1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_MSHYPER_H 3 #define _ASM_X86_MSHYPER_H 4 5 #include <linux/types.h> 6 #include <linux/atomic.h> 7 #include <linux/nmi.h> 8 #include <asm/io.h> 9 #include <asm/hyperv.h> 10 11 /* 12 * The below CPUID leaves are present if VersionAndFeatures.HypervisorPresent 13 * is set by CPUID(HVCPUID_VERSION_FEATURES). 14 */ 15 enum hv_cpuid_function { 16 HVCPUID_VERSION_FEATURES = 0x00000001, 17 HVCPUID_VENDOR_MAXFUNCTION = 0x40000000, 18 HVCPUID_INTERFACE = 0x40000001, 19 20 /* 21 * The remaining functions depend on the value of 22 * HVCPUID_INTERFACE 23 */ 24 HVCPUID_VERSION = 0x40000002, 25 HVCPUID_FEATURES = 0x40000003, 26 HVCPUID_ENLIGHTENMENT_INFO = 0x40000004, 27 HVCPUID_IMPLEMENTATION_LIMITS = 0x40000005, 28 }; 29 30 struct ms_hyperv_info { 31 u32 features; 32 u32 misc_features; 33 u32 hints; 34 u32 max_vp_index; 35 u32 max_lp_index; 36 }; 37 38 extern struct ms_hyperv_info ms_hyperv; 39 40 /* 41 * Declare the MSR used to setup pages used to communicate with the hypervisor. 42 */ 43 union hv_x64_msr_hypercall_contents { 44 u64 as_uint64; 45 struct { 46 u64 enable:1; 47 u64 reserved:11; 48 u64 guest_physical_address:52; 49 }; 50 }; 51 52 /* 53 * TSC page layout. 54 */ 55 56 struct ms_hyperv_tsc_page { 57 volatile u32 tsc_sequence; 58 u32 reserved1; 59 volatile u64 tsc_scale; 60 volatile s64 tsc_offset; 61 u64 reserved2[509]; 62 }; 63 64 /* 65 * The guest OS needs to register the guest ID with the hypervisor. 66 * The guest ID is a 64 bit entity and the structure of this ID is 67 * specified in the Hyper-V specification: 68 * 69 * msdn.microsoft.com/en-us/library/windows/hardware/ff542653%28v=vs.85%29.aspx 70 * 71 * While the current guideline does not specify how Linux guest ID(s) 72 * need to be generated, our plan is to publish the guidelines for 73 * Linux and other guest operating systems that currently are hosted 74 * on Hyper-V. The implementation here conforms to this yet 75 * unpublished guidelines. 76 * 77 * 78 * Bit(s) 79 * 63 - Indicates if the OS is Open Source or not; 1 is Open Source 80 * 62:56 - Os Type; Linux is 0x100 81 * 55:48 - Distro specific identification 82 * 47:16 - Linux kernel version number 83 * 15:0 - Distro specific identification 84 * 85 * 86 */ 87 88 #define HV_LINUX_VENDOR_ID 0x8100 89 90 /* 91 * Generate the guest ID based on the guideline described above. 92 */ 93 94 static inline __u64 generate_guest_id(__u64 d_info1, __u64 kernel_version, 95 __u64 d_info2) 96 { 97 __u64 guest_id = 0; 98 99 guest_id = (((__u64)HV_LINUX_VENDOR_ID) << 48); 100 guest_id |= (d_info1 << 48); 101 guest_id |= (kernel_version << 16); 102 guest_id |= d_info2; 103 104 return guest_id; 105 } 106 107 108 /* Free the message slot and signal end-of-message if required */ 109 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type) 110 { 111 /* 112 * On crash we're reading some other CPU's message page and we need 113 * to be careful: this other CPU may already had cleared the header 114 * and the host may already had delivered some other message there. 115 * In case we blindly write msg->header.message_type we're going 116 * to lose it. We can still lose a message of the same type but 117 * we count on the fact that there can only be one 118 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages 119 * on crash. 120 */ 121 if (cmpxchg(&msg->header.message_type, old_msg_type, 122 HVMSG_NONE) != old_msg_type) 123 return; 124 125 /* 126 * Make sure the write to MessageType (ie set to 127 * HVMSG_NONE) happens before we read the 128 * MessagePending and EOMing. Otherwise, the EOMing 129 * will not deliver any more messages since there is 130 * no empty slot 131 */ 132 mb(); 133 134 if (msg->header.message_flags.msg_pending) { 135 /* 136 * This will cause message queue rescan to 137 * possibly deliver another msg from the 138 * hypervisor 139 */ 140 wrmsrl(HV_X64_MSR_EOM, 0); 141 } 142 } 143 144 #define hv_init_timer(timer, tick) wrmsrl(timer, tick) 145 #define hv_init_timer_config(config, val) wrmsrl(config, val) 146 147 #define hv_get_simp(val) rdmsrl(HV_X64_MSR_SIMP, val) 148 #define hv_set_simp(val) wrmsrl(HV_X64_MSR_SIMP, val) 149 150 #define hv_get_siefp(val) rdmsrl(HV_X64_MSR_SIEFP, val) 151 #define hv_set_siefp(val) wrmsrl(HV_X64_MSR_SIEFP, val) 152 153 #define hv_get_synic_state(val) rdmsrl(HV_X64_MSR_SCONTROL, val) 154 #define hv_set_synic_state(val) wrmsrl(HV_X64_MSR_SCONTROL, val) 155 156 #define hv_get_vp_index(index) rdmsrl(HV_X64_MSR_VP_INDEX, index) 157 158 #define hv_get_synint_state(int_num, val) rdmsrl(int_num, val) 159 #define hv_set_synint_state(int_num, val) wrmsrl(int_num, val) 160 161 void hyperv_callback_vector(void); 162 #ifdef CONFIG_TRACING 163 #define trace_hyperv_callback_vector hyperv_callback_vector 164 #endif 165 void hyperv_vector_handler(struct pt_regs *regs); 166 void hv_setup_vmbus_irq(void (*handler)(void)); 167 void hv_remove_vmbus_irq(void); 168 169 void hv_setup_kexec_handler(void (*handler)(void)); 170 void hv_remove_kexec_handler(void); 171 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs)); 172 void hv_remove_crash_handler(void); 173 174 #if IS_ENABLED(CONFIG_HYPERV) 175 extern struct clocksource *hyperv_cs; 176 extern void *hv_hypercall_pg; 177 178 static inline u64 hv_do_hypercall(u64 control, void *input, void *output) 179 { 180 u64 input_address = input ? virt_to_phys(input) : 0; 181 u64 output_address = output ? virt_to_phys(output) : 0; 182 u64 hv_status; 183 184 #ifdef CONFIG_X86_64 185 if (!hv_hypercall_pg) 186 return U64_MAX; 187 188 __asm__ __volatile__("mov %4, %%r8\n" 189 "call *%5" 190 : "=a" (hv_status), ASM_CALL_CONSTRAINT, 191 "+c" (control), "+d" (input_address) 192 : "r" (output_address), "m" (hv_hypercall_pg) 193 : "cc", "memory", "r8", "r9", "r10", "r11"); 194 #else 195 u32 input_address_hi = upper_32_bits(input_address); 196 u32 input_address_lo = lower_32_bits(input_address); 197 u32 output_address_hi = upper_32_bits(output_address); 198 u32 output_address_lo = lower_32_bits(output_address); 199 200 if (!hv_hypercall_pg) 201 return U64_MAX; 202 203 __asm__ __volatile__("call *%7" 204 : "=A" (hv_status), 205 "+c" (input_address_lo), ASM_CALL_CONSTRAINT 206 : "A" (control), 207 "b" (input_address_hi), 208 "D"(output_address_hi), "S"(output_address_lo), 209 "m" (hv_hypercall_pg) 210 : "cc", "memory"); 211 #endif /* !x86_64 */ 212 return hv_status; 213 } 214 215 #define HV_HYPERCALL_RESULT_MASK GENMASK_ULL(15, 0) 216 #define HV_HYPERCALL_FAST_BIT BIT(16) 217 #define HV_HYPERCALL_VARHEAD_OFFSET 17 218 #define HV_HYPERCALL_REP_COMP_OFFSET 32 219 #define HV_HYPERCALL_REP_COMP_MASK GENMASK_ULL(43, 32) 220 #define HV_HYPERCALL_REP_START_OFFSET 48 221 #define HV_HYPERCALL_REP_START_MASK GENMASK_ULL(59, 48) 222 223 /* Fast hypercall with 8 bytes of input and no output */ 224 static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1) 225 { 226 u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT; 227 228 #ifdef CONFIG_X86_64 229 { 230 __asm__ __volatile__("call *%4" 231 : "=a" (hv_status), ASM_CALL_CONSTRAINT, 232 "+c" (control), "+d" (input1) 233 : "m" (hv_hypercall_pg) 234 : "cc", "r8", "r9", "r10", "r11"); 235 } 236 #else 237 { 238 u32 input1_hi = upper_32_bits(input1); 239 u32 input1_lo = lower_32_bits(input1); 240 241 __asm__ __volatile__ ("call *%5" 242 : "=A"(hv_status), 243 "+c"(input1_lo), 244 ASM_CALL_CONSTRAINT 245 : "A" (control), 246 "b" (input1_hi), 247 "m" (hv_hypercall_pg) 248 : "cc", "edi", "esi"); 249 } 250 #endif 251 return hv_status; 252 } 253 254 /* 255 * Rep hypercalls. Callers of this functions are supposed to ensure that 256 * rep_count and varhead_size comply with Hyper-V hypercall definition. 257 */ 258 static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size, 259 void *input, void *output) 260 { 261 u64 control = code; 262 u64 status; 263 u16 rep_comp; 264 265 control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET; 266 control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET; 267 268 do { 269 status = hv_do_hypercall(control, input, output); 270 if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS) 271 return status; 272 273 /* Bits 32-43 of status have 'Reps completed' data. */ 274 rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >> 275 HV_HYPERCALL_REP_COMP_OFFSET; 276 277 control &= ~HV_HYPERCALL_REP_START_MASK; 278 control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET; 279 280 touch_nmi_watchdog(); 281 } while (rep_comp < rep_count); 282 283 return status; 284 } 285 286 /* 287 * Hypervisor's notion of virtual processor ID is different from 288 * Linux' notion of CPU ID. This information can only be retrieved 289 * in the context of the calling CPU. Setup a map for easy access 290 * to this information. 291 */ 292 extern u32 *hv_vp_index; 293 extern u32 hv_max_vp_index; 294 295 /** 296 * hv_cpu_number_to_vp_number() - Map CPU to VP. 297 * @cpu_number: CPU number in Linux terms 298 * 299 * This function returns the mapping between the Linux processor 300 * number and the hypervisor's virtual processor number, useful 301 * in making hypercalls and such that talk about specific 302 * processors. 303 * 304 * Return: Virtual processor number in Hyper-V terms 305 */ 306 static inline int hv_cpu_number_to_vp_number(int cpu_number) 307 { 308 return hv_vp_index[cpu_number]; 309 } 310 311 void hyperv_init(void); 312 void hyperv_setup_mmu_ops(void); 313 void hyper_alloc_mmu(void); 314 void hyperv_report_panic(struct pt_regs *regs, long err); 315 bool hv_is_hypercall_page_setup(void); 316 void hyperv_cleanup(void); 317 #else /* CONFIG_HYPERV */ 318 static inline void hyperv_init(void) {} 319 static inline bool hv_is_hypercall_page_setup(void) { return false; } 320 static inline void hyperv_cleanup(void) {} 321 static inline void hyperv_setup_mmu_ops(void) {} 322 #endif /* CONFIG_HYPERV */ 323 324 #ifdef CONFIG_HYPERV_TSCPAGE 325 struct ms_hyperv_tsc_page *hv_get_tsc_page(void); 326 static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg) 327 { 328 u64 scale, offset, cur_tsc; 329 u32 sequence; 330 331 /* 332 * The protocol for reading Hyper-V TSC page is specified in Hypervisor 333 * Top-Level Functional Specification ver. 3.0 and above. To get the 334 * reference time we must do the following: 335 * - READ ReferenceTscSequence 336 * A special '0' value indicates the time source is unreliable and we 337 * need to use something else. The currently published specification 338 * versions (up to 4.0b) contain a mistake and wrongly claim '-1' 339 * instead of '0' as the special value, see commit c35b82ef0294. 340 * - ReferenceTime = 341 * ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset 342 * - READ ReferenceTscSequence again. In case its value has changed 343 * since our first reading we need to discard ReferenceTime and repeat 344 * the whole sequence as the hypervisor was updating the page in 345 * between. 346 */ 347 do { 348 sequence = READ_ONCE(tsc_pg->tsc_sequence); 349 if (!sequence) 350 return U64_MAX; 351 /* 352 * Make sure we read sequence before we read other values from 353 * TSC page. 354 */ 355 smp_rmb(); 356 357 scale = READ_ONCE(tsc_pg->tsc_scale); 358 offset = READ_ONCE(tsc_pg->tsc_offset); 359 cur_tsc = rdtsc_ordered(); 360 361 /* 362 * Make sure we read sequence after we read all other values 363 * from TSC page. 364 */ 365 smp_rmb(); 366 367 } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence); 368 369 return mul_u64_u64_shr(cur_tsc, scale, 64) + offset; 370 } 371 372 #else 373 static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void) 374 { 375 return NULL; 376 } 377 #endif 378 #endif 379