xref: /linux/arch/x86/kvm/vmx/vmx.h (revision 2c97b5ae83dca56718774e7b4bf9640f05d11867)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __KVM_X86_VMX_H
3 #define __KVM_X86_VMX_H
4 
5 #include <linux/kvm_host.h>
6 
7 #include <asm/kvm.h>
8 #include <asm/intel_pt.h>
9 
10 #include "capabilities.h"
11 #include "ops.h"
12 #include "vmcs.h"
13 
14 extern const u32 vmx_msr_index[];
15 extern u64 host_efer;
16 
17 extern u32 get_umwait_control_msr(void);
18 
19 #define MSR_TYPE_R	1
20 #define MSR_TYPE_W	2
21 #define MSR_TYPE_RW	3
22 
23 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
24 
25 #define NR_LOADSTORE_MSRS 8
26 
27 struct vmx_msrs {
28 	unsigned int		nr;
29 	struct vmx_msr_entry	val[NR_LOADSTORE_MSRS];
30 };
31 
32 struct shared_msr_entry {
33 	unsigned index;
34 	u64 data;
35 	u64 mask;
36 };
37 
38 enum segment_cache_field {
39 	SEG_FIELD_SEL = 0,
40 	SEG_FIELD_BASE = 1,
41 	SEG_FIELD_LIMIT = 2,
42 	SEG_FIELD_AR = 3,
43 
44 	SEG_FIELD_NR = 4
45 };
46 
47 /* Posted-Interrupt Descriptor */
48 struct pi_desc {
49 	u32 pir[8];     /* Posted interrupt requested */
50 	union {
51 		struct {
52 				/* bit 256 - Outstanding Notification */
53 			u16	on	: 1,
54 				/* bit 257 - Suppress Notification */
55 				sn	: 1,
56 				/* bit 271:258 - Reserved */
57 				rsvd_1	: 14;
58 				/* bit 279:272 - Notification Vector */
59 			u8	nv;
60 				/* bit 287:280 - Reserved */
61 			u8	rsvd_2;
62 				/* bit 319:288 - Notification Destination */
63 			u32	ndst;
64 		};
65 		u64 control;
66 	};
67 	u32 rsvd[6];
68 } __aligned(64);
69 
70 #define RTIT_ADDR_RANGE		4
71 
72 struct pt_ctx {
73 	u64 ctl;
74 	u64 status;
75 	u64 output_base;
76 	u64 output_mask;
77 	u64 cr3_match;
78 	u64 addr_a[RTIT_ADDR_RANGE];
79 	u64 addr_b[RTIT_ADDR_RANGE];
80 };
81 
82 struct pt_desc {
83 	u64 ctl_bitmask;
84 	u32 addr_range;
85 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
86 	struct pt_ctx host;
87 	struct pt_ctx guest;
88 };
89 
90 /*
91  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
92  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
93  */
94 struct nested_vmx {
95 	/* Has the level1 guest done vmxon? */
96 	bool vmxon;
97 	gpa_t vmxon_ptr;
98 	bool pml_full;
99 
100 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
101 	gpa_t current_vmptr;
102 	/*
103 	 * Cache of the guest's VMCS, existing outside of guest memory.
104 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
105 	 * memory during VMCLEAR and VMPTRLD.
106 	 */
107 	struct vmcs12 *cached_vmcs12;
108 	/*
109 	 * Cache of the guest's shadow VMCS, existing outside of guest
110 	 * memory. Loaded from guest memory during VM entry. Flushed
111 	 * to guest memory during VM exit.
112 	 */
113 	struct vmcs12 *cached_shadow_vmcs12;
114 
115 	/*
116 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
117 	 * with the data held by struct vmcs12.
118 	 */
119 	bool need_vmcs12_to_shadow_sync;
120 	bool dirty_vmcs12;
121 
122 	/*
123 	 * Indicates lazily loaded guest state has not yet been decached from
124 	 * vmcs02.
125 	 */
126 	bool need_sync_vmcs02_to_vmcs12_rare;
127 
128 	/*
129 	 * vmcs02 has been initialized, i.e. state that is constant for
130 	 * vmcs02 has been written to the backing VMCS.  Initialization
131 	 * is delayed until L1 actually attempts to run a nested VM.
132 	 */
133 	bool vmcs02_initialized;
134 
135 	bool change_vmcs01_virtual_apic_mode;
136 
137 	/*
138 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
139 	 * use it. However, VMX features available to L1 will be limited based
140 	 * on what the enlightened VMCS supports.
141 	 */
142 	bool enlightened_vmcs_enabled;
143 
144 	/* L2 must run next, and mustn't decide to exit to L1. */
145 	bool nested_run_pending;
146 
147 	struct loaded_vmcs vmcs02;
148 
149 	/*
150 	 * Guest pages referred to in the vmcs02 with host-physical
151 	 * pointers, so we must keep them pinned while L2 runs.
152 	 */
153 	struct page *apic_access_page;
154 	struct kvm_host_map virtual_apic_map;
155 	struct kvm_host_map pi_desc_map;
156 
157 	struct kvm_host_map msr_bitmap_map;
158 
159 	struct pi_desc *pi_desc;
160 	bool pi_pending;
161 	u16 posted_intr_nv;
162 
163 	struct hrtimer preemption_timer;
164 	bool preemption_timer_expired;
165 
166 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
167 	u64 vmcs01_debugctl;
168 	u64 vmcs01_guest_bndcfgs;
169 
170 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
171 	int l1_tpr_threshold;
172 
173 	u16 vpid02;
174 	u16 last_vpid;
175 
176 	struct nested_vmx_msrs msrs;
177 
178 	/* SMM related state */
179 	struct {
180 		/* in VMX operation on SMM entry? */
181 		bool vmxon;
182 		/* in guest mode on SMM entry? */
183 		bool guest_mode;
184 	} smm;
185 
186 	gpa_t hv_evmcs_vmptr;
187 	struct kvm_host_map hv_evmcs_map;
188 	struct hv_enlightened_vmcs *hv_evmcs;
189 };
190 
191 struct vcpu_vmx {
192 	struct kvm_vcpu       vcpu;
193 	u8                    fail;
194 	u8		      msr_bitmap_mode;
195 
196 	/*
197 	 * If true, host state has been stored in vmx->loaded_vmcs for
198 	 * the CPU registers that only need to be switched when transitioning
199 	 * to/from the kernel, and the registers have been loaded with guest
200 	 * values.  If false, host state is loaded in the CPU registers
201 	 * and vmx->loaded_vmcs->host_state is invalid.
202 	 */
203 	bool		      guest_state_loaded;
204 
205 	u32                   exit_intr_info;
206 	u32                   idt_vectoring_info;
207 	ulong                 rflags;
208 
209 	struct shared_msr_entry *guest_msrs;
210 	int                   nmsrs;
211 	int                   save_nmsrs;
212 	bool                  guest_msrs_ready;
213 #ifdef CONFIG_X86_64
214 	u64		      msr_host_kernel_gs_base;
215 	u64		      msr_guest_kernel_gs_base;
216 #endif
217 
218 	u64		      spec_ctrl;
219 	u32		      msr_ia32_umwait_control;
220 
221 	u32 secondary_exec_control;
222 
223 	/*
224 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
225 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
226 	 * guest (L2), it points to a different VMCS.
227 	 */
228 	struct loaded_vmcs    vmcs01;
229 	struct loaded_vmcs   *loaded_vmcs;
230 
231 	struct msr_autoload {
232 		struct vmx_msrs guest;
233 		struct vmx_msrs host;
234 	} msr_autoload;
235 
236 	struct msr_autostore {
237 		struct vmx_msrs guest;
238 	} msr_autostore;
239 
240 	struct {
241 		int vm86_active;
242 		ulong save_rflags;
243 		struct kvm_segment segs[8];
244 	} rmode;
245 	struct {
246 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
247 		struct kvm_save_segment {
248 			u16 selector;
249 			unsigned long base;
250 			u32 limit;
251 			u32 ar;
252 		} seg[8];
253 	} segment_cache;
254 	int vpid;
255 	bool emulation_required;
256 
257 	u32 exit_reason;
258 
259 	/* Posted interrupt descriptor */
260 	struct pi_desc pi_desc;
261 
262 	/* Support for a guest hypervisor (nested VMX) */
263 	struct nested_vmx nested;
264 
265 	/* Dynamic PLE window. */
266 	unsigned int ple_window;
267 	bool ple_window_dirty;
268 
269 	bool req_immediate_exit;
270 
271 	/* Support for PML */
272 #define PML_ENTITY_NUM		512
273 	struct page *pml_pg;
274 
275 	/* apic deadline value in host tsc */
276 	u64 hv_deadline_tsc;
277 
278 	u64 current_tsc_ratio;
279 
280 	u32 host_pkru;
281 
282 	unsigned long host_debugctlmsr;
283 
284 	/*
285 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
286 	 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
287 	 * in msr_ia32_feature_control_valid_bits.
288 	 */
289 	u64 msr_ia32_feature_control;
290 	u64 msr_ia32_feature_control_valid_bits;
291 	u64 ept_pointer;
292 
293 	struct pt_desc pt_desc;
294 };
295 
296 enum ept_pointers_status {
297 	EPT_POINTERS_CHECK = 0,
298 	EPT_POINTERS_MATCH = 1,
299 	EPT_POINTERS_MISMATCH = 2
300 };
301 
302 struct kvm_vmx {
303 	struct kvm kvm;
304 
305 	unsigned int tss_addr;
306 	bool ept_identity_pagetable_done;
307 	gpa_t ept_identity_map_addr;
308 
309 	enum ept_pointers_status ept_pointers_match;
310 	spinlock_t ept_pointer_lock;
311 };
312 
313 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
314 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu);
315 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
316 int allocate_vpid(void);
317 void free_vpid(int vpid);
318 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
319 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
320 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
321 			unsigned long fs_base, unsigned long gs_base);
322 int vmx_get_cpl(struct kvm_vcpu *vcpu);
323 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
324 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
325 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
326 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
327 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
328 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
329 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
330 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
331 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
332 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
333 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
334 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
335 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
336 void update_exception_bitmap(struct kvm_vcpu *vcpu);
337 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
338 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
339 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
340 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
341 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
342 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
343 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
344 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr);
345 
346 #define POSTED_INTR_ON  0
347 #define POSTED_INTR_SN  1
348 
349 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
350 {
351 	return test_and_set_bit(POSTED_INTR_ON,
352 			(unsigned long *)&pi_desc->control);
353 }
354 
355 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
356 {
357 	return test_and_clear_bit(POSTED_INTR_ON,
358 			(unsigned long *)&pi_desc->control);
359 }
360 
361 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
362 {
363 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
364 }
365 
366 static inline bool pi_is_pir_empty(struct pi_desc *pi_desc)
367 {
368 	return bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS);
369 }
370 
371 static inline void pi_set_sn(struct pi_desc *pi_desc)
372 {
373 	set_bit(POSTED_INTR_SN,
374 		(unsigned long *)&pi_desc->control);
375 }
376 
377 static inline void pi_set_on(struct pi_desc *pi_desc)
378 {
379 	set_bit(POSTED_INTR_ON,
380 		(unsigned long *)&pi_desc->control);
381 }
382 
383 static inline void pi_clear_on(struct pi_desc *pi_desc)
384 {
385 	clear_bit(POSTED_INTR_ON,
386 		(unsigned long *)&pi_desc->control);
387 }
388 
389 static inline void pi_clear_sn(struct pi_desc *pi_desc)
390 {
391 	clear_bit(POSTED_INTR_SN,
392 		(unsigned long *)&pi_desc->control);
393 }
394 
395 static inline int pi_test_on(struct pi_desc *pi_desc)
396 {
397 	return test_bit(POSTED_INTR_ON,
398 			(unsigned long *)&pi_desc->control);
399 }
400 
401 static inline int pi_test_sn(struct pi_desc *pi_desc)
402 {
403 	return test_bit(POSTED_INTR_SN,
404 			(unsigned long *)&pi_desc->control);
405 }
406 
407 static inline u8 vmx_get_rvi(void)
408 {
409 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
410 }
411 
412 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
413 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
414 {									    \
415 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
416 		vmcs_write32(uname, val);				    \
417 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
418 	}								    \
419 }									    \
420 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
421 {									    \
422 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
423 }									    \
424 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
425 {									    \
426 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
427 }									    \
428 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
429 {									    \
430 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
431 }
432 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
433 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
434 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
435 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
436 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
437 
438 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
439 {
440 	vmx->segment_cache.bitmask = 0;
441 }
442 
443 static inline u32 vmx_vmentry_ctrl(void)
444 {
445 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
446 	if (pt_mode == PT_MODE_SYSTEM)
447 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
448 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
449 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
450 	return vmentry_ctrl &
451 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
452 }
453 
454 static inline u32 vmx_vmexit_ctrl(void)
455 {
456 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
457 	if (pt_mode == PT_MODE_SYSTEM)
458 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
459 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
460 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
461 	return vmexit_ctrl &
462 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
463 }
464 
465 u32 vmx_exec_control(struct vcpu_vmx *vmx);
466 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
467 
468 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
469 {
470 	return container_of(kvm, struct kvm_vmx, kvm);
471 }
472 
473 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
474 {
475 	return container_of(vcpu, struct vcpu_vmx, vcpu);
476 }
477 
478 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
479 {
480 	return &(to_vmx(vcpu)->pi_desc);
481 }
482 
483 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
484 void free_vmcs(struct vmcs *vmcs);
485 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
486 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
487 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs);
488 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
489 
490 static inline struct vmcs *alloc_vmcs(bool shadow)
491 {
492 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
493 			      GFP_KERNEL_ACCOUNT);
494 }
495 
496 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
497 
498 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
499 				bool invalidate_gpa)
500 {
501 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
502 		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
503 			return;
504 		ept_sync_context(construct_eptp(vcpu,
505 						vcpu->arch.mmu->root_hpa));
506 	} else {
507 		vpid_sync_context(vpid);
508 	}
509 }
510 
511 static inline void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
512 {
513 	__vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
514 }
515 
516 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
517 {
518 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
519 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
520 }
521 
522 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
523 {
524 	return vmx->secondary_exec_control &
525 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
526 }
527 
528 void dump_vmcs(void);
529 
530 #endif /* __KVM_X86_VMX_H */
531