xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision aa0dc6a7)
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 	bool load_into_hardware;
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 	bool update_vmcs01_cpu_dirty_logging;
169 
170 	/*
171 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
172 	 * use it. However, VMX features available to L1 will be limited based
173 	 * on what the enlightened VMCS supports.
174 	 */
175 	bool enlightened_vmcs_enabled;
176 
177 	/* L2 must run next, and mustn't decide to exit to L1. */
178 	bool nested_run_pending;
179 
180 	/* Pending MTF VM-exit into L1.  */
181 	bool mtf_pending;
182 
183 	struct loaded_vmcs vmcs02;
184 
185 	/*
186 	 * Guest pages referred to in the vmcs02 with host-physical
187 	 * pointers, so we must keep them pinned while L2 runs.
188 	 */
189 	struct page *apic_access_page;
190 	struct kvm_host_map virtual_apic_map;
191 	struct kvm_host_map pi_desc_map;
192 
193 	struct kvm_host_map msr_bitmap_map;
194 
195 	struct pi_desc *pi_desc;
196 	bool pi_pending;
197 	u16 posted_intr_nv;
198 
199 	struct hrtimer preemption_timer;
200 	u64 preemption_timer_deadline;
201 	bool has_preemption_timer_deadline;
202 	bool preemption_timer_expired;
203 
204 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
205 	u64 vmcs01_debugctl;
206 	u64 vmcs01_guest_bndcfgs;
207 
208 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
209 	int l1_tpr_threshold;
210 
211 	u16 vpid02;
212 	u16 last_vpid;
213 
214 	struct nested_vmx_msrs msrs;
215 
216 	/* SMM related state */
217 	struct {
218 		/* in VMX operation on SMM entry? */
219 		bool vmxon;
220 		/* in guest mode on SMM entry? */
221 		bool guest_mode;
222 	} smm;
223 
224 	gpa_t hv_evmcs_vmptr;
225 	struct kvm_host_map hv_evmcs_map;
226 	struct hv_enlightened_vmcs *hv_evmcs;
227 };
228 
229 struct vcpu_vmx {
230 	struct kvm_vcpu       vcpu;
231 	u8                    fail;
232 	u8		      msr_bitmap_mode;
233 
234 	/*
235 	 * If true, host state has been stored in vmx->loaded_vmcs for
236 	 * the CPU registers that only need to be switched when transitioning
237 	 * to/from the kernel, and the registers have been loaded with guest
238 	 * values.  If false, host state is loaded in the CPU registers
239 	 * and vmx->loaded_vmcs->host_state is invalid.
240 	 */
241 	bool		      guest_state_loaded;
242 
243 	unsigned long         exit_qualification;
244 	u32                   exit_intr_info;
245 	u32                   idt_vectoring_info;
246 	ulong                 rflags;
247 
248 	/*
249 	 * User return MSRs are always emulated when enabled in the guest, but
250 	 * only loaded into hardware when necessary, e.g. SYSCALL #UDs outside
251 	 * of 64-bit mode or if EFER.SCE=1, thus the SYSCALL MSRs don't need to
252 	 * be loaded into hardware if those conditions aren't met.
253 	 * nr_active_uret_msrs tracks the number of MSRs that need to be loaded
254 	 * into hardware when running the guest.  guest_uret_msrs[] is resorted
255 	 * whenever the number of "active" uret MSRs is modified.
256 	 */
257 	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
258 	int                   nr_active_uret_msrs;
259 	bool                  guest_uret_msrs_loaded;
260 #ifdef CONFIG_X86_64
261 	u64		      msr_host_kernel_gs_base;
262 	u64		      msr_guest_kernel_gs_base;
263 #endif
264 
265 	u64		      spec_ctrl;
266 	u32		      msr_ia32_umwait_control;
267 
268 	u32 secondary_exec_control;
269 
270 	/*
271 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
272 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
273 	 * guest (L2), it points to a different VMCS.
274 	 */
275 	struct loaded_vmcs    vmcs01;
276 	struct loaded_vmcs   *loaded_vmcs;
277 
278 	struct msr_autoload {
279 		struct vmx_msrs guest;
280 		struct vmx_msrs host;
281 	} msr_autoload;
282 
283 	struct msr_autostore {
284 		struct vmx_msrs guest;
285 	} msr_autostore;
286 
287 	struct {
288 		int vm86_active;
289 		ulong save_rflags;
290 		struct kvm_segment segs[8];
291 	} rmode;
292 	struct {
293 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
294 		struct kvm_save_segment {
295 			u16 selector;
296 			unsigned long base;
297 			u32 limit;
298 			u32 ar;
299 		} seg[8];
300 	} segment_cache;
301 	int vpid;
302 	bool emulation_required;
303 
304 	union vmx_exit_reason exit_reason;
305 
306 	/* Posted interrupt descriptor */
307 	struct pi_desc pi_desc;
308 
309 	/* Support for a guest hypervisor (nested VMX) */
310 	struct nested_vmx nested;
311 
312 	/* Dynamic PLE window. */
313 	unsigned int ple_window;
314 	bool ple_window_dirty;
315 
316 	bool req_immediate_exit;
317 
318 	/* Support for PML */
319 #define PML_ENTITY_NUM		512
320 	struct page *pml_pg;
321 
322 	/* apic deadline value in host tsc */
323 	u64 hv_deadline_tsc;
324 
325 	unsigned long host_debugctlmsr;
326 
327 	/*
328 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
329 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
330 	 * in msr_ia32_feature_control_valid_bits.
331 	 */
332 	u64 msr_ia32_feature_control;
333 	u64 msr_ia32_feature_control_valid_bits;
334 	/* SGX Launch Control public key hash */
335 	u64 msr_ia32_sgxlepubkeyhash[4];
336 
337 	struct pt_desc pt_desc;
338 	struct lbr_desc lbr_desc;
339 
340 	/* Save desired MSR intercept (read: pass-through) state */
341 #define MAX_POSSIBLE_PASSTHROUGH_MSRS	13
342 	struct {
343 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
344 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
345 	} shadow_msr_intercept;
346 };
347 
348 struct kvm_vmx {
349 	struct kvm kvm;
350 
351 	unsigned int tss_addr;
352 	bool ept_identity_pagetable_done;
353 	gpa_t ept_identity_map_addr;
354 };
355 
356 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
357 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
358 			struct loaded_vmcs *buddy);
359 int allocate_vpid(void);
360 void free_vpid(int vpid);
361 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
362 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
363 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
364 			unsigned long fs_base, unsigned long gs_base);
365 int vmx_get_cpl(struct kvm_vcpu *vcpu);
366 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
367 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
368 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
369 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
370 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
371 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
372 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
373 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
374 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
375 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
376 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
377 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level);
378 
379 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu);
380 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu);
381 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
382 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
383 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
384 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
385 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
386 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
387 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
388 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
389 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
390 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
391 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
392 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
393 
394 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
395 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
396 
397 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu);
398 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu);
399 
400 static inline void vmx_set_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr,
401 					     int type, bool value)
402 {
403 	if (value)
404 		vmx_enable_intercept_for_msr(vcpu, msr, type);
405 	else
406 		vmx_disable_intercept_for_msr(vcpu, msr, type);
407 }
408 
409 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu);
410 
411 static inline u8 vmx_get_rvi(void)
412 {
413 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
414 }
415 
416 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
417 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
418 {									    \
419 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
420 		vmcs_write32(uname, val);				    \
421 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
422 	}								    \
423 }									    \
424 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
425 {									    \
426 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
427 }									    \
428 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
429 {									    \
430 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
431 }									    \
432 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
433 {									    \
434 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
435 }
436 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
437 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
438 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
439 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
440 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
441 
442 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
443 {
444 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
445 				  | (1 << VCPU_EXREG_RFLAGS)
446 				  | (1 << VCPU_EXREG_PDPTR)
447 				  | (1 << VCPU_EXREG_SEGMENTS)
448 				  | (1 << VCPU_EXREG_CR0)
449 				  | (1 << VCPU_EXREG_CR3)
450 				  | (1 << VCPU_EXREG_CR4)
451 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
452 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
453 	vcpu->arch.regs_dirty = 0;
454 }
455 
456 static inline u32 vmx_vmentry_ctrl(void)
457 {
458 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
459 	if (vmx_pt_mode_is_system())
460 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
461 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
462 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
463 	return vmentry_ctrl &
464 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
465 }
466 
467 static inline u32 vmx_vmexit_ctrl(void)
468 {
469 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
470 	if (vmx_pt_mode_is_system())
471 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
472 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
473 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
474 	return vmexit_ctrl &
475 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
476 }
477 
478 u32 vmx_exec_control(struct vcpu_vmx *vmx);
479 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
480 
481 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
482 {
483 	return container_of(kvm, struct kvm_vmx, kvm);
484 }
485 
486 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
487 {
488 	return container_of(vcpu, struct vcpu_vmx, vcpu);
489 }
490 
491 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
492 {
493 	struct vcpu_vmx *vmx = to_vmx(vcpu);
494 
495 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
496 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
497 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
498 	}
499 	return vmx->exit_qualification;
500 }
501 
502 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
503 {
504 	struct vcpu_vmx *vmx = to_vmx(vcpu);
505 
506 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
507 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
508 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
509 	}
510 	return vmx->exit_intr_info;
511 }
512 
513 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
514 void free_vmcs(struct vmcs *vmcs);
515 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
516 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
517 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
518 
519 static inline struct vmcs *alloc_vmcs(bool shadow)
520 {
521 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
522 			      GFP_KERNEL_ACCOUNT);
523 }
524 
525 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
526 {
527 	return vmx->secondary_exec_control &
528 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
529 }
530 
531 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
532 {
533 	if (!enable_ept)
534 		return true;
535 
536 	return allow_smaller_maxphyaddr && cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
537 }
538 
539 static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
540 {
541 	return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
542 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
543 	    SECONDARY_EXEC_UNRESTRICTED_GUEST));
544 }
545 
546 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
547 static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
548 {
549 	return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
550 }
551 
552 void dump_vmcs(struct kvm_vcpu *vcpu);
553 
554 #endif /* __KVM_X86_VMX_H */
555