xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision e0d07278)
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 "ops.h"
13 #include "vmcs.h"
14 #include "cpuid.h"
15 
16 extern const u32 vmx_msr_index[];
17 
18 #define MSR_TYPE_R	1
19 #define MSR_TYPE_W	2
20 #define MSR_TYPE_RW	3
21 
22 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
23 
24 #ifdef CONFIG_X86_64
25 #define NR_SHARED_MSRS	7
26 #else
27 #define NR_SHARED_MSRS	4
28 #endif
29 
30 #define NR_LOADSTORE_MSRS 8
31 
32 struct vmx_msrs {
33 	unsigned int		nr;
34 	struct vmx_msr_entry	val[NR_LOADSTORE_MSRS];
35 };
36 
37 struct shared_msr_entry {
38 	unsigned index;
39 	u64 data;
40 	u64 mask;
41 };
42 
43 enum segment_cache_field {
44 	SEG_FIELD_SEL = 0,
45 	SEG_FIELD_BASE = 1,
46 	SEG_FIELD_LIMIT = 2,
47 	SEG_FIELD_AR = 3,
48 
49 	SEG_FIELD_NR = 4
50 };
51 
52 /* Posted-Interrupt Descriptor */
53 struct pi_desc {
54 	u32 pir[8];     /* Posted interrupt requested */
55 	union {
56 		struct {
57 				/* bit 256 - Outstanding Notification */
58 			u16	on	: 1,
59 				/* bit 257 - Suppress Notification */
60 				sn	: 1,
61 				/* bit 271:258 - Reserved */
62 				rsvd_1	: 14;
63 				/* bit 279:272 - Notification Vector */
64 			u8	nv;
65 				/* bit 287:280 - Reserved */
66 			u8	rsvd_2;
67 				/* bit 319:288 - Notification Destination */
68 			u32	ndst;
69 		};
70 		u64 control;
71 	};
72 	u32 rsvd[6];
73 } __aligned(64);
74 
75 #define RTIT_ADDR_RANGE		4
76 
77 struct pt_ctx {
78 	u64 ctl;
79 	u64 status;
80 	u64 output_base;
81 	u64 output_mask;
82 	u64 cr3_match;
83 	u64 addr_a[RTIT_ADDR_RANGE];
84 	u64 addr_b[RTIT_ADDR_RANGE];
85 };
86 
87 struct pt_desc {
88 	u64 ctl_bitmask;
89 	u32 addr_range;
90 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
91 	struct pt_ctx host;
92 	struct pt_ctx guest;
93 };
94 
95 /*
96  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
97  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
98  */
99 struct nested_vmx {
100 	/* Has the level1 guest done vmxon? */
101 	bool vmxon;
102 	gpa_t vmxon_ptr;
103 	bool pml_full;
104 
105 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
106 	gpa_t current_vmptr;
107 	/*
108 	 * Cache of the guest's VMCS, existing outside of guest memory.
109 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
110 	 * memory during VMCLEAR and VMPTRLD.
111 	 */
112 	struct vmcs12 *cached_vmcs12;
113 	/*
114 	 * Cache of the guest's shadow VMCS, existing outside of guest
115 	 * memory. Loaded from guest memory during VM entry. Flushed
116 	 * to guest memory during VM exit.
117 	 */
118 	struct vmcs12 *cached_shadow_vmcs12;
119 
120 	/*
121 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
122 	 * with the data held by struct vmcs12.
123 	 */
124 	bool need_vmcs12_to_shadow_sync;
125 	bool dirty_vmcs12;
126 
127 	/*
128 	 * Indicates lazily loaded guest state has not yet been decached from
129 	 * vmcs02.
130 	 */
131 	bool need_sync_vmcs02_to_vmcs12_rare;
132 
133 	/*
134 	 * vmcs02 has been initialized, i.e. state that is constant for
135 	 * vmcs02 has been written to the backing VMCS.  Initialization
136 	 * is delayed until L1 actually attempts to run a nested VM.
137 	 */
138 	bool vmcs02_initialized;
139 
140 	bool change_vmcs01_virtual_apic_mode;
141 	bool reload_vmcs01_apic_access_page;
142 
143 	/*
144 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
145 	 * use it. However, VMX features available to L1 will be limited based
146 	 * on what the enlightened VMCS supports.
147 	 */
148 	bool enlightened_vmcs_enabled;
149 
150 	/* L2 must run next, and mustn't decide to exit to L1. */
151 	bool nested_run_pending;
152 
153 	/* Pending MTF VM-exit into L1.  */
154 	bool mtf_pending;
155 
156 	struct loaded_vmcs vmcs02;
157 
158 	/*
159 	 * Guest pages referred to in the vmcs02 with host-physical
160 	 * pointers, so we must keep them pinned while L2 runs.
161 	 */
162 	struct page *apic_access_page;
163 	struct kvm_host_map virtual_apic_map;
164 	struct kvm_host_map pi_desc_map;
165 
166 	struct kvm_host_map msr_bitmap_map;
167 
168 	struct pi_desc *pi_desc;
169 	bool pi_pending;
170 	u16 posted_intr_nv;
171 
172 	struct hrtimer preemption_timer;
173 	u64 preemption_timer_deadline;
174 	bool has_preemption_timer_deadline;
175 	bool preemption_timer_expired;
176 
177 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
178 	u64 vmcs01_debugctl;
179 	u64 vmcs01_guest_bndcfgs;
180 
181 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
182 	int l1_tpr_threshold;
183 
184 	u16 vpid02;
185 	u16 last_vpid;
186 
187 	struct nested_vmx_msrs msrs;
188 
189 	/* SMM related state */
190 	struct {
191 		/* in VMX operation on SMM entry? */
192 		bool vmxon;
193 		/* in guest mode on SMM entry? */
194 		bool guest_mode;
195 	} smm;
196 
197 	gpa_t hv_evmcs_vmptr;
198 	struct kvm_host_map hv_evmcs_map;
199 	struct hv_enlightened_vmcs *hv_evmcs;
200 };
201 
202 struct vcpu_vmx {
203 	struct kvm_vcpu       vcpu;
204 	u8                    fail;
205 	u8		      msr_bitmap_mode;
206 
207 	/*
208 	 * If true, host state has been stored in vmx->loaded_vmcs for
209 	 * the CPU registers that only need to be switched when transitioning
210 	 * to/from the kernel, and the registers have been loaded with guest
211 	 * values.  If false, host state is loaded in the CPU registers
212 	 * and vmx->loaded_vmcs->host_state is invalid.
213 	 */
214 	bool		      guest_state_loaded;
215 
216 	unsigned long         exit_qualification;
217 	u32                   exit_intr_info;
218 	u32                   idt_vectoring_info;
219 	ulong                 rflags;
220 
221 	struct shared_msr_entry guest_msrs[NR_SHARED_MSRS];
222 	int                   nmsrs;
223 	int                   save_nmsrs;
224 	bool                  guest_msrs_ready;
225 #ifdef CONFIG_X86_64
226 	u64		      msr_host_kernel_gs_base;
227 	u64		      msr_guest_kernel_gs_base;
228 #endif
229 
230 	u64		      spec_ctrl;
231 	u32		      msr_ia32_umwait_control;
232 
233 	u32 secondary_exec_control;
234 
235 	/*
236 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
237 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
238 	 * guest (L2), it points to a different VMCS.
239 	 */
240 	struct loaded_vmcs    vmcs01;
241 	struct loaded_vmcs   *loaded_vmcs;
242 
243 	struct msr_autoload {
244 		struct vmx_msrs guest;
245 		struct vmx_msrs host;
246 	} msr_autoload;
247 
248 	struct msr_autostore {
249 		struct vmx_msrs guest;
250 	} msr_autostore;
251 
252 	struct {
253 		int vm86_active;
254 		ulong save_rflags;
255 		struct kvm_segment segs[8];
256 	} rmode;
257 	struct {
258 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
259 		struct kvm_save_segment {
260 			u16 selector;
261 			unsigned long base;
262 			u32 limit;
263 			u32 ar;
264 		} seg[8];
265 	} segment_cache;
266 	int vpid;
267 	bool emulation_required;
268 
269 	u32 exit_reason;
270 
271 	/* Posted interrupt descriptor */
272 	struct pi_desc pi_desc;
273 
274 	/* Support for a guest hypervisor (nested VMX) */
275 	struct nested_vmx nested;
276 
277 	/* Dynamic PLE window. */
278 	unsigned int ple_window;
279 	bool ple_window_dirty;
280 
281 	bool req_immediate_exit;
282 
283 	/* Support for PML */
284 #define PML_ENTITY_NUM		512
285 	struct page *pml_pg;
286 
287 	/* apic deadline value in host tsc */
288 	u64 hv_deadline_tsc;
289 
290 	u64 current_tsc_ratio;
291 
292 	unsigned long host_debugctlmsr;
293 
294 	/*
295 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
296 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
297 	 * in msr_ia32_feature_control_valid_bits.
298 	 */
299 	u64 msr_ia32_feature_control;
300 	u64 msr_ia32_feature_control_valid_bits;
301 	u64 ept_pointer;
302 
303 	struct pt_desc pt_desc;
304 };
305 
306 enum ept_pointers_status {
307 	EPT_POINTERS_CHECK = 0,
308 	EPT_POINTERS_MATCH = 1,
309 	EPT_POINTERS_MISMATCH = 2
310 };
311 
312 struct kvm_vmx {
313 	struct kvm kvm;
314 
315 	unsigned int tss_addr;
316 	bool ept_identity_pagetable_done;
317 	gpa_t ept_identity_map_addr;
318 
319 	enum ept_pointers_status ept_pointers_match;
320 	spinlock_t ept_pointer_lock;
321 };
322 
323 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
324 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
325 			struct loaded_vmcs *buddy);
326 int allocate_vpid(void);
327 void free_vpid(int vpid);
328 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
329 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
330 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
331 			unsigned long fs_base, unsigned long gs_base);
332 int vmx_get_cpl(struct kvm_vcpu *vcpu);
333 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
334 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
335 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
336 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
337 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
338 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
339 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
340 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
341 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
342 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
343 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
344 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa,
345 		   int root_level);
346 void update_exception_bitmap(struct kvm_vcpu *vcpu);
347 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
348 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
349 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
350 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
351 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
352 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
353 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
354 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
355 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
356 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr);
357 int vmx_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
358 			      struct x86_exception *e);
359 
360 #define POSTED_INTR_ON  0
361 #define POSTED_INTR_SN  1
362 
363 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
364 {
365 	return test_and_set_bit(POSTED_INTR_ON,
366 			(unsigned long *)&pi_desc->control);
367 }
368 
369 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
370 {
371 	return test_and_clear_bit(POSTED_INTR_ON,
372 			(unsigned long *)&pi_desc->control);
373 }
374 
375 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
376 {
377 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
378 }
379 
380 static inline bool pi_is_pir_empty(struct pi_desc *pi_desc)
381 {
382 	return bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS);
383 }
384 
385 static inline void pi_set_sn(struct pi_desc *pi_desc)
386 {
387 	set_bit(POSTED_INTR_SN,
388 		(unsigned long *)&pi_desc->control);
389 }
390 
391 static inline void pi_set_on(struct pi_desc *pi_desc)
392 {
393 	set_bit(POSTED_INTR_ON,
394 		(unsigned long *)&pi_desc->control);
395 }
396 
397 static inline void pi_clear_on(struct pi_desc *pi_desc)
398 {
399 	clear_bit(POSTED_INTR_ON,
400 		(unsigned long *)&pi_desc->control);
401 }
402 
403 static inline void pi_clear_sn(struct pi_desc *pi_desc)
404 {
405 	clear_bit(POSTED_INTR_SN,
406 		(unsigned long *)&pi_desc->control);
407 }
408 
409 static inline int pi_test_on(struct pi_desc *pi_desc)
410 {
411 	return test_bit(POSTED_INTR_ON,
412 			(unsigned long *)&pi_desc->control);
413 }
414 
415 static inline int pi_test_sn(struct pi_desc *pi_desc)
416 {
417 	return test_bit(POSTED_INTR_SN,
418 			(unsigned long *)&pi_desc->control);
419 }
420 
421 static inline u8 vmx_get_rvi(void)
422 {
423 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
424 }
425 
426 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
427 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
428 {									    \
429 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
430 		vmcs_write32(uname, val);				    \
431 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
432 	}								    \
433 }									    \
434 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
435 {									    \
436 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
437 }									    \
438 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
439 {									    \
440 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
441 }									    \
442 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
443 {									    \
444 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
445 }
446 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
447 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
448 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
449 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
450 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
451 
452 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
453 {
454 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
455 				  | (1 << VCPU_EXREG_RFLAGS)
456 				  | (1 << VCPU_EXREG_PDPTR)
457 				  | (1 << VCPU_EXREG_SEGMENTS)
458 				  | (1 << VCPU_EXREG_CR0)
459 				  | (1 << VCPU_EXREG_CR3)
460 				  | (1 << VCPU_EXREG_CR4)
461 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
462 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
463 	vcpu->arch.regs_dirty = 0;
464 }
465 
466 static inline u32 vmx_vmentry_ctrl(void)
467 {
468 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
469 	if (vmx_pt_mode_is_system())
470 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
471 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
472 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
473 	return vmentry_ctrl &
474 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
475 }
476 
477 static inline u32 vmx_vmexit_ctrl(void)
478 {
479 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
480 	if (vmx_pt_mode_is_system())
481 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
482 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
483 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
484 	return vmexit_ctrl &
485 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
486 }
487 
488 u32 vmx_exec_control(struct vcpu_vmx *vmx);
489 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
490 
491 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
492 {
493 	return container_of(kvm, struct kvm_vmx, kvm);
494 }
495 
496 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
497 {
498 	return container_of(vcpu, struct vcpu_vmx, vcpu);
499 }
500 
501 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
502 {
503 	return &(to_vmx(vcpu)->pi_desc);
504 }
505 
506 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
507 {
508 	struct vcpu_vmx *vmx = to_vmx(vcpu);
509 
510 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
511 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
512 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
513 	}
514 	return vmx->exit_qualification;
515 }
516 
517 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
518 {
519 	struct vcpu_vmx *vmx = to_vmx(vcpu);
520 
521 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
522 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
523 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
524 	}
525 	return vmx->exit_intr_info;
526 }
527 
528 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
529 void free_vmcs(struct vmcs *vmcs);
530 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
531 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
532 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
533 
534 static inline struct vmcs *alloc_vmcs(bool shadow)
535 {
536 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
537 			      GFP_KERNEL_ACCOUNT);
538 }
539 
540 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
541 {
542 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
543 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
544 }
545 
546 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
547 {
548 	return vmx->secondary_exec_control &
549 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
550 }
551 
552 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
553 {
554 	return !enable_ept || cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
555 }
556 
557 void dump_vmcs(void);
558 
559 #endif /* __KVM_X86_VMX_H */
560