arch/arm/xen/enlighten.c

   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 int xen_starting_cpu(unsigned int cpu)
 165 {
 166         struct vcpu_register_vcpu_info info;
 167         struct vcpu_info *vcpup;
 168         int err;
 169
 170         /*
 171          * VCPUOP_register_vcpu_info cannot be called twice for the same
 172          * vcpu, so if vcpu_info is already registered, just get out. This
 173          * can happen with cpu-hotplug.
 174          */
 175         if (per_cpu(xen_vcpu, cpu) != NULL)
 176                 goto after_register_vcpu_info;
 177
 178         pr_info("Xen: initializing cpu%d\n", cpu);
 179         vcpup = per_cpu_ptr(xen_vcpu_info, cpu);
 180
 181         info.mfn = virt_to_gfn(vcpup);
 182         info.offset = xen_offset_in_page(vcpup);
 183
 184         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 185         BUG_ON(err);
 186         per_cpu(xen_vcpu, cpu) = vcpup;
 187
 188         xen_setup_runstate_info(cpu);
 189
 190 after_register_vcpu_info:
 191         enable_percpu_irq(xen_events_irq, 0);
 192         return 0;
 193 }
 194
 195 static int xen_dying_cpu(unsigned int cpu)
 196 {
 197         disable_percpu_irq(xen_events_irq);
 198         return 0;
 199 }
 200
 201 static void xen_restart(enum reboot_mode reboot_mode, const char *cmd)
 202 {
 203         struct sched_shutdown r = { .reason = SHUTDOWN_reboot };
 204         int rc;
 205         rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
 206         BUG_ON(rc);
 207 }
 208
 209 static void xen_power_off(void)
 210 {
 211         struct sched_shutdown r = { .reason = SHUTDOWN_poweroff };
 212         int rc;
 213         rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
 214         BUG_ON(rc);
 215 }
 216
 217 static irqreturn_t xen_arm_callback(int irq, void *arg)
 218 {
 219         xen_hvm_evtchn_do_upcall();
 220         return IRQ_HANDLED;
 221 }
 222
 223 /*
 224  * see Documentation/devicetree/bindings/arm/xen.txt for the
 225  * documentation of the Xen Device Tree format.
 226  */
 227 #define GRANT_TABLE_PHYSADDR 0
 228 void __init xen_early_init(void)
 229 {
 230         int len;
 231         const char *s = NULL;
 232         const char *version = NULL;
 233         const char *xen_prefix = "xen,xen-";
 234
 235         xen_node = of_find_compatible_node(NULL, NULL, "xen,xen");
 236         if (!xen_node) {
 237                 pr_debug("No Xen support\n");
 238                 return;
 239         }
 240         s = of_get_property(xen_node, "compatible", &len);
 241         if (strlen(xen_prefix) + 3  < len &&
 242                         !strncmp(xen_prefix, s, strlen(xen_prefix)))
 243                 version = s + strlen(xen_prefix);
 244         if (version == NULL) {
 245                 pr_debug("Xen version not found\n");
 246                 return;
 247         }
 248
 249         pr_info("Xen %s support found\n", version);
 250
 251         xen_domain_type = XEN_HVM_DOMAIN;
 252
 253         xen_setup_features();
 254
 255         if (xen_feature(XENFEAT_dom0))
 256                 xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
 257         else
 258                 xen_start_info->flags &= ~(SIF_INITDOMAIN|SIF_PRIVILEGED);
 259
 260         if (!console_set_on_cmdline && !xen_initial_domain())
 261                 add_preferred_console("hvc", 0, NULL);
 262 }
 263
 264 static int __init xen_guest_init(void)
 265 {
 266         struct xen_add_to_physmap xatp;
 267         struct shared_info *shared_info_page = NULL;
 268         struct resource res;
 269         phys_addr_t grant_frames;
 270
 271         if (!xen_domain())
 272                 return 0;
 273
 274         if (of_address_to_resource(xen_node, GRANT_TABLE_PHYSADDR, &res)) {
 275                 pr_err("Xen grant table base address not found\n");
 276                 return -ENODEV;
 277         }
 278         grant_frames = res.start;
 279
 280         xen_events_irq = irq_of_parse_and_map(xen_node, 0);
 281         if (!xen_events_irq) {
 282                 pr_err("Xen event channel interrupt not found\n");
 283                 return -ENODEV;
 284         }
 285
 286         shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL);
 287
 288         if (!shared_info_page) {
 289                 pr_err("not enough memory\n");
 290                 return -ENOMEM;
 291         }
 292         xatp.domid = DOMID_SELF;
 293         xatp.idx = 0;
 294         xatp.space = XENMAPSPACE_shared_info;
 295         xatp.gpfn = virt_to_gfn(shared_info_page);
 296         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
 297                 BUG();
 298
 299         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
 300
 301         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
 302          * page, we use it in the event channel upcall and in some pvclock
 303          * related functions.
 304          * The shared info contains exactly 1 CPU (the boot CPU). The guest
 305          * is required to use VCPUOP_register_vcpu_info to place vcpu info
 306          * for secondary CPUs as they are brought up.
 307          * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
 308          */
 309         xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info),
 310                                                sizeof(struct vcpu_info));
 311         if (xen_vcpu_info == NULL)
 312                 return -ENOMEM;
 313
 314         if (gnttab_setup_auto_xlat_frames(grant_frames)) {
 315                 free_percpu(xen_vcpu_info);
 316                 return -ENOMEM;
 317         }
 318         gnttab_init();
 319         if (!xen_initial_domain())
 320                 xenbus_probe(NULL);
 321
 322         /*
 323          * Making sure board specific code will not set up ops for
 324          * cpu idle and cpu freq.
 325          */
 326         disable_cpuidle();
 327         disable_cpufreq();
 328
 329         xen_init_IRQ();
 330
 331         if (request_percpu_irq(xen_events_irq, xen_arm_callback,
 332                                "events", &xen_vcpu)) {
 333                 pr_err("Error request IRQ %d\n", xen_events_irq);
 334                 return -EINVAL;
 335         }
 336
 337         pv_time_ops.steal_clock = xen_stolen_accounting;
 338         static_key_slow_inc(&paravirt_steal_enabled);
 339         if (xen_initial_domain())
 340                 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
 341
 342         return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
 343                                  "AP_ARM_XEN_STARTING", xen_starting_cpu,
 344                                  xen_dying_cpu);
 345 }
 346 early_initcall(xen_guest_init);
 347
 348 static int __init xen_pm_init(void)
 349 {
 350         if (!xen_domain())
 351                 return -ENODEV;
 352
 353         pm_power_off = xen_power_off;
 354         arm_pm_restart = xen_restart;
 355         if (!xen_initial_domain()) {
 356                 struct timespec64 ts;
 357                 xen_read_wallclock(&ts);
 358                 do_settimeofday64(&ts);
 359         }
 360
 361         return 0;
 362 }
 363 late_initcall(xen_pm_init);
 364
 365
 366 /* empty stubs */
 367 void xen_arch_pre_suspend(void) { }
 368 void xen_arch_post_suspend(int suspend_cancelled) { }
 369 void xen_timer_resume(void) { }
 370 void xen_arch_resume(void) { }
 371 void xen_arch_suspend(void) { }
 372
 373
 374 /* In the hypercall.S file. */
 375 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op);
 376 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op);
 377 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version);
 378 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io);
 379 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op);
 380 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op);
 381 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op);
 382 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op);
 383 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op);
 384 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op);
 385 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op);
 386 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall);
 387 EXPORT_SYMBOL_GPL(privcmd_call);