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 38 #include <linux/mm.h> 39 40 static struct start_info _xen_start_info; 41 struct start_info *xen_start_info = &_xen_start_info; 42 EXPORT_SYMBOL(xen_start_info); 43 44 enum xen_domain_type xen_domain_type = XEN_NATIVE; 45 EXPORT_SYMBOL(xen_domain_type); 46 47 struct shared_info xen_dummy_shared_info; 48 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 49 50 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 51 static struct vcpu_info __percpu *xen_vcpu_info; 52 53 /* Linux <-> Xen vCPU id mapping */ 54 DEFINE_PER_CPU(uint32_t, xen_vcpu_id); 55 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id); 56 57 /* These are unused until we support booting "pre-ballooned" */ 58 unsigned long xen_released_pages; 59 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; 60 61 static __read_mostly unsigned int xen_events_irq; 62 static __read_mostly phys_addr_t xen_grant_frames; 63 64 #define GRANT_TABLE_INDEX 0 65 #define EXT_REGION_INDEX 1 66 67 uint32_t xen_start_flags; 68 EXPORT_SYMBOL(xen_start_flags); 69 70 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma, 71 int nr, struct page **pages) 72 { 73 return xen_xlate_unmap_gfn_range(vma, nr, pages); 74 } 75 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range); 76 77 static void xen_read_wallclock(struct timespec64 *ts) 78 { 79 u32 version; 80 struct timespec64 now, ts_monotonic; 81 struct shared_info *s = HYPERVISOR_shared_info; 82 struct pvclock_wall_clock *wall_clock = &(s->wc); 83 84 /* get wallclock at system boot */ 85 do { 86 version = wall_clock->version; 87 rmb(); /* fetch version before time */ 88 now.tv_sec = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec; 89 now.tv_nsec = wall_clock->nsec; 90 rmb(); /* fetch time before checking version */ 91 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 92 93 /* time since system boot */ 94 ktime_get_ts64(&ts_monotonic); 95 *ts = timespec64_add(now, ts_monotonic); 96 } 97 98 static int xen_pvclock_gtod_notify(struct notifier_block *nb, 99 unsigned long was_set, void *priv) 100 { 101 /* Protected by the calling core code serialization */ 102 static struct timespec64 next_sync; 103 104 struct xen_platform_op op; 105 struct timespec64 now, system_time; 106 struct timekeeper *tk = priv; 107 108 now.tv_sec = tk->xtime_sec; 109 now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 110 system_time = timespec64_add(now, tk->wall_to_monotonic); 111 112 /* 113 * We only take the expensive HV call when the clock was set 114 * or when the 11 minutes RTC synchronization time elapsed. 115 */ 116 if (!was_set && timespec64_compare(&now, &next_sync) < 0) 117 return NOTIFY_OK; 118 119 op.cmd = XENPF_settime64; 120 op.u.settime64.mbz = 0; 121 op.u.settime64.secs = now.tv_sec; 122 op.u.settime64.nsecs = now.tv_nsec; 123 op.u.settime64.system_time = timespec64_to_ns(&system_time); 124 (void)HYPERVISOR_platform_op(&op); 125 126 /* 127 * Move the next drift compensation time 11 minutes 128 * ahead. That's emulating the sync_cmos_clock() update for 129 * the hardware RTC. 130 */ 131 next_sync = now; 132 next_sync.tv_sec += 11 * 60; 133 134 return NOTIFY_OK; 135 } 136 137 static struct notifier_block xen_pvclock_gtod_notifier = { 138 .notifier_call = xen_pvclock_gtod_notify, 139 }; 140 141 static int xen_starting_cpu(unsigned int cpu) 142 { 143 struct vcpu_register_vcpu_info info; 144 struct vcpu_info *vcpup; 145 int err; 146 147 /* 148 * VCPUOP_register_vcpu_info cannot be called twice for the same 149 * vcpu, so if vcpu_info is already registered, just get out. This 150 * can happen with cpu-hotplug. 151 */ 152 if (per_cpu(xen_vcpu, cpu) != NULL) 153 goto after_register_vcpu_info; 154 155 pr_info("Xen: initializing cpu%d\n", cpu); 156 vcpup = per_cpu_ptr(xen_vcpu_info, cpu); 157 158 info.mfn = percpu_to_gfn(vcpup); 159 info.offset = xen_offset_in_page(vcpup); 160 161 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu), 162 &info); 163 BUG_ON(err); 164 per_cpu(xen_vcpu, cpu) = vcpup; 165 166 if (!xen_kernel_unmapped_at_usr()) 167 xen_setup_runstate_info(cpu); 168 169 after_register_vcpu_info: 170 enable_percpu_irq(xen_events_irq, 0); 171 return 0; 172 } 173 174 static int xen_dying_cpu(unsigned int cpu) 175 { 176 disable_percpu_irq(xen_events_irq); 177 return 0; 178 } 179 180 void xen_reboot(int reason) 181 { 182 struct sched_shutdown r = { .reason = reason }; 183 int rc; 184 185 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r); 186 BUG_ON(rc); 187 } 188 189 static int xen_restart(struct notifier_block *nb, unsigned long action, 190 void *data) 191 { 192 xen_reboot(SHUTDOWN_reboot); 193 194 return NOTIFY_DONE; 195 } 196 197 static struct notifier_block xen_restart_nb = { 198 .notifier_call = xen_restart, 199 .priority = 192, 200 }; 201 202 static void xen_power_off(void) 203 { 204 xen_reboot(SHUTDOWN_poweroff); 205 } 206 207 static irqreturn_t xen_arm_callback(int irq, void *arg) 208 { 209 xen_hvm_evtchn_do_upcall(); 210 return IRQ_HANDLED; 211 } 212 213 static __initdata struct { 214 const char *compat; 215 const char *prefix; 216 const char *version; 217 bool found; 218 } hyper_node = {"xen,xen", "xen,xen-", NULL, false}; 219 220 static int __init fdt_find_hyper_node(unsigned long node, const char *uname, 221 int depth, void *data) 222 { 223 const void *s = NULL; 224 int len; 225 226 if (depth != 1 || strcmp(uname, "hypervisor") != 0) 227 return 0; 228 229 if (of_flat_dt_is_compatible(node, hyper_node.compat)) 230 hyper_node.found = true; 231 232 s = of_get_flat_dt_prop(node, "compatible", &len); 233 if (strlen(hyper_node.prefix) + 3 < len && 234 !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix))) 235 hyper_node.version = s + strlen(hyper_node.prefix); 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 = kcalloc(nr_reg, sizeof(*regs), GFP_KERNEL); 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(sizeof(*tmp_res), GFP_KERNEL); 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 xen_set_restricted_virtio_memory_access(); 447 448 if (!acpi_disabled) 449 xen_acpi_guest_init(); 450 else 451 xen_dt_guest_init(); 452 453 if (!xen_events_irq) { 454 pr_err("Xen event channel interrupt not found\n"); 455 return -ENODEV; 456 } 457 458 /* 459 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI 460 * parameters are found. Force enable runtime services. 461 */ 462 if (efi_enabled(EFI_RUNTIME_SERVICES)) 463 xen_efi_runtime_setup(); 464 465 shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL); 466 467 if (!shared_info_page) { 468 pr_err("not enough memory\n"); 469 return -ENOMEM; 470 } 471 xatp.domid = DOMID_SELF; 472 xatp.idx = 0; 473 xatp.space = XENMAPSPACE_shared_info; 474 xatp.gpfn = virt_to_gfn(shared_info_page); 475 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 476 BUG(); 477 478 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 479 480 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 481 * page, we use it in the event channel upcall and in some pvclock 482 * related functions. 483 * The shared info contains exactly 1 CPU (the boot CPU). The guest 484 * is required to use VCPUOP_register_vcpu_info to place vcpu info 485 * for secondary CPUs as they are brought up. 486 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0. 487 */ 488 xen_vcpu_info = alloc_percpu(struct vcpu_info); 489 if (xen_vcpu_info == NULL) 490 return -ENOMEM; 491 492 /* Direct vCPU id mapping for ARM guests. */ 493 for_each_possible_cpu(cpu) 494 per_cpu(xen_vcpu_id, cpu) = cpu; 495 496 if (!xen_grant_frames) { 497 xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames(); 498 rc = xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn, 499 &xen_auto_xlat_grant_frames.vaddr, 500 xen_auto_xlat_grant_frames.count); 501 } else 502 rc = gnttab_setup_auto_xlat_frames(xen_grant_frames); 503 if (rc) { 504 free_percpu(xen_vcpu_info); 505 return rc; 506 } 507 gnttab_init(); 508 509 /* 510 * Making sure board specific code will not set up ops for 511 * cpu idle and cpu freq. 512 */ 513 disable_cpuidle(); 514 disable_cpufreq(); 515 516 xen_init_IRQ(); 517 518 if (request_percpu_irq(xen_events_irq, xen_arm_callback, 519 "events", &xen_vcpu)) { 520 pr_err("Error request IRQ %d\n", xen_events_irq); 521 return -EINVAL; 522 } 523 524 if (!xen_kernel_unmapped_at_usr()) 525 xen_time_setup_guest(); 526 527 if (xen_initial_domain()) 528 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier); 529 530 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING, 531 "arm/xen:starting", xen_starting_cpu, 532 xen_dying_cpu); 533 } 534 early_initcall(xen_guest_init); 535 536 static int __init xen_pm_init(void) 537 { 538 if (!xen_domain()) 539 return -ENODEV; 540 541 pm_power_off = xen_power_off; 542 register_restart_handler(&xen_restart_nb); 543 if (!xen_initial_domain()) { 544 struct timespec64 ts; 545 xen_read_wallclock(&ts); 546 do_settimeofday64(&ts); 547 } 548 549 return 0; 550 } 551 late_initcall(xen_pm_init); 552 553 554 /* empty stubs */ 555 void xen_arch_pre_suspend(void) { } 556 void xen_arch_post_suspend(int suspend_cancelled) { } 557 void xen_timer_resume(void) { } 558 void xen_arch_resume(void) { } 559 void xen_arch_suspend(void) { } 560 561 562 /* In the hypercall.S file. */ 563 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op); 564 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op); 565 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version); 566 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io); 567 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op); 568 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op); 569 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op); 570 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op); 571 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op); 572 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op_raw); 573 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall); 574 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist); 575 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op); 576 EXPORT_SYMBOL_GPL(privcmd_call); 577