1 // SPDX-License-Identifier: GPL-2.0-only 2 #include <xen/xen.h> 3 #include <xen/events.h> 4 #include <xen/grant_table.h> 5 #include <xen/hvm.h> 6 #include <xen/interface/vcpu.h> 7 #include <xen/interface/xen.h> 8 #include <xen/interface/memory.h> 9 #include <xen/interface/hvm/params.h> 10 #include <xen/features.h> 11 #include <xen/platform_pci.h> 12 #include <xen/xenbus.h> 13 #include <xen/page.h> 14 #include <xen/interface/sched.h> 15 #include <xen/xen-ops.h> 16 #include <asm/xen/hypervisor.h> 17 #include <asm/xen/hypercall.h> 18 #include <asm/system_misc.h> 19 #include <asm/efi.h> 20 #include <linux/interrupt.h> 21 #include <linux/irqreturn.h> 22 #include <linux/module.h> 23 #include <linux/of.h> 24 #include <linux/of_fdt.h> 25 #include <linux/of_irq.h> 26 #include <linux/of_address.h> 27 #include <linux/cpuidle.h> 28 #include <linux/cpufreq.h> 29 #include <linux/cpu.h> 30 #include <linux/console.h> 31 #include <linux/pvclock_gtod.h> 32 #include <linux/reboot.h> 33 #include <linux/time64.h> 34 #include <linux/timekeeping.h> 35 #include <linux/timekeeper_internal.h> 36 #include <linux/acpi.h> 37 #include <linux/virtio_anchor.h> 38 39 #include <linux/mm.h> 40 41 static struct start_info _xen_start_info; 42 struct start_info *xen_start_info = &_xen_start_info; 43 EXPORT_SYMBOL(xen_start_info); 44 45 enum xen_domain_type xen_domain_type = XEN_NATIVE; 46 EXPORT_SYMBOL(xen_domain_type); 47 48 struct shared_info xen_dummy_shared_info; 49 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 50 51 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 52 static struct vcpu_info __percpu *xen_vcpu_info; 53 54 /* Linux <-> Xen vCPU id mapping */ 55 DEFINE_PER_CPU(uint32_t, xen_vcpu_id); 56 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id); 57 58 /* These are unused until we support booting "pre-ballooned" */ 59 unsigned long xen_released_pages; 60 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; 61 62 static __read_mostly unsigned int xen_events_irq; 63 static __read_mostly phys_addr_t xen_grant_frames; 64 65 #define GRANT_TABLE_INDEX 0 66 #define EXT_REGION_INDEX 1 67 68 uint32_t xen_start_flags; 69 EXPORT_SYMBOL(xen_start_flags); 70 71 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma, 72 int nr, struct page **pages) 73 { 74 return xen_xlate_unmap_gfn_range(vma, nr, pages); 75 } 76 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range); 77 78 static void xen_read_wallclock(struct timespec64 *ts) 79 { 80 u32 version; 81 struct timespec64 now, ts_monotonic; 82 struct shared_info *s = HYPERVISOR_shared_info; 83 struct pvclock_wall_clock *wall_clock = &(s->wc); 84 85 /* get wallclock at system boot */ 86 do { 87 version = wall_clock->version; 88 rmb(); /* fetch version before time */ 89 now.tv_sec = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec; 90 now.tv_nsec = wall_clock->nsec; 91 rmb(); /* fetch time before checking version */ 92 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 93 94 /* time since system boot */ 95 ktime_get_ts64(&ts_monotonic); 96 *ts = timespec64_add(now, ts_monotonic); 97 } 98 99 static int xen_pvclock_gtod_notify(struct notifier_block *nb, 100 unsigned long was_set, void *priv) 101 { 102 /* Protected by the calling core code serialization */ 103 static struct timespec64 next_sync; 104 105 struct xen_platform_op op; 106 struct timespec64 now, system_time; 107 struct timekeeper *tk = priv; 108 109 now.tv_sec = tk->xtime_sec; 110 now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 111 system_time = timespec64_add(now, tk->wall_to_monotonic); 112 113 /* 114 * We only take the expensive HV call when the clock was set 115 * or when the 11 minutes RTC synchronization time elapsed. 116 */ 117 if (!was_set && timespec64_compare(&now, &next_sync) < 0) 118 return NOTIFY_OK; 119 120 op.cmd = XENPF_settime64; 121 op.u.settime64.mbz = 0; 122 op.u.settime64.secs = now.tv_sec; 123 op.u.settime64.nsecs = now.tv_nsec; 124 op.u.settime64.system_time = timespec64_to_ns(&system_time); 125 (void)HYPERVISOR_platform_op(&op); 126 127 /* 128 * Move the next drift compensation time 11 minutes 129 * ahead. That's emulating the sync_cmos_clock() update for 130 * the hardware RTC. 131 */ 132 next_sync = now; 133 next_sync.tv_sec += 11 * 60; 134 135 return NOTIFY_OK; 136 } 137 138 static struct notifier_block xen_pvclock_gtod_notifier = { 139 .notifier_call = xen_pvclock_gtod_notify, 140 }; 141 142 static int xen_starting_cpu(unsigned int cpu) 143 { 144 struct vcpu_register_vcpu_info info; 145 struct vcpu_info *vcpup; 146 int err; 147 148 /* 149 * VCPUOP_register_vcpu_info cannot be called twice for the same 150 * vcpu, so if vcpu_info is already registered, just get out. This 151 * can happen with cpu-hotplug. 152 */ 153 if (per_cpu(xen_vcpu, cpu) != NULL) 154 goto after_register_vcpu_info; 155 156 pr_info("Xen: initializing cpu%d\n", cpu); 157 vcpup = per_cpu_ptr(xen_vcpu_info, cpu); 158 159 info.mfn = percpu_to_gfn(vcpup); 160 info.offset = xen_offset_in_page(vcpup); 161 162 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu), 163 &info); 164 BUG_ON(err); 165 per_cpu(xen_vcpu, cpu) = vcpup; 166 167 after_register_vcpu_info: 168 enable_percpu_irq(xen_events_irq, 0); 169 return 0; 170 } 171 172 static int xen_dying_cpu(unsigned int cpu) 173 { 174 disable_percpu_irq(xen_events_irq); 175 return 0; 176 } 177 178 void xen_reboot(int reason) 179 { 180 struct sched_shutdown r = { .reason = reason }; 181 int rc; 182 183 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r); 184 BUG_ON(rc); 185 } 186 187 static int xen_restart(struct notifier_block *nb, unsigned long action, 188 void *data) 189 { 190 xen_reboot(SHUTDOWN_reboot); 191 192 return NOTIFY_DONE; 193 } 194 195 static struct notifier_block xen_restart_nb = { 196 .notifier_call = xen_restart, 197 .priority = 192, 198 }; 199 200 static void xen_power_off(void) 201 { 202 xen_reboot(SHUTDOWN_poweroff); 203 } 204 205 static irqreturn_t xen_arm_callback(int irq, void *arg) 206 { 207 xen_evtchn_do_upcall(); 208 return IRQ_HANDLED; 209 } 210 211 static __initdata struct { 212 const char *compat; 213 const char *prefix; 214 const char *version; 215 bool found; 216 } hyper_node = {"xen,xen", "xen,xen-", NULL, false}; 217 218 static int __init fdt_find_hyper_node(unsigned long node, const char *uname, 219 int depth, void *data) 220 { 221 const char *s = NULL; 222 int len; 223 size_t prefix_len = strlen(hyper_node.prefix); 224 225 if (depth != 1 || strcmp(uname, "hypervisor") != 0) 226 return 0; 227 228 if (of_flat_dt_is_compatible(node, hyper_node.compat)) 229 hyper_node.found = true; 230 231 s = of_get_flat_dt_prop(node, "compatible", &len); 232 if (s && len > 0 && strnlen(s, len) < len && 233 len > prefix_len + 3 && 234 !strncmp(hyper_node.prefix, s, prefix_len)) 235 hyper_node.version = s + prefix_len; 236 237 /* 238 * Check if Xen supports EFI by checking whether there is the 239 * "/hypervisor/uefi" node in DT. If so, runtime services are available 240 * through proxy functions (e.g. in case of Xen dom0 EFI implementation 241 * they call special hypercall which executes relevant EFI functions) 242 * and that is why they are always enabled. 243 */ 244 if (IS_ENABLED(CONFIG_XEN_EFI)) { 245 if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) && 246 !efi_runtime_disabled()) 247 set_bit(EFI_RUNTIME_SERVICES, &efi.flags); 248 } 249 250 return 0; 251 } 252 253 /* 254 * see Documentation/devicetree/bindings/arm/xen.txt for the 255 * documentation of the Xen Device Tree format. 256 */ 257 void __init xen_early_init(void) 258 { 259 of_scan_flat_dt(fdt_find_hyper_node, NULL); 260 if (!hyper_node.found) { 261 pr_debug("No Xen support\n"); 262 return; 263 } 264 265 if (hyper_node.version == NULL) { 266 pr_debug("Xen version not found\n"); 267 return; 268 } 269 270 pr_info("Xen %s support found\n", hyper_node.version); 271 272 xen_domain_type = XEN_HVM_DOMAIN; 273 274 xen_setup_features(); 275 276 if (xen_feature(XENFEAT_dom0)) 277 xen_start_flags |= SIF_INITDOMAIN|SIF_PRIVILEGED; 278 279 if (!console_set_on_cmdline && !xen_initial_domain()) 280 add_preferred_console("hvc", 0, NULL); 281 } 282 283 static void __init xen_acpi_guest_init(void) 284 { 285 #ifdef CONFIG_ACPI 286 struct xen_hvm_param a; 287 int interrupt, trigger, polarity; 288 289 a.domid = DOMID_SELF; 290 a.index = HVM_PARAM_CALLBACK_IRQ; 291 292 if (HYPERVISOR_hvm_op(HVMOP_get_param, &a) 293 || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) { 294 xen_events_irq = 0; 295 return; 296 } 297 298 interrupt = a.value & 0xff; 299 trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE 300 : ACPI_LEVEL_SENSITIVE; 301 polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW 302 : ACPI_ACTIVE_HIGH; 303 xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity); 304 #endif 305 } 306 307 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC 308 /* 309 * A type-less specific Xen resource which contains extended regions 310 * (unused regions of guest physical address space provided by the hypervisor). 311 */ 312 static struct resource xen_resource = { 313 .name = "Xen unused space", 314 }; 315 316 int __init arch_xen_unpopulated_init(struct resource **res) 317 { 318 struct device_node *np; 319 struct resource *regs, *tmp_res; 320 uint64_t min_gpaddr = -1, max_gpaddr = 0; 321 unsigned int i, nr_reg = 0; 322 int rc; 323 324 if (!xen_domain()) 325 return -ENODEV; 326 327 if (!acpi_disabled) 328 return -ENODEV; 329 330 np = of_find_compatible_node(NULL, NULL, "xen,xen"); 331 if (WARN_ON(!np)) 332 return -ENODEV; 333 334 /* Skip region 0 which is reserved for grant table space */ 335 while (of_get_address(np, nr_reg + EXT_REGION_INDEX, NULL, NULL)) 336 nr_reg++; 337 338 if (!nr_reg) { 339 pr_err("No extended regions are found\n"); 340 of_node_put(np); 341 return -EINVAL; 342 } 343 344 regs = kzalloc_objs(*regs, nr_reg); 345 if (!regs) { 346 of_node_put(np); 347 return -ENOMEM; 348 } 349 350 /* 351 * Create resource from extended regions provided by the hypervisor to be 352 * used as unused address space for Xen scratch pages. 353 */ 354 for (i = 0; i < nr_reg; i++) { 355 rc = of_address_to_resource(np, i + EXT_REGION_INDEX, ®s[i]); 356 if (rc) 357 goto err; 358 359 if (max_gpaddr < regs[i].end) 360 max_gpaddr = regs[i].end; 361 if (min_gpaddr > regs[i].start) 362 min_gpaddr = regs[i].start; 363 } 364 365 xen_resource.start = min_gpaddr; 366 xen_resource.end = max_gpaddr; 367 368 /* 369 * Mark holes between extended regions as unavailable. The rest of that 370 * address space will be available for the allocation. 371 */ 372 for (i = 1; i < nr_reg; i++) { 373 resource_size_t start, end; 374 375 /* There is an overlap between regions */ 376 if (regs[i - 1].end + 1 > regs[i].start) { 377 rc = -EINVAL; 378 goto err; 379 } 380 381 /* There is no hole between regions */ 382 if (regs[i - 1].end + 1 == regs[i].start) 383 continue; 384 385 start = regs[i - 1].end + 1; 386 end = regs[i].start - 1; 387 388 tmp_res = kzalloc_obj(*tmp_res); 389 if (!tmp_res) { 390 rc = -ENOMEM; 391 goto err; 392 } 393 394 tmp_res->name = "Unavailable space"; 395 tmp_res->start = start; 396 tmp_res->end = end; 397 398 rc = insert_resource(&xen_resource, tmp_res); 399 if (rc) { 400 pr_err("Cannot insert resource %pR (%d)\n", tmp_res, rc); 401 kfree(tmp_res); 402 goto err; 403 } 404 } 405 406 *res = &xen_resource; 407 408 err: 409 of_node_put(np); 410 kfree(regs); 411 return rc; 412 } 413 #endif 414 415 static void __init xen_dt_guest_init(void) 416 { 417 struct device_node *xen_node; 418 struct resource res; 419 420 xen_node = of_find_compatible_node(NULL, NULL, "xen,xen"); 421 if (!xen_node) { 422 pr_err("Xen support was detected before, but it has disappeared\n"); 423 return; 424 } 425 426 xen_events_irq = irq_of_parse_and_map(xen_node, 0); 427 428 if (of_address_to_resource(xen_node, GRANT_TABLE_INDEX, &res)) { 429 pr_err("Xen grant table region is not found\n"); 430 of_node_put(xen_node); 431 return; 432 } 433 of_node_put(xen_node); 434 xen_grant_frames = res.start; 435 } 436 437 static int __init xen_guest_init(void) 438 { 439 struct xen_add_to_physmap xatp; 440 struct shared_info *shared_info_page = NULL; 441 int rc, cpu; 442 443 if (!xen_domain()) 444 return 0; 445 446 if (IS_ENABLED(CONFIG_XEN_VIRTIO)) 447 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc); 448 449 if (!acpi_disabled) 450 xen_acpi_guest_init(); 451 else 452 xen_dt_guest_init(); 453 454 if (!xen_events_irq) { 455 pr_err("Xen event channel interrupt not found\n"); 456 return -ENODEV; 457 } 458 459 /* 460 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI 461 * parameters are found. Force enable runtime services. 462 */ 463 if (efi_enabled(EFI_RUNTIME_SERVICES)) 464 xen_efi_runtime_setup(); 465 466 shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL); 467 468 if (!shared_info_page) { 469 pr_err("not enough memory\n"); 470 return -ENOMEM; 471 } 472 xatp.domid = DOMID_SELF; 473 xatp.idx = 0; 474 xatp.space = XENMAPSPACE_shared_info; 475 xatp.gpfn = virt_to_gfn(shared_info_page); 476 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 477 BUG(); 478 479 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 480 481 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 482 * page, we use it in the event channel upcall and in some pvclock 483 * related functions. 484 * The shared info contains exactly 1 CPU (the boot CPU). The guest 485 * is required to use VCPUOP_register_vcpu_info to place vcpu info 486 * for secondary CPUs as they are brought up. 487 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0. 488 */ 489 xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info), 490 1 << fls(sizeof(struct vcpu_info) - 1)); 491 if (xen_vcpu_info == NULL) 492 return -ENOMEM; 493 494 /* Direct vCPU id mapping for ARM guests. */ 495 for_each_possible_cpu(cpu) 496 per_cpu(xen_vcpu_id, cpu) = cpu; 497 498 if (!xen_grant_frames) { 499 xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames(); 500 rc = xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn, 501 &xen_auto_xlat_grant_frames.vaddr, 502 xen_auto_xlat_grant_frames.count); 503 } else 504 rc = gnttab_setup_auto_xlat_frames(xen_grant_frames); 505 if (rc) { 506 free_percpu(xen_vcpu_info); 507 return rc; 508 } 509 gnttab_init(); 510 511 /* 512 * Making sure board specific code will not set up ops for 513 * cpu idle and cpu freq. 514 */ 515 disable_cpuidle(); 516 disable_cpufreq(); 517 518 xen_init_IRQ(); 519 520 if (request_percpu_irq(xen_events_irq, xen_arm_callback, 521 "events", &xen_vcpu)) { 522 pr_err("Error request IRQ %d\n", xen_events_irq); 523 return -EINVAL; 524 } 525 526 if (xen_initial_domain()) 527 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier); 528 529 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING, 530 "arm/xen:starting", xen_starting_cpu, 531 xen_dying_cpu); 532 } 533 early_initcall(xen_guest_init); 534 535 static int xen_starting_runstate_cpu(unsigned int cpu) 536 { 537 xen_setup_runstate_info(cpu); 538 return 0; 539 } 540 541 static int __init xen_late_init(void) 542 { 543 if (!xen_domain()) 544 return -ENODEV; 545 546 register_platform_power_off(xen_power_off); 547 register_restart_handler(&xen_restart_nb); 548 if (!xen_initial_domain()) { 549 struct timespec64 ts; 550 xen_read_wallclock(&ts); 551 do_settimeofday64(&ts); 552 } 553 554 if (xen_kernel_unmapped_at_usr()) 555 return 0; 556 557 xen_time_setup_guest(); 558 559 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_RUNSTATE_STARTING, 560 "arm/xen_runstate:starting", 561 xen_starting_runstate_cpu, NULL); 562 } 563 late_initcall(xen_late_init); 564 565 566 /* empty stubs */ 567 void xen_arch_pre_suspend(void) { } 568 void xen_arch_post_suspend(int suspend_cancelled) { } 569 void xen_timer_resume(void) { } 570 void xen_arch_resume(void) { } 571 void xen_arch_suspend(void) { } 572 573 574 /* In the hypercall.S file. */ 575 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op); 576 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op); 577 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version); 578 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io); 579 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op); 580 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op); 581 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op); 582 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op); 583 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op); 584 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op_raw); 585 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall); 586 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist); 587 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op); 588 EXPORT_SYMBOL_GPL(privcmd_call); 589