xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.h (revision 7e60e389)
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 
17 extern const u32 vmx_msr_index[];
18 
19 #define MSR_TYPE_R	1
20 #define MSR_TYPE_W	2
21 #define MSR_TYPE_RW	3
22 
23 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
24 
25 #ifdef CONFIG_X86_64
26 #define MAX_NR_USER_RETURN_MSRS	7
27 #else
28 #define MAX_NR_USER_RETURN_MSRS	4
29 #endif
30 
31 #define MAX_NR_LOADSTORE_MSRS	8
32 
33 struct vmx_msrs {
34 	unsigned int		nr;
35 	struct vmx_msr_entry	val[MAX_NR_LOADSTORE_MSRS];
36 };
37 
38 struct vmx_uret_msr {
39 	unsigned int slot; /* The MSR's slot in kvm_user_return_msrs. */
40 	u64 data;
41 	u64 mask;
42 };
43 
44 enum segment_cache_field {
45 	SEG_FIELD_SEL = 0,
46 	SEG_FIELD_BASE = 1,
47 	SEG_FIELD_LIMIT = 2,
48 	SEG_FIELD_AR = 3,
49 
50 	SEG_FIELD_NR = 4
51 };
52 
53 #define RTIT_ADDR_RANGE		4
54 
55 struct pt_ctx {
56 	u64 ctl;
57 	u64 status;
58 	u64 output_base;
59 	u64 output_mask;
60 	u64 cr3_match;
61 	u64 addr_a[RTIT_ADDR_RANGE];
62 	u64 addr_b[RTIT_ADDR_RANGE];
63 };
64 
65 struct pt_desc {
66 	u64 ctl_bitmask;
67 	u32 addr_range;
68 	u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
69 	struct pt_ctx host;
70 	struct pt_ctx guest;
71 };
72 
73 /*
74  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
75  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
76  */
77 struct nested_vmx {
78 	/* Has the level1 guest done vmxon? */
79 	bool vmxon;
80 	gpa_t vmxon_ptr;
81 	bool pml_full;
82 
83 	/* The guest-physical address of the current VMCS L1 keeps for L2 */
84 	gpa_t current_vmptr;
85 	/*
86 	 * Cache of the guest's VMCS, existing outside of guest memory.
87 	 * Loaded from guest memory during VMPTRLD. Flushed to guest
88 	 * memory during VMCLEAR and VMPTRLD.
89 	 */
90 	struct vmcs12 *cached_vmcs12;
91 	/*
92 	 * Cache of the guest's shadow VMCS, existing outside of guest
93 	 * memory. Loaded from guest memory during VM entry. Flushed
94 	 * to guest memory during VM exit.
95 	 */
96 	struct vmcs12 *cached_shadow_vmcs12;
97 
98 	/*
99 	 * Indicates if the shadow vmcs or enlightened vmcs must be updated
100 	 * with the data held by struct vmcs12.
101 	 */
102 	bool need_vmcs12_to_shadow_sync;
103 	bool dirty_vmcs12;
104 
105 	/*
106 	 * Indicates lazily loaded guest state has not yet been decached from
107 	 * vmcs02.
108 	 */
109 	bool need_sync_vmcs02_to_vmcs12_rare;
110 
111 	/*
112 	 * vmcs02 has been initialized, i.e. state that is constant for
113 	 * vmcs02 has been written to the backing VMCS.  Initialization
114 	 * is delayed until L1 actually attempts to run a nested VM.
115 	 */
116 	bool vmcs02_initialized;
117 
118 	bool change_vmcs01_virtual_apic_mode;
119 	bool reload_vmcs01_apic_access_page;
120 
121 	/*
122 	 * Enlightened VMCS has been enabled. It does not mean that L1 has to
123 	 * use it. However, VMX features available to L1 will be limited based
124 	 * on what the enlightened VMCS supports.
125 	 */
126 	bool enlightened_vmcs_enabled;
127 
128 	/* L2 must run next, and mustn't decide to exit to L1. */
129 	bool nested_run_pending;
130 
131 	/* Pending MTF VM-exit into L1.  */
132 	bool mtf_pending;
133 
134 	struct loaded_vmcs vmcs02;
135 
136 	/*
137 	 * Guest pages referred to in the vmcs02 with host-physical
138 	 * pointers, so we must keep them pinned while L2 runs.
139 	 */
140 	struct page *apic_access_page;
141 	struct kvm_host_map virtual_apic_map;
142 	struct kvm_host_map pi_desc_map;
143 
144 	struct kvm_host_map msr_bitmap_map;
145 
146 	struct pi_desc *pi_desc;
147 	bool pi_pending;
148 	u16 posted_intr_nv;
149 
150 	struct hrtimer preemption_timer;
151 	u64 preemption_timer_deadline;
152 	bool has_preemption_timer_deadline;
153 	bool preemption_timer_expired;
154 
155 	/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
156 	u64 vmcs01_debugctl;
157 	u64 vmcs01_guest_bndcfgs;
158 
159 	/* to migrate it to L1 if L2 writes to L1's CR8 directly */
160 	int l1_tpr_threshold;
161 
162 	u16 vpid02;
163 	u16 last_vpid;
164 
165 	struct nested_vmx_msrs msrs;
166 
167 	/* SMM related state */
168 	struct {
169 		/* in VMX operation on SMM entry? */
170 		bool vmxon;
171 		/* in guest mode on SMM entry? */
172 		bool guest_mode;
173 	} smm;
174 
175 	gpa_t hv_evmcs_vmptr;
176 	struct kvm_host_map hv_evmcs_map;
177 	struct hv_enlightened_vmcs *hv_evmcs;
178 };
179 
180 struct vcpu_vmx {
181 	struct kvm_vcpu       vcpu;
182 	u8                    fail;
183 	u8		      msr_bitmap_mode;
184 
185 	/*
186 	 * If true, host state has been stored in vmx->loaded_vmcs for
187 	 * the CPU registers that only need to be switched when transitioning
188 	 * to/from the kernel, and the registers have been loaded with guest
189 	 * values.  If false, host state is loaded in the CPU registers
190 	 * and vmx->loaded_vmcs->host_state is invalid.
191 	 */
192 	bool		      guest_state_loaded;
193 
194 	unsigned long         exit_qualification;
195 	u32                   exit_intr_info;
196 	u32                   idt_vectoring_info;
197 	ulong                 rflags;
198 
199 	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
200 	int                   nr_uret_msrs;
201 	int                   nr_active_uret_msrs;
202 	bool                  guest_uret_msrs_loaded;
203 #ifdef CONFIG_X86_64
204 	u64		      msr_host_kernel_gs_base;
205 	u64		      msr_guest_kernel_gs_base;
206 #endif
207 
208 	u64		      spec_ctrl;
209 	u32		      msr_ia32_umwait_control;
210 
211 	u32 secondary_exec_control;
212 
213 	/*
214 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
215 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
216 	 * guest (L2), it points to a different VMCS.
217 	 */
218 	struct loaded_vmcs    vmcs01;
219 	struct loaded_vmcs   *loaded_vmcs;
220 
221 	struct msr_autoload {
222 		struct vmx_msrs guest;
223 		struct vmx_msrs host;
224 	} msr_autoload;
225 
226 	struct msr_autostore {
227 		struct vmx_msrs guest;
228 	} msr_autostore;
229 
230 	struct {
231 		int vm86_active;
232 		ulong save_rflags;
233 		struct kvm_segment segs[8];
234 	} rmode;
235 	struct {
236 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
237 		struct kvm_save_segment {
238 			u16 selector;
239 			unsigned long base;
240 			u32 limit;
241 			u32 ar;
242 		} seg[8];
243 	} segment_cache;
244 	int vpid;
245 	bool emulation_required;
246 
247 	u32 exit_reason;
248 
249 	/* Posted interrupt descriptor */
250 	struct pi_desc pi_desc;
251 
252 	/* Support for a guest hypervisor (nested VMX) */
253 	struct nested_vmx nested;
254 
255 	/* Dynamic PLE window. */
256 	unsigned int ple_window;
257 	bool ple_window_dirty;
258 
259 	bool req_immediate_exit;
260 
261 	/* Support for PML */
262 #define PML_ENTITY_NUM		512
263 	struct page *pml_pg;
264 
265 	/* apic deadline value in host tsc */
266 	u64 hv_deadline_tsc;
267 
268 	u64 current_tsc_ratio;
269 
270 	unsigned long host_debugctlmsr;
271 
272 	/*
273 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
274 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
275 	 * in msr_ia32_feature_control_valid_bits.
276 	 */
277 	u64 msr_ia32_feature_control;
278 	u64 msr_ia32_feature_control_valid_bits;
279 	u64 ept_pointer;
280 
281 	struct pt_desc pt_desc;
282 
283 	/* Save desired MSR intercept (read: pass-through) state */
284 #define MAX_POSSIBLE_PASSTHROUGH_MSRS	13
285 	struct {
286 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
287 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
288 	} shadow_msr_intercept;
289 };
290 
291 enum ept_pointers_status {
292 	EPT_POINTERS_CHECK = 0,
293 	EPT_POINTERS_MATCH = 1,
294 	EPT_POINTERS_MISMATCH = 2
295 };
296 
297 struct kvm_vmx {
298 	struct kvm kvm;
299 
300 	unsigned int tss_addr;
301 	bool ept_identity_pagetable_done;
302 	gpa_t ept_identity_map_addr;
303 
304 	enum ept_pointers_status ept_pointers_match;
305 	spinlock_t ept_pointer_lock;
306 };
307 
308 bool nested_vmx_allowed(struct kvm_vcpu *vcpu);
309 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
310 			struct loaded_vmcs *buddy);
311 int allocate_vpid(void);
312 void free_vpid(int vpid);
313 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
314 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
315 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
316 			unsigned long fs_base, unsigned long gs_base);
317 int vmx_get_cpl(struct kvm_vcpu *vcpu);
318 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
319 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
320 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
321 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
322 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
323 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
324 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
325 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
326 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
327 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
328 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
329 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa,
330 		   int root_level);
331 
332 void update_exception_bitmap(struct kvm_vcpu *vcpu);
333 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
334 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
335 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
336 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
337 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
338 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
339 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
340 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
341 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
342 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
343 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
344 
345 static inline u8 vmx_get_rvi(void)
346 {
347 	return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
348 }
349 
350 #define BUILD_CONTROLS_SHADOW(lname, uname)				    \
351 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u32 val)	    \
352 {									    \
353 	if (vmx->loaded_vmcs->controls_shadow.lname != val) {		    \
354 		vmcs_write32(uname, val);				    \
355 		vmx->loaded_vmcs->controls_shadow.lname = val;		    \
356 	}								    \
357 }									    \
358 static inline u32 lname##_controls_get(struct vcpu_vmx *vmx)		    \
359 {									    \
360 	return vmx->loaded_vmcs->controls_shadow.lname;			    \
361 }									    \
362 static inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u32 val)   \
363 {									    \
364 	lname##_controls_set(vmx, lname##_controls_get(vmx) | val);	    \
365 }									    \
366 static inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u32 val) \
367 {									    \
368 	lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);	    \
369 }
370 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS)
371 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS)
372 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL)
373 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL)
374 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL)
375 
376 static inline void vmx_register_cache_reset(struct kvm_vcpu *vcpu)
377 {
378 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
379 				  | (1 << VCPU_EXREG_RFLAGS)
380 				  | (1 << VCPU_EXREG_PDPTR)
381 				  | (1 << VCPU_EXREG_SEGMENTS)
382 				  | (1 << VCPU_EXREG_CR0)
383 				  | (1 << VCPU_EXREG_CR3)
384 				  | (1 << VCPU_EXREG_CR4)
385 				  | (1 << VCPU_EXREG_EXIT_INFO_1)
386 				  | (1 << VCPU_EXREG_EXIT_INFO_2));
387 	vcpu->arch.regs_dirty = 0;
388 }
389 
390 static inline u32 vmx_vmentry_ctrl(void)
391 {
392 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
393 	if (vmx_pt_mode_is_system())
394 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
395 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
396 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
397 	return vmentry_ctrl &
398 		~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
399 }
400 
401 static inline u32 vmx_vmexit_ctrl(void)
402 {
403 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
404 	if (vmx_pt_mode_is_system())
405 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
406 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
407 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
408 	return vmexit_ctrl &
409 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
410 }
411 
412 u32 vmx_exec_control(struct vcpu_vmx *vmx);
413 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx);
414 
415 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
416 {
417 	return container_of(kvm, struct kvm_vmx, kvm);
418 }
419 
420 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
421 {
422 	return container_of(vcpu, struct vcpu_vmx, vcpu);
423 }
424 
425 static inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
426 {
427 	struct vcpu_vmx *vmx = to_vmx(vcpu);
428 
429 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_1)) {
430 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
431 		vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
432 	}
433 	return vmx->exit_qualification;
434 }
435 
436 static inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
437 {
438 	struct vcpu_vmx *vmx = to_vmx(vcpu);
439 
440 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_EXIT_INFO_2)) {
441 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
442 		vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
443 	}
444 	return vmx->exit_intr_info;
445 }
446 
447 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
448 void free_vmcs(struct vmcs *vmcs);
449 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
450 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
451 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
452 
453 static inline struct vmcs *alloc_vmcs(bool shadow)
454 {
455 	return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
456 			      GFP_KERNEL_ACCOUNT);
457 }
458 
459 static inline void decache_tsc_multiplier(struct vcpu_vmx *vmx)
460 {
461 	vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
462 	vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
463 }
464 
465 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
466 {
467 	return vmx->secondary_exec_control &
468 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
469 }
470 
471 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
472 {
473 	if (!enable_ept)
474 		return true;
475 
476 	return allow_smaller_maxphyaddr && cpuid_maxphyaddr(vcpu) < boot_cpu_data.x86_phys_bits;
477 }
478 
479 static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
480 {
481 	return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
482 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
483 	    SECONDARY_EXEC_UNRESTRICTED_GUEST));
484 }
485 
486 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
487 static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
488 {
489 	return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
490 }
491 
492 void dump_vmcs(void);
493 
494 #endif /* __KVM_X86_VMX_H */
495