1 // SPDX-License-Identifier: GPL-2.0 2 3 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG 4 #include <linux/memblock.h> 5 #endif 6 #include <linux/console.h> 7 #include <linux/cpu.h> 8 #include <linux/kexec.h> 9 #include <linux/slab.h> 10 #include <linux/panic_notifier.h> 11 12 #include <xen/xen.h> 13 #include <xen/features.h> 14 #include <xen/interface/sched.h> 15 #include <xen/interface/version.h> 16 #include <xen/page.h> 17 18 #include <asm/xen/hypercall.h> 19 #include <asm/xen/hypervisor.h> 20 #include <asm/cpu.h> 21 #include <asm/e820/api.h> 22 #include <asm/setup.h> 23 24 #include "xen-ops.h" 25 #include "smp.h" 26 #include "pmu.h" 27 28 EXPORT_SYMBOL_GPL(hypercall_page); 29 30 /* 31 * Pointer to the xen_vcpu_info structure or 32 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info 33 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info 34 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point 35 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to 36 * acknowledge pending events. 37 * Also more subtly it is used by the patched version of irq enable/disable 38 * e.g. xen_irq_enable_direct and xen_iret in PV mode. 39 * 40 * The desire to be able to do those mask/unmask operations as a single 41 * instruction by using the per-cpu offset held in %gs is the real reason 42 * vcpu info is in a per-cpu pointer and the original reason for this 43 * hypercall. 44 * 45 */ 46 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 47 48 /* 49 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info 50 * hypercall. This can be used both in PV and PVHVM mode. The structure 51 * overrides the default per_cpu(xen_vcpu, cpu) value. 52 */ 53 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 54 55 /* Linux <-> Xen vCPU id mapping */ 56 DEFINE_PER_CPU(uint32_t, xen_vcpu_id); 57 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id); 58 59 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START; 60 EXPORT_SYMBOL(machine_to_phys_mapping); 61 unsigned long machine_to_phys_nr; 62 EXPORT_SYMBOL(machine_to_phys_nr); 63 64 struct start_info *xen_start_info; 65 EXPORT_SYMBOL_GPL(xen_start_info); 66 67 struct shared_info xen_dummy_shared_info; 68 69 __read_mostly int xen_have_vector_callback; 70 EXPORT_SYMBOL_GPL(xen_have_vector_callback); 71 72 /* 73 * NB: These need to live in .data or alike because they're used by 74 * xen_prepare_pvh() which runs before clearing the bss. 75 */ 76 enum xen_domain_type __ro_after_init xen_domain_type = XEN_NATIVE; 77 EXPORT_SYMBOL_GPL(xen_domain_type); 78 uint32_t __ro_after_init xen_start_flags; 79 EXPORT_SYMBOL(xen_start_flags); 80 81 /* 82 * Point at some empty memory to start with. We map the real shared_info 83 * page as soon as fixmap is up and running. 84 */ 85 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info; 86 87 /* 88 * Flag to determine whether vcpu info placement is available on all 89 * VCPUs. We assume it is to start with, and then set it to zero on 90 * the first failure. This is because it can succeed on some VCPUs 91 * and not others, since it can involve hypervisor memory allocation, 92 * or because the guest failed to guarantee all the appropriate 93 * constraints on all VCPUs (ie buffer can't cross a page boundary). 94 * 95 * Note that any particular CPU may be using a placed vcpu structure, 96 * but we can only optimise if the all are. 97 * 98 * 0: not available, 1: available 99 */ 100 int xen_have_vcpu_info_placement = 1; 101 102 static int xen_cpu_up_online(unsigned int cpu) 103 { 104 xen_init_lock_cpu(cpu); 105 return 0; 106 } 107 108 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int), 109 int (*cpu_dead_cb)(unsigned int)) 110 { 111 int rc; 112 113 rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE, 114 "x86/xen/guest:prepare", 115 cpu_up_prepare_cb, cpu_dead_cb); 116 if (rc >= 0) { 117 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 118 "x86/xen/guest:online", 119 xen_cpu_up_online, NULL); 120 if (rc < 0) 121 cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE); 122 } 123 124 return rc >= 0 ? 0 : rc; 125 } 126 127 static int xen_vcpu_setup_restore(int cpu) 128 { 129 int rc = 0; 130 131 /* Any per_cpu(xen_vcpu) is stale, so reset it */ 132 xen_vcpu_info_reset(cpu); 133 134 /* 135 * For PVH and PVHVM, setup online VCPUs only. The rest will 136 * be handled by hotplug. 137 */ 138 if (xen_pv_domain() || 139 (xen_hvm_domain() && cpu_online(cpu))) { 140 rc = xen_vcpu_setup(cpu); 141 } 142 143 return rc; 144 } 145 146 /* 147 * On restore, set the vcpu placement up again. 148 * If it fails, then we're in a bad state, since 149 * we can't back out from using it... 150 */ 151 void xen_vcpu_restore(void) 152 { 153 int cpu, rc; 154 155 for_each_possible_cpu(cpu) { 156 bool other_cpu = (cpu != smp_processor_id()); 157 bool is_up; 158 159 if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID) 160 continue; 161 162 /* Only Xen 4.5 and higher support this. */ 163 is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, 164 xen_vcpu_nr(cpu), NULL) > 0; 165 166 if (other_cpu && is_up && 167 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL)) 168 BUG(); 169 170 if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock)) 171 xen_setup_runstate_info(cpu); 172 173 rc = xen_vcpu_setup_restore(cpu); 174 if (rc) 175 pr_emerg_once("vcpu restore failed for cpu=%d err=%d. " 176 "System will hang.\n", cpu, rc); 177 /* 178 * In case xen_vcpu_setup_restore() fails, do not bring up the 179 * VCPU. This helps us avoid the resulting OOPS when the VCPU 180 * accesses pvclock_vcpu_time via xen_vcpu (which is NULL.) 181 * Note that this does not improve the situation much -- now the 182 * VM hangs instead of OOPSing -- with the VCPUs that did not 183 * fail, spinning in stop_machine(), waiting for the failed 184 * VCPUs to come up. 185 */ 186 if (other_cpu && is_up && (rc == 0) && 187 HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL)) 188 BUG(); 189 } 190 } 191 192 void xen_vcpu_info_reset(int cpu) 193 { 194 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) { 195 per_cpu(xen_vcpu, cpu) = 196 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)]; 197 } else { 198 /* Set to NULL so that if somebody accesses it we get an OOPS */ 199 per_cpu(xen_vcpu, cpu) = NULL; 200 } 201 } 202 203 int xen_vcpu_setup(int cpu) 204 { 205 struct vcpu_register_vcpu_info info; 206 int err; 207 struct vcpu_info *vcpup; 208 209 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); 210 211 /* 212 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu) 213 * and at restore (xen_vcpu_restore). Also called for hotplugged 214 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm). 215 * However, the hypercall can only be done once (see below) so if a VCPU 216 * is offlined and comes back online then let's not redo the hypercall. 217 * 218 * For PV it is called during restore (xen_vcpu_restore) and bootup 219 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not 220 * use this function. 221 */ 222 if (xen_hvm_domain()) { 223 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu)) 224 return 0; 225 } 226 227 if (xen_have_vcpu_info_placement) { 228 vcpup = &per_cpu(xen_vcpu_info, cpu); 229 info.mfn = arbitrary_virt_to_mfn(vcpup); 230 info.offset = offset_in_page(vcpup); 231 232 /* 233 * Check to see if the hypervisor will put the vcpu_info 234 * structure where we want it, which allows direct access via 235 * a percpu-variable. 236 * N.B. This hypercall can _only_ be called once per CPU. 237 * Subsequent calls will error out with -EINVAL. This is due to 238 * the fact that hypervisor has no unregister variant and this 239 * hypercall does not allow to over-write info.mfn and 240 * info.offset. 241 */ 242 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, 243 xen_vcpu_nr(cpu), &info); 244 245 if (err) { 246 pr_warn_once("register_vcpu_info failed: cpu=%d err=%d\n", 247 cpu, err); 248 xen_have_vcpu_info_placement = 0; 249 } else { 250 /* 251 * This cpu is using the registered vcpu info, even if 252 * later ones fail to. 253 */ 254 per_cpu(xen_vcpu, cpu) = vcpup; 255 } 256 } 257 258 if (!xen_have_vcpu_info_placement) 259 xen_vcpu_info_reset(cpu); 260 261 return ((per_cpu(xen_vcpu, cpu) == NULL) ? -ENODEV : 0); 262 } 263 264 void __init xen_banner(void) 265 { 266 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); 267 struct xen_extraversion extra; 268 269 HYPERVISOR_xen_version(XENVER_extraversion, &extra); 270 271 pr_info("Booting kernel on %s\n", pv_info.name); 272 pr_info("Xen version: %u.%u%s%s\n", 273 version >> 16, version & 0xffff, extra.extraversion, 274 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) 275 ? " (preserve-AD)" : ""); 276 } 277 278 /* Check if running on Xen version (major, minor) or later */ 279 bool xen_running_on_version_or_later(unsigned int major, unsigned int minor) 280 { 281 unsigned int version; 282 283 if (!xen_domain()) 284 return false; 285 286 version = HYPERVISOR_xen_version(XENVER_version, NULL); 287 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) || 288 ((version >> 16) > major)) 289 return true; 290 return false; 291 } 292 293 void __init xen_add_preferred_consoles(void) 294 { 295 add_preferred_console("xenboot", 0, NULL); 296 if (!boot_params.screen_info.orig_video_isVGA) 297 add_preferred_console("tty", 0, NULL); 298 add_preferred_console("hvc", 0, NULL); 299 if (boot_params.screen_info.orig_video_isVGA) 300 add_preferred_console("tty", 0, NULL); 301 } 302 303 void xen_reboot(int reason) 304 { 305 struct sched_shutdown r = { .reason = reason }; 306 int cpu; 307 308 for_each_online_cpu(cpu) 309 xen_pmu_finish(cpu); 310 311 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) 312 BUG(); 313 } 314 315 static int reboot_reason = SHUTDOWN_reboot; 316 static bool xen_legacy_crash; 317 void xen_emergency_restart(void) 318 { 319 xen_reboot(reboot_reason); 320 } 321 322 static int 323 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr) 324 { 325 if (!kexec_crash_loaded()) { 326 if (xen_legacy_crash) 327 xen_reboot(SHUTDOWN_crash); 328 329 reboot_reason = SHUTDOWN_crash; 330 331 /* 332 * If panic_timeout==0 then we are supposed to wait forever. 333 * However, to preserve original dom0 behavior we have to drop 334 * into hypervisor. (domU behavior is controlled by its 335 * config file) 336 */ 337 if (panic_timeout == 0) 338 panic_timeout = -1; 339 } 340 return NOTIFY_DONE; 341 } 342 343 static int __init parse_xen_legacy_crash(char *arg) 344 { 345 xen_legacy_crash = true; 346 return 0; 347 } 348 early_param("xen_legacy_crash", parse_xen_legacy_crash); 349 350 static struct notifier_block xen_panic_block = { 351 .notifier_call = xen_panic_event, 352 .priority = INT_MIN 353 }; 354 355 int xen_panic_handler_init(void) 356 { 357 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block); 358 return 0; 359 } 360 361 void xen_pin_vcpu(int cpu) 362 { 363 static bool disable_pinning; 364 struct sched_pin_override pin_override; 365 int ret; 366 367 if (disable_pinning) 368 return; 369 370 pin_override.pcpu = cpu; 371 ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override); 372 373 /* Ignore errors when removing override. */ 374 if (cpu < 0) 375 return; 376 377 switch (ret) { 378 case -ENOSYS: 379 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n", 380 cpu); 381 disable_pinning = true; 382 break; 383 case -EPERM: 384 WARN(1, "Trying to pin vcpu without having privilege to do so\n"); 385 disable_pinning = true; 386 break; 387 case -EINVAL: 388 case -EBUSY: 389 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n", 390 cpu); 391 break; 392 case 0: 393 break; 394 default: 395 WARN(1, "rc %d while trying to pin vcpu\n", ret); 396 disable_pinning = true; 397 } 398 } 399 400 #ifdef CONFIG_HOTPLUG_CPU 401 void xen_arch_register_cpu(int num) 402 { 403 arch_register_cpu(num); 404 } 405 EXPORT_SYMBOL(xen_arch_register_cpu); 406 407 void xen_arch_unregister_cpu(int num) 408 { 409 arch_unregister_cpu(num); 410 } 411 EXPORT_SYMBOL(xen_arch_unregister_cpu); 412 #endif 413