1 /* 2 * Core of Xen paravirt_ops implementation. 3 * 4 * This file contains the xen_paravirt_ops structure itself, and the 5 * implementations for: 6 * - privileged instructions 7 * - interrupt flags 8 * - segment operations 9 * - booting and setup 10 * 11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 12 */ 13 14 #include <linux/cpu.h> 15 #include <linux/kernel.h> 16 #include <linux/init.h> 17 #include <linux/smp.h> 18 #include <linux/preempt.h> 19 #include <linux/hardirq.h> 20 #include <linux/percpu.h> 21 #include <linux/delay.h> 22 #include <linux/start_kernel.h> 23 #include <linux/sched.h> 24 #include <linux/kprobes.h> 25 #include <linux/bootmem.h> 26 #include <linux/module.h> 27 #include <linux/mm.h> 28 #include <linux/page-flags.h> 29 #include <linux/highmem.h> 30 #include <linux/console.h> 31 #include <linux/pci.h> 32 #include <linux/gfp.h> 33 #include <linux/memblock.h> 34 #include <linux/edd.h> 35 36 #include <xen/xen.h> 37 #include <xen/events.h> 38 #include <xen/interface/xen.h> 39 #include <xen/interface/version.h> 40 #include <xen/interface/physdev.h> 41 #include <xen/interface/vcpu.h> 42 #include <xen/interface/memory.h> 43 #include <xen/interface/xen-mca.h> 44 #include <xen/features.h> 45 #include <xen/page.h> 46 #include <xen/hvm.h> 47 #include <xen/hvc-console.h> 48 #include <xen/acpi.h> 49 50 #include <asm/paravirt.h> 51 #include <asm/apic.h> 52 #include <asm/page.h> 53 #include <asm/xen/pci.h> 54 #include <asm/xen/hypercall.h> 55 #include <asm/xen/hypervisor.h> 56 #include <asm/fixmap.h> 57 #include <asm/processor.h> 58 #include <asm/proto.h> 59 #include <asm/msr-index.h> 60 #include <asm/traps.h> 61 #include <asm/setup.h> 62 #include <asm/desc.h> 63 #include <asm/pgalloc.h> 64 #include <asm/pgtable.h> 65 #include <asm/tlbflush.h> 66 #include <asm/reboot.h> 67 #include <asm/stackprotector.h> 68 #include <asm/hypervisor.h> 69 #include <asm/mwait.h> 70 #include <asm/pci_x86.h> 71 #include <asm/pat.h> 72 73 #ifdef CONFIG_ACPI 74 #include <linux/acpi.h> 75 #include <asm/acpi.h> 76 #include <acpi/pdc_intel.h> 77 #include <acpi/processor.h> 78 #include <xen/interface/platform.h> 79 #endif 80 81 #include "xen-ops.h" 82 #include "mmu.h" 83 #include "smp.h" 84 #include "multicalls.h" 85 86 EXPORT_SYMBOL_GPL(hypercall_page); 87 88 /* 89 * Pointer to the xen_vcpu_info structure or 90 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info 91 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info 92 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point 93 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to 94 * acknowledge pending events. 95 * Also more subtly it is used by the patched version of irq enable/disable 96 * e.g. xen_irq_enable_direct and xen_iret in PV mode. 97 * 98 * The desire to be able to do those mask/unmask operations as a single 99 * instruction by using the per-cpu offset held in %gs is the real reason 100 * vcpu info is in a per-cpu pointer and the original reason for this 101 * hypercall. 102 * 103 */ 104 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 105 106 /* 107 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info 108 * hypercall. This can be used both in PV and PVHVM mode. The structure 109 * overrides the default per_cpu(xen_vcpu, cpu) value. 110 */ 111 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 112 113 enum xen_domain_type xen_domain_type = XEN_NATIVE; 114 EXPORT_SYMBOL_GPL(xen_domain_type); 115 116 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START; 117 EXPORT_SYMBOL(machine_to_phys_mapping); 118 unsigned long machine_to_phys_nr; 119 EXPORT_SYMBOL(machine_to_phys_nr); 120 121 struct start_info *xen_start_info; 122 EXPORT_SYMBOL_GPL(xen_start_info); 123 124 struct shared_info xen_dummy_shared_info; 125 126 void *xen_initial_gdt; 127 128 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE); 129 __read_mostly int xen_have_vector_callback; 130 EXPORT_SYMBOL_GPL(xen_have_vector_callback); 131 132 /* 133 * Point at some empty memory to start with. We map the real shared_info 134 * page as soon as fixmap is up and running. 135 */ 136 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info; 137 138 /* 139 * Flag to determine whether vcpu info placement is available on all 140 * VCPUs. We assume it is to start with, and then set it to zero on 141 * the first failure. This is because it can succeed on some VCPUs 142 * and not others, since it can involve hypervisor memory allocation, 143 * or because the guest failed to guarantee all the appropriate 144 * constraints on all VCPUs (ie buffer can't cross a page boundary). 145 * 146 * Note that any particular CPU may be using a placed vcpu structure, 147 * but we can only optimise if the all are. 148 * 149 * 0: not available, 1: available 150 */ 151 static int have_vcpu_info_placement = 1; 152 153 struct tls_descs { 154 struct desc_struct desc[3]; 155 }; 156 157 /* 158 * Updating the 3 TLS descriptors in the GDT on every task switch is 159 * surprisingly expensive so we avoid updating them if they haven't 160 * changed. Since Xen writes different descriptors than the one 161 * passed in the update_descriptor hypercall we keep shadow copies to 162 * compare against. 163 */ 164 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc); 165 166 static void clamp_max_cpus(void) 167 { 168 #ifdef CONFIG_SMP 169 if (setup_max_cpus > MAX_VIRT_CPUS) 170 setup_max_cpus = MAX_VIRT_CPUS; 171 #endif 172 } 173 174 static void xen_vcpu_setup(int cpu) 175 { 176 struct vcpu_register_vcpu_info info; 177 int err; 178 struct vcpu_info *vcpup; 179 180 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); 181 182 /* 183 * This path is called twice on PVHVM - first during bootup via 184 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being 185 * hotplugged: cpu_up -> xen_hvm_cpu_notify. 186 * As we can only do the VCPUOP_register_vcpu_info once lets 187 * not over-write its result. 188 * 189 * For PV it is called during restore (xen_vcpu_restore) and bootup 190 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not 191 * use this function. 192 */ 193 if (xen_hvm_domain()) { 194 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu)) 195 return; 196 } 197 if (cpu < MAX_VIRT_CPUS) 198 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 199 200 if (!have_vcpu_info_placement) { 201 if (cpu >= MAX_VIRT_CPUS) 202 clamp_max_cpus(); 203 return; 204 } 205 206 vcpup = &per_cpu(xen_vcpu_info, cpu); 207 info.mfn = arbitrary_virt_to_mfn(vcpup); 208 info.offset = offset_in_page(vcpup); 209 210 /* Check to see if the hypervisor will put the vcpu_info 211 structure where we want it, which allows direct access via 212 a percpu-variable. 213 N.B. This hypercall can _only_ be called once per CPU. Subsequent 214 calls will error out with -EINVAL. This is due to the fact that 215 hypervisor has no unregister variant and this hypercall does not 216 allow to over-write info.mfn and info.offset. 217 */ 218 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 219 220 if (err) { 221 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); 222 have_vcpu_info_placement = 0; 223 clamp_max_cpus(); 224 } else { 225 /* This cpu is using the registered vcpu info, even if 226 later ones fail to. */ 227 per_cpu(xen_vcpu, cpu) = vcpup; 228 } 229 } 230 231 /* 232 * On restore, set the vcpu placement up again. 233 * If it fails, then we're in a bad state, since 234 * we can't back out from using it... 235 */ 236 void xen_vcpu_restore(void) 237 { 238 int cpu; 239 240 for_each_possible_cpu(cpu) { 241 bool other_cpu = (cpu != smp_processor_id()); 242 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL); 243 244 if (other_cpu && is_up && 245 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) 246 BUG(); 247 248 xen_setup_runstate_info(cpu); 249 250 if (have_vcpu_info_placement) 251 xen_vcpu_setup(cpu); 252 253 if (other_cpu && is_up && 254 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) 255 BUG(); 256 } 257 } 258 259 static void __init xen_banner(void) 260 { 261 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); 262 struct xen_extraversion extra; 263 HYPERVISOR_xen_version(XENVER_extraversion, &extra); 264 265 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 266 pv_info.name); 267 printk(KERN_INFO "Xen version: %d.%d%s%s\n", 268 version >> 16, version & 0xffff, extra.extraversion, 269 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); 270 } 271 /* Check if running on Xen version (major, minor) or later */ 272 bool 273 xen_running_on_version_or_later(unsigned int major, unsigned int minor) 274 { 275 unsigned int version; 276 277 if (!xen_domain()) 278 return false; 279 280 version = HYPERVISOR_xen_version(XENVER_version, NULL); 281 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) || 282 ((version >> 16) > major)) 283 return true; 284 return false; 285 } 286 287 #define CPUID_THERM_POWER_LEAF 6 288 #define APERFMPERF_PRESENT 0 289 290 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0; 291 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0; 292 293 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask; 294 static __read_mostly unsigned int cpuid_leaf5_ecx_val; 295 static __read_mostly unsigned int cpuid_leaf5_edx_val; 296 297 static void xen_cpuid(unsigned int *ax, unsigned int *bx, 298 unsigned int *cx, unsigned int *dx) 299 { 300 unsigned maskebx = ~0; 301 unsigned maskecx = ~0; 302 unsigned maskedx = ~0; 303 unsigned setecx = 0; 304 /* 305 * Mask out inconvenient features, to try and disable as many 306 * unsupported kernel subsystems as possible. 307 */ 308 switch (*ax) { 309 case 1: 310 maskecx = cpuid_leaf1_ecx_mask; 311 setecx = cpuid_leaf1_ecx_set_mask; 312 maskedx = cpuid_leaf1_edx_mask; 313 break; 314 315 case CPUID_MWAIT_LEAF: 316 /* Synthesize the values.. */ 317 *ax = 0; 318 *bx = 0; 319 *cx = cpuid_leaf5_ecx_val; 320 *dx = cpuid_leaf5_edx_val; 321 return; 322 323 case CPUID_THERM_POWER_LEAF: 324 /* Disabling APERFMPERF for kernel usage */ 325 maskecx = ~(1 << APERFMPERF_PRESENT); 326 break; 327 328 case 0xb: 329 /* Suppress extended topology stuff */ 330 maskebx = 0; 331 break; 332 } 333 334 asm(XEN_EMULATE_PREFIX "cpuid" 335 : "=a" (*ax), 336 "=b" (*bx), 337 "=c" (*cx), 338 "=d" (*dx) 339 : "0" (*ax), "2" (*cx)); 340 341 *bx &= maskebx; 342 *cx &= maskecx; 343 *cx |= setecx; 344 *dx &= maskedx; 345 346 } 347 348 static bool __init xen_check_mwait(void) 349 { 350 #ifdef CONFIG_ACPI 351 struct xen_platform_op op = { 352 .cmd = XENPF_set_processor_pminfo, 353 .u.set_pminfo.id = -1, 354 .u.set_pminfo.type = XEN_PM_PDC, 355 }; 356 uint32_t buf[3]; 357 unsigned int ax, bx, cx, dx; 358 unsigned int mwait_mask; 359 360 /* We need to determine whether it is OK to expose the MWAIT 361 * capability to the kernel to harvest deeper than C3 states from ACPI 362 * _CST using the processor_harvest_xen.c module. For this to work, we 363 * need to gather the MWAIT_LEAF values (which the cstate.c code 364 * checks against). The hypervisor won't expose the MWAIT flag because 365 * it would break backwards compatibility; so we will find out directly 366 * from the hardware and hypercall. 367 */ 368 if (!xen_initial_domain()) 369 return false; 370 371 /* 372 * When running under platform earlier than Xen4.2, do not expose 373 * mwait, to avoid the risk of loading native acpi pad driver 374 */ 375 if (!xen_running_on_version_or_later(4, 2)) 376 return false; 377 378 ax = 1; 379 cx = 0; 380 381 native_cpuid(&ax, &bx, &cx, &dx); 382 383 mwait_mask = (1 << (X86_FEATURE_EST % 32)) | 384 (1 << (X86_FEATURE_MWAIT % 32)); 385 386 if ((cx & mwait_mask) != mwait_mask) 387 return false; 388 389 /* We need to emulate the MWAIT_LEAF and for that we need both 390 * ecx and edx. The hypercall provides only partial information. 391 */ 392 393 ax = CPUID_MWAIT_LEAF; 394 bx = 0; 395 cx = 0; 396 dx = 0; 397 398 native_cpuid(&ax, &bx, &cx, &dx); 399 400 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so, 401 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3. 402 */ 403 buf[0] = ACPI_PDC_REVISION_ID; 404 buf[1] = 1; 405 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP); 406 407 set_xen_guest_handle(op.u.set_pminfo.pdc, buf); 408 409 if ((HYPERVISOR_dom0_op(&op) == 0) && 410 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) { 411 cpuid_leaf5_ecx_val = cx; 412 cpuid_leaf5_edx_val = dx; 413 } 414 return true; 415 #else 416 return false; 417 #endif 418 } 419 static void __init xen_init_cpuid_mask(void) 420 { 421 unsigned int ax, bx, cx, dx; 422 unsigned int xsave_mask; 423 424 cpuid_leaf1_edx_mask = 425 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */ 426 (1 << X86_FEATURE_ACC)); /* thermal monitoring */ 427 428 if (!xen_initial_domain()) 429 cpuid_leaf1_edx_mask &= 430 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */ 431 432 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32)); 433 434 ax = 1; 435 cx = 0; 436 xen_cpuid(&ax, &bx, &cx, &dx); 437 438 xsave_mask = 439 (1 << (X86_FEATURE_XSAVE % 32)) | 440 (1 << (X86_FEATURE_OSXSAVE % 32)); 441 442 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */ 443 if ((cx & xsave_mask) != xsave_mask) 444 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */ 445 if (xen_check_mwait()) 446 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32)); 447 } 448 449 static void xen_set_debugreg(int reg, unsigned long val) 450 { 451 HYPERVISOR_set_debugreg(reg, val); 452 } 453 454 static unsigned long xen_get_debugreg(int reg) 455 { 456 return HYPERVISOR_get_debugreg(reg); 457 } 458 459 static void xen_end_context_switch(struct task_struct *next) 460 { 461 xen_mc_flush(); 462 paravirt_end_context_switch(next); 463 } 464 465 static unsigned long xen_store_tr(void) 466 { 467 return 0; 468 } 469 470 /* 471 * Set the page permissions for a particular virtual address. If the 472 * address is a vmalloc mapping (or other non-linear mapping), then 473 * find the linear mapping of the page and also set its protections to 474 * match. 475 */ 476 static void set_aliased_prot(void *v, pgprot_t prot) 477 { 478 int level; 479 pte_t *ptep; 480 pte_t pte; 481 unsigned long pfn; 482 struct page *page; 483 484 ptep = lookup_address((unsigned long)v, &level); 485 BUG_ON(ptep == NULL); 486 487 pfn = pte_pfn(*ptep); 488 page = pfn_to_page(pfn); 489 490 pte = pfn_pte(pfn, prot); 491 492 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) 493 BUG(); 494 495 if (!PageHighMem(page)) { 496 void *av = __va(PFN_PHYS(pfn)); 497 498 if (av != v) 499 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) 500 BUG(); 501 } else 502 kmap_flush_unused(); 503 } 504 505 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) 506 { 507 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 508 int i; 509 510 for(i = 0; i < entries; i += entries_per_page) 511 set_aliased_prot(ldt + i, PAGE_KERNEL_RO); 512 } 513 514 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) 515 { 516 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 517 int i; 518 519 for(i = 0; i < entries; i += entries_per_page) 520 set_aliased_prot(ldt + i, PAGE_KERNEL); 521 } 522 523 static void xen_set_ldt(const void *addr, unsigned entries) 524 { 525 struct mmuext_op *op; 526 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 527 528 trace_xen_cpu_set_ldt(addr, entries); 529 530 op = mcs.args; 531 op->cmd = MMUEXT_SET_LDT; 532 op->arg1.linear_addr = (unsigned long)addr; 533 op->arg2.nr_ents = entries; 534 535 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 536 537 xen_mc_issue(PARAVIRT_LAZY_CPU); 538 } 539 540 static void xen_load_gdt(const struct desc_ptr *dtr) 541 { 542 unsigned long va = dtr->address; 543 unsigned int size = dtr->size + 1; 544 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 545 unsigned long frames[pages]; 546 int f; 547 548 /* 549 * A GDT can be up to 64k in size, which corresponds to 8192 550 * 8-byte entries, or 16 4k pages.. 551 */ 552 553 BUG_ON(size > 65536); 554 BUG_ON(va & ~PAGE_MASK); 555 556 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 557 int level; 558 pte_t *ptep; 559 unsigned long pfn, mfn; 560 void *virt; 561 562 /* 563 * The GDT is per-cpu and is in the percpu data area. 564 * That can be virtually mapped, so we need to do a 565 * page-walk to get the underlying MFN for the 566 * hypercall. The page can also be in the kernel's 567 * linear range, so we need to RO that mapping too. 568 */ 569 ptep = lookup_address(va, &level); 570 BUG_ON(ptep == NULL); 571 572 pfn = pte_pfn(*ptep); 573 mfn = pfn_to_mfn(pfn); 574 virt = __va(PFN_PHYS(pfn)); 575 576 frames[f] = mfn; 577 578 make_lowmem_page_readonly((void *)va); 579 make_lowmem_page_readonly(virt); 580 } 581 582 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 583 BUG(); 584 } 585 586 /* 587 * load_gdt for early boot, when the gdt is only mapped once 588 */ 589 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr) 590 { 591 unsigned long va = dtr->address; 592 unsigned int size = dtr->size + 1; 593 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 594 unsigned long frames[pages]; 595 int f; 596 597 /* 598 * A GDT can be up to 64k in size, which corresponds to 8192 599 * 8-byte entries, or 16 4k pages.. 600 */ 601 602 BUG_ON(size > 65536); 603 BUG_ON(va & ~PAGE_MASK); 604 605 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 606 pte_t pte; 607 unsigned long pfn, mfn; 608 609 pfn = virt_to_pfn(va); 610 mfn = pfn_to_mfn(pfn); 611 612 pte = pfn_pte(pfn, PAGE_KERNEL_RO); 613 614 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 615 BUG(); 616 617 frames[f] = mfn; 618 } 619 620 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 621 BUG(); 622 } 623 624 static inline bool desc_equal(const struct desc_struct *d1, 625 const struct desc_struct *d2) 626 { 627 return d1->a == d2->a && d1->b == d2->b; 628 } 629 630 static void load_TLS_descriptor(struct thread_struct *t, 631 unsigned int cpu, unsigned int i) 632 { 633 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i]; 634 struct desc_struct *gdt; 635 xmaddr_t maddr; 636 struct multicall_space mc; 637 638 if (desc_equal(shadow, &t->tls_array[i])) 639 return; 640 641 *shadow = t->tls_array[i]; 642 643 gdt = get_cpu_gdt_table(cpu); 644 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 645 mc = __xen_mc_entry(0); 646 647 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 648 } 649 650 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 651 { 652 /* 653 * XXX sleazy hack: If we're being called in a lazy-cpu zone 654 * and lazy gs handling is enabled, it means we're in a 655 * context switch, and %gs has just been saved. This means we 656 * can zero it out to prevent faults on exit from the 657 * hypervisor if the next process has no %gs. Either way, it 658 * has been saved, and the new value will get loaded properly. 659 * This will go away as soon as Xen has been modified to not 660 * save/restore %gs for normal hypercalls. 661 * 662 * On x86_64, this hack is not used for %gs, because gs points 663 * to KERNEL_GS_BASE (and uses it for PDA references), so we 664 * must not zero %gs on x86_64 665 * 666 * For x86_64, we need to zero %fs, otherwise we may get an 667 * exception between the new %fs descriptor being loaded and 668 * %fs being effectively cleared at __switch_to(). 669 */ 670 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { 671 #ifdef CONFIG_X86_32 672 lazy_load_gs(0); 673 #else 674 loadsegment(fs, 0); 675 #endif 676 } 677 678 xen_mc_batch(); 679 680 load_TLS_descriptor(t, cpu, 0); 681 load_TLS_descriptor(t, cpu, 1); 682 load_TLS_descriptor(t, cpu, 2); 683 684 xen_mc_issue(PARAVIRT_LAZY_CPU); 685 } 686 687 #ifdef CONFIG_X86_64 688 static void xen_load_gs_index(unsigned int idx) 689 { 690 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) 691 BUG(); 692 } 693 #endif 694 695 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 696 const void *ptr) 697 { 698 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); 699 u64 entry = *(u64 *)ptr; 700 701 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry); 702 703 preempt_disable(); 704 705 xen_mc_flush(); 706 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 707 BUG(); 708 709 preempt_enable(); 710 } 711 712 static int cvt_gate_to_trap(int vector, const gate_desc *val, 713 struct trap_info *info) 714 { 715 unsigned long addr; 716 717 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT) 718 return 0; 719 720 info->vector = vector; 721 722 addr = gate_offset(*val); 723 #ifdef CONFIG_X86_64 724 /* 725 * Look for known traps using IST, and substitute them 726 * appropriately. The debugger ones are the only ones we care 727 * about. Xen will handle faults like double_fault, 728 * so we should never see them. Warn if 729 * there's an unexpected IST-using fault handler. 730 */ 731 if (addr == (unsigned long)debug) 732 addr = (unsigned long)xen_debug; 733 else if (addr == (unsigned long)int3) 734 addr = (unsigned long)xen_int3; 735 else if (addr == (unsigned long)stack_segment) 736 addr = (unsigned long)xen_stack_segment; 737 else if (addr == (unsigned long)double_fault) { 738 /* Don't need to handle these */ 739 return 0; 740 #ifdef CONFIG_X86_MCE 741 } else if (addr == (unsigned long)machine_check) { 742 /* 743 * when xen hypervisor inject vMCE to guest, 744 * use native mce handler to handle it 745 */ 746 ; 747 #endif 748 } else if (addr == (unsigned long)nmi) 749 /* 750 * Use the native version as well. 751 */ 752 ; 753 else { 754 /* Some other trap using IST? */ 755 if (WARN_ON(val->ist != 0)) 756 return 0; 757 } 758 #endif /* CONFIG_X86_64 */ 759 info->address = addr; 760 761 info->cs = gate_segment(*val); 762 info->flags = val->dpl; 763 /* interrupt gates clear IF */ 764 if (val->type == GATE_INTERRUPT) 765 info->flags |= 1 << 2; 766 767 return 1; 768 } 769 770 /* Locations of each CPU's IDT */ 771 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 772 773 /* Set an IDT entry. If the entry is part of the current IDT, then 774 also update Xen. */ 775 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 776 { 777 unsigned long p = (unsigned long)&dt[entrynum]; 778 unsigned long start, end; 779 780 trace_xen_cpu_write_idt_entry(dt, entrynum, g); 781 782 preempt_disable(); 783 784 start = __this_cpu_read(idt_desc.address); 785 end = start + __this_cpu_read(idt_desc.size) + 1; 786 787 xen_mc_flush(); 788 789 native_write_idt_entry(dt, entrynum, g); 790 791 if (p >= start && (p + 8) <= end) { 792 struct trap_info info[2]; 793 794 info[1].address = 0; 795 796 if (cvt_gate_to_trap(entrynum, g, &info[0])) 797 if (HYPERVISOR_set_trap_table(info)) 798 BUG(); 799 } 800 801 preempt_enable(); 802 } 803 804 static void xen_convert_trap_info(const struct desc_ptr *desc, 805 struct trap_info *traps) 806 { 807 unsigned in, out, count; 808 809 count = (desc->size+1) / sizeof(gate_desc); 810 BUG_ON(count > 256); 811 812 for (in = out = 0; in < count; in++) { 813 gate_desc *entry = (gate_desc*)(desc->address) + in; 814 815 if (cvt_gate_to_trap(in, entry, &traps[out])) 816 out++; 817 } 818 traps[out].address = 0; 819 } 820 821 void xen_copy_trap_info(struct trap_info *traps) 822 { 823 const struct desc_ptr *desc = &__get_cpu_var(idt_desc); 824 825 xen_convert_trap_info(desc, traps); 826 } 827 828 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 829 hold a spinlock to protect the static traps[] array (static because 830 it avoids allocation, and saves stack space). */ 831 static void xen_load_idt(const struct desc_ptr *desc) 832 { 833 static DEFINE_SPINLOCK(lock); 834 static struct trap_info traps[257]; 835 836 trace_xen_cpu_load_idt(desc); 837 838 spin_lock(&lock); 839 840 __get_cpu_var(idt_desc) = *desc; 841 842 xen_convert_trap_info(desc, traps); 843 844 xen_mc_flush(); 845 if (HYPERVISOR_set_trap_table(traps)) 846 BUG(); 847 848 spin_unlock(&lock); 849 } 850 851 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 852 they're handled differently. */ 853 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 854 const void *desc, int type) 855 { 856 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 857 858 preempt_disable(); 859 860 switch (type) { 861 case DESC_LDT: 862 case DESC_TSS: 863 /* ignore */ 864 break; 865 866 default: { 867 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); 868 869 xen_mc_flush(); 870 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 871 BUG(); 872 } 873 874 } 875 876 preempt_enable(); 877 } 878 879 /* 880 * Version of write_gdt_entry for use at early boot-time needed to 881 * update an entry as simply as possible. 882 */ 883 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, 884 const void *desc, int type) 885 { 886 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 887 888 switch (type) { 889 case DESC_LDT: 890 case DESC_TSS: 891 /* ignore */ 892 break; 893 894 default: { 895 xmaddr_t maddr = virt_to_machine(&dt[entry]); 896 897 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 898 dt[entry] = *(struct desc_struct *)desc; 899 } 900 901 } 902 } 903 904 static void xen_load_sp0(struct tss_struct *tss, 905 struct thread_struct *thread) 906 { 907 struct multicall_space mcs; 908 909 mcs = xen_mc_entry(0); 910 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 911 xen_mc_issue(PARAVIRT_LAZY_CPU); 912 } 913 914 static void xen_set_iopl_mask(unsigned mask) 915 { 916 struct physdev_set_iopl set_iopl; 917 918 /* Force the change at ring 0. */ 919 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; 920 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 921 } 922 923 static void xen_io_delay(void) 924 { 925 } 926 927 #ifdef CONFIG_X86_LOCAL_APIC 928 static unsigned long xen_set_apic_id(unsigned int x) 929 { 930 WARN_ON(1); 931 return x; 932 } 933 static unsigned int xen_get_apic_id(unsigned long x) 934 { 935 return ((x)>>24) & 0xFFu; 936 } 937 static u32 xen_apic_read(u32 reg) 938 { 939 struct xen_platform_op op = { 940 .cmd = XENPF_get_cpuinfo, 941 .interface_version = XENPF_INTERFACE_VERSION, 942 .u.pcpu_info.xen_cpuid = 0, 943 }; 944 int ret = 0; 945 946 /* Shouldn't need this as APIC is turned off for PV, and we only 947 * get called on the bootup processor. But just in case. */ 948 if (!xen_initial_domain() || smp_processor_id()) 949 return 0; 950 951 if (reg == APIC_LVR) 952 return 0x10; 953 954 if (reg != APIC_ID) 955 return 0; 956 957 ret = HYPERVISOR_dom0_op(&op); 958 if (ret) 959 return 0; 960 961 return op.u.pcpu_info.apic_id << 24; 962 } 963 964 static void xen_apic_write(u32 reg, u32 val) 965 { 966 /* Warn to see if there's any stray references */ 967 WARN_ON(1); 968 } 969 970 static u64 xen_apic_icr_read(void) 971 { 972 return 0; 973 } 974 975 static void xen_apic_icr_write(u32 low, u32 id) 976 { 977 /* Warn to see if there's any stray references */ 978 WARN_ON(1); 979 } 980 981 static void xen_apic_wait_icr_idle(void) 982 { 983 return; 984 } 985 986 static u32 xen_safe_apic_wait_icr_idle(void) 987 { 988 return 0; 989 } 990 991 static void set_xen_basic_apic_ops(void) 992 { 993 apic->read = xen_apic_read; 994 apic->write = xen_apic_write; 995 apic->icr_read = xen_apic_icr_read; 996 apic->icr_write = xen_apic_icr_write; 997 apic->wait_icr_idle = xen_apic_wait_icr_idle; 998 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; 999 apic->set_apic_id = xen_set_apic_id; 1000 apic->get_apic_id = xen_get_apic_id; 1001 1002 #ifdef CONFIG_SMP 1003 apic->send_IPI_allbutself = xen_send_IPI_allbutself; 1004 apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself; 1005 apic->send_IPI_mask = xen_send_IPI_mask; 1006 apic->send_IPI_all = xen_send_IPI_all; 1007 apic->send_IPI_self = xen_send_IPI_self; 1008 #endif 1009 } 1010 1011 #endif 1012 1013 static void xen_clts(void) 1014 { 1015 struct multicall_space mcs; 1016 1017 mcs = xen_mc_entry(0); 1018 1019 MULTI_fpu_taskswitch(mcs.mc, 0); 1020 1021 xen_mc_issue(PARAVIRT_LAZY_CPU); 1022 } 1023 1024 static DEFINE_PER_CPU(unsigned long, xen_cr0_value); 1025 1026 static unsigned long xen_read_cr0(void) 1027 { 1028 unsigned long cr0 = this_cpu_read(xen_cr0_value); 1029 1030 if (unlikely(cr0 == 0)) { 1031 cr0 = native_read_cr0(); 1032 this_cpu_write(xen_cr0_value, cr0); 1033 } 1034 1035 return cr0; 1036 } 1037 1038 static void xen_write_cr0(unsigned long cr0) 1039 { 1040 struct multicall_space mcs; 1041 1042 this_cpu_write(xen_cr0_value, cr0); 1043 1044 /* Only pay attention to cr0.TS; everything else is 1045 ignored. */ 1046 mcs = xen_mc_entry(0); 1047 1048 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); 1049 1050 xen_mc_issue(PARAVIRT_LAZY_CPU); 1051 } 1052 1053 static void xen_write_cr4(unsigned long cr4) 1054 { 1055 cr4 &= ~X86_CR4_PGE; 1056 cr4 &= ~X86_CR4_PSE; 1057 1058 native_write_cr4(cr4); 1059 } 1060 #ifdef CONFIG_X86_64 1061 static inline unsigned long xen_read_cr8(void) 1062 { 1063 return 0; 1064 } 1065 static inline void xen_write_cr8(unsigned long val) 1066 { 1067 BUG_ON(val); 1068 } 1069 #endif 1070 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) 1071 { 1072 int ret; 1073 1074 ret = 0; 1075 1076 switch (msr) { 1077 #ifdef CONFIG_X86_64 1078 unsigned which; 1079 u64 base; 1080 1081 case MSR_FS_BASE: which = SEGBASE_FS; goto set; 1082 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; 1083 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; 1084 1085 set: 1086 base = ((u64)high << 32) | low; 1087 if (HYPERVISOR_set_segment_base(which, base) != 0) 1088 ret = -EIO; 1089 break; 1090 #endif 1091 1092 case MSR_STAR: 1093 case MSR_CSTAR: 1094 case MSR_LSTAR: 1095 case MSR_SYSCALL_MASK: 1096 case MSR_IA32_SYSENTER_CS: 1097 case MSR_IA32_SYSENTER_ESP: 1098 case MSR_IA32_SYSENTER_EIP: 1099 /* Fast syscall setup is all done in hypercalls, so 1100 these are all ignored. Stub them out here to stop 1101 Xen console noise. */ 1102 break; 1103 1104 case MSR_IA32_CR_PAT: 1105 if (smp_processor_id() == 0) 1106 xen_set_pat(((u64)high << 32) | low); 1107 break; 1108 1109 default: 1110 ret = native_write_msr_safe(msr, low, high); 1111 } 1112 1113 return ret; 1114 } 1115 1116 void xen_setup_shared_info(void) 1117 { 1118 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 1119 set_fixmap(FIX_PARAVIRT_BOOTMAP, 1120 xen_start_info->shared_info); 1121 1122 HYPERVISOR_shared_info = 1123 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); 1124 } else 1125 HYPERVISOR_shared_info = 1126 (struct shared_info *)__va(xen_start_info->shared_info); 1127 1128 #ifndef CONFIG_SMP 1129 /* In UP this is as good a place as any to set up shared info */ 1130 xen_setup_vcpu_info_placement(); 1131 #endif 1132 1133 xen_setup_mfn_list_list(); 1134 } 1135 1136 /* This is called once we have the cpu_possible_mask */ 1137 void xen_setup_vcpu_info_placement(void) 1138 { 1139 int cpu; 1140 1141 for_each_possible_cpu(cpu) 1142 xen_vcpu_setup(cpu); 1143 1144 /* xen_vcpu_setup managed to place the vcpu_info within the 1145 percpu area for all cpus, so make use of it */ 1146 if (have_vcpu_info_placement) { 1147 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); 1148 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); 1149 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); 1150 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); 1151 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 1152 } 1153 } 1154 1155 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 1156 unsigned long addr, unsigned len) 1157 { 1158 char *start, *end, *reloc; 1159 unsigned ret; 1160 1161 start = end = reloc = NULL; 1162 1163 #define SITE(op, x) \ 1164 case PARAVIRT_PATCH(op.x): \ 1165 if (have_vcpu_info_placement) { \ 1166 start = (char *)xen_##x##_direct; \ 1167 end = xen_##x##_direct_end; \ 1168 reloc = xen_##x##_direct_reloc; \ 1169 } \ 1170 goto patch_site 1171 1172 switch (type) { 1173 SITE(pv_irq_ops, irq_enable); 1174 SITE(pv_irq_ops, irq_disable); 1175 SITE(pv_irq_ops, save_fl); 1176 SITE(pv_irq_ops, restore_fl); 1177 #undef SITE 1178 1179 patch_site: 1180 if (start == NULL || (end-start) > len) 1181 goto default_patch; 1182 1183 ret = paravirt_patch_insns(insnbuf, len, start, end); 1184 1185 /* Note: because reloc is assigned from something that 1186 appears to be an array, gcc assumes it's non-null, 1187 but doesn't know its relationship with start and 1188 end. */ 1189 if (reloc > start && reloc < end) { 1190 int reloc_off = reloc - start; 1191 long *relocp = (long *)(insnbuf + reloc_off); 1192 long delta = start - (char *)addr; 1193 1194 *relocp += delta; 1195 } 1196 break; 1197 1198 default_patch: 1199 default: 1200 ret = paravirt_patch_default(type, clobbers, insnbuf, 1201 addr, len); 1202 break; 1203 } 1204 1205 return ret; 1206 } 1207 1208 static const struct pv_info xen_info __initconst = { 1209 .paravirt_enabled = 1, 1210 .shared_kernel_pmd = 0, 1211 1212 #ifdef CONFIG_X86_64 1213 .extra_user_64bit_cs = FLAT_USER_CS64, 1214 #endif 1215 1216 .name = "Xen", 1217 }; 1218 1219 static const struct pv_init_ops xen_init_ops __initconst = { 1220 .patch = xen_patch, 1221 }; 1222 1223 static const struct pv_cpu_ops xen_cpu_ops __initconst = { 1224 .cpuid = xen_cpuid, 1225 1226 .set_debugreg = xen_set_debugreg, 1227 .get_debugreg = xen_get_debugreg, 1228 1229 .clts = xen_clts, 1230 1231 .read_cr0 = xen_read_cr0, 1232 .write_cr0 = xen_write_cr0, 1233 1234 .read_cr4 = native_read_cr4, 1235 .read_cr4_safe = native_read_cr4_safe, 1236 .write_cr4 = xen_write_cr4, 1237 1238 #ifdef CONFIG_X86_64 1239 .read_cr8 = xen_read_cr8, 1240 .write_cr8 = xen_write_cr8, 1241 #endif 1242 1243 .wbinvd = native_wbinvd, 1244 1245 .read_msr = native_read_msr_safe, 1246 .write_msr = xen_write_msr_safe, 1247 1248 .read_tsc = native_read_tsc, 1249 .read_pmc = native_read_pmc, 1250 1251 .read_tscp = native_read_tscp, 1252 1253 .iret = xen_iret, 1254 .irq_enable_sysexit = xen_sysexit, 1255 #ifdef CONFIG_X86_64 1256 .usergs_sysret32 = xen_sysret32, 1257 .usergs_sysret64 = xen_sysret64, 1258 #endif 1259 1260 .load_tr_desc = paravirt_nop, 1261 .set_ldt = xen_set_ldt, 1262 .load_gdt = xen_load_gdt, 1263 .load_idt = xen_load_idt, 1264 .load_tls = xen_load_tls, 1265 #ifdef CONFIG_X86_64 1266 .load_gs_index = xen_load_gs_index, 1267 #endif 1268 1269 .alloc_ldt = xen_alloc_ldt, 1270 .free_ldt = xen_free_ldt, 1271 1272 .store_idt = native_store_idt, 1273 .store_tr = xen_store_tr, 1274 1275 .write_ldt_entry = xen_write_ldt_entry, 1276 .write_gdt_entry = xen_write_gdt_entry, 1277 .write_idt_entry = xen_write_idt_entry, 1278 .load_sp0 = xen_load_sp0, 1279 1280 .set_iopl_mask = xen_set_iopl_mask, 1281 .io_delay = xen_io_delay, 1282 1283 /* Xen takes care of %gs when switching to usermode for us */ 1284 .swapgs = paravirt_nop, 1285 1286 .start_context_switch = paravirt_start_context_switch, 1287 .end_context_switch = xen_end_context_switch, 1288 }; 1289 1290 static const struct pv_apic_ops xen_apic_ops __initconst = { 1291 #ifdef CONFIG_X86_LOCAL_APIC 1292 .startup_ipi_hook = paravirt_nop, 1293 #endif 1294 }; 1295 1296 static void xen_reboot(int reason) 1297 { 1298 struct sched_shutdown r = { .reason = reason }; 1299 1300 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) 1301 BUG(); 1302 } 1303 1304 static void xen_restart(char *msg) 1305 { 1306 xen_reboot(SHUTDOWN_reboot); 1307 } 1308 1309 static void xen_emergency_restart(void) 1310 { 1311 xen_reboot(SHUTDOWN_reboot); 1312 } 1313 1314 static void xen_machine_halt(void) 1315 { 1316 xen_reboot(SHUTDOWN_poweroff); 1317 } 1318 1319 static void xen_machine_power_off(void) 1320 { 1321 if (pm_power_off) 1322 pm_power_off(); 1323 xen_reboot(SHUTDOWN_poweroff); 1324 } 1325 1326 static void xen_crash_shutdown(struct pt_regs *regs) 1327 { 1328 xen_reboot(SHUTDOWN_crash); 1329 } 1330 1331 static int 1332 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr) 1333 { 1334 xen_reboot(SHUTDOWN_crash); 1335 return NOTIFY_DONE; 1336 } 1337 1338 static struct notifier_block xen_panic_block = { 1339 .notifier_call= xen_panic_event, 1340 }; 1341 1342 int xen_panic_handler_init(void) 1343 { 1344 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block); 1345 return 0; 1346 } 1347 1348 static const struct machine_ops xen_machine_ops __initconst = { 1349 .restart = xen_restart, 1350 .halt = xen_machine_halt, 1351 .power_off = xen_machine_power_off, 1352 .shutdown = xen_machine_halt, 1353 .crash_shutdown = xen_crash_shutdown, 1354 .emergency_restart = xen_emergency_restart, 1355 }; 1356 1357 static void __init xen_boot_params_init_edd(void) 1358 { 1359 #if IS_ENABLED(CONFIG_EDD) 1360 struct xen_platform_op op; 1361 struct edd_info *edd_info; 1362 u32 *mbr_signature; 1363 unsigned nr; 1364 int ret; 1365 1366 edd_info = boot_params.eddbuf; 1367 mbr_signature = boot_params.edd_mbr_sig_buffer; 1368 1369 op.cmd = XENPF_firmware_info; 1370 1371 op.u.firmware_info.type = XEN_FW_DISK_INFO; 1372 for (nr = 0; nr < EDDMAXNR; nr++) { 1373 struct edd_info *info = edd_info + nr; 1374 1375 op.u.firmware_info.index = nr; 1376 info->params.length = sizeof(info->params); 1377 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params, 1378 &info->params); 1379 ret = HYPERVISOR_dom0_op(&op); 1380 if (ret) 1381 break; 1382 1383 #define C(x) info->x = op.u.firmware_info.u.disk_info.x 1384 C(device); 1385 C(version); 1386 C(interface_support); 1387 C(legacy_max_cylinder); 1388 C(legacy_max_head); 1389 C(legacy_sectors_per_track); 1390 #undef C 1391 } 1392 boot_params.eddbuf_entries = nr; 1393 1394 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE; 1395 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) { 1396 op.u.firmware_info.index = nr; 1397 ret = HYPERVISOR_dom0_op(&op); 1398 if (ret) 1399 break; 1400 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature; 1401 } 1402 boot_params.edd_mbr_sig_buf_entries = nr; 1403 #endif 1404 } 1405 1406 /* 1407 * Set up the GDT and segment registers for -fstack-protector. Until 1408 * we do this, we have to be careful not to call any stack-protected 1409 * function, which is most of the kernel. 1410 */ 1411 static void __init xen_setup_stackprotector(void) 1412 { 1413 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; 1414 pv_cpu_ops.load_gdt = xen_load_gdt_boot; 1415 1416 setup_stack_canary_segment(0); 1417 switch_to_new_gdt(0); 1418 1419 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; 1420 pv_cpu_ops.load_gdt = xen_load_gdt; 1421 } 1422 1423 /* First C function to be called on Xen boot */ 1424 asmlinkage void __init xen_start_kernel(void) 1425 { 1426 struct physdev_set_iopl set_iopl; 1427 int rc; 1428 1429 if (!xen_start_info) 1430 return; 1431 1432 xen_domain_type = XEN_PV_DOMAIN; 1433 1434 xen_setup_machphys_mapping(); 1435 1436 /* Install Xen paravirt ops */ 1437 pv_info = xen_info; 1438 pv_init_ops = xen_init_ops; 1439 pv_cpu_ops = xen_cpu_ops; 1440 pv_apic_ops = xen_apic_ops; 1441 1442 x86_init.resources.memory_setup = xen_memory_setup; 1443 x86_init.oem.arch_setup = xen_arch_setup; 1444 x86_init.oem.banner = xen_banner; 1445 1446 xen_init_time_ops(); 1447 1448 /* 1449 * Set up some pagetable state before starting to set any ptes. 1450 */ 1451 1452 xen_init_mmu_ops(); 1453 1454 /* Prevent unwanted bits from being set in PTEs. */ 1455 __supported_pte_mask &= ~_PAGE_GLOBAL; 1456 #if 0 1457 if (!xen_initial_domain()) 1458 #endif 1459 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); 1460 1461 __supported_pte_mask |= _PAGE_IOMAP; 1462 1463 /* 1464 * Prevent page tables from being allocated in highmem, even 1465 * if CONFIG_HIGHPTE is enabled. 1466 */ 1467 __userpte_alloc_gfp &= ~__GFP_HIGHMEM; 1468 1469 /* Work out if we support NX */ 1470 x86_configure_nx(); 1471 1472 xen_setup_features(); 1473 1474 /* Get mfn list */ 1475 if (!xen_feature(XENFEAT_auto_translated_physmap)) 1476 xen_build_dynamic_phys_to_machine(); 1477 1478 /* 1479 * Set up kernel GDT and segment registers, mainly so that 1480 * -fstack-protector code can be executed. 1481 */ 1482 xen_setup_stackprotector(); 1483 1484 xen_init_irq_ops(); 1485 xen_init_cpuid_mask(); 1486 1487 #ifdef CONFIG_X86_LOCAL_APIC 1488 /* 1489 * set up the basic apic ops. 1490 */ 1491 set_xen_basic_apic_ops(); 1492 #endif 1493 1494 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { 1495 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; 1496 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; 1497 } 1498 1499 machine_ops = xen_machine_ops; 1500 1501 /* 1502 * The only reliable way to retain the initial address of the 1503 * percpu gdt_page is to remember it here, so we can go and 1504 * mark it RW later, when the initial percpu area is freed. 1505 */ 1506 xen_initial_gdt = &per_cpu(gdt_page, 0); 1507 1508 xen_smp_init(); 1509 1510 #ifdef CONFIG_ACPI_NUMA 1511 /* 1512 * The pages we from Xen are not related to machine pages, so 1513 * any NUMA information the kernel tries to get from ACPI will 1514 * be meaningless. Prevent it from trying. 1515 */ 1516 acpi_numa = -1; 1517 #endif 1518 #ifdef CONFIG_X86_PAT 1519 /* 1520 * For right now disable the PAT. We should remove this once 1521 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1 1522 * (xen/pat: Disable PAT support for now) is reverted. 1523 */ 1524 pat_enabled = 0; 1525 #endif 1526 /* Don't do the full vcpu_info placement stuff until we have a 1527 possible map and a non-dummy shared_info. */ 1528 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; 1529 1530 local_irq_disable(); 1531 early_boot_irqs_disabled = true; 1532 1533 xen_raw_console_write("mapping kernel into physical memory\n"); 1534 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, xen_start_info->nr_pages); 1535 1536 /* Allocate and initialize top and mid mfn levels for p2m structure */ 1537 xen_build_mfn_list_list(); 1538 1539 /* keep using Xen gdt for now; no urgent need to change it */ 1540 1541 #ifdef CONFIG_X86_32 1542 pv_info.kernel_rpl = 1; 1543 if (xen_feature(XENFEAT_supervisor_mode_kernel)) 1544 pv_info.kernel_rpl = 0; 1545 #else 1546 pv_info.kernel_rpl = 0; 1547 #endif 1548 /* set the limit of our address space */ 1549 xen_reserve_top(); 1550 1551 /* We used to do this in xen_arch_setup, but that is too late on AMD 1552 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init 1553 * which pokes 0xcf8 port. 1554 */ 1555 set_iopl.iopl = 1; 1556 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 1557 if (rc != 0) 1558 xen_raw_printk("physdev_op failed %d\n", rc); 1559 1560 #ifdef CONFIG_X86_32 1561 /* set up basic CPUID stuff */ 1562 cpu_detect(&new_cpu_data); 1563 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU); 1564 new_cpu_data.wp_works_ok = 1; 1565 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1566 #endif 1567 1568 /* Poke various useful things into boot_params */ 1569 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1570 boot_params.hdr.ramdisk_image = xen_start_info->mod_start 1571 ? __pa(xen_start_info->mod_start) : 0; 1572 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1573 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1574 1575 if (!xen_initial_domain()) { 1576 add_preferred_console("xenboot", 0, NULL); 1577 add_preferred_console("tty", 0, NULL); 1578 add_preferred_console("hvc", 0, NULL); 1579 if (pci_xen) 1580 x86_init.pci.arch_init = pci_xen_init; 1581 } else { 1582 const struct dom0_vga_console_info *info = 1583 (void *)((char *)xen_start_info + 1584 xen_start_info->console.dom0.info_off); 1585 struct xen_platform_op op = { 1586 .cmd = XENPF_firmware_info, 1587 .interface_version = XENPF_INTERFACE_VERSION, 1588 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS, 1589 }; 1590 1591 xen_init_vga(info, xen_start_info->console.dom0.info_size); 1592 xen_start_info->console.domU.mfn = 0; 1593 xen_start_info->console.domU.evtchn = 0; 1594 1595 if (HYPERVISOR_dom0_op(&op) == 0) 1596 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags; 1597 1598 xen_init_apic(); 1599 1600 /* Make sure ACS will be enabled */ 1601 pci_request_acs(); 1602 1603 xen_acpi_sleep_register(); 1604 1605 /* Avoid searching for BIOS MP tables */ 1606 x86_init.mpparse.find_smp_config = x86_init_noop; 1607 x86_init.mpparse.get_smp_config = x86_init_uint_noop; 1608 1609 xen_boot_params_init_edd(); 1610 } 1611 #ifdef CONFIG_PCI 1612 /* PCI BIOS service won't work from a PV guest. */ 1613 pci_probe &= ~PCI_PROBE_BIOS; 1614 #endif 1615 xen_raw_console_write("about to get started...\n"); 1616 1617 xen_setup_runstate_info(0); 1618 1619 /* Start the world */ 1620 #ifdef CONFIG_X86_32 1621 i386_start_kernel(); 1622 #else 1623 x86_64_start_reservations((char *)__pa_symbol(&boot_params)); 1624 #endif 1625 } 1626 1627 void __ref xen_hvm_init_shared_info(void) 1628 { 1629 int cpu; 1630 struct xen_add_to_physmap xatp; 1631 static struct shared_info *shared_info_page = 0; 1632 1633 if (!shared_info_page) 1634 shared_info_page = (struct shared_info *) 1635 extend_brk(PAGE_SIZE, PAGE_SIZE); 1636 xatp.domid = DOMID_SELF; 1637 xatp.idx = 0; 1638 xatp.space = XENMAPSPACE_shared_info; 1639 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT; 1640 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 1641 BUG(); 1642 1643 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 1644 1645 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 1646 * page, we use it in the event channel upcall and in some pvclock 1647 * related functions. We don't need the vcpu_info placement 1648 * optimizations because we don't use any pv_mmu or pv_irq op on 1649 * HVM. 1650 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is 1651 * online but xen_hvm_init_shared_info is run at resume time too and 1652 * in that case multiple vcpus might be online. */ 1653 for_each_online_cpu(cpu) { 1654 /* Leave it to be NULL. */ 1655 if (cpu >= MAX_VIRT_CPUS) 1656 continue; 1657 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 1658 } 1659 } 1660 1661 #ifdef CONFIG_XEN_PVHVM 1662 static void __init init_hvm_pv_info(void) 1663 { 1664 int major, minor; 1665 uint32_t eax, ebx, ecx, edx, pages, msr, base; 1666 u64 pfn; 1667 1668 base = xen_cpuid_base(); 1669 cpuid(base + 1, &eax, &ebx, &ecx, &edx); 1670 1671 major = eax >> 16; 1672 minor = eax & 0xffff; 1673 printk(KERN_INFO "Xen version %d.%d.\n", major, minor); 1674 1675 cpuid(base + 2, &pages, &msr, &ecx, &edx); 1676 1677 pfn = __pa(hypercall_page); 1678 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32)); 1679 1680 xen_setup_features(); 1681 1682 pv_info.name = "Xen HVM"; 1683 1684 xen_domain_type = XEN_HVM_DOMAIN; 1685 } 1686 1687 static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action, 1688 void *hcpu) 1689 { 1690 int cpu = (long)hcpu; 1691 switch (action) { 1692 case CPU_UP_PREPARE: 1693 xen_vcpu_setup(cpu); 1694 if (xen_have_vector_callback) { 1695 if (xen_feature(XENFEAT_hvm_safe_pvclock)) 1696 xen_setup_timer(cpu); 1697 } 1698 break; 1699 default: 1700 break; 1701 } 1702 return NOTIFY_OK; 1703 } 1704 1705 static struct notifier_block xen_hvm_cpu_notifier = { 1706 .notifier_call = xen_hvm_cpu_notify, 1707 }; 1708 1709 static void __init xen_hvm_guest_init(void) 1710 { 1711 init_hvm_pv_info(); 1712 1713 xen_hvm_init_shared_info(); 1714 1715 xen_panic_handler_init(); 1716 1717 if (xen_feature(XENFEAT_hvm_callback_vector)) 1718 xen_have_vector_callback = 1; 1719 xen_hvm_smp_init(); 1720 register_cpu_notifier(&xen_hvm_cpu_notifier); 1721 xen_unplug_emulated_devices(); 1722 x86_init.irqs.intr_init = xen_init_IRQ; 1723 xen_hvm_init_time_ops(); 1724 xen_hvm_init_mmu_ops(); 1725 } 1726 1727 static uint32_t __init xen_hvm_platform(void) 1728 { 1729 if (xen_pv_domain()) 1730 return 0; 1731 1732 return xen_cpuid_base(); 1733 } 1734 1735 bool xen_hvm_need_lapic(void) 1736 { 1737 if (xen_pv_domain()) 1738 return false; 1739 if (!xen_hvm_domain()) 1740 return false; 1741 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback) 1742 return false; 1743 return true; 1744 } 1745 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic); 1746 1747 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = { 1748 .name = "Xen HVM", 1749 .detect = xen_hvm_platform, 1750 .init_platform = xen_hvm_guest_init, 1751 .x2apic_available = xen_x2apic_para_available, 1752 }; 1753 EXPORT_SYMBOL(x86_hyper_xen_hvm); 1754 #endif 1755