xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision 0661cb2a)
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 #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 MAX_NR_USER_RETURN_MSRS	7
25 #else
26 #define MAX_NR_USER_RETURN_MSRS	4
27 #endif
28 
29 #define MAX_NR_LOADSTORE_MSRS	8
30 
31 struct vmx_msrs {
32 	unsigned int		nr;
33 	struct vmx_msr_entry	val[MAX_NR_LOADSTORE_MSRS];
34 };
35 
36 struct vmx_uret_msr {
37 	bool load_into_hardware;
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 #define RTIT_ADDR_RANGE		4
52 
53 struct pt_ctx {
54 	u64 ctl;
55 	u64 status;
56 	u64 output_base;
57 	u64 output_mask;
58 	u64 cr3_match;
59 	u64 addr_a[RTIT_ADDR_RANGE];
60 	u64 addr_b[RTIT_ADDR_RANGE];
61 };
62 
63 struct pt_desc {
64 	u64 ctl_bitmask;
65 	u32 addr_range;
66 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
67 	struct pt_ctx host;
68 	struct pt_ctx guest;
69 };
70 
71 union vmx_exit_reason {
72 	struct {
73 		u32	basic			: 16;
74 		u32	reserved16		: 1;
75 		u32	reserved17		: 1;
76 		u32	reserved18		: 1;
77 		u32	reserved19		: 1;
78 		u32	reserved20		: 1;
79 		u32	reserved21		: 1;
80 		u32	reserved22		: 1;
81 		u32	reserved23		: 1;
82 		u32	reserved24		: 1;
83 		u32	reserved25		: 1;
84 		u32	bus_lock_detected	: 1;
85 		u32	enclave_mode		: 1;
86 		u32	smi_pending_mtf		: 1;
87 		u32	smi_from_vmx_root	: 1;
88 		u32	reserved30		: 1;
89 		u32	failed_vmentry		: 1;
90 	};
91 	u32 full;
92 };
93 
94 #define vcpu_to_lbr_desc(vcpu) (&to_vmx(vcpu)->lbr_desc)
95 #define vcpu_to_lbr_records(vcpu) (&to_vmx(vcpu)->lbr_desc.records)
96 
97 bool intel_pmu_lbr_is_compatible(struct kvm_vcpu *vcpu);
98 bool intel_pmu_lbr_is_enabled(struct kvm_vcpu *vcpu);
99 
100 int intel_pmu_create_guest_lbr_event(struct kvm_vcpu *vcpu);
101 void vmx_passthrough_lbr_msrs(struct kvm_vcpu *vcpu);
102 
103 struct lbr_desc {
104 	/* Basic info about guest LBR records. */
105 	struct x86_pmu_lbr records;
106 
107 	/*
108 	 * Emulate LBR feature via passthrough LBR registers when the
109 	 * per-vcpu guest LBR event is scheduled on the current pcpu.
110 	 *
111 	 * The records may be inaccurate if the host reclaims the LBR.
112 	 */
113 	struct perf_event *event;
114 
115 	/* True if LBRs are marked as not intercepted in the MSR bitmap */
116 	bool msr_passthrough;
117 };
118 
119 /*
120  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
121  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
122  */
123 struct nested_vmx {
124 	/* Has the level1 guest done vmxon? */
125 	bool vmxon;
126 	gpa_t vmxon_ptr;
127 	bool pml_full;
128 
129 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
130 	gpa_t current_vmptr;
131 	/*
132 	 * Cache of the guest's VMCS, existing outside of guest memory.
133 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
134 	 * memory during VMCLEAR and VMPTRLD.
135 	 */
136 	struct vmcs12 *cached_vmcs12;
137 	/*
138 	 * Cache of the guest's shadow VMCS, existing outside of guest
139 	 * memory. Loaded from guest memory during VM entry. Flushed
140 	 * to guest memory during VM exit.
141 	 */
142 	struct vmcs12 *cached_shadow_vmcs12;
143 
144 	/*
145 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
146 	 * with the data held by struct vmcs12.
147 	 */
148 	bool need_vmcs12_to_shadow_sync;
149 	bool dirty_vmcs12;
150 
151 	/*
152 	 * Indicates lazily loaded guest state has not yet been decached from
153 	 * vmcs02.
154 	 */
155 	bool need_sync_vmcs02_to_vmcs12_rare;
156 
157 	/*
158 	 * vmcs02 has been initialized, i.e. state that is constant for
159 	 * vmcs02 has been written to the backing VMCS.  Initialization
160 	 * is delayed until L1 actually attempts to run a nested VM.
161 	 */
162 	bool vmcs02_initialized;
163 
164 	bool change_vmcs01_virtual_apic_mode;
165 	bool reload_vmcs01_apic_access_page;
166 	bool update_vmcs01_cpu_dirty_logging;
167 
168 	/*
169 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
170 	 * use it. However, VMX features available to L1 will be limited based
171 	 * on what the enlightened VMCS supports.
172 	 */
173 	bool enlightened_vmcs_enabled;
174 
175 	/* L2 must run next, and mustn't decide to exit to L1. */
176 	bool nested_run_pending;
177 
178 	/* Pending MTF VM-exit into L1.  */
179 	bool mtf_pending;
180 
181 	struct loaded_vmcs vmcs02;
182 
183 	/*
184 	 * Guest pages referred to in the vmcs02 with host-physical
185 	 * pointers, so we must keep them pinned while L2 runs.
186 	 */
187 	struct page *apic_access_page;
188 	struct kvm_host_map virtual_apic_map;
189 	struct kvm_host_map pi_desc_map;
190 
191 	struct kvm_host_map msr_bitmap_map;
192 
193 	struct pi_desc *pi_desc;
194 	bool pi_pending;
195 	u16 posted_intr_nv;
196 
197 	struct hrtimer preemption_timer;
198 	u64 preemption_timer_deadline;
199 	bool has_preemption_timer_deadline;
200 	bool preemption_timer_expired;
201 
202 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
203 	u64 vmcs01_debugctl;
204 	u64 vmcs01_guest_bndcfgs;
205 
206 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
207 	int l1_tpr_threshold;
208 
209 	u16 vpid02;
210 	u16 last_vpid;
211 
212 	struct nested_vmx_msrs msrs;
213 
214 	/* SMM related state */
215 	struct {
216 		/* in VMX operation on SMM entry? */
217 		bool vmxon;
218 		/* in guest mode on SMM entry? */
219 		bool guest_mode;
220 	} smm;
221 
222 	gpa_t hv_evmcs_vmptr;
223 	struct kvm_host_map hv_evmcs_map;
224 	struct hv_enlightened_vmcs *hv_evmcs;
225 };
226 
227 struct vcpu_vmx {
228 	struct kvm_vcpu       vcpu;
229 	u8                    fail;
230 	u8		      msr_bitmap_mode;
231 
232 	/*
233 	 * If true, host state has been stored in vmx->loaded_vmcs for
234 	 * the CPU registers that only need to be switched when transitioning
235 	 * to/from the kernel, and the registers have been loaded with guest
236 	 * values.  If false, host state is loaded in the CPU registers
237 	 * and vmx->loaded_vmcs->host_state is invalid.
238 	 */
239 	bool		      guest_state_loaded;
240 
241 	unsigned long         exit_qualification;
242 	u32                   exit_intr_info;
243 	u32                   idt_vectoring_info;
244 	ulong                 rflags;
245 
246 	/*
247 	 * User return MSRs are always emulated when enabled in the guest, but
248 	 * only loaded into hardware when necessary, e.g. SYSCALL #UDs outside
249 	 * of 64-bit mode or if EFER.SCE=1, thus the SYSCALL MSRs don't need to
250 	 * be loaded into hardware if those conditions aren't met.
251 	 * nr_active_uret_msrs tracks the number of MSRs that need to be loaded
252 	 * into hardware when running the guest.  guest_uret_msrs[] is resorted
253 	 * whenever the number of "active" uret MSRs is modified.
254 	 */
255 	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
256 	int                   nr_active_uret_msrs;
257 	bool                  guest_uret_msrs_loaded;
258 #ifdef CONFIG_X86_64
259 	u64		      msr_host_kernel_gs_base;
260 	u64		      msr_guest_kernel_gs_base;
261 #endif
262 
263 	u64		      spec_ctrl;
264 	u32		      msr_ia32_umwait_control;
265 
266 	u32 secondary_exec_control;
267 
268 	/*
269 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
270 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
271 	 * guest (L2), it points to a different VMCS.
272 	 */
273 	struct loaded_vmcs    vmcs01;
274 	struct loaded_vmcs   *loaded_vmcs;
275 
276 	struct msr_autoload {
277 		struct vmx_msrs guest;
278 		struct vmx_msrs host;
279 	} msr_autoload;
280 
281 	struct msr_autostore {
282 		struct vmx_msrs guest;
283 	} msr_autostore;
284 
285 	struct {
286 		int vm86_active;
287 		ulong save_rflags;
288 		struct kvm_segment segs[8];
289 	} rmode;
290 	struct {
291 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
292 		struct kvm_save_segment {
293 			u16 selector;
294 			unsigned long base;
295 			u32 limit;
296 			u32 ar;
297 		} seg[8];
298 	} segment_cache;
299 	int vpid;
300 	bool emulation_required;
301 
302 	union vmx_exit_reason exit_reason;
303 
304 	/* Posted interrupt descriptor */
305 	struct pi_desc pi_desc;
306 
307 	/* Support for a guest hypervisor (nested VMX) */
308 	struct nested_vmx nested;
309 
310 	/* Dynamic PLE window. */
311 	unsigned int ple_window;
312 	bool ple_window_dirty;
313 
314 	bool req_immediate_exit;
315 
316 	/* Support for PML */
317 #define PML_ENTITY_NUM		512
318 	struct page *pml_pg;
319 
320 	/* apic deadline value in host tsc */
321 	u64 hv_deadline_tsc;
322 
323 	unsigned long host_debugctlmsr;
324 
325 	/*
326 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
327 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
328 	 * in msr_ia32_feature_control_valid_bits.
329 	 */
330 	u64 msr_ia32_feature_control;
331 	u64 msr_ia32_feature_control_valid_bits;
332 	/* SGX Launch Control public key hash */
333 	u64 msr_ia32_sgxlepubkeyhash[4];
334 
335 	struct pt_desc pt_desc;
336 	struct lbr_desc lbr_desc;
337 
338 	/* Save desired MSR intercept (read: pass-through) state */
339 #define MAX_POSSIBLE_PASSTHROUGH_MSRS	13
340 	struct {
341 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
342 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
343 	} shadow_msr_intercept;
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 
354 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
355 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
356 			struct loaded_vmcs *buddy);
357 int allocate_vpid(void);
358 void free_vpid(int vpid);
359 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
360 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
361 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
362 			unsigned long fs_base, unsigned long gs_base);
363 int vmx_get_cpl(struct kvm_vcpu *vcpu);
364 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
365 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
366 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
367 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
368 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
369 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
370 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
371 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
372 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
373 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
374 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
375 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level);
376 
377 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu);
378 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu);
379 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
380 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
381 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
382 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
383 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
384 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
385 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
386 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
387 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
388 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
389 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
390 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
391 
392 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
393 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
394 
395 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu);
396 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu);
397 
398 static inline void vmx_set_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr,
399 					     int type, bool value)
400 {
401 	if (value)
402 		vmx_enable_intercept_for_msr(vcpu, msr, type);
403 	else
404 		vmx_disable_intercept_for_msr(vcpu, msr, type);
405 }
406 
407 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu);
408 
409 static inline u8 vmx_get_rvi(void)
410 {
411 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
412 }
413 
414 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
415 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
416 {									    \
417 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
418 		vmcs_write32(uname, val);				    \
419 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
420 	}								    \
421 }									    \
422 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
423 {									    \
424 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
425 }									    \
426 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
427 {									    \
428 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
429 }									    \
430 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
431 {									    \
432 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
433 }
434 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
435 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
436 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
437 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
438 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
439 
440 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
441 {
442 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
443 				  | (1 << VCPU_EXREG_RFLAGS)
444 				  | (1 << VCPU_EXREG_PDPTR)
445 				  | (1 << VCPU_EXREG_SEGMENTS)
446 				  | (1 << VCPU_EXREG_CR0)
447 				  | (1 << VCPU_EXREG_CR3)
448 				  | (1 << VCPU_EXREG_CR4)
449 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
450 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
451 	vcpu->arch.regs_dirty = 0;
452 }
453 
454 static inline u32 vmx_vmentry_ctrl(void)
455 {
456 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
457 	if (vmx_pt_mode_is_system())
458 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
459 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
460 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
461 	return vmentry_ctrl &
462 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
463 }
464 
465 static inline u32 vmx_vmexit_ctrl(void)
466 {
467 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
468 	if (vmx_pt_mode_is_system())
469 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
470 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
471 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
472 	return vmexit_ctrl &
473 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
474 }
475 
476 u32 vmx_exec_control(struct vcpu_vmx *vmx);
477 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
478 
479 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
480 {
481 	return container_of(kvm, struct kvm_vmx, kvm);
482 }
483 
484 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
485 {
486 	return container_of(vcpu, struct vcpu_vmx, vcpu);
487 }
488 
489 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
490 {
491 	struct vcpu_vmx *vmx = to_vmx(vcpu);
492 
493 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
494 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
495 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
496 	}
497 	return vmx->exit_qualification;
498 }
499 
500 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
501 {
502 	struct vcpu_vmx *vmx = to_vmx(vcpu);
503 
504 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
505 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
506 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
507 	}
508 	return vmx->exit_intr_info;
509 }
510 
511 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
512 void free_vmcs(struct vmcs *vmcs);
513 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
514 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
515 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
516 
517 static inline struct vmcs *alloc_vmcs(bool shadow)
518 {
519 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
520 			      GFP_KERNEL_ACCOUNT);
521 }
522 
523 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
524 {
525 	return vmx->secondary_exec_control &
526 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
527 }
528 
529 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
530 {
531 	if (!enable_ept)
532 		return true;
533 
534 	return allow_smaller_maxphyaddr && cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
535 }
536 
537 static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
538 {
539 	return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
540 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
541 	    SECONDARY_EXEC_UNRESTRICTED_GUEST));
542 }
543 
544 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
545 static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
546 {
547 	return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
548 }
549 
550 void dump_vmcs(struct kvm_vcpu *vcpu);
551 
552 #endif /* __KVM_X86_VMX_H */
553