blob: feb5ac986c5d0eb23480512193a706f247ccd4b3 [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.
10 *
11 * Authors:
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
17 *
18 */
Avi Kivity6aa8b732006-12-10 02:21:36 -080019
20#include "vmx.h"
21#include "kvm.h"
22
Avi Kivitye4956062007-06-28 14:15:57 -040023#include <linux/types.h>
24#include <linux/string.h>
25#include <linux/mm.h>
26#include <linux/highmem.h>
27#include <linux/module.h>
28
29#include <asm/page.h>
30#include <asm/cmpxchg.h>
31
Avi Kivity37a7d8b2007-01-05 16:36:56 -080032#undef MMU_DEBUG
33
34#undef AUDIT
35
36#ifdef AUDIT
37static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg);
38#else
39static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {}
40#endif
41
42#ifdef MMU_DEBUG
43
44#define pgprintk(x...) do { if (dbg) printk(x); } while (0)
45#define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
46
47#else
48
49#define pgprintk(x...) do { } while (0)
50#define rmap_printk(x...) do { } while (0)
51
52#endif
53
54#if defined(MMU_DEBUG) || defined(AUDIT)
55static int dbg = 1;
56#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Yaozu Dongd6c69ee2007-04-25 14:17:25 +080058#ifndef MMU_DEBUG
59#define ASSERT(x) do { } while (0)
60#else
Avi Kivity6aa8b732006-12-10 02:21:36 -080061#define ASSERT(x) \
62 if (!(x)) { \
63 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
64 __FILE__, __LINE__, #x); \
65 }
Yaozu Dongd6c69ee2007-04-25 14:17:25 +080066#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080067
Avi Kivitycea0f0e2007-01-05 16:36:43 -080068#define PT64_PT_BITS 9
69#define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
70#define PT32_PT_BITS 10
71#define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
Avi Kivity6aa8b732006-12-10 02:21:36 -080072
73#define PT_WRITABLE_SHIFT 1
74
75#define PT_PRESENT_MASK (1ULL << 0)
76#define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
77#define PT_USER_MASK (1ULL << 2)
78#define PT_PWT_MASK (1ULL << 3)
79#define PT_PCD_MASK (1ULL << 4)
80#define PT_ACCESSED_MASK (1ULL << 5)
81#define PT_DIRTY_MASK (1ULL << 6)
82#define PT_PAGE_SIZE_MASK (1ULL << 7)
83#define PT_PAT_MASK (1ULL << 7)
84#define PT_GLOBAL_MASK (1ULL << 8)
85#define PT64_NX_MASK (1ULL << 63)
86
87#define PT_PAT_SHIFT 7
88#define PT_DIR_PAT_SHIFT 12
89#define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
90
91#define PT32_DIR_PSE36_SIZE 4
92#define PT32_DIR_PSE36_SHIFT 13
93#define PT32_DIR_PSE36_MASK (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
94
95
Avi Kivity6aa8b732006-12-10 02:21:36 -080096#define PT_FIRST_AVAIL_BITS_SHIFT 9
97#define PT64_SECOND_AVAIL_BITS_SHIFT 52
98
Avi Kivity6aa8b732006-12-10 02:21:36 -080099#define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
100
Avi Kivity6aa8b732006-12-10 02:21:36 -0800101#define VALID_PAGE(x) ((x) != INVALID_PAGE)
102
103#define PT64_LEVEL_BITS 9
104
105#define PT64_LEVEL_SHIFT(level) \
106 ( PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS )
107
108#define PT64_LEVEL_MASK(level) \
109 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
110
111#define PT64_INDEX(address, level)\
112 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
113
114
115#define PT32_LEVEL_BITS 10
116
117#define PT32_LEVEL_SHIFT(level) \
118 ( PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS )
119
120#define PT32_LEVEL_MASK(level) \
121 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
122
123#define PT32_INDEX(address, level)\
124 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
125
126
Avi Kivity27aba762007-03-09 13:04:31 +0200127#define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
Avi Kivity6aa8b732006-12-10 02:21:36 -0800128#define PT64_DIR_BASE_ADDR_MASK \
129 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
130
131#define PT32_BASE_ADDR_MASK PAGE_MASK
132#define PT32_DIR_BASE_ADDR_MASK \
133 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
134
135
136#define PFERR_PRESENT_MASK (1U << 0)
137#define PFERR_WRITE_MASK (1U << 1)
138#define PFERR_USER_MASK (1U << 2)
Avi Kivity73b10872007-01-26 00:56:41 -0800139#define PFERR_FETCH_MASK (1U << 4)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800140
141#define PT64_ROOT_LEVEL 4
142#define PT32_ROOT_LEVEL 2
143#define PT32E_ROOT_LEVEL 3
144
145#define PT_DIRECTORY_LEVEL 2
146#define PT_PAGE_TABLE_LEVEL 1
147
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800148#define RMAP_EXT 4
149
150struct kvm_rmap_desc {
151 u64 *shadow_ptes[RMAP_EXT];
152 struct kvm_rmap_desc *more;
153};
154
Avi Kivityb5a33a72007-04-15 16:31:09 +0300155static struct kmem_cache *pte_chain_cache;
156static struct kmem_cache *rmap_desc_cache;
Avi Kivityd3d25b02007-05-30 12:34:53 +0300157static struct kmem_cache *mmu_page_header_cache;
Avi Kivityb5a33a72007-04-15 16:31:09 +0300158
Avi Kivity6aa8b732006-12-10 02:21:36 -0800159static int is_write_protection(struct kvm_vcpu *vcpu)
160{
Rusty Russell707d92f2007-07-17 23:19:08 +1000161 return vcpu->cr0 & X86_CR0_WP;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800162}
163
164static int is_cpuid_PSE36(void)
165{
166 return 1;
167}
168
Avi Kivity73b10872007-01-26 00:56:41 -0800169static int is_nx(struct kvm_vcpu *vcpu)
170{
171 return vcpu->shadow_efer & EFER_NX;
172}
173
Avi Kivity6aa8b732006-12-10 02:21:36 -0800174static int is_present_pte(unsigned long pte)
175{
176 return pte & PT_PRESENT_MASK;
177}
178
179static int is_writeble_pte(unsigned long pte)
180{
181 return pte & PT_WRITABLE_MASK;
182}
183
184static int is_io_pte(unsigned long pte)
185{
186 return pte & PT_SHADOW_IO_MARK;
187}
188
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800189static int is_rmap_pte(u64 pte)
190{
191 return (pte & (PT_WRITABLE_MASK | PT_PRESENT_MASK))
192 == (PT_WRITABLE_MASK | PT_PRESENT_MASK);
193}
194
Avi Kivitye663ee62007-05-31 15:46:04 +0300195static void set_shadow_pte(u64 *sptep, u64 spte)
196{
197#ifdef CONFIG_X86_64
198 set_64bit((unsigned long *)sptep, spte);
199#else
200 set_64bit((unsigned long long *)sptep, spte);
201#endif
202}
203
Avi Kivitye2dec932007-01-05 16:36:54 -0800204static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
Avi Kivity2e3e5882007-09-10 11:28:17 +0300205 struct kmem_cache *base_cache, int min)
Avi Kivity714b93d2007-01-05 16:36:53 -0800206{
207 void *obj;
208
209 if (cache->nobjs >= min)
Avi Kivitye2dec932007-01-05 16:36:54 -0800210 return 0;
Avi Kivity714b93d2007-01-05 16:36:53 -0800211 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
Avi Kivity2e3e5882007-09-10 11:28:17 +0300212 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
Avi Kivity714b93d2007-01-05 16:36:53 -0800213 if (!obj)
Avi Kivitye2dec932007-01-05 16:36:54 -0800214 return -ENOMEM;
Avi Kivity714b93d2007-01-05 16:36:53 -0800215 cache->objects[cache->nobjs++] = obj;
216 }
Avi Kivitye2dec932007-01-05 16:36:54 -0800217 return 0;
Avi Kivity714b93d2007-01-05 16:36:53 -0800218}
219
220static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
221{
222 while (mc->nobjs)
223 kfree(mc->objects[--mc->nobjs]);
224}
225
Avi Kivityc1158e62007-07-20 08:18:27 +0300226static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
Avi Kivity2e3e5882007-09-10 11:28:17 +0300227 int min)
Avi Kivityc1158e62007-07-20 08:18:27 +0300228{
229 struct page *page;
230
231 if (cache->nobjs >= min)
232 return 0;
233 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
Avi Kivity2e3e5882007-09-10 11:28:17 +0300234 page = alloc_page(GFP_KERNEL);
Avi Kivityc1158e62007-07-20 08:18:27 +0300235 if (!page)
236 return -ENOMEM;
237 set_page_private(page, 0);
238 cache->objects[cache->nobjs++] = page_address(page);
239 }
240 return 0;
241}
242
243static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
244{
245 while (mc->nobjs)
Avi Kivityc4d198d2007-07-21 09:06:46 +0300246 free_page((unsigned long)mc->objects[--mc->nobjs]);
Avi Kivityc1158e62007-07-20 08:18:27 +0300247}
248
Avi Kivity8c438502007-04-16 11:53:17 +0300249static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
250{
251 int r;
252
Avi Kivity22d95b12007-09-14 20:26:06 +0300253 kvm_mmu_free_some_pages(vcpu);
Avi Kivity2e3e5882007-09-10 11:28:17 +0300254 r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache,
255 pte_chain_cache, 4);
256 if (r)
257 goto out;
258 r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache,
259 rmap_desc_cache, 1);
260 if (r)
261 goto out;
262 r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 4);
263 if (r)
264 goto out;
265 r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache,
266 mmu_page_header_cache, 4);
267out:
Avi Kivity8c438502007-04-16 11:53:17 +0300268 return r;
269}
270
Avi Kivity714b93d2007-01-05 16:36:53 -0800271static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
272{
273 mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache);
274 mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache);
Avi Kivityc1158e62007-07-20 08:18:27 +0300275 mmu_free_memory_cache_page(&vcpu->mmu_page_cache);
Avi Kivityd3d25b02007-05-30 12:34:53 +0300276 mmu_free_memory_cache(&vcpu->mmu_page_header_cache);
Avi Kivity714b93d2007-01-05 16:36:53 -0800277}
278
279static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
280 size_t size)
281{
282 void *p;
283
284 BUG_ON(!mc->nobjs);
285 p = mc->objects[--mc->nobjs];
286 memset(p, 0, size);
287 return p;
288}
289
Avi Kivity714b93d2007-01-05 16:36:53 -0800290static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
291{
292 return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache,
293 sizeof(struct kvm_pte_chain));
294}
295
Avi Kivity90cb0522007-07-17 13:04:56 +0300296static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
Avi Kivity714b93d2007-01-05 16:36:53 -0800297{
Avi Kivity90cb0522007-07-17 13:04:56 +0300298 kfree(pc);
Avi Kivity714b93d2007-01-05 16:36:53 -0800299}
300
301static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
302{
303 return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache,
304 sizeof(struct kvm_rmap_desc));
305}
306
Avi Kivity90cb0522007-07-17 13:04:56 +0300307static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
Avi Kivity714b93d2007-01-05 16:36:53 -0800308{
Avi Kivity90cb0522007-07-17 13:04:56 +0300309 kfree(rd);
Avi Kivity714b93d2007-01-05 16:36:53 -0800310}
311
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800312/*
313 * Reverse mapping data structures:
314 *
315 * If page->private bit zero is zero, then page->private points to the
316 * shadow page table entry that points to page_address(page).
317 *
318 * If page->private bit zero is one, (then page->private & ~1) points
319 * to a struct kvm_rmap_desc containing more mappings.
320 */
Avi Kivity714b93d2007-01-05 16:36:53 -0800321static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte)
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800322{
323 struct page *page;
324 struct kvm_rmap_desc *desc;
325 int i;
326
327 if (!is_rmap_pte(*spte))
328 return;
329 page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
Markus Rechberger5972e952007-02-19 14:37:47 +0200330 if (!page_private(page)) {
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800331 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
Markus Rechberger5972e952007-02-19 14:37:47 +0200332 set_page_private(page,(unsigned long)spte);
333 } else if (!(page_private(page) & 1)) {
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800334 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
Avi Kivity714b93d2007-01-05 16:36:53 -0800335 desc = mmu_alloc_rmap_desc(vcpu);
Markus Rechberger5972e952007-02-19 14:37:47 +0200336 desc->shadow_ptes[0] = (u64 *)page_private(page);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800337 desc->shadow_ptes[1] = spte;
Markus Rechberger5972e952007-02-19 14:37:47 +0200338 set_page_private(page,(unsigned long)desc | 1);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800339 } else {
340 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
Markus Rechberger5972e952007-02-19 14:37:47 +0200341 desc = (struct kvm_rmap_desc *)(page_private(page) & ~1ul);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800342 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
343 desc = desc->more;
344 if (desc->shadow_ptes[RMAP_EXT-1]) {
Avi Kivity714b93d2007-01-05 16:36:53 -0800345 desc->more = mmu_alloc_rmap_desc(vcpu);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800346 desc = desc->more;
347 }
348 for (i = 0; desc->shadow_ptes[i]; ++i)
349 ;
350 desc->shadow_ptes[i] = spte;
351 }
352}
353
Avi Kivity90cb0522007-07-17 13:04:56 +0300354static void rmap_desc_remove_entry(struct page *page,
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800355 struct kvm_rmap_desc *desc,
356 int i,
357 struct kvm_rmap_desc *prev_desc)
358{
359 int j;
360
361 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
362 ;
363 desc->shadow_ptes[i] = desc->shadow_ptes[j];
Al Viro11718b4d2007-02-09 16:39:20 +0000364 desc->shadow_ptes[j] = NULL;
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800365 if (j != 0)
366 return;
367 if (!prev_desc && !desc->more)
Markus Rechberger5972e952007-02-19 14:37:47 +0200368 set_page_private(page,(unsigned long)desc->shadow_ptes[0]);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800369 else
370 if (prev_desc)
371 prev_desc->more = desc->more;
372 else
Markus Rechberger5972e952007-02-19 14:37:47 +0200373 set_page_private(page,(unsigned long)desc->more | 1);
Avi Kivity90cb0522007-07-17 13:04:56 +0300374 mmu_free_rmap_desc(desc);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800375}
376
Avi Kivity90cb0522007-07-17 13:04:56 +0300377static void rmap_remove(u64 *spte)
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800378{
379 struct page *page;
380 struct kvm_rmap_desc *desc;
381 struct kvm_rmap_desc *prev_desc;
382 int i;
383
384 if (!is_rmap_pte(*spte))
385 return;
386 page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
Markus Rechberger5972e952007-02-19 14:37:47 +0200387 if (!page_private(page)) {
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800388 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
389 BUG();
Markus Rechberger5972e952007-02-19 14:37:47 +0200390 } else if (!(page_private(page) & 1)) {
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800391 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
Markus Rechberger5972e952007-02-19 14:37:47 +0200392 if ((u64 *)page_private(page) != spte) {
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800393 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
394 spte, *spte);
395 BUG();
396 }
Markus Rechberger5972e952007-02-19 14:37:47 +0200397 set_page_private(page,0);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800398 } else {
399 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
Markus Rechberger5972e952007-02-19 14:37:47 +0200400 desc = (struct kvm_rmap_desc *)(page_private(page) & ~1ul);
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800401 prev_desc = NULL;
402 while (desc) {
403 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
404 if (desc->shadow_ptes[i] == spte) {
Avi Kivity90cb0522007-07-17 13:04:56 +0300405 rmap_desc_remove_entry(page,
Avi Kivity714b93d2007-01-05 16:36:53 -0800406 desc, i,
Avi Kivitycd4a4e52007-01-05 16:36:38 -0800407 prev_desc);
408 return;
409 }
410 prev_desc = desc;
411 desc = desc->more;
412 }
413 BUG();
414 }
415}
416
Avi Kivity714b93d2007-01-05 16:36:53 -0800417static void rmap_write_protect(struct kvm_vcpu *vcpu, u64 gfn)
Avi Kivity374cbac2007-01-05 16:36:43 -0800418{
Avi Kivity714b93d2007-01-05 16:36:53 -0800419 struct kvm *kvm = vcpu->kvm;
Avi Kivity374cbac2007-01-05 16:36:43 -0800420 struct page *page;
Avi Kivity374cbac2007-01-05 16:36:43 -0800421 struct kvm_rmap_desc *desc;
422 u64 *spte;
423
Avi Kivity954bbbc22007-03-30 14:02:32 +0300424 page = gfn_to_page(kvm, gfn);
425 BUG_ON(!page);
Avi Kivity374cbac2007-01-05 16:36:43 -0800426
Markus Rechberger5972e952007-02-19 14:37:47 +0200427 while (page_private(page)) {
428 if (!(page_private(page) & 1))
429 spte = (u64 *)page_private(page);
Avi Kivity374cbac2007-01-05 16:36:43 -0800430 else {
Markus Rechberger5972e952007-02-19 14:37:47 +0200431 desc = (struct kvm_rmap_desc *)(page_private(page) & ~1ul);
Avi Kivity374cbac2007-01-05 16:36:43 -0800432 spte = desc->shadow_ptes[0];
433 }
434 BUG_ON(!spte);
Avi Kivity27aba762007-03-09 13:04:31 +0200435 BUG_ON((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT
436 != page_to_pfn(page));
Avi Kivity374cbac2007-01-05 16:36:43 -0800437 BUG_ON(!(*spte & PT_PRESENT_MASK));
438 BUG_ON(!(*spte & PT_WRITABLE_MASK));
439 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
Avi Kivity90cb0522007-07-17 13:04:56 +0300440 rmap_remove(spte);
Avi Kivitye663ee62007-05-31 15:46:04 +0300441 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
Shaohua Li88a97f02007-06-20 17:13:26 +0800442 kvm_flush_remote_tlbs(vcpu->kvm);
Avi Kivity374cbac2007-01-05 16:36:43 -0800443 }
444}
445
Yaozu Dongd6c69ee2007-04-25 14:17:25 +0800446#ifdef MMU_DEBUG
Avi Kivity47ad8e62007-05-06 15:50:58 +0300447static int is_empty_shadow_page(u64 *spt)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800448{
Avi Kivity139bdb22007-01-05 16:36:50 -0800449 u64 *pos;
450 u64 *end;
451
Avi Kivity47ad8e62007-05-06 15:50:58 +0300452 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
Avi Kivity139bdb22007-01-05 16:36:50 -0800453 if (*pos != 0) {
454 printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__,
455 pos, *pos);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800456 return 0;
Avi Kivity139bdb22007-01-05 16:36:50 -0800457 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800458 return 1;
459}
Yaozu Dongd6c69ee2007-04-25 14:17:25 +0800460#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800461
Avi Kivity90cb0522007-07-17 13:04:56 +0300462static void kvm_mmu_free_page(struct kvm *kvm,
Avi Kivity4b02d6d2007-05-06 15:36:30 +0300463 struct kvm_mmu_page *page_head)
Avi Kivity260746c2007-01-05 16:36:49 -0800464{
Avi Kivity47ad8e62007-05-06 15:50:58 +0300465 ASSERT(is_empty_shadow_page(page_head->spt));
Avi Kivityd3d25b02007-05-30 12:34:53 +0300466 list_del(&page_head->link);
Avi Kivityc1158e62007-07-20 08:18:27 +0300467 __free_page(virt_to_page(page_head->spt));
Avi Kivity90cb0522007-07-17 13:04:56 +0300468 kfree(page_head);
469 ++kvm->n_free_mmu_pages;
Avi Kivity260746c2007-01-05 16:36:49 -0800470}
471
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800472static unsigned kvm_page_table_hashfn(gfn_t gfn)
473{
474 return gfn;
475}
476
Avi Kivity25c0de22007-01-05 16:36:42 -0800477static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
478 u64 *parent_pte)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800479{
480 struct kvm_mmu_page *page;
481
Avi Kivityd3d25b02007-05-30 12:34:53 +0300482 if (!vcpu->kvm->n_free_mmu_pages)
Avi Kivity25c0de22007-01-05 16:36:42 -0800483 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800484
Avi Kivityd3d25b02007-05-30 12:34:53 +0300485 page = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache,
486 sizeof *page);
487 page->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
488 set_page_private(virt_to_page(page->spt), (unsigned long)page);
489 list_add(&page->link, &vcpu->kvm->active_mmu_pages);
Avi Kivity47ad8e62007-05-06 15:50:58 +0300490 ASSERT(is_empty_shadow_page(page->spt));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800491 page->slot_bitmap = 0;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800492 page->multimapped = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800493 page->parent_pte = parent_pte;
Avi Kivityebeace82007-01-05 16:36:47 -0800494 --vcpu->kvm->n_free_mmu_pages;
Avi Kivity25c0de22007-01-05 16:36:42 -0800495 return page;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800496}
497
Avi Kivity714b93d2007-01-05 16:36:53 -0800498static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
499 struct kvm_mmu_page *page, u64 *parent_pte)
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800500{
501 struct kvm_pte_chain *pte_chain;
502 struct hlist_node *node;
503 int i;
504
505 if (!parent_pte)
506 return;
507 if (!page->multimapped) {
508 u64 *old = page->parent_pte;
509
510 if (!old) {
511 page->parent_pte = parent_pte;
512 return;
513 }
514 page->multimapped = 1;
Avi Kivity714b93d2007-01-05 16:36:53 -0800515 pte_chain = mmu_alloc_pte_chain(vcpu);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800516 INIT_HLIST_HEAD(&page->parent_ptes);
517 hlist_add_head(&pte_chain->link, &page->parent_ptes);
518 pte_chain->parent_ptes[0] = old;
519 }
520 hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) {
521 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
522 continue;
523 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
524 if (!pte_chain->parent_ptes[i]) {
525 pte_chain->parent_ptes[i] = parent_pte;
526 return;
527 }
528 }
Avi Kivity714b93d2007-01-05 16:36:53 -0800529 pte_chain = mmu_alloc_pte_chain(vcpu);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800530 BUG_ON(!pte_chain);
531 hlist_add_head(&pte_chain->link, &page->parent_ptes);
532 pte_chain->parent_ptes[0] = parent_pte;
533}
534
Avi Kivity90cb0522007-07-17 13:04:56 +0300535static void mmu_page_remove_parent_pte(struct kvm_mmu_page *page,
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800536 u64 *parent_pte)
537{
538 struct kvm_pte_chain *pte_chain;
539 struct hlist_node *node;
540 int i;
541
542 if (!page->multimapped) {
543 BUG_ON(page->parent_pte != parent_pte);
544 page->parent_pte = NULL;
545 return;
546 }
547 hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link)
548 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
549 if (!pte_chain->parent_ptes[i])
550 break;
551 if (pte_chain->parent_ptes[i] != parent_pte)
552 continue;
Avi Kivity697fe2e2007-01-05 16:36:46 -0800553 while (i + 1 < NR_PTE_CHAIN_ENTRIES
554 && pte_chain->parent_ptes[i + 1]) {
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800555 pte_chain->parent_ptes[i]
556 = pte_chain->parent_ptes[i + 1];
557 ++i;
558 }
559 pte_chain->parent_ptes[i] = NULL;
Avi Kivity697fe2e2007-01-05 16:36:46 -0800560 if (i == 0) {
561 hlist_del(&pte_chain->link);
Avi Kivity90cb0522007-07-17 13:04:56 +0300562 mmu_free_pte_chain(pte_chain);
Avi Kivity697fe2e2007-01-05 16:36:46 -0800563 if (hlist_empty(&page->parent_ptes)) {
564 page->multimapped = 0;
565 page->parent_pte = NULL;
566 }
567 }
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800568 return;
569 }
570 BUG();
571}
572
573static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm_vcpu *vcpu,
574 gfn_t gfn)
575{
576 unsigned index;
577 struct hlist_head *bucket;
578 struct kvm_mmu_page *page;
579 struct hlist_node *node;
580
581 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
582 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
583 bucket = &vcpu->kvm->mmu_page_hash[index];
584 hlist_for_each_entry(page, node, bucket, hash_link)
585 if (page->gfn == gfn && !page->role.metaphysical) {
586 pgprintk("%s: found role %x\n",
587 __FUNCTION__, page->role.word);
588 return page;
589 }
590 return NULL;
591}
592
593static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
594 gfn_t gfn,
595 gva_t gaddr,
596 unsigned level,
597 int metaphysical,
Avi Kivityd28c6cf2007-03-23 09:55:25 +0200598 unsigned hugepage_access,
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800599 u64 *parent_pte)
600{
601 union kvm_mmu_page_role role;
602 unsigned index;
603 unsigned quadrant;
604 struct hlist_head *bucket;
605 struct kvm_mmu_page *page;
606 struct hlist_node *node;
607
608 role.word = 0;
609 role.glevels = vcpu->mmu.root_level;
610 role.level = level;
611 role.metaphysical = metaphysical;
Avi Kivityd28c6cf2007-03-23 09:55:25 +0200612 role.hugepage_access = hugepage_access;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800613 if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) {
614 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
615 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
616 role.quadrant = quadrant;
617 }
618 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__,
619 gfn, role.word);
620 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
621 bucket = &vcpu->kvm->mmu_page_hash[index];
622 hlist_for_each_entry(page, node, bucket, hash_link)
623 if (page->gfn == gfn && page->role.word == role.word) {
Avi Kivity714b93d2007-01-05 16:36:53 -0800624 mmu_page_add_parent_pte(vcpu, page, parent_pte);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800625 pgprintk("%s: found\n", __FUNCTION__);
626 return page;
627 }
628 page = kvm_mmu_alloc_page(vcpu, parent_pte);
629 if (!page)
630 return page;
631 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word);
632 page->gfn = gfn;
633 page->role = role;
634 hlist_add_head(&page->hash_link, bucket);
Avi Kivity374cbac2007-01-05 16:36:43 -0800635 if (!metaphysical)
Avi Kivity714b93d2007-01-05 16:36:53 -0800636 rmap_write_protect(vcpu, gfn);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800637 return page;
638}
639
Avi Kivity90cb0522007-07-17 13:04:56 +0300640static void kvm_mmu_page_unlink_children(struct kvm *kvm,
Avi Kivitya4360362007-01-05 16:36:45 -0800641 struct kvm_mmu_page *page)
642{
Avi Kivity697fe2e2007-01-05 16:36:46 -0800643 unsigned i;
644 u64 *pt;
645 u64 ent;
646
Avi Kivity47ad8e62007-05-06 15:50:58 +0300647 pt = page->spt;
Avi Kivity697fe2e2007-01-05 16:36:46 -0800648
649 if (page->role.level == PT_PAGE_TABLE_LEVEL) {
650 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
651 if (pt[i] & PT_PRESENT_MASK)
Avi Kivity90cb0522007-07-17 13:04:56 +0300652 rmap_remove(&pt[i]);
Avi Kivity697fe2e2007-01-05 16:36:46 -0800653 pt[i] = 0;
654 }
Avi Kivity90cb0522007-07-17 13:04:56 +0300655 kvm_flush_remote_tlbs(kvm);
Avi Kivity697fe2e2007-01-05 16:36:46 -0800656 return;
657 }
658
659 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
660 ent = pt[i];
661
662 pt[i] = 0;
663 if (!(ent & PT_PRESENT_MASK))
664 continue;
665 ent &= PT64_BASE_ADDR_MASK;
Avi Kivity90cb0522007-07-17 13:04:56 +0300666 mmu_page_remove_parent_pte(page_header(ent), &pt[i]);
Avi Kivity697fe2e2007-01-05 16:36:46 -0800667 }
Avi Kivity90cb0522007-07-17 13:04:56 +0300668 kvm_flush_remote_tlbs(kvm);
Avi Kivitya4360362007-01-05 16:36:45 -0800669}
670
Avi Kivity90cb0522007-07-17 13:04:56 +0300671static void kvm_mmu_put_page(struct kvm_mmu_page *page,
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800672 u64 *parent_pte)
673{
Avi Kivity90cb0522007-07-17 13:04:56 +0300674 mmu_page_remove_parent_pte(page, parent_pte);
Avi Kivitya4360362007-01-05 16:36:45 -0800675}
676
Avi Kivity90cb0522007-07-17 13:04:56 +0300677static void kvm_mmu_zap_page(struct kvm *kvm,
Avi Kivitya4360362007-01-05 16:36:45 -0800678 struct kvm_mmu_page *page)
679{
680 u64 *parent_pte;
681
682 while (page->multimapped || page->parent_pte) {
683 if (!page->multimapped)
684 parent_pte = page->parent_pte;
685 else {
686 struct kvm_pte_chain *chain;
687
688 chain = container_of(page->parent_ptes.first,
689 struct kvm_pte_chain, link);
690 parent_pte = chain->parent_ptes[0];
691 }
Avi Kivity697fe2e2007-01-05 16:36:46 -0800692 BUG_ON(!parent_pte);
Avi Kivity90cb0522007-07-17 13:04:56 +0300693 kvm_mmu_put_page(page, parent_pte);
Avi Kivitye663ee62007-05-31 15:46:04 +0300694 set_shadow_pte(parent_pte, 0);
Avi Kivitya4360362007-01-05 16:36:45 -0800695 }
Avi Kivity90cb0522007-07-17 13:04:56 +0300696 kvm_mmu_page_unlink_children(kvm, page);
Avi Kivity3bb65a22007-01-05 16:36:51 -0800697 if (!page->root_count) {
698 hlist_del(&page->hash_link);
Avi Kivity90cb0522007-07-17 13:04:56 +0300699 kvm_mmu_free_page(kvm, page);
Avi Kivity36868f72007-03-26 19:31:52 +0200700 } else
Avi Kivity90cb0522007-07-17 13:04:56 +0300701 list_move(&page->link, &kvm->active_mmu_pages);
Avi Kivitya4360362007-01-05 16:36:45 -0800702}
703
704static int kvm_mmu_unprotect_page(struct kvm_vcpu *vcpu, gfn_t gfn)
705{
706 unsigned index;
707 struct hlist_head *bucket;
708 struct kvm_mmu_page *page;
709 struct hlist_node *node, *n;
710 int r;
711
712 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
713 r = 0;
714 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
715 bucket = &vcpu->kvm->mmu_page_hash[index];
716 hlist_for_each_entry_safe(page, node, n, bucket, hash_link)
717 if (page->gfn == gfn && !page->role.metaphysical) {
Avi Kivity697fe2e2007-01-05 16:36:46 -0800718 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn,
719 page->role.word);
Avi Kivity90cb0522007-07-17 13:04:56 +0300720 kvm_mmu_zap_page(vcpu->kvm, page);
Avi Kivitya4360362007-01-05 16:36:45 -0800721 r = 1;
722 }
723 return r;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800724}
725
Avi Kivity97a0a012007-05-31 15:08:29 +0300726static void mmu_unshadow(struct kvm_vcpu *vcpu, gfn_t gfn)
727{
728 struct kvm_mmu_page *page;
729
730 while ((page = kvm_mmu_lookup_page(vcpu, gfn)) != NULL) {
731 pgprintk("%s: zap %lx %x\n",
732 __FUNCTION__, gfn, page->role.word);
Avi Kivity90cb0522007-07-17 13:04:56 +0300733 kvm_mmu_zap_page(vcpu->kvm, page);
Avi Kivity97a0a012007-05-31 15:08:29 +0300734 }
735}
736
Avi Kivity6aa8b732006-12-10 02:21:36 -0800737static void page_header_update_slot(struct kvm *kvm, void *pte, gpa_t gpa)
738{
739 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gpa >> PAGE_SHIFT));
740 struct kvm_mmu_page *page_head = page_header(__pa(pte));
741
742 __set_bit(slot, &page_head->slot_bitmap);
743}
744
745hpa_t safe_gpa_to_hpa(struct kvm_vcpu *vcpu, gpa_t gpa)
746{
747 hpa_t hpa = gpa_to_hpa(vcpu, gpa);
748
749 return is_error_hpa(hpa) ? bad_page_address | (gpa & ~PAGE_MASK): hpa;
750}
751
752hpa_t gpa_to_hpa(struct kvm_vcpu *vcpu, gpa_t gpa)
753{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800754 struct page *page;
755
756 ASSERT((gpa & HPA_ERR_MASK) == 0);
Avi Kivity954bbbc22007-03-30 14:02:32 +0300757 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
758 if (!page)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800759 return gpa | HPA_ERR_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800760 return ((hpa_t)page_to_pfn(page) << PAGE_SHIFT)
761 | (gpa & (PAGE_SIZE-1));
762}
763
764hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva)
765{
766 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
767
768 if (gpa == UNMAPPED_GVA)
769 return UNMAPPED_GVA;
770 return gpa_to_hpa(vcpu, gpa);
771}
772
Avi Kivity039576c2007-03-20 12:46:50 +0200773struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
774{
775 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
776
777 if (gpa == UNMAPPED_GVA)
778 return NULL;
779 return pfn_to_page(gpa_to_hpa(vcpu, gpa) >> PAGE_SHIFT);
780}
781
Avi Kivity6aa8b732006-12-10 02:21:36 -0800782static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
783{
784}
785
786static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p)
787{
788 int level = PT32E_ROOT_LEVEL;
789 hpa_t table_addr = vcpu->mmu.root_hpa;
790
791 for (; ; level--) {
792 u32 index = PT64_INDEX(v, level);
793 u64 *table;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800794 u64 pte;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800795
796 ASSERT(VALID_PAGE(table_addr));
797 table = __va(table_addr);
798
799 if (level == 1) {
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800800 pte = table[index];
801 if (is_present_pte(pte) && is_writeble_pte(pte))
802 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800803 mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT);
804 page_header_update_slot(vcpu->kvm, table, v);
805 table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK |
806 PT_USER_MASK;
Avi Kivity714b93d2007-01-05 16:36:53 -0800807 rmap_add(vcpu, &table[index]);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800808 return 0;
809 }
810
811 if (table[index] == 0) {
Avi Kivity25c0de22007-01-05 16:36:42 -0800812 struct kvm_mmu_page *new_table;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800813 gfn_t pseudo_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800814
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800815 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
816 >> PAGE_SHIFT;
817 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
818 v, level - 1,
Avi Kivityd28c6cf2007-03-23 09:55:25 +0200819 1, 0, &table[index]);
Avi Kivity25c0de22007-01-05 16:36:42 -0800820 if (!new_table) {
Avi Kivity6aa8b732006-12-10 02:21:36 -0800821 pgprintk("nonpaging_map: ENOMEM\n");
822 return -ENOMEM;
823 }
824
Avi Kivity47ad8e62007-05-06 15:50:58 +0300825 table[index] = __pa(new_table->spt) | PT_PRESENT_MASK
Avi Kivity25c0de22007-01-05 16:36:42 -0800826 | PT_WRITABLE_MASK | PT_USER_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800827 }
828 table_addr = table[index] & PT64_BASE_ADDR_MASK;
829 }
830}
831
Avi Kivity17ac10a2007-01-05 16:36:40 -0800832static void mmu_free_roots(struct kvm_vcpu *vcpu)
833{
834 int i;
Avi Kivity3bb65a22007-01-05 16:36:51 -0800835 struct kvm_mmu_page *page;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800836
Avi Kivity7b53aa52007-06-05 12:17:03 +0300837 if (!VALID_PAGE(vcpu->mmu.root_hpa))
838 return;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800839#ifdef CONFIG_X86_64
840 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
841 hpa_t root = vcpu->mmu.root_hpa;
842
Avi Kivity3bb65a22007-01-05 16:36:51 -0800843 page = page_header(root);
844 --page->root_count;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800845 vcpu->mmu.root_hpa = INVALID_PAGE;
846 return;
847 }
848#endif
849 for (i = 0; i < 4; ++i) {
850 hpa_t root = vcpu->mmu.pae_root[i];
851
Avi Kivity417726a2007-04-12 17:35:58 +0300852 if (root) {
Avi Kivity417726a2007-04-12 17:35:58 +0300853 root &= PT64_BASE_ADDR_MASK;
854 page = page_header(root);
855 --page->root_count;
856 }
Avi Kivity17ac10a2007-01-05 16:36:40 -0800857 vcpu->mmu.pae_root[i] = INVALID_PAGE;
858 }
859 vcpu->mmu.root_hpa = INVALID_PAGE;
860}
861
862static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
863{
864 int i;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800865 gfn_t root_gfn;
Avi Kivity3bb65a22007-01-05 16:36:51 -0800866 struct kvm_mmu_page *page;
867
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800868 root_gfn = vcpu->cr3 >> PAGE_SHIFT;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800869
870#ifdef CONFIG_X86_64
871 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
872 hpa_t root = vcpu->mmu.root_hpa;
873
874 ASSERT(!VALID_PAGE(root));
Ingo Molnar68a99f62007-01-05 16:36:59 -0800875 page = kvm_mmu_get_page(vcpu, root_gfn, 0,
Avi Kivityd28c6cf2007-03-23 09:55:25 +0200876 PT64_ROOT_LEVEL, 0, 0, NULL);
Avi Kivity47ad8e62007-05-06 15:50:58 +0300877 root = __pa(page->spt);
Avi Kivity3bb65a22007-01-05 16:36:51 -0800878 ++page->root_count;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800879 vcpu->mmu.root_hpa = root;
880 return;
881 }
882#endif
883 for (i = 0; i < 4; ++i) {
884 hpa_t root = vcpu->mmu.pae_root[i];
885
886 ASSERT(!VALID_PAGE(root));
Avi Kivity417726a2007-04-12 17:35:58 +0300887 if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) {
888 if (!is_present_pte(vcpu->pdptrs[i])) {
889 vcpu->mmu.pae_root[i] = 0;
890 continue;
891 }
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800892 root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT;
Avi Kivity417726a2007-04-12 17:35:58 +0300893 } else if (vcpu->mmu.root_level == 0)
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800894 root_gfn = 0;
Ingo Molnar68a99f62007-01-05 16:36:59 -0800895 page = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800896 PT32_ROOT_LEVEL, !is_paging(vcpu),
Avi Kivityd28c6cf2007-03-23 09:55:25 +0200897 0, NULL);
Avi Kivity47ad8e62007-05-06 15:50:58 +0300898 root = __pa(page->spt);
Avi Kivity3bb65a22007-01-05 16:36:51 -0800899 ++page->root_count;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800900 vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK;
901 }
902 vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root);
903}
904
Avi Kivity6aa8b732006-12-10 02:21:36 -0800905static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
906{
907 return vaddr;
908}
909
910static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
911 u32 error_code)
912{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800913 gpa_t addr = gva;
Avi Kivityebeace82007-01-05 16:36:47 -0800914 hpa_t paddr;
Avi Kivitye2dec932007-01-05 16:36:54 -0800915 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916
Avi Kivitye2dec932007-01-05 16:36:54 -0800917 r = mmu_topup_memory_caches(vcpu);
918 if (r)
919 return r;
Avi Kivity714b93d2007-01-05 16:36:53 -0800920
Avi Kivity6aa8b732006-12-10 02:21:36 -0800921 ASSERT(vcpu);
922 ASSERT(VALID_PAGE(vcpu->mmu.root_hpa));
923
Avi Kivity6aa8b732006-12-10 02:21:36 -0800924
Avi Kivityebeace82007-01-05 16:36:47 -0800925 paddr = gpa_to_hpa(vcpu , addr & PT64_BASE_ADDR_MASK);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800926
Avi Kivityebeace82007-01-05 16:36:47 -0800927 if (is_error_hpa(paddr))
928 return 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800929
Avi Kivityebeace82007-01-05 16:36:47 -0800930 return nonpaging_map(vcpu, addr & PAGE_MASK, paddr);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800931}
932
Avi Kivity6aa8b732006-12-10 02:21:36 -0800933static void nonpaging_free(struct kvm_vcpu *vcpu)
934{
Avi Kivity17ac10a2007-01-05 16:36:40 -0800935 mmu_free_roots(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800936}
937
938static int nonpaging_init_context(struct kvm_vcpu *vcpu)
939{
940 struct kvm_mmu *context = &vcpu->mmu;
941
942 context->new_cr3 = nonpaging_new_cr3;
943 context->page_fault = nonpaging_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800944 context->gva_to_gpa = nonpaging_gva_to_gpa;
945 context->free = nonpaging_free;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800946 context->root_level = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800947 context->shadow_root_level = PT32E_ROOT_LEVEL;
Avi Kivity17c3ba92007-06-04 15:58:30 +0300948 context->root_hpa = INVALID_PAGE;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800949 return 0;
950}
951
Avi Kivity6aa8b732006-12-10 02:21:36 -0800952static void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
953{
Avi Kivity1165f5f2007-04-19 17:27:43 +0300954 ++vcpu->stat.tlb_flush;
Christian Ehrhardtcbdd1be2007-09-09 15:41:59 +0300955 kvm_x86_ops->tlb_flush(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800956}
957
958static void paging_new_cr3(struct kvm_vcpu *vcpu)
959{
Avi Kivity374cbac2007-01-05 16:36:43 -0800960 pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800961 mmu_free_roots(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800962}
963
Avi Kivity6aa8b732006-12-10 02:21:36 -0800964static void inject_page_fault(struct kvm_vcpu *vcpu,
965 u64 addr,
966 u32 err_code)
967{
Christian Ehrhardtcbdd1be2007-09-09 15:41:59 +0300968 kvm_x86_ops->inject_page_fault(vcpu, addr, err_code);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969}
970
Avi Kivity6aa8b732006-12-10 02:21:36 -0800971static void paging_free(struct kvm_vcpu *vcpu)
972{
973 nonpaging_free(vcpu);
974}
975
976#define PTTYPE 64
977#include "paging_tmpl.h"
978#undef PTTYPE
979
980#define PTTYPE 32
981#include "paging_tmpl.h"
982#undef PTTYPE
983
Avi Kivity17ac10a2007-01-05 16:36:40 -0800984static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800985{
986 struct kvm_mmu *context = &vcpu->mmu;
987
988 ASSERT(is_pae(vcpu));
989 context->new_cr3 = paging_new_cr3;
990 context->page_fault = paging64_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800991 context->gva_to_gpa = paging64_gva_to_gpa;
992 context->free = paging_free;
Avi Kivity17ac10a2007-01-05 16:36:40 -0800993 context->root_level = level;
994 context->shadow_root_level = level;
Avi Kivity17c3ba92007-06-04 15:58:30 +0300995 context->root_hpa = INVALID_PAGE;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800996 return 0;
997}
998
Avi Kivity17ac10a2007-01-05 16:36:40 -0800999static int paging64_init_context(struct kvm_vcpu *vcpu)
1000{
1001 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1002}
1003
Avi Kivity6aa8b732006-12-10 02:21:36 -08001004static int paging32_init_context(struct kvm_vcpu *vcpu)
1005{
1006 struct kvm_mmu *context = &vcpu->mmu;
1007
1008 context->new_cr3 = paging_new_cr3;
1009 context->page_fault = paging32_page_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001010 context->gva_to_gpa = paging32_gva_to_gpa;
1011 context->free = paging_free;
1012 context->root_level = PT32_ROOT_LEVEL;
1013 context->shadow_root_level = PT32E_ROOT_LEVEL;
Avi Kivity17c3ba92007-06-04 15:58:30 +03001014 context->root_hpa = INVALID_PAGE;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001015 return 0;
1016}
1017
1018static int paging32E_init_context(struct kvm_vcpu *vcpu)
1019{
Avi Kivity17ac10a2007-01-05 16:36:40 -08001020 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001021}
1022
1023static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1024{
1025 ASSERT(vcpu);
1026 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1027
1028 if (!is_paging(vcpu))
1029 return nonpaging_init_context(vcpu);
Avi Kivitya9058ec2006-12-29 16:49:37 -08001030 else if (is_long_mode(vcpu))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001031 return paging64_init_context(vcpu);
1032 else if (is_pae(vcpu))
1033 return paging32E_init_context(vcpu);
1034 else
1035 return paging32_init_context(vcpu);
1036}
1037
1038static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1039{
1040 ASSERT(vcpu);
1041 if (VALID_PAGE(vcpu->mmu.root_hpa)) {
1042 vcpu->mmu.free(vcpu);
1043 vcpu->mmu.root_hpa = INVALID_PAGE;
1044 }
1045}
1046
1047int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1048{
Avi Kivity17c3ba92007-06-04 15:58:30 +03001049 destroy_kvm_mmu(vcpu);
1050 return init_kvm_mmu(vcpu);
1051}
Eddie Dong8668a3c2007-10-10 14:26:45 +08001052EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
Avi Kivity17c3ba92007-06-04 15:58:30 +03001053
1054int kvm_mmu_load(struct kvm_vcpu *vcpu)
1055{
Avi Kivity714b93d2007-01-05 16:36:53 -08001056 int r;
1057
Shaohua Li11ec2802007-07-23 14:51:37 +08001058 mutex_lock(&vcpu->kvm->lock);
Avi Kivitye2dec932007-01-05 16:36:54 -08001059 r = mmu_topup_memory_caches(vcpu);
Avi Kivity17c3ba92007-06-04 15:58:30 +03001060 if (r)
1061 goto out;
1062 mmu_alloc_roots(vcpu);
Christian Ehrhardtcbdd1be2007-09-09 15:41:59 +03001063 kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa);
Avi Kivity17c3ba92007-06-04 15:58:30 +03001064 kvm_mmu_flush_tlb(vcpu);
Avi Kivity714b93d2007-01-05 16:36:53 -08001065out:
Shaohua Li11ec2802007-07-23 14:51:37 +08001066 mutex_unlock(&vcpu->kvm->lock);
Avi Kivity714b93d2007-01-05 16:36:53 -08001067 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001068}
Avi Kivity17c3ba92007-06-04 15:58:30 +03001069EXPORT_SYMBOL_GPL(kvm_mmu_load);
1070
1071void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1072{
1073 mmu_free_roots(vcpu);
1074}
Avi Kivity6aa8b732006-12-10 02:21:36 -08001075
Avi Kivity09072da2007-05-01 14:16:52 +03001076static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
Avi Kivityac1b7142007-03-08 17:13:32 +02001077 struct kvm_mmu_page *page,
1078 u64 *spte)
1079{
1080 u64 pte;
1081 struct kvm_mmu_page *child;
1082
1083 pte = *spte;
1084 if (is_present_pte(pte)) {
1085 if (page->role.level == PT_PAGE_TABLE_LEVEL)
Avi Kivity90cb0522007-07-17 13:04:56 +03001086 rmap_remove(spte);
Avi Kivityac1b7142007-03-08 17:13:32 +02001087 else {
1088 child = page_header(pte & PT64_BASE_ADDR_MASK);
Avi Kivity90cb0522007-07-17 13:04:56 +03001089 mmu_page_remove_parent_pte(child, spte);
Avi Kivityac1b7142007-03-08 17:13:32 +02001090 }
1091 }
Izik Eidus7f2145a2007-09-23 12:30:19 +02001092 set_shadow_pte(spte, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +03001093 kvm_flush_remote_tlbs(vcpu->kvm);
Avi Kivityac1b7142007-03-08 17:13:32 +02001094}
1095
Avi Kivity00284252007-05-01 16:53:31 +03001096static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1097 struct kvm_mmu_page *page,
1098 u64 *spte,
1099 const void *new, int bytes)
1100{
1101 if (page->role.level != PT_PAGE_TABLE_LEVEL)
1102 return;
1103
1104 if (page->role.glevels == PT32_ROOT_LEVEL)
1105 paging32_update_pte(vcpu, page, spte, new, bytes);
1106 else
1107 paging64_update_pte(vcpu, page, spte, new, bytes);
1108}
1109
Avi Kivity09072da2007-05-01 14:16:52 +03001110void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
Shaohua Life551882007-07-23 14:51:39 +08001111 const u8 *new, int bytes)
Avi Kivityda4a00f2007-01-05 16:36:44 -08001112{
Avi Kivity9b7a0322007-01-05 16:36:45 -08001113 gfn_t gfn = gpa >> PAGE_SHIFT;
1114 struct kvm_mmu_page *page;
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001115 struct hlist_node *node, *n;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001116 struct hlist_head *bucket;
1117 unsigned index;
1118 u64 *spte;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001119 unsigned offset = offset_in_page(gpa);
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001120 unsigned pte_size;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001121 unsigned page_offset;
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001122 unsigned misaligned;
Avi Kivityfce06572007-05-01 16:44:05 +03001123 unsigned quadrant;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001124 int level;
Avi Kivity86a5ba02007-01-05 16:36:50 -08001125 int flooded = 0;
Avi Kivityac1b7142007-03-08 17:13:32 +02001126 int npte;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001127
Avi Kivityda4a00f2007-01-05 16:36:44 -08001128 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes);
Avi Kivity86a5ba02007-01-05 16:36:50 -08001129 if (gfn == vcpu->last_pt_write_gfn) {
1130 ++vcpu->last_pt_write_count;
1131 if (vcpu->last_pt_write_count >= 3)
1132 flooded = 1;
1133 } else {
1134 vcpu->last_pt_write_gfn = gfn;
1135 vcpu->last_pt_write_count = 1;
1136 }
Avi Kivity9b7a0322007-01-05 16:36:45 -08001137 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
1138 bucket = &vcpu->kvm->mmu_page_hash[index];
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001139 hlist_for_each_entry_safe(page, node, n, bucket, hash_link) {
Avi Kivity9b7a0322007-01-05 16:36:45 -08001140 if (page->gfn != gfn || page->role.metaphysical)
1141 continue;
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001142 pte_size = page->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1143 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
Avi Kivitye925c5b2007-04-30 14:47:02 +03001144 misaligned |= bytes < 4;
Avi Kivity86a5ba02007-01-05 16:36:50 -08001145 if (misaligned || flooded) {
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001146 /*
1147 * Misaligned accesses are too much trouble to fix
1148 * up; also, they usually indicate a page is not used
1149 * as a page table.
Avi Kivity86a5ba02007-01-05 16:36:50 -08001150 *
1151 * If we're seeing too many writes to a page,
1152 * it may no longer be a page table, or we may be
1153 * forking, in which case it is better to unmap the
1154 * page.
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001155 */
1156 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1157 gpa, bytes, page->role.word);
Avi Kivity90cb0522007-07-17 13:04:56 +03001158 kvm_mmu_zap_page(vcpu->kvm, page);
Avi Kivity0e7bc4b2007-01-05 16:36:48 -08001159 continue;
1160 }
Avi Kivity9b7a0322007-01-05 16:36:45 -08001161 page_offset = offset;
1162 level = page->role.level;
Avi Kivityac1b7142007-03-08 17:13:32 +02001163 npte = 1;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001164 if (page->role.glevels == PT32_ROOT_LEVEL) {
Avi Kivityac1b7142007-03-08 17:13:32 +02001165 page_offset <<= 1; /* 32->64 */
1166 /*
1167 * A 32-bit pde maps 4MB while the shadow pdes map
1168 * only 2MB. So we need to double the offset again
1169 * and zap two pdes instead of one.
1170 */
1171 if (level == PT32_ROOT_LEVEL) {
Avi Kivity6b8d0f92007-04-18 11:18:18 +03001172 page_offset &= ~7; /* kill rounding error */
Avi Kivityac1b7142007-03-08 17:13:32 +02001173 page_offset <<= 1;
1174 npte = 2;
1175 }
Avi Kivityfce06572007-05-01 16:44:05 +03001176 quadrant = page_offset >> PAGE_SHIFT;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001177 page_offset &= ~PAGE_MASK;
Avi Kivityfce06572007-05-01 16:44:05 +03001178 if (quadrant != page->role.quadrant)
1179 continue;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001180 }
Avi Kivity47ad8e62007-05-06 15:50:58 +03001181 spte = &page->spt[page_offset / sizeof(*spte)];
Avi Kivityac1b7142007-03-08 17:13:32 +02001182 while (npte--) {
Avi Kivity09072da2007-05-01 14:16:52 +03001183 mmu_pte_write_zap_pte(vcpu, page, spte);
Avi Kivity00284252007-05-01 16:53:31 +03001184 mmu_pte_write_new_pte(vcpu, page, spte, new, bytes);
Avi Kivityac1b7142007-03-08 17:13:32 +02001185 ++spte;
Avi Kivity9b7a0322007-01-05 16:36:45 -08001186 }
Avi Kivity9b7a0322007-01-05 16:36:45 -08001187 }
Avi Kivityda4a00f2007-01-05 16:36:44 -08001188}
1189
Avi Kivitya4360362007-01-05 16:36:45 -08001190int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1191{
1192 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
1193
1194 return kvm_mmu_unprotect_page(vcpu, gpa >> PAGE_SHIFT);
1195}
1196
Avi Kivity22d95b12007-09-14 20:26:06 +03001197void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
Avi Kivityebeace82007-01-05 16:36:47 -08001198{
1199 while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) {
1200 struct kvm_mmu_page *page;
1201
1202 page = container_of(vcpu->kvm->active_mmu_pages.prev,
1203 struct kvm_mmu_page, link);
Avi Kivity90cb0522007-07-17 13:04:56 +03001204 kvm_mmu_zap_page(vcpu->kvm, page);
Avi Kivityebeace82007-01-05 16:36:47 -08001205 }
1206}
Avi Kivityebeace82007-01-05 16:36:47 -08001207
Avi Kivity6aa8b732006-12-10 02:21:36 -08001208static void free_mmu_pages(struct kvm_vcpu *vcpu)
1209{
Avi Kivityf51234c2007-01-05 16:36:52 -08001210 struct kvm_mmu_page *page;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001211
Avi Kivityf51234c2007-01-05 16:36:52 -08001212 while (!list_empty(&vcpu->kvm->active_mmu_pages)) {
1213 page = container_of(vcpu->kvm->active_mmu_pages.next,
1214 struct kvm_mmu_page, link);
Avi Kivity90cb0522007-07-17 13:04:56 +03001215 kvm_mmu_zap_page(vcpu->kvm, page);
Avi Kivityf51234c2007-01-05 16:36:52 -08001216 }
Avi Kivity17ac10a2007-01-05 16:36:40 -08001217 free_page((unsigned long)vcpu->mmu.pae_root);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001218}
1219
1220static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
1221{
Avi Kivity17ac10a2007-01-05 16:36:40 -08001222 struct page *page;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001223 int i;
1224
1225 ASSERT(vcpu);
1226
Avi Kivityd3d25b02007-05-30 12:34:53 +03001227 vcpu->kvm->n_free_mmu_pages = KVM_NUM_MMU_PAGES;
Avi Kivity17ac10a2007-01-05 16:36:40 -08001228
1229 /*
1230 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1231 * Therefore we need to allocate shadow page tables in the first
1232 * 4GB of memory, which happens to fit the DMA32 zone.
1233 */
1234 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
1235 if (!page)
1236 goto error_1;
1237 vcpu->mmu.pae_root = page_address(page);
1238 for (i = 0; i < 4; ++i)
1239 vcpu->mmu.pae_root[i] = INVALID_PAGE;
1240
Avi Kivity6aa8b732006-12-10 02:21:36 -08001241 return 0;
1242
1243error_1:
1244 free_mmu_pages(vcpu);
1245 return -ENOMEM;
1246}
1247
Ingo Molnar8018c272006-12-29 16:50:01 -08001248int kvm_mmu_create(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001249{
Avi Kivity6aa8b732006-12-10 02:21:36 -08001250 ASSERT(vcpu);
1251 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
Avi Kivity6aa8b732006-12-10 02:21:36 -08001252
Ingo Molnar8018c272006-12-29 16:50:01 -08001253 return alloc_mmu_pages(vcpu);
1254}
Avi Kivity6aa8b732006-12-10 02:21:36 -08001255
Ingo Molnar8018c272006-12-29 16:50:01 -08001256int kvm_mmu_setup(struct kvm_vcpu *vcpu)
1257{
1258 ASSERT(vcpu);
1259 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
Avi Kivity2c264952006-12-22 01:05:28 -08001260
Ingo Molnar8018c272006-12-29 16:50:01 -08001261 return init_kvm_mmu(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001262}
1263
1264void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
1265{
1266 ASSERT(vcpu);
1267
1268 destroy_kvm_mmu(vcpu);
1269 free_mmu_pages(vcpu);
Avi Kivity714b93d2007-01-05 16:36:53 -08001270 mmu_free_memory_caches(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001271}
1272
Avi Kivity90cb0522007-07-17 13:04:56 +03001273void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001274{
1275 struct kvm_mmu_page *page;
1276
1277 list_for_each_entry(page, &kvm->active_mmu_pages, link) {
1278 int i;
1279 u64 *pt;
1280
1281 if (!test_bit(slot, &page->slot_bitmap))
1282 continue;
1283
Avi Kivity47ad8e62007-05-06 15:50:58 +03001284 pt = page->spt;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001285 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
1286 /* avoid RMW */
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001287 if (pt[i] & PT_WRITABLE_MASK) {
Avi Kivity90cb0522007-07-17 13:04:56 +03001288 rmap_remove(&pt[i]);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001289 pt[i] &= ~PT_WRITABLE_MASK;
Avi Kivitycd4a4e52007-01-05 16:36:38 -08001290 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001291 }
1292}
Avi Kivity37a7d8b2007-01-05 16:36:56 -08001293
Avi Kivity90cb0522007-07-17 13:04:56 +03001294void kvm_mmu_zap_all(struct kvm *kvm)
Dor Laore0fa8262007-03-30 13:06:33 +03001295{
Avi Kivity90cb0522007-07-17 13:04:56 +03001296 struct kvm_mmu_page *page, *node;
Dor Laore0fa8262007-03-30 13:06:33 +03001297
Avi Kivity90cb0522007-07-17 13:04:56 +03001298 list_for_each_entry_safe(page, node, &kvm->active_mmu_pages, link)
1299 kvm_mmu_zap_page(kvm, page);
Dor Laore0fa8262007-03-30 13:06:33 +03001300
Avi Kivity90cb0522007-07-17 13:04:56 +03001301 kvm_flush_remote_tlbs(kvm);
Dor Laore0fa8262007-03-30 13:06:33 +03001302}
1303
Avi Kivityb5a33a72007-04-15 16:31:09 +03001304void kvm_mmu_module_exit(void)
1305{
1306 if (pte_chain_cache)
1307 kmem_cache_destroy(pte_chain_cache);
1308 if (rmap_desc_cache)
1309 kmem_cache_destroy(rmap_desc_cache);
Avi Kivityd3d25b02007-05-30 12:34:53 +03001310 if (mmu_page_header_cache)
1311 kmem_cache_destroy(mmu_page_header_cache);
Avi Kivityb5a33a72007-04-15 16:31:09 +03001312}
1313
1314int kvm_mmu_module_init(void)
1315{
1316 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
1317 sizeof(struct kvm_pte_chain),
Paul Mundt20c2df82007-07-20 10:11:58 +09001318 0, 0, NULL);
Avi Kivityb5a33a72007-04-15 16:31:09 +03001319 if (!pte_chain_cache)
1320 goto nomem;
1321 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
1322 sizeof(struct kvm_rmap_desc),
Paul Mundt20c2df82007-07-20 10:11:58 +09001323 0, 0, NULL);
Avi Kivityb5a33a72007-04-15 16:31:09 +03001324 if (!rmap_desc_cache)
1325 goto nomem;
1326
Avi Kivityd3d25b02007-05-30 12:34:53 +03001327 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
1328 sizeof(struct kvm_mmu_page),
Paul Mundt20c2df82007-07-20 10:11:58 +09001329 0, 0, NULL);
Avi Kivityd3d25b02007-05-30 12:34:53 +03001330 if (!mmu_page_header_cache)
1331 goto nomem;
1332
Avi Kivityb5a33a72007-04-15 16:31:09 +03001333 return 0;
1334
1335nomem:
1336 kvm_mmu_module_exit();
1337 return -ENOMEM;
1338}
1339
Avi Kivity37a7d8b2007-01-05 16:36:56 -08001340#ifdef AUDIT
1341
1342static const char *audit_msg;
1343
1344static gva_t canonicalize(gva_t gva)
1345{
1346#ifdef CONFIG_X86_64
1347 gva = (long long)(gva << 16) >> 16;
1348#endif
1349 return gva;
1350}
1351
1352static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
1353 gva_t va, int level)
1354{
1355 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
1356 int i;
1357 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
1358
1359 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
1360 u64 ent = pt[i];
1361
Adrian Bunk28076962007-04-28 21:20:48 +02001362 if (!(ent & PT_PRESENT_MASK))
Avi Kivity37a7d8b2007-01-05 16:36:56 -08001363 continue;
1364
1365 va = canonicalize(va);
1366 if (level > 1)
1367 audit_mappings_page(vcpu, ent, va, level - 1);
1368 else {
1369 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va);
1370 hpa_t hpa = gpa_to_hpa(vcpu, gpa);
1371
1372 if ((ent & PT_PRESENT_MASK)
1373 && (ent & PT64_BASE_ADDR_MASK) != hpa)
1374 printk(KERN_ERR "audit error: (%s) levels %d"
1375 " gva %lx gpa %llx hpa %llx ent %llx\n",
1376 audit_msg, vcpu->mmu.root_level,
1377 va, gpa, hpa, ent);
1378 }
1379 }
1380}
1381
1382static void audit_mappings(struct kvm_vcpu *vcpu)
1383{
Avi Kivity1ea252a2007-03-08 11:48:09 +02001384 unsigned i;
Avi Kivity37a7d8b2007-01-05 16:36:56 -08001385
1386 if (vcpu->mmu.root_level == 4)
1387 audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4);
1388 else
1389 for (i = 0; i < 4; ++i)
1390 if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK)
1391 audit_mappings_page(vcpu,
1392 vcpu->mmu.pae_root[i],
1393 i << 30,
1394 2);
1395}
1396
1397static int count_rmaps(struct kvm_vcpu *vcpu)
1398{
1399 int nmaps = 0;
1400 int i, j, k;
1401
1402 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1403 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
1404 struct kvm_rmap_desc *d;
1405
1406 for (j = 0; j < m->npages; ++j) {
1407 struct page *page = m->phys_mem[j];
1408
1409 if (!page->private)
1410 continue;
1411 if (!(page->private & 1)) {
1412 ++nmaps;
1413 continue;
1414 }
1415 d = (struct kvm_rmap_desc *)(page->private & ~1ul);
1416 while (d) {
1417 for (k = 0; k < RMAP_EXT; ++k)
1418 if (d->shadow_ptes[k])
1419 ++nmaps;
1420 else
1421 break;
1422 d = d->more;
1423 }
1424 }
1425 }
1426 return nmaps;
1427}
1428
1429static int count_writable_mappings(struct kvm_vcpu *vcpu)
1430{
1431 int nmaps = 0;
1432 struct kvm_mmu_page *page;
1433 int i;
1434
1435 list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) {
Avi Kivity47ad8e62007-05-06 15:50:58 +03001436 u64 *pt = page->spt;
Avi Kivity37a7d8b2007-01-05 16:36:56 -08001437
1438 if (page->role.level != PT_PAGE_TABLE_LEVEL)
1439 continue;
1440
1441 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1442 u64 ent = pt[i];
1443
1444 if (!(ent & PT_PRESENT_MASK))
1445 continue;
1446 if (!(ent & PT_WRITABLE_MASK))
1447 continue;
1448 ++nmaps;
1449 }
1450 }
1451 return nmaps;
1452}
1453
1454static void audit_rmap(struct kvm_vcpu *vcpu)
1455{
1456 int n_rmap = count_rmaps(vcpu);
1457 int n_actual = count_writable_mappings(vcpu);
1458
1459 if (n_rmap != n_actual)
1460 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
1461 __FUNCTION__, audit_msg, n_rmap, n_actual);
1462}
1463
1464static void audit_write_protection(struct kvm_vcpu *vcpu)
1465{
1466 struct kvm_mmu_page *page;
1467
1468 list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) {
1469 hfn_t hfn;
1470 struct page *pg;
1471
1472 if (page->role.metaphysical)
1473 continue;
1474
1475 hfn = gpa_to_hpa(vcpu, (gpa_t)page->gfn << PAGE_SHIFT)
1476 >> PAGE_SHIFT;
1477 pg = pfn_to_page(hfn);
1478 if (pg->private)
1479 printk(KERN_ERR "%s: (%s) shadow page has writable"
1480 " mappings: gfn %lx role %x\n",
1481 __FUNCTION__, audit_msg, page->gfn,
1482 page->role.word);
1483 }
1484}
1485
1486static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
1487{
1488 int olddbg = dbg;
1489
1490 dbg = 0;
1491 audit_msg = msg;
1492 audit_rmap(vcpu);
1493 audit_write_protection(vcpu);
1494 audit_mappings(vcpu);
1495 dbg = olddbg;
1496}
1497
1498#endif