arch/x86/xen/enlighten.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <linux/console.h>
   4 #include <linux/cpu.h>
   5 #include <linux/kexec.h>
   6 #include <linux/memblock.h>
   7 #include <linux/slab.h>
   8 #include <linux/panic_notifier.h>
   9
  10 #include <xen/xen.h>
  11 #include <xen/features.h>
  12 #include <xen/interface/sched.h>
  13 #include <xen/interface/version.h>
  14 #include <xen/page.h>
  15
  16 #include <asm/xen/hypercall.h>
  17 #include <asm/xen/hypervisor.h>
  18 #include <asm/cpu.h>
  19 #include <asm/e820/api.h>
  20 #include <asm/setup.h>
  21
  22 #include "xen-ops.h"
  23
  24 EXPORT_SYMBOL_GPL(hypercall_page);
  25
  26 /*
  27  * Pointer to the xen_vcpu_info structure or
  28  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
  29  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
  30  * but during boot it is switched to point to xen_vcpu_info.
  31  * The pointer is used in xen_evtchn_do_upcall to acknowledge pending events.
  32  * Make sure that xen_vcpu_info doesn't cross a page boundary by making it
  33  * cache-line aligned (the struct is guaranteed to have a size of 64 bytes,
  34  * which matches the cache line size of 64-bit x86 processors).
  35  */
  36 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  37 DEFINE_PER_CPU_ALIGNED(struct vcpu_info, xen_vcpu_info);
  38
  39 /* Linux <-> Xen vCPU id mapping */
  40 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
  41 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
  42
  43 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
  44 EXPORT_SYMBOL(machine_to_phys_mapping);
  45 unsigned long  machine_to_phys_nr;
  46 EXPORT_SYMBOL(machine_to_phys_nr);
  47
  48 struct start_info *xen_start_info;
  49 EXPORT_SYMBOL_GPL(xen_start_info);
  50
  51 struct shared_info xen_dummy_shared_info;
  52
  53 __read_mostly bool xen_have_vector_callback = true;
  54 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
  55
  56 /*
  57  * NB: These need to live in .data or alike because they're used by
  58  * xen_prepare_pvh() which runs before clearing the bss.
  59  */
  60 enum xen_domain_type __ro_after_init xen_domain_type = XEN_NATIVE;
  61 EXPORT_SYMBOL_GPL(xen_domain_type);
  62 uint32_t __ro_after_init xen_start_flags;
  63 EXPORT_SYMBOL(xen_start_flags);
  64
  65 /*
  66  * Point at some empty memory to start with. We map the real shared_info
  67  * page as soon as fixmap is up and running.
  68  */
  69 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
  70
  71 static int xen_cpu_up_online(unsigned int cpu)
  72 {
  73         xen_init_lock_cpu(cpu);
  74         return 0;
  75 }
  76
  77 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
  78                     int (*cpu_dead_cb)(unsigned int))
  79 {
  80         int rc;
  81
  82         rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
  83                                        "x86/xen/guest:prepare",
  84                                        cpu_up_prepare_cb, cpu_dead_cb);
  85         if (rc >= 0) {
  86                 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
  87                                                "x86/xen/guest:online",
  88                                                xen_cpu_up_online, NULL);
  89                 if (rc < 0)
  90                         cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
  91         }
  92
  93         return rc >= 0 ? 0 : rc;
  94 }
  95
  96 static void xen_vcpu_setup_restore(int cpu)
  97 {
  98         /* Any per_cpu(xen_vcpu) is stale, so reset it */
  99         xen_vcpu_info_reset(cpu);
 100
 101         /*
 102          * For PVH and PVHVM, setup online VCPUs only. The rest will
 103          * be handled by hotplug.
 104          */
 105         if (xen_pv_domain() ||
 106             (xen_hvm_domain() && cpu_online(cpu)))
 107                 xen_vcpu_setup(cpu);
 108 }
 109
 110 /*
 111  * On restore, set the vcpu placement up again.
 112  * If it fails, then we're in a bad state, since
 113  * we can't back out from using it...
 114  */
 115 void xen_vcpu_restore(void)
 116 {
 117         int cpu;
 118
 119         for_each_possible_cpu(cpu) {
 120                 bool other_cpu = (cpu != smp_processor_id());
 121                 bool is_up;
 122
 123                 if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
 124                         continue;
 125
 126                 /* Only Xen 4.5 and higher support this. */
 127                 is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
 128                                            xen_vcpu_nr(cpu), NULL) > 0;
 129
 130                 if (other_cpu && is_up &&
 131                     HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
 132                         BUG();
 133
 134                 if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
 135                         xen_setup_runstate_info(cpu);
 136
 137                 xen_vcpu_setup_restore(cpu);
 138
 139                 if (other_cpu && is_up &&
 140                     HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
 141                         BUG();
 142         }
 143 }
 144
 145 void xen_vcpu_info_reset(int cpu)
 146 {
 147         if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
 148                 per_cpu(xen_vcpu, cpu) =
 149                         &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
 150         } else {
 151                 /* Set to NULL so that if somebody accesses it we get an OOPS */
 152                 per_cpu(xen_vcpu, cpu) = NULL;
 153         }
 154 }
 155
 156 void xen_vcpu_setup(int cpu)
 157 {
 158         struct vcpu_register_vcpu_info info;
 159         int err;
 160         struct vcpu_info *vcpup;
 161
 162         BUILD_BUG_ON(sizeof(*vcpup) > SMP_CACHE_BYTES);
 163         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
 164
 165         /*
 166          * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
 167          * and at restore (xen_vcpu_restore). Also called for hotplugged
 168          * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
 169          * However, the hypercall can only be done once (see below) so if a VCPU
 170          * is offlined and comes back online then let's not redo the hypercall.
 171          *
 172          * For PV it is called during restore (xen_vcpu_restore) and bootup
 173          * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
 174          * use this function.
 175          */
 176         if (xen_hvm_domain()) {
 177                 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
 178                         return;
 179         }
 180
 181         vcpup = &per_cpu(xen_vcpu_info, cpu);
 182         info.mfn = arbitrary_virt_to_mfn(vcpup);
 183         info.offset = offset_in_page(vcpup);
 184
 185         /*
 186          * N.B. This hypercall can _only_ be called once per CPU.
 187          * Subsequent calls will error out with -EINVAL. This is due to
 188          * the fact that hypervisor has no unregister variant and this
 189          * hypercall does not allow to over-write info.mfn and
 190          * info.offset.
 191          */
 192         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
 193                                  &info);
 194         if (err)
 195                 panic("register_vcpu_info failed: cpu=%d err=%d\n", cpu, err);
 196
 197         per_cpu(xen_vcpu, cpu) = vcpup;
 198 }
 199
 200 void __init xen_banner(void)
 201 {
 202         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
 203         struct xen_extraversion extra;
 204
 205         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
 206
 207         pr_info("Booting kernel on %s\n", pv_info.name);
 208         pr_info("Xen version: %u.%u%s%s\n",
 209                 version >> 16, version & 0xffff, extra.extraversion,
 210                 xen_feature(XENFEAT_mmu_pt_update_preserve_ad)
 211                 ? " (preserve-AD)" : "");
 212 }
 213
 214 /* Check if running on Xen version (major, minor) or later */
 215 bool xen_running_on_version_or_later(unsigned int major, unsigned int minor)
 216 {
 217         unsigned int version;
 218
 219         if (!xen_domain())
 220                 return false;
 221
 222         version = HYPERVISOR_xen_version(XENVER_version, NULL);
 223         if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
 224                 ((version >> 16) > major))
 225                 return true;
 226         return false;
 227 }
 228
 229 void __init xen_add_preferred_consoles(void)
 230 {
 231         add_preferred_console("xenboot", 0, NULL);
 232         if (!boot_params.screen_info.orig_video_isVGA)
 233                 add_preferred_console("tty", 0, NULL);
 234         add_preferred_console("hvc", 0, NULL);
 235         if (boot_params.screen_info.orig_video_isVGA)
 236                 add_preferred_console("tty", 0, NULL);
 237 }
 238
 239 void xen_reboot(int reason)
 240 {
 241         struct sched_shutdown r = { .reason = reason };
 242         int cpu;
 243
 244         for_each_online_cpu(cpu)
 245                 xen_pmu_finish(cpu);
 246
 247         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
 248                 BUG();
 249 }
 250
 251 static int reboot_reason = SHUTDOWN_reboot;
 252 static bool xen_legacy_crash;
 253 void xen_emergency_restart(void)
 254 {
 255         xen_reboot(reboot_reason);
 256 }
 257
 258 static int
 259 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
 260 {
 261         if (!kexec_crash_loaded()) {
 262                 if (xen_legacy_crash)
 263                         xen_reboot(SHUTDOWN_crash);
 264
 265                 reboot_reason = SHUTDOWN_crash;
 266
 267                 /*
 268                  * If panic_timeout==0 then we are supposed to wait forever.
 269                  * However, to preserve original dom0 behavior we have to drop
 270                  * into hypervisor. (domU behavior is controlled by its
 271                  * config file)
 272                  */
 273                 if (panic_timeout == 0)
 274                         panic_timeout = -1;
 275         }
 276         return NOTIFY_DONE;
 277 }
 278
 279 static int __init parse_xen_legacy_crash(char *arg)
 280 {
 281         xen_legacy_crash = true;
 282         return 0;
 283 }
 284 early_param("xen_legacy_crash", parse_xen_legacy_crash);
 285
 286 static struct notifier_block xen_panic_block = {
 287         .notifier_call = xen_panic_event,
 288         .priority = INT_MIN
 289 };
 290
 291 int xen_panic_handler_init(void)
 292 {
 293         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
 294         return 0;
 295 }
 296
 297 void xen_pin_vcpu(int cpu)
 298 {
 299         static bool disable_pinning;
 300         struct sched_pin_override pin_override;
 301         int ret;
 302
 303         if (disable_pinning)
 304                 return;
 305
 306         pin_override.pcpu = cpu;
 307         ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
 308
 309         /* Ignore errors when removing override. */
 310         if (cpu < 0)
 311                 return;
 312
 313         switch (ret) {
 314         case -ENOSYS:
 315                 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
 316                         cpu);
 317                 disable_pinning = true;
 318                 break;
 319         case -EPERM:
 320                 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
 321                 disable_pinning = true;
 322                 break;
 323         case -EINVAL:
 324         case -EBUSY:
 325                 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
 326                         cpu);
 327                 break;
 328         case 0:
 329                 break;
 330         default:
 331                 WARN(1, "rc %d while trying to pin vcpu\n", ret);
 332                 disable_pinning = true;
 333         }
 334 }
 335
 336 #ifdef CONFIG_HOTPLUG_CPU
 337 void xen_arch_register_cpu(int num)
 338 {
 339         arch_register_cpu(num);
 340 }
 341 EXPORT_SYMBOL(xen_arch_register_cpu);
 342
 343 void xen_arch_unregister_cpu(int num)
 344 {
 345         arch_unregister_cpu(num);
 346 }
 347 EXPORT_SYMBOL(xen_arch_unregister_cpu);
 348 #endif
 349
 350 /* Amount of extra memory space we add to the e820 ranges */
 351 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
 352
 353 void __init xen_add_extra_mem(unsigned long start_pfn, unsigned long n_pfns)
 354 {
 355         unsigned int i;
 356
 357         /*
 358          * No need to check for zero size, should happen rarely and will only
 359          * write a new entry regarded to be unused due to zero size.
 360          */
 361         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 362                 /* Add new region. */
 363                 if (xen_extra_mem[i].n_pfns == 0) {
 364                         xen_extra_mem[i].start_pfn = start_pfn;
 365                         xen_extra_mem[i].n_pfns = n_pfns;
 366                         break;
 367                 }
 368                 /* Append to existing region. */
 369                 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
 370                     start_pfn) {
 371                         xen_extra_mem[i].n_pfns += n_pfns;
 372                         break;
 373                 }
 374         }
 375         if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 376                 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 377
 378         memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 379 }
 380
 381 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC
 382 int __init arch_xen_unpopulated_init(struct resource **res)
 383 {
 384         unsigned int i;
 385
 386         if (!xen_domain())
 387                 return -ENODEV;
 388
 389         /* Must be set strictly before calling xen_free_unpopulated_pages(). */
 390         *res = &iomem_resource;
 391
 392         /*
 393          * Initialize with pages from the extra memory regions (see
 394          * arch/x86/xen/setup.c).
 395          */
 396         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 397                 unsigned int j;
 398
 399                 for (j = 0; j < xen_extra_mem[i].n_pfns; j++) {
 400                         struct page *pg =
 401                                 pfn_to_page(xen_extra_mem[i].start_pfn + j);
 402
 403                         xen_free_unpopulated_pages(1, &pg);
 404                 }
 405
 406                 /* Zero so region is not also added to the balloon driver. */
 407                 xen_extra_mem[i].n_pfns = 0;
 408         }
 409
 410         return 0;
 411 }
 412 #endif