xref: /linux/arch/x86/kvm/vmx/vmx.h (revision c79c3c34f75d72a066e292b10aa50fc758c97c89)
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 "kvm_cache_regs.h"
12 #include "posted_intr.h"
13 #include "vmcs.h"
14 #include "vmx_ops.h"
15 #include "cpuid.h"
16 
17 extern const u32 vmx_msr_index[];
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 #ifdef CONFIG_X86_64
26 #define MAX_NR_USER_RETURN_MSRS	7
27 #else
28 #define MAX_NR_USER_RETURN_MSRS	4
29 #endif
30 
31 #define MAX_NR_LOADSTORE_MSRS	8
32 
33 struct vmx_msrs {
34 	unsigned int		nr;
35 	struct vmx_msr_entry	val[MAX_NR_LOADSTORE_MSRS];
36 };
37 
38 struct vmx_uret_msr {
39 	unsigned int slot; /* The MSR's slot in kvm_user_return_msrs. */
40 	u64 data;
41 	u64 mask;
42 };
43 
44 enum segment_cache_field {
45 	SEG_FIELD_SEL = 0,
46 	SEG_FIELD_BASE = 1,
47 	SEG_FIELD_LIMIT = 2,
48 	SEG_FIELD_AR = 3,
49 
50 	SEG_FIELD_NR = 4
51 };
52 
53 #define RTIT_ADDR_RANGE		4
54 
55 struct pt_ctx {
56 	u64 ctl;
57 	u64 status;
58 	u64 output_base;
59 	u64 output_mask;
60 	u64 cr3_match;
61 	u64 addr_a[RTIT_ADDR_RANGE];
62 	u64 addr_b[RTIT_ADDR_RANGE];
63 };
64 
65 struct pt_desc {
66 	u64 ctl_bitmask;
67 	u32 addr_range;
68 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
69 	struct pt_ctx host;
70 	struct pt_ctx guest;
71 };
72 
73 union vmx_exit_reason {
74 	struct {
75 		u32	basic			: 16;
76 		u32	reserved16		: 1;
77 		u32	reserved17		: 1;
78 		u32	reserved18		: 1;
79 		u32	reserved19		: 1;
80 		u32	reserved20		: 1;
81 		u32	reserved21		: 1;
82 		u32	reserved22		: 1;
83 		u32	reserved23		: 1;
84 		u32	reserved24		: 1;
85 		u32	reserved25		: 1;
86 		u32	bus_lock_detected	: 1;
87 		u32	enclave_mode		: 1;
88 		u32	smi_pending_mtf		: 1;
89 		u32	smi_from_vmx_root	: 1;
90 		u32	reserved30		: 1;
91 		u32	failed_vmentry		: 1;
92 	};
93 	u32 full;
94 };
95 
96 #define vcpu_to_lbr_desc(vcpu) (&to_vmx(vcpu)->lbr_desc)
97 #define vcpu_to_lbr_records(vcpu) (&to_vmx(vcpu)->lbr_desc.records)
98 
99 bool intel_pmu_lbr_is_compatible(struct kvm_vcpu *vcpu);
100 bool intel_pmu_lbr_is_enabled(struct kvm_vcpu *vcpu);
101 
102 int intel_pmu_create_guest_lbr_event(struct kvm_vcpu *vcpu);
103 void vmx_passthrough_lbr_msrs(struct kvm_vcpu *vcpu);
104 
105 struct lbr_desc {
106 	/* Basic info about guest LBR records. */
107 	struct x86_pmu_lbr records;
108 
109 	/*
110 	 * Emulate LBR feature via passthrough LBR registers when the
111 	 * per-vcpu guest LBR event is scheduled on the current pcpu.
112 	 *
113 	 * The records may be inaccurate if the host reclaims the LBR.
114 	 */
115 	struct perf_event *event;
116 
117 	/* True if LBRs are marked as not intercepted in the MSR bitmap */
118 	bool msr_passthrough;
119 };
120 
121 /*
122  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
123  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
124  */
125 struct nested_vmx {
126 	/* Has the level1 guest done vmxon? */
127 	bool vmxon;
128 	gpa_t vmxon_ptr;
129 	bool pml_full;
130 
131 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
132 	gpa_t current_vmptr;
133 	/*
134 	 * Cache of the guest's VMCS, existing outside of guest memory.
135 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
136 	 * memory during VMCLEAR and VMPTRLD.
137 	 */
138 	struct vmcs12 *cached_vmcs12;
139 	/*
140 	 * Cache of the guest's shadow VMCS, existing outside of guest
141 	 * memory. Loaded from guest memory during VM entry. Flushed
142 	 * to guest memory during VM exit.
143 	 */
144 	struct vmcs12 *cached_shadow_vmcs12;
145 
146 	/*
147 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
148 	 * with the data held by struct vmcs12.
149 	 */
150 	bool need_vmcs12_to_shadow_sync;
151 	bool dirty_vmcs12;
152 
153 	/*
154 	 * Indicates lazily loaded guest state has not yet been decached from
155 	 * vmcs02.
156 	 */
157 	bool need_sync_vmcs02_to_vmcs12_rare;
158 
159 	/*
160 	 * vmcs02 has been initialized, i.e. state that is constant for
161 	 * vmcs02 has been written to the backing VMCS.  Initialization
162 	 * is delayed until L1 actually attempts to run a nested VM.
163 	 */
164 	bool vmcs02_initialized;
165 
166 	bool change_vmcs01_virtual_apic_mode;
167 	bool reload_vmcs01_apic_access_page;
168 
169 	/*
170 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
171 	 * use it. However, VMX features available to L1 will be limited based
172 	 * on what the enlightened VMCS supports.
173 	 */
174 	bool enlightened_vmcs_enabled;
175 
176 	/* L2 must run next, and mustn't decide to exit to L1. */
177 	bool nested_run_pending;
178 
179 	/* Pending MTF VM-exit into L1.  */
180 	bool mtf_pending;
181 
182 	struct loaded_vmcs vmcs02;
183 
184 	/*
185 	 * Guest pages referred to in the vmcs02 with host-physical
186 	 * pointers, so we must keep them pinned while L2 runs.
187 	 */
188 	struct page *apic_access_page;
189 	struct kvm_host_map virtual_apic_map;
190 	struct kvm_host_map pi_desc_map;
191 
192 	struct kvm_host_map msr_bitmap_map;
193 
194 	struct pi_desc *pi_desc;
195 	bool pi_pending;
196 	u16 posted_intr_nv;
197 
198 	struct hrtimer preemption_timer;
199 	u64 preemption_timer_deadline;
200 	bool has_preemption_timer_deadline;
201 	bool preemption_timer_expired;
202 
203 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
204 	u64 vmcs01_debugctl;
205 	u64 vmcs01_guest_bndcfgs;
206 
207 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
208 	int l1_tpr_threshold;
209 
210 	u16 vpid02;
211 	u16 last_vpid;
212 
213 	struct nested_vmx_msrs msrs;
214 
215 	/* SMM related state */
216 	struct {
217 		/* in VMX operation on SMM entry? */
218 		bool vmxon;
219 		/* in guest mode on SMM entry? */
220 		bool guest_mode;
221 	} smm;
222 
223 	gpa_t hv_evmcs_vmptr;
224 	struct kvm_host_map hv_evmcs_map;
225 	struct hv_enlightened_vmcs *hv_evmcs;
226 };
227 
228 struct vcpu_vmx {
229 	struct kvm_vcpu       vcpu;
230 	u8                    fail;
231 	u8		      msr_bitmap_mode;
232 
233 	/*
234 	 * If true, host state has been stored in vmx->loaded_vmcs for
235 	 * the CPU registers that only need to be switched when transitioning
236 	 * to/from the kernel, and the registers have been loaded with guest
237 	 * values.  If false, host state is loaded in the CPU registers
238 	 * and vmx->loaded_vmcs->host_state is invalid.
239 	 */
240 	bool		      guest_state_loaded;
241 
242 	unsigned long         exit_qualification;
243 	u32                   exit_intr_info;
244 	u32                   idt_vectoring_info;
245 	ulong                 rflags;
246 
247 	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
248 	int                   nr_uret_msrs;
249 	int                   nr_active_uret_msrs;
250 	bool                  guest_uret_msrs_loaded;
251 #ifdef CONFIG_X86_64
252 	u64		      msr_host_kernel_gs_base;
253 	u64		      msr_guest_kernel_gs_base;
254 #endif
255 
256 	u64		      spec_ctrl;
257 	u32		      msr_ia32_umwait_control;
258 
259 	u32 secondary_exec_control;
260 
261 	/*
262 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
263 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
264 	 * guest (L2), it points to a different VMCS.
265 	 */
266 	struct loaded_vmcs    vmcs01;
267 	struct loaded_vmcs   *loaded_vmcs;
268 
269 	struct msr_autoload {
270 		struct vmx_msrs guest;
271 		struct vmx_msrs host;
272 	} msr_autoload;
273 
274 	struct msr_autostore {
275 		struct vmx_msrs guest;
276 	} msr_autostore;
277 
278 	struct {
279 		int vm86_active;
280 		ulong save_rflags;
281 		struct kvm_segment segs[8];
282 	} rmode;
283 	struct {
284 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
285 		struct kvm_save_segment {
286 			u16 selector;
287 			unsigned long base;
288 			u32 limit;
289 			u32 ar;
290 		} seg[8];
291 	} segment_cache;
292 	int vpid;
293 	bool emulation_required;
294 
295 	union vmx_exit_reason exit_reason;
296 
297 	/* Posted interrupt descriptor */
298 	struct pi_desc pi_desc;
299 
300 	/* Support for a guest hypervisor (nested VMX) */
301 	struct nested_vmx nested;
302 
303 	/* Dynamic PLE window. */
304 	unsigned int ple_window;
305 	bool ple_window_dirty;
306 
307 	bool req_immediate_exit;
308 
309 	/* Support for PML */
310 #define PML_ENTITY_NUM		512
311 	struct page *pml_pg;
312 
313 	/* apic deadline value in host tsc */
314 	u64 hv_deadline_tsc;
315 
316 	u64 current_tsc_ratio;
317 
318 	unsigned long host_debugctlmsr;
319 
320 	/*
321 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
322 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
323 	 * in msr_ia32_feature_control_valid_bits.
324 	 */
325 	u64 msr_ia32_feature_control;
326 	u64 msr_ia32_feature_control_valid_bits;
327 	u64 ept_pointer;
328 
329 	struct pt_desc pt_desc;
330 	struct lbr_desc lbr_desc;
331 
332 	/* Save desired MSR intercept (read: pass-through) state */
333 #define MAX_POSSIBLE_PASSTHROUGH_MSRS	13
334 	struct {
335 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
336 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
337 	} shadow_msr_intercept;
338 };
339 
340 enum ept_pointers_status {
341 	EPT_POINTERS_CHECK = 0,
342 	EPT_POINTERS_MATCH = 1,
343 	EPT_POINTERS_MISMATCH = 2
344 };
345 
346 struct kvm_vmx {
347 	struct kvm kvm;
348 
349 	unsigned int tss_addr;
350 	bool ept_identity_pagetable_done;
351 	gpa_t ept_identity_map_addr;
352 
353 	enum ept_pointers_status ept_pointers_match;
354 	spinlock_t ept_pointer_lock;
355 };
356 
357 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
358 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
359 			struct loaded_vmcs *buddy);
360 int allocate_vpid(void);
361 void free_vpid(int vpid);
362 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
363 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
364 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
365 			unsigned long fs_base, unsigned long gs_base);
366 int vmx_get_cpl(struct kvm_vcpu *vcpu);
367 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
368 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
369 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
370 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
371 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
372 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
373 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
374 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
375 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
376 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
377 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
378 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa,
379 		   int root_level);
380 
381 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu);
382 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
383 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
384 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
385 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
386 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
387 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
388 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
389 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
390 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
391 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
392 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
393 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
394 void vmx_set_intercept_for_msr(struct kvm_vcpu *vcpu,
395 	u32 msr, int type, bool value);
396 
397 static inline u8 vmx_get_rvi(void)
398 {
399 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
400 }
401 
402 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
403 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
404 {									    \
405 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
406 		vmcs_write32(uname, val);				    \
407 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
408 	}								    \
409 }									    \
410 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
411 {									    \
412 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
413 }									    \
414 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
415 {									    \
416 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
417 }									    \
418 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
419 {									    \
420 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
421 }
422 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
423 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
424 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
425 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
426 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
427 
428 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
429 {
430 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
431 				  | (1 << VCPU_EXREG_RFLAGS)
432 				  | (1 << VCPU_EXREG_PDPTR)
433 				  | (1 << VCPU_EXREG_SEGMENTS)
434 				  | (1 << VCPU_EXREG_CR0)
435 				  | (1 << VCPU_EXREG_CR3)
436 				  | (1 << VCPU_EXREG_CR4)
437 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
438 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
439 	vcpu->arch.regs_dirty = 0;
440 }
441 
442 static inline u32 vmx_vmentry_ctrl(void)
443 {
444 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
445 	if (vmx_pt_mode_is_system())
446 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
447 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
448 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
449 	return vmentry_ctrl &
450 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
451 }
452 
453 static inline u32 vmx_vmexit_ctrl(void)
454 {
455 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
456 	if (vmx_pt_mode_is_system())
457 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
458 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
459 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
460 	return vmexit_ctrl &
461 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
462 }
463 
464 u32 vmx_exec_control(struct vcpu_vmx *vmx);
465 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
466 
467 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
468 {
469 	return container_of(kvm, struct kvm_vmx, kvm);
470 }
471 
472 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
473 {
474 	return container_of(vcpu, struct vcpu_vmx, vcpu);
475 }
476 
477 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
478 {
479 	struct vcpu_vmx *vmx = to_vmx(vcpu);
480 
481 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
482 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
483 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
484 	}
485 	return vmx->exit_qualification;
486 }
487 
488 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
489 {
490 	struct vcpu_vmx *vmx = to_vmx(vcpu);
491 
492 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
493 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
494 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
495 	}
496 	return vmx->exit_intr_info;
497 }
498 
499 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
500 void free_vmcs(struct vmcs *vmcs);
501 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
502 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
503 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
504 
505 static inline struct vmcs *alloc_vmcs(bool shadow)
506 {
507 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
508 			      GFP_KERNEL_ACCOUNT);
509 }
510 
511 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
512 {
513 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
514 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
515 }
516 
517 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
518 {
519 	return vmx->secondary_exec_control &
520 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
521 }
522 
523 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
524 {
525 	if (!enable_ept)
526 		return true;
527 
528 	return allow_smaller_maxphyaddr && cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
529 }
530 
531 static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
532 {
533 	return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
534 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
535 	    SECONDARY_EXEC_UNRESTRICTED_GUEST));
536 }
537 
538 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
539 static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
540 {
541 	return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
542 }
543 
544 void dump_vmcs(void);
545 
546 #endif /* __KVM_X86_VMX_H */
547