1 #include <xen/xen.h> 2 #include <xen/events.h> 3 #include <xen/grant_table.h> 4 #include <xen/hvm.h> 5 #include <xen/interface/vcpu.h> 6 #include <xen/interface/xen.h> 7 #include <xen/interface/memory.h> 8 #include <xen/interface/hvm/params.h> 9 #include <xen/features.h> 10 #include <xen/platform_pci.h> 11 #include <xen/xenbus.h> 12 #include <xen/page.h> 13 #include <xen/interface/sched.h> 14 #include <xen/xen-ops.h> 15 #include <asm/xen/hypervisor.h> 16 #include <asm/xen/hypercall.h> 17 #include <asm/xen/xen-ops.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/time64.h> 33 #include <linux/timekeeping.h> 34 #include <linux/timekeeper_internal.h> 35 #include <linux/acpi.h> 36 37 #include <linux/mm.h> 38 39 struct start_info _xen_start_info; 40 struct start_info *xen_start_info = &_xen_start_info; 41 EXPORT_SYMBOL(xen_start_info); 42 43 enum xen_domain_type xen_domain_type = XEN_NATIVE; 44 EXPORT_SYMBOL(xen_domain_type); 45 46 struct shared_info xen_dummy_shared_info; 47 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 48 49 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 50 static struct vcpu_info __percpu *xen_vcpu_info; 51 52 /* Linux <-> Xen vCPU id mapping */ 53 DEFINE_PER_CPU(uint32_t, xen_vcpu_id); 54 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id); 55 56 /* These are unused until we support booting "pre-ballooned" */ 57 unsigned long xen_released_pages; 58 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; 59 60 static __read_mostly unsigned int xen_events_irq; 61 62 int xen_remap_domain_gfn_array(struct vm_area_struct *vma, 63 unsigned long addr, 64 xen_pfn_t *gfn, int nr, 65 int *err_ptr, pgprot_t prot, 66 unsigned domid, 67 struct page **pages) 68 { 69 return xen_xlate_remap_gfn_array(vma, addr, gfn, nr, err_ptr, 70 prot, domid, pages); 71 } 72 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array); 73 74 /* Not used by XENFEAT_auto_translated guests. */ 75 int xen_remap_domain_gfn_range(struct vm_area_struct *vma, 76 unsigned long addr, 77 xen_pfn_t gfn, int nr, 78 pgprot_t prot, unsigned domid, 79 struct page **pages) 80 { 81 return -ENOSYS; 82 } 83 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range); 84 85 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma, 86 int nr, struct page **pages) 87 { 88 return xen_xlate_unmap_gfn_range(vma, nr, pages); 89 } 90 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range); 91 92 /* Not used by XENFEAT_auto_translated guests. */ 93 int xen_remap_domain_mfn_array(struct vm_area_struct *vma, 94 unsigned long addr, 95 xen_pfn_t *mfn, int nr, 96 int *err_ptr, pgprot_t prot, 97 unsigned int domid, struct page **pages) 98 { 99 return -ENOSYS; 100 } 101 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_array); 102 103 static void xen_read_wallclock(struct timespec64 *ts) 104 { 105 u32 version; 106 struct timespec64 now, ts_monotonic; 107 struct shared_info *s = HYPERVISOR_shared_info; 108 struct pvclock_wall_clock *wall_clock = &(s->wc); 109 110 /* get wallclock at system boot */ 111 do { 112 version = wall_clock->version; 113 rmb(); /* fetch version before time */ 114 now.tv_sec = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec; 115 now.tv_nsec = wall_clock->nsec; 116 rmb(); /* fetch time before checking version */ 117 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 118 119 /* time since system boot */ 120 ktime_get_ts64(&ts_monotonic); 121 *ts = timespec64_add(now, ts_monotonic); 122 } 123 124 static int xen_pvclock_gtod_notify(struct notifier_block *nb, 125 unsigned long was_set, void *priv) 126 { 127 /* Protected by the calling core code serialization */ 128 static struct timespec64 next_sync; 129 130 struct xen_platform_op op; 131 struct timespec64 now, system_time; 132 struct timekeeper *tk = priv; 133 134 now.tv_sec = tk->xtime_sec; 135 now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 136 system_time = timespec64_add(now, tk->wall_to_monotonic); 137 138 /* 139 * We only take the expensive HV call when the clock was set 140 * or when the 11 minutes RTC synchronization time elapsed. 141 */ 142 if (!was_set && timespec64_compare(&now, &next_sync) < 0) 143 return NOTIFY_OK; 144 145 op.cmd = XENPF_settime64; 146 op.u.settime64.mbz = 0; 147 op.u.settime64.secs = now.tv_sec; 148 op.u.settime64.nsecs = now.tv_nsec; 149 op.u.settime64.system_time = timespec64_to_ns(&system_time); 150 (void)HYPERVISOR_platform_op(&op); 151 152 /* 153 * Move the next drift compensation time 11 minutes 154 * ahead. That's emulating the sync_cmos_clock() update for 155 * the hardware RTC. 156 */ 157 next_sync = now; 158 next_sync.tv_sec += 11 * 60; 159 160 return NOTIFY_OK; 161 } 162 163 static struct notifier_block xen_pvclock_gtod_notifier = { 164 .notifier_call = xen_pvclock_gtod_notify, 165 }; 166 167 static int xen_starting_cpu(unsigned int cpu) 168 { 169 struct vcpu_register_vcpu_info info; 170 struct vcpu_info *vcpup; 171 int err; 172 173 /* 174 * VCPUOP_register_vcpu_info cannot be called twice for the same 175 * vcpu, so if vcpu_info is already registered, just get out. This 176 * can happen with cpu-hotplug. 177 */ 178 if (per_cpu(xen_vcpu, cpu) != NULL) 179 goto after_register_vcpu_info; 180 181 pr_info("Xen: initializing cpu%d\n", cpu); 182 vcpup = per_cpu_ptr(xen_vcpu_info, cpu); 183 184 info.mfn = virt_to_gfn(vcpup); 185 info.offset = xen_offset_in_page(vcpup); 186 187 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu), 188 &info); 189 BUG_ON(err); 190 per_cpu(xen_vcpu, cpu) = vcpup; 191 192 xen_setup_runstate_info(cpu); 193 194 after_register_vcpu_info: 195 enable_percpu_irq(xen_events_irq, 0); 196 return 0; 197 } 198 199 static int xen_dying_cpu(unsigned int cpu) 200 { 201 disable_percpu_irq(xen_events_irq); 202 return 0; 203 } 204 205 void xen_reboot(int reason) 206 { 207 struct sched_shutdown r = { .reason = reason }; 208 int rc; 209 210 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r); 211 BUG_ON(rc); 212 } 213 214 static void xen_restart(enum reboot_mode reboot_mode, const char *cmd) 215 { 216 xen_reboot(SHUTDOWN_reboot); 217 } 218 219 220 static void xen_power_off(void) 221 { 222 xen_reboot(SHUTDOWN_poweroff); 223 } 224 225 static irqreturn_t xen_arm_callback(int irq, void *arg) 226 { 227 xen_hvm_evtchn_do_upcall(); 228 return IRQ_HANDLED; 229 } 230 231 static __initdata struct { 232 const char *compat; 233 const char *prefix; 234 const char *version; 235 bool found; 236 } hyper_node = {"xen,xen", "xen,xen-", NULL, false}; 237 238 static int __init fdt_find_hyper_node(unsigned long node, const char *uname, 239 int depth, void *data) 240 { 241 const void *s = NULL; 242 int len; 243 244 if (depth != 1 || strcmp(uname, "hypervisor") != 0) 245 return 0; 246 247 if (of_flat_dt_is_compatible(node, hyper_node.compat)) 248 hyper_node.found = true; 249 250 s = of_get_flat_dt_prop(node, "compatible", &len); 251 if (strlen(hyper_node.prefix) + 3 < len && 252 !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix))) 253 hyper_node.version = s + strlen(hyper_node.prefix); 254 255 /* 256 * Check if Xen supports EFI by checking whether there is the 257 * "/hypervisor/uefi" node in DT. If so, runtime services are available 258 * through proxy functions (e.g. in case of Xen dom0 EFI implementation 259 * they call special hypercall which executes relevant EFI functions) 260 * and that is why they are always enabled. 261 */ 262 if (IS_ENABLED(CONFIG_XEN_EFI)) { 263 if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) && 264 !efi_runtime_disabled()) 265 set_bit(EFI_RUNTIME_SERVICES, &efi.flags); 266 } 267 268 return 0; 269 } 270 271 /* 272 * see Documentation/devicetree/bindings/arm/xen.txt for the 273 * documentation of the Xen Device Tree format. 274 */ 275 #define GRANT_TABLE_PHYSADDR 0 276 void __init xen_early_init(void) 277 { 278 of_scan_flat_dt(fdt_find_hyper_node, NULL); 279 if (!hyper_node.found) { 280 pr_debug("No Xen support\n"); 281 return; 282 } 283 284 if (hyper_node.version == NULL) { 285 pr_debug("Xen version not found\n"); 286 return; 287 } 288 289 pr_info("Xen %s support found\n", hyper_node.version); 290 291 xen_domain_type = XEN_HVM_DOMAIN; 292 293 xen_setup_features(); 294 295 if (xen_feature(XENFEAT_dom0)) 296 xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED; 297 else 298 xen_start_info->flags &= ~(SIF_INITDOMAIN|SIF_PRIVILEGED); 299 300 if (!console_set_on_cmdline && !xen_initial_domain()) 301 add_preferred_console("hvc", 0, NULL); 302 } 303 304 static void __init xen_acpi_guest_init(void) 305 { 306 #ifdef CONFIG_ACPI 307 struct xen_hvm_param a; 308 int interrupt, trigger, polarity; 309 310 a.domid = DOMID_SELF; 311 a.index = HVM_PARAM_CALLBACK_IRQ; 312 313 if (HYPERVISOR_hvm_op(HVMOP_get_param, &a) 314 || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) { 315 xen_events_irq = 0; 316 return; 317 } 318 319 interrupt = a.value & 0xff; 320 trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE 321 : ACPI_LEVEL_SENSITIVE; 322 polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW 323 : ACPI_ACTIVE_HIGH; 324 xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity); 325 #endif 326 } 327 328 static void __init xen_dt_guest_init(void) 329 { 330 struct device_node *xen_node; 331 332 xen_node = of_find_compatible_node(NULL, NULL, "xen,xen"); 333 if (!xen_node) { 334 pr_err("Xen support was detected before, but it has disappeared\n"); 335 return; 336 } 337 338 xen_events_irq = irq_of_parse_and_map(xen_node, 0); 339 } 340 341 static int __init xen_guest_init(void) 342 { 343 struct xen_add_to_physmap xatp; 344 struct shared_info *shared_info_page = NULL; 345 int cpu; 346 347 if (!xen_domain()) 348 return 0; 349 350 if (!acpi_disabled) 351 xen_acpi_guest_init(); 352 else 353 xen_dt_guest_init(); 354 355 if (!xen_events_irq) { 356 pr_err("Xen event channel interrupt not found\n"); 357 return -ENODEV; 358 } 359 360 /* 361 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI 362 * parameters are found. Force enable runtime services. 363 */ 364 if (efi_enabled(EFI_RUNTIME_SERVICES)) 365 xen_efi_runtime_setup(); 366 367 shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL); 368 369 if (!shared_info_page) { 370 pr_err("not enough memory\n"); 371 return -ENOMEM; 372 } 373 xatp.domid = DOMID_SELF; 374 xatp.idx = 0; 375 xatp.space = XENMAPSPACE_shared_info; 376 xatp.gpfn = virt_to_gfn(shared_info_page); 377 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 378 BUG(); 379 380 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 381 382 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 383 * page, we use it in the event channel upcall and in some pvclock 384 * related functions. 385 * The shared info contains exactly 1 CPU (the boot CPU). The guest 386 * is required to use VCPUOP_register_vcpu_info to place vcpu info 387 * for secondary CPUs as they are brought up. 388 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0. 389 */ 390 xen_vcpu_info = alloc_percpu(struct vcpu_info); 391 if (xen_vcpu_info == NULL) 392 return -ENOMEM; 393 394 /* Direct vCPU id mapping for ARM guests. */ 395 for_each_possible_cpu(cpu) 396 per_cpu(xen_vcpu_id, cpu) = cpu; 397 398 xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames(); 399 if (xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn, 400 &xen_auto_xlat_grant_frames.vaddr, 401 xen_auto_xlat_grant_frames.count)) { 402 free_percpu(xen_vcpu_info); 403 return -ENOMEM; 404 } 405 gnttab_init(); 406 if (!xen_initial_domain()) 407 xenbus_probe(NULL); 408 409 /* 410 * Making sure board specific code will not set up ops for 411 * cpu idle and cpu freq. 412 */ 413 disable_cpuidle(); 414 disable_cpufreq(); 415 416 xen_init_IRQ(); 417 418 if (request_percpu_irq(xen_events_irq, xen_arm_callback, 419 "events", &xen_vcpu)) { 420 pr_err("Error request IRQ %d\n", xen_events_irq); 421 return -EINVAL; 422 } 423 424 xen_time_setup_guest(); 425 426 if (xen_initial_domain()) 427 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier); 428 429 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING, 430 "arm/xen:starting", xen_starting_cpu, 431 xen_dying_cpu); 432 } 433 early_initcall(xen_guest_init); 434 435 static int __init xen_pm_init(void) 436 { 437 if (!xen_domain()) 438 return -ENODEV; 439 440 pm_power_off = xen_power_off; 441 arm_pm_restart = xen_restart; 442 if (!xen_initial_domain()) { 443 struct timespec64 ts; 444 xen_read_wallclock(&ts); 445 do_settimeofday64(&ts); 446 } 447 448 return 0; 449 } 450 late_initcall(xen_pm_init); 451 452 453 /* empty stubs */ 454 void xen_arch_pre_suspend(void) { } 455 void xen_arch_post_suspend(int suspend_cancelled) { } 456 void xen_timer_resume(void) { } 457 void xen_arch_resume(void) { } 458 void xen_arch_suspend(void) { } 459 460 461 /* In the hypercall.S file. */ 462 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op); 463 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op); 464 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version); 465 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io); 466 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op); 467 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op); 468 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op); 469 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op); 470 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op); 471 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op); 472 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op); 473 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall); 474 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist); 475 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op); 476 EXPORT_SYMBOL_GPL(privcmd_call); 477