1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef ARCH_X86_KVM_X86_H
3 #define ARCH_X86_KVM_X86_H
5 #include <linux/kvm_host.h>
7 #include <asm/pvclock.h>
8 #include "kvm_cache_regs.h"
9 #include "kvm_emulate.h"
11 void kvm_spurious_fault(void);
13 static __always_inline void kvm_guest_enter_irqoff(void)
16 * VMENTER enables interrupts (host state), but the kernel state is
17 * interrupts disabled when this is invoked. Also tell RCU about
18 * it. This is the same logic as for exit_to_user_mode().
20 * This ensures that e.g. latency analysis on the host observes
21 * guest mode as interrupt enabled.
23 * guest_enter_irqoff() informs context tracking about the
24 * transition to guest mode and if enabled adjusts RCU state
27 instrumentation_begin();
28 trace_hardirqs_on_prepare();
29 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
30 instrumentation_end();
33 lockdep_hardirqs_on(CALLER_ADDR0);
36 static __always_inline void kvm_guest_exit_irqoff(void)
39 * VMEXIT disables interrupts (host state), but tracing and lockdep
40 * have them in state 'on' as recorded before entering guest mode.
41 * Same as enter_from_user_mode().
43 * context_tracking_guest_exit() restores host context and reinstates
44 * RCU if enabled and required.
46 * This needs to be done immediately after VM-Exit, before any code
47 * that might contain tracepoints or call out to the greater world,
48 * e.g. before x86_spec_ctrl_restore_host().
50 lockdep_hardirqs_off(CALLER_ADDR0);
51 context_tracking_guest_exit();
53 instrumentation_begin();
54 trace_hardirqs_off_finish();
55 instrumentation_end();
58 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check) \
60 bool failed = (consistency_check); \
62 trace_kvm_nested_vmenter_failed(#consistency_check, 0); \
66 #define KVM_DEFAULT_PLE_GAP 128
67 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
68 #define KVM_DEFAULT_PLE_WINDOW_GROW 2
69 #define KVM_DEFAULT_PLE_WINDOW_SHRINK 0
70 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX UINT_MAX
71 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX USHRT_MAX
72 #define KVM_SVM_DEFAULT_PLE_WINDOW 3000
74 static inline unsigned int __grow_ple_window(unsigned int val,
75 unsigned int base, unsigned int modifier, unsigned int max)
87 return min(ret, (u64)max);
90 static inline unsigned int __shrink_ple_window(unsigned int val,
91 unsigned int base, unsigned int modifier, unsigned int min)
101 return max(val, min);
104 #define MSR_IA32_CR_PAT_DEFAULT 0x0007040600070406ULL
106 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu);
107 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
109 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
111 vcpu->arch.exception.pending = false;
112 vcpu->arch.exception.injected = false;
115 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
118 vcpu->arch.interrupt.injected = true;
119 vcpu->arch.interrupt.soft = soft;
120 vcpu->arch.interrupt.nr = vector;
123 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
125 vcpu->arch.interrupt.injected = false;
128 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
130 return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
131 vcpu->arch.nmi_injected;
134 static inline bool kvm_exception_is_soft(unsigned int nr)
136 return (nr == BP_VECTOR) || (nr == OF_VECTOR);
139 static inline bool is_protmode(struct kvm_vcpu *vcpu)
141 return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
144 static inline int is_long_mode(struct kvm_vcpu *vcpu)
147 return vcpu->arch.efer & EFER_LMA;
153 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
157 WARN_ON_ONCE(vcpu->arch.guest_state_protected);
159 if (!is_long_mode(vcpu))
161 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
165 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu)
168 * If running with protected guest state, the CS register is not
169 * accessible. The hypercall register values will have had to been
170 * provided in 64-bit mode, so assume the guest is in 64-bit.
172 return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu);
175 static inline bool x86_exception_has_error_code(unsigned int vector)
177 static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
178 BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
179 BIT(PF_VECTOR) | BIT(AC_VECTOR);
181 return (1U << vector) & exception_has_error_code;
184 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
186 return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
189 static inline int is_pae(struct kvm_vcpu *vcpu)
191 return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
194 static inline int is_pse(struct kvm_vcpu *vcpu)
196 return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
199 static inline int is_paging(struct kvm_vcpu *vcpu)
201 return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
204 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
206 return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
209 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
211 return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
214 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
216 return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
219 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
221 return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
224 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
225 gva_t gva, gfn_t gfn, unsigned access)
227 u64 gen = kvm_memslots(vcpu->kvm)->generation;
229 if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
233 * If this is a shadow nested page table, the "GVA" is
236 vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
237 vcpu->arch.mmio_access = access;
238 vcpu->arch.mmio_gfn = gfn;
239 vcpu->arch.mmio_gen = gen;
242 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
244 return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
248 * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
249 * clear all mmio cache info.
251 #define MMIO_GVA_ANY (~(gva_t)0)
253 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
255 if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
258 vcpu->arch.mmio_gva = 0;
261 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
263 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
264 vcpu->arch.mmio_gva == (gva & PAGE_MASK))
270 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
272 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
273 vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
279 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
281 unsigned long val = kvm_register_read_raw(vcpu, reg);
283 return is_64_bit_mode(vcpu) ? val : (u32)val;
286 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
287 int reg, unsigned long val)
289 if (!is_64_bit_mode(vcpu))
291 return kvm_register_write_raw(vcpu, reg, val);
294 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
296 return !(kvm->arch.disabled_quirks & quirk);
299 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
301 return is_smm(vcpu) || static_call(kvm_x86_apic_init_signal_blocked)(vcpu);
304 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
306 u64 get_kvmclock_ns(struct kvm *kvm);
308 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
309 gva_t addr, void *val, unsigned int bytes,
310 struct x86_exception *exception);
312 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
313 gva_t addr, void *val, unsigned int bytes,
314 struct x86_exception *exception);
316 int handle_ud(struct kvm_vcpu *vcpu);
318 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
320 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
321 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
322 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
323 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
324 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
325 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
327 bool kvm_vector_hashing_enabled(void);
328 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
329 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
330 void *insn, int insn_len);
331 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
332 int emulation_type, void *insn, int insn_len);
333 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
335 extern u64 host_xcr0;
336 extern u64 supported_xcr0;
338 extern u64 supported_xss;
339 extern bool enable_pmu;
341 static inline bool kvm_mpx_supported(void)
343 return (supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
344 == (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
347 extern unsigned int min_timer_period_us;
349 extern bool enable_vmware_backdoor;
351 extern int pi_inject_timer;
353 extern bool report_ignored_msrs;
355 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
357 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
358 vcpu->arch.virtual_tsc_shift);
361 /* Same "calling convention" as do_div:
362 * - divide (n << 32) by base
366 #define do_shl32_div32(n, base) \
369 asm("divl %2" : "=a" (__quot), "=d" (__rem) \
370 : "rm" (base), "0" (0), "1" ((u32) n)); \
375 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
377 return kvm->arch.mwait_in_guest;
380 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
382 return kvm->arch.hlt_in_guest;
385 static inline bool kvm_pause_in_guest(struct kvm *kvm)
387 return kvm->arch.pause_in_guest;
390 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
392 return kvm->arch.cstate_in_guest;
396 /* Values are arbitrary, but must be non-zero. */
397 KVM_HANDLING_IRQ = 1,
401 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,
402 enum kvm_intr_type intr)
404 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr);
407 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
409 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0);
412 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu)
414 return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI;
417 static inline bool kvm_pat_valid(u64 data)
419 if (data & 0xF8F8F8F8F8F8F8F8ull)
421 /* 0, 1, 4, 5, 6, 7 are valid values. */
422 return (data | ((data & 0x0202020202020202ull) << 1)) == data;
425 static inline bool kvm_dr7_valid(u64 data)
427 /* Bits [63:32] are reserved */
428 return !(data >> 32);
430 static inline bool kvm_dr6_valid(u64 data)
432 /* Bits [63:32] are reserved */
433 return !(data >> 32);
437 * Trigger machine check on the host. We assume all the MSRs are already set up
438 * by the CPU and that we still run on the same CPU as the MCE occurred on.
439 * We pass a fake environment to the machine check handler because we want
440 * the guest to be always treated like user space, no matter what context
441 * it used internally.
443 static inline void kvm_machine_check(void)
445 #if defined(CONFIG_X86_MCE)
446 struct pt_regs regs = {
447 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
448 .flags = X86_EFLAGS_IF,
451 do_machine_check(®s);
455 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
456 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
457 int kvm_spec_ctrl_test_value(u64 value);
458 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
459 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
460 struct x86_exception *e);
461 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
462 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
465 * Internal error codes that are used to indicate that MSR emulation encountered
466 * an error that should result in #GP in the guest, unless userspace
469 #define KVM_MSR_RET_INVALID 2 /* in-kernel MSR emulation #GP condition */
470 #define KVM_MSR_RET_FILTERED 3 /* #GP due to userspace MSR filter */
472 #define __cr4_reserved_bits(__cpu_has, __c) \
474 u64 __reserved_bits = CR4_RESERVED_BITS; \
476 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
477 __reserved_bits |= X86_CR4_OSXSAVE; \
478 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
479 __reserved_bits |= X86_CR4_SMEP; \
480 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
481 __reserved_bits |= X86_CR4_SMAP; \
482 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
483 __reserved_bits |= X86_CR4_FSGSBASE; \
484 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
485 __reserved_bits |= X86_CR4_PKE; \
486 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
487 __reserved_bits |= X86_CR4_LA57; \
488 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
489 __reserved_bits |= X86_CR4_UMIP; \
490 if (!__cpu_has(__c, X86_FEATURE_VMX)) \
491 __reserved_bits |= X86_CR4_VMXE; \
492 if (!__cpu_has(__c, X86_FEATURE_PCID)) \
493 __reserved_bits |= X86_CR4_PCIDE; \
497 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
499 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
501 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
502 unsigned int port, void *data, unsigned int count,