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/kernel.h> 15 #include <linux/init.h> 16 #include <linux/smp.h> 17 #include <linux/preempt.h> 18 #include <linux/hardirq.h> 19 #include <linux/percpu.h> 20 #include <linux/delay.h> 21 #include <linux/start_kernel.h> 22 #include <linux/sched.h> 23 #include <linux/kprobes.h> 24 #include <linux/bootmem.h> 25 #include <linux/module.h> 26 #include <linux/mm.h> 27 #include <linux/page-flags.h> 28 #include <linux/highmem.h> 29 #include <linux/console.h> 30 31 #include <xen/interface/xen.h> 32 #include <xen/interface/version.h> 33 #include <xen/interface/physdev.h> 34 #include <xen/interface/vcpu.h> 35 #include <xen/features.h> 36 #include <xen/page.h> 37 #include <xen/hvc-console.h> 38 39 #include <asm/paravirt.h> 40 #include <asm/apic.h> 41 #include <asm/page.h> 42 #include <asm/xen/hypercall.h> 43 #include <asm/xen/hypervisor.h> 44 #include <asm/fixmap.h> 45 #include <asm/processor.h> 46 #include <asm/proto.h> 47 #include <asm/msr-index.h> 48 #include <asm/traps.h> 49 #include <asm/setup.h> 50 #include <asm/desc.h> 51 #include <asm/pgtable.h> 52 #include <asm/tlbflush.h> 53 #include <asm/reboot.h> 54 #include <asm/stackprotector.h> 55 56 #include "xen-ops.h" 57 #include "mmu.h" 58 #include "multicalls.h" 59 60 EXPORT_SYMBOL_GPL(hypercall_page); 61 62 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 63 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 64 65 enum xen_domain_type xen_domain_type = XEN_NATIVE; 66 EXPORT_SYMBOL_GPL(xen_domain_type); 67 68 struct start_info *xen_start_info; 69 EXPORT_SYMBOL_GPL(xen_start_info); 70 71 struct shared_info xen_dummy_shared_info; 72 73 void *xen_initial_gdt; 74 75 /* 76 * Point at some empty memory to start with. We map the real shared_info 77 * page as soon as fixmap is up and running. 78 */ 79 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 80 81 /* 82 * Flag to determine whether vcpu info placement is available on all 83 * VCPUs. We assume it is to start with, and then set it to zero on 84 * the first failure. This is because it can succeed on some VCPUs 85 * and not others, since it can involve hypervisor memory allocation, 86 * or because the guest failed to guarantee all the appropriate 87 * constraints on all VCPUs (ie buffer can't cross a page boundary). 88 * 89 * Note that any particular CPU may be using a placed vcpu structure, 90 * but we can only optimise if the all are. 91 * 92 * 0: not available, 1: available 93 */ 94 static int have_vcpu_info_placement = 1; 95 96 static void xen_vcpu_setup(int cpu) 97 { 98 struct vcpu_register_vcpu_info info; 99 int err; 100 struct vcpu_info *vcpup; 101 102 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); 103 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 104 105 if (!have_vcpu_info_placement) 106 return; /* already tested, not available */ 107 108 vcpup = &per_cpu(xen_vcpu_info, cpu); 109 110 info.mfn = arbitrary_virt_to_mfn(vcpup); 111 info.offset = offset_in_page(vcpup); 112 113 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", 114 cpu, vcpup, info.mfn, info.offset); 115 116 /* Check to see if the hypervisor will put the vcpu_info 117 structure where we want it, which allows direct access via 118 a percpu-variable. */ 119 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 120 121 if (err) { 122 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); 123 have_vcpu_info_placement = 0; 124 } else { 125 /* This cpu is using the registered vcpu info, even if 126 later ones fail to. */ 127 per_cpu(xen_vcpu, cpu) = vcpup; 128 129 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", 130 cpu, vcpup); 131 } 132 } 133 134 /* 135 * On restore, set the vcpu placement up again. 136 * If it fails, then we're in a bad state, since 137 * we can't back out from using it... 138 */ 139 void xen_vcpu_restore(void) 140 { 141 if (have_vcpu_info_placement) { 142 int cpu; 143 144 for_each_online_cpu(cpu) { 145 bool other_cpu = (cpu != smp_processor_id()); 146 147 if (other_cpu && 148 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) 149 BUG(); 150 151 xen_vcpu_setup(cpu); 152 153 if (other_cpu && 154 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) 155 BUG(); 156 } 157 158 BUG_ON(!have_vcpu_info_placement); 159 } 160 } 161 162 static void __init xen_banner(void) 163 { 164 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); 165 struct xen_extraversion extra; 166 HYPERVISOR_xen_version(XENVER_extraversion, &extra); 167 168 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 169 pv_info.name); 170 printk(KERN_INFO "Xen version: %d.%d%s%s\n", 171 version >> 16, version & 0xffff, extra.extraversion, 172 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); 173 } 174 175 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0; 176 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0; 177 178 static void xen_cpuid(unsigned int *ax, unsigned int *bx, 179 unsigned int *cx, unsigned int *dx) 180 { 181 unsigned maskecx = ~0; 182 unsigned maskedx = ~0; 183 184 /* 185 * Mask out inconvenient features, to try and disable as many 186 * unsupported kernel subsystems as possible. 187 */ 188 if (*ax == 1) { 189 maskecx = cpuid_leaf1_ecx_mask; 190 maskedx = cpuid_leaf1_edx_mask; 191 } 192 193 asm(XEN_EMULATE_PREFIX "cpuid" 194 : "=a" (*ax), 195 "=b" (*bx), 196 "=c" (*cx), 197 "=d" (*dx) 198 : "0" (*ax), "2" (*cx)); 199 200 *cx &= maskecx; 201 *dx &= maskedx; 202 } 203 204 static __init void xen_init_cpuid_mask(void) 205 { 206 unsigned int ax, bx, cx, dx; 207 208 cpuid_leaf1_edx_mask = 209 ~((1 << X86_FEATURE_MCE) | /* disable MCE */ 210 (1 << X86_FEATURE_MCA) | /* disable MCA */ 211 (1 << X86_FEATURE_ACC)); /* thermal monitoring */ 212 213 if (!xen_initial_domain()) 214 cpuid_leaf1_edx_mask &= 215 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */ 216 (1 << X86_FEATURE_ACPI)); /* disable ACPI */ 217 218 ax = 1; 219 cx = 0; 220 xen_cpuid(&ax, &bx, &cx, &dx); 221 222 /* cpuid claims we support xsave; try enabling it to see what happens */ 223 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) { 224 unsigned long cr4; 225 226 set_in_cr4(X86_CR4_OSXSAVE); 227 228 cr4 = read_cr4(); 229 230 if ((cr4 & X86_CR4_OSXSAVE) == 0) 231 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32)); 232 233 clear_in_cr4(X86_CR4_OSXSAVE); 234 } 235 } 236 237 static void xen_set_debugreg(int reg, unsigned long val) 238 { 239 HYPERVISOR_set_debugreg(reg, val); 240 } 241 242 static unsigned long xen_get_debugreg(int reg) 243 { 244 return HYPERVISOR_get_debugreg(reg); 245 } 246 247 static void xen_end_context_switch(struct task_struct *next) 248 { 249 xen_mc_flush(); 250 paravirt_end_context_switch(next); 251 } 252 253 static unsigned long xen_store_tr(void) 254 { 255 return 0; 256 } 257 258 /* 259 * Set the page permissions for a particular virtual address. If the 260 * address is a vmalloc mapping (or other non-linear mapping), then 261 * find the linear mapping of the page and also set its protections to 262 * match. 263 */ 264 static void set_aliased_prot(void *v, pgprot_t prot) 265 { 266 int level; 267 pte_t *ptep; 268 pte_t pte; 269 unsigned long pfn; 270 struct page *page; 271 272 ptep = lookup_address((unsigned long)v, &level); 273 BUG_ON(ptep == NULL); 274 275 pfn = pte_pfn(*ptep); 276 page = pfn_to_page(pfn); 277 278 pte = pfn_pte(pfn, prot); 279 280 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) 281 BUG(); 282 283 if (!PageHighMem(page)) { 284 void *av = __va(PFN_PHYS(pfn)); 285 286 if (av != v) 287 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) 288 BUG(); 289 } else 290 kmap_flush_unused(); 291 } 292 293 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) 294 { 295 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 296 int i; 297 298 for(i = 0; i < entries; i += entries_per_page) 299 set_aliased_prot(ldt + i, PAGE_KERNEL_RO); 300 } 301 302 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) 303 { 304 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 305 int i; 306 307 for(i = 0; i < entries; i += entries_per_page) 308 set_aliased_prot(ldt + i, PAGE_KERNEL); 309 } 310 311 static void xen_set_ldt(const void *addr, unsigned entries) 312 { 313 struct mmuext_op *op; 314 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 315 316 op = mcs.args; 317 op->cmd = MMUEXT_SET_LDT; 318 op->arg1.linear_addr = (unsigned long)addr; 319 op->arg2.nr_ents = entries; 320 321 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 322 323 xen_mc_issue(PARAVIRT_LAZY_CPU); 324 } 325 326 static void xen_load_gdt(const struct desc_ptr *dtr) 327 { 328 unsigned long va = dtr->address; 329 unsigned int size = dtr->size + 1; 330 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 331 unsigned long frames[pages]; 332 int f; 333 334 /* 335 * A GDT can be up to 64k in size, which corresponds to 8192 336 * 8-byte entries, or 16 4k pages.. 337 */ 338 339 BUG_ON(size > 65536); 340 BUG_ON(va & ~PAGE_MASK); 341 342 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 343 int level; 344 pte_t *ptep; 345 unsigned long pfn, mfn; 346 void *virt; 347 348 /* 349 * The GDT is per-cpu and is in the percpu data area. 350 * That can be virtually mapped, so we need to do a 351 * page-walk to get the underlying MFN for the 352 * hypercall. The page can also be in the kernel's 353 * linear range, so we need to RO that mapping too. 354 */ 355 ptep = lookup_address(va, &level); 356 BUG_ON(ptep == NULL); 357 358 pfn = pte_pfn(*ptep); 359 mfn = pfn_to_mfn(pfn); 360 virt = __va(PFN_PHYS(pfn)); 361 362 frames[f] = mfn; 363 364 make_lowmem_page_readonly((void *)va); 365 make_lowmem_page_readonly(virt); 366 } 367 368 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 369 BUG(); 370 } 371 372 /* 373 * load_gdt for early boot, when the gdt is only mapped once 374 */ 375 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr) 376 { 377 unsigned long va = dtr->address; 378 unsigned int size = dtr->size + 1; 379 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 380 unsigned long frames[pages]; 381 int f; 382 383 /* 384 * A GDT can be up to 64k in size, which corresponds to 8192 385 * 8-byte entries, or 16 4k pages.. 386 */ 387 388 BUG_ON(size > 65536); 389 BUG_ON(va & ~PAGE_MASK); 390 391 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 392 pte_t pte; 393 unsigned long pfn, mfn; 394 395 pfn = virt_to_pfn(va); 396 mfn = pfn_to_mfn(pfn); 397 398 pte = pfn_pte(pfn, PAGE_KERNEL_RO); 399 400 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 401 BUG(); 402 403 frames[f] = mfn; 404 } 405 406 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 407 BUG(); 408 } 409 410 static void load_TLS_descriptor(struct thread_struct *t, 411 unsigned int cpu, unsigned int i) 412 { 413 struct desc_struct *gdt = get_cpu_gdt_table(cpu); 414 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 415 struct multicall_space mc = __xen_mc_entry(0); 416 417 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 418 } 419 420 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 421 { 422 /* 423 * XXX sleazy hack: If we're being called in a lazy-cpu zone 424 * and lazy gs handling is enabled, it means we're in a 425 * context switch, and %gs has just been saved. This means we 426 * can zero it out to prevent faults on exit from the 427 * hypervisor if the next process has no %gs. Either way, it 428 * has been saved, and the new value will get loaded properly. 429 * This will go away as soon as Xen has been modified to not 430 * save/restore %gs for normal hypercalls. 431 * 432 * On x86_64, this hack is not used for %gs, because gs points 433 * to KERNEL_GS_BASE (and uses it for PDA references), so we 434 * must not zero %gs on x86_64 435 * 436 * For x86_64, we need to zero %fs, otherwise we may get an 437 * exception between the new %fs descriptor being loaded and 438 * %fs being effectively cleared at __switch_to(). 439 */ 440 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { 441 #ifdef CONFIG_X86_32 442 lazy_load_gs(0); 443 #else 444 loadsegment(fs, 0); 445 #endif 446 } 447 448 xen_mc_batch(); 449 450 load_TLS_descriptor(t, cpu, 0); 451 load_TLS_descriptor(t, cpu, 1); 452 load_TLS_descriptor(t, cpu, 2); 453 454 xen_mc_issue(PARAVIRT_LAZY_CPU); 455 } 456 457 #ifdef CONFIG_X86_64 458 static void xen_load_gs_index(unsigned int idx) 459 { 460 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) 461 BUG(); 462 } 463 #endif 464 465 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 466 const void *ptr) 467 { 468 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); 469 u64 entry = *(u64 *)ptr; 470 471 preempt_disable(); 472 473 xen_mc_flush(); 474 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 475 BUG(); 476 477 preempt_enable(); 478 } 479 480 static int cvt_gate_to_trap(int vector, const gate_desc *val, 481 struct trap_info *info) 482 { 483 unsigned long addr; 484 485 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT) 486 return 0; 487 488 info->vector = vector; 489 490 addr = gate_offset(*val); 491 #ifdef CONFIG_X86_64 492 /* 493 * Look for known traps using IST, and substitute them 494 * appropriately. The debugger ones are the only ones we care 495 * about. Xen will handle faults like double_fault and 496 * machine_check, so we should never see them. Warn if 497 * there's an unexpected IST-using fault handler. 498 */ 499 if (addr == (unsigned long)debug) 500 addr = (unsigned long)xen_debug; 501 else if (addr == (unsigned long)int3) 502 addr = (unsigned long)xen_int3; 503 else if (addr == (unsigned long)stack_segment) 504 addr = (unsigned long)xen_stack_segment; 505 else if (addr == (unsigned long)double_fault || 506 addr == (unsigned long)nmi) { 507 /* Don't need to handle these */ 508 return 0; 509 #ifdef CONFIG_X86_MCE 510 } else if (addr == (unsigned long)machine_check) { 511 return 0; 512 #endif 513 } else { 514 /* Some other trap using IST? */ 515 if (WARN_ON(val->ist != 0)) 516 return 0; 517 } 518 #endif /* CONFIG_X86_64 */ 519 info->address = addr; 520 521 info->cs = gate_segment(*val); 522 info->flags = val->dpl; 523 /* interrupt gates clear IF */ 524 if (val->type == GATE_INTERRUPT) 525 info->flags |= 1 << 2; 526 527 return 1; 528 } 529 530 /* Locations of each CPU's IDT */ 531 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 532 533 /* Set an IDT entry. If the entry is part of the current IDT, then 534 also update Xen. */ 535 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 536 { 537 unsigned long p = (unsigned long)&dt[entrynum]; 538 unsigned long start, end; 539 540 preempt_disable(); 541 542 start = __get_cpu_var(idt_desc).address; 543 end = start + __get_cpu_var(idt_desc).size + 1; 544 545 xen_mc_flush(); 546 547 native_write_idt_entry(dt, entrynum, g); 548 549 if (p >= start && (p + 8) <= end) { 550 struct trap_info info[2]; 551 552 info[1].address = 0; 553 554 if (cvt_gate_to_trap(entrynum, g, &info[0])) 555 if (HYPERVISOR_set_trap_table(info)) 556 BUG(); 557 } 558 559 preempt_enable(); 560 } 561 562 static void xen_convert_trap_info(const struct desc_ptr *desc, 563 struct trap_info *traps) 564 { 565 unsigned in, out, count; 566 567 count = (desc->size+1) / sizeof(gate_desc); 568 BUG_ON(count > 256); 569 570 for (in = out = 0; in < count; in++) { 571 gate_desc *entry = (gate_desc*)(desc->address) + in; 572 573 if (cvt_gate_to_trap(in, entry, &traps[out])) 574 out++; 575 } 576 traps[out].address = 0; 577 } 578 579 void xen_copy_trap_info(struct trap_info *traps) 580 { 581 const struct desc_ptr *desc = &__get_cpu_var(idt_desc); 582 583 xen_convert_trap_info(desc, traps); 584 } 585 586 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 587 hold a spinlock to protect the static traps[] array (static because 588 it avoids allocation, and saves stack space). */ 589 static void xen_load_idt(const struct desc_ptr *desc) 590 { 591 static DEFINE_SPINLOCK(lock); 592 static struct trap_info traps[257]; 593 594 spin_lock(&lock); 595 596 __get_cpu_var(idt_desc) = *desc; 597 598 xen_convert_trap_info(desc, traps); 599 600 xen_mc_flush(); 601 if (HYPERVISOR_set_trap_table(traps)) 602 BUG(); 603 604 spin_unlock(&lock); 605 } 606 607 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 608 they're handled differently. */ 609 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 610 const void *desc, int type) 611 { 612 preempt_disable(); 613 614 switch (type) { 615 case DESC_LDT: 616 case DESC_TSS: 617 /* ignore */ 618 break; 619 620 default: { 621 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); 622 623 xen_mc_flush(); 624 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 625 BUG(); 626 } 627 628 } 629 630 preempt_enable(); 631 } 632 633 /* 634 * Version of write_gdt_entry for use at early boot-time needed to 635 * update an entry as simply as possible. 636 */ 637 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, 638 const void *desc, int type) 639 { 640 switch (type) { 641 case DESC_LDT: 642 case DESC_TSS: 643 /* ignore */ 644 break; 645 646 default: { 647 xmaddr_t maddr = virt_to_machine(&dt[entry]); 648 649 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 650 dt[entry] = *(struct desc_struct *)desc; 651 } 652 653 } 654 } 655 656 static void xen_load_sp0(struct tss_struct *tss, 657 struct thread_struct *thread) 658 { 659 struct multicall_space mcs = xen_mc_entry(0); 660 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 661 xen_mc_issue(PARAVIRT_LAZY_CPU); 662 } 663 664 static void xen_set_iopl_mask(unsigned mask) 665 { 666 struct physdev_set_iopl set_iopl; 667 668 /* Force the change at ring 0. */ 669 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; 670 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 671 } 672 673 static void xen_io_delay(void) 674 { 675 } 676 677 #ifdef CONFIG_X86_LOCAL_APIC 678 static u32 xen_apic_read(u32 reg) 679 { 680 return 0; 681 } 682 683 static void xen_apic_write(u32 reg, u32 val) 684 { 685 /* Warn to see if there's any stray references */ 686 WARN_ON(1); 687 } 688 689 static u64 xen_apic_icr_read(void) 690 { 691 return 0; 692 } 693 694 static void xen_apic_icr_write(u32 low, u32 id) 695 { 696 /* Warn to see if there's any stray references */ 697 WARN_ON(1); 698 } 699 700 static void xen_apic_wait_icr_idle(void) 701 { 702 return; 703 } 704 705 static u32 xen_safe_apic_wait_icr_idle(void) 706 { 707 return 0; 708 } 709 710 static void set_xen_basic_apic_ops(void) 711 { 712 apic->read = xen_apic_read; 713 apic->write = xen_apic_write; 714 apic->icr_read = xen_apic_icr_read; 715 apic->icr_write = xen_apic_icr_write; 716 apic->wait_icr_idle = xen_apic_wait_icr_idle; 717 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; 718 } 719 720 #endif 721 722 723 static void xen_clts(void) 724 { 725 struct multicall_space mcs; 726 727 mcs = xen_mc_entry(0); 728 729 MULTI_fpu_taskswitch(mcs.mc, 0); 730 731 xen_mc_issue(PARAVIRT_LAZY_CPU); 732 } 733 734 static DEFINE_PER_CPU(unsigned long, xen_cr0_value); 735 736 static unsigned long xen_read_cr0(void) 737 { 738 unsigned long cr0 = percpu_read(xen_cr0_value); 739 740 if (unlikely(cr0 == 0)) { 741 cr0 = native_read_cr0(); 742 percpu_write(xen_cr0_value, cr0); 743 } 744 745 return cr0; 746 } 747 748 static void xen_write_cr0(unsigned long cr0) 749 { 750 struct multicall_space mcs; 751 752 percpu_write(xen_cr0_value, cr0); 753 754 /* Only pay attention to cr0.TS; everything else is 755 ignored. */ 756 mcs = xen_mc_entry(0); 757 758 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); 759 760 xen_mc_issue(PARAVIRT_LAZY_CPU); 761 } 762 763 static void xen_write_cr4(unsigned long cr4) 764 { 765 cr4 &= ~X86_CR4_PGE; 766 cr4 &= ~X86_CR4_PSE; 767 768 native_write_cr4(cr4); 769 } 770 771 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) 772 { 773 int ret; 774 775 ret = 0; 776 777 switch (msr) { 778 #ifdef CONFIG_X86_64 779 unsigned which; 780 u64 base; 781 782 case MSR_FS_BASE: which = SEGBASE_FS; goto set; 783 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; 784 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; 785 786 set: 787 base = ((u64)high << 32) | low; 788 if (HYPERVISOR_set_segment_base(which, base) != 0) 789 ret = -EFAULT; 790 break; 791 #endif 792 793 case MSR_STAR: 794 case MSR_CSTAR: 795 case MSR_LSTAR: 796 case MSR_SYSCALL_MASK: 797 case MSR_IA32_SYSENTER_CS: 798 case MSR_IA32_SYSENTER_ESP: 799 case MSR_IA32_SYSENTER_EIP: 800 /* Fast syscall setup is all done in hypercalls, so 801 these are all ignored. Stub them out here to stop 802 Xen console noise. */ 803 break; 804 805 default: 806 ret = native_write_msr_safe(msr, low, high); 807 } 808 809 return ret; 810 } 811 812 void xen_setup_shared_info(void) 813 { 814 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 815 set_fixmap(FIX_PARAVIRT_BOOTMAP, 816 xen_start_info->shared_info); 817 818 HYPERVISOR_shared_info = 819 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); 820 } else 821 HYPERVISOR_shared_info = 822 (struct shared_info *)__va(xen_start_info->shared_info); 823 824 #ifndef CONFIG_SMP 825 /* In UP this is as good a place as any to set up shared info */ 826 xen_setup_vcpu_info_placement(); 827 #endif 828 829 xen_setup_mfn_list_list(); 830 } 831 832 /* This is called once we have the cpu_possible_map */ 833 void xen_setup_vcpu_info_placement(void) 834 { 835 int cpu; 836 837 for_each_possible_cpu(cpu) 838 xen_vcpu_setup(cpu); 839 840 /* xen_vcpu_setup managed to place the vcpu_info within the 841 percpu area for all cpus, so make use of it */ 842 if (have_vcpu_info_placement) { 843 printk(KERN_INFO "Xen: using vcpu_info placement\n"); 844 845 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); 846 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); 847 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); 848 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); 849 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 850 } 851 } 852 853 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 854 unsigned long addr, unsigned len) 855 { 856 char *start, *end, *reloc; 857 unsigned ret; 858 859 start = end = reloc = NULL; 860 861 #define SITE(op, x) \ 862 case PARAVIRT_PATCH(op.x): \ 863 if (have_vcpu_info_placement) { \ 864 start = (char *)xen_##x##_direct; \ 865 end = xen_##x##_direct_end; \ 866 reloc = xen_##x##_direct_reloc; \ 867 } \ 868 goto patch_site 869 870 switch (type) { 871 SITE(pv_irq_ops, irq_enable); 872 SITE(pv_irq_ops, irq_disable); 873 SITE(pv_irq_ops, save_fl); 874 SITE(pv_irq_ops, restore_fl); 875 #undef SITE 876 877 patch_site: 878 if (start == NULL || (end-start) > len) 879 goto default_patch; 880 881 ret = paravirt_patch_insns(insnbuf, len, start, end); 882 883 /* Note: because reloc is assigned from something that 884 appears to be an array, gcc assumes it's non-null, 885 but doesn't know its relationship with start and 886 end. */ 887 if (reloc > start && reloc < end) { 888 int reloc_off = reloc - start; 889 long *relocp = (long *)(insnbuf + reloc_off); 890 long delta = start - (char *)addr; 891 892 *relocp += delta; 893 } 894 break; 895 896 default_patch: 897 default: 898 ret = paravirt_patch_default(type, clobbers, insnbuf, 899 addr, len); 900 break; 901 } 902 903 return ret; 904 } 905 906 static const struct pv_info xen_info __initdata = { 907 .paravirt_enabled = 1, 908 .shared_kernel_pmd = 0, 909 910 .name = "Xen", 911 }; 912 913 static const struct pv_init_ops xen_init_ops __initdata = { 914 .patch = xen_patch, 915 916 .banner = xen_banner, 917 .memory_setup = xen_memory_setup, 918 .arch_setup = xen_arch_setup, 919 .post_allocator_init = xen_post_allocator_init, 920 }; 921 922 static const struct pv_time_ops xen_time_ops __initdata = { 923 .time_init = xen_time_init, 924 925 .set_wallclock = xen_set_wallclock, 926 .get_wallclock = xen_get_wallclock, 927 .get_tsc_khz = xen_tsc_khz, 928 .sched_clock = xen_sched_clock, 929 }; 930 931 static const struct pv_cpu_ops xen_cpu_ops __initdata = { 932 .cpuid = xen_cpuid, 933 934 .set_debugreg = xen_set_debugreg, 935 .get_debugreg = xen_get_debugreg, 936 937 .clts = xen_clts, 938 939 .read_cr0 = xen_read_cr0, 940 .write_cr0 = xen_write_cr0, 941 942 .read_cr4 = native_read_cr4, 943 .read_cr4_safe = native_read_cr4_safe, 944 .write_cr4 = xen_write_cr4, 945 946 .wbinvd = native_wbinvd, 947 948 .read_msr = native_read_msr_safe, 949 .write_msr = xen_write_msr_safe, 950 .read_tsc = native_read_tsc, 951 .read_pmc = native_read_pmc, 952 953 .iret = xen_iret, 954 .irq_enable_sysexit = xen_sysexit, 955 #ifdef CONFIG_X86_64 956 .usergs_sysret32 = xen_sysret32, 957 .usergs_sysret64 = xen_sysret64, 958 #endif 959 960 .load_tr_desc = paravirt_nop, 961 .set_ldt = xen_set_ldt, 962 .load_gdt = xen_load_gdt, 963 .load_idt = xen_load_idt, 964 .load_tls = xen_load_tls, 965 #ifdef CONFIG_X86_64 966 .load_gs_index = xen_load_gs_index, 967 #endif 968 969 .alloc_ldt = xen_alloc_ldt, 970 .free_ldt = xen_free_ldt, 971 972 .store_gdt = native_store_gdt, 973 .store_idt = native_store_idt, 974 .store_tr = xen_store_tr, 975 976 .write_ldt_entry = xen_write_ldt_entry, 977 .write_gdt_entry = xen_write_gdt_entry, 978 .write_idt_entry = xen_write_idt_entry, 979 .load_sp0 = xen_load_sp0, 980 981 .set_iopl_mask = xen_set_iopl_mask, 982 .io_delay = xen_io_delay, 983 984 /* Xen takes care of %gs when switching to usermode for us */ 985 .swapgs = paravirt_nop, 986 987 .start_context_switch = paravirt_start_context_switch, 988 .end_context_switch = xen_end_context_switch, 989 }; 990 991 static const struct pv_apic_ops xen_apic_ops __initdata = { 992 #ifdef CONFIG_X86_LOCAL_APIC 993 .setup_boot_clock = paravirt_nop, 994 .setup_secondary_clock = paravirt_nop, 995 .startup_ipi_hook = paravirt_nop, 996 #endif 997 }; 998 999 static void xen_reboot(int reason) 1000 { 1001 struct sched_shutdown r = { .reason = reason }; 1002 1003 #ifdef CONFIG_SMP 1004 smp_send_stop(); 1005 #endif 1006 1007 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) 1008 BUG(); 1009 } 1010 1011 static void xen_restart(char *msg) 1012 { 1013 xen_reboot(SHUTDOWN_reboot); 1014 } 1015 1016 static void xen_emergency_restart(void) 1017 { 1018 xen_reboot(SHUTDOWN_reboot); 1019 } 1020 1021 static void xen_machine_halt(void) 1022 { 1023 xen_reboot(SHUTDOWN_poweroff); 1024 } 1025 1026 static void xen_crash_shutdown(struct pt_regs *regs) 1027 { 1028 xen_reboot(SHUTDOWN_crash); 1029 } 1030 1031 static const struct machine_ops __initdata xen_machine_ops = { 1032 .restart = xen_restart, 1033 .halt = xen_machine_halt, 1034 .power_off = xen_machine_halt, 1035 .shutdown = xen_machine_halt, 1036 .crash_shutdown = xen_crash_shutdown, 1037 .emergency_restart = xen_emergency_restart, 1038 }; 1039 1040 /* 1041 * Set up the GDT and segment registers for -fstack-protector. Until 1042 * we do this, we have to be careful not to call any stack-protected 1043 * function, which is most of the kernel. 1044 */ 1045 static void __init xen_setup_stackprotector(void) 1046 { 1047 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; 1048 pv_cpu_ops.load_gdt = xen_load_gdt_boot; 1049 1050 setup_stack_canary_segment(0); 1051 switch_to_new_gdt(0); 1052 1053 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; 1054 pv_cpu_ops.load_gdt = xen_load_gdt; 1055 } 1056 1057 /* First C function to be called on Xen boot */ 1058 asmlinkage void __init xen_start_kernel(void) 1059 { 1060 pgd_t *pgd; 1061 1062 if (!xen_start_info) 1063 return; 1064 1065 xen_domain_type = XEN_PV_DOMAIN; 1066 1067 /* Install Xen paravirt ops */ 1068 pv_info = xen_info; 1069 pv_init_ops = xen_init_ops; 1070 pv_time_ops = xen_time_ops; 1071 pv_cpu_ops = xen_cpu_ops; 1072 pv_apic_ops = xen_apic_ops; 1073 pv_mmu_ops = xen_mmu_ops; 1074 1075 /* 1076 * Set up some pagetable state before starting to set any ptes. 1077 */ 1078 1079 /* Prevent unwanted bits from being set in PTEs. */ 1080 __supported_pte_mask &= ~_PAGE_GLOBAL; 1081 if (!xen_initial_domain()) 1082 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); 1083 1084 __supported_pte_mask |= _PAGE_IOMAP; 1085 1086 xen_setup_features(); 1087 1088 /* Get mfn list */ 1089 if (!xen_feature(XENFEAT_auto_translated_physmap)) 1090 xen_build_dynamic_phys_to_machine(); 1091 1092 /* 1093 * Set up kernel GDT and segment registers, mainly so that 1094 * -fstack-protector code can be executed. 1095 */ 1096 xen_setup_stackprotector(); 1097 1098 xen_init_irq_ops(); 1099 xen_init_cpuid_mask(); 1100 1101 #ifdef CONFIG_X86_LOCAL_APIC 1102 /* 1103 * set up the basic apic ops. 1104 */ 1105 set_xen_basic_apic_ops(); 1106 #endif 1107 1108 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { 1109 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; 1110 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; 1111 } 1112 1113 machine_ops = xen_machine_ops; 1114 1115 /* 1116 * The only reliable way to retain the initial address of the 1117 * percpu gdt_page is to remember it here, so we can go and 1118 * mark it RW later, when the initial percpu area is freed. 1119 */ 1120 xen_initial_gdt = &per_cpu(gdt_page, 0); 1121 1122 xen_smp_init(); 1123 1124 pgd = (pgd_t *)xen_start_info->pt_base; 1125 1126 #ifdef CONFIG_X86_64 1127 /* Work out if we support NX */ 1128 check_efer(); 1129 #endif 1130 1131 /* Don't do the full vcpu_info placement stuff until we have a 1132 possible map and a non-dummy shared_info. */ 1133 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; 1134 1135 local_irq_disable(); 1136 early_boot_irqs_off(); 1137 1138 xen_raw_console_write("mapping kernel into physical memory\n"); 1139 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); 1140 1141 init_mm.pgd = pgd; 1142 1143 /* keep using Xen gdt for now; no urgent need to change it */ 1144 1145 pv_info.kernel_rpl = 1; 1146 if (xen_feature(XENFEAT_supervisor_mode_kernel)) 1147 pv_info.kernel_rpl = 0; 1148 1149 /* set the limit of our address space */ 1150 xen_reserve_top(); 1151 1152 #ifdef CONFIG_X86_32 1153 /* set up basic CPUID stuff */ 1154 cpu_detect(&new_cpu_data); 1155 new_cpu_data.hard_math = 1; 1156 new_cpu_data.wp_works_ok = 1; 1157 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1158 #endif 1159 1160 /* Poke various useful things into boot_params */ 1161 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1162 boot_params.hdr.ramdisk_image = xen_start_info->mod_start 1163 ? __pa(xen_start_info->mod_start) : 0; 1164 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1165 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1166 1167 if (!xen_initial_domain()) { 1168 add_preferred_console("xenboot", 0, NULL); 1169 add_preferred_console("tty", 0, NULL); 1170 add_preferred_console("hvc", 0, NULL); 1171 } 1172 1173 xen_raw_console_write("about to get started...\n"); 1174 1175 /* Start the world */ 1176 #ifdef CONFIG_X86_32 1177 i386_start_kernel(); 1178 #else 1179 x86_64_start_reservations((char *)__pa_symbol(&boot_params)); 1180 #endif 1181 } 1182