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