blob: fc72bad878eda6568b4ab82e9e4f5246f59d69dc [file] [log] [blame]
Carsten Otte043405e2007-10-10 17:16:19 +02001#/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This header defines architecture specific interfaces, x86 version
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2. See
7 * the COPYING file in the top-level directory.
8 *
9 */
10
Avi Kivityedf88412007-12-16 11:02:48 +020011#ifndef ASM_KVM_HOST_H
12#define ASM_KVM_HOST_H
Carsten Otte043405e2007-10-10 17:16:19 +020013
Zhang Xiantao34c16ee2007-10-20 15:34:38 +080014#include <linux/types.h>
15#include <linux/mm.h>
16
17#include <linux/kvm.h>
18#include <linux/kvm_para.h>
Avi Kivityedf88412007-12-16 11:02:48 +020019#include <linux/kvm_types.h>
Zhang Xiantao34c16ee2007-10-20 15:34:38 +080020
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +020021#include <asm/pvclock-abi.h>
Hollis Blancharde01a1b52007-12-03 15:30:25 -060022#include <asm/desc.h>
23
Avi Kivity69a9f692008-03-21 12:38:23 +020024#define KVM_MAX_VCPUS 16
25#define KVM_MEMORY_SLOTS 32
26/* memory slots that does not exposed to userspace */
27#define KVM_PRIVATE_MEM_SLOTS 4
28
29#define KVM_PIO_PAGE_OFFSET 1
30
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080031#define CR3_PAE_RESERVED_BITS ((X86_CR3_PWT | X86_CR3_PCD) - 1)
32#define CR3_NONPAE_RESERVED_BITS ((PAGE_SIZE-1) & ~(X86_CR3_PWT | X86_CR3_PCD))
Joe Perches7d76b4d2008-03-23 01:02:34 -070033#define CR3_L_MODE_RESERVED_BITS (CR3_NONPAE_RESERVED_BITS | \
34 0xFFFFFF0000000000ULL)
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080035
Joe Perches7d76b4d2008-03-23 01:02:34 -070036#define KVM_GUEST_CR0_MASK \
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080037 (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE \
38 | X86_CR0_NW | X86_CR0_CD)
Joe Perches7d76b4d2008-03-23 01:02:34 -070039#define KVM_VM_CR0_ALWAYS_ON \
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080040 (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE | X86_CR0_TS \
41 | X86_CR0_MP)
Joe Perches7d76b4d2008-03-23 01:02:34 -070042#define KVM_GUEST_CR4_MASK \
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080043 (X86_CR4_VME | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_PGE | X86_CR4_VMXE)
44#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
45#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
46
47#define INVALID_PAGE (~(hpa_t)0)
48#define UNMAPPED_GVA (~(gpa_t)0)
49
Marcelo Tosatti05da4552008-02-23 11:44:30 -030050/* shadow tables are PAE even on non-PAE hosts */
51#define KVM_HPAGE_SHIFT 21
52#define KVM_HPAGE_SIZE (1UL << KVM_HPAGE_SHIFT)
53#define KVM_HPAGE_MASK (~(KVM_HPAGE_SIZE - 1))
54
55#define KVM_PAGES_PER_HPAGE (KVM_HPAGE_SIZE / PAGE_SIZE)
56
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080057#define DE_VECTOR 0
58#define UD_VECTOR 6
59#define NM_VECTOR 7
60#define DF_VECTOR 8
61#define TS_VECTOR 10
62#define NP_VECTOR 11
63#define SS_VECTOR 12
64#define GP_VECTOR 13
65#define PF_VECTOR 14
Joerg Roedel53371b52008-04-09 14:15:30 +020066#define MC_VECTOR 18
Zhang Xiantaocd6e8f82007-11-19 14:33:37 +080067
68#define SELECTOR_TI_MASK (1 << 2)
69#define SELECTOR_RPL_MASK 0x03
70
71#define IOPL_SHIFT 12
72
Zhang Xiantaod69fb812007-12-14 09:54:20 +080073#define KVM_ALIAS_SLOTS 4
74
Zhang Xiantaod657a982007-12-14 09:41:22 +080075#define KVM_PERMILLE_MMU_PAGES 20
76#define KVM_MIN_ALLOC_MMU_PAGES 64
Dong, Eddie1ae0a132008-01-07 13:20:25 +020077#define KVM_MMU_HASH_SHIFT 10
78#define KVM_NUM_MMU_PAGES (1 << KVM_MMU_HASH_SHIFT)
Zhang Xiantaod657a982007-12-14 09:41:22 +080079#define KVM_MIN_FREE_MMU_PAGES 5
80#define KVM_REFILL_PAGES 25
81#define KVM_MAX_CPUID_ENTRIES 40
Avi Kivity9ba075a2008-05-26 20:06:35 +030082#define KVM_NR_VAR_MTRR 8
Zhang Xiantaod657a982007-12-14 09:41:22 +080083
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080084extern spinlock_t kvm_lock;
85extern struct list_head vm_list;
86
Zhang Xiantaod657a982007-12-14 09:41:22 +080087struct kvm_vcpu;
88struct kvm;
89
Zhang Xiantao2b3ccfa2007-11-19 14:56:05 +080090enum {
91 VCPU_REGS_RAX = 0,
92 VCPU_REGS_RCX = 1,
93 VCPU_REGS_RDX = 2,
94 VCPU_REGS_RBX = 3,
95 VCPU_REGS_RSP = 4,
96 VCPU_REGS_RBP = 5,
97 VCPU_REGS_RSI = 6,
98 VCPU_REGS_RDI = 7,
99#ifdef CONFIG_X86_64
100 VCPU_REGS_R8 = 8,
101 VCPU_REGS_R9 = 9,
102 VCPU_REGS_R10 = 10,
103 VCPU_REGS_R11 = 11,
104 VCPU_REGS_R12 = 12,
105 VCPU_REGS_R13 = 13,
106 VCPU_REGS_R14 = 14,
107 VCPU_REGS_R15 = 15,
108#endif
109 NR_VCPU_REGS
110};
111
112enum {
Zhang Xiantao2b3ccfa2007-11-19 14:56:05 +0800113 VCPU_SREG_ES,
Avi Kivity81609e32008-05-27 16:26:01 +0300114 VCPU_SREG_CS,
115 VCPU_SREG_SS,
116 VCPU_SREG_DS,
Zhang Xiantao2b3ccfa2007-11-19 14:56:05 +0800117 VCPU_SREG_FS,
118 VCPU_SREG_GS,
Zhang Xiantao2b3ccfa2007-11-19 14:56:05 +0800119 VCPU_SREG_TR,
120 VCPU_SREG_LDTR,
121};
122
Avi Kivityedf88412007-12-16 11:02:48 +0200123#include <asm/kvm_x86_emulate.h>
Zhang Xiantao2b3ccfa2007-11-19 14:56:05 +0800124
Zhang Xiantaod657a982007-12-14 09:41:22 +0800125#define KVM_NR_MEM_OBJS 40
126
Avi Kivity69a9f692008-03-21 12:38:23 +0200127struct kvm_guest_debug {
128 int enabled;
129 unsigned long bp[4];
130 int singlestep;
131};
132
Zhang Xiantaod657a982007-12-14 09:41:22 +0800133/*
134 * We don't want allocation failures within the mmu code, so we preallocate
135 * enough memory for a single page fault in a cache.
136 */
137struct kvm_mmu_memory_cache {
138 int nobjs;
139 void *objects[KVM_NR_MEM_OBJS];
140};
141
142#define NR_PTE_CHAIN_ENTRIES 5
143
144struct kvm_pte_chain {
145 u64 *parent_ptes[NR_PTE_CHAIN_ENTRIES];
146 struct hlist_node link;
147};
148
149/*
150 * kvm_mmu_page_role, below, is defined as:
151 *
152 * bits 0:3 - total guest paging levels (2-4, or zero for real mode)
153 * bits 4:7 - page table level for this shadow (1-4)
154 * bits 8:9 - page table quadrant for 2-level guests
155 * bit 16 - "metaphysical" - gfn is not a real page (huge page/real mode)
156 * bits 17:19 - common access permissions for all ptes in this shadow page
157 */
158union kvm_mmu_page_role {
159 unsigned word;
160 struct {
Joe Perches7d76b4d2008-03-23 01:02:34 -0700161 unsigned glevels:4;
162 unsigned level:4;
163 unsigned quadrant:2;
164 unsigned pad_for_nice_hex_output:6;
165 unsigned metaphysical:1;
166 unsigned access:3;
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500167 unsigned invalid:1;
Zhang Xiantaod657a982007-12-14 09:41:22 +0800168 };
169};
170
171struct kvm_mmu_page {
172 struct list_head link;
173 struct hlist_node hash_link;
174
175 /*
176 * The following two entries are used to key the shadow page in the
177 * hash table.
178 */
179 gfn_t gfn;
180 union kvm_mmu_page_role role;
181
182 u64 *spt;
183 /* hold the gfn of each spte inside spt */
184 gfn_t *gfns;
185 unsigned long slot_bitmap; /* One bit set per slot which has memory
186 * in this shadow page.
187 */
188 int multimapped; /* More than one parent_pte? */
189 int root_count; /* Currently serving as active root */
190 union {
191 u64 *parent_pte; /* !multimapped */
192 struct hlist_head parent_ptes; /* multimapped, kvm_pte_chain */
193 };
194};
195
196/*
197 * x86 supports 3 paging modes (4-level 64-bit, 3-level 64-bit, and 2-level
198 * 32-bit). The kvm_mmu structure abstracts the details of the current mmu
199 * mode.
200 */
201struct kvm_mmu {
202 void (*new_cr3)(struct kvm_vcpu *vcpu);
203 int (*page_fault)(struct kvm_vcpu *vcpu, gva_t gva, u32 err);
204 void (*free)(struct kvm_vcpu *vcpu);
205 gpa_t (*gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t gva);
206 void (*prefetch_page)(struct kvm_vcpu *vcpu,
207 struct kvm_mmu_page *page);
208 hpa_t root_hpa;
209 int root_level;
210 int shadow_root_level;
211
212 u64 *pae_root;
213};
214
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800215struct kvm_vcpu_arch {
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800216 u64 host_tsc;
217 int interrupt_window_open;
218 unsigned long irq_summary; /* bit vector: 1 per word in irq_pending */
219 DECLARE_BITMAP(irq_pending, KVM_NR_INTERRUPTS);
220 unsigned long regs[NR_VCPU_REGS]; /* for rsp: vcpu_load_rsp_rip() */
221 unsigned long rip; /* needs vcpu_load_rsp_rip() */
222
223 unsigned long cr0;
224 unsigned long cr2;
225 unsigned long cr3;
226 unsigned long cr4;
227 unsigned long cr8;
228 u64 pdptrs[4]; /* pae */
229 u64 shadow_efer;
230 u64 apic_base;
231 struct kvm_lapic *apic; /* kernel irqchip context */
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800232 int mp_state;
233 int sipi_vector;
234 u64 ia32_misc_enable_msr;
Avi Kivityb209749f2007-10-22 16:50:39 +0200235 bool tpr_access_reporting;
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800236
237 struct kvm_mmu mmu;
238
239 struct kvm_mmu_memory_cache mmu_pte_chain_cache;
240 struct kvm_mmu_memory_cache mmu_rmap_desc_cache;
241 struct kvm_mmu_memory_cache mmu_page_cache;
242 struct kvm_mmu_memory_cache mmu_page_header_cache;
243
244 gfn_t last_pt_write_gfn;
245 int last_pt_write_count;
246 u64 *last_pte_updated;
Avi Kivity1b7fcd32008-05-15 13:51:35 +0300247 gfn_t last_pte_gfn;
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800248
Avi Kivityd7824ff2007-12-30 12:29:05 +0200249 struct {
Anthony Liguori35149e22008-04-02 14:46:56 -0500250 gfn_t gfn; /* presumed gfn during guest pte update */
251 pfn_t pfn; /* pfn corresponding to that gfn */
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300252 int largepage;
Avi Kivityd7824ff2007-12-30 12:29:05 +0200253 } update_pte;
254
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800255 struct i387_fxsave_struct host_fx_image;
256 struct i387_fxsave_struct guest_fx_image;
257
258 gva_t mmio_fault_cr2;
259 struct kvm_pio_request pio;
260 void *pio_data;
261
Avi Kivity298101d2007-11-25 13:41:11 +0200262 struct kvm_queued_exception {
263 bool pending;
264 bool has_error_code;
265 u8 nr;
266 u32 error_code;
267 } exception;
268
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800269 struct {
270 int active;
271 u8 save_iopl;
272 struct kvm_save_segment {
273 u16 selector;
274 unsigned long base;
275 u32 limit;
276 u32 ar;
277 } tr, es, ds, fs, gs;
278 } rmode;
279 int halt_request; /* real mode on Intel only */
280
281 int cpuid_nent;
Dan Kenigsberg07716712007-11-21 17:10:04 +0200282 struct kvm_cpuid_entry2 cpuid_entries[KVM_MAX_CPUID_ENTRIES];
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800283 /* emulate context */
284
285 struct x86_emulate_ctxt emulate_ctxt;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200286
287 gpa_t time;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200288 struct pvclock_vcpu_time_info hv_clock;
289 unsigned int hv_clock_tsc_khz;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200290 unsigned int time_offset;
291 struct page *time_page;
Sheng Yang3419ffc2008-05-15 09:52:48 +0800292
293 bool nmi_pending;
Avi Kivity9ba075a2008-05-26 20:06:35 +0300294
295 u64 mtrr[0x100];
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800296};
297
Zhang Xiantaod69fb812007-12-14 09:54:20 +0800298struct kvm_mem_alias {
299 gfn_t base_gfn;
300 unsigned long npages;
301 gfn_t target_gfn;
302};
303
304struct kvm_arch{
305 int naliases;
306 struct kvm_mem_alias aliases[KVM_ALIAS_SLOTS];
Zhang Xiantaof05e70a2007-12-14 10:01:48 +0800307
308 unsigned int n_free_mmu_pages;
309 unsigned int n_requested_mmu_pages;
310 unsigned int n_alloc_mmu_pages;
311 struct hlist_head mmu_page_hash[KVM_NUM_MMU_PAGES];
312 /*
313 * Hash table of struct kvm_mmu_page.
314 */
315 struct list_head active_mmu_pages;
Zhang Xiantaod7deeeb02007-12-14 10:17:34 +0800316 struct kvm_pic *vpic;
317 struct kvm_ioapic *vioapic;
Sheng Yang78376992008-01-28 05:10:22 +0800318 struct kvm_pit *vpit;
Zhang Xiantaobfc6d222007-12-14 10:20:16 +0800319
320 int round_robin_prev_vcpu;
321 unsigned int tss_addr;
322 struct page *apic_access_page;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200323
324 gpa_t wall_clock;
Sheng Yangb7ebfb02008-04-25 21:44:52 +0800325
326 struct page *ept_identity_pagetable;
327 bool ept_identity_pagetable_done;
Zhang Xiantaod69fb812007-12-14 09:54:20 +0800328};
329
Zhang Xiantao07114562007-12-14 10:23:23 +0800330struct kvm_vm_stat {
331 u32 mmu_shadow_zapped;
332 u32 mmu_pte_write;
333 u32 mmu_pte_updated;
334 u32 mmu_pde_zapped;
335 u32 mmu_flooded;
336 u32 mmu_recycled;
Avi Kivitydfc5aa02007-12-18 19:47:18 +0200337 u32 mmu_cache_miss;
Zhang Xiantao07114562007-12-14 10:23:23 +0800338 u32 remote_tlb_flush;
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300339 u32 lpages;
Zhang Xiantao07114562007-12-14 10:23:23 +0800340};
341
Zhang Xiantao77b4c252007-12-14 09:49:26 +0800342struct kvm_vcpu_stat {
343 u32 pf_fixed;
344 u32 pf_guest;
345 u32 tlb_flush;
346 u32 invlpg;
347
348 u32 exits;
349 u32 io_exits;
350 u32 mmio_exits;
351 u32 signal_exits;
352 u32 irq_window_exits;
Sheng Yangf08864b2008-05-15 18:23:25 +0800353 u32 nmi_window_exits;
Zhang Xiantao77b4c252007-12-14 09:49:26 +0800354 u32 halt_exits;
355 u32 halt_wakeup;
356 u32 request_irq_exits;
357 u32 irq_exits;
358 u32 host_state_reload;
359 u32 efer_reload;
360 u32 fpu_reload;
361 u32 insn_emulation;
362 u32 insn_emulation_fail;
Amit Shahf11c3a82008-02-21 01:00:30 +0530363 u32 hypercalls;
Zhang Xiantao77b4c252007-12-14 09:49:26 +0800364};
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800365
Hollis Blancharde01a1b52007-12-03 15:30:25 -0600366struct descriptor_table {
367 u16 limit;
368 unsigned long base;
369} __attribute__((packed));
370
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800371struct kvm_x86_ops {
372 int (*cpu_has_kvm_support)(void); /* __init */
373 int (*disabled_by_bios)(void); /* __init */
374 void (*hardware_enable)(void *dummy); /* __init */
375 void (*hardware_disable)(void *dummy);
376 void (*check_processor_compatibility)(void *rtn);
377 int (*hardware_setup)(void); /* __init */
378 void (*hardware_unsetup)(void); /* __exit */
Avi Kivity774ead32007-12-26 13:57:04 +0200379 bool (*cpu_has_accelerated_tpr)(void);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800380
381 /* Create, but do not attach this VCPU */
382 struct kvm_vcpu *(*vcpu_create)(struct kvm *kvm, unsigned id);
383 void (*vcpu_free)(struct kvm_vcpu *vcpu);
384 int (*vcpu_reset)(struct kvm_vcpu *vcpu);
385
386 void (*prepare_guest_switch)(struct kvm_vcpu *vcpu);
387 void (*vcpu_load)(struct kvm_vcpu *vcpu, int cpu);
388 void (*vcpu_put)(struct kvm_vcpu *vcpu);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800389
390 int (*set_guest_debug)(struct kvm_vcpu *vcpu,
391 struct kvm_debug_guest *dbg);
392 void (*guest_debug_pre)(struct kvm_vcpu *vcpu);
393 int (*get_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata);
394 int (*set_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
395 u64 (*get_segment_base)(struct kvm_vcpu *vcpu, int seg);
396 void (*get_segment)(struct kvm_vcpu *vcpu,
397 struct kvm_segment *var, int seg);
Izik Eidus2e4d2652008-03-24 19:38:34 +0200398 int (*get_cpl)(struct kvm_vcpu *vcpu);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800399 void (*set_segment)(struct kvm_vcpu *vcpu,
400 struct kvm_segment *var, int seg);
401 void (*get_cs_db_l_bits)(struct kvm_vcpu *vcpu, int *db, int *l);
402 void (*decache_cr4_guest_bits)(struct kvm_vcpu *vcpu);
403 void (*set_cr0)(struct kvm_vcpu *vcpu, unsigned long cr0);
404 void (*set_cr3)(struct kvm_vcpu *vcpu, unsigned long cr3);
405 void (*set_cr4)(struct kvm_vcpu *vcpu, unsigned long cr4);
406 void (*set_efer)(struct kvm_vcpu *vcpu, u64 efer);
407 void (*get_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
408 void (*set_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
409 void (*get_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
410 void (*set_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
411 unsigned long (*get_dr)(struct kvm_vcpu *vcpu, int dr);
412 void (*set_dr)(struct kvm_vcpu *vcpu, int dr, unsigned long value,
413 int *exception);
414 void (*cache_regs)(struct kvm_vcpu *vcpu);
415 void (*decache_regs)(struct kvm_vcpu *vcpu);
416 unsigned long (*get_rflags)(struct kvm_vcpu *vcpu);
417 void (*set_rflags)(struct kvm_vcpu *vcpu, unsigned long rflags);
418
419 void (*tlb_flush)(struct kvm_vcpu *vcpu);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800420
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800421 void (*run)(struct kvm_vcpu *vcpu, struct kvm_run *run);
422 int (*handle_exit)(struct kvm_run *run, struct kvm_vcpu *vcpu);
423 void (*skip_emulated_instruction)(struct kvm_vcpu *vcpu);
424 void (*patch_hypercall)(struct kvm_vcpu *vcpu,
425 unsigned char *hypercall_addr);
426 int (*get_irq)(struct kvm_vcpu *vcpu);
427 void (*set_irq)(struct kvm_vcpu *vcpu, int vec);
Avi Kivity298101d2007-11-25 13:41:11 +0200428 void (*queue_exception)(struct kvm_vcpu *vcpu, unsigned nr,
429 bool has_error_code, u32 error_code);
430 bool (*exception_injected)(struct kvm_vcpu *vcpu);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800431 void (*inject_pending_irq)(struct kvm_vcpu *vcpu);
432 void (*inject_pending_vectors)(struct kvm_vcpu *vcpu,
433 struct kvm_run *run);
434
435 int (*set_tss_addr)(struct kvm *kvm, unsigned int addr);
Sheng Yang67253af2008-04-25 10:20:22 +0800436 int (*get_tdp_level)(void);
Zhang Xiantaoea4a5ff2007-11-19 14:40:47 +0800437};
438
Zhang Xiantao97896d02007-11-14 20:09:30 +0800439extern struct kvm_x86_ops *kvm_x86_ops;
440
Zhang Xiantao54f15852007-11-19 15:24:28 +0800441int kvm_mmu_module_init(void);
442void kvm_mmu_module_exit(void);
443
444void kvm_mmu_destroy(struct kvm_vcpu *vcpu);
445int kvm_mmu_create(struct kvm_vcpu *vcpu);
446int kvm_mmu_setup(struct kvm_vcpu *vcpu);
447void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte);
Sheng Yang7b523452008-04-25 21:13:50 +0800448void kvm_mmu_set_base_ptes(u64 base_pte);
449void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
450 u64 dirty_mask, u64 nx_mask, u64 x_mask);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800451
452int kvm_mmu_reset_context(struct kvm_vcpu *vcpu);
453void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot);
454void kvm_mmu_zap_all(struct kvm *kvm);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800455unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800456void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages);
457
Joerg Roedelcc4b6872008-02-07 13:47:43 +0100458int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
459
Marcelo Tosatti3200f402008-03-29 20:17:59 -0300460int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
Avi Kivity9f811282008-03-02 14:06:05 +0200461 const void *val, int bytes);
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -0500462int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
463 gpa_t addr, unsigned long *ret);
464
465extern bool tdp_enabled;
Avi Kivity9f811282008-03-02 14:06:05 +0200466
Zhang Xiantao54f15852007-11-19 15:24:28 +0800467enum emulation_result {
468 EMULATE_DONE, /* no further processing */
469 EMULATE_DO_MMIO, /* kvm_run filled with mmio request */
470 EMULATE_FAIL, /* can't emulate this instruction */
471};
472
Sheng Yang571008d2008-01-02 14:49:22 +0800473#define EMULTYPE_NO_DECODE (1 << 0)
474#define EMULTYPE_TRAP_UD (1 << 1)
Zhang Xiantao54f15852007-11-19 15:24:28 +0800475int emulate_instruction(struct kvm_vcpu *vcpu, struct kvm_run *run,
Sheng Yang571008d2008-01-02 14:49:22 +0800476 unsigned long cr2, u16 error_code, int emulation_type);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800477void kvm_report_emulation_failure(struct kvm_vcpu *cvpu, const char *context);
478void realmode_lgdt(struct kvm_vcpu *vcpu, u16 size, unsigned long address);
479void realmode_lidt(struct kvm_vcpu *vcpu, u16 size, unsigned long address);
480void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
481 unsigned long *rflags);
482
483unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr);
484void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long value,
485 unsigned long *rflags);
Joerg Roedelf2b4b7d2008-01-31 14:57:37 +0100486void kvm_enable_efer_bits(u64);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800487int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *data);
488int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
489
490struct x86_emulate_ctxt;
491
492int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
493 int size, unsigned port);
494int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
495 int size, unsigned long count, int down,
496 gva_t address, int rep, unsigned port);
497void kvm_emulate_cpuid(struct kvm_vcpu *vcpu);
498int kvm_emulate_halt(struct kvm_vcpu *vcpu);
499int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address);
500int emulate_clts(struct kvm_vcpu *vcpu);
501int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
502 unsigned long *dest);
503int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
504 unsigned long value);
505
Izik Eidus37817f22008-03-24 23:14:53 +0200506int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason);
507
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200508void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
Joerg Roedel9c204562008-04-01 16:44:56 +0200509void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
510void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
511void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8);
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200512unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu);
513void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800514void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l);
515
516int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
517int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data);
518
Avi Kivity298101d2007-11-25 13:41:11 +0200519void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr);
520void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200521void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long cr2,
522 u32 error_code);
Avi Kivity298101d2007-11-25 13:41:11 +0200523
Sheng Yang3419ffc2008-05-15 09:52:48 +0800524void kvm_inject_nmi(struct kvm_vcpu *vcpu);
525
Zhang Xiantao54f15852007-11-19 15:24:28 +0800526void fx_init(struct kvm_vcpu *vcpu);
527
528int emulator_read_std(unsigned long addr,
529 void *val,
530 unsigned int bytes,
531 struct kvm_vcpu *vcpu);
532int emulator_write_emulated(unsigned long addr,
533 const void *val,
534 unsigned int bytes,
535 struct kvm_vcpu *vcpu);
536
537unsigned long segment_base(u16 selector);
538
Avi Kivityd835dfe2007-11-21 02:57:59 +0200539void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu);
Zhang Xiantao54f15852007-11-19 15:24:28 +0800540void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
541 const u8 *new, int bytes);
542int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva);
543void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu);
544int kvm_mmu_load(struct kvm_vcpu *vcpu);
545void kvm_mmu_unload(struct kvm_vcpu *vcpu);
546
547int kvm_emulate_hypercall(struct kvm_vcpu *vcpu);
548
549int kvm_fix_hypercall(struct kvm_vcpu *vcpu);
550
Avi Kivity30677142007-10-28 18:48:59 +0200551int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t gva, u32 error_code);
Zhang Xiantao34c16ee2007-10-20 15:34:38 +0800552
Joerg Roedel18552672008-02-07 13:47:41 +0100553void kvm_enable_tdp(void);
554
Carsten Ottea03490e2007-10-29 16:09:35 +0100555int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
Carsten Ottede7d7892007-10-30 18:44:25 +0100556int complete_pio(struct kvm_vcpu *vcpu);
Zhang Xiantaoec6d2732007-11-19 15:08:31 +0800557
558static inline struct kvm_mmu_page *page_header(hpa_t shadow_page)
559{
560 struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
561
562 return (struct kvm_mmu_page *)page_private(page);
563}
564
565static inline u16 read_fs(void)
566{
567 u16 seg;
568 asm("mov %%fs, %0" : "=g"(seg));
569 return seg;
570}
571
572static inline u16 read_gs(void)
573{
574 u16 seg;
575 asm("mov %%gs, %0" : "=g"(seg));
576 return seg;
577}
578
579static inline u16 read_ldt(void)
580{
581 u16 ldt;
582 asm("sldt %0" : "=g"(ldt));
583 return ldt;
584}
585
586static inline void load_fs(u16 sel)
587{
588 asm("mov %0, %%fs" : : "rm"(sel));
589}
590
591static inline void load_gs(u16 sel)
592{
593 asm("mov %0, %%gs" : : "rm"(sel));
594}
595
596#ifndef load_ldt
597static inline void load_ldt(u16 sel)
598{
599 asm("lldt %0" : : "rm"(sel));
600}
601#endif
602
603static inline void get_idt(struct descriptor_table *table)
604{
605 asm("sidt %0" : "=m"(*table));
606}
607
608static inline void get_gdt(struct descriptor_table *table)
609{
610 asm("sgdt %0" : "=m"(*table));
611}
612
613static inline unsigned long read_tr_base(void)
614{
615 u16 tr;
616 asm("str %0" : "=g"(tr));
617 return segment_base(tr);
618}
619
620#ifdef CONFIG_X86_64
621static inline unsigned long read_msr(unsigned long msr)
622{
623 u64 value;
624
625 rdmsrl(msr, value);
626 return value;
627}
628#endif
629
630static inline void fx_save(struct i387_fxsave_struct *image)
631{
632 asm("fxsave (%0)":: "r" (image));
633}
634
635static inline void fx_restore(struct i387_fxsave_struct *image)
636{
637 asm("fxrstor (%0)":: "r" (image));
638}
639
Andrea Arcangelibc1a34f2008-05-01 18:43:33 +0200640static inline void fx_finit(void)
Zhang Xiantaoec6d2732007-11-19 15:08:31 +0800641{
642 asm("finit");
643}
644
645static inline u32 get_rdx_init_val(void)
646{
647 return 0x600; /* P6 family */
648}
649
Avi Kivityc1a5d4f2007-11-25 14:12:03 +0200650static inline void kvm_inject_gp(struct kvm_vcpu *vcpu, u32 error_code)
651{
652 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
653}
654
Zhang Xiantaoec6d2732007-11-19 15:08:31 +0800655#define ASM_VMX_VMCLEAR_RAX ".byte 0x66, 0x0f, 0xc7, 0x30"
656#define ASM_VMX_VMLAUNCH ".byte 0x0f, 0x01, 0xc2"
657#define ASM_VMX_VMRESUME ".byte 0x0f, 0x01, 0xc3"
658#define ASM_VMX_VMPTRLD_RAX ".byte 0x0f, 0xc7, 0x30"
659#define ASM_VMX_VMREAD_RDX_RAX ".byte 0x0f, 0x78, 0xd0"
660#define ASM_VMX_VMWRITE_RAX_RDX ".byte 0x0f, 0x79, 0xd0"
661#define ASM_VMX_VMWRITE_RSP_RDX ".byte 0x0f, 0x79, 0xd4"
662#define ASM_VMX_VMXOFF ".byte 0x0f, 0x01, 0xc4"
663#define ASM_VMX_VMXON_RAX ".byte 0xf3, 0x0f, 0xc7, 0x30"
Sheng Yang14394422008-04-28 12:24:45 +0800664#define ASM_VMX_INVEPT ".byte 0x66, 0x0f, 0x38, 0x80, 0x08"
Sheng Yang2384d2b2008-01-17 15:14:33 +0800665#define ASM_VMX_INVVPID ".byte 0x66, 0x0f, 0x38, 0x81, 0x08"
Zhang Xiantaoec6d2732007-11-19 15:08:31 +0800666
667#define MSR_IA32_TIME_STAMP_COUNTER 0x010
668
669#define TSS_IOPB_BASE_OFFSET 0x66
670#define TSS_BASE_SIZE 0x68
671#define TSS_IOPB_SIZE (65536 / 8)
672#define TSS_REDIRECTION_SIZE (256 / 8)
Joe Perches7d76b4d2008-03-23 01:02:34 -0700673#define RMODE_TSS_SIZE \
674 (TSS_BASE_SIZE + TSS_REDIRECTION_SIZE + TSS_IOPB_SIZE + 1)
Hollis Blanchard53e0aa72007-12-03 16:15:26 -0600675
Izik Eidus37817f22008-03-24 23:14:53 +0200676enum {
677 TASK_SWITCH_CALL = 0,
678 TASK_SWITCH_IRET = 1,
679 TASK_SWITCH_JMP = 2,
680 TASK_SWITCH_GATE = 3,
681};
682
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -0400683#define KVMTRACE_5D(evt, vcpu, d1, d2, d3, d4, d5, name) \
684 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
685 vcpu, 5, d1, d2, d3, d4, d5)
686#define KVMTRACE_4D(evt, vcpu, d1, d2, d3, d4, name) \
687 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
688 vcpu, 4, d1, d2, d3, d4, 0)
689#define KVMTRACE_3D(evt, vcpu, d1, d2, d3, name) \
690 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
691 vcpu, 3, d1, d2, d3, 0, 0)
692#define KVMTRACE_2D(evt, vcpu, d1, d2, name) \
693 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
694 vcpu, 2, d1, d2, 0, 0, 0)
695#define KVMTRACE_1D(evt, vcpu, d1, name) \
696 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
697 vcpu, 1, d1, 0, 0, 0, 0)
698#define KVMTRACE_0D(evt, vcpu, name) \
699 trace_mark(kvm_trace_##name, "%u %p %u %u %u %u %u %u", KVM_TRC_##evt, \
700 vcpu, 0, 0, 0, 0, 0, 0)
701
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300702#ifdef CONFIG_64BIT
703#define KVM_EX_ENTRY ".quad"
704#else
705#define KVM_EX_ENTRY ".long"
706#endif
707
708/*
709 * Hardware virtualization extension instructions may fault if a
710 * reboot turns off virtualization while processes are running.
711 * Trap the fault and ignore the instruction if that happens.
712 */
713asmlinkage void kvm_handle_fault_on_reboot(void);
714
715#define __kvm_handle_fault_on_reboot(insn) \
716 "666: " insn "\n\t" \
717 ".pushsection .text.fixup, \"ax\" \n" \
718 "667: \n\t" \
719 "push $666b \n\t" \
720 "jmp kvm_handle_fault_on_reboot \n\t" \
721 ".popsection \n\t" \
722 ".pushsection __ex_table, \"a\" \n\t" \
723 KVM_EX_ENTRY " 666b, 667b \n\t" \
724 ".popsection"
725
Carsten Otte043405e2007-10-10 17:16:19 +0200726#endif