xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision 74a22e8f)
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 "ops.h"
12 #include "vmcs.h"
13 
14 extern const u32 vmx_msr_index[];
15 extern u64 host_efer;
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 #define NR_AUTOLOAD_MSRS 8
24 
25 struct vmx_msrs {
26 	unsigned int		nr;
27 	struct vmx_msr_entry	val[NR_AUTOLOAD_MSRS];
28 };
29 
30 struct shared_msr_entry {
31 	unsigned index;
32 	u64 data;
33 	u64 mask;
34 };
35 
36 enum segment_cache_field {
37 	SEG_FIELD_SEL = 0,
38 	SEG_FIELD_BASE = 1,
39 	SEG_FIELD_LIMIT = 2,
40 	SEG_FIELD_AR = 3,
41 
42 	SEG_FIELD_NR = 4
43 };
44 
45 /* Posted-Interrupt Descriptor */
46 struct pi_desc {
47 	u32 pir[8];     /* Posted interrupt requested */
48 	union {
49 		struct {
50 				/* bit 256 - Outstanding Notification */
51 			u16	on	: 1,
52 				/* bit 257 - Suppress Notification */
53 				sn	: 1,
54 				/* bit 271:258 - Reserved */
55 				rsvd_1	: 14;
56 				/* bit 279:272 - Notification Vector */
57 			u8	nv;
58 				/* bit 287:280 - Reserved */
59 			u8	rsvd_2;
60 				/* bit 319:288 - Notification Destination */
61 			u32	ndst;
62 		};
63 		u64 control;
64 	};
65 	u32 rsvd[6];
66 } __aligned(64);
67 
68 #define RTIT_ADDR_RANGE		4
69 
70 struct pt_ctx {
71 	u64 ctl;
72 	u64 status;
73 	u64 output_base;
74 	u64 output_mask;
75 	u64 cr3_match;
76 	u64 addr_a[RTIT_ADDR_RANGE];
77 	u64 addr_b[RTIT_ADDR_RANGE];
78 };
79 
80 struct pt_desc {
81 	u64 ctl_bitmask;
82 	u32 addr_range;
83 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
84 	struct pt_ctx host;
85 	struct pt_ctx guest;
86 };
87 
88 /*
89  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
90  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
91  */
92 struct nested_vmx {
93 	/* Has the level1 guest done vmxon? */
94 	bool vmxon;
95 	gpa_t vmxon_ptr;
96 	bool pml_full;
97 
98 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
99 	gpa_t current_vmptr;
100 	/*
101 	 * Cache of the guest's VMCS, existing outside of guest memory.
102 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
103 	 * memory during VMCLEAR and VMPTRLD.
104 	 */
105 	struct vmcs12 *cached_vmcs12;
106 	/*
107 	 * Cache of the guest's shadow VMCS, existing outside of guest
108 	 * memory. Loaded from guest memory during VM entry. Flushed
109 	 * to guest memory during VM exit.
110 	 */
111 	struct vmcs12 *cached_shadow_vmcs12;
112 	/*
113 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
114 	 * with the data held by struct vmcs12.
115 	 */
116 	bool need_vmcs12_sync;
117 	bool dirty_vmcs12;
118 
119 	/*
120 	 * vmcs02 has been initialized, i.e. state that is constant for
121 	 * vmcs02 has been written to the backing VMCS.  Initialization
122 	 * is delayed until L1 actually attempts to run a nested VM.
123 	 */
124 	bool vmcs02_initialized;
125 
126 	bool change_vmcs01_virtual_apic_mode;
127 
128 	/*
129 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
130 	 * use it. However, VMX features available to L1 will be limited based
131 	 * on what the enlightened VMCS supports.
132 	 */
133 	bool enlightened_vmcs_enabled;
134 
135 	/* L2 must run next, and mustn't decide to exit to L1. */
136 	bool nested_run_pending;
137 
138 	struct loaded_vmcs vmcs02;
139 
140 	/*
141 	 * Guest pages referred to in the vmcs02 with host-physical
142 	 * pointers, so we must keep them pinned while L2 runs.
143 	 */
144 	struct page *apic_access_page;
145 	struct kvm_host_map virtual_apic_map;
146 	struct kvm_host_map pi_desc_map;
147 
148 	struct kvm_host_map msr_bitmap_map;
149 
150 	struct pi_desc *pi_desc;
151 	bool pi_pending;
152 	u16 posted_intr_nv;
153 
154 	struct hrtimer preemption_timer;
155 	bool preemption_timer_expired;
156 
157 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
158 	u64 vmcs01_debugctl;
159 	u64 vmcs01_guest_bndcfgs;
160 
161 	u16 vpid02;
162 	u16 last_vpid;
163 
164 	struct nested_vmx_msrs msrs;
165 
166 	/* SMM related state */
167 	struct {
168 		/* in VMX operation on SMM entry? */
169 		bool vmxon;
170 		/* in guest mode on SMM entry? */
171 		bool guest_mode;
172 	} smm;
173 
174 	gpa_t hv_evmcs_vmptr;
175 	struct kvm_host_map hv_evmcs_map;
176 	struct hv_enlightened_vmcs *hv_evmcs;
177 };
178 
179 struct vcpu_vmx {
180 	struct kvm_vcpu       vcpu;
181 	u8                    fail;
182 	u8		      msr_bitmap_mode;
183 	u32                   exit_intr_info;
184 	u32                   idt_vectoring_info;
185 	ulong                 rflags;
186 	struct shared_msr_entry *guest_msrs;
187 	int                   nmsrs;
188 	int                   save_nmsrs;
189 	bool                  guest_msrs_dirty;
190 	unsigned long	      host_idt_base;
191 #ifdef CONFIG_X86_64
192 	u64		      msr_host_kernel_gs_base;
193 	u64		      msr_guest_kernel_gs_base;
194 #endif
195 
196 	u64		      spec_ctrl;
197 
198 	u32 vm_entry_controls_shadow;
199 	u32 vm_exit_controls_shadow;
200 	u32 secondary_exec_control;
201 
202 	/*
203 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
204 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
205 	 * guest (L2), it points to a different VMCS.  loaded_cpu_state points
206 	 * to the VMCS whose state is loaded into the CPU registers that only
207 	 * need to be switched when transitioning to/from the kernel; a NULL
208 	 * value indicates that host state is loaded.
209 	 */
210 	struct loaded_vmcs    vmcs01;
211 	struct loaded_vmcs   *loaded_vmcs;
212 	struct loaded_vmcs   *loaded_cpu_state;
213 
214 	struct msr_autoload {
215 		struct vmx_msrs guest;
216 		struct vmx_msrs host;
217 	} msr_autoload;
218 
219 	struct {
220 		int vm86_active;
221 		ulong save_rflags;
222 		struct kvm_segment segs[8];
223 	} rmode;
224 	struct {
225 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
226 		struct kvm_save_segment {
227 			u16 selector;
228 			unsigned long base;
229 			u32 limit;
230 			u32 ar;
231 		} seg[8];
232 	} segment_cache;
233 	int vpid;
234 	bool emulation_required;
235 
236 	u32 exit_reason;
237 
238 	/* Posted interrupt descriptor */
239 	struct pi_desc pi_desc;
240 
241 	/* Support for a guest hypervisor (nested VMX) */
242 	struct nested_vmx nested;
243 
244 	/* Dynamic PLE window. */
245 	int ple_window;
246 	bool ple_window_dirty;
247 
248 	bool req_immediate_exit;
249 
250 	/* Support for PML */
251 #define PML_ENTITY_NUM		512
252 	struct page *pml_pg;
253 
254 	/* apic deadline value in host tsc */
255 	u64 hv_deadline_tsc;
256 
257 	u64 current_tsc_ratio;
258 
259 	u32 host_pkru;
260 
261 	unsigned long host_debugctlmsr;
262 
263 	u64 msr_ia32_power_ctl;
264 
265 	/*
266 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
267 	 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
268 	 * in msr_ia32_feature_control_valid_bits.
269 	 */
270 	u64 msr_ia32_feature_control;
271 	u64 msr_ia32_feature_control_valid_bits;
272 	u64 ept_pointer;
273 
274 	struct pt_desc pt_desc;
275 };
276 
277 enum ept_pointers_status {
278 	EPT_POINTERS_CHECK = 0,
279 	EPT_POINTERS_MATCH = 1,
280 	EPT_POINTERS_MISMATCH = 2
281 };
282 
283 struct kvm_vmx {
284 	struct kvm kvm;
285 
286 	unsigned int tss_addr;
287 	bool ept_identity_pagetable_done;
288 	gpa_t ept_identity_map_addr;
289 
290 	enum ept_pointers_status ept_pointers_match;
291 	spinlock_t ept_pointer_lock;
292 };
293 
294 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
295 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
296 void vmx_vcpu_put(struct kvm_vcpu *vcpu);
297 int allocate_vpid(void);
298 void free_vpid(int vpid);
299 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
300 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
301 int vmx_get_cpl(struct kvm_vcpu *vcpu);
302 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
303 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
304 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
305 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
306 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
307 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
308 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
309 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
310 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
311 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
312 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
313 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
314 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
315 void update_exception_bitmap(struct kvm_vcpu *vcpu);
316 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
317 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
318 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
319 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
320 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
321 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
322 
323 #define POSTED_INTR_ON  0
324 #define POSTED_INTR_SN  1
325 
326 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
327 {
328 	return test_and_set_bit(POSTED_INTR_ON,
329 			(unsigned long *)&pi_desc->control);
330 }
331 
332 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
333 {
334 	return test_and_clear_bit(POSTED_INTR_ON,
335 			(unsigned long *)&pi_desc->control);
336 }
337 
338 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
339 {
340 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
341 }
342 
343 static inline void pi_set_sn(struct pi_desc *pi_desc)
344 {
345 	set_bit(POSTED_INTR_SN,
346 		(unsigned long *)&pi_desc->control);
347 }
348 
349 static inline void pi_set_on(struct pi_desc *pi_desc)
350 {
351 	set_bit(POSTED_INTR_ON,
352 		(unsigned long *)&pi_desc->control);
353 }
354 
355 static inline void pi_clear_on(struct pi_desc *pi_desc)
356 {
357 	clear_bit(POSTED_INTR_ON,
358 		(unsigned long *)&pi_desc->control);
359 }
360 
361 static inline int pi_test_on(struct pi_desc *pi_desc)
362 {
363 	return test_bit(POSTED_INTR_ON,
364 			(unsigned long *)&pi_desc->control);
365 }
366 
367 static inline int pi_test_sn(struct pi_desc *pi_desc)
368 {
369 	return test_bit(POSTED_INTR_SN,
370 			(unsigned long *)&pi_desc->control);
371 }
372 
373 static inline u8 vmx_get_rvi(void)
374 {
375 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
376 }
377 
378 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
379 {
380 	vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
381 }
382 
383 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
384 {
385 	vmcs_write32(VM_ENTRY_CONTROLS, val);
386 	vmx->vm_entry_controls_shadow = val;
387 }
388 
389 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
390 {
391 	if (vmx->vm_entry_controls_shadow != val)
392 		vm_entry_controls_init(vmx, val);
393 }
394 
395 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
396 {
397 	return vmx->vm_entry_controls_shadow;
398 }
399 
400 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
401 {
402 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
403 }
404 
405 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
406 {
407 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
408 }
409 
410 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
411 {
412 	vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
413 }
414 
415 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
416 {
417 	vmcs_write32(VM_EXIT_CONTROLS, val);
418 	vmx->vm_exit_controls_shadow = val;
419 }
420 
421 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
422 {
423 	if (vmx->vm_exit_controls_shadow != val)
424 		vm_exit_controls_init(vmx, val);
425 }
426 
427 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
428 {
429 	return vmx->vm_exit_controls_shadow;
430 }
431 
432 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
433 {
434 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
435 }
436 
437 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
438 {
439 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
440 }
441 
442 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
443 {
444 	vmx->segment_cache.bitmask = 0;
445 }
446 
447 static inline u32 vmx_vmentry_ctrl(void)
448 {
449 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
450 	if (pt_mode == PT_MODE_SYSTEM)
451 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
452 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
453 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
454 	return vmentry_ctrl &
455 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
456 }
457 
458 static inline u32 vmx_vmexit_ctrl(void)
459 {
460 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
461 	if (pt_mode == PT_MODE_SYSTEM)
462 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
463 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
464 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
465 	return vmexit_ctrl &
466 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
467 }
468 
469 u32 vmx_exec_control(struct vcpu_vmx *vmx);
470 
471 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
472 {
473 	return container_of(kvm, struct kvm_vmx, kvm);
474 }
475 
476 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
477 {
478 	return container_of(vcpu, struct vcpu_vmx, vcpu);
479 }
480 
481 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
482 {
483 	return &(to_vmx(vcpu)->pi_desc);
484 }
485 
486 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
487 void free_vmcs(struct vmcs *vmcs);
488 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
489 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
490 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs);
491 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
492 
493 static inline struct vmcs *alloc_vmcs(bool shadow)
494 {
495 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
496 			      GFP_KERNEL_ACCOUNT);
497 }
498 
499 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
500 
501 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
502 				bool invalidate_gpa)
503 {
504 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
505 		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
506 			return;
507 		ept_sync_context(construct_eptp(vcpu,
508 						vcpu->arch.mmu->root_hpa));
509 	} else {
510 		vpid_sync_context(vpid);
511 	}
512 }
513 
514 static inline void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
515 {
516 	__vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
517 }
518 
519 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
520 {
521 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
522 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
523 }
524 
525 void dump_vmcs(void);
526 
527 #endif /* __KVM_X86_VMX_H */
528