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_APIC) | /* disable local APIC */ 431 (1 << X86_FEATURE_ACPI)); /* disable ACPI */ 432 433 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32)); 434 435 ax = 1; 436 cx = 0; 437 xen_cpuid(&ax, &bx, &cx, &dx); 438 439 xsave_mask = 440 (1 << (X86_FEATURE_XSAVE % 32)) | 441 (1 << (X86_FEATURE_OSXSAVE % 32)); 442 443 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */ 444 if ((cx & xsave_mask) != xsave_mask) 445 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */ 446 if (xen_check_mwait()) 447 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32)); 448 } 449 450 static void xen_set_debugreg(int reg, unsigned long val) 451 { 452 HYPERVISOR_set_debugreg(reg, val); 453 } 454 455 static unsigned long xen_get_debugreg(int reg) 456 { 457 return HYPERVISOR_get_debugreg(reg); 458 } 459 460 static void xen_end_context_switch(struct task_struct *next) 461 { 462 xen_mc_flush(); 463 paravirt_end_context_switch(next); 464 } 465 466 static unsigned long xen_store_tr(void) 467 { 468 return 0; 469 } 470 471 /* 472 * Set the page permissions for a particular virtual address. If the 473 * address is a vmalloc mapping (or other non-linear mapping), then 474 * find the linear mapping of the page and also set its protections to 475 * match. 476 */ 477 static void set_aliased_prot(void *v, pgprot_t prot) 478 { 479 int level; 480 pte_t *ptep; 481 pte_t pte; 482 unsigned long pfn; 483 struct page *page; 484 485 ptep = lookup_address((unsigned long)v, &level); 486 BUG_ON(ptep == NULL); 487 488 pfn = pte_pfn(*ptep); 489 page = pfn_to_page(pfn); 490 491 pte = pfn_pte(pfn, prot); 492 493 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) 494 BUG(); 495 496 if (!PageHighMem(page)) { 497 void *av = __va(PFN_PHYS(pfn)); 498 499 if (av != v) 500 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) 501 BUG(); 502 } else 503 kmap_flush_unused(); 504 } 505 506 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) 507 { 508 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 509 int i; 510 511 for(i = 0; i < entries; i += entries_per_page) 512 set_aliased_prot(ldt + i, PAGE_KERNEL_RO); 513 } 514 515 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) 516 { 517 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 518 int i; 519 520 for(i = 0; i < entries; i += entries_per_page) 521 set_aliased_prot(ldt + i, PAGE_KERNEL); 522 } 523 524 static void xen_set_ldt(const void *addr, unsigned entries) 525 { 526 struct mmuext_op *op; 527 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 528 529 trace_xen_cpu_set_ldt(addr, entries); 530 531 op = mcs.args; 532 op->cmd = MMUEXT_SET_LDT; 533 op->arg1.linear_addr = (unsigned long)addr; 534 op->arg2.nr_ents = entries; 535 536 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 537 538 xen_mc_issue(PARAVIRT_LAZY_CPU); 539 } 540 541 static void xen_load_gdt(const struct desc_ptr *dtr) 542 { 543 unsigned long va = dtr->address; 544 unsigned int size = dtr->size + 1; 545 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 546 unsigned long frames[pages]; 547 int f; 548 549 /* 550 * A GDT can be up to 64k in size, which corresponds to 8192 551 * 8-byte entries, or 16 4k pages.. 552 */ 553 554 BUG_ON(size > 65536); 555 BUG_ON(va & ~PAGE_MASK); 556 557 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 558 int level; 559 pte_t *ptep; 560 unsigned long pfn, mfn; 561 void *virt; 562 563 /* 564 * The GDT is per-cpu and is in the percpu data area. 565 * That can be virtually mapped, so we need to do a 566 * page-walk to get the underlying MFN for the 567 * hypercall. The page can also be in the kernel's 568 * linear range, so we need to RO that mapping too. 569 */ 570 ptep = lookup_address(va, &level); 571 BUG_ON(ptep == NULL); 572 573 pfn = pte_pfn(*ptep); 574 mfn = pfn_to_mfn(pfn); 575 virt = __va(PFN_PHYS(pfn)); 576 577 frames[f] = mfn; 578 579 make_lowmem_page_readonly((void *)va); 580 make_lowmem_page_readonly(virt); 581 } 582 583 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 584 BUG(); 585 } 586 587 /* 588 * load_gdt for early boot, when the gdt is only mapped once 589 */ 590 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr) 591 { 592 unsigned long va = dtr->address; 593 unsigned int size = dtr->size + 1; 594 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 595 unsigned long frames[pages]; 596 int f; 597 598 /* 599 * A GDT can be up to 64k in size, which corresponds to 8192 600 * 8-byte entries, or 16 4k pages.. 601 */ 602 603 BUG_ON(size > 65536); 604 BUG_ON(va & ~PAGE_MASK); 605 606 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 607 pte_t pte; 608 unsigned long pfn, mfn; 609 610 pfn = virt_to_pfn(va); 611 mfn = pfn_to_mfn(pfn); 612 613 pte = pfn_pte(pfn, PAGE_KERNEL_RO); 614 615 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 616 BUG(); 617 618 frames[f] = mfn; 619 } 620 621 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 622 BUG(); 623 } 624 625 static inline bool desc_equal(const struct desc_struct *d1, 626 const struct desc_struct *d2) 627 { 628 return d1->a == d2->a && d1->b == d2->b; 629 } 630 631 static void load_TLS_descriptor(struct thread_struct *t, 632 unsigned int cpu, unsigned int i) 633 { 634 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i]; 635 struct desc_struct *gdt; 636 xmaddr_t maddr; 637 struct multicall_space mc; 638 639 if (desc_equal(shadow, &t->tls_array[i])) 640 return; 641 642 *shadow = t->tls_array[i]; 643 644 gdt = get_cpu_gdt_table(cpu); 645 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 646 mc = __xen_mc_entry(0); 647 648 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 649 } 650 651 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 652 { 653 /* 654 * XXX sleazy hack: If we're being called in a lazy-cpu zone 655 * and lazy gs handling is enabled, it means we're in a 656 * context switch, and %gs has just been saved. This means we 657 * can zero it out to prevent faults on exit from the 658 * hypervisor if the next process has no %gs. Either way, it 659 * has been saved, and the new value will get loaded properly. 660 * This will go away as soon as Xen has been modified to not 661 * save/restore %gs for normal hypercalls. 662 * 663 * On x86_64, this hack is not used for %gs, because gs points 664 * to KERNEL_GS_BASE (and uses it for PDA references), so we 665 * must not zero %gs on x86_64 666 * 667 * For x86_64, we need to zero %fs, otherwise we may get an 668 * exception between the new %fs descriptor being loaded and 669 * %fs being effectively cleared at __switch_to(). 670 */ 671 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { 672 #ifdef CONFIG_X86_32 673 lazy_load_gs(0); 674 #else 675 loadsegment(fs, 0); 676 #endif 677 } 678 679 xen_mc_batch(); 680 681 load_TLS_descriptor(t, cpu, 0); 682 load_TLS_descriptor(t, cpu, 1); 683 load_TLS_descriptor(t, cpu, 2); 684 685 xen_mc_issue(PARAVIRT_LAZY_CPU); 686 } 687 688 #ifdef CONFIG_X86_64 689 static void xen_load_gs_index(unsigned int idx) 690 { 691 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) 692 BUG(); 693 } 694 #endif 695 696 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 697 const void *ptr) 698 { 699 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); 700 u64 entry = *(u64 *)ptr; 701 702 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry); 703 704 preempt_disable(); 705 706 xen_mc_flush(); 707 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 708 BUG(); 709 710 preempt_enable(); 711 } 712 713 static int cvt_gate_to_trap(int vector, const gate_desc *val, 714 struct trap_info *info) 715 { 716 unsigned long addr; 717 718 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT) 719 return 0; 720 721 info->vector = vector; 722 723 addr = gate_offset(*val); 724 #ifdef CONFIG_X86_64 725 /* 726 * Look for known traps using IST, and substitute them 727 * appropriately. The debugger ones are the only ones we care 728 * about. Xen will handle faults like double_fault, 729 * so we should never see them. Warn if 730 * there's an unexpected IST-using fault handler. 731 */ 732 if (addr == (unsigned long)debug) 733 addr = (unsigned long)xen_debug; 734 else if (addr == (unsigned long)int3) 735 addr = (unsigned long)xen_int3; 736 else if (addr == (unsigned long)stack_segment) 737 addr = (unsigned long)xen_stack_segment; 738 else if (addr == (unsigned long)double_fault || 739 addr == (unsigned long)nmi) { 740 /* Don't need to handle these */ 741 return 0; 742 #ifdef CONFIG_X86_MCE 743 } else if (addr == (unsigned long)machine_check) { 744 /* 745 * when xen hypervisor inject vMCE to guest, 746 * use native mce handler to handle it 747 */ 748 ; 749 #endif 750 } else { 751 /* Some other trap using IST? */ 752 if (WARN_ON(val->ist != 0)) 753 return 0; 754 } 755 #endif /* CONFIG_X86_64 */ 756 info->address = addr; 757 758 info->cs = gate_segment(*val); 759 info->flags = val->dpl; 760 /* interrupt gates clear IF */ 761 if (val->type == GATE_INTERRUPT) 762 info->flags |= 1 << 2; 763 764 return 1; 765 } 766 767 /* Locations of each CPU's IDT */ 768 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 769 770 /* Set an IDT entry. If the entry is part of the current IDT, then 771 also update Xen. */ 772 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 773 { 774 unsigned long p = (unsigned long)&dt[entrynum]; 775 unsigned long start, end; 776 777 trace_xen_cpu_write_idt_entry(dt, entrynum, g); 778 779 preempt_disable(); 780 781 start = __this_cpu_read(idt_desc.address); 782 end = start + __this_cpu_read(idt_desc.size) + 1; 783 784 xen_mc_flush(); 785 786 native_write_idt_entry(dt, entrynum, g); 787 788 if (p >= start && (p + 8) <= end) { 789 struct trap_info info[2]; 790 791 info[1].address = 0; 792 793 if (cvt_gate_to_trap(entrynum, g, &info[0])) 794 if (HYPERVISOR_set_trap_table(info)) 795 BUG(); 796 } 797 798 preempt_enable(); 799 } 800 801 static void xen_convert_trap_info(const struct desc_ptr *desc, 802 struct trap_info *traps) 803 { 804 unsigned in, out, count; 805 806 count = (desc->size+1) / sizeof(gate_desc); 807 BUG_ON(count > 256); 808 809 for (in = out = 0; in < count; in++) { 810 gate_desc *entry = (gate_desc*)(desc->address) + in; 811 812 if (cvt_gate_to_trap(in, entry, &traps[out])) 813 out++; 814 } 815 traps[out].address = 0; 816 } 817 818 void xen_copy_trap_info(struct trap_info *traps) 819 { 820 const struct desc_ptr *desc = &__get_cpu_var(idt_desc); 821 822 xen_convert_trap_info(desc, traps); 823 } 824 825 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 826 hold a spinlock to protect the static traps[] array (static because 827 it avoids allocation, and saves stack space). */ 828 static void xen_load_idt(const struct desc_ptr *desc) 829 { 830 static DEFINE_SPINLOCK(lock); 831 static struct trap_info traps[257]; 832 833 trace_xen_cpu_load_idt(desc); 834 835 spin_lock(&lock); 836 837 __get_cpu_var(idt_desc) = *desc; 838 839 xen_convert_trap_info(desc, traps); 840 841 xen_mc_flush(); 842 if (HYPERVISOR_set_trap_table(traps)) 843 BUG(); 844 845 spin_unlock(&lock); 846 } 847 848 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 849 they're handled differently. */ 850 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 851 const void *desc, int type) 852 { 853 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 854 855 preempt_disable(); 856 857 switch (type) { 858 case DESC_LDT: 859 case DESC_TSS: 860 /* ignore */ 861 break; 862 863 default: { 864 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); 865 866 xen_mc_flush(); 867 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 868 BUG(); 869 } 870 871 } 872 873 preempt_enable(); 874 } 875 876 /* 877 * Version of write_gdt_entry for use at early boot-time needed to 878 * update an entry as simply as possible. 879 */ 880 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, 881 const void *desc, int type) 882 { 883 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 884 885 switch (type) { 886 case DESC_LDT: 887 case DESC_TSS: 888 /* ignore */ 889 break; 890 891 default: { 892 xmaddr_t maddr = virt_to_machine(&dt[entry]); 893 894 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 895 dt[entry] = *(struct desc_struct *)desc; 896 } 897 898 } 899 } 900 901 static void xen_load_sp0(struct tss_struct *tss, 902 struct thread_struct *thread) 903 { 904 struct multicall_space mcs; 905 906 mcs = xen_mc_entry(0); 907 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 908 xen_mc_issue(PARAVIRT_LAZY_CPU); 909 } 910 911 static void xen_set_iopl_mask(unsigned mask) 912 { 913 struct physdev_set_iopl set_iopl; 914 915 /* Force the change at ring 0. */ 916 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; 917 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 918 } 919 920 static void xen_io_delay(void) 921 { 922 } 923 924 #ifdef CONFIG_X86_LOCAL_APIC 925 static unsigned long xen_set_apic_id(unsigned int x) 926 { 927 WARN_ON(1); 928 return x; 929 } 930 static unsigned int xen_get_apic_id(unsigned long x) 931 { 932 return ((x)>>24) & 0xFFu; 933 } 934 static u32 xen_apic_read(u32 reg) 935 { 936 struct xen_platform_op op = { 937 .cmd = XENPF_get_cpuinfo, 938 .interface_version = XENPF_INTERFACE_VERSION, 939 .u.pcpu_info.xen_cpuid = 0, 940 }; 941 int ret = 0; 942 943 /* Shouldn't need this as APIC is turned off for PV, and we only 944 * get called on the bootup processor. But just in case. */ 945 if (!xen_initial_domain() || smp_processor_id()) 946 return 0; 947 948 if (reg == APIC_LVR) 949 return 0x10; 950 951 if (reg != APIC_ID) 952 return 0; 953 954 ret = HYPERVISOR_dom0_op(&op); 955 if (ret) 956 return 0; 957 958 return op.u.pcpu_info.apic_id << 24; 959 } 960 961 static void xen_apic_write(u32 reg, u32 val) 962 { 963 /* Warn to see if there's any stray references */ 964 WARN_ON(1); 965 } 966 967 static u64 xen_apic_icr_read(void) 968 { 969 return 0; 970 } 971 972 static void xen_apic_icr_write(u32 low, u32 id) 973 { 974 /* Warn to see if there's any stray references */ 975 WARN_ON(1); 976 } 977 978 static void xen_apic_wait_icr_idle(void) 979 { 980 return; 981 } 982 983 static u32 xen_safe_apic_wait_icr_idle(void) 984 { 985 return 0; 986 } 987 988 static void set_xen_basic_apic_ops(void) 989 { 990 apic->read = xen_apic_read; 991 apic->write = xen_apic_write; 992 apic->icr_read = xen_apic_icr_read; 993 apic->icr_write = xen_apic_icr_write; 994 apic->wait_icr_idle = xen_apic_wait_icr_idle; 995 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; 996 apic->set_apic_id = xen_set_apic_id; 997 apic->get_apic_id = xen_get_apic_id; 998 999 #ifdef CONFIG_SMP 1000 apic->send_IPI_allbutself = xen_send_IPI_allbutself; 1001 apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself; 1002 apic->send_IPI_mask = xen_send_IPI_mask; 1003 apic->send_IPI_all = xen_send_IPI_all; 1004 apic->send_IPI_self = xen_send_IPI_self; 1005 #endif 1006 } 1007 1008 #endif 1009 1010 static void xen_clts(void) 1011 { 1012 struct multicall_space mcs; 1013 1014 mcs = xen_mc_entry(0); 1015 1016 MULTI_fpu_taskswitch(mcs.mc, 0); 1017 1018 xen_mc_issue(PARAVIRT_LAZY_CPU); 1019 } 1020 1021 static DEFINE_PER_CPU(unsigned long, xen_cr0_value); 1022 1023 static unsigned long xen_read_cr0(void) 1024 { 1025 unsigned long cr0 = this_cpu_read(xen_cr0_value); 1026 1027 if (unlikely(cr0 == 0)) { 1028 cr0 = native_read_cr0(); 1029 this_cpu_write(xen_cr0_value, cr0); 1030 } 1031 1032 return cr0; 1033 } 1034 1035 static void xen_write_cr0(unsigned long cr0) 1036 { 1037 struct multicall_space mcs; 1038 1039 this_cpu_write(xen_cr0_value, cr0); 1040 1041 /* Only pay attention to cr0.TS; everything else is 1042 ignored. */ 1043 mcs = xen_mc_entry(0); 1044 1045 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); 1046 1047 xen_mc_issue(PARAVIRT_LAZY_CPU); 1048 } 1049 1050 static void xen_write_cr4(unsigned long cr4) 1051 { 1052 cr4 &= ~X86_CR4_PGE; 1053 cr4 &= ~X86_CR4_PSE; 1054 1055 native_write_cr4(cr4); 1056 } 1057 #ifdef CONFIG_X86_64 1058 static inline unsigned long xen_read_cr8(void) 1059 { 1060 return 0; 1061 } 1062 static inline void xen_write_cr8(unsigned long val) 1063 { 1064 BUG_ON(val); 1065 } 1066 #endif 1067 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) 1068 { 1069 int ret; 1070 1071 ret = 0; 1072 1073 switch (msr) { 1074 #ifdef CONFIG_X86_64 1075 unsigned which; 1076 u64 base; 1077 1078 case MSR_FS_BASE: which = SEGBASE_FS; goto set; 1079 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; 1080 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; 1081 1082 set: 1083 base = ((u64)high << 32) | low; 1084 if (HYPERVISOR_set_segment_base(which, base) != 0) 1085 ret = -EIO; 1086 break; 1087 #endif 1088 1089 case MSR_STAR: 1090 case MSR_CSTAR: 1091 case MSR_LSTAR: 1092 case MSR_SYSCALL_MASK: 1093 case MSR_IA32_SYSENTER_CS: 1094 case MSR_IA32_SYSENTER_ESP: 1095 case MSR_IA32_SYSENTER_EIP: 1096 /* Fast syscall setup is all done in hypercalls, so 1097 these are all ignored. Stub them out here to stop 1098 Xen console noise. */ 1099 break; 1100 1101 case MSR_IA32_CR_PAT: 1102 if (smp_processor_id() == 0) 1103 xen_set_pat(((u64)high << 32) | low); 1104 break; 1105 1106 default: 1107 ret = native_write_msr_safe(msr, low, high); 1108 } 1109 1110 return ret; 1111 } 1112 1113 void xen_setup_shared_info(void) 1114 { 1115 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 1116 set_fixmap(FIX_PARAVIRT_BOOTMAP, 1117 xen_start_info->shared_info); 1118 1119 HYPERVISOR_shared_info = 1120 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); 1121 } else 1122 HYPERVISOR_shared_info = 1123 (struct shared_info *)__va(xen_start_info->shared_info); 1124 1125 #ifndef CONFIG_SMP 1126 /* In UP this is as good a place as any to set up shared info */ 1127 xen_setup_vcpu_info_placement(); 1128 #endif 1129 1130 xen_setup_mfn_list_list(); 1131 } 1132 1133 /* This is called once we have the cpu_possible_mask */ 1134 void xen_setup_vcpu_info_placement(void) 1135 { 1136 int cpu; 1137 1138 for_each_possible_cpu(cpu) 1139 xen_vcpu_setup(cpu); 1140 1141 /* xen_vcpu_setup managed to place the vcpu_info within the 1142 percpu area for all cpus, so make use of it */ 1143 if (have_vcpu_info_placement) { 1144 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); 1145 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); 1146 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); 1147 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); 1148 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 1149 } 1150 } 1151 1152 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 1153 unsigned long addr, unsigned len) 1154 { 1155 char *start, *end, *reloc; 1156 unsigned ret; 1157 1158 start = end = reloc = NULL; 1159 1160 #define SITE(op, x) \ 1161 case PARAVIRT_PATCH(op.x): \ 1162 if (have_vcpu_info_placement) { \ 1163 start = (char *)xen_##x##_direct; \ 1164 end = xen_##x##_direct_end; \ 1165 reloc = xen_##x##_direct_reloc; \ 1166 } \ 1167 goto patch_site 1168 1169 switch (type) { 1170 SITE(pv_irq_ops, irq_enable); 1171 SITE(pv_irq_ops, irq_disable); 1172 SITE(pv_irq_ops, save_fl); 1173 SITE(pv_irq_ops, restore_fl); 1174 #undef SITE 1175 1176 patch_site: 1177 if (start == NULL || (end-start) > len) 1178 goto default_patch; 1179 1180 ret = paravirt_patch_insns(insnbuf, len, start, end); 1181 1182 /* Note: because reloc is assigned from something that 1183 appears to be an array, gcc assumes it's non-null, 1184 but doesn't know its relationship with start and 1185 end. */ 1186 if (reloc > start && reloc < end) { 1187 int reloc_off = reloc - start; 1188 long *relocp = (long *)(insnbuf + reloc_off); 1189 long delta = start - (char *)addr; 1190 1191 *relocp += delta; 1192 } 1193 break; 1194 1195 default_patch: 1196 default: 1197 ret = paravirt_patch_default(type, clobbers, insnbuf, 1198 addr, len); 1199 break; 1200 } 1201 1202 return ret; 1203 } 1204 1205 static const struct pv_info xen_info __initconst = { 1206 .paravirt_enabled = 1, 1207 .shared_kernel_pmd = 0, 1208 1209 #ifdef CONFIG_X86_64 1210 .extra_user_64bit_cs = FLAT_USER_CS64, 1211 #endif 1212 1213 .name = "Xen", 1214 }; 1215 1216 static const struct pv_init_ops xen_init_ops __initconst = { 1217 .patch = xen_patch, 1218 }; 1219 1220 static const struct pv_cpu_ops xen_cpu_ops __initconst = { 1221 .cpuid = xen_cpuid, 1222 1223 .set_debugreg = xen_set_debugreg, 1224 .get_debugreg = xen_get_debugreg, 1225 1226 .clts = xen_clts, 1227 1228 .read_cr0 = xen_read_cr0, 1229 .write_cr0 = xen_write_cr0, 1230 1231 .read_cr4 = native_read_cr4, 1232 .read_cr4_safe = native_read_cr4_safe, 1233 .write_cr4 = xen_write_cr4, 1234 1235 #ifdef CONFIG_X86_64 1236 .read_cr8 = xen_read_cr8, 1237 .write_cr8 = xen_write_cr8, 1238 #endif 1239 1240 .wbinvd = native_wbinvd, 1241 1242 .read_msr = native_read_msr_safe, 1243 .write_msr = xen_write_msr_safe, 1244 1245 .read_tsc = native_read_tsc, 1246 .read_pmc = native_read_pmc, 1247 1248 .read_tscp = native_read_tscp, 1249 1250 .iret = xen_iret, 1251 .irq_enable_sysexit = xen_sysexit, 1252 #ifdef CONFIG_X86_64 1253 .usergs_sysret32 = xen_sysret32, 1254 .usergs_sysret64 = xen_sysret64, 1255 #endif 1256 1257 .load_tr_desc = paravirt_nop, 1258 .set_ldt = xen_set_ldt, 1259 .load_gdt = xen_load_gdt, 1260 .load_idt = xen_load_idt, 1261 .load_tls = xen_load_tls, 1262 #ifdef CONFIG_X86_64 1263 .load_gs_index = xen_load_gs_index, 1264 #endif 1265 1266 .alloc_ldt = xen_alloc_ldt, 1267 .free_ldt = xen_free_ldt, 1268 1269 .store_idt = native_store_idt, 1270 .store_tr = xen_store_tr, 1271 1272 .write_ldt_entry = xen_write_ldt_entry, 1273 .write_gdt_entry = xen_write_gdt_entry, 1274 .write_idt_entry = xen_write_idt_entry, 1275 .load_sp0 = xen_load_sp0, 1276 1277 .set_iopl_mask = xen_set_iopl_mask, 1278 .io_delay = xen_io_delay, 1279 1280 /* Xen takes care of %gs when switching to usermode for us */ 1281 .swapgs = paravirt_nop, 1282 1283 .start_context_switch = paravirt_start_context_switch, 1284 .end_context_switch = xen_end_context_switch, 1285 }; 1286 1287 static const struct pv_apic_ops xen_apic_ops __initconst = { 1288 #ifdef CONFIG_X86_LOCAL_APIC 1289 .startup_ipi_hook = paravirt_nop, 1290 #endif 1291 }; 1292 1293 static void xen_reboot(int reason) 1294 { 1295 struct sched_shutdown r = { .reason = reason }; 1296 1297 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) 1298 BUG(); 1299 } 1300 1301 static void xen_restart(char *msg) 1302 { 1303 xen_reboot(SHUTDOWN_reboot); 1304 } 1305 1306 static void xen_emergency_restart(void) 1307 { 1308 xen_reboot(SHUTDOWN_reboot); 1309 } 1310 1311 static void xen_machine_halt(void) 1312 { 1313 xen_reboot(SHUTDOWN_poweroff); 1314 } 1315 1316 static void xen_machine_power_off(void) 1317 { 1318 if (pm_power_off) 1319 pm_power_off(); 1320 xen_reboot(SHUTDOWN_poweroff); 1321 } 1322 1323 static void xen_crash_shutdown(struct pt_regs *regs) 1324 { 1325 xen_reboot(SHUTDOWN_crash); 1326 } 1327 1328 static int 1329 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr) 1330 { 1331 xen_reboot(SHUTDOWN_crash); 1332 return NOTIFY_DONE; 1333 } 1334 1335 static struct notifier_block xen_panic_block = { 1336 .notifier_call= xen_panic_event, 1337 }; 1338 1339 int xen_panic_handler_init(void) 1340 { 1341 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block); 1342 return 0; 1343 } 1344 1345 static const struct machine_ops xen_machine_ops __initconst = { 1346 .restart = xen_restart, 1347 .halt = xen_machine_halt, 1348 .power_off = xen_machine_power_off, 1349 .shutdown = xen_machine_halt, 1350 .crash_shutdown = xen_crash_shutdown, 1351 .emergency_restart = xen_emergency_restart, 1352 }; 1353 1354 static void __init xen_boot_params_init_edd(void) 1355 { 1356 #if IS_ENABLED(CONFIG_EDD) 1357 struct xen_platform_op op; 1358 struct edd_info *edd_info; 1359 u32 *mbr_signature; 1360 unsigned nr; 1361 int ret; 1362 1363 edd_info = boot_params.eddbuf; 1364 mbr_signature = boot_params.edd_mbr_sig_buffer; 1365 1366 op.cmd = XENPF_firmware_info; 1367 1368 op.u.firmware_info.type = XEN_FW_DISK_INFO; 1369 for (nr = 0; nr < EDDMAXNR; nr++) { 1370 struct edd_info *info = edd_info + nr; 1371 1372 op.u.firmware_info.index = nr; 1373 info->params.length = sizeof(info->params); 1374 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params, 1375 &info->params); 1376 ret = HYPERVISOR_dom0_op(&op); 1377 if (ret) 1378 break; 1379 1380 #define C(x) info->x = op.u.firmware_info.u.disk_info.x 1381 C(device); 1382 C(version); 1383 C(interface_support); 1384 C(legacy_max_cylinder); 1385 C(legacy_max_head); 1386 C(legacy_sectors_per_track); 1387 #undef C 1388 } 1389 boot_params.eddbuf_entries = nr; 1390 1391 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE; 1392 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) { 1393 op.u.firmware_info.index = nr; 1394 ret = HYPERVISOR_dom0_op(&op); 1395 if (ret) 1396 break; 1397 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature; 1398 } 1399 boot_params.edd_mbr_sig_buf_entries = nr; 1400 #endif 1401 } 1402 1403 /* 1404 * Set up the GDT and segment registers for -fstack-protector. Until 1405 * we do this, we have to be careful not to call any stack-protected 1406 * function, which is most of the kernel. 1407 */ 1408 static void __init xen_setup_stackprotector(void) 1409 { 1410 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; 1411 pv_cpu_ops.load_gdt = xen_load_gdt_boot; 1412 1413 setup_stack_canary_segment(0); 1414 switch_to_new_gdt(0); 1415 1416 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; 1417 pv_cpu_ops.load_gdt = xen_load_gdt; 1418 } 1419 1420 /* First C function to be called on Xen boot */ 1421 asmlinkage void __init xen_start_kernel(void) 1422 { 1423 struct physdev_set_iopl set_iopl; 1424 int rc; 1425 1426 if (!xen_start_info) 1427 return; 1428 1429 xen_domain_type = XEN_PV_DOMAIN; 1430 1431 xen_setup_machphys_mapping(); 1432 1433 /* Install Xen paravirt ops */ 1434 pv_info = xen_info; 1435 pv_init_ops = xen_init_ops; 1436 pv_cpu_ops = xen_cpu_ops; 1437 pv_apic_ops = xen_apic_ops; 1438 1439 x86_init.resources.memory_setup = xen_memory_setup; 1440 x86_init.oem.arch_setup = xen_arch_setup; 1441 x86_init.oem.banner = xen_banner; 1442 1443 xen_init_time_ops(); 1444 1445 /* 1446 * Set up some pagetable state before starting to set any ptes. 1447 */ 1448 1449 xen_init_mmu_ops(); 1450 1451 /* Prevent unwanted bits from being set in PTEs. */ 1452 __supported_pte_mask &= ~_PAGE_GLOBAL; 1453 #if 0 1454 if (!xen_initial_domain()) 1455 #endif 1456 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); 1457 1458 __supported_pte_mask |= _PAGE_IOMAP; 1459 1460 /* 1461 * Prevent page tables from being allocated in highmem, even 1462 * if CONFIG_HIGHPTE is enabled. 1463 */ 1464 __userpte_alloc_gfp &= ~__GFP_HIGHMEM; 1465 1466 /* Work out if we support NX */ 1467 x86_configure_nx(); 1468 1469 xen_setup_features(); 1470 1471 /* Get mfn list */ 1472 if (!xen_feature(XENFEAT_auto_translated_physmap)) 1473 xen_build_dynamic_phys_to_machine(); 1474 1475 /* 1476 * Set up kernel GDT and segment registers, mainly so that 1477 * -fstack-protector code can be executed. 1478 */ 1479 xen_setup_stackprotector(); 1480 1481 xen_init_irq_ops(); 1482 xen_init_cpuid_mask(); 1483 1484 #ifdef CONFIG_X86_LOCAL_APIC 1485 /* 1486 * set up the basic apic ops. 1487 */ 1488 set_xen_basic_apic_ops(); 1489 #endif 1490 1491 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { 1492 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; 1493 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; 1494 } 1495 1496 machine_ops = xen_machine_ops; 1497 1498 /* 1499 * The only reliable way to retain the initial address of the 1500 * percpu gdt_page is to remember it here, so we can go and 1501 * mark it RW later, when the initial percpu area is freed. 1502 */ 1503 xen_initial_gdt = &per_cpu(gdt_page, 0); 1504 1505 xen_smp_init(); 1506 1507 #ifdef CONFIG_ACPI_NUMA 1508 /* 1509 * The pages we from Xen are not related to machine pages, so 1510 * any NUMA information the kernel tries to get from ACPI will 1511 * be meaningless. Prevent it from trying. 1512 */ 1513 acpi_numa = -1; 1514 #endif 1515 #ifdef CONFIG_X86_PAT 1516 /* 1517 * For right now disable the PAT. We should remove this once 1518 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1 1519 * (xen/pat: Disable PAT support for now) is reverted. 1520 */ 1521 pat_enabled = 0; 1522 #endif 1523 /* Don't do the full vcpu_info placement stuff until we have a 1524 possible map and a non-dummy shared_info. */ 1525 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; 1526 1527 local_irq_disable(); 1528 early_boot_irqs_disabled = true; 1529 1530 xen_raw_console_write("mapping kernel into physical memory\n"); 1531 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, xen_start_info->nr_pages); 1532 1533 /* Allocate and initialize top and mid mfn levels for p2m structure */ 1534 xen_build_mfn_list_list(); 1535 1536 /* keep using Xen gdt for now; no urgent need to change it */ 1537 1538 #ifdef CONFIG_X86_32 1539 pv_info.kernel_rpl = 1; 1540 if (xen_feature(XENFEAT_supervisor_mode_kernel)) 1541 pv_info.kernel_rpl = 0; 1542 #else 1543 pv_info.kernel_rpl = 0; 1544 #endif 1545 /* set the limit of our address space */ 1546 xen_reserve_top(); 1547 1548 /* We used to do this in xen_arch_setup, but that is too late on AMD 1549 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init 1550 * which pokes 0xcf8 port. 1551 */ 1552 set_iopl.iopl = 1; 1553 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 1554 if (rc != 0) 1555 xen_raw_printk("physdev_op failed %d\n", rc); 1556 1557 #ifdef CONFIG_X86_32 1558 /* set up basic CPUID stuff */ 1559 cpu_detect(&new_cpu_data); 1560 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU); 1561 new_cpu_data.wp_works_ok = 1; 1562 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1563 #endif 1564 1565 /* Poke various useful things into boot_params */ 1566 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1567 boot_params.hdr.ramdisk_image = xen_start_info->mod_start 1568 ? __pa(xen_start_info->mod_start) : 0; 1569 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1570 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1571 1572 if (!xen_initial_domain()) { 1573 add_preferred_console("xenboot", 0, NULL); 1574 add_preferred_console("tty", 0, NULL); 1575 add_preferred_console("hvc", 0, NULL); 1576 if (pci_xen) 1577 x86_init.pci.arch_init = pci_xen_init; 1578 } else { 1579 const struct dom0_vga_console_info *info = 1580 (void *)((char *)xen_start_info + 1581 xen_start_info->console.dom0.info_off); 1582 struct xen_platform_op op = { 1583 .cmd = XENPF_firmware_info, 1584 .interface_version = XENPF_INTERFACE_VERSION, 1585 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS, 1586 }; 1587 1588 xen_init_vga(info, xen_start_info->console.dom0.info_size); 1589 xen_start_info->console.domU.mfn = 0; 1590 xen_start_info->console.domU.evtchn = 0; 1591 1592 if (HYPERVISOR_dom0_op(&op) == 0) 1593 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags; 1594 1595 xen_init_apic(); 1596 1597 /* Make sure ACS will be enabled */ 1598 pci_request_acs(); 1599 1600 xen_acpi_sleep_register(); 1601 1602 /* Avoid searching for BIOS MP tables */ 1603 x86_init.mpparse.find_smp_config = x86_init_noop; 1604 x86_init.mpparse.get_smp_config = x86_init_uint_noop; 1605 1606 xen_boot_params_init_edd(); 1607 } 1608 #ifdef CONFIG_PCI 1609 /* PCI BIOS service won't work from a PV guest. */ 1610 pci_probe &= ~PCI_PROBE_BIOS; 1611 #endif 1612 xen_raw_console_write("about to get started...\n"); 1613 1614 xen_setup_runstate_info(0); 1615 1616 /* Start the world */ 1617 #ifdef CONFIG_X86_32 1618 i386_start_kernel(); 1619 #else 1620 x86_64_start_reservations((char *)__pa_symbol(&boot_params)); 1621 #endif 1622 } 1623 1624 void __ref xen_hvm_init_shared_info(void) 1625 { 1626 int cpu; 1627 struct xen_add_to_physmap xatp; 1628 static struct shared_info *shared_info_page = 0; 1629 1630 if (!shared_info_page) 1631 shared_info_page = (struct shared_info *) 1632 extend_brk(PAGE_SIZE, PAGE_SIZE); 1633 xatp.domid = DOMID_SELF; 1634 xatp.idx = 0; 1635 xatp.space = XENMAPSPACE_shared_info; 1636 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT; 1637 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 1638 BUG(); 1639 1640 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page; 1641 1642 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 1643 * page, we use it in the event channel upcall and in some pvclock 1644 * related functions. We don't need the vcpu_info placement 1645 * optimizations because we don't use any pv_mmu or pv_irq op on 1646 * HVM. 1647 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is 1648 * online but xen_hvm_init_shared_info is run at resume time too and 1649 * in that case multiple vcpus might be online. */ 1650 for_each_online_cpu(cpu) { 1651 /* Leave it to be NULL. */ 1652 if (cpu >= MAX_VIRT_CPUS) 1653 continue; 1654 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 1655 } 1656 } 1657 1658 #ifdef CONFIG_XEN_PVHVM 1659 static void __init init_hvm_pv_info(void) 1660 { 1661 int major, minor; 1662 uint32_t eax, ebx, ecx, edx, pages, msr, base; 1663 u64 pfn; 1664 1665 base = xen_cpuid_base(); 1666 cpuid(base + 1, &eax, &ebx, &ecx, &edx); 1667 1668 major = eax >> 16; 1669 minor = eax & 0xffff; 1670 printk(KERN_INFO "Xen version %d.%d.\n", major, minor); 1671 1672 cpuid(base + 2, &pages, &msr, &ecx, &edx); 1673 1674 pfn = __pa(hypercall_page); 1675 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32)); 1676 1677 xen_setup_features(); 1678 1679 pv_info.name = "Xen HVM"; 1680 1681 xen_domain_type = XEN_HVM_DOMAIN; 1682 } 1683 1684 static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action, 1685 void *hcpu) 1686 { 1687 int cpu = (long)hcpu; 1688 switch (action) { 1689 case CPU_UP_PREPARE: 1690 xen_vcpu_setup(cpu); 1691 if (xen_have_vector_callback) { 1692 xen_init_lock_cpu(cpu); 1693 if (xen_feature(XENFEAT_hvm_safe_pvclock)) 1694 xen_setup_timer(cpu); 1695 } 1696 break; 1697 default: 1698 break; 1699 } 1700 return NOTIFY_OK; 1701 } 1702 1703 static struct notifier_block xen_hvm_cpu_notifier = { 1704 .notifier_call = xen_hvm_cpu_notify, 1705 }; 1706 1707 static void __init xen_hvm_guest_init(void) 1708 { 1709 init_hvm_pv_info(); 1710 1711 xen_hvm_init_shared_info(); 1712 1713 if (xen_feature(XENFEAT_hvm_callback_vector)) 1714 xen_have_vector_callback = 1; 1715 xen_hvm_smp_init(); 1716 register_cpu_notifier(&xen_hvm_cpu_notifier); 1717 xen_unplug_emulated_devices(); 1718 x86_init.irqs.intr_init = xen_init_IRQ; 1719 xen_hvm_init_time_ops(); 1720 xen_hvm_init_mmu_ops(); 1721 } 1722 1723 static bool __init xen_hvm_platform(void) 1724 { 1725 if (xen_pv_domain()) 1726 return false; 1727 1728 if (!xen_cpuid_base()) 1729 return false; 1730 1731 return true; 1732 } 1733 1734 bool xen_hvm_need_lapic(void) 1735 { 1736 if (xen_pv_domain()) 1737 return false; 1738 if (!xen_hvm_domain()) 1739 return false; 1740 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback) 1741 return false; 1742 return true; 1743 } 1744 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic); 1745 1746 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = { 1747 .name = "Xen HVM", 1748 .detect = xen_hvm_platform, 1749 .init_platform = xen_hvm_guest_init, 1750 .x2apic_available = xen_x2apic_para_available, 1751 }; 1752 EXPORT_SYMBOL(x86_hyper_xen_hvm); 1753 #endif 1754