xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision c8ec3743)
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 page *virtual_apic_page;
146 	struct page *pi_desc_page;
147 	struct pi_desc *pi_desc;
148 	bool pi_pending;
149 	u16 posted_intr_nv;
150 
151 	struct hrtimer preemption_timer;
152 	bool preemption_timer_expired;
153 
154 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
155 	u64 vmcs01_debugctl;
156 	u64 vmcs01_guest_bndcfgs;
157 
158 	u16 vpid02;
159 	u16 last_vpid;
160 
161 	struct nested_vmx_msrs msrs;
162 
163 	/* SMM related state */
164 	struct {
165 		/* in VMX operation on SMM entry? */
166 		bool vmxon;
167 		/* in guest mode on SMM entry? */
168 		bool guest_mode;
169 	} smm;
170 
171 	gpa_t hv_evmcs_vmptr;
172 	struct page *hv_evmcs_page;
173 	struct hv_enlightened_vmcs *hv_evmcs;
174 };
175 
176 struct vcpu_vmx {
177 	struct kvm_vcpu       vcpu;
178 	unsigned long         host_rsp;
179 	u8                    fail;
180 	u8		      msr_bitmap_mode;
181 	u32                   exit_intr_info;
182 	u32                   idt_vectoring_info;
183 	ulong                 rflags;
184 	struct shared_msr_entry *guest_msrs;
185 	int                   nmsrs;
186 	int                   save_nmsrs;
187 	bool                  guest_msrs_dirty;
188 	unsigned long	      host_idt_base;
189 #ifdef CONFIG_X86_64
190 	u64		      msr_host_kernel_gs_base;
191 	u64		      msr_guest_kernel_gs_base;
192 #endif
193 
194 	u64		      arch_capabilities;
195 	u64		      spec_ctrl;
196 
197 	u32 vm_entry_controls_shadow;
198 	u32 vm_exit_controls_shadow;
199 	u32 secondary_exec_control;
200 
201 	/*
202 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
203 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
204 	 * guest (L2), it points to a different VMCS.  loaded_cpu_state points
205 	 * to the VMCS whose state is loaded into the CPU registers that only
206 	 * need to be switched when transitioning to/from the kernel; a NULL
207 	 * value indicates that host state is loaded.
208 	 */
209 	struct loaded_vmcs    vmcs01;
210 	struct loaded_vmcs   *loaded_vmcs;
211 	struct loaded_vmcs   *loaded_cpu_state;
212 	bool                  __launched; /* temporary, used in vmx_vcpu_run */
213 	struct msr_autoload {
214 		struct vmx_msrs guest;
215 		struct vmx_msrs host;
216 	} msr_autoload;
217 
218 	struct {
219 		int vm86_active;
220 		ulong save_rflags;
221 		struct kvm_segment segs[8];
222 	} rmode;
223 	struct {
224 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
225 		struct kvm_save_segment {
226 			u16 selector;
227 			unsigned long base;
228 			u32 limit;
229 			u32 ar;
230 		} seg[8];
231 	} segment_cache;
232 	int vpid;
233 	bool emulation_required;
234 
235 	u32 exit_reason;
236 
237 	/* Posted interrupt descriptor */
238 	struct pi_desc pi_desc;
239 
240 	/* Support for a guest hypervisor (nested VMX) */
241 	struct nested_vmx nested;
242 
243 	/* Dynamic PLE window. */
244 	int ple_window;
245 	bool ple_window_dirty;
246 
247 	bool req_immediate_exit;
248 
249 	/* Support for PML */
250 #define PML_ENTITY_NUM		512
251 	struct page *pml_pg;
252 
253 	/* apic deadline value in host tsc */
254 	u64 hv_deadline_tsc;
255 
256 	u64 current_tsc_ratio;
257 
258 	u32 host_pkru;
259 
260 	unsigned long host_debugctlmsr;
261 
262 	/*
263 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
264 	 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
265 	 * in msr_ia32_feature_control_valid_bits.
266 	 */
267 	u64 msr_ia32_feature_control;
268 	u64 msr_ia32_feature_control_valid_bits;
269 	u64 ept_pointer;
270 
271 	struct pt_desc pt_desc;
272 };
273 
274 enum ept_pointers_status {
275 	EPT_POINTERS_CHECK = 0,
276 	EPT_POINTERS_MATCH = 1,
277 	EPT_POINTERS_MISMATCH = 2
278 };
279 
280 struct kvm_vmx {
281 	struct kvm kvm;
282 
283 	unsigned int tss_addr;
284 	bool ept_identity_pagetable_done;
285 	gpa_t ept_identity_map_addr;
286 
287 	enum ept_pointers_status ept_pointers_match;
288 	spinlock_t ept_pointer_lock;
289 };
290 
291 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
292 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
293 void vmx_vcpu_put(struct kvm_vcpu *vcpu);
294 int allocate_vpid(void);
295 void free_vpid(int vpid);
296 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
297 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
298 int vmx_get_cpl(struct kvm_vcpu *vcpu);
299 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
300 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
301 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
302 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
303 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
304 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
305 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
306 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
307 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
308 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
309 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
310 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
311 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
312 void update_exception_bitmap(struct kvm_vcpu *vcpu);
313 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
314 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
315 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
316 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
317 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
318 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
319 
320 #define POSTED_INTR_ON  0
321 #define POSTED_INTR_SN  1
322 
323 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
324 {
325 	return test_and_set_bit(POSTED_INTR_ON,
326 			(unsigned long *)&pi_desc->control);
327 }
328 
329 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
330 {
331 	return test_and_clear_bit(POSTED_INTR_ON,
332 			(unsigned long *)&pi_desc->control);
333 }
334 
335 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
336 {
337 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
338 }
339 
340 static inline void pi_clear_sn(struct pi_desc *pi_desc)
341 {
342 	return clear_bit(POSTED_INTR_SN,
343 			(unsigned long *)&pi_desc->control);
344 }
345 
346 static inline void pi_set_sn(struct pi_desc *pi_desc)
347 {
348 	return set_bit(POSTED_INTR_SN,
349 			(unsigned long *)&pi_desc->control);
350 }
351 
352 static inline void pi_clear_on(struct pi_desc *pi_desc)
353 {
354 	clear_bit(POSTED_INTR_ON,
355 		(unsigned long *)&pi_desc->control);
356 }
357 
358 static inline int pi_test_on(struct pi_desc *pi_desc)
359 {
360 	return test_bit(POSTED_INTR_ON,
361 			(unsigned long *)&pi_desc->control);
362 }
363 
364 static inline int pi_test_sn(struct pi_desc *pi_desc)
365 {
366 	return test_bit(POSTED_INTR_SN,
367 			(unsigned long *)&pi_desc->control);
368 }
369 
370 static inline u8 vmx_get_rvi(void)
371 {
372 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
373 }
374 
375 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
376 {
377 	vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
378 }
379 
380 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
381 {
382 	vmcs_write32(VM_ENTRY_CONTROLS, val);
383 	vmx->vm_entry_controls_shadow = val;
384 }
385 
386 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
387 {
388 	if (vmx->vm_entry_controls_shadow != val)
389 		vm_entry_controls_init(vmx, val);
390 }
391 
392 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
393 {
394 	return vmx->vm_entry_controls_shadow;
395 }
396 
397 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
398 {
399 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
400 }
401 
402 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
403 {
404 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
405 }
406 
407 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
408 {
409 	vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
410 }
411 
412 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
413 {
414 	vmcs_write32(VM_EXIT_CONTROLS, val);
415 	vmx->vm_exit_controls_shadow = val;
416 }
417 
418 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
419 {
420 	if (vmx->vm_exit_controls_shadow != val)
421 		vm_exit_controls_init(vmx, val);
422 }
423 
424 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
425 {
426 	return vmx->vm_exit_controls_shadow;
427 }
428 
429 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
430 {
431 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
432 }
433 
434 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
435 {
436 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
437 }
438 
439 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
440 {
441 	vmx->segment_cache.bitmask = 0;
442 }
443 
444 static inline u32 vmx_vmentry_ctrl(void)
445 {
446 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
447 	if (pt_mode == PT_MODE_SYSTEM)
448 		vmentry_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP | VM_EXIT_CLEAR_IA32_RTIT_CTL);
449 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
450 	return vmentry_ctrl &
451 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
452 }
453 
454 static inline u32 vmx_vmexit_ctrl(void)
455 {
456 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
457 	if (pt_mode == PT_MODE_SYSTEM)
458 		vmexit_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP | VM_ENTRY_LOAD_IA32_RTIT_CTL);
459 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
460 	return vmcs_config.vmexit_ctrl &
461 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
462 }
463 
464 u32 vmx_exec_control(struct vcpu_vmx *vmx);
465 
466 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
467 {
468 	return container_of(kvm, struct kvm_vmx, kvm);
469 }
470 
471 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
472 {
473 	return container_of(vcpu, struct vcpu_vmx, vcpu);
474 }
475 
476 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
477 {
478 	return &(to_vmx(vcpu)->pi_desc);
479 }
480 
481 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu);
482 void free_vmcs(struct vmcs *vmcs);
483 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
484 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
485 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs);
486 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
487 
488 static inline struct vmcs *alloc_vmcs(bool shadow)
489 {
490 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id());
491 }
492 
493 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
494 
495 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
496 				bool invalidate_gpa)
497 {
498 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
499 		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
500 			return;
501 		ept_sync_context(construct_eptp(vcpu,
502 						vcpu->arch.mmu->root_hpa));
503 	} else {
504 		vpid_sync_context(vpid);
505 	}
506 }
507 
508 static inline void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
509 {
510 	__vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
511 }
512 
513 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
514 {
515 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
516 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
517 }
518 
519 #endif /* __KVM_X86_VMX_H */
520