xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision 5b628549)
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		      arch_capabilities;
194 	u64		      spec_ctrl;
195 
196 	u32 vm_entry_controls_shadow;
197 	u32 vm_exit_controls_shadow;
198 	u32 secondary_exec_control;
199 
200 	/*
201 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
202 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
203 	 * guest (L2), it points to a different VMCS.  loaded_cpu_state points
204 	 * to the VMCS whose state is loaded into the CPU registers that only
205 	 * need to be switched when transitioning to/from the kernel; a NULL
206 	 * value indicates that host state is loaded.
207 	 */
208 	struct loaded_vmcs    vmcs01;
209 	struct loaded_vmcs   *loaded_vmcs;
210 	struct loaded_vmcs   *loaded_cpu_state;
211 
212 	struct msr_autoload {
213 		struct vmx_msrs guest;
214 		struct vmx_msrs host;
215 	} msr_autoload;
216 
217 	struct {
218 		int vm86_active;
219 		ulong save_rflags;
220 		struct kvm_segment segs[8];
221 	} rmode;
222 	struct {
223 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
224 		struct kvm_save_segment {
225 			u16 selector;
226 			unsigned long base;
227 			u32 limit;
228 			u32 ar;
229 		} seg[8];
230 	} segment_cache;
231 	int vpid;
232 	bool emulation_required;
233 
234 	u32 exit_reason;
235 
236 	/* Posted interrupt descriptor */
237 	struct pi_desc pi_desc;
238 
239 	/* Support for a guest hypervisor (nested VMX) */
240 	struct nested_vmx nested;
241 
242 	/* Dynamic PLE window. */
243 	int ple_window;
244 	bool ple_window_dirty;
245 
246 	bool req_immediate_exit;
247 
248 	/* Support for PML */
249 #define PML_ENTITY_NUM		512
250 	struct page *pml_pg;
251 
252 	/* apic deadline value in host tsc */
253 	u64 hv_deadline_tsc;
254 
255 	u64 current_tsc_ratio;
256 
257 	u32 host_pkru;
258 
259 	unsigned long host_debugctlmsr;
260 
261 	/*
262 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
263 	 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
264 	 * in msr_ia32_feature_control_valid_bits.
265 	 */
266 	u64 msr_ia32_feature_control;
267 	u64 msr_ia32_feature_control_valid_bits;
268 	u64 ept_pointer;
269 
270 	struct pt_desc pt_desc;
271 };
272 
273 enum ept_pointers_status {
274 	EPT_POINTERS_CHECK = 0,
275 	EPT_POINTERS_MATCH = 1,
276 	EPT_POINTERS_MISMATCH = 2
277 };
278 
279 struct kvm_vmx {
280 	struct kvm kvm;
281 
282 	unsigned int tss_addr;
283 	bool ept_identity_pagetable_done;
284 	gpa_t ept_identity_map_addr;
285 
286 	enum ept_pointers_status ept_pointers_match;
287 	spinlock_t ept_pointer_lock;
288 };
289 
290 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
291 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
292 void vmx_vcpu_put(struct kvm_vcpu *vcpu);
293 int allocate_vpid(void);
294 void free_vpid(int vpid);
295 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
296 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
297 int vmx_get_cpl(struct kvm_vcpu *vcpu);
298 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
299 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
300 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
301 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
302 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
303 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
304 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
305 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
306 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
307 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
308 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
309 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
310 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
311 void update_exception_bitmap(struct kvm_vcpu *vcpu);
312 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
313 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
314 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
315 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
316 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr);
317 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx);
318 
319 #define POSTED_INTR_ON  0
320 #define POSTED_INTR_SN  1
321 
322 static inline bool pi_test_and_set_on(struct pi_desc *pi_desc)
323 {
324 	return test_and_set_bit(POSTED_INTR_ON,
325 			(unsigned long *)&pi_desc->control);
326 }
327 
328 static inline bool pi_test_and_clear_on(struct pi_desc *pi_desc)
329 {
330 	return test_and_clear_bit(POSTED_INTR_ON,
331 			(unsigned long *)&pi_desc->control);
332 }
333 
334 static inline int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
335 {
336 	return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
337 }
338 
339 static inline void pi_set_sn(struct pi_desc *pi_desc)
340 {
341 	set_bit(POSTED_INTR_SN,
342 		(unsigned long *)&pi_desc->control);
343 }
344 
345 static inline void pi_set_on(struct pi_desc *pi_desc)
346 {
347 	set_bit(POSTED_INTR_ON,
348 		(unsigned long *)&pi_desc->control);
349 }
350 
351 static inline void pi_clear_on(struct pi_desc *pi_desc)
352 {
353 	clear_bit(POSTED_INTR_ON,
354 		(unsigned long *)&pi_desc->control);
355 }
356 
357 static inline int pi_test_on(struct pi_desc *pi_desc)
358 {
359 	return test_bit(POSTED_INTR_ON,
360 			(unsigned long *)&pi_desc->control);
361 }
362 
363 static inline int pi_test_sn(struct pi_desc *pi_desc)
364 {
365 	return test_bit(POSTED_INTR_SN,
366 			(unsigned long *)&pi_desc->control);
367 }
368 
369 static inline u8 vmx_get_rvi(void)
370 {
371 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
372 }
373 
374 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
375 {
376 	vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
377 }
378 
379 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
380 {
381 	vmcs_write32(VM_ENTRY_CONTROLS, val);
382 	vmx->vm_entry_controls_shadow = val;
383 }
384 
385 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
386 {
387 	if (vmx->vm_entry_controls_shadow != val)
388 		vm_entry_controls_init(vmx, val);
389 }
390 
391 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
392 {
393 	return vmx->vm_entry_controls_shadow;
394 }
395 
396 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
397 {
398 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
399 }
400 
401 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
402 {
403 	vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
404 }
405 
406 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
407 {
408 	vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
409 }
410 
411 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
412 {
413 	vmcs_write32(VM_EXIT_CONTROLS, val);
414 	vmx->vm_exit_controls_shadow = val;
415 }
416 
417 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
418 {
419 	if (vmx->vm_exit_controls_shadow != val)
420 		vm_exit_controls_init(vmx, val);
421 }
422 
423 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
424 {
425 	return vmx->vm_exit_controls_shadow;
426 }
427 
428 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
429 {
430 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
431 }
432 
433 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
434 {
435 	vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
436 }
437 
438 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
439 {
440 	vmx->segment_cache.bitmask = 0;
441 }
442 
443 static inline u32 vmx_vmentry_ctrl(void)
444 {
445 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
446 	if (pt_mode == PT_MODE_SYSTEM)
447 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
448 				  VM_ENTRY_LOAD_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_EXIT_PT_CONCEAL_PIP |
459 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
460 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
461 	return vmexit_ctrl &
462 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
463 }
464 
465 u32 vmx_exec_control(struct vcpu_vmx *vmx);
466 
467 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
468 {
469 	return container_of(kvm, struct kvm_vmx, kvm);
470 }
471 
472 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
473 {
474 	return container_of(vcpu, struct vcpu_vmx, vcpu);
475 }
476 
477 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
478 {
479 	return &(to_vmx(vcpu)->pi_desc);
480 }
481 
482 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
483 void free_vmcs(struct vmcs *vmcs);
484 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
485 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
486 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs);
487 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
488 
489 static inline struct vmcs *alloc_vmcs(bool shadow)
490 {
491 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
492 			      GFP_KERNEL_ACCOUNT);
493 }
494 
495 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
496 
497 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
498 				bool invalidate_gpa)
499 {
500 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
501 		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
502 			return;
503 		ept_sync_context(construct_eptp(vcpu,
504 						vcpu->arch.mmu->root_hpa));
505 	} else {
506 		vpid_sync_context(vpid);
507 	}
508 }
509 
510 static inline void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
511 {
512 	__vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
513 }
514 
515 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
516 {
517 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
518 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
519 }
520 
521 #endif /* __KVM_X86_VMX_H */
522