xref: /openbmc/linux/arch/x86/kvm/x86.h (revision f5ad1c74)
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/pvclock.h>
7 #include "kvm_cache_regs.h"
8 #include "kvm_emulate.h"
9 
10 #define KVM_DEFAULT_PLE_GAP		128
11 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
12 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
13 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
14 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
15 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
16 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
17 
18 static inline unsigned int __grow_ple_window(unsigned int val,
19 		unsigned int base, unsigned int modifier, unsigned int max)
20 {
21 	u64 ret = val;
22 
23 	if (modifier < 1)
24 		return base;
25 
26 	if (modifier < base)
27 		ret *= modifier;
28 	else
29 		ret += modifier;
30 
31 	return min(ret, (u64)max);
32 }
33 
34 static inline unsigned int __shrink_ple_window(unsigned int val,
35 		unsigned int base, unsigned int modifier, unsigned int min)
36 {
37 	if (modifier < 1)
38 		return base;
39 
40 	if (modifier < base)
41 		val /= modifier;
42 	else
43 		val -= modifier;
44 
45 	return max(val, min);
46 }
47 
48 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
49 
50 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
51 {
52 	vcpu->arch.exception.pending = false;
53 	vcpu->arch.exception.injected = false;
54 }
55 
56 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
57 	bool soft)
58 {
59 	vcpu->arch.interrupt.injected = true;
60 	vcpu->arch.interrupt.soft = soft;
61 	vcpu->arch.interrupt.nr = vector;
62 }
63 
64 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
65 {
66 	vcpu->arch.interrupt.injected = false;
67 }
68 
69 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
70 {
71 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
72 		vcpu->arch.nmi_injected;
73 }
74 
75 static inline bool kvm_exception_is_soft(unsigned int nr)
76 {
77 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
78 }
79 
80 static inline bool is_protmode(struct kvm_vcpu *vcpu)
81 {
82 	return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
83 }
84 
85 static inline int is_long_mode(struct kvm_vcpu *vcpu)
86 {
87 #ifdef CONFIG_X86_64
88 	return vcpu->arch.efer & EFER_LMA;
89 #else
90 	return 0;
91 #endif
92 }
93 
94 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
95 {
96 	int cs_db, cs_l;
97 
98 	if (!is_long_mode(vcpu))
99 		return false;
100 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
101 	return cs_l;
102 }
103 
104 static inline bool is_la57_mode(struct kvm_vcpu *vcpu)
105 {
106 #ifdef CONFIG_X86_64
107 	return (vcpu->arch.efer & EFER_LMA) &&
108 		 kvm_read_cr4_bits(vcpu, X86_CR4_LA57);
109 #else
110 	return 0;
111 #endif
112 }
113 
114 static inline bool x86_exception_has_error_code(unsigned int vector)
115 {
116 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
117 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
118 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
119 
120 	return (1U << vector) & exception_has_error_code;
121 }
122 
123 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
124 {
125 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
126 }
127 
128 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
129 {
130 	++vcpu->stat.tlb_flush;
131 	kvm_x86_ops.tlb_flush_current(vcpu);
132 }
133 
134 static inline int is_pae(struct kvm_vcpu *vcpu)
135 {
136 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
137 }
138 
139 static inline int is_pse(struct kvm_vcpu *vcpu)
140 {
141 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
142 }
143 
144 static inline int is_paging(struct kvm_vcpu *vcpu)
145 {
146 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
147 }
148 
149 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
150 {
151 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
152 }
153 
154 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
155 {
156 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
157 }
158 
159 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
160 {
161 	return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
162 }
163 
164 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
165 {
166 	return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
167 }
168 
169 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
170 					gva_t gva, gfn_t gfn, unsigned access)
171 {
172 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
173 
174 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
175 		return;
176 
177 	/*
178 	 * If this is a shadow nested page table, the "GVA" is
179 	 * actually a nGPA.
180 	 */
181 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
182 	vcpu->arch.mmio_access = access;
183 	vcpu->arch.mmio_gfn = gfn;
184 	vcpu->arch.mmio_gen = gen;
185 }
186 
187 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
188 {
189 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
190 }
191 
192 /*
193  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
194  * clear all mmio cache info.
195  */
196 #define MMIO_GVA_ANY (~(gva_t)0)
197 
198 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
199 {
200 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
201 		return;
202 
203 	vcpu->arch.mmio_gva = 0;
204 }
205 
206 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
207 {
208 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
209 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
210 		return true;
211 
212 	return false;
213 }
214 
215 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
216 {
217 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
218 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
219 		return true;
220 
221 	return false;
222 }
223 
224 static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu, int reg)
225 {
226 	unsigned long val = kvm_register_read(vcpu, reg);
227 
228 	return is_64_bit_mode(vcpu) ? val : (u32)val;
229 }
230 
231 static inline void kvm_register_writel(struct kvm_vcpu *vcpu,
232 				       int reg, unsigned long val)
233 {
234 	if (!is_64_bit_mode(vcpu))
235 		val = (u32)val;
236 	return kvm_register_write(vcpu, reg, val);
237 }
238 
239 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
240 {
241 	return !(kvm->arch.disabled_quirks & quirk);
242 }
243 
244 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
245 {
246 	return is_smm(vcpu) || kvm_x86_ops.apic_init_signal_blocked(vcpu);
247 }
248 
249 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
250 
251 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr);
252 u64 get_kvmclock_ns(struct kvm *kvm);
253 
254 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
255 	gva_t addr, void *val, unsigned int bytes,
256 	struct x86_exception *exception);
257 
258 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
259 	gva_t addr, void *val, unsigned int bytes,
260 	struct x86_exception *exception);
261 
262 int handle_ud(struct kvm_vcpu *vcpu);
263 
264 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
265 
266 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
267 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
268 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
269 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
270 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
271 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
272 					  int page_num);
273 bool kvm_vector_hashing_enabled(void);
274 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
275 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
276 			    int emulation_type, void *insn, int insn_len);
277 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
278 
279 extern u64 host_xcr0;
280 extern u64 supported_xcr0;
281 extern u64 supported_xss;
282 
283 static inline bool kvm_mpx_supported(void)
284 {
285 	return (supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
286 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
287 }
288 
289 extern unsigned int min_timer_period_us;
290 
291 extern bool enable_vmware_backdoor;
292 
293 extern int pi_inject_timer;
294 
295 extern struct static_key kvm_no_apic_vcpu;
296 
297 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
298 {
299 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
300 				   vcpu->arch.virtual_tsc_shift);
301 }
302 
303 /* Same "calling convention" as do_div:
304  * - divide (n << 32) by base
305  * - put result in n
306  * - return remainder
307  */
308 #define do_shl32_div32(n, base)					\
309 	({							\
310 	    u32 __quot, __rem;					\
311 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
312 			: "rm" (base), "0" (0), "1" ((u32) n));	\
313 	    n = __quot;						\
314 	    __rem;						\
315 	 })
316 
317 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
318 {
319 	return kvm->arch.mwait_in_guest;
320 }
321 
322 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
323 {
324 	return kvm->arch.hlt_in_guest;
325 }
326 
327 static inline bool kvm_pause_in_guest(struct kvm *kvm)
328 {
329 	return kvm->arch.pause_in_guest;
330 }
331 
332 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
333 {
334 	return kvm->arch.cstate_in_guest;
335 }
336 
337 DECLARE_PER_CPU(struct kvm_vcpu *, current_vcpu);
338 
339 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu)
340 {
341 	__this_cpu_write(current_vcpu, vcpu);
342 }
343 
344 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
345 {
346 	__this_cpu_write(current_vcpu, NULL);
347 }
348 
349 
350 static inline bool kvm_pat_valid(u64 data)
351 {
352 	if (data & 0xF8F8F8F8F8F8F8F8ull)
353 		return false;
354 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
355 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
356 }
357 
358 static inline bool kvm_dr7_valid(u64 data)
359 {
360 	/* Bits [63:32] are reserved */
361 	return !(data >> 32);
362 }
363 static inline bool kvm_dr6_valid(u64 data)
364 {
365 	/* Bits [63:32] are reserved */
366 	return !(data >> 32);
367 }
368 
369 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
370 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
371 int kvm_spec_ctrl_test_value(u64 value);
372 int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
373 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu);
374 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
375 			      struct x86_exception *e);
376 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
377 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
378 
379 /*
380  * Internal error codes that are used to indicate that MSR emulation encountered
381  * an error that should result in #GP in the guest, unless userspace
382  * handles it.
383  */
384 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
385 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
386 
387 #define __cr4_reserved_bits(__cpu_has, __c)             \
388 ({                                                      \
389 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
390                                                         \
391 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
392 		__reserved_bits |= X86_CR4_OSXSAVE;     \
393 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
394 		__reserved_bits |= X86_CR4_SMEP;        \
395 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
396 		__reserved_bits |= X86_CR4_SMAP;        \
397 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
398 		__reserved_bits |= X86_CR4_FSGSBASE;    \
399 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
400 		__reserved_bits |= X86_CR4_PKE;         \
401 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
402 		__reserved_bits |= X86_CR4_LA57;        \
403 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
404 		__reserved_bits |= X86_CR4_UMIP;        \
405 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
406 		__reserved_bits |= X86_CR4_VMXE;        \
407 	__reserved_bits;                                \
408 })
409 
410 #endif
411