xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision b8265621)
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 
15 extern const u32 vmx_msr_index[];
16 
17 #define MSR_TYPE_R	1
18 #define MSR_TYPE_W	2
19 #define MSR_TYPE_RW	3
20 
21 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
22 
23 #ifdef CONFIG_X86_64
24 #define NR_SHARED_MSRS	7
25 #else
26 #define NR_SHARED_MSRS	4
27 #endif
28 
29 #define NR_LOADSTORE_MSRS 8
30 
31 struct vmx_msrs {
32 	unsigned int		nr;
33 	struct vmx_msr_entry	val[NR_LOADSTORE_MSRS];
34 };
35 
36 struct shared_msr_entry {
37 	unsigned index;
38 	u64 data;
39 	u64 mask;
40 };
41 
42 enum segment_cache_field {
43 	SEG_FIELD_SEL = 0,
44 	SEG_FIELD_BASE = 1,
45 	SEG_FIELD_LIMIT = 2,
46 	SEG_FIELD_AR = 3,
47 
48 	SEG_FIELD_NR = 4
49 };
50 
51 /* Posted-Interrupt Descriptor */
52 struct pi_desc {
53 	u32 pir[8];     /* Posted interrupt requested */
54 	union {
55 		struct {
56 				/* bit 256 - Outstanding Notification */
57 			u16	on	: 1,
58 				/* bit 257 - Suppress Notification */
59 				sn	: 1,
60 				/* bit 271:258 - Reserved */
61 				rsvd_1	: 14;
62 				/* bit 279:272 - Notification Vector */
63 			u8	nv;
64 				/* bit 287:280 - Reserved */
65 			u8	rsvd_2;
66 				/* bit 319:288 - Notification Destination */
67 			u32	ndst;
68 		};
69 		u64 control;
70 	};
71 	u32 rsvd[6];
72 } __aligned(64);
73 
74 #define RTIT_ADDR_RANGE		4
75 
76 struct pt_ctx {
77 	u64 ctl;
78 	u64 status;
79 	u64 output_base;
80 	u64 output_mask;
81 	u64 cr3_match;
82 	u64 addr_a[RTIT_ADDR_RANGE];
83 	u64 addr_b[RTIT_ADDR_RANGE];
84 };
85 
86 struct pt_desc {
87 	u64 ctl_bitmask;
88 	u32 addr_range;
89 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
90 	struct pt_ctx host;
91 	struct pt_ctx guest;
92 };
93 
94 /*
95  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
96  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
97  */
98 struct nested_vmx {
99 	/* Has the level1 guest done vmxon? */
100 	bool vmxon;
101 	gpa_t vmxon_ptr;
102 	bool pml_full;
103 
104 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
105 	gpa_t current_vmptr;
106 	/*
107 	 * Cache of the guest's VMCS, existing outside of guest memory.
108 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
109 	 * memory during VMCLEAR and VMPTRLD.
110 	 */
111 	struct vmcs12 *cached_vmcs12;
112 	/*
113 	 * Cache of the guest's shadow VMCS, existing outside of guest
114 	 * memory. Loaded from guest memory during VM entry. Flushed
115 	 * to guest memory during VM exit.
116 	 */
117 	struct vmcs12 *cached_shadow_vmcs12;
118 
119 	/*
120 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
121 	 * with the data held by struct vmcs12.
122 	 */
123 	bool need_vmcs12_to_shadow_sync;
124 	bool dirty_vmcs12;
125 
126 	/*
127 	 * Indicates lazily loaded guest state has not yet been decached from
128 	 * vmcs02.
129 	 */
130 	bool need_sync_vmcs02_to_vmcs12_rare;
131 
132 	/*
133 	 * vmcs02 has been initialized, i.e. state that is constant for
134 	 * vmcs02 has been written to the backing VMCS.  Initialization
135 	 * is delayed until L1 actually attempts to run a nested VM.
136 	 */
137 	bool vmcs02_initialized;
138 
139 	bool change_vmcs01_virtual_apic_mode;
140 	bool reload_vmcs01_apic_access_page;
141 
142 	/*
143 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
144 	 * use it. However, VMX features available to L1 will be limited based
145 	 * on what the enlightened VMCS supports.
146 	 */
147 	bool enlightened_vmcs_enabled;
148 
149 	/* L2 must run next, and mustn't decide to exit to L1. */
150 	bool nested_run_pending;
151 
152 	/* Pending MTF VM-exit into L1.  */
153 	bool mtf_pending;
154 
155 	struct loaded_vmcs vmcs02;
156 
157 	/*
158 	 * Guest pages referred to in the vmcs02 with host-physical
159 	 * pointers, so we must keep them pinned while L2 runs.
160 	 */
161 	struct page *apic_access_page;
162 	struct kvm_host_map virtual_apic_map;
163 	struct kvm_host_map pi_desc_map;
164 
165 	struct kvm_host_map msr_bitmap_map;
166 
167 	struct pi_desc *pi_desc;
168 	bool pi_pending;
169 	u16 posted_intr_nv;
170 
171 	struct hrtimer preemption_timer;
172 	u64 preemption_timer_deadline;
173 	bool has_preemption_timer_deadline;
174 	bool preemption_timer_expired;
175 
176 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
177 	u64 vmcs01_debugctl;
178 	u64 vmcs01_guest_bndcfgs;
179 
180 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
181 	int l1_tpr_threshold;
182 
183 	u16 vpid02;
184 	u16 last_vpid;
185 
186 	struct nested_vmx_msrs msrs;
187 
188 	/* SMM related state */
189 	struct {
190 		/* in VMX operation on SMM entry? */
191 		bool vmxon;
192 		/* in guest mode on SMM entry? */
193 		bool guest_mode;
194 	} smm;
195 
196 	gpa_t hv_evmcs_vmptr;
197 	struct kvm_host_map hv_evmcs_map;
198 	struct hv_enlightened_vmcs *hv_evmcs;
199 };
200 
201 struct vcpu_vmx {
202 	struct kvm_vcpu       vcpu;
203 	u8                    fail;
204 	u8		      msr_bitmap_mode;
205 
206 	/*
207 	 * If true, host state has been stored in vmx->loaded_vmcs for
208 	 * the CPU registers that only need to be switched when transitioning
209 	 * to/from the kernel, and the registers have been loaded with guest
210 	 * values.  If false, host state is loaded in the CPU registers
211 	 * and vmx->loaded_vmcs->host_state is invalid.
212 	 */
213 	bool		      guest_state_loaded;
214 
215 	unsigned long         exit_qualification;
216 	u32                   exit_intr_info;
217 	u32                   idt_vectoring_info;
218 	ulong                 rflags;
219 
220 	struct shared_msr_entry guest_msrs[NR_SHARED_MSRS];
221 	int                   nmsrs;
222 	int                   save_nmsrs;
223 	bool                  guest_msrs_ready;
224 #ifdef CONFIG_X86_64
225 	u64		      msr_host_kernel_gs_base;
226 	u64		      msr_guest_kernel_gs_base;
227 #endif
228 
229 	u64		      spec_ctrl;
230 	u32		      msr_ia32_umwait_control;
231 
232 	u32 secondary_exec_control;
233 
234 	/*
235 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
236 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
237 	 * guest (L2), it points to a different VMCS.
238 	 */
239 	struct loaded_vmcs    vmcs01;
240 	struct loaded_vmcs   *loaded_vmcs;
241 
242 	struct msr_autoload {
243 		struct vmx_msrs guest;
244 		struct vmx_msrs host;
245 	} msr_autoload;
246 
247 	struct msr_autostore {
248 		struct vmx_msrs guest;
249 	} msr_autostore;
250 
251 	struct {
252 		int vm86_active;
253 		ulong save_rflags;
254 		struct kvm_segment segs[8];
255 	} rmode;
256 	struct {
257 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
258 		struct kvm_save_segment {
259 			u16 selector;
260 			unsigned long base;
261 			u32 limit;
262 			u32 ar;
263 		} seg[8];
264 	} segment_cache;
265 	int vpid;
266 	bool emulation_required;
267 
268 	u32 exit_reason;
269 
270 	/* Posted interrupt descriptor */
271 	struct pi_desc pi_desc;
272 
273 	/* Support for a guest hypervisor (nested VMX) */
274 	struct nested_vmx nested;
275 
276 	/* Dynamic PLE window. */
277 	unsigned int ple_window;
278 	bool ple_window_dirty;
279 
280 	bool req_immediate_exit;
281 
282 	/* Support for PML */
283 #define PML_ENTITY_NUM		512
284 	struct page *pml_pg;
285 
286 	/* apic deadline value in host tsc */
287 	u64 hv_deadline_tsc;
288 
289 	u64 current_tsc_ratio;
290 
291 	unsigned long host_debugctlmsr;
292 
293 	/*
294 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
295 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
296 	 * in msr_ia32_feature_control_valid_bits.
297 	 */
298 	u64 msr_ia32_feature_control;
299 	u64 msr_ia32_feature_control_valid_bits;
300 	u64 ept_pointer;
301 
302 	struct pt_desc pt_desc;
303 };
304 
305 enum ept_pointers_status {
306 	EPT_POINTERS_CHECK = 0,
307 	EPT_POINTERS_MATCH = 1,
308 	EPT_POINTERS_MISMATCH = 2
309 };
310 
311 struct kvm_vmx {
312 	struct kvm kvm;
313 
314 	unsigned int tss_addr;
315 	bool ept_identity_pagetable_done;
316 	gpa_t ept_identity_map_addr;
317 
318 	enum ept_pointers_status ept_pointers_match;
319 	spinlock_t ept_pointer_lock;
320 };
321 
322 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
323 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
324 			struct loaded_vmcs *buddy);
325 int allocate_vpid(void);
326 void free_vpid(int vpid);
327 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
328 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
329 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
330 			unsigned long fs_base, unsigned long gs_base);
331 int vmx_get_cpl(struct kvm_vcpu *vcpu);
332 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
333 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
334 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
335 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
336 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
337 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
338 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
339 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
340 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long cr3);
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 void update_exception_bitmap(struct kvm_vcpu *vcpu);
346 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
347 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
348 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
349 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
350 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
351 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
352 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
353 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
354 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
355 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr);
356 int vmx_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
357 			      struct x86_exception *e);
358 
359 #define POSTED_INTR_ON  0
360 #define POSTED_INTR_SN  1
361 
362 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
363 {
364 	return test_and_set_bit(POSTED_INTR_ON,
365 			(unsigned long *)&pi_desc->control);
366 }
367 
368 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
369 {
370 	return test_and_clear_bit(POSTED_INTR_ON,
371 			(unsigned long *)&pi_desc->control);
372 }
373 
374 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
375 {
376 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
377 }
378 
379 static inline bool pi_is_pir_empty(struct pi_desc *pi_desc)
380 {
381 	return bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS);
382 }
383 
384 static inline void pi_set_sn(struct pi_desc *pi_desc)
385 {
386 	set_bit(POSTED_INTR_SN,
387 		(unsigned long *)&pi_desc->control);
388 }
389 
390 static inline void pi_set_on(struct pi_desc *pi_desc)
391 {
392 	set_bit(POSTED_INTR_ON,
393 		(unsigned long *)&pi_desc->control);
394 }
395 
396 static inline void pi_clear_on(struct pi_desc *pi_desc)
397 {
398 	clear_bit(POSTED_INTR_ON,
399 		(unsigned long *)&pi_desc->control);
400 }
401 
402 static inline void pi_clear_sn(struct pi_desc *pi_desc)
403 {
404 	clear_bit(POSTED_INTR_SN,
405 		(unsigned long *)&pi_desc->control);
406 }
407 
408 static inline int pi_test_on(struct pi_desc *pi_desc)
409 {
410 	return test_bit(POSTED_INTR_ON,
411 			(unsigned long *)&pi_desc->control);
412 }
413 
414 static inline int pi_test_sn(struct pi_desc *pi_desc)
415 {
416 	return test_bit(POSTED_INTR_SN,
417 			(unsigned long *)&pi_desc->control);
418 }
419 
420 static inline u8 vmx_get_rvi(void)
421 {
422 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
423 }
424 
425 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
426 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
427 {									    \
428 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
429 		vmcs_write32(uname, val);				    \
430 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
431 	}								    \
432 }									    \
433 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
434 {									    \
435 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
436 }									    \
437 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
438 {									    \
439 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
440 }									    \
441 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
442 {									    \
443 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
444 }
445 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
446 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
447 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
448 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
449 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
450 
451 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
452 {
453 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
454 				  | (1 << VCPU_EXREG_RFLAGS)
455 				  | (1 << VCPU_EXREG_PDPTR)
456 				  | (1 << VCPU_EXREG_SEGMENTS)
457 				  | (1 << VCPU_EXREG_CR0)
458 				  | (1 << VCPU_EXREG_CR3)
459 				  | (1 << VCPU_EXREG_CR4)
460 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
461 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
462 	vcpu->arch.regs_dirty = 0;
463 }
464 
465 static inline u32 vmx_vmentry_ctrl(void)
466 {
467 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
468 	if (vmx_pt_mode_is_system())
469 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
470 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
471 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
472 	return vmentry_ctrl &
473 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
474 }
475 
476 static inline u32 vmx_vmexit_ctrl(void)
477 {
478 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
479 	if (vmx_pt_mode_is_system())
480 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
481 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
482 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
483 	return vmexit_ctrl &
484 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
485 }
486 
487 u32 vmx_exec_control(struct vcpu_vmx *vmx);
488 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
489 
490 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
491 {
492 	return container_of(kvm, struct kvm_vmx, kvm);
493 }
494 
495 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
496 {
497 	return container_of(vcpu, struct vcpu_vmx, vcpu);
498 }
499 
500 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
501 {
502 	return &(to_vmx(vcpu)->pi_desc);
503 }
504 
505 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
506 {
507 	struct vcpu_vmx *vmx = to_vmx(vcpu);
508 
509 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
510 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
511 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
512 	}
513 	return vmx->exit_qualification;
514 }
515 
516 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
517 {
518 	struct vcpu_vmx *vmx = to_vmx(vcpu);
519 
520 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
521 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
522 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
523 	}
524 	return vmx->exit_intr_info;
525 }
526 
527 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
528 void free_vmcs(struct vmcs *vmcs);
529 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
530 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
531 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
532 
533 static inline struct vmcs *alloc_vmcs(bool shadow)
534 {
535 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
536 			      GFP_KERNEL_ACCOUNT);
537 }
538 
539 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
540 
541 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
542 {
543 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
544 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
545 }
546 
547 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
548 {
549 	return vmx->secondary_exec_control &
550 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
551 }
552 
553 void dump_vmcs(void);
554 
555 #endif /* __KVM_X86_VMX_H */
556