xref: /openbmc/linux/arch/x86/kvm/x86.h (revision 5086ea4b)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef ARCH_X86_KVM_X86_H
3 #define ARCH_X86_KVM_X86_H
4 
5 #include <linux/kvm_host.h>
6 #include <asm/mce.h>
7 #include <asm/pvclock.h>
8 #include "kvm_cache_regs.h"
9 #include "kvm_emulate.h"
10 
11 #define KVM_DEFAULT_PLE_GAP		128
12 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
13 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
14 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
15 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
16 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
17 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
18 
19 static inline unsigned int __grow_ple_window(unsigned int val,
20 		unsigned int base, unsigned int modifier, unsigned int max)
21 {
22 	u64 ret = val;
23 
24 	if (modifier < 1)
25 		return base;
26 
27 	if (modifier < base)
28 		ret *= modifier;
29 	else
30 		ret += modifier;
31 
32 	return min(ret, (u64)max);
33 }
34 
35 static inline unsigned int __shrink_ple_window(unsigned int val,
36 		unsigned int base, unsigned int modifier, unsigned int min)
37 {
38 	if (modifier < 1)
39 		return base;
40 
41 	if (modifier < base)
42 		val /= modifier;
43 	else
44 		val -= modifier;
45 
46 	return max(val, min);
47 }
48 
49 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
50 
51 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
52 {
53 	vcpu->arch.exception.pending = false;
54 	vcpu->arch.exception.injected = false;
55 }
56 
57 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
58 	bool soft)
59 {
60 	vcpu->arch.interrupt.injected = true;
61 	vcpu->arch.interrupt.soft = soft;
62 	vcpu->arch.interrupt.nr = vector;
63 }
64 
65 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
66 {
67 	vcpu->arch.interrupt.injected = false;
68 }
69 
70 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
71 {
72 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
73 		vcpu->arch.nmi_injected;
74 }
75 
76 static inline bool kvm_exception_is_soft(unsigned int nr)
77 {
78 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
79 }
80 
81 static inline bool is_protmode(struct kvm_vcpu *vcpu)
82 {
83 	return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
84 }
85 
86 static inline int is_long_mode(struct kvm_vcpu *vcpu)
87 {
88 #ifdef CONFIG_X86_64
89 	return vcpu->arch.efer & EFER_LMA;
90 #else
91 	return 0;
92 #endif
93 }
94 
95 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
96 {
97 	int cs_db, cs_l;
98 
99 	if (!is_long_mode(vcpu))
100 		return false;
101 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
102 	return cs_l;
103 }
104 
105 static inline bool is_la57_mode(struct kvm_vcpu *vcpu)
106 {
107 #ifdef CONFIG_X86_64
108 	return (vcpu->arch.efer & EFER_LMA) &&
109 		 kvm_read_cr4_bits(vcpu, X86_CR4_LA57);
110 #else
111 	return 0;
112 #endif
113 }
114 
115 static inline bool x86_exception_has_error_code(unsigned int vector)
116 {
117 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
118 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
119 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
120 
121 	return (1U << vector) & exception_has_error_code;
122 }
123 
124 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
125 {
126 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
127 }
128 
129 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
130 {
131 	++vcpu->stat.tlb_flush;
132 	static_call(kvm_x86_tlb_flush_current)(vcpu);
133 }
134 
135 static inline int is_pae(struct kvm_vcpu *vcpu)
136 {
137 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
138 }
139 
140 static inline int is_pse(struct kvm_vcpu *vcpu)
141 {
142 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
143 }
144 
145 static inline int is_paging(struct kvm_vcpu *vcpu)
146 {
147 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
148 }
149 
150 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
151 {
152 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
153 }
154 
155 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
156 {
157 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
158 }
159 
160 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
161 {
162 	return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
163 }
164 
165 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
166 {
167 	return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
168 }
169 
170 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
171 					gva_t gva, gfn_t gfn, unsigned access)
172 {
173 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
174 
175 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
176 		return;
177 
178 	/*
179 	 * If this is a shadow nested page table, the "GVA" is
180 	 * actually a nGPA.
181 	 */
182 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
183 	vcpu->arch.mmio_access = access;
184 	vcpu->arch.mmio_gfn = gfn;
185 	vcpu->arch.mmio_gen = gen;
186 }
187 
188 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
189 {
190 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
191 }
192 
193 /*
194  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
195  * clear all mmio cache info.
196  */
197 #define MMIO_GVA_ANY (~(gva_t)0)
198 
199 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
200 {
201 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
202 		return;
203 
204 	vcpu->arch.mmio_gva = 0;
205 }
206 
207 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
208 {
209 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
210 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
211 		return true;
212 
213 	return false;
214 }
215 
216 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
217 {
218 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
219 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
220 		return true;
221 
222 	return false;
223 }
224 
225 static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu, int reg)
226 {
227 	unsigned long val = kvm_register_read(vcpu, reg);
228 
229 	return is_64_bit_mode(vcpu) ? val : (u32)val;
230 }
231 
232 static inline void kvm_register_writel(struct kvm_vcpu *vcpu,
233 				       int reg, unsigned long val)
234 {
235 	if (!is_64_bit_mode(vcpu))
236 		val = (u32)val;
237 	return kvm_register_write(vcpu, reg, val);
238 }
239 
240 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
241 {
242 	return !(kvm->arch.disabled_quirks & quirk);
243 }
244 
245 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
246 {
247 	return is_smm(vcpu) || static_call(kvm_x86_apic_init_signal_blocked)(vcpu);
248 }
249 
250 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs);
251 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
252 
253 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr);
254 u64 get_kvmclock_ns(struct kvm *kvm);
255 
256 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
257 	gva_t addr, void *val, unsigned int bytes,
258 	struct x86_exception *exception);
259 
260 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
261 	gva_t addr, void *val, unsigned int bytes,
262 	struct x86_exception *exception);
263 
264 int handle_ud(struct kvm_vcpu *vcpu);
265 
266 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
267 
268 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
269 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
270 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
271 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
272 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
273 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
274 					  int page_num);
275 bool kvm_vector_hashing_enabled(void);
276 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
277 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
278 				    void *insn, int insn_len);
279 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
280 			    int emulation_type, void *insn, int insn_len);
281 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
282 
283 extern u64 host_xcr0;
284 extern u64 supported_xcr0;
285 extern u64 host_xss;
286 extern u64 supported_xss;
287 
288 static inline bool kvm_mpx_supported(void)
289 {
290 	return (supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
291 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
292 }
293 
294 extern unsigned int min_timer_period_us;
295 
296 extern bool enable_vmware_backdoor;
297 
298 extern int pi_inject_timer;
299 
300 extern struct static_key kvm_no_apic_vcpu;
301 
302 extern bool report_ignored_msrs;
303 
304 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
305 {
306 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
307 				   vcpu->arch.virtual_tsc_shift);
308 }
309 
310 /* Same "calling convention" as do_div:
311  * - divide (n << 32) by base
312  * - put result in n
313  * - return remainder
314  */
315 #define do_shl32_div32(n, base)					\
316 	({							\
317 	    u32 __quot, __rem;					\
318 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
319 			: "rm" (base), "0" (0), "1" ((u32) n));	\
320 	    n = __quot;						\
321 	    __rem;						\
322 	 })
323 
324 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
325 {
326 	return kvm->arch.mwait_in_guest;
327 }
328 
329 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
330 {
331 	return kvm->arch.hlt_in_guest;
332 }
333 
334 static inline bool kvm_pause_in_guest(struct kvm *kvm)
335 {
336 	return kvm->arch.pause_in_guest;
337 }
338 
339 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
340 {
341 	return kvm->arch.cstate_in_guest;
342 }
343 
344 DECLARE_PER_CPU(struct kvm_vcpu *, current_vcpu);
345 
346 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu)
347 {
348 	__this_cpu_write(current_vcpu, vcpu);
349 }
350 
351 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
352 {
353 	__this_cpu_write(current_vcpu, NULL);
354 }
355 
356 
357 static inline bool kvm_pat_valid(u64 data)
358 {
359 	if (data & 0xF8F8F8F8F8F8F8F8ull)
360 		return false;
361 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
362 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
363 }
364 
365 static inline bool kvm_dr7_valid(u64 data)
366 {
367 	/* Bits [63:32] are reserved */
368 	return !(data >> 32);
369 }
370 static inline bool kvm_dr6_valid(u64 data)
371 {
372 	/* Bits [63:32] are reserved */
373 	return !(data >> 32);
374 }
375 
376 /*
377  * Trigger machine check on the host. We assume all the MSRs are already set up
378  * by the CPU and that we still run on the same CPU as the MCE occurred on.
379  * We pass a fake environment to the machine check handler because we want
380  * the guest to be always treated like user space, no matter what context
381  * it used internally.
382  */
383 static inline void kvm_machine_check(void)
384 {
385 #if defined(CONFIG_X86_MCE)
386 	struct pt_regs regs = {
387 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
388 		.flags = X86_EFLAGS_IF,
389 	};
390 
391 	do_machine_check(&regs);
392 #endif
393 }
394 
395 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
396 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
397 int kvm_spec_ctrl_test_value(u64 value);
398 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
399 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
400 			      struct x86_exception *e);
401 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
402 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
403 
404 /*
405  * Internal error codes that are used to indicate that MSR emulation encountered
406  * an error that should result in #GP in the guest, unless userspace
407  * handles it.
408  */
409 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
410 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
411 
412 #define __cr4_reserved_bits(__cpu_has, __c)             \
413 ({                                                      \
414 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
415                                                         \
416 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
417 		__reserved_bits |= X86_CR4_OSXSAVE;     \
418 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
419 		__reserved_bits |= X86_CR4_SMEP;        \
420 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
421 		__reserved_bits |= X86_CR4_SMAP;        \
422 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
423 		__reserved_bits |= X86_CR4_FSGSBASE;    \
424 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
425 		__reserved_bits |= X86_CR4_PKE;         \
426 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
427 		__reserved_bits |= X86_CR4_LA57;        \
428 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
429 		__reserved_bits |= X86_CR4_UMIP;        \
430 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
431 		__reserved_bits |= X86_CR4_VMXE;        \
432 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
433 		__reserved_bits |= X86_CR4_PCIDE;       \
434 	__reserved_bits;                                \
435 })
436 
437 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
438 			  void *dst);
439 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
440 			 void *dst);
441 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
442 			 unsigned int port, void *data,  unsigned int count,
443 			 int in);
444 
445 #endif
446