xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision 367e5927)
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 	u8                    fail;
179 	u8		      msr_bitmap_mode;
180 	u32                   exit_intr_info;
181 	u32                   idt_vectoring_info;
182 	ulong                 rflags;
183 	struct shared_msr_entry *guest_msrs;
184 	int                   nmsrs;
185 	int                   save_nmsrs;
186 	bool                  guest_msrs_dirty;
187 	unsigned long	      host_idt_base;
188 #ifdef CONFIG_X86_64
189 	u64		      msr_host_kernel_gs_base;
190 	u64		      msr_guest_kernel_gs_base;
191 #endif
192 
193 	u64		      spec_ctrl;
194 
195 	u32 vm_entry_controls_shadow;
196 	u32 vm_exit_controls_shadow;
197 	u32 secondary_exec_control;
198 
199 	/*
200 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
201 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
202 	 * guest (L2), it points to a different VMCS.  loaded_cpu_state points
203 	 * to the VMCS whose state is loaded into the CPU registers that only
204 	 * need to be switched when transitioning to/from the kernel; a NULL
205 	 * value indicates that host state is loaded.
206 	 */
207 	struct loaded_vmcs    vmcs01;
208 	struct loaded_vmcs   *loaded_vmcs;
209 	struct loaded_vmcs   *loaded_cpu_state;
210 
211 	struct msr_autoload {
212 		struct vmx_msrs guest;
213 		struct vmx_msrs host;
214 	} msr_autoload;
215 
216 	struct {
217 		int vm86_active;
218 		ulong save_rflags;
219 		struct kvm_segment segs[8];
220 	} rmode;
221 	struct {
222 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
223 		struct kvm_save_segment {
224 			u16 selector;
225 			unsigned long base;
226 			u32 limit;
227 			u32 ar;
228 		} seg[8];
229 	} segment_cache;
230 	int vpid;
231 	bool emulation_required;
232 
233 	u32 exit_reason;
234 
235 	/* Posted interrupt descriptor */
236 	struct pi_desc pi_desc;
237 
238 	/* Support for a guest hypervisor (nested VMX) */
239 	struct nested_vmx nested;
240 
241 	/* Dynamic PLE window. */
242 	int ple_window;
243 	bool ple_window_dirty;
244 
245 	bool req_immediate_exit;
246 
247 	/* Support for PML */
248 #define PML_ENTITY_NUM		512
249 	struct page *pml_pg;
250 
251 	/* apic deadline value in host tsc */
252 	u64 hv_deadline_tsc;
253 
254 	u64 current_tsc_ratio;
255 
256 	u32 host_pkru;
257 
258 	unsigned long host_debugctlmsr;
259 
260 	/*
261 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
262 	 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
263 	 * in msr_ia32_feature_control_valid_bits.
264 	 */
265 	u64 msr_ia32_feature_control;
266 	u64 msr_ia32_feature_control_valid_bits;
267 	u64 ept_pointer;
268 
269 	struct pt_desc pt_desc;
270 };
271 
272 enum ept_pointers_status {
273 	EPT_POINTERS_CHECK = 0,
274 	EPT_POINTERS_MATCH = 1,
275 	EPT_POINTERS_MISMATCH = 2
276 };
277 
278 struct kvm_vmx {
279 	struct kvm kvm;
280 
281 	unsigned int tss_addr;
282 	bool ept_identity_pagetable_done;
283 	gpa_t ept_identity_map_addr;
284 
285 	enum ept_pointers_status ept_pointers_match;
286 	spinlock_t ept_pointer_lock;
287 };
288 
289 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
290 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
291 void vmx_vcpu_put(struct kvm_vcpu *vcpu);
292 int allocate_vpid(void);
293 void free_vpid(int vpid);
294 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
295 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
296 int vmx_get_cpl(struct kvm_vcpu *vcpu);
297 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
298 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
299 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
300 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
301 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
302 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
303 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
304 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
305 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
306 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
307 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
308 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
309 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
310 void update_exception_bitmap(struct kvm_vcpu *vcpu);
311 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
312 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
313 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
314 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
315 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
316 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
317 
318 #define POSTED_INTR_ON  0
319 #define POSTED_INTR_SN  1
320 
321 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
322 {
323 	return test_and_set_bit(POSTED_INTR_ON,
324 			(unsigned long *)&pi_desc->control);
325 }
326 
327 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
328 {
329 	return test_and_clear_bit(POSTED_INTR_ON,
330 			(unsigned long *)&pi_desc->control);
331 }
332 
333 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
334 {
335 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
336 }
337 
338 static inline void pi_set_sn(struct pi_desc *pi_desc)
339 {
340 	set_bit(POSTED_INTR_SN,
341 		(unsigned long *)&pi_desc->control);
342 }
343 
344 static inline void pi_set_on(struct pi_desc *pi_desc)
345 {
346 	set_bit(POSTED_INTR_ON,
347 		(unsigned long *)&pi_desc->control);
348 }
349 
350 static inline void pi_clear_on(struct pi_desc *pi_desc)
351 {
352 	clear_bit(POSTED_INTR_ON,
353 		(unsigned long *)&pi_desc->control);
354 }
355 
356 static inline int pi_test_on(struct pi_desc *pi_desc)
357 {
358 	return test_bit(POSTED_INTR_ON,
359 			(unsigned long *)&pi_desc->control);
360 }
361 
362 static inline int pi_test_sn(struct pi_desc *pi_desc)
363 {
364 	return test_bit(POSTED_INTR_SN,
365 			(unsigned long *)&pi_desc->control);
366 }
367 
368 static inline u8 vmx_get_rvi(void)
369 {
370 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
371 }
372 
373 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
374 {
375 	vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
376 }
377 
378 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
379 {
380 	vmcs_write32(VM_ENTRY_CONTROLS, val);
381 	vmx->vm_entry_controls_shadow = val;
382 }
383 
384 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
385 {
386 	if (vmx->vm_entry_controls_shadow != val)
387 		vm_entry_controls_init(vmx, val);
388 }
389 
390 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
391 {
392 	return vmx->vm_entry_controls_shadow;
393 }
394 
395 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
396 {
397 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
398 }
399 
400 static inline void vm_entry_controls_clearbit(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_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
406 {
407 	vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
408 }
409 
410 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
411 {
412 	vmcs_write32(VM_EXIT_CONTROLS, val);
413 	vmx->vm_exit_controls_shadow = val;
414 }
415 
416 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
417 {
418 	if (vmx->vm_exit_controls_shadow != val)
419 		vm_exit_controls_init(vmx, val);
420 }
421 
422 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
423 {
424 	return vmx->vm_exit_controls_shadow;
425 }
426 
427 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
428 {
429 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
430 }
431 
432 static inline void vm_exit_controls_clearbit(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 vmx_segment_cache_clear(struct vcpu_vmx *vmx)
438 {
439 	vmx->segment_cache.bitmask = 0;
440 }
441 
442 static inline u32 vmx_vmentry_ctrl(void)
443 {
444 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
445 	if (pt_mode == PT_MODE_SYSTEM)
446 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
447 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
448 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
449 	return vmentry_ctrl &
450 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
451 }
452 
453 static inline u32 vmx_vmexit_ctrl(void)
454 {
455 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
456 	if (pt_mode == PT_MODE_SYSTEM)
457 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
458 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
459 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
460 	return 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, gfp_t flags);
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 			      GFP_KERNEL_ACCOUNT);
492 }
493 
494 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
495 
496 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
497 				bool invalidate_gpa)
498 {
499 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
500 		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
501 			return;
502 		ept_sync_context(construct_eptp(vcpu,
503 						vcpu->arch.mmu->root_hpa));
504 	} else {
505 		vpid_sync_context(vpid);
506 	}
507 }
508 
509 static inline void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
510 {
511 	__vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
512 }
513 
514 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
515 {
516 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
517 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
518 }
519 
520 #endif /* __KVM_X86_VMX_H */
521