arch/x86/include/asm/mshyperv.h

   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 #include <asm/nospec-branch.h>
  11
  12 /*
  13  * The below CPUID leaves are present if VersionAndFeatures.HypervisorPresent
  14  * is set by CPUID(HVCPUID_VERSION_FEATURES).
  15  */
  16 enum hv_cpuid_function {
  17         HVCPUID_VERSION_FEATURES                = 0x00000001,
  18         HVCPUID_VENDOR_MAXFUNCTION              = 0x40000000,
  19         HVCPUID_INTERFACE                       = 0x40000001,
  20
  21         /*
  22          * The remaining functions depend on the value of
  23          * HVCPUID_INTERFACE
  24          */
  25         HVCPUID_VERSION                         = 0x40000002,
  26         HVCPUID_FEATURES                        = 0x40000003,
  27         HVCPUID_ENLIGHTENMENT_INFO              = 0x40000004,
  28         HVCPUID_IMPLEMENTATION_LIMITS           = 0x40000005,
  29 };
  30
  31 struct ms_hyperv_info {
  32         u32 features;
  33         u32 misc_features;
  34         u32 hints;
  35         u32 max_vp_index;
  36         u32 max_lp_index;
  37 };
  38
  39 extern struct ms_hyperv_info ms_hyperv;
  40
  41 /*
  42  * Declare the MSR used to setup pages used to communicate with the hypervisor.
  43  */
  44 union hv_x64_msr_hypercall_contents {
  45         u64 as_uint64;
  46         struct {
  47                 u64 enable:1;
  48                 u64 reserved:11;
  49                 u64 guest_physical_address:52;
  50         };
  51 };
  52
  53 /*
  54  * TSC page layout.
  55  */
  56
  57 struct ms_hyperv_tsc_page {
  58         volatile u32 tsc_sequence;
  59         u32 reserved1;
  60         volatile u64 tsc_scale;
  61         volatile s64 tsc_offset;
  62         u64 reserved2[509];
  63 };
  64
  65 /*
  66  * The guest OS needs to register the guest ID with the hypervisor.
  67  * The guest ID is a 64 bit entity and the structure of this ID is
  68  * specified in the Hyper-V specification:
  69  *
  70  * msdn.microsoft.com/en-us/library/windows/hardware/ff542653%28v=vs.85%29.aspx
  71  *
  72  * While the current guideline does not specify how Linux guest ID(s)
  73  * need to be generated, our plan is to publish the guidelines for
  74  * Linux and other guest operating systems that currently are hosted
  75  * on Hyper-V. The implementation here conforms to this yet
  76  * unpublished guidelines.
  77  *
  78  *
  79  * Bit(s)
  80  * 63 - Indicates if the OS is Open Source or not; 1 is Open Source
  81  * 62:56 - Os Type; Linux is 0x100
  82  * 55:48 - Distro specific identification
  83  * 47:16 - Linux kernel version number
  84  * 15:0  - Distro specific identification
  85  *
  86  *
  87  */
  88
  89 #define HV_LINUX_VENDOR_ID              0x8100
  90
  91 /*
  92  * Generate the guest ID based on the guideline described above.
  93  */
  94
  95 static inline  __u64 generate_guest_id(__u64 d_info1, __u64 kernel_version,
  96                                        __u64 d_info2)
  97 {
  98         __u64 guest_id = 0;
  99
 100         guest_id = (((__u64)HV_LINUX_VENDOR_ID) << 48);
 101         guest_id |= (d_info1 << 48);
 102         guest_id |= (kernel_version << 16);
 103         guest_id |= d_info2;
 104
 105         return guest_id;
 106 }
 107
 108
 109 /* Free the message slot and signal end-of-message if required */
 110 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type)
 111 {
 112         /*
 113          * On crash we're reading some other CPU's message page and we need
 114          * to be careful: this other CPU may already had cleared the header
 115          * and the host may already had delivered some other message there.
 116          * In case we blindly write msg->header.message_type we're going
 117          * to lose it. We can still lose a message of the same type but
 118          * we count on the fact that there can only be one
 119          * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages
 120          * on crash.
 121          */
 122         if (cmpxchg(&msg->header.message_type, old_msg_type,
 123                     HVMSG_NONE) != old_msg_type)
 124                 return;
 125
 126         /*
 127          * Make sure the write to MessageType (ie set to
 128          * HVMSG_NONE) happens before we read the
 129          * MessagePending and EOMing. Otherwise, the EOMing
 130          * will not deliver any more messages since there is
 131          * no empty slot
 132          */
 133         mb();
 134
 135         if (msg->header.message_flags.msg_pending) {
 136                 /*
 137                  * This will cause message queue rescan to
 138                  * possibly deliver another msg from the
 139                  * hypervisor
 140                  */
 141                 wrmsrl(HV_X64_MSR_EOM, 0);
 142         }
 143 }
 144
 145 #define hv_init_timer(timer, tick) wrmsrl(timer, tick)
 146 #define hv_init_timer_config(config, val) wrmsrl(config, val)
 147
 148 #define hv_get_simp(val) rdmsrl(HV_X64_MSR_SIMP, val)
 149 #define hv_set_simp(val) wrmsrl(HV_X64_MSR_SIMP, val)
 150
 151 #define hv_get_siefp(val) rdmsrl(HV_X64_MSR_SIEFP, val)
 152 #define hv_set_siefp(val) wrmsrl(HV_X64_MSR_SIEFP, val)
 153
 154 #define hv_get_synic_state(val) rdmsrl(HV_X64_MSR_SCONTROL, val)
 155 #define hv_set_synic_state(val) wrmsrl(HV_X64_MSR_SCONTROL, val)
 156
 157 #define hv_get_vp_index(index) rdmsrl(HV_X64_MSR_VP_INDEX, index)
 158
 159 #define hv_get_synint_state(int_num, val) rdmsrl(int_num, val)
 160 #define hv_set_synint_state(int_num, val) wrmsrl(int_num, val)
 161
 162 void hyperv_callback_vector(void);
 163 void hyperv_reenlightenment_vector(void);
 164 #ifdef CONFIG_TRACING
 165 #define trace_hyperv_callback_vector hyperv_callback_vector
 166 #endif
 167 void hyperv_vector_handler(struct pt_regs *regs);
 168 void hv_setup_vmbus_irq(void (*handler)(void));
 169 void hv_remove_vmbus_irq(void);
 170
 171 void hv_setup_kexec_handler(void (*handler)(void));
 172 void hv_remove_kexec_handler(void);
 173 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs));
 174 void hv_remove_crash_handler(void);
 175
 176 #if IS_ENABLED(CONFIG_HYPERV)
 177 extern struct clocksource *hyperv_cs;
 178 extern void *hv_hypercall_pg;
 179
 180 static inline u64 hv_do_hypercall(u64 control, void *input, void *output)
 181 {
 182         u64 input_address = input ? virt_to_phys(input) : 0;
 183         u64 output_address = output ? virt_to_phys(output) : 0;
 184         u64 hv_status;
 185
 186 #ifdef CONFIG_X86_64
 187         if (!hv_hypercall_pg)
 188                 return U64_MAX;
 189
 190         __asm__ __volatile__("mov %4, %%r8\n"
 191                              CALL_NOSPEC
 192                              : "=a" (hv_status), ASM_CALL_CONSTRAINT,
 193                                "+c" (control), "+d" (input_address)
 194                              :  "r" (output_address),
 195                                 THUNK_TARGET(hv_hypercall_pg)
 196                              : "cc", "memory", "r8", "r9", "r10", "r11");
 197 #else
 198         u32 input_address_hi = upper_32_bits(input_address);
 199         u32 input_address_lo = lower_32_bits(input_address);
 200         u32 output_address_hi = upper_32_bits(output_address);
 201         u32 output_address_lo = lower_32_bits(output_address);
 202
 203         if (!hv_hypercall_pg)
 204                 return U64_MAX;
 205
 206         __asm__ __volatile__(CALL_NOSPEC
 207                              : "=A" (hv_status),
 208                                "+c" (input_address_lo), ASM_CALL_CONSTRAINT
 209                              : "A" (control),
 210                                "b" (input_address_hi),
 211                                "D"(output_address_hi), "S"(output_address_lo),
 212                                THUNK_TARGET(hv_hypercall_pg)
 213                              : "cc", "memory");
 214 #endif /* !x86_64 */
 215         return hv_status;
 216 }
 217
 218 #define HV_HYPERCALL_RESULT_MASK        GENMASK_ULL(15, 0)
 219 #define HV_HYPERCALL_FAST_BIT           BIT(16)
 220 #define HV_HYPERCALL_VARHEAD_OFFSET     17
 221 #define HV_HYPERCALL_REP_COMP_OFFSET    32
 222 #define HV_HYPERCALL_REP_COMP_MASK      GENMASK_ULL(43, 32)
 223 #define HV_HYPERCALL_REP_START_OFFSET   48
 224 #define HV_HYPERCALL_REP_START_MASK     GENMASK_ULL(59, 48)
 225
 226 /* Fast hypercall with 8 bytes of input and no output */
 227 static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1)
 228 {
 229         u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT;
 230
 231 #ifdef CONFIG_X86_64
 232         {
 233                 __asm__ __volatile__(CALL_NOSPEC
 234                                      : "=a" (hv_status), ASM_CALL_CONSTRAINT,
 235                                        "+c" (control), "+d" (input1)
 236                                      : THUNK_TARGET(hv_hypercall_pg)
 237                                      : "cc", "r8", "r9", "r10", "r11");
 238         }
 239 #else
 240         {
 241                 u32 input1_hi = upper_32_bits(input1);
 242                 u32 input1_lo = lower_32_bits(input1);
 243
 244                 __asm__ __volatile__ (CALL_NOSPEC
 245                                       : "=A"(hv_status),
 246                                         "+c"(input1_lo),
 247                                         ASM_CALL_CONSTRAINT
 248                                       : "A" (control),
 249                                         "b" (input1_hi),
 250                                         THUNK_TARGET(hv_hypercall_pg)
 251                                       : "cc", "edi", "esi");
 252         }
 253 #endif
 254                 return hv_status;
 255 }
 256
 257 /*
 258  * Rep hypercalls. Callers of this functions are supposed to ensure that
 259  * rep_count and varhead_size comply with Hyper-V hypercall definition.
 260  */
 261 static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size,
 262                                       void *input, void *output)
 263 {
 264         u64 control = code;
 265         u64 status;
 266         u16 rep_comp;
 267
 268         control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET;
 269         control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET;
 270
 271         do {
 272                 status = hv_do_hypercall(control, input, output);
 273                 if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS)
 274                         return status;
 275
 276                 /* Bits 32-43 of status have 'Reps completed' data. */
 277                 rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >>
 278                         HV_HYPERCALL_REP_COMP_OFFSET;
 279
 280                 control &= ~HV_HYPERCALL_REP_START_MASK;
 281                 control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET;
 282
 283                 touch_nmi_watchdog();
 284         } while (rep_comp < rep_count);
 285
 286         return status;
 287 }
 288
 289 /*
 290  * Hypervisor's notion of virtual processor ID is different from
 291  * Linux' notion of CPU ID. This information can only be retrieved
 292  * in the context of the calling CPU. Setup a map for easy access
 293  * to this information.
 294  */
 295 extern u32 *hv_vp_index;
 296 extern u32 hv_max_vp_index;
 297
 298 /**
 299  * hv_cpu_number_to_vp_number() - Map CPU to VP.
 300  * @cpu_number: CPU number in Linux terms
 301  *
 302  * This function returns the mapping between the Linux processor
 303  * number and the hypervisor's virtual processor number, useful
 304  * in making hypercalls and such that talk about specific
 305  * processors.
 306  *
 307  * Return: Virtual processor number in Hyper-V terms
 308  */
 309 static inline int hv_cpu_number_to_vp_number(int cpu_number)
 310 {
 311         return hv_vp_index[cpu_number];
 312 }
 313
 314 void hyperv_init(void);
 315 void hyperv_setup_mmu_ops(void);
 316 void hyper_alloc_mmu(void);
 317 void hyperv_report_panic(struct pt_regs *regs, long err);
 318 bool hv_is_hyperv_initialized(void);
 319 void hyperv_cleanup(void);
 320
 321 void hyperv_reenlightenment_intr(struct pt_regs *regs);
 322 void set_hv_tscchange_cb(void (*cb)(void));
 323 void clear_hv_tscchange_cb(void);
 324 void hyperv_stop_tsc_emulation(void);
 325 #else /* CONFIG_HYPERV */
 326 static inline void hyperv_init(void) {}
 327 static inline bool hv_is_hyperv_initialized(void) { return false; }
 328 static inline void hyperv_cleanup(void) {}
 329 static inline void hyperv_setup_mmu_ops(void) {}
 330 static inline void set_hv_tscchange_cb(void (*cb)(void)) {}
 331 static inline void clear_hv_tscchange_cb(void) {}
 332 static inline void hyperv_stop_tsc_emulation(void) {};
 333 #endif /* CONFIG_HYPERV */
 334
 335 #ifdef CONFIG_HYPERV_TSCPAGE
 336 struct ms_hyperv_tsc_page *hv_get_tsc_page(void);
 337 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
 338                                        u64 *cur_tsc)
 339 {
 340         u64 scale, offset;
 341         u32 sequence;
 342
 343         /*
 344          * The protocol for reading Hyper-V TSC page is specified in Hypervisor
 345          * Top-Level Functional Specification ver. 3.0 and above. To get the
 346          * reference time we must do the following:
 347          * - READ ReferenceTscSequence
 348          *   A special '0' value indicates the time source is unreliable and we
 349          *   need to use something else. The currently published specification
 350          *   versions (up to 4.0b) contain a mistake and wrongly claim '-1'
 351          *   instead of '0' as the special value, see commit c35b82ef0294.
 352          * - ReferenceTime =
 353          *        ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset
 354          * - READ ReferenceTscSequence again. In case its value has changed
 355          *   since our first reading we need to discard ReferenceTime and repeat
 356          *   the whole sequence as the hypervisor was updating the page in
 357          *   between.
 358          */
 359         do {
 360                 sequence = READ_ONCE(tsc_pg->tsc_sequence);
 361                 if (!sequence)
 362                         return U64_MAX;
 363                 /*
 364                  * Make sure we read sequence before we read other values from
 365                  * TSC page.
 366                  */
 367                 smp_rmb();
 368
 369                 scale = READ_ONCE(tsc_pg->tsc_scale);
 370                 offset = READ_ONCE(tsc_pg->tsc_offset);
 371                 *cur_tsc = rdtsc_ordered();
 372
 373                 /*
 374                  * Make sure we read sequence after we read all other values
 375                  * from TSC page.
 376                  */
 377                 smp_rmb();
 378
 379         } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
 380
 381         return mul_u64_u64_shr(*cur_tsc, scale, 64) + offset;
 382 }
 383
 384 static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
 385 {
 386         u64 cur_tsc;
 387
 388         return hv_read_tsc_page_tsc(tsc_pg, &cur_tsc);
 389 }
 390
 391 #else
 392 static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
 393 {
 394         return NULL;
 395 }
 396
 397 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
 398                                        u64 *cur_tsc)
 399 {
 400         BUG();
 401         return U64_MAX;
 402 }
 403 #endif
 404 #endif