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/paravirt.h> 16 #include <asm/xen/hypervisor.h> 17 #include <asm/xen/hypercall.h> 18 #include <asm/system_misc.h> 19 #include <linux/interrupt.h> 20 #include <linux/irqreturn.h> 21 #include <linux/module.h> 22 #include <linux/of.h> 23 #include <linux/of_irq.h> 24 #include <linux/of_address.h> 25 #include <linux/cpuidle.h> 26 #include <linux/cpufreq.h> 27 #include <linux/cpu.h> 28 #include <linux/console.h> 29 #include <linux/pvclock_gtod.h> 30 #include <linux/time64.h> 31 #include <linux/timekeeping.h> 32 #include <linux/timekeeper_internal.h> 33 34 #include <linux/mm.h> 35 36 struct start_info _xen_start_info; 37 struct start_info *xen_start_info = &_xen_start_info; 38 EXPORT_SYMBOL(xen_start_info); 39 40 enum xen_domain_type xen_domain_type = XEN_NATIVE; 41 EXPORT_SYMBOL(xen_domain_type); 42 43 struct shared_info xen_dummy_shared_info; 44 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 45 46 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 47 static struct vcpu_info __percpu *xen_vcpu_info; 48 49 /* These are unused until we support booting "pre-ballooned" */ 50 unsigned long xen_released_pages; 51 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; 52 53 static __read_mostly unsigned int xen_events_irq; 54 55 static __initdata struct device_node *xen_node; 56 57 int xen_remap_domain_gfn_array(struct vm_area_struct *vma, 58 unsigned long addr, 59 xen_pfn_t *gfn, int nr, 60 int *err_ptr, pgprot_t prot, 61 unsigned domid, 62 struct page **pages) 63 { 64 return xen_xlate_remap_gfn_array(vma, addr, gfn, nr, err_ptr, 65 prot, domid, pages); 66 } 67 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array); 68 69 /* Not used by XENFEAT_auto_translated guests. */ 70 int xen_remap_domain_gfn_range(struct vm_area_struct *vma, 71 unsigned long addr, 72 xen_pfn_t gfn, int nr, 73 pgprot_t prot, unsigned domid, 74 struct page **pages) 75 { 76 return -ENOSYS; 77 } 78 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range); 79 80 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma, 81 int nr, struct page **pages) 82 { 83 return xen_xlate_unmap_gfn_range(vma, nr, pages); 84 } 85 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range); 86 87 static unsigned long long xen_stolen_accounting(int cpu) 88 { 89 struct vcpu_runstate_info state; 90 91 BUG_ON(cpu != smp_processor_id()); 92 93 xen_get_runstate_snapshot(&state); 94 95 WARN_ON(state.state != RUNSTATE_running); 96 97 return state.time[RUNSTATE_runnable] + state.time[RUNSTATE_offline]; 98 } 99 100 static void xen_read_wallclock(struct timespec64 *ts) 101 { 102 u32 version; 103 struct timespec64 now, ts_monotonic; 104 struct shared_info *s = HYPERVISOR_shared_info; 105 struct pvclock_wall_clock *wall_clock = &(s->wc); 106 107 /* get wallclock at system boot */ 108 do { 109 version = wall_clock->version; 110 rmb(); /* fetch version before time */ 111 now.tv_sec = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec; 112 now.tv_nsec = wall_clock->nsec; 113 rmb(); /* fetch time before checking version */ 114 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 115 116 /* time since system boot */ 117 ktime_get_ts64(&ts_monotonic); 118 *ts = timespec64_add(now, ts_monotonic); 119 } 120 121 static int xen_pvclock_gtod_notify(struct notifier_block *nb, 122 unsigned long was_set, void *priv) 123 { 124 /* Protected by the calling core code serialization */ 125 static struct timespec64 next_sync; 126 127 struct xen_platform_op op; 128 struct timespec64 now, system_time; 129 struct timekeeper *tk = priv; 130 131 now.tv_sec = tk->xtime_sec; 132 now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 133 system_time = timespec64_add(now, tk->wall_to_monotonic); 134 135 /* 136 * We only take the expensive HV call when the clock was set 137 * or when the 11 minutes RTC synchronization time elapsed. 138 */ 139 if (!was_set && timespec64_compare(&now, &next_sync) < 0) 140 return NOTIFY_OK; 141 142 op.cmd = XENPF_settime64; 143 op.u.settime64.mbz = 0; 144 op.u.settime64.secs = now.tv_sec; 145 op.u.settime64.nsecs = now.tv_nsec; 146 op.u.settime64.system_time = timespec64_to_ns(&system_time); 147 (void)HYPERVISOR_platform_op(&op); 148 149 /* 150 * Move the next drift compensation time 11 minutes 151 * ahead. That's emulating the sync_cmos_clock() update for 152 * the hardware RTC. 153 */ 154 next_sync = now; 155 next_sync.tv_sec += 11 * 60; 156 157 return NOTIFY_OK; 158 } 159 160 static struct notifier_block xen_pvclock_gtod_notifier = { 161 .notifier_call = xen_pvclock_gtod_notify, 162 }; 163 164 static void xen_percpu_init(void) 165 { 166 struct vcpu_register_vcpu_info info; 167 struct vcpu_info *vcpup; 168 int err; 169 int cpu = get_cpu(); 170 171 /* 172 * VCPUOP_register_vcpu_info cannot be called twice for the same 173 * vcpu, so if vcpu_info is already registered, just get out. This 174 * can happen with cpu-hotplug. 175 */ 176 if (per_cpu(xen_vcpu, cpu) != NULL) 177 goto after_register_vcpu_info; 178 179 pr_info("Xen: initializing cpu%d\n", cpu); 180 vcpup = per_cpu_ptr(xen_vcpu_info, cpu); 181 182 info.mfn = virt_to_gfn(vcpup); 183 info.offset = xen_offset_in_page(vcpup); 184 185 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 186 BUG_ON(err); 187 per_cpu(xen_vcpu, cpu) = vcpup; 188 189 xen_setup_runstate_info(cpu); 190 191 after_register_vcpu_info: 192 enable_percpu_irq(xen_events_irq, 0); 193 put_cpu(); 194 } 195 196 static void xen_restart(enum reboot_mode reboot_mode, const char *cmd) 197 { 198 struct sched_shutdown r = { .reason = SHUTDOWN_reboot }; 199 int rc; 200 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r); 201 BUG_ON(rc); 202 } 203 204 static void xen_power_off(void) 205 { 206 struct sched_shutdown r = { .reason = SHUTDOWN_poweroff }; 207 int rc; 208 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r); 209 BUG_ON(rc); 210 } 211 212 static int xen_cpu_notification(struct notifier_block *self, 213 unsigned long action, 214 void *hcpu) 215 { 216 switch (action) { 217 case CPU_STARTING: 218 xen_percpu_init(); 219 break; 220 case CPU_DYING: 221 disable_percpu_irq(xen_events_irq); 222 break; 223 default: 224 break; 225 } 226 227 return NOTIFY_OK; 228 } 229 230 static struct notifier_block xen_cpu_notifier = { 231 .notifier_call = xen_cpu_notification, 232 }; 233 234 static irqreturn_t xen_arm_callback(int irq, void *arg) 235 { 236 xen_hvm_evtchn_do_upcall(); 237 return IRQ_HANDLED; 238 } 239 240 /* 241 * see Documentation/devicetree/bindings/arm/xen.txt for the 242 * documentation of the Xen Device Tree format. 243 */ 244 #define GRANT_TABLE_PHYSADDR 0 245 void __init xen_early_init(void) 246 { 247 int len; 248 const char *s = NULL; 249 const char *version = NULL; 250 const char *xen_prefix = "xen,xen-"; 251 252 xen_node = of_find_compatible_node(NULL, NULL, "xen,xen"); 253 if (!xen_node) { 254 pr_debug("No Xen support\n"); 255 return; 256 } 257 s = of_get_property(xen_node, "compatible", &len); 258 if (strlen(xen_prefix) + 3 < len && 259 !strncmp(xen_prefix, s, strlen(xen_prefix))) 260 version = s + strlen(xen_prefix); 261 if (version == NULL) { 262 pr_debug("Xen version not found\n"); 263 return; 264 } 265 266 pr_info("Xen %s support found\n", version); 267 268 xen_domain_type = XEN_HVM_DOMAIN; 269 270 xen_setup_features(); 271 272 if (xen_feature(XENFEAT_dom0)) 273 xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED; 274 else 275 xen_start_info->flags &= ~(SIF_INITDOMAIN|SIF_PRIVILEGED); 276 277 if (!console_set_on_cmdline && !xen_initial_domain()) 278 add_preferred_console("hvc", 0, NULL); 279 } 280 281 static int __init xen_guest_init(void) 282 { 283 struct xen_add_to_physmap xatp; 284 struct shared_info *shared_info_page = NULL; 285 struct resource res; 286 phys_addr_t grant_frames; 287 288 if (!xen_domain()) 289 return 0; 290 291 if (of_address_to_resource(xen_node, GRANT_TABLE_PHYSADDR, &res)) { 292 pr_err("Xen grant table base address not found\n"); 293 return -ENODEV; 294 } 295 grant_frames = res.start; 296 297 xen_events_irq = irq_of_parse_and_map(xen_node, 0); 298 if (!xen_events_irq) { 299 pr_err("Xen event channel interrupt not found\n"); 300 return -ENODEV; 301 } 302 303 shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL); 304 305 if (!shared_info_page) { 306 pr_err("not enough memory\n"); 307 return -ENOMEM; 308 } 309 xatp.domid = DOMID_SELF; 310 xatp.idx = 0; 311 xatp.space = XENMAPSPACE_shared_info; 312 xatp.gpfn = virt_to_gfn(shared_info_page); 313 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 314 BUG(); 315 316 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 317 318 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 319 * page, we use it in the event channel upcall and in some pvclock 320 * related functions. 321 * The shared info contains exactly 1 CPU (the boot CPU). The guest 322 * is required to use VCPUOP_register_vcpu_info to place vcpu info 323 * for secondary CPUs as they are brought up. 324 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0. 325 */ 326 xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info), 327 sizeof(struct vcpu_info)); 328 if (xen_vcpu_info == NULL) 329 return -ENOMEM; 330 331 if (gnttab_setup_auto_xlat_frames(grant_frames)) { 332 free_percpu(xen_vcpu_info); 333 return -ENOMEM; 334 } 335 gnttab_init(); 336 if (!xen_initial_domain()) 337 xenbus_probe(NULL); 338 339 /* 340 * Making sure board specific code will not set up ops for 341 * cpu idle and cpu freq. 342 */ 343 disable_cpuidle(); 344 disable_cpufreq(); 345 346 xen_init_IRQ(); 347 348 if (request_percpu_irq(xen_events_irq, xen_arm_callback, 349 "events", &xen_vcpu)) { 350 pr_err("Error request IRQ %d\n", xen_events_irq); 351 return -EINVAL; 352 } 353 354 xen_percpu_init(); 355 356 register_cpu_notifier(&xen_cpu_notifier); 357 358 pv_time_ops.steal_clock = xen_stolen_accounting; 359 static_key_slow_inc(¶virt_steal_enabled); 360 if (xen_initial_domain()) 361 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier); 362 363 return 0; 364 } 365 early_initcall(xen_guest_init); 366 367 static int __init xen_pm_init(void) 368 { 369 if (!xen_domain()) 370 return -ENODEV; 371 372 pm_power_off = xen_power_off; 373 arm_pm_restart = xen_restart; 374 if (!xen_initial_domain()) { 375 struct timespec64 ts; 376 xen_read_wallclock(&ts); 377 do_settimeofday64(&ts); 378 } 379 380 return 0; 381 } 382 late_initcall(xen_pm_init); 383 384 385 /* empty stubs */ 386 void xen_arch_pre_suspend(void) { } 387 void xen_arch_post_suspend(int suspend_cancelled) { } 388 void xen_timer_resume(void) { } 389 void xen_arch_resume(void) { } 390 void xen_arch_suspend(void) { } 391 392 393 /* In the hypercall.S file. */ 394 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op); 395 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op); 396 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version); 397 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io); 398 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op); 399 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op); 400 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op); 401 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op); 402 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op); 403 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op); 404 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op); 405 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall); 406 EXPORT_SYMBOL_GPL(privcmd_call); 407