xref: /kernel/linux/linux-5.10/arch/x86/kvm/vmx/vmx.h (revision 8c2ecf20)
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#include "run_flags.h"
17
18extern const u32 vmx_msr_index[];
19
20#define MSR_TYPE_R	1
21#define MSR_TYPE_W	2
22#define MSR_TYPE_RW	3
23
24#define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
25
26#ifdef CONFIG_X86_64
27#define MAX_NR_USER_RETURN_MSRS	7
28#else
29#define MAX_NR_USER_RETURN_MSRS	4
30#endif
31
32#define MAX_NR_LOADSTORE_MSRS	8
33
34struct vmx_msrs {
35	unsigned int		nr;
36	struct vmx_msr_entry	val[MAX_NR_LOADSTORE_MSRS];
37};
38
39struct vmx_uret_msr {
40	unsigned int slot; /* The MSR's slot in kvm_user_return_msrs. */
41	u64 data;
42	u64 mask;
43};
44
45enum segment_cache_field {
46	SEG_FIELD_SEL = 0,
47	SEG_FIELD_BASE = 1,
48	SEG_FIELD_LIMIT = 2,
49	SEG_FIELD_AR = 3,
50
51	SEG_FIELD_NR = 4
52};
53
54#define RTIT_ADDR_RANGE		4
55
56struct pt_ctx {
57	u64 ctl;
58	u64 status;
59	u64 output_base;
60	u64 output_mask;
61	u64 cr3_match;
62	u64 addr_a[RTIT_ADDR_RANGE];
63	u64 addr_b[RTIT_ADDR_RANGE];
64};
65
66struct pt_desc {
67	u64 ctl_bitmask;
68	u32 addr_range;
69	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
70	struct pt_ctx host;
71	struct pt_ctx guest;
72};
73
74union vmx_exit_reason {
75	struct {
76		u32	basic			: 16;
77		u32	reserved16		: 1;
78		u32	reserved17		: 1;
79		u32	reserved18		: 1;
80		u32	reserved19		: 1;
81		u32	reserved20		: 1;
82		u32	reserved21		: 1;
83		u32	reserved22		: 1;
84		u32	reserved23		: 1;
85		u32	reserved24		: 1;
86		u32	reserved25		: 1;
87		u32	reserved26		: 1;
88		u32	enclave_mode		: 1;
89		u32	smi_pending_mtf		: 1;
90		u32	smi_from_vmx_root	: 1;
91		u32	reserved30		: 1;
92		u32	failed_vmentry		: 1;
93	};
94	u32 full;
95};
96
97/*
98 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
99 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
100 */
101struct nested_vmx {
102	/* Has the level1 guest done vmxon? */
103	bool vmxon;
104	gpa_t vmxon_ptr;
105	bool pml_full;
106
107	/* The guest-physical address of the current VMCS L1 keeps for L2 */
108	gpa_t current_vmptr;
109	/*
110	 * Cache of the guest's VMCS, existing outside of guest memory.
111	 * Loaded from guest memory during VMPTRLD. Flushed to guest
112	 * memory during VMCLEAR and VMPTRLD.
113	 */
114	struct vmcs12 *cached_vmcs12;
115	/*
116	 * Cache of the guest's shadow VMCS, existing outside of guest
117	 * memory. Loaded from guest memory during VM entry. Flushed
118	 * to guest memory during VM exit.
119	 */
120	struct vmcs12 *cached_shadow_vmcs12;
121
122	/*
123	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
124	 * with the data held by struct vmcs12.
125	 */
126	bool need_vmcs12_to_shadow_sync;
127	bool dirty_vmcs12;
128
129	/*
130	 * Indicates lazily loaded guest state has not yet been decached from
131	 * vmcs02.
132	 */
133	bool need_sync_vmcs02_to_vmcs12_rare;
134
135	/*
136	 * vmcs02 has been initialized, i.e. state that is constant for
137	 * vmcs02 has been written to the backing VMCS.  Initialization
138	 * is delayed until L1 actually attempts to run a nested VM.
139	 */
140	bool vmcs02_initialized;
141
142	bool change_vmcs01_virtual_apic_mode;
143	bool reload_vmcs01_apic_access_page;
144
145	/*
146	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
147	 * use it. However, VMX features available to L1 will be limited based
148	 * on what the enlightened VMCS supports.
149	 */
150	bool enlightened_vmcs_enabled;
151
152	/* L2 must run next, and mustn't decide to exit to L1. */
153	bool nested_run_pending;
154
155	/* Pending MTF VM-exit into L1.  */
156	bool mtf_pending;
157
158	struct loaded_vmcs vmcs02;
159
160	/*
161	 * Guest pages referred to in the vmcs02 with host-physical
162	 * pointers, so we must keep them pinned while L2 runs.
163	 */
164	struct page *apic_access_page;
165	struct kvm_host_map virtual_apic_map;
166	struct kvm_host_map pi_desc_map;
167
168	struct kvm_host_map msr_bitmap_map;
169
170	struct pi_desc *pi_desc;
171	bool pi_pending;
172	u16 posted_intr_nv;
173
174	struct hrtimer preemption_timer;
175	u64 preemption_timer_deadline;
176	bool has_preemption_timer_deadline;
177	bool preemption_timer_expired;
178
179	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
180	u64 vmcs01_debugctl;
181	u64 vmcs01_guest_bndcfgs;
182
183	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
184	int l1_tpr_threshold;
185
186	u16 vpid02;
187	u16 last_vpid;
188
189	struct nested_vmx_msrs msrs;
190
191	/* SMM related state */
192	struct {
193		/* in VMX operation on SMM entry? */
194		bool vmxon;
195		/* in guest mode on SMM entry? */
196		bool guest_mode;
197	} smm;
198
199	gpa_t hv_evmcs_vmptr;
200	struct kvm_host_map hv_evmcs_map;
201	struct hv_enlightened_vmcs *hv_evmcs;
202};
203
204struct vcpu_vmx {
205	struct kvm_vcpu       vcpu;
206	u8                    fail;
207	u8		      msr_bitmap_mode;
208
209	/*
210	 * If true, host state has been stored in vmx->loaded_vmcs for
211	 * the CPU registers that only need to be switched when transitioning
212	 * to/from the kernel, and the registers have been loaded with guest
213	 * values.  If false, host state is loaded in the CPU registers
214	 * and vmx->loaded_vmcs->host_state is invalid.
215	 */
216	bool		      guest_state_loaded;
217
218	unsigned long         exit_qualification;
219	u32                   exit_intr_info;
220	u32                   idt_vectoring_info;
221	ulong                 rflags;
222
223	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
224	int                   nr_uret_msrs;
225	int                   nr_active_uret_msrs;
226	bool                  guest_uret_msrs_loaded;
227#ifdef CONFIG_X86_64
228	u64		      msr_host_kernel_gs_base;
229	u64		      msr_guest_kernel_gs_base;
230#endif
231
232	u64		      spec_ctrl;
233	u32		      msr_ia32_umwait_control;
234
235	u32 secondary_exec_control;
236
237	/*
238	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
239	 * non-nested (L1) guest, it always points to vmcs01. For a nested
240	 * guest (L2), it points to a different VMCS.
241	 */
242	struct loaded_vmcs    vmcs01;
243	struct loaded_vmcs   *loaded_vmcs;
244
245	struct msr_autoload {
246		struct vmx_msrs guest;
247		struct vmx_msrs host;
248	} msr_autoload;
249
250	struct msr_autostore {
251		struct vmx_msrs guest;
252	} msr_autostore;
253
254	struct {
255		int vm86_active;
256		ulong save_rflags;
257		struct kvm_segment segs[8];
258	} rmode;
259	struct {
260		u32 bitmask; /* 4 bits per segment (1 bit per field) */
261		struct kvm_save_segment {
262			u16 selector;
263			unsigned long base;
264			u32 limit;
265			u32 ar;
266		} seg[8];
267	} segment_cache;
268	int vpid;
269	bool emulation_required;
270
271	union vmx_exit_reason exit_reason;
272
273	/* Posted interrupt descriptor */
274	struct pi_desc pi_desc;
275
276	/* Support for a guest hypervisor (nested VMX) */
277	struct nested_vmx nested;
278
279	/* Dynamic PLE window. */
280	unsigned int ple_window;
281	bool ple_window_dirty;
282
283	bool req_immediate_exit;
284
285	/* Support for PML */
286#define PML_ENTITY_NUM		512
287	struct page *pml_pg;
288
289	/* apic deadline value in host tsc */
290	u64 hv_deadline_tsc;
291
292	u64 current_tsc_ratio;
293
294	unsigned long host_debugctlmsr;
295
296	/*
297	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
298	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
299	 * in msr_ia32_feature_control_valid_bits.
300	 */
301	u64 msr_ia32_feature_control;
302	u64 msr_ia32_feature_control_valid_bits;
303	u64 ept_pointer;
304	u64 msr_ia32_mcu_opt_ctrl;
305	bool disable_fb_clear;
306
307	struct pt_desc pt_desc;
308
309	/* Save desired MSR intercept (read: pass-through) state */
310#define MAX_POSSIBLE_PASSTHROUGH_MSRS	13
311	struct {
312		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
313		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
314	} shadow_msr_intercept;
315};
316
317enum ept_pointers_status {
318	EPT_POINTERS_CHECK = 0,
319	EPT_POINTERS_MATCH = 1,
320	EPT_POINTERS_MISMATCH = 2
321};
322
323struct kvm_vmx {
324	struct kvm kvm;
325
326	unsigned int tss_addr;
327	bool ept_identity_pagetable_done;
328	gpa_t ept_identity_map_addr;
329
330	enum ept_pointers_status ept_pointers_match;
331	spinlock_t ept_pointer_lock;
332};
333
334bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
335void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
336			struct loaded_vmcs *buddy);
337int allocate_vpid(void);
338void free_vpid(int vpid);
339void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
340void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
341void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
342			unsigned long fs_base, unsigned long gs_base);
343int vmx_get_cpl(struct kvm_vcpu *vcpu);
344unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
345void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
346u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
347void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
348int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
349void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
350void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
351void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
352void ept_save_pdptrs(struct kvm_vcpu *vcpu);
353void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
354void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
355u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa,
356		   int root_level);
357
358bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu);
359void update_exception_bitmap(struct kvm_vcpu *vcpu);
360void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
361bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
362bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
363bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
364void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
365void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
366struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
367void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
368void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
369void vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, unsigned int flags);
370unsigned int __vmx_vcpu_run_flags(struct vcpu_vmx *vmx);
371bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs,
372		    unsigned int flags);
373int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
374void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
375
376static inline u8 vmx_get_rvi(void)
377{
378	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
379}
380
381#define BUILD_CONTROLS_SHADOW(lname, uname)				    \
382static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
383{									    \
384	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
385		vmcs_write32(uname, val);				    \
386		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
387	}								    \
388}									    \
389static inline u32 __##lname##_controls_get(struct loaded_vmcs *vmcs)	    \
390{									    \
391	return vmcs->controls_shadow.lname;				    \
392}									    \
393static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
394{									    \
395	return __##lname##_controls_get(vmx->loaded_vmcs);		    \
396}									    \
397static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
398{									    \
399	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
400}									    \
401static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
402{									    \
403	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
404}
405BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
406BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
407BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
408BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
409BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
410
411static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
412{
413	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
414				  | (1 << VCPU_EXREG_RFLAGS)
415				  | (1 << VCPU_EXREG_PDPTR)
416				  | (1 << VCPU_EXREG_SEGMENTS)
417				  | (1 << VCPU_EXREG_CR0)
418				  | (1 << VCPU_EXREG_CR3)
419				  | (1 << VCPU_EXREG_CR4)
420				  | (1 << VCPU_EXREG_EXIT_INFO_1)
421				  | (1 << VCPU_EXREG_EXIT_INFO_2));
422	vcpu->arch.regs_dirty = 0;
423}
424
425static inline u32 vmx_vmentry_ctrl(void)
426{
427	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
428	if (vmx_pt_mode_is_system())
429		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
430				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
431	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
432	return vmentry_ctrl &
433		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
434}
435
436static inline u32 vmx_vmexit_ctrl(void)
437{
438	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
439	if (vmx_pt_mode_is_system())
440		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
441				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
442	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
443	return vmexit_ctrl &
444		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
445}
446
447u32 vmx_exec_control(struct vcpu_vmx *vmx);
448u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
449
450static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
451{
452	return container_of(kvm, struct kvm_vmx, kvm);
453}
454
455static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
456{
457	return container_of(vcpu, struct vcpu_vmx, vcpu);
458}
459
460static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
461{
462	struct vcpu_vmx *vmx = to_vmx(vcpu);
463
464	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
465		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
466		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
467	}
468	return vmx->exit_qualification;
469}
470
471static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
472{
473	struct vcpu_vmx *vmx = to_vmx(vcpu);
474
475	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
476		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
477		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
478	}
479	return vmx->exit_intr_info;
480}
481
482struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
483void free_vmcs(struct vmcs *vmcs);
484int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
485void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
486void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
487
488static inline struct vmcs *alloc_vmcs(bool shadow)
489{
490	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
491			      GFP_KERNEL_ACCOUNT);
492}
493
494static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
495{
496	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
497	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
498}
499
500static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
501{
502	return secondary_exec_controls_get(vmx) &
503		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
504}
505
506static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
507{
508	if (!enable_ept)
509		return true;
510
511	return allow_smaller_maxphyaddr && cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
512}
513
514static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
515{
516	return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
517	    (secondary_exec_controls_get(to_vmx(vcpu)) &
518	    SECONDARY_EXEC_UNRESTRICTED_GUEST));
519}
520
521bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
522static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
523{
524	return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
525}
526
527void dump_vmcs(void);
528
529#endif /* __KVM_X86_VMX_H */
530