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 <linux/version.h> 19 #include <linux/vmalloc.h> 20 #include <linux/mm.h> 21 #include <linux/hyperv.h> 22 #include <linux/slab.h> 23 #include <linux/kernel.h> 24 #include <linux/cpuhotplug.h> 25 #include <linux/syscore_ops.h> 26 #include <clocksource/hyperv_timer.h> 27 28 void *hv_hypercall_pg; 29 EXPORT_SYMBOL_GPL(hv_hypercall_pg); 30 31 /* Storage to save the hypercall page temporarily for hibernation */ 32 static void *hv_hypercall_pg_saved; 33 34 u32 *hv_vp_index; 35 EXPORT_SYMBOL_GPL(hv_vp_index); 36 37 struct hv_vp_assist_page **hv_vp_assist_page; 38 EXPORT_SYMBOL_GPL(hv_vp_assist_page); 39 40 void __percpu **hyperv_pcpu_input_arg; 41 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg); 42 43 u32 hv_max_vp_index; 44 EXPORT_SYMBOL_GPL(hv_max_vp_index); 45 46 void *hv_alloc_hyperv_page(void) 47 { 48 BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE); 49 50 return (void *)__get_free_page(GFP_KERNEL); 51 } 52 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_page); 53 54 void *hv_alloc_hyperv_zeroed_page(void) 55 { 56 BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE); 57 58 return (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO); 59 } 60 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_zeroed_page); 61 62 void hv_free_hyperv_page(unsigned long addr) 63 { 64 free_page(addr); 65 } 66 EXPORT_SYMBOL_GPL(hv_free_hyperv_page); 67 68 static int hv_cpu_init(unsigned int cpu) 69 { 70 u64 msr_vp_index; 71 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()]; 72 void **input_arg; 73 struct page *pg; 74 75 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg); 76 /* hv_cpu_init() can be called with IRQs disabled from hv_resume() */ 77 pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL); 78 if (unlikely(!pg)) 79 return -ENOMEM; 80 *input_arg = page_address(pg); 81 82 hv_get_vp_index(msr_vp_index); 83 84 hv_vp_index[smp_processor_id()] = msr_vp_index; 85 86 if (msr_vp_index > hv_max_vp_index) 87 hv_max_vp_index = msr_vp_index; 88 89 if (!hv_vp_assist_page) 90 return 0; 91 92 /* 93 * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section 94 * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure 95 * we always write the EOI MSR in hv_apic_eoi_write() *after* the 96 * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may 97 * not be stopped in the case of CPU offlining and the VM will hang. 98 */ 99 if (!*hvp) { 100 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO, 101 PAGE_KERNEL); 102 } 103 104 if (*hvp) { 105 u64 val; 106 107 val = vmalloc_to_pfn(*hvp); 108 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) | 109 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE; 110 111 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val); 112 } 113 114 return 0; 115 } 116 117 static void (*hv_reenlightenment_cb)(void); 118 119 static void hv_reenlightenment_notify(struct work_struct *dummy) 120 { 121 struct hv_tsc_emulation_status emu_status; 122 123 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 124 125 /* Don't issue the callback if TSC accesses are not emulated */ 126 if (hv_reenlightenment_cb && emu_status.inprogress) 127 hv_reenlightenment_cb(); 128 } 129 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify); 130 131 void hyperv_stop_tsc_emulation(void) 132 { 133 u64 freq; 134 struct hv_tsc_emulation_status emu_status; 135 136 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 137 emu_status.inprogress = 0; 138 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status); 139 140 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq); 141 tsc_khz = div64_u64(freq, 1000); 142 } 143 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation); 144 145 static inline bool hv_reenlightenment_available(void) 146 { 147 /* 148 * Check for required features and priviliges to make TSC frequency 149 * change notifications work. 150 */ 151 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS && 152 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE && 153 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT; 154 } 155 156 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs) 157 { 158 entering_ack_irq(); 159 160 inc_irq_stat(irq_hv_reenlightenment_count); 161 162 schedule_delayed_work(&hv_reenlightenment_work, HZ/10); 163 164 exiting_irq(); 165 } 166 167 void set_hv_tscchange_cb(void (*cb)(void)) 168 { 169 struct hv_reenlightenment_control re_ctrl = { 170 .vector = HYPERV_REENLIGHTENMENT_VECTOR, 171 .enabled = 1, 172 .target_vp = hv_vp_index[smp_processor_id()] 173 }; 174 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1}; 175 176 if (!hv_reenlightenment_available()) { 177 pr_warn("Hyper-V: reenlightenment support is unavailable\n"); 178 return; 179 } 180 181 hv_reenlightenment_cb = cb; 182 183 /* Make sure callback is registered before we write to MSRs */ 184 wmb(); 185 186 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 187 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl)); 188 } 189 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb); 190 191 void clear_hv_tscchange_cb(void) 192 { 193 struct hv_reenlightenment_control re_ctrl; 194 195 if (!hv_reenlightenment_available()) 196 return; 197 198 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl); 199 re_ctrl.enabled = 0; 200 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl); 201 202 hv_reenlightenment_cb = NULL; 203 } 204 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb); 205 206 static int hv_cpu_die(unsigned int cpu) 207 { 208 struct hv_reenlightenment_control re_ctrl; 209 unsigned int new_cpu; 210 unsigned long flags; 211 void **input_arg; 212 void *input_pg = NULL; 213 214 local_irq_save(flags); 215 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg); 216 input_pg = *input_arg; 217 *input_arg = NULL; 218 local_irq_restore(flags); 219 free_page((unsigned long)input_pg); 220 221 if (hv_vp_assist_page && hv_vp_assist_page[cpu]) 222 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0); 223 224 if (hv_reenlightenment_cb == NULL) 225 return 0; 226 227 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 228 if (re_ctrl.target_vp == hv_vp_index[cpu]) { 229 /* 230 * Reassign reenlightenment notifications to some other online 231 * CPU or just disable the feature if there are no online CPUs 232 * left (happens on hibernation). 233 */ 234 new_cpu = cpumask_any_but(cpu_online_mask, cpu); 235 236 if (new_cpu < nr_cpu_ids) 237 re_ctrl.target_vp = hv_vp_index[new_cpu]; 238 else 239 re_ctrl.enabled = 0; 240 241 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl)); 242 } 243 244 return 0; 245 } 246 247 static int __init hv_pci_init(void) 248 { 249 int gen2vm = efi_enabled(EFI_BOOT); 250 251 /* 252 * For Generation-2 VM, we exit from pci_arch_init() by returning 0. 253 * The purpose is to suppress the harmless warning: 254 * "PCI: Fatal: No config space access function found" 255 */ 256 if (gen2vm) 257 return 0; 258 259 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */ 260 return 1; 261 } 262 263 static int hv_suspend(void) 264 { 265 union hv_x64_msr_hypercall_contents hypercall_msr; 266 int ret; 267 268 /* 269 * Reset the hypercall page as it is going to be invalidated 270 * accross hibernation. Setting hv_hypercall_pg to NULL ensures 271 * that any subsequent hypercall operation fails safely instead of 272 * crashing due to an access of an invalid page. The hypercall page 273 * pointer is restored on resume. 274 */ 275 hv_hypercall_pg_saved = hv_hypercall_pg; 276 hv_hypercall_pg = NULL; 277 278 /* Disable the hypercall page in the hypervisor */ 279 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 280 hypercall_msr.enable = 0; 281 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 282 283 ret = hv_cpu_die(0); 284 return ret; 285 } 286 287 static void hv_resume(void) 288 { 289 union hv_x64_msr_hypercall_contents hypercall_msr; 290 int ret; 291 292 ret = hv_cpu_init(0); 293 WARN_ON(ret); 294 295 /* Re-enable the hypercall page */ 296 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 297 hypercall_msr.enable = 1; 298 hypercall_msr.guest_physical_address = 299 vmalloc_to_pfn(hv_hypercall_pg_saved); 300 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 301 302 hv_hypercall_pg = hv_hypercall_pg_saved; 303 hv_hypercall_pg_saved = NULL; 304 305 /* 306 * Reenlightenment notifications are disabled by hv_cpu_die(0), 307 * reenable them here if hv_reenlightenment_cb was previously set. 308 */ 309 if (hv_reenlightenment_cb) 310 set_hv_tscchange_cb(hv_reenlightenment_cb); 311 } 312 313 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */ 314 static struct syscore_ops hv_syscore_ops = { 315 .suspend = hv_suspend, 316 .resume = hv_resume, 317 }; 318 319 /* 320 * This function is to be invoked early in the boot sequence after the 321 * hypervisor has been detected. 322 * 323 * 1. Setup the hypercall page. 324 * 2. Register Hyper-V specific clocksource. 325 * 3. Setup Hyper-V specific APIC entry points. 326 */ 327 void __init hyperv_init(void) 328 { 329 u64 guest_id, required_msrs; 330 union hv_x64_msr_hypercall_contents hypercall_msr; 331 int cpuhp, i; 332 333 if (x86_hyper_type != X86_HYPER_MS_HYPERV) 334 return; 335 336 /* Absolutely required MSRs */ 337 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE | 338 HV_X64_MSR_VP_INDEX_AVAILABLE; 339 340 if ((ms_hyperv.features & required_msrs) != required_msrs) 341 return; 342 343 /* 344 * Allocate the per-CPU state for the hypercall input arg. 345 * If this allocation fails, we will not be able to setup 346 * (per-CPU) hypercall input page and thus this failure is 347 * fatal on Hyper-V. 348 */ 349 hyperv_pcpu_input_arg = alloc_percpu(void *); 350 351 BUG_ON(hyperv_pcpu_input_arg == NULL); 352 353 /* Allocate percpu VP index */ 354 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index), 355 GFP_KERNEL); 356 if (!hv_vp_index) 357 return; 358 359 for (i = 0; i < num_possible_cpus(); i++) 360 hv_vp_index[i] = VP_INVAL; 361 362 hv_vp_assist_page = kcalloc(num_possible_cpus(), 363 sizeof(*hv_vp_assist_page), GFP_KERNEL); 364 if (!hv_vp_assist_page) { 365 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED; 366 goto free_vp_index; 367 } 368 369 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online", 370 hv_cpu_init, hv_cpu_die); 371 if (cpuhp < 0) 372 goto free_vp_assist_page; 373 374 /* 375 * Setup the hypercall page and enable hypercalls. 376 * 1. Register the guest ID 377 * 2. Enable the hypercall and register the hypercall page 378 */ 379 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0); 380 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id); 381 382 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX); 383 if (hv_hypercall_pg == NULL) { 384 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 385 goto remove_cpuhp_state; 386 } 387 388 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 389 hypercall_msr.enable = 1; 390 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg); 391 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 392 393 /* 394 * Ignore any errors in setting up stimer clockevents 395 * as we can run with the LAPIC timer as a fallback. 396 */ 397 (void)hv_stimer_alloc(); 398 399 hv_apic_init(); 400 401 x86_init.pci.arch_init = hv_pci_init; 402 403 register_syscore_ops(&hv_syscore_ops); 404 405 return; 406 407 remove_cpuhp_state: 408 cpuhp_remove_state(cpuhp); 409 free_vp_assist_page: 410 kfree(hv_vp_assist_page); 411 hv_vp_assist_page = NULL; 412 free_vp_index: 413 kfree(hv_vp_index); 414 hv_vp_index = NULL; 415 } 416 417 /* 418 * This routine is called before kexec/kdump, it does the required cleanup. 419 */ 420 void hyperv_cleanup(void) 421 { 422 union hv_x64_msr_hypercall_contents hypercall_msr; 423 424 unregister_syscore_ops(&hv_syscore_ops); 425 426 /* Reset our OS id */ 427 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 428 429 /* 430 * Reset hypercall page reference before reset the page, 431 * let hypercall operations fail safely rather than 432 * panic the kernel for using invalid hypercall page 433 */ 434 hv_hypercall_pg = NULL; 435 436 /* Reset the hypercall page */ 437 hypercall_msr.as_uint64 = 0; 438 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 439 440 /* Reset the TSC page */ 441 hypercall_msr.as_uint64 = 0; 442 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64); 443 } 444 EXPORT_SYMBOL_GPL(hyperv_cleanup); 445 446 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die) 447 { 448 static bool panic_reported; 449 u64 guest_id; 450 451 if (in_die && !panic_on_oops) 452 return; 453 454 /* 455 * We prefer to report panic on 'die' chain as we have proper 456 * registers to report, but if we miss it (e.g. on BUG()) we need 457 * to report it on 'panic'. 458 */ 459 if (panic_reported) 460 return; 461 panic_reported = true; 462 463 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id); 464 465 wrmsrl(HV_X64_MSR_CRASH_P0, err); 466 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id); 467 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip); 468 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax); 469 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp); 470 471 /* 472 * Let Hyper-V know there is crash data available 473 */ 474 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY); 475 } 476 EXPORT_SYMBOL_GPL(hyperv_report_panic); 477 478 /** 479 * hyperv_report_panic_msg - report panic message to Hyper-V 480 * @pa: physical address of the panic page containing the message 481 * @size: size of the message in the page 482 */ 483 void hyperv_report_panic_msg(phys_addr_t pa, size_t size) 484 { 485 /* 486 * P3 to contain the physical address of the panic page & P4 to 487 * contain the size of the panic data in that page. Rest of the 488 * registers are no-op when the NOTIFY_MSG flag is set. 489 */ 490 wrmsrl(HV_X64_MSR_CRASH_P0, 0); 491 wrmsrl(HV_X64_MSR_CRASH_P1, 0); 492 wrmsrl(HV_X64_MSR_CRASH_P2, 0); 493 wrmsrl(HV_X64_MSR_CRASH_P3, pa); 494 wrmsrl(HV_X64_MSR_CRASH_P4, size); 495 496 /* 497 * Let Hyper-V know there is crash data available along with 498 * the panic message. 499 */ 500 wrmsrl(HV_X64_MSR_CRASH_CTL, 501 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG)); 502 } 503 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg); 504 505 bool hv_is_hyperv_initialized(void) 506 { 507 union hv_x64_msr_hypercall_contents hypercall_msr; 508 509 /* 510 * Ensure that we're really on Hyper-V, and not a KVM or Xen 511 * emulation of Hyper-V 512 */ 513 if (x86_hyper_type != X86_HYPER_MS_HYPERV) 514 return false; 515 516 /* 517 * Verify that earlier initialization succeeded by checking 518 * that the hypercall page is setup 519 */ 520 hypercall_msr.as_uint64 = 0; 521 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 522 523 return hypercall_msr.enable; 524 } 525 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized); 526 527 bool hv_is_hibernation_supported(void) 528 { 529 return acpi_sleep_state_supported(ACPI_STATE_S4); 530 } 531 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported); 532