rsvd_bits_mask[0][0] = exb_bit_rsvd |
3549 rsvd_bits(maxphyaddr, 62); /* PTE */
3550 context->rsvd_bits_mask[1][1] = exb_bit_rsvd |
3551 rsvd_bits(maxphyaddr, 62) |
3552 rsvd_bits(13, 20); /* large page */
Xiao Guangrongf815bce2010-03-19 17:58:53 +08003553 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0];
Dong, Eddie82725b22009-03-30 16:21:08 +08003554 break;
3555 case PT64_ROOT_LEVEL:
3556 context->rsvd_bits_mask[0][3] = exb_bit_rsvd |
Nadav Amitcd9ae5f2014-04-04 06:31:04 +03003557 rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 7);
Dong, Eddie82725b22009-03-30 16:21:08 +08003558 context->rsvd_bits_mask[0][2] = exb_bit_rsvd |
Nadav Amit5f7dde72014-05-07 15:32:50 +03003559 gbpages_bit_rsvd | rsvd_bits(maxphyaddr, 51);
Dong, Eddie82725b22009-03-30 16:21:08 +08003560 context->rsvd_bits_mask[0][1] = exb_bit_rsvd |
Sheng Yang4c26b4c2009-04-02 10:28:37 +08003561 rsvd_bits(maxphyaddr, 51);
Dong, Eddie82725b22009-03-30 16:21:08 +08003562 context->rsvd_bits_mask[0][0] = exb_bit_rsvd |
3563 rsvd_bits(maxphyaddr, 51);
3564 context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3];
Joerg Roedele04da982009-07-27 16:30:45 +02003565 context->rsvd_bits_mask[1][2] = exb_bit_rsvd |
Nadav Amit5f7dde72014-05-07 15:32:50 +03003566 gbpages_bit_rsvd | rsvd_bits(maxphyaddr, 51) |
Joerg Roedele04da982009-07-27 16:30:45 +02003567 rsvd_bits(13, 29);
Dong, Eddie82725b22009-03-30 16:21:08 +08003568 context->rsvd_bits_mask[1][1] = exb_bit_rsvd |
Sheng Yang4c26b4c2009-04-02 10:28:37 +08003569 rsvd_bits(maxphyaddr, 51) |
3570 rsvd_bits(13, 20); /* large page */
Xiao Guangrongf815bce2010-03-19 17:58:53 +08003571 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0];
Dong, Eddie82725b22009-03-30 16:21:08 +08003572 break;
3573 }
3574}
3575
Yang Zhang25d92082013-08-06 12:00:32 +03003576static void reset_rsvds_bits_mask_ept(struct kvm_vcpu *vcpu,
3577 struct kvm_mmu *context, bool execonly)
3578{
3579 int maxphyaddr = cpuid_maxphyaddr(vcpu);
3580 int pte;
3581
3582 context->rsvd_bits_mask[0][3] =
3583 rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 7);
3584 context->rsvd_bits_mask[0][2] =
3585 rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6);
3586 context->rsvd_bits_mask[0][1] =
3587 rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6);
3588 context->rsvd_bits_mask[0][0] = rsvd_bits(maxphyaddr, 51);
3589
3590 /* large page */
3591 context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3];
3592 context->rsvd_bits_mask[1][2] =
3593 rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 29);
3594 context->rsvd_bits_mask[1][1] =
3595 rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 20);
3596 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0];
3597
3598 for (pte = 0; pte < 64; pte++) {
3599 int rwx_bits = pte & 7;
3600 int mt = pte >> 3;
3601 if (mt == 0x2 || mt == 0x3 || mt == 0x7 ||
3602 rwx_bits == 0x2 || rwx_bits == 0x6 ||
3603 (rwx_bits == 0x4 && !execonly))
3604 context->bad_mt_xwr |= (1ull << pte);
3605 }
3606}
3607
Feng Wu97ec8c02014-04-01 17:46:34 +08003608void update_permission_bitmask(struct kvm_vcpu *vcpu,
Yang Zhang25d92082013-08-06 12:00:32 +03003609 struct kvm_mmu *mmu, bool ept)
Avi Kivity97d64b72012-09-12 14:52:00 +03003610{
3611 unsigned bit, byte, pfec;
3612 u8 map;
Feng Wu66386ad2014-04-01 17:56:48 +08003613 bool fault, x, w, u, wf, uf, ff, smapf, cr4_smap, cr4_smep, smap = 0;
Avi Kivity97d64b72012-09-12 14:52:00 +03003614
Feng Wu66386ad2014-04-01 17:56:48 +08003615 cr4_smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP);
Feng Wu97ec8c02014-04-01 17:46:34 +08003616 cr4_smap = kvm_read_cr4_bits(vcpu, X86_CR4_SMAP);
Avi Kivity97d64b72012-09-12 14:52:00 +03003617 for (byte = 0; byte < ARRAY_SIZE(mmu->permissions); ++byte) {
3618 pfec = byte << 1;
3619 map = 0;
3620 wf = pfec & PFERR_WRITE_MASK;
3621 uf = pfec & PFERR_USER_MASK;
3622 ff = pfec & PFERR_FETCH_MASK;
Feng Wu97ec8c02014-04-01 17:46:34 +08003623 /*
3624 * PFERR_RSVD_MASK bit is set in PFEC if the access is not
3625 * subject to SMAP restrictions, and cleared otherwise. The
3626 * bit is only meaningful if the SMAP bit is set in CR4.
3627 */
3628 smapf = !(pfec & PFERR_RSVD_MASK);
Avi Kivity97d64b72012-09-12 14:52:00 +03003629 for (bit = 0; bit < 8; ++bit) {
3630 x = bit & ACC_EXEC_MASK;
3631 w = bit & ACC_WRITE_MASK;
3632 u = bit & ACC_USER_MASK;
3633
Yang Zhang25d92082013-08-06 12:00:32 +03003634 if (!ept) {
3635 /* Not really needed: !nx will cause pte.nx to fault */
3636 x |= !mmu->nx;
3637 /* Allow supervisor writes if !cr0.wp */
3638 w |= !is_write_protection(vcpu) && !uf;
3639 /* Disallow supervisor fetches of user code if cr4.smep */
Feng Wu66386ad2014-04-01 17:56:48 +08003640 x &= !(cr4_smep && u && !uf);
Feng Wu97ec8c02014-04-01 17:46:34 +08003641
3642 /*
3643 * SMAP:kernel-mode data accesses from user-mode
3644 * mappings should fault. A fault is considered
3645 * as a SMAP violation if all of the following
3646 * conditions are ture:
3647 * - X86_CR4_SMAP is set in CR4
3648 * - An user page is accessed
3649 * - Page fault in kernel mode
3650 * - if CPL = 3 or X86_EFLAGS_AC is clear
3651 *
3652 * Here, we cover the first three conditions.
3653 * The fourth is computed dynamically in
3654 * permission_fault() and is in smapf.
3655 *
3656 * Also, SMAP does not affect instruction
3657 * fetches, add the !ff check here to make it
3658 * clearer.
3659 */
3660 smap = cr4_smap && u && !uf && !ff;
Yang Zhang25d92082013-08-06 12:00:32 +03003661 } else
3662 /* Not really needed: no U/S accesses on ept */
3663 u = 1;
Avi Kivity97d64b72012-09-12 14:52:00 +03003664
Feng Wu97ec8c02014-04-01 17:46:34 +08003665 fault = (ff && !x) || (uf && !u) || (wf && !w) ||
3666 (smapf && smap);
Avi Kivity97d64b72012-09-12 14:52:00 +03003667 map |= fault << bit;
3668 }
3669 mmu->permissions[byte] = map;
3670 }
3671}
3672
Avi Kivity6fd01b72012-09-12 20:46:56 +03003673static void update_last_pte_bitmap(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu)
3674{
3675 u8 map;
3676 unsigned level, root_level = mmu->root_level;
3677 const unsigned ps_set_index = 1 << 2; /* bit 2 of index: ps */
3678
3679 if (root_level == PT32E_ROOT_LEVEL)
3680 --root_level;
3681 /* PT_PAGE_TABLE_LEVEL always terminates */
3682 map = 1 | (1 << ps_set_index);
3683 for (level = PT_DIRECTORY_LEVEL; level <= root_level; ++level) {
3684 if (level <= PT_PDPE_LEVEL
3685 && (mmu->root_level >= PT32E_ROOT_LEVEL || is_pse(vcpu)))
3686 map |= 1 << (ps_set_index | (level - 1));
3687 }
3688 mmu->last_pte_bitmap = map;
3689}
3690
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003691static void paging64_init_context_common(struct kvm_vcpu *vcpu,
3692 struct kvm_mmu *context,
3693 int level)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003694{
Joerg Roedel2d48a982010-09-10 17:31:01 +02003695 context->nx = is_nx(vcpu);
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003696 context->root_level = level;
Joerg Roedel2d48a982010-09-10 17:31:01 +02003697
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003698 reset_rsvds_bits_mask(vcpu, context);
Yang Zhang25d92082013-08-06 12:00:32 +03003699 update_permission_bitmask(vcpu, context, false);
Avi Kivity6fd01b72012-09-12 20:46:56 +03003700 update_last_pte_bitmap(vcpu, context);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003701
3702 ASSERT(is_pae(vcpu));
Avi Kivity6aa8b732006-12-10 02:21:36 -08003703 context->page_fault = paging64_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003704 context->gva_to_gpa = paging64_gva_to_gpa;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03003705 context->sync_page = paging64_sync_page;
Marcelo Tosattia7052892008-09-23 13:18:35 -03003706 context->invlpg = paging64_invlpg;
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08003707 context->update_pte = paging64_update_pte;
Avi Kivity17ac10a2007-01-05 16:36:40 -08003708 context->shadow_root_level = level;
Avi Kivity17c3ba92007-06-04 15:58:30 +03003709 context->root_hpa = INVALID_PAGE;
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02003710 context->direct_map = false;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003711}
3712
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003713static void paging64_init_context(struct kvm_vcpu *vcpu,
3714 struct kvm_mmu *context)
Avi Kivity17ac10a2007-01-05 16:36:40 -08003715{
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003716 paging64_init_context_common(vcpu, context, PT64_ROOT_LEVEL);
Avi Kivity17ac10a2007-01-05 16:36:40 -08003717}
3718
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003719static void paging32_init_context(struct kvm_vcpu *vcpu,
3720 struct kvm_mmu *context)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003721{
Joerg Roedel2d48a982010-09-10 17:31:01 +02003722 context->nx = false;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003723 context->root_level = PT32_ROOT_LEVEL;
Joerg Roedel2d48a982010-09-10 17:31:01 +02003724
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003725 reset_rsvds_bits_mask(vcpu, context);
Yang Zhang25d92082013-08-06 12:00:32 +03003726 update_permission_bitmask(vcpu, context, false);
Avi Kivity6fd01b72012-09-12 20:46:56 +03003727 update_last_pte_bitmap(vcpu, context);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003728
Avi Kivity6aa8b732006-12-10 02:21:36 -08003729 context->page_fault = paging32_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003730 context->gva_to_gpa = paging32_gva_to_gpa;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03003731 context->sync_page = paging32_sync_page;
Marcelo Tosattia7052892008-09-23 13:18:35 -03003732 context->invlpg = paging32_invlpg;
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08003733 context->update_pte = paging32_update_pte;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003734 context->shadow_root_level = PT32E_ROOT_LEVEL;
Avi Kivity17c3ba92007-06-04 15:58:30 +03003735 context->root_hpa = INVALID_PAGE;
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02003736 context->direct_map = false;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003737}
3738
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003739static void paging32E_init_context(struct kvm_vcpu *vcpu,
3740 struct kvm_mmu *context)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003741{
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003742 paging64_init_context_common(vcpu, context, PT32E_ROOT_LEVEL);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003743}
3744
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003745static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
Joerg Roedelfb72d162008-02-07 13:47:44 +01003746{
Joerg Roedel14dfe852010-09-10 17:30:49 +02003747 struct kvm_mmu *context = vcpu->arch.walk_mmu;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003748
Avi Kivityc445f8e2010-12-21 16:26:01 +02003749 context->base_role.word = 0;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003750 context->page_fault = tdp_page_fault;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -03003751 context->sync_page = nonpaging_sync_page;
Marcelo Tosattia7052892008-09-23 13:18:35 -03003752 context->invlpg = nonpaging_invlpg;
Xiao Guangrong0f53b5b2011-03-09 15:43:51 +08003753 context->update_pte = nonpaging_update_pte;
Sheng Yang67253af2008-04-25 10:20:22 +08003754 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
Joerg Roedelfb72d162008-02-07 13:47:44 +01003755 context->root_hpa = INVALID_PAGE;
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02003756 context->direct_map = true;
Joerg Roedel1c97f0a2010-09-10 17:30:41 +02003757 context->set_cr3 = kvm_x86_ops->set_tdp_cr3;
Joerg Roedel5777ed32010-09-10 17:30:42 +02003758 context->get_cr3 = get_cr3;
Avi Kivitye4e517b2011-07-28 11:36:17 +03003759 context->get_pdptr = kvm_pdptr_read;
Joerg Roedelcb659db2010-09-10 17:30:43 +02003760 context->inject_page_fault = kvm_inject_page_fault;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003761
3762 if (!is_paging(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003763 context->nx = false;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003764 context->gva_to_gpa = nonpaging_gva_to_gpa;
3765 context->root_level = 0;
3766 } else if (is_long_mode(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003767 context->nx = is_nx(vcpu);
Joerg Roedelfb72d162008-02-07 13:47:44 +01003768 context->root_level = PT64_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003769 reset_rsvds_bits_mask(vcpu, context);
3770 context->gva_to_gpa = paging64_gva_to_gpa;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003771 } else if (is_pae(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003772 context->nx = is_nx(vcpu);
Joerg Roedelfb72d162008-02-07 13:47:44 +01003773 context->root_level = PT32E_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003774 reset_rsvds_bits_mask(vcpu, context);
3775 context->gva_to_gpa = paging64_gva_to_gpa;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003776 } else {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003777 context->nx = false;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003778 context->root_level = PT32_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003779 reset_rsvds_bits_mask(vcpu, context);
3780 context->gva_to_gpa = paging32_gva_to_gpa;
Joerg Roedelfb72d162008-02-07 13:47:44 +01003781 }
3782
Yang Zhang25d92082013-08-06 12:00:32 +03003783 update_permission_bitmask(vcpu, context, false);
Avi Kivity6fd01b72012-09-12 20:46:56 +03003784 update_last_pte_bitmap(vcpu, context);
Joerg Roedelfb72d162008-02-07 13:47:44 +01003785}
3786
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003787void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *context)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003788{
Avi Kivity411c5882011-06-06 16:11:54 +03003789 bool smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003790 ASSERT(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003791 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
Avi Kivity6aa8b732006-12-10 02:21:36 -08003792
3793 if (!is_paging(vcpu))
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003794 nonpaging_init_context(vcpu, context);
Avi Kivitya9058ec2006-12-29 16:49:37 -08003795 else if (is_long_mode(vcpu))
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003796 paging64_init_context(vcpu, context);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003797 else if (is_pae(vcpu))
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003798 paging32E_init_context(vcpu, context);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003799 else
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003800 paging32_init_context(vcpu, context);
Avi Kivitya770f6f2008-12-21 19:20:09 +02003801
Gleb Natapov2c9afa52013-01-30 16:45:02 +02003802 vcpu->arch.mmu.base_role.nxe = is_nx(vcpu);
Avi Kivity5b7e0102010-04-14 19:20:03 +03003803 vcpu->arch.mmu.base_role.cr4_pae = !!is_pae(vcpu);
Joerg Roedelf43addd2010-09-10 17:30:40 +02003804 vcpu->arch.mmu.base_role.cr0_wp = is_write_protection(vcpu);
Avi Kivity411c5882011-06-06 16:11:54 +03003805 vcpu->arch.mmu.base_role.smep_andnot_wp
3806 = smep && !is_write_protection(vcpu);
Joerg Roedel52fde8d2010-09-10 17:30:44 +02003807}
3808EXPORT_SYMBOL_GPL(kvm_init_shadow_mmu);
3809
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003810void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *context,
Nadav Har'El155a97a2013-08-05 11:07:16 +03003811 bool execonly)
3812{
3813 ASSERT(vcpu);
3814 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
3815
3816 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
3817
3818 context->nx = true;
Nadav Har'El155a97a2013-08-05 11:07:16 +03003819 context->page_fault = ept_page_fault;
3820 context->gva_to_gpa = ept_gva_to_gpa;
3821 context->sync_page = ept_sync_page;
3822 context->invlpg = ept_invlpg;
3823 context->update_pte = ept_update_pte;
Nadav Har'El155a97a2013-08-05 11:07:16 +03003824 context->root_level = context->shadow_root_level;
3825 context->root_hpa = INVALID_PAGE;
3826 context->direct_map = false;
3827
3828 update_permission_bitmask(vcpu, context, true);
3829 reset_rsvds_bits_mask_ept(vcpu, context, execonly);
Nadav Har'El155a97a2013-08-05 11:07:16 +03003830}
3831EXPORT_SYMBOL_GPL(kvm_init_shadow_ept_mmu);
3832
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003833static void init_kvm_softmmu(struct kvm_vcpu *vcpu)
Joerg Roedel52fde8d2010-09-10 17:30:44 +02003834{
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003835 kvm_init_shadow_mmu(vcpu, vcpu->arch.walk_mmu);
Joerg Roedel14dfe852010-09-10 17:30:49 +02003836 vcpu->arch.walk_mmu->set_cr3 = kvm_x86_ops->set_cr3;
3837 vcpu->arch.walk_mmu->get_cr3 = get_cr3;
Avi Kivitye4e517b2011-07-28 11:36:17 +03003838 vcpu->arch.walk_mmu->get_pdptr = kvm_pdptr_read;
Joerg Roedel14dfe852010-09-10 17:30:49 +02003839 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003840}
3841
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003842static void init_kvm_nested_mmu(struct kvm_vcpu *vcpu)
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003843{
3844 struct kvm_mmu *g_context = &vcpu->arch.nested_mmu;
3845
3846 g_context->get_cr3 = get_cr3;
Avi Kivitye4e517b2011-07-28 11:36:17 +03003847 g_context->get_pdptr = kvm_pdptr_read;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003848 g_context->inject_page_fault = kvm_inject_page_fault;
3849
3850 /*
3851 * Note that arch.mmu.gva_to_gpa translates l2_gva to l1_gpa. The
3852 * translation of l2_gpa to l1_gpa addresses is done using the
3853 * arch.nested_mmu.gva_to_gpa function. Basically the gva_to_gpa
3854 * functions between mmu and nested_mmu are swapped.
3855 */
3856 if (!is_paging(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003857 g_context->nx = false;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003858 g_context->root_level = 0;
3859 g_context->gva_to_gpa = nonpaging_gva_to_gpa_nested;
3860 } else if (is_long_mode(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003861 g_context->nx = is_nx(vcpu);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003862 g_context->root_level = PT64_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003863 reset_rsvds_bits_mask(vcpu, g_context);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003864 g_context->gva_to_gpa = paging64_gva_to_gpa_nested;
3865 } else if (is_pae(vcpu)) {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003866 g_context->nx = is_nx(vcpu);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003867 g_context->root_level = PT32E_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003868 reset_rsvds_bits_mask(vcpu, g_context);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003869 g_context->gva_to_gpa = paging64_gva_to_gpa_nested;
3870 } else {
Joerg Roedel2d48a982010-09-10 17:31:01 +02003871 g_context->nx = false;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003872 g_context->root_level = PT32_ROOT_LEVEL;
Davidlohr Bueso4d6931c2012-03-05 16:53:06 +01003873 reset_rsvds_bits_mask(vcpu, g_context);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003874 g_context->gva_to_gpa = paging32_gva_to_gpa_nested;
3875 }
3876
Yang Zhang25d92082013-08-06 12:00:32 +03003877 update_permission_bitmask(vcpu, g_context, false);
Avi Kivity6fd01b72012-09-12 20:46:56 +03003878 update_last_pte_bitmap(vcpu, g_context);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003879}
3880
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003881static void init_kvm_mmu(struct kvm_vcpu *vcpu)
Joerg Roedelfb72d162008-02-07 13:47:44 +01003882{
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003883 if (mmu_is_nested(vcpu))
3884 return init_kvm_nested_mmu(vcpu);
3885 else if (tdp_enabled)
Joerg Roedelfb72d162008-02-07 13:47:44 +01003886 return init_kvm_tdp_mmu(vcpu);
3887 else
3888 return init_kvm_softmmu(vcpu);
3889}
3890
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003891void kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003892{
Paolo Bonzini95f93af2013-10-02 16:56:12 +02003893 ASSERT(vcpu);
3894
3895 kvm_mmu_unload(vcpu);
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02003896 init_kvm_mmu(vcpu);
Avi Kivity17c3ba92007-06-04 15:58:30 +03003897}
Eddie Dong8668a3c2007-10-10 14:26:45 +08003898EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
Avi Kivity17c3ba92007-06-04 15:58:30 +03003899
3900int kvm_mmu_load(struct kvm_vcpu *vcpu)
3901{
Avi Kivity714b93d2007-01-05 16:36:53 -08003902 int r;
3903
Avi Kivitye2dec932007-01-05 16:36:54 -08003904 r = mmu_topup_memory_caches(vcpu);
Avi Kivity17c3ba92007-06-04 15:58:30 +03003905 if (r)
3906 goto out;
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003907 r = mmu_alloc_roots(vcpu);
Takuya Yoshikawae2858b42013-05-09 15:45:12 +09003908 kvm_mmu_sync_roots(vcpu);
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03003909 if (r)
3910 goto out;
Sheng Yang3662cb12009-07-09 17:00:42 +08003911 /* set_cr3() should ensure TLB has been flushed */
Joerg Roedelf43addd2010-09-10 17:30:40 +02003912 vcpu->arch.mmu.set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
Avi Kivity714b93d2007-01-05 16:36:53 -08003913out:
3914 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003915}
Avi Kivity17c3ba92007-06-04 15:58:30 +03003916EXPORT_SYMBOL_GPL(kvm_mmu_load);
3917
3918void kvm_mmu_unload(struct kvm_vcpu *vcpu)
3919{
3920 mmu_free_roots(vcpu);
Paolo Bonzini95f93af2013-10-02 16:56:12 +02003921 WARN_ON(VALID_PAGE(vcpu->arch.mmu.root_hpa));
Avi Kivity17c3ba92007-06-04 15:58:30 +03003922}
Joerg Roedel4b161842010-09-10 17:31:03 +02003923EXPORT_SYMBOL_GPL(kvm_mmu_unload);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003924
Avi Kivity00284252007-05-01 16:53:31 +03003925static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
Xiao Guangrong7c562522011-03-28 10:29:27 +08003926 struct kvm_mmu_page *sp, u64 *spte,
3927 const void *new)
Avi Kivity00284252007-05-01 16:53:31 +03003928{
Marcelo Tosatti30945382008-06-11 20:32:40 -03003929 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
Joerg Roedel7e4e4052009-07-27 16:30:46 +02003930 ++vcpu->kvm->stat.mmu_pde_zapped;
3931 return;
Marcelo Tosatti30945382008-06-11 20:32:40 -03003932 }
Avi Kivity00284252007-05-01 16:53:31 +03003933
Avi Kivity4cee5762007-11-18 16:37:07 +02003934 ++vcpu->kvm->stat.mmu_pte_updated;
Xiao Guangrong7c562522011-03-28 10:29:27 +08003935 vcpu->arch.mmu.update_pte(vcpu, sp, spte, new);
Avi Kivity00284252007-05-01 16:53:31 +03003936}
3937
Avi Kivity79539ce2007-11-21 02:06:21 +02003938static bool need_remote_flush(u64 old, u64 new)
3939{
3940 if (!is_shadow_present_pte(old))
3941 return false;
3942 if (!is_shadow_present_pte(new))
3943 return true;
3944 if ((old ^ new) & PT64_BASE_ADDR_MASK)
3945 return true;
Gleb Natapov53166222013-08-05 11:07:14 +03003946 old ^= shadow_nx_mask;
3947 new ^= shadow_nx_mask;
Avi Kivity79539ce2007-11-21 02:06:21 +02003948 return (old & ~new & PT64_PERM_MASK) != 0;
3949}
3950
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08003951static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, bool zap_page,
3952 bool remote_flush, bool local_flush)
Avi Kivity79539ce2007-11-21 02:06:21 +02003953{
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08003954 if (zap_page)
3955 return;
3956
3957 if (remote_flush)
Avi Kivity79539ce2007-11-21 02:06:21 +02003958 kvm_flush_remote_tlbs(vcpu->kvm);
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08003959 else if (local_flush)
Avi Kivity79539ce2007-11-21 02:06:21 +02003960 kvm_mmu_flush_tlb(vcpu);
3961}
3962
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003963static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa,
3964 const u8 *new, int *bytes)
Avi Kivityda4a00f2007-01-05 16:36:44 -08003965{
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003966 u64 gentry;
3967 int r;
Avi Kivity72016f32010-03-15 13:59:53 +02003968
Avi Kivity08e850c2010-03-15 13:59:57 +02003969 /*
3970 * Assume that the pte write on a page table of the same type
Xiao Guangrong49b26e22011-03-04 19:00:00 +08003971 * as the current vcpu paging mode since we update the sptes only
3972 * when they have the same mode.
Avi Kivity08e850c2010-03-15 13:59:57 +02003973 */
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003974 if (is_pae(vcpu) && *bytes == 4) {
Avi Kivity08e850c2010-03-15 13:59:57 +02003975 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003976 *gpa &= ~(gpa_t)7;
3977 *bytes = 8;
Gleb Natapov116eb3d2013-01-30 16:45:03 +02003978 r = kvm_read_guest(vcpu->kvm, *gpa, &gentry, 8);
Avi Kivity08e850c2010-03-15 13:59:57 +02003979 if (r)
3980 gentry = 0;
3981 new = (const u8 *)&gentry;
3982 }
3983
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003984 switch (*bytes) {
Avi Kivity72016f32010-03-15 13:59:53 +02003985 case 4:
3986 gentry = *(const u32 *)new;
3987 break;
3988 case 8:
3989 gentry = *(const u64 *)new;
3990 break;
3991 default:
3992 gentry = 0;
3993 break;
3994 }
3995
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08003996 return gentry;
3997}
3998
3999/*
4000 * If we're seeing too many writes to a page, it may no longer be a page table,
4001 * or we may be forking, in which case it is better to unmap the page.
4002 */
Xiao Guangronga138fe72011-12-16 18:18:10 +08004003static bool detect_write_flooding(struct kvm_mmu_page *sp)
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004004{
Xiao Guangronga30f47c2011-09-22 16:58:36 +08004005 /*
4006 * Skip write-flooding detected for the sp whose level is 1, because
4007 * it can become unsync, then the guest page is not write-protected.
4008 */
Davidlohr Buesof71fa312012-04-18 12:24:29 +02004009 if (sp->role.level == PT_PAGE_TABLE_LEVEL)
Xiao Guangronga30f47c2011-09-22 16:58:36 +08004010 return false;
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004011
Xiao Guangronga30f47c2011-09-22 16:58:36 +08004012 return ++sp->write_flooding_count >= 3;
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004013}
4014
4015/*
4016 * Misaligned accesses are too much trouble to fix up; also, they usually
4017 * indicate a page is not used as a page table.
4018 */
4019static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa,
4020 int bytes)
4021{
4022 unsigned offset, pte_size, misaligned;
4023
4024 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
4025 gpa, bytes, sp->role.word);
4026
4027 offset = offset_in_page(gpa);
4028 pte_size = sp->role.cr4_pae ? 8 : 4;
Xiao Guangrong5d9ca302011-09-22 16:57:55 +08004029
4030 /*
4031 * Sometimes, the OS only writes the last one bytes to update status
4032 * bits, for example, in linux, andb instruction is used in clear_bit().
4033 */
4034 if (!(offset & (pte_size - 1)) && bytes == 1)
4035 return false;
4036
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004037 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
4038 misaligned |= bytes < 4;
4039
4040 return misaligned;
4041}
4042
4043static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte)
4044{
4045 unsigned page_offset, quadrant;
4046 u64 *spte;
4047 int level;
4048
4049 page_offset = offset_in_page(gpa);
4050 level = sp->role.level;
4051 *nspte = 1;
4052 if (!sp->role.cr4_pae) {
4053 page_offset <<= 1; /* 32->64 */
4054 /*
4055 * A 32-bit pde maps 4MB while the shadow pdes map
4056 * only 2MB. So we need to double the offset again
4057 * and zap two pdes instead of one.
4058 */
4059 if (level == PT32_ROOT_LEVEL) {
4060 page_offset &= ~7; /* kill rounding error */
4061 page_offset <<= 1;
4062 *nspte = 2;
4063 }
4064 quadrant = page_offset >> PAGE_SHIFT;
4065 page_offset &= ~PAGE_MASK;
4066 if (quadrant != sp->role.quadrant)
4067 return NULL;
4068 }
4069
4070 spte = &sp->spt[page_offset / sizeof(*spte)];
4071 return spte;
4072}
4073
4074void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
4075 const u8 *new, int bytes)
4076{
4077 gfn_t gfn = gpa >> PAGE_SHIFT;
4078 union kvm_mmu_page_role mask = { .word = 0 };
4079 struct kvm_mmu_page *sp;
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004080 LIST_HEAD(invalid_list);
4081 u64 entry, gentry, *spte;
4082 int npte;
Xiao Guangronga30f47c2011-09-22 16:58:36 +08004083 bool remote_flush, local_flush, zap_page;
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004084
4085 /*
4086 * If we don't have indirect shadow pages, it means no page is
4087 * write-protected, so we can exit simply.
4088 */
4089 if (!ACCESS_ONCE(vcpu->kvm->arch.indirect_shadow_pages))
4090 return;
4091
4092 zap_page = remote_flush = local_flush = false;
4093
4094 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes);
4095
4096 gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, new, &bytes);
4097
4098 /*
4099 * No need to care whether allocation memory is successful
4100 * or not since pte prefetch is skiped if it does not have
4101 * enough objects in the cache.
4102 */
4103 mmu_topup_memory_caches(vcpu);
4104
4105 spin_lock(&vcpu->kvm->mmu_lock);
4106 ++vcpu->kvm->stat.mmu_pte_write;
Xiao Guangrong0375f7f2011-11-28 20:41:00 +08004107 kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE);
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004108
Xiao Guangrongfa1de2b2010-07-16 11:19:51 +08004109 mask.cr0_wp = mask.cr4_pae = mask.nxe = 1;
Sasha Levinb67bfe02013-02-27 17:06:00 -08004110 for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn) {
Xiao Guangronga30f47c2011-09-22 16:58:36 +08004111 if (detect_write_misaligned(sp, gpa, bytes) ||
Xiao Guangronga138fe72011-12-16 18:18:10 +08004112 detect_write_flooding(sp)) {
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08004113 zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp,
Xiao Guangrongf41d3352010-06-04 21:56:11 +08004114 &invalid_list);
Avi Kivity4cee5762007-11-18 16:37:07 +02004115 ++vcpu->kvm->stat.mmu_flooded;
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08004116 continue;
4117 }
Xiao Guangrong889e5cb2011-09-22 16:57:23 +08004118
4119 spte = get_written_sptes(sp, gpa, &npte);
4120 if (!spte)
4121 continue;
4122
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08004123 local_flush = true;
Avi Kivityac1b7142007-03-08 17:13:32 +02004124 while (npte--) {
Avi Kivity79539ce2007-11-21 02:06:21 +02004125 entry = *spte;
Xiao Guangrong38e3b2b2011-05-15 23:27:52 +08004126 mmu_page_zap_pte(vcpu->kvm, sp, spte);
Xiao Guangrongfa1de2b2010-07-16 11:19:51 +08004127 if (gentry &&
4128 !((sp->role.word ^ vcpu->arch.mmu.base_role.word)
Xiao Guangrongf759e2b2011-09-22 16:53:17 +08004129 & mask.word) && rmap_can_add(vcpu))
Xiao Guangrong7c562522011-03-28 10:29:27 +08004130 mmu_pte_write_new_pte(vcpu, sp, spte, &gentry);
Gleb Natapov9bb4f6b2013-01-30 16:45:01 +02004131 if (need_remote_flush(entry, *spte))
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08004132 remote_flush = true;
Avi Kivityac1b7142007-03-08 17:13:32 +02004133 ++spte;
Avi Kivity9b7a0322007-01-05 16:36:45 -08004134 }
Avi Kivity9b7a0322007-01-05 16:36:45 -08004135 }
Xiao Guangrong0671a8e2010-06-04 21:56:59 +08004136 mmu_pte_write_flush_tlb(vcpu, zap_page, remote_flush, local_flush);
Xiao Guangrongd98ba052010-06-04 21:55:29 +08004137 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
Xiao Guangrong0375f7f2011-11-28 20:41:00 +08004138 kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -05004139 spin_unlock(&vcpu->kvm->mmu_lock);
Avi Kivityda4a00f2007-01-05 16:36:44 -08004140}
4141
Avi Kivitya4360362007-01-05 16:36:45 -08004142int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
4143{
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004144 gpa_t gpa;
4145 int r;
Avi Kivitya4360362007-01-05 16:36:45 -08004146
Joerg Roedelc5a78f2b2010-09-10 17:30:39 +02004147 if (vcpu->arch.mmu.direct_map)
Avi Kivity60f24782009-08-27 13:37:06 +03004148 return 0;
4149
Gleb Natapov1871c602010-02-10 14:21:32 +02004150 gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004151
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004152 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004153
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004154 return r;
Avi Kivitya4360362007-01-05 16:36:45 -08004155}
Avi Kivity577bdc42008-07-19 08:57:05 +03004156EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
Avi Kivitya4360362007-01-05 16:36:45 -08004157
Takuya Yoshikawa81f4f762013-03-21 19:34:27 +09004158static void make_mmu_pages_available(struct kvm_vcpu *vcpu)
Avi Kivityebeace82007-01-05 16:36:47 -08004159{
Xiao Guangrongd98ba052010-06-04 21:55:29 +08004160 LIST_HEAD(invalid_list);
Xiao Guangrong103ad252010-06-04 21:54:38 +08004161
Takuya Yoshikawa81f4f762013-03-21 19:34:27 +09004162 if (likely(kvm_mmu_available_pages(vcpu->kvm) >= KVM_MIN_FREE_MMU_PAGES))
4163 return;
4164
Takuya Yoshikawa5da59602013-03-06 16:06:58 +09004165 while (kvm_mmu_available_pages(vcpu->kvm) < KVM_REFILL_PAGES) {
4166 if (!prepare_zap_oldest_mmu_page(vcpu->kvm, &invalid_list))
4167 break;
Avi Kivityebeace82007-01-05 16:36:47 -08004168
Avi Kivity4cee5762007-11-18 16:37:07 +02004169 ++vcpu->kvm->stat.mmu_recycled;
Avi Kivityebeace82007-01-05 16:36:47 -08004170 }
Xiao Guangrongaa6bd182011-07-12 03:26:40 +08004171 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
Avi Kivityebeace82007-01-05 16:36:47 -08004172}
Avi Kivityebeace82007-01-05 16:36:47 -08004173
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004174static bool is_mmio_page_fault(struct kvm_vcpu *vcpu, gva_t addr)
4175{
4176 if (vcpu->arch.mmu.direct_map || mmu_is_nested(vcpu))
4177 return vcpu_match_mmio_gpa(vcpu, addr);
4178
4179 return vcpu_match_mmio_gva(vcpu, addr);
4180}
4181
Andre Przywaradc25e892010-12-21 11:12:07 +01004182int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code,
4183 void *insn, int insn_len)
Avi Kivity30677142007-10-28 18:48:59 +02004184{
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004185 int r, emulation_type = EMULTYPE_RETRY;
Avi Kivity30677142007-10-28 18:48:59 +02004186 enum emulation_result er;
4187
Gleb Natapov56028d02010-10-17 18:13:42 +02004188 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code, false);
Avi Kivity30677142007-10-28 18:48:59 +02004189 if (r < 0)
4190 goto out;
4191
4192 if (!r) {
4193 r = 1;
4194 goto out;
4195 }
4196
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004197 if (is_mmio_page_fault(vcpu, cr2))
4198 emulation_type = 0;
4199
4200 er = x86_emulate_instruction(vcpu, cr2, emulation_type, insn, insn_len);
Avi Kivity30677142007-10-28 18:48:59 +02004201
4202 switch (er) {
4203 case EMULATE_DONE:
4204 return 1;
Paolo Bonziniac0a48c2013-06-25 18:24:41 +02004205 case EMULATE_USER_EXIT:
Avi Kivity30677142007-10-28 18:48:59 +02004206 ++vcpu->stat.mmio_exits;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004207 /* fall through */
Avi Kivity30677142007-10-28 18:48:59 +02004208 case EMULATE_FAIL:
Avi Kivity3f5d18a2009-06-11 15:43:28 +03004209 return 0;
Avi Kivity30677142007-10-28 18:48:59 +02004210 default:
4211 BUG();
4212 }
4213out:
Avi Kivity30677142007-10-28 18:48:59 +02004214 return r;
4215}
4216EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
4217
Marcelo Tosattia7052892008-09-23 13:18:35 -03004218void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
4219{
Marcelo Tosattia7052892008-09-23 13:18:35 -03004220 vcpu->arch.mmu.invlpg(vcpu, gva);
Marcelo Tosattia7052892008-09-23 13:18:35 -03004221 kvm_mmu_flush_tlb(vcpu);
4222 ++vcpu->stat.invlpg;
4223}
4224EXPORT_SYMBOL_GPL(kvm_mmu_invlpg);
4225
Joerg Roedel18552672008-02-07 13:47:41 +01004226void kvm_enable_tdp(void)
4227{
4228 tdp_enabled = true;
4229}
4230EXPORT_SYMBOL_GPL(kvm_enable_tdp);
4231
Joerg Roedel5f4cb662008-07-14 20:36:36 +02004232void kvm_disable_tdp(void)
4233{
4234 tdp_enabled = false;
4235}
4236EXPORT_SYMBOL_GPL(kvm_disable_tdp);
4237
Avi Kivity6aa8b732006-12-10 02:21:36 -08004238static void free_mmu_pages(struct kvm_vcpu *vcpu)
4239{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08004240 free_page((unsigned long)vcpu->arch.mmu.pae_root);
Joerg Roedel81407ca2010-09-10 17:31:00 +02004241 if (vcpu->arch.mmu.lm_root != NULL)
4242 free_page((unsigned long)vcpu->arch.mmu.lm_root);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004243}
4244
4245static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
4246{
Avi Kivity17ac10a2007-01-05 16:36:40 -08004247 struct page *page;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004248 int i;
4249
4250 ASSERT(vcpu);
4251
Avi Kivity17ac10a2007-01-05 16:36:40 -08004252 /*
4253 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
4254 * Therefore we need to allocate shadow page tables in the first
4255 * 4GB of memory, which happens to fit the DMA32 zone.
4256 */
4257 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
4258 if (!page)
Wei Yongjund7fa6ab2010-01-22 16:55:05 +08004259 return -ENOMEM;
4260
Zhang Xiantaoad312c72007-12-13 23:50:52 +08004261 vcpu->arch.mmu.pae_root = page_address(page);
Avi Kivity17ac10a2007-01-05 16:36:40 -08004262 for (i = 0; i < 4; ++i)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08004263 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
Avi Kivity17ac10a2007-01-05 16:36:40 -08004264
Avi Kivity6aa8b732006-12-10 02:21:36 -08004265 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004266}
4267
Ingo Molnar8018c272006-12-29 16:50:01 -08004268int kvm_mmu_create(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08004269{
Avi Kivity6aa8b732006-12-10 02:21:36 -08004270 ASSERT(vcpu);
Xiao Guangronge459e322011-11-28 20:42:16 +08004271
4272 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
4273 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4274 vcpu->arch.mmu.translate_gpa = translate_gpa;
4275 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004276
Ingo Molnar8018c272006-12-29 16:50:01 -08004277 return alloc_mmu_pages(vcpu);
4278}
Avi Kivity6aa8b732006-12-10 02:21:36 -08004279
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02004280void kvm_mmu_setup(struct kvm_vcpu *vcpu)
Ingo Molnar8018c272006-12-29 16:50:01 -08004281{
4282 ASSERT(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08004283 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
Avi Kivity2c264952006-12-22 01:05:28 -08004284
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02004285 init_kvm_mmu(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004286}
4287
Avi Kivity90cb0522007-07-17 13:04:56 +03004288void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
Avi Kivity6aa8b732006-12-10 02:21:36 -08004289{
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004290 struct kvm_memory_slot *memslot;
4291 gfn_t last_gfn;
4292 int i;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004293
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004294 memslot = id_to_memslot(kvm->memslots, slot);
4295 last_gfn = memslot->base_gfn + memslot->npages - 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004296
Takuya Yoshikawa9d1beef2013-01-08 19:46:48 +09004297 spin_lock(&kvm->mmu_lock);
4298
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004299 for (i = PT_PAGE_TABLE_LEVEL;
4300 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
4301 unsigned long *rmapp;
4302 unsigned long last_index, index;
Avi Kivity6aa8b732006-12-10 02:21:36 -08004303
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004304 rmapp = memslot->arch.rmap[i - PT_PAGE_TABLE_LEVEL];
4305 last_index = gfn_to_index(last_gfn, memslot->base_gfn, i);
Xiao Guangrongda8dc752011-03-04 18:56:41 +08004306
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004307 for (index = 0; index <= last_index; ++index, ++rmapp) {
4308 if (*rmapp)
4309 __rmap_write_protect(kvm, rmapp, false);
Takuya Yoshikawa6b81b052013-01-08 19:47:33 +09004310
Xiao Guangrong198c74f2014-04-17 17:06:16 +08004311 if (need_resched() || spin_needbreak(&kvm->mmu_lock))
Takuya Yoshikawa6b81b052013-01-08 19:47:33 +09004312 cond_resched_lock(&kvm->mmu_lock);
Avi Kivity8234b222010-12-27 12:08:45 +02004313 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08004314 }
Takuya Yoshikawab99db1d2013-01-08 19:44:48 +09004315
Takuya Yoshikawa9d1beef2013-01-08 19:46:48 +09004316 spin_unlock(&kvm->mmu_lock);
Xiao Guangrong198c74f2014-04-17 17:06:16 +08004317
4318 /*
4319 * kvm_mmu_slot_remove_write_access() and kvm_vm_ioctl_get_dirty_log()
4320 * which do tlb flush out of mmu-lock should be serialized by
4321 * kvm->slots_lock otherwise tlb flush would be missed.
4322 */
4323 lockdep_assert_held(&kvm->slots_lock);
4324
4325 /*
4326 * We can flush all the TLBs out of the mmu lock without TLB
4327 * corruption since we just change the spte from writable to
4328 * readonly so that we only need to care the case of changing
4329 * spte from present to present (changing the spte from present
4330 * to nonpresent will flush all the TLBs immediately), in other
4331 * words, the only case we care is mmu_spte_update() where we
4332 * haved checked SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE
4333 * instead of PT_WRITABLE_MASK, that means it does not depend
4334 * on PT_WRITABLE_MASK anymore.
4335 */
4336 kvm_flush_remote_tlbs(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004337}
Avi Kivity37a7d8b2007-01-05 16:36:56 -08004338
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004339#define BATCH_ZAP_PAGES 10
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004340static void kvm_zap_obsolete_pages(struct kvm *kvm)
4341{
4342 struct kvm_mmu_page *sp, *node;
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004343 int batch = 0;
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004344
4345restart:
4346 list_for_each_entry_safe_reverse(sp, node,
4347 &kvm->arch.active_mmu_pages, link) {
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004348 int ret;
4349
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004350 /*
4351 * No obsolete page exists before new created page since
4352 * active_mmu_pages is the FIFO list.
4353 */
4354 if (!is_obsolete_sp(kvm, sp))
4355 break;
4356
4357 /*
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004358 * Since we are reversely walking the list and the invalid
4359 * list will be moved to the head, skip the invalid page
4360 * can help us to avoid the infinity list walking.
4361 */
4362 if (sp->role.invalid)
4363 continue;
4364
Xiao Guangrongf34d2512013-05-31 08:36:28 +08004365 /*
4366 * Need not flush tlb since we only zap the sp with invalid
4367 * generation number.
4368 */
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004369 if (batch >= BATCH_ZAP_PAGES &&
Xiao Guangrongf34d2512013-05-31 08:36:28 +08004370 cond_resched_lock(&kvm->mmu_lock)) {
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004371 batch = 0;
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004372 goto restart;
4373 }
4374
Xiao Guangrong365c8862013-05-31 08:36:29 +08004375 ret = kvm_mmu_prepare_zap_page(kvm, sp,
4376 &kvm->arch.zapped_obsolete_pages);
Xiao Guangronge7d11c72013-05-31 08:36:27 +08004377 batch += ret;
4378
4379 if (ret)
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004380 goto restart;
4381 }
4382
Xiao Guangrongf34d2512013-05-31 08:36:28 +08004383 /*
4384 * Should flush tlb before free page tables since lockless-walking
4385 * may use the pages.
4386 */
Xiao Guangrong365c8862013-05-31 08:36:29 +08004387 kvm_mmu_commit_zap_page(kvm, &kvm->arch.zapped_obsolete_pages);
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004388}
4389
4390/*
4391 * Fast invalidate all shadow pages and use lock-break technique
4392 * to zap obsolete pages.
4393 *
4394 * It's required when memslot is being deleted or VM is being
4395 * destroyed, in these cases, we should ensure that KVM MMU does
4396 * not use any resource of the being-deleted slot or all slots
4397 * after calling the function.
4398 */
4399void kvm_mmu_invalidate_zap_all_pages(struct kvm *kvm)
4400{
4401 spin_lock(&kvm->mmu_lock);
Xiao Guangrong35006122013-05-31 08:36:25 +08004402 trace_kvm_mmu_invalidate_zap_all_pages(kvm);
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004403 kvm->arch.mmu_valid_gen++;
4404
Xiao Guangrongf34d2512013-05-31 08:36:28 +08004405 /*
4406 * Notify all vcpus to reload its shadow page table
4407 * and flush TLB. Then all vcpus will switch to new
4408 * shadow page table with the new mmu_valid_gen.
4409 *
4410 * Note: we should do this under the protection of
4411 * mmu-lock, otherwise, vcpu would purge shadow page
4412 * but miss tlb flush.
4413 */
4414 kvm_reload_remote_mmus(kvm);
4415
Xiao Guangrong5304b8d2013-05-31 08:36:22 +08004416 kvm_zap_obsolete_pages(kvm);
4417 spin_unlock(&kvm->mmu_lock);
4418}
4419
Xiao Guangrong365c8862013-05-31 08:36:29 +08004420static bool kvm_has_zapped_obsolete_pages(struct kvm *kvm)
4421{
4422 return unlikely(!list_empty_careful(&kvm->arch.zapped_obsolete_pages));
4423}
4424
Xiao Guangrongf8f55942013-06-07 16:51:26 +08004425void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm)
4426{
4427 /*
4428 * The very rare case: if the generation-number is round,
4429 * zap all shadow pages.
Xiao Guangrongf8f55942013-06-07 16:51:26 +08004430 */
Takuya Yoshikawae6dff7d2013-07-04 13:41:26 +09004431 if (unlikely(kvm_current_mmio_generation(kvm) >= MMIO_MAX_GEN)) {
Takuya Yoshikawa7a2e8aa2013-06-21 01:34:31 +09004432 printk_ratelimited(KERN_INFO "kvm: zapping shadow pages for mmio generation wraparound\n");
Xiao Guangronga8eca9d2013-06-10 16:28:55 +08004433 kvm_mmu_invalidate_zap_all_pages(kvm);
Takuya Yoshikawa7a2e8aa2013-06-21 01:34:31 +09004434 }
Xiao Guangrongf8f55942013-06-07 16:51:26 +08004435}
4436
Dave Chinner70534a72013-08-28 10:18:14 +10004437static unsigned long
4438mmu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
Izik Eidus3ee16c82008-03-30 15:17:21 +03004439{
4440 struct kvm *kvm;
Ying Han1495f232011-05-24 17:12:27 -07004441 int nr_to_scan = sc->nr_to_scan;
Dave Chinner70534a72013-08-28 10:18:14 +10004442 unsigned long freed = 0;
Izik Eidus3ee16c82008-03-30 15:17:21 +03004443
Paolo Bonzini2f303b72013-09-25 13:53:07 +02004444 spin_lock(&kvm_lock);
Izik Eidus3ee16c82008-03-30 15:17:21 +03004445
4446 list_for_each_entry(kvm, &vm_list, vm_list) {
Jan Kiszka3d56cbd2011-12-02 18:35:24 +01004447 int idx;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08004448 LIST_HEAD(invalid_list);
Izik Eidus3ee16c82008-03-30 15:17:21 +03004449
Gleb Natapov19526392012-06-04 14:53:23 +03004450 /*
Takuya Yoshikawa35f2d162012-08-20 18:35:39 +09004451 * Never scan more than sc->nr_to_scan VM instances.
4452 * Will not hit this condition practically since we do not try
4453 * to shrink more than one VM and it is very unlikely to see
4454 * !n_used_mmu_pages so many times.
4455 */
4456 if (!nr_to_scan--)
4457 break;
4458 /*
Gleb Natapov19526392012-06-04 14:53:23 +03004459 * n_used_mmu_pages is accessed without holding kvm->mmu_lock
4460 * here. We may skip a VM instance errorneosly, but we do not
4461 * want to shrink a VM that only started to populate its MMU
4462 * anyway.
4463 */
Xiao Guangrong365c8862013-05-31 08:36:29 +08004464 if (!kvm->arch.n_used_mmu_pages &&
4465 !kvm_has_zapped_obsolete_pages(kvm))
Gleb Natapov19526392012-06-04 14:53:23 +03004466 continue;
Gleb Natapov19526392012-06-04 14:53:23 +03004467
Marcelo Tosattif656ce02009-12-23 14:35:25 -02004468 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3ee16c82008-03-30 15:17:21 +03004469 spin_lock(&kvm->mmu_lock);
Izik Eidus3ee16c82008-03-30 15:17:21 +03004470
Xiao Guangrong365c8862013-05-31 08:36:29 +08004471 if (kvm_has_zapped_obsolete_pages(kvm)) {
4472 kvm_mmu_commit_zap_page(kvm,
4473 &kvm->arch.zapped_obsolete_pages);
4474 goto unlock;
4475 }
4476
Dave Chinner70534a72013-08-28 10:18:14 +10004477 if (prepare_zap_oldest_mmu_page(kvm, &invalid_list))
4478 freed++;
Xiao Guangrongd98ba052010-06-04 21:55:29 +08004479 kvm_mmu_commit_zap_page(kvm, &invalid_list);
Gleb Natapov19526392012-06-04 14:53:23 +03004480
Xiao Guangrong365c8862013-05-31 08:36:29 +08004481unlock:
Izik Eidus3ee16c82008-03-30 15:17:21 +03004482 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02004483 srcu_read_unlock(&kvm->srcu, idx);
Gleb Natapov19526392012-06-04 14:53:23 +03004484
Dave Chinner70534a72013-08-28 10:18:14 +10004485 /*
4486 * unfair on small ones
4487 * per-vm shrinkers cry out
4488 * sadness comes quickly
4489 */
Gleb Natapov19526392012-06-04 14:53:23 +03004490 list_move_tail(&kvm->vm_list, &vm_list);
4491 break;
Izik Eidus3ee16c82008-03-30 15:17:21 +03004492 }
Izik Eidus3ee16c82008-03-30 15:17:21 +03004493
Paolo Bonzini2f303b72013-09-25 13:53:07 +02004494 spin_unlock(&kvm_lock);
Dave Chinner70534a72013-08-28 10:18:14 +10004495 return freed;
Dave Chinner70534a72013-08-28 10:18:14 +10004496}
4497
4498static unsigned long
4499mmu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
4500{
Dave Hansen45221ab2010-08-19 18:11:37 -07004501 return percpu_counter_read_positive(&kvm_total_used_mmu_pages);
Izik Eidus3ee16c82008-03-30 15:17:21 +03004502}
4503
4504static struct shrinker mmu_shrinker = {
Dave Chinner70534a72013-08-28 10:18:14 +10004505 .count_objects = mmu_shrink_count,
4506 .scan_objects = mmu_shrink_scan,
Izik Eidus3ee16c82008-03-30 15:17:21 +03004507 .seeks = DEFAULT_SEEKS * 10,
4508};
4509
Ingo Molnar2ddfd202008-05-22 10:37:48 +02004510static void mmu_destroy_caches(void)
Avi Kivityb5a33a72007-04-15 16:31:09 +03004511{
Xiao Guangrong53c07b12011-05-15 23:26:20 +08004512 if (pte_list_desc_cache)
4513 kmem_cache_destroy(pte_list_desc_cache);
Avi Kivityd3d25b02007-05-30 12:34:53 +03004514 if (mmu_page_header_cache)
4515 kmem_cache_destroy(mmu_page_header_cache);
Avi Kivityb5a33a72007-04-15 16:31:09 +03004516}
4517
4518int kvm_mmu_module_init(void)
4519{
Xiao Guangrong53c07b12011-05-15 23:26:20 +08004520 pte_list_desc_cache = kmem_cache_create("pte_list_desc",
4521 sizeof(struct pte_list_desc),
Paul Mundt20c2df82007-07-20 10:11:58 +09004522 0, 0, NULL);
Xiao Guangrong53c07b12011-05-15 23:26:20 +08004523 if (!pte_list_desc_cache)
Avi Kivityb5a33a72007-04-15 16:31:09 +03004524 goto nomem;
4525
Avi Kivityd3d25b02007-05-30 12:34:53 +03004526 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
4527 sizeof(struct kvm_mmu_page),
Paul Mundt20c2df82007-07-20 10:11:58 +09004528 0, 0, NULL);
Avi Kivityd3d25b02007-05-30 12:34:53 +03004529 if (!mmu_page_header_cache)
4530 goto nomem;
4531
Wei Yongjun45bf21a2010-08-23 16:13:15 +08004532 if (percpu_counter_init(&kvm_total_used_mmu_pages, 0))
4533 goto nomem;
4534
Izik Eidus3ee16c82008-03-30 15:17:21 +03004535 register_shrinker(&mmu_shrinker);
4536
Avi Kivityb5a33a72007-04-15 16:31:09 +03004537 return 0;
4538
4539nomem:
Izik Eidus3ee16c82008-03-30 15:17:21 +03004540 mmu_destroy_caches();
Avi Kivityb5a33a72007-04-15 16:31:09 +03004541 return -ENOMEM;
4542}
4543
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08004544/*
4545 * Caculate mmu pages needed for kvm.
4546 */
4547unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
4548{
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08004549 unsigned int nr_mmu_pages;
4550 unsigned int nr_pages = 0;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02004551 struct kvm_memslots *slots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +08004552 struct kvm_memory_slot *memslot;
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08004553
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08004554 slots = kvm_memslots(kvm);
4555
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +08004556 kvm_for_each_memslot(memslot, slots)
4557 nr_pages += memslot->npages;
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08004558
4559 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
4560 nr_mmu_pages = max(nr_mmu_pages,
4561 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
4562
4563 return nr_mmu_pages;
4564}
4565
Marcelo Tosatti94d8b052009-06-11 12:07:42 -03004566int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4])
4567{
4568 struct kvm_shadow_walk_iterator iterator;
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08004569 u64 spte;
Marcelo Tosatti94d8b052009-06-11 12:07:42 -03004570 int nr_sptes = 0;
4571
Marcelo Tosatti37f6a4e2014-01-03 17:09:32 -02004572 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4573 return nr_sptes;
4574
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08004575 walk_shadow_page_lockless_begin(vcpu);
4576 for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) {
4577 sptes[iterator.level-1] = spte;
Marcelo Tosatti94d8b052009-06-11 12:07:42 -03004578 nr_sptes++;
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08004579 if (!is_shadow_present_pte(spte))
Marcelo Tosatti94d8b052009-06-11 12:07:42 -03004580 break;
4581 }
Xiao Guangrongc2a2ac22011-07-12 03:32:13 +08004582 walk_shadow_page_lockless_end(vcpu);
Marcelo Tosatti94d8b052009-06-11 12:07:42 -03004583
4584 return nr_sptes;
4585}
4586EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy);
4587
Xiao Guangrongc42fffe2010-09-27 18:07:07 +08004588void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
4589{
4590 ASSERT(vcpu);
4591
Paolo Bonzini95f93af2013-10-02 16:56:12 +02004592 kvm_mmu_unload(vcpu);
Xiao Guangrongc42fffe2010-09-27 18:07:07 +08004593 free_mmu_pages(vcpu);
4594 mmu_free_memory_caches(vcpu);
Xiao Guangrongb034cf02010-12-23 16:08:35 +08004595}
4596
Xiao Guangrongb034cf02010-12-23 16:08:35 +08004597void kvm_mmu_module_exit(void)
4598{
4599 mmu_destroy_caches();
4600 percpu_counter_destroy(&kvm_total_used_mmu_pages);
4601 unregister_shrinker(&mmu_shrinker);
Xiao Guangrongc42fffe2010-09-27 18:07:07 +08004602 mmu_audit_disable();
4603}