1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * X86 specific Hyper-V initialization code. 4 * 5 * Copyright (C) 2016, Microsoft, Inc. 6 * 7 * Author : K. Y. Srinivasan <kys@microsoft.com> 8 */ 9 10 #include <linux/acpi.h> 11 #include <linux/efi.h> 12 #include <linux/types.h> 13 #include <asm/apic.h> 14 #include <asm/desc.h> 15 #include <asm/hypervisor.h> 16 #include <asm/hyperv-tlfs.h> 17 #include <asm/mshyperv.h> 18 #include <asm/idtentry.h> 19 #include <linux/kexec.h> 20 #include <linux/version.h> 21 #include <linux/vmalloc.h> 22 #include <linux/mm.h> 23 #include <linux/hyperv.h> 24 #include <linux/slab.h> 25 #include <linux/kernel.h> 26 #include <linux/cpuhotplug.h> 27 #include <linux/syscore_ops.h> 28 #include <clocksource/hyperv_timer.h> 29 30 int hyperv_init_cpuhp; 31 32 void *hv_hypercall_pg; 33 EXPORT_SYMBOL_GPL(hv_hypercall_pg); 34 35 /* Storage to save the hypercall page temporarily for hibernation */ 36 static void *hv_hypercall_pg_saved; 37 38 u32 *hv_vp_index; 39 EXPORT_SYMBOL_GPL(hv_vp_index); 40 41 struct hv_vp_assist_page **hv_vp_assist_page; 42 EXPORT_SYMBOL_GPL(hv_vp_assist_page); 43 44 void __percpu **hyperv_pcpu_input_arg; 45 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg); 46 47 u32 hv_max_vp_index; 48 EXPORT_SYMBOL_GPL(hv_max_vp_index); 49 50 void *hv_alloc_hyperv_page(void) 51 { 52 BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE); 53 54 return (void *)__get_free_page(GFP_KERNEL); 55 } 56 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_page); 57 58 void *hv_alloc_hyperv_zeroed_page(void) 59 { 60 BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE); 61 62 return (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO); 63 } 64 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_zeroed_page); 65 66 void hv_free_hyperv_page(unsigned long addr) 67 { 68 free_page(addr); 69 } 70 EXPORT_SYMBOL_GPL(hv_free_hyperv_page); 71 72 static int hv_cpu_init(unsigned int cpu) 73 { 74 u64 msr_vp_index; 75 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()]; 76 void **input_arg; 77 struct page *pg; 78 79 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg); 80 /* hv_cpu_init() can be called with IRQs disabled from hv_resume() */ 81 pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL); 82 if (unlikely(!pg)) 83 return -ENOMEM; 84 *input_arg = page_address(pg); 85 86 hv_get_vp_index(msr_vp_index); 87 88 hv_vp_index[smp_processor_id()] = msr_vp_index; 89 90 if (msr_vp_index > hv_max_vp_index) 91 hv_max_vp_index = msr_vp_index; 92 93 if (!hv_vp_assist_page) 94 return 0; 95 96 /* 97 * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section 98 * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure 99 * we always write the EOI MSR in hv_apic_eoi_write() *after* the 100 * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may 101 * not be stopped in the case of CPU offlining and the VM will hang. 102 */ 103 if (!*hvp) { 104 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO); 105 } 106 107 if (*hvp) { 108 u64 val; 109 110 val = vmalloc_to_pfn(*hvp); 111 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) | 112 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE; 113 114 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val); 115 } 116 117 return 0; 118 } 119 120 static void (*hv_reenlightenment_cb)(void); 121 122 static void hv_reenlightenment_notify(struct work_struct *dummy) 123 { 124 struct hv_tsc_emulation_status emu_status; 125 126 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 127 128 /* Don't issue the callback if TSC accesses are not emulated */ 129 if (hv_reenlightenment_cb && emu_status.inprogress) 130 hv_reenlightenment_cb(); 131 } 132 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify); 133 134 void hyperv_stop_tsc_emulation(void) 135 { 136 u64 freq; 137 struct hv_tsc_emulation_status emu_status; 138 139 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 140 emu_status.inprogress = 0; 141 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 142 143 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq); 144 tsc_khz = div64_u64(freq, 1000); 145 } 146 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation); 147 148 static inline bool hv_reenlightenment_available(void) 149 { 150 /* 151 * Check for required features and priviliges to make TSC frequency 152 * change notifications work. 153 */ 154 return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 155 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE && 156 ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT; 157 } 158 159 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment) 160 { 161 ack_APIC_irq(); 162 inc_irq_stat(irq_hv_reenlightenment_count); 163 schedule_delayed_work(&hv_reenlightenment_work, HZ/10); 164 } 165 166 void set_hv_tscchange_cb(void (*cb)(void)) 167 { 168 struct hv_reenlightenment_control re_ctrl = { 169 .vector = HYPERV_REENLIGHTENMENT_VECTOR, 170 .enabled = 1, 171 .target_vp = hv_vp_index[smp_processor_id()] 172 }; 173 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1}; 174 175 if (!hv_reenlightenment_available()) { 176 pr_warn("Hyper-V: reenlightenment support is unavailable\n"); 177 return; 178 } 179 180 hv_reenlightenment_cb = cb; 181 182 /* Make sure callback is registered before we write to MSRs */ 183 wmb(); 184 185 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 186 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl)); 187 } 188 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb); 189 190 void clear_hv_tscchange_cb(void) 191 { 192 struct hv_reenlightenment_control re_ctrl; 193 194 if (!hv_reenlightenment_available()) 195 return; 196 197 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl); 198 re_ctrl.enabled = 0; 199 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl); 200 201 hv_reenlightenment_cb = NULL; 202 } 203 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb); 204 205 static int hv_cpu_die(unsigned int cpu) 206 { 207 struct hv_reenlightenment_control re_ctrl; 208 unsigned int new_cpu; 209 unsigned long flags; 210 void **input_arg; 211 void *input_pg = NULL; 212 213 local_irq_save(flags); 214 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg); 215 input_pg = *input_arg; 216 *input_arg = NULL; 217 local_irq_restore(flags); 218 free_page((unsigned long)input_pg); 219 220 if (hv_vp_assist_page && hv_vp_assist_page[cpu]) 221 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0); 222 223 if (hv_reenlightenment_cb == NULL) 224 return 0; 225 226 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 227 if (re_ctrl.target_vp == hv_vp_index[cpu]) { 228 /* 229 * Reassign reenlightenment notifications to some other online 230 * CPU or just disable the feature if there are no online CPUs 231 * left (happens on hibernation). 232 */ 233 new_cpu = cpumask_any_but(cpu_online_mask, cpu); 234 235 if (new_cpu < nr_cpu_ids) 236 re_ctrl.target_vp = hv_vp_index[new_cpu]; 237 else 238 re_ctrl.enabled = 0; 239 240 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 241 } 242 243 return 0; 244 } 245 246 static int __init hv_pci_init(void) 247 { 248 int gen2vm = efi_enabled(EFI_BOOT); 249 250 /* 251 * For Generation-2 VM, we exit from pci_arch_init() by returning 0. 252 * The purpose is to suppress the harmless warning: 253 * "PCI: Fatal: No config space access function found" 254 */ 255 if (gen2vm) 256 return 0; 257 258 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */ 259 return 1; 260 } 261 262 static int hv_suspend(void) 263 { 264 union hv_x64_msr_hypercall_contents hypercall_msr; 265 int ret; 266 267 /* 268 * Reset the hypercall page as it is going to be invalidated 269 * accross hibernation. Setting hv_hypercall_pg to NULL ensures 270 * that any subsequent hypercall operation fails safely instead of 271 * crashing due to an access of an invalid page. The hypercall page 272 * pointer is restored on resume. 273 */ 274 hv_hypercall_pg_saved = hv_hypercall_pg; 275 hv_hypercall_pg = NULL; 276 277 /* Disable the hypercall page in the hypervisor */ 278 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 279 hypercall_msr.enable = 0; 280 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 281 282 ret = hv_cpu_die(0); 283 return ret; 284 } 285 286 static void hv_resume(void) 287 { 288 union hv_x64_msr_hypercall_contents hypercall_msr; 289 int ret; 290 291 ret = hv_cpu_init(0); 292 WARN_ON(ret); 293 294 /* Re-enable the hypercall page */ 295 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 296 hypercall_msr.enable = 1; 297 hypercall_msr.guest_physical_address = 298 vmalloc_to_pfn(hv_hypercall_pg_saved); 299 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 300 301 hv_hypercall_pg = hv_hypercall_pg_saved; 302 hv_hypercall_pg_saved = NULL; 303 304 /* 305 * Reenlightenment notifications are disabled by hv_cpu_die(0), 306 * reenable them here if hv_reenlightenment_cb was previously set. 307 */ 308 if (hv_reenlightenment_cb) 309 set_hv_tscchange_cb(hv_reenlightenment_cb); 310 } 311 312 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */ 313 static struct syscore_ops hv_syscore_ops = { 314 .suspend = hv_suspend, 315 .resume = hv_resume, 316 }; 317 318 /* 319 * This function is to be invoked early in the boot sequence after the 320 * hypervisor has been detected. 321 * 322 * 1. Setup the hypercall page. 323 * 2. Register Hyper-V specific clocksource. 324 * 3. Setup Hyper-V specific APIC entry points. 325 */ 326 void __init hyperv_init(void) 327 { 328 u64 guest_id, required_msrs; 329 union hv_x64_msr_hypercall_contents hypercall_msr; 330 int cpuhp, i; 331 332 if (x86_hyper_type != X86_HYPER_MS_HYPERV) 333 return; 334 335 /* Absolutely required MSRs */ 336 required_msrs = HV_MSR_HYPERCALL_AVAILABLE | 337 HV_MSR_VP_INDEX_AVAILABLE; 338 339 if ((ms_hyperv.features & required_msrs) != required_msrs) 340 return; 341 342 /* 343 * Allocate the per-CPU state for the hypercall input arg. 344 * If this allocation fails, we will not be able to setup 345 * (per-CPU) hypercall input page and thus this failure is 346 * fatal on Hyper-V. 347 */ 348 hyperv_pcpu_input_arg = alloc_percpu(void *); 349 350 BUG_ON(hyperv_pcpu_input_arg == NULL); 351 352 /* Allocate percpu VP index */ 353 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index), 354 GFP_KERNEL); 355 if (!hv_vp_index) 356 return; 357 358 for (i = 0; i < num_possible_cpus(); i++) 359 hv_vp_index[i] = VP_INVAL; 360 361 hv_vp_assist_page = kcalloc(num_possible_cpus(), 362 sizeof(*hv_vp_assist_page), GFP_KERNEL); 363 if (!hv_vp_assist_page) { 364 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED; 365 goto free_vp_index; 366 } 367 368 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online", 369 hv_cpu_init, hv_cpu_die); 370 if (cpuhp < 0) 371 goto free_vp_assist_page; 372 373 /* 374 * Setup the hypercall page and enable hypercalls. 375 * 1. Register the guest ID 376 * 2. Enable the hypercall and register the hypercall page 377 */ 378 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0); 379 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id); 380 381 hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START, 382 VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX, 383 VM_FLUSH_RESET_PERMS, NUMA_NO_NODE, 384 __builtin_return_address(0)); 385 if (hv_hypercall_pg == NULL) { 386 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 387 goto remove_cpuhp_state; 388 } 389 390 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 391 hypercall_msr.enable = 1; 392 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg); 393 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 394 395 /* 396 * Ignore any errors in setting up stimer clockevents 397 * as we can run with the LAPIC timer as a fallback. 398 */ 399 (void)hv_stimer_alloc(); 400 401 hv_apic_init(); 402 403 x86_init.pci.arch_init = hv_pci_init; 404 405 register_syscore_ops(&hv_syscore_ops); 406 407 hyperv_init_cpuhp = cpuhp; 408 return; 409 410 remove_cpuhp_state: 411 cpuhp_remove_state(cpuhp); 412 free_vp_assist_page: 413 kfree(hv_vp_assist_page); 414 hv_vp_assist_page = NULL; 415 free_vp_index: 416 kfree(hv_vp_index); 417 hv_vp_index = NULL; 418 } 419 420 /* 421 * This routine is called before kexec/kdump, it does the required cleanup. 422 */ 423 void hyperv_cleanup(void) 424 { 425 union hv_x64_msr_hypercall_contents hypercall_msr; 426 427 unregister_syscore_ops(&hv_syscore_ops); 428 429 /* Reset our OS id */ 430 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 431 432 /* 433 * Reset hypercall page reference before reset the page, 434 * let hypercall operations fail safely rather than 435 * panic the kernel for using invalid hypercall page 436 */ 437 hv_hypercall_pg = NULL; 438 439 /* Reset the hypercall page */ 440 hypercall_msr.as_uint64 = 0; 441 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 442 443 /* Reset the TSC page */ 444 hypercall_msr.as_uint64 = 0; 445 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64); 446 } 447 EXPORT_SYMBOL_GPL(hyperv_cleanup); 448 449 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die) 450 { 451 static bool panic_reported; 452 u64 guest_id; 453 454 if (in_die && !panic_on_oops) 455 return; 456 457 /* 458 * We prefer to report panic on 'die' chain as we have proper 459 * registers to report, but if we miss it (e.g. on BUG()) we need 460 * to report it on 'panic'. 461 */ 462 if (panic_reported) 463 return; 464 panic_reported = true; 465 466 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id); 467 468 wrmsrl(HV_X64_MSR_CRASH_P0, err); 469 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id); 470 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip); 471 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax); 472 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp); 473 474 /* 475 * Let Hyper-V know there is crash data available 476 */ 477 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY); 478 } 479 EXPORT_SYMBOL_GPL(hyperv_report_panic); 480 481 /** 482 * hyperv_report_panic_msg - report panic message to Hyper-V 483 * @pa: physical address of the panic page containing the message 484 * @size: size of the message in the page 485 */ 486 void hyperv_report_panic_msg(phys_addr_t pa, size_t size) 487 { 488 /* 489 * P3 to contain the physical address of the panic page & P4 to 490 * contain the size of the panic data in that page. Rest of the 491 * registers are no-op when the NOTIFY_MSG flag is set. 492 */ 493 wrmsrl(HV_X64_MSR_CRASH_P0, 0); 494 wrmsrl(HV_X64_MSR_CRASH_P1, 0); 495 wrmsrl(HV_X64_MSR_CRASH_P2, 0); 496 wrmsrl(HV_X64_MSR_CRASH_P3, pa); 497 wrmsrl(HV_X64_MSR_CRASH_P4, size); 498 499 /* 500 * Let Hyper-V know there is crash data available along with 501 * the panic message. 502 */ 503 wrmsrl(HV_X64_MSR_CRASH_CTL, 504 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG)); 505 } 506 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg); 507 508 bool hv_is_hyperv_initialized(void) 509 { 510 union hv_x64_msr_hypercall_contents hypercall_msr; 511 512 /* 513 * Ensure that we're really on Hyper-V, and not a KVM or Xen 514 * emulation of Hyper-V 515 */ 516 if (x86_hyper_type != X86_HYPER_MS_HYPERV) 517 return false; 518 519 /* 520 * Verify that earlier initialization succeeded by checking 521 * that the hypercall page is setup 522 */ 523 hypercall_msr.as_uint64 = 0; 524 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 525 526 return hypercall_msr.enable; 527 } 528 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized); 529 530 bool hv_is_hibernation_supported(void) 531 { 532 return acpi_sleep_state_supported(ACPI_STATE_S4); 533 } 534 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported); 535