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 #include <linux/pci.h> 31 32 #include <xen/xen.h> 33 #include <xen/interface/xen.h> 34 #include <xen/interface/version.h> 35 #include <xen/interface/physdev.h> 36 #include <xen/interface/vcpu.h> 37 #include <xen/features.h> 38 #include <xen/page.h> 39 #include <xen/hvc-console.h> 40 41 #include <asm/paravirt.h> 42 #include <asm/apic.h> 43 #include <asm/page.h> 44 #include <asm/xen/hypercall.h> 45 #include <asm/xen/hypervisor.h> 46 #include <asm/fixmap.h> 47 #include <asm/processor.h> 48 #include <asm/proto.h> 49 #include <asm/msr-index.h> 50 #include <asm/traps.h> 51 #include <asm/setup.h> 52 #include <asm/desc.h> 53 #include <asm/pgtable.h> 54 #include <asm/tlbflush.h> 55 #include <asm/reboot.h> 56 #include <asm/stackprotector.h> 57 58 #include "xen-ops.h" 59 #include "mmu.h" 60 #include "multicalls.h" 61 62 EXPORT_SYMBOL_GPL(hypercall_page); 63 64 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 65 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 66 67 enum xen_domain_type xen_domain_type = XEN_NATIVE; 68 EXPORT_SYMBOL_GPL(xen_domain_type); 69 70 struct start_info *xen_start_info; 71 EXPORT_SYMBOL_GPL(xen_start_info); 72 73 struct shared_info xen_dummy_shared_info; 74 75 void *xen_initial_gdt; 76 77 /* 78 * Point at some empty memory to start with. We map the real shared_info 79 * page as soon as fixmap is up and running. 80 */ 81 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; 82 83 /* 84 * Flag to determine whether vcpu info placement is available on all 85 * VCPUs. We assume it is to start with, and then set it to zero on 86 * the first failure. This is because it can succeed on some VCPUs 87 * and not others, since it can involve hypervisor memory allocation, 88 * or because the guest failed to guarantee all the appropriate 89 * constraints on all VCPUs (ie buffer can't cross a page boundary). 90 * 91 * Note that any particular CPU may be using a placed vcpu structure, 92 * but we can only optimise if the all are. 93 * 94 * 0: not available, 1: available 95 */ 96 static int have_vcpu_info_placement = 1; 97 98 static void xen_vcpu_setup(int cpu) 99 { 100 struct vcpu_register_vcpu_info info; 101 int err; 102 struct vcpu_info *vcpup; 103 104 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); 105 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 106 107 if (!have_vcpu_info_placement) 108 return; /* already tested, not available */ 109 110 vcpup = &per_cpu(xen_vcpu_info, cpu); 111 112 info.mfn = arbitrary_virt_to_mfn(vcpup); 113 info.offset = offset_in_page(vcpup); 114 115 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", 116 cpu, vcpup, info.mfn, info.offset); 117 118 /* Check to see if the hypervisor will put the vcpu_info 119 structure where we want it, which allows direct access via 120 a percpu-variable. */ 121 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 122 123 if (err) { 124 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); 125 have_vcpu_info_placement = 0; 126 } else { 127 /* This cpu is using the registered vcpu info, even if 128 later ones fail to. */ 129 per_cpu(xen_vcpu, cpu) = vcpup; 130 131 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", 132 cpu, vcpup); 133 } 134 } 135 136 /* 137 * On restore, set the vcpu placement up again. 138 * If it fails, then we're in a bad state, since 139 * we can't back out from using it... 140 */ 141 void xen_vcpu_restore(void) 142 { 143 int cpu; 144 145 for_each_online_cpu(cpu) { 146 bool other_cpu = (cpu != smp_processor_id()); 147 148 if (other_cpu && 149 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) 150 BUG(); 151 152 xen_setup_runstate_info(cpu); 153 154 if (have_vcpu_info_placement) 155 xen_vcpu_setup(cpu); 156 157 if (other_cpu && 158 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) 159 BUG(); 160 } 161 } 162 163 static void __init xen_banner(void) 164 { 165 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); 166 struct xen_extraversion extra; 167 HYPERVISOR_xen_version(XENVER_extraversion, &extra); 168 169 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 170 pv_info.name); 171 printk(KERN_INFO "Xen version: %d.%d%s%s\n", 172 version >> 16, version & 0xffff, extra.extraversion, 173 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); 174 } 175 176 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0; 177 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0; 178 179 static void xen_cpuid(unsigned int *ax, unsigned int *bx, 180 unsigned int *cx, unsigned int *dx) 181 { 182 unsigned maskebx = ~0; 183 unsigned maskecx = ~0; 184 unsigned maskedx = ~0; 185 186 /* 187 * Mask out inconvenient features, to try and disable as many 188 * unsupported kernel subsystems as possible. 189 */ 190 switch (*ax) { 191 case 1: 192 maskecx = cpuid_leaf1_ecx_mask; 193 maskedx = cpuid_leaf1_edx_mask; 194 break; 195 196 case 0xb: 197 /* Suppress extended topology stuff */ 198 maskebx = 0; 199 break; 200 } 201 202 asm(XEN_EMULATE_PREFIX "cpuid" 203 : "=a" (*ax), 204 "=b" (*bx), 205 "=c" (*cx), 206 "=d" (*dx) 207 : "0" (*ax), "2" (*cx)); 208 209 *bx &= maskebx; 210 *cx &= maskecx; 211 *dx &= maskedx; 212 } 213 214 static __init void xen_init_cpuid_mask(void) 215 { 216 unsigned int ax, bx, cx, dx; 217 218 cpuid_leaf1_edx_mask = 219 ~((1 << X86_FEATURE_MCE) | /* disable MCE */ 220 (1 << X86_FEATURE_MCA) | /* disable MCA */ 221 (1 << X86_FEATURE_ACC)); /* thermal monitoring */ 222 223 if (!xen_initial_domain()) 224 cpuid_leaf1_edx_mask &= 225 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */ 226 (1 << X86_FEATURE_ACPI)); /* disable ACPI */ 227 228 ax = 1; 229 cx = 0; 230 xen_cpuid(&ax, &bx, &cx, &dx); 231 232 /* cpuid claims we support xsave; try enabling it to see what happens */ 233 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) { 234 unsigned long cr4; 235 236 set_in_cr4(X86_CR4_OSXSAVE); 237 238 cr4 = read_cr4(); 239 240 if ((cr4 & X86_CR4_OSXSAVE) == 0) 241 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32)); 242 243 clear_in_cr4(X86_CR4_OSXSAVE); 244 } 245 } 246 247 static void xen_set_debugreg(int reg, unsigned long val) 248 { 249 HYPERVISOR_set_debugreg(reg, val); 250 } 251 252 static unsigned long xen_get_debugreg(int reg) 253 { 254 return HYPERVISOR_get_debugreg(reg); 255 } 256 257 static void xen_end_context_switch(struct task_struct *next) 258 { 259 xen_mc_flush(); 260 paravirt_end_context_switch(next); 261 } 262 263 static unsigned long xen_store_tr(void) 264 { 265 return 0; 266 } 267 268 /* 269 * Set the page permissions for a particular virtual address. If the 270 * address is a vmalloc mapping (or other non-linear mapping), then 271 * find the linear mapping of the page and also set its protections to 272 * match. 273 */ 274 static void set_aliased_prot(void *v, pgprot_t prot) 275 { 276 int level; 277 pte_t *ptep; 278 pte_t pte; 279 unsigned long pfn; 280 struct page *page; 281 282 ptep = lookup_address((unsigned long)v, &level); 283 BUG_ON(ptep == NULL); 284 285 pfn = pte_pfn(*ptep); 286 page = pfn_to_page(pfn); 287 288 pte = pfn_pte(pfn, prot); 289 290 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) 291 BUG(); 292 293 if (!PageHighMem(page)) { 294 void *av = __va(PFN_PHYS(pfn)); 295 296 if (av != v) 297 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) 298 BUG(); 299 } else 300 kmap_flush_unused(); 301 } 302 303 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) 304 { 305 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 306 int i; 307 308 for(i = 0; i < entries; i += entries_per_page) 309 set_aliased_prot(ldt + i, PAGE_KERNEL_RO); 310 } 311 312 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) 313 { 314 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 315 int i; 316 317 for(i = 0; i < entries; i += entries_per_page) 318 set_aliased_prot(ldt + i, PAGE_KERNEL); 319 } 320 321 static void xen_set_ldt(const void *addr, unsigned entries) 322 { 323 struct mmuext_op *op; 324 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 325 326 op = mcs.args; 327 op->cmd = MMUEXT_SET_LDT; 328 op->arg1.linear_addr = (unsigned long)addr; 329 op->arg2.nr_ents = entries; 330 331 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 332 333 xen_mc_issue(PARAVIRT_LAZY_CPU); 334 } 335 336 static void xen_load_gdt(const struct desc_ptr *dtr) 337 { 338 unsigned long va = dtr->address; 339 unsigned int size = dtr->size + 1; 340 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 341 unsigned long frames[pages]; 342 int f; 343 344 /* 345 * A GDT can be up to 64k in size, which corresponds to 8192 346 * 8-byte entries, or 16 4k pages.. 347 */ 348 349 BUG_ON(size > 65536); 350 BUG_ON(va & ~PAGE_MASK); 351 352 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 353 int level; 354 pte_t *ptep; 355 unsigned long pfn, mfn; 356 void *virt; 357 358 /* 359 * The GDT is per-cpu and is in the percpu data area. 360 * That can be virtually mapped, so we need to do a 361 * page-walk to get the underlying MFN for the 362 * hypercall. The page can also be in the kernel's 363 * linear range, so we need to RO that mapping too. 364 */ 365 ptep = lookup_address(va, &level); 366 BUG_ON(ptep == NULL); 367 368 pfn = pte_pfn(*ptep); 369 mfn = pfn_to_mfn(pfn); 370 virt = __va(PFN_PHYS(pfn)); 371 372 frames[f] = mfn; 373 374 make_lowmem_page_readonly((void *)va); 375 make_lowmem_page_readonly(virt); 376 } 377 378 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 379 BUG(); 380 } 381 382 /* 383 * load_gdt for early boot, when the gdt is only mapped once 384 */ 385 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr) 386 { 387 unsigned long va = dtr->address; 388 unsigned int size = dtr->size + 1; 389 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 390 unsigned long frames[pages]; 391 int f; 392 393 /* 394 * A GDT can be up to 64k in size, which corresponds to 8192 395 * 8-byte entries, or 16 4k pages.. 396 */ 397 398 BUG_ON(size > 65536); 399 BUG_ON(va & ~PAGE_MASK); 400 401 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 402 pte_t pte; 403 unsigned long pfn, mfn; 404 405 pfn = virt_to_pfn(va); 406 mfn = pfn_to_mfn(pfn); 407 408 pte = pfn_pte(pfn, PAGE_KERNEL_RO); 409 410 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 411 BUG(); 412 413 frames[f] = mfn; 414 } 415 416 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) 417 BUG(); 418 } 419 420 static void load_TLS_descriptor(struct thread_struct *t, 421 unsigned int cpu, unsigned int i) 422 { 423 struct desc_struct *gdt = get_cpu_gdt_table(cpu); 424 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 425 struct multicall_space mc = __xen_mc_entry(0); 426 427 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 428 } 429 430 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 431 { 432 /* 433 * XXX sleazy hack: If we're being called in a lazy-cpu zone 434 * and lazy gs handling is enabled, it means we're in a 435 * context switch, and %gs has just been saved. This means we 436 * can zero it out to prevent faults on exit from the 437 * hypervisor if the next process has no %gs. Either way, it 438 * has been saved, and the new value will get loaded properly. 439 * This will go away as soon as Xen has been modified to not 440 * save/restore %gs for normal hypercalls. 441 * 442 * On x86_64, this hack is not used for %gs, because gs points 443 * to KERNEL_GS_BASE (and uses it for PDA references), so we 444 * must not zero %gs on x86_64 445 * 446 * For x86_64, we need to zero %fs, otherwise we may get an 447 * exception between the new %fs descriptor being loaded and 448 * %fs being effectively cleared at __switch_to(). 449 */ 450 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { 451 #ifdef CONFIG_X86_32 452 lazy_load_gs(0); 453 #else 454 loadsegment(fs, 0); 455 #endif 456 } 457 458 xen_mc_batch(); 459 460 load_TLS_descriptor(t, cpu, 0); 461 load_TLS_descriptor(t, cpu, 1); 462 load_TLS_descriptor(t, cpu, 2); 463 464 xen_mc_issue(PARAVIRT_LAZY_CPU); 465 } 466 467 #ifdef CONFIG_X86_64 468 static void xen_load_gs_index(unsigned int idx) 469 { 470 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) 471 BUG(); 472 } 473 #endif 474 475 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 476 const void *ptr) 477 { 478 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); 479 u64 entry = *(u64 *)ptr; 480 481 preempt_disable(); 482 483 xen_mc_flush(); 484 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 485 BUG(); 486 487 preempt_enable(); 488 } 489 490 static int cvt_gate_to_trap(int vector, const gate_desc *val, 491 struct trap_info *info) 492 { 493 unsigned long addr; 494 495 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT) 496 return 0; 497 498 info->vector = vector; 499 500 addr = gate_offset(*val); 501 #ifdef CONFIG_X86_64 502 /* 503 * Look for known traps using IST, and substitute them 504 * appropriately. The debugger ones are the only ones we care 505 * about. Xen will handle faults like double_fault and 506 * machine_check, so we should never see them. Warn if 507 * there's an unexpected IST-using fault handler. 508 */ 509 if (addr == (unsigned long)debug) 510 addr = (unsigned long)xen_debug; 511 else if (addr == (unsigned long)int3) 512 addr = (unsigned long)xen_int3; 513 else if (addr == (unsigned long)stack_segment) 514 addr = (unsigned long)xen_stack_segment; 515 else if (addr == (unsigned long)double_fault || 516 addr == (unsigned long)nmi) { 517 /* Don't need to handle these */ 518 return 0; 519 #ifdef CONFIG_X86_MCE 520 } else if (addr == (unsigned long)machine_check) { 521 return 0; 522 #endif 523 } else { 524 /* Some other trap using IST? */ 525 if (WARN_ON(val->ist != 0)) 526 return 0; 527 } 528 #endif /* CONFIG_X86_64 */ 529 info->address = addr; 530 531 info->cs = gate_segment(*val); 532 info->flags = val->dpl; 533 /* interrupt gates clear IF */ 534 if (val->type == GATE_INTERRUPT) 535 info->flags |= 1 << 2; 536 537 return 1; 538 } 539 540 /* Locations of each CPU's IDT */ 541 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 542 543 /* Set an IDT entry. If the entry is part of the current IDT, then 544 also update Xen. */ 545 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 546 { 547 unsigned long p = (unsigned long)&dt[entrynum]; 548 unsigned long start, end; 549 550 preempt_disable(); 551 552 start = __get_cpu_var(idt_desc).address; 553 end = start + __get_cpu_var(idt_desc).size + 1; 554 555 xen_mc_flush(); 556 557 native_write_idt_entry(dt, entrynum, g); 558 559 if (p >= start && (p + 8) <= end) { 560 struct trap_info info[2]; 561 562 info[1].address = 0; 563 564 if (cvt_gate_to_trap(entrynum, g, &info[0])) 565 if (HYPERVISOR_set_trap_table(info)) 566 BUG(); 567 } 568 569 preempt_enable(); 570 } 571 572 static void xen_convert_trap_info(const struct desc_ptr *desc, 573 struct trap_info *traps) 574 { 575 unsigned in, out, count; 576 577 count = (desc->size+1) / sizeof(gate_desc); 578 BUG_ON(count > 256); 579 580 for (in = out = 0; in < count; in++) { 581 gate_desc *entry = (gate_desc*)(desc->address) + in; 582 583 if (cvt_gate_to_trap(in, entry, &traps[out])) 584 out++; 585 } 586 traps[out].address = 0; 587 } 588 589 void xen_copy_trap_info(struct trap_info *traps) 590 { 591 const struct desc_ptr *desc = &__get_cpu_var(idt_desc); 592 593 xen_convert_trap_info(desc, traps); 594 } 595 596 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 597 hold a spinlock to protect the static traps[] array (static because 598 it avoids allocation, and saves stack space). */ 599 static void xen_load_idt(const struct desc_ptr *desc) 600 { 601 static DEFINE_SPINLOCK(lock); 602 static struct trap_info traps[257]; 603 604 spin_lock(&lock); 605 606 __get_cpu_var(idt_desc) = *desc; 607 608 xen_convert_trap_info(desc, traps); 609 610 xen_mc_flush(); 611 if (HYPERVISOR_set_trap_table(traps)) 612 BUG(); 613 614 spin_unlock(&lock); 615 } 616 617 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 618 they're handled differently. */ 619 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 620 const void *desc, int type) 621 { 622 preempt_disable(); 623 624 switch (type) { 625 case DESC_LDT: 626 case DESC_TSS: 627 /* ignore */ 628 break; 629 630 default: { 631 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); 632 633 xen_mc_flush(); 634 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 635 BUG(); 636 } 637 638 } 639 640 preempt_enable(); 641 } 642 643 /* 644 * Version of write_gdt_entry for use at early boot-time needed to 645 * update an entry as simply as possible. 646 */ 647 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, 648 const void *desc, int type) 649 { 650 switch (type) { 651 case DESC_LDT: 652 case DESC_TSS: 653 /* ignore */ 654 break; 655 656 default: { 657 xmaddr_t maddr = virt_to_machine(&dt[entry]); 658 659 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 660 dt[entry] = *(struct desc_struct *)desc; 661 } 662 663 } 664 } 665 666 static void xen_load_sp0(struct tss_struct *tss, 667 struct thread_struct *thread) 668 { 669 struct multicall_space mcs = xen_mc_entry(0); 670 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 671 xen_mc_issue(PARAVIRT_LAZY_CPU); 672 } 673 674 static void xen_set_iopl_mask(unsigned mask) 675 { 676 struct physdev_set_iopl set_iopl; 677 678 /* Force the change at ring 0. */ 679 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; 680 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 681 } 682 683 static void xen_io_delay(void) 684 { 685 } 686 687 #ifdef CONFIG_X86_LOCAL_APIC 688 static u32 xen_apic_read(u32 reg) 689 { 690 return 0; 691 } 692 693 static void xen_apic_write(u32 reg, u32 val) 694 { 695 /* Warn to see if there's any stray references */ 696 WARN_ON(1); 697 } 698 699 static u64 xen_apic_icr_read(void) 700 { 701 return 0; 702 } 703 704 static void xen_apic_icr_write(u32 low, u32 id) 705 { 706 /* Warn to see if there's any stray references */ 707 WARN_ON(1); 708 } 709 710 static void xen_apic_wait_icr_idle(void) 711 { 712 return; 713 } 714 715 static u32 xen_safe_apic_wait_icr_idle(void) 716 { 717 return 0; 718 } 719 720 static void set_xen_basic_apic_ops(void) 721 { 722 apic->read = xen_apic_read; 723 apic->write = xen_apic_write; 724 apic->icr_read = xen_apic_icr_read; 725 apic->icr_write = xen_apic_icr_write; 726 apic->wait_icr_idle = xen_apic_wait_icr_idle; 727 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; 728 } 729 730 #endif 731 732 733 static void xen_clts(void) 734 { 735 struct multicall_space mcs; 736 737 mcs = xen_mc_entry(0); 738 739 MULTI_fpu_taskswitch(mcs.mc, 0); 740 741 xen_mc_issue(PARAVIRT_LAZY_CPU); 742 } 743 744 static DEFINE_PER_CPU(unsigned long, xen_cr0_value); 745 746 static unsigned long xen_read_cr0(void) 747 { 748 unsigned long cr0 = percpu_read(xen_cr0_value); 749 750 if (unlikely(cr0 == 0)) { 751 cr0 = native_read_cr0(); 752 percpu_write(xen_cr0_value, cr0); 753 } 754 755 return cr0; 756 } 757 758 static void xen_write_cr0(unsigned long cr0) 759 { 760 struct multicall_space mcs; 761 762 percpu_write(xen_cr0_value, cr0); 763 764 /* Only pay attention to cr0.TS; everything else is 765 ignored. */ 766 mcs = xen_mc_entry(0); 767 768 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); 769 770 xen_mc_issue(PARAVIRT_LAZY_CPU); 771 } 772 773 static void xen_write_cr4(unsigned long cr4) 774 { 775 cr4 &= ~X86_CR4_PGE; 776 cr4 &= ~X86_CR4_PSE; 777 778 native_write_cr4(cr4); 779 } 780 781 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) 782 { 783 int ret; 784 785 ret = 0; 786 787 switch (msr) { 788 #ifdef CONFIG_X86_64 789 unsigned which; 790 u64 base; 791 792 case MSR_FS_BASE: which = SEGBASE_FS; goto set; 793 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; 794 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; 795 796 set: 797 base = ((u64)high << 32) | low; 798 if (HYPERVISOR_set_segment_base(which, base) != 0) 799 ret = -EIO; 800 break; 801 #endif 802 803 case MSR_STAR: 804 case MSR_CSTAR: 805 case MSR_LSTAR: 806 case MSR_SYSCALL_MASK: 807 case MSR_IA32_SYSENTER_CS: 808 case MSR_IA32_SYSENTER_ESP: 809 case MSR_IA32_SYSENTER_EIP: 810 /* Fast syscall setup is all done in hypercalls, so 811 these are all ignored. Stub them out here to stop 812 Xen console noise. */ 813 break; 814 815 default: 816 ret = native_write_msr_safe(msr, low, high); 817 } 818 819 return ret; 820 } 821 822 void xen_setup_shared_info(void) 823 { 824 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 825 set_fixmap(FIX_PARAVIRT_BOOTMAP, 826 xen_start_info->shared_info); 827 828 HYPERVISOR_shared_info = 829 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); 830 } else 831 HYPERVISOR_shared_info = 832 (struct shared_info *)__va(xen_start_info->shared_info); 833 834 #ifndef CONFIG_SMP 835 /* In UP this is as good a place as any to set up shared info */ 836 xen_setup_vcpu_info_placement(); 837 #endif 838 839 xen_setup_mfn_list_list(); 840 } 841 842 /* This is called once we have the cpu_possible_map */ 843 void xen_setup_vcpu_info_placement(void) 844 { 845 int cpu; 846 847 for_each_possible_cpu(cpu) 848 xen_vcpu_setup(cpu); 849 850 /* xen_vcpu_setup managed to place the vcpu_info within the 851 percpu area for all cpus, so make use of it */ 852 if (have_vcpu_info_placement) { 853 printk(KERN_INFO "Xen: using vcpu_info placement\n"); 854 855 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); 856 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); 857 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); 858 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); 859 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 860 } 861 } 862 863 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 864 unsigned long addr, unsigned len) 865 { 866 char *start, *end, *reloc; 867 unsigned ret; 868 869 start = end = reloc = NULL; 870 871 #define SITE(op, x) \ 872 case PARAVIRT_PATCH(op.x): \ 873 if (have_vcpu_info_placement) { \ 874 start = (char *)xen_##x##_direct; \ 875 end = xen_##x##_direct_end; \ 876 reloc = xen_##x##_direct_reloc; \ 877 } \ 878 goto patch_site 879 880 switch (type) { 881 SITE(pv_irq_ops, irq_enable); 882 SITE(pv_irq_ops, irq_disable); 883 SITE(pv_irq_ops, save_fl); 884 SITE(pv_irq_ops, restore_fl); 885 #undef SITE 886 887 patch_site: 888 if (start == NULL || (end-start) > len) 889 goto default_patch; 890 891 ret = paravirt_patch_insns(insnbuf, len, start, end); 892 893 /* Note: because reloc is assigned from something that 894 appears to be an array, gcc assumes it's non-null, 895 but doesn't know its relationship with start and 896 end. */ 897 if (reloc > start && reloc < end) { 898 int reloc_off = reloc - start; 899 long *relocp = (long *)(insnbuf + reloc_off); 900 long delta = start - (char *)addr; 901 902 *relocp += delta; 903 } 904 break; 905 906 default_patch: 907 default: 908 ret = paravirt_patch_default(type, clobbers, insnbuf, 909 addr, len); 910 break; 911 } 912 913 return ret; 914 } 915 916 static const struct pv_info xen_info __initdata = { 917 .paravirt_enabled = 1, 918 .shared_kernel_pmd = 0, 919 920 .name = "Xen", 921 }; 922 923 static const struct pv_init_ops xen_init_ops __initdata = { 924 .patch = xen_patch, 925 }; 926 927 static const struct pv_time_ops xen_time_ops __initdata = { 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 .startup_ipi_hook = paravirt_nop, 994 #endif 995 }; 996 997 static void xen_reboot(int reason) 998 { 999 struct sched_shutdown r = { .reason = reason }; 1000 1001 #ifdef CONFIG_SMP 1002 smp_send_stop(); 1003 #endif 1004 1005 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) 1006 BUG(); 1007 } 1008 1009 static void xen_restart(char *msg) 1010 { 1011 xen_reboot(SHUTDOWN_reboot); 1012 } 1013 1014 static void xen_emergency_restart(void) 1015 { 1016 xen_reboot(SHUTDOWN_reboot); 1017 } 1018 1019 static void xen_machine_halt(void) 1020 { 1021 xen_reboot(SHUTDOWN_poweroff); 1022 } 1023 1024 static void xen_crash_shutdown(struct pt_regs *regs) 1025 { 1026 xen_reboot(SHUTDOWN_crash); 1027 } 1028 1029 static const struct machine_ops __initdata xen_machine_ops = { 1030 .restart = xen_restart, 1031 .halt = xen_machine_halt, 1032 .power_off = xen_machine_halt, 1033 .shutdown = xen_machine_halt, 1034 .crash_shutdown = xen_crash_shutdown, 1035 .emergency_restart = xen_emergency_restart, 1036 }; 1037 1038 /* 1039 * Set up the GDT and segment registers for -fstack-protector. Until 1040 * we do this, we have to be careful not to call any stack-protected 1041 * function, which is most of the kernel. 1042 */ 1043 static void __init xen_setup_stackprotector(void) 1044 { 1045 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; 1046 pv_cpu_ops.load_gdt = xen_load_gdt_boot; 1047 1048 setup_stack_canary_segment(0); 1049 switch_to_new_gdt(0); 1050 1051 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; 1052 pv_cpu_ops.load_gdt = xen_load_gdt; 1053 } 1054 1055 /* First C function to be called on Xen boot */ 1056 asmlinkage void __init xen_start_kernel(void) 1057 { 1058 pgd_t *pgd; 1059 1060 if (!xen_start_info) 1061 return; 1062 1063 xen_domain_type = XEN_PV_DOMAIN; 1064 1065 /* Install Xen paravirt ops */ 1066 pv_info = xen_info; 1067 pv_init_ops = xen_init_ops; 1068 pv_time_ops = xen_time_ops; 1069 pv_cpu_ops = xen_cpu_ops; 1070 pv_apic_ops = xen_apic_ops; 1071 1072 x86_init.resources.memory_setup = xen_memory_setup; 1073 x86_init.oem.arch_setup = xen_arch_setup; 1074 x86_init.oem.banner = xen_banner; 1075 1076 x86_init.timers.timer_init = xen_time_init; 1077 x86_init.timers.setup_percpu_clockev = x86_init_noop; 1078 x86_cpuinit.setup_percpu_clockev = x86_init_noop; 1079 1080 x86_platform.calibrate_tsc = xen_tsc_khz; 1081 x86_platform.get_wallclock = xen_get_wallclock; 1082 x86_platform.set_wallclock = xen_set_wallclock; 1083 1084 /* 1085 * Set up some pagetable state before starting to set any ptes. 1086 */ 1087 1088 xen_init_mmu_ops(); 1089 1090 /* Prevent unwanted bits from being set in PTEs. */ 1091 __supported_pte_mask &= ~_PAGE_GLOBAL; 1092 if (!xen_initial_domain()) 1093 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); 1094 1095 __supported_pte_mask |= _PAGE_IOMAP; 1096 1097 /* Work out if we support NX */ 1098 x86_configure_nx(); 1099 1100 xen_setup_features(); 1101 1102 /* Get mfn list */ 1103 if (!xen_feature(XENFEAT_auto_translated_physmap)) 1104 xen_build_dynamic_phys_to_machine(); 1105 1106 /* 1107 * Set up kernel GDT and segment registers, mainly so that 1108 * -fstack-protector code can be executed. 1109 */ 1110 xen_setup_stackprotector(); 1111 1112 xen_init_irq_ops(); 1113 xen_init_cpuid_mask(); 1114 1115 #ifdef CONFIG_X86_LOCAL_APIC 1116 /* 1117 * set up the basic apic ops. 1118 */ 1119 set_xen_basic_apic_ops(); 1120 #endif 1121 1122 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { 1123 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; 1124 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; 1125 } 1126 1127 machine_ops = xen_machine_ops; 1128 1129 /* 1130 * The only reliable way to retain the initial address of the 1131 * percpu gdt_page is to remember it here, so we can go and 1132 * mark it RW later, when the initial percpu area is freed. 1133 */ 1134 xen_initial_gdt = &per_cpu(gdt_page, 0); 1135 1136 xen_smp_init(); 1137 1138 pgd = (pgd_t *)xen_start_info->pt_base; 1139 1140 /* Don't do the full vcpu_info placement stuff until we have a 1141 possible map and a non-dummy shared_info. */ 1142 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; 1143 1144 local_irq_disable(); 1145 early_boot_irqs_off(); 1146 1147 xen_raw_console_write("mapping kernel into physical memory\n"); 1148 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); 1149 1150 init_mm.pgd = pgd; 1151 1152 /* keep using Xen gdt for now; no urgent need to change it */ 1153 1154 pv_info.kernel_rpl = 1; 1155 if (xen_feature(XENFEAT_supervisor_mode_kernel)) 1156 pv_info.kernel_rpl = 0; 1157 1158 /* set the limit of our address space */ 1159 xen_reserve_top(); 1160 1161 #ifdef CONFIG_X86_32 1162 /* set up basic CPUID stuff */ 1163 cpu_detect(&new_cpu_data); 1164 new_cpu_data.hard_math = 1; 1165 new_cpu_data.wp_works_ok = 1; 1166 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1167 #endif 1168 1169 /* Poke various useful things into boot_params */ 1170 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1171 boot_params.hdr.ramdisk_image = xen_start_info->mod_start 1172 ? __pa(xen_start_info->mod_start) : 0; 1173 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1174 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1175 1176 if (!xen_initial_domain()) { 1177 add_preferred_console("xenboot", 0, NULL); 1178 add_preferred_console("tty", 0, NULL); 1179 add_preferred_console("hvc", 0, NULL); 1180 } else { 1181 /* Make sure ACS will be enabled */ 1182 pci_request_acs(); 1183 } 1184 1185 1186 xen_raw_console_write("about to get started...\n"); 1187 1188 xen_setup_runstate_info(0); 1189 1190 /* Start the world */ 1191 #ifdef CONFIG_X86_32 1192 i386_start_kernel(); 1193 #else 1194 x86_64_start_reservations((char *)__pa_symbol(&boot_params)); 1195 #endif 1196 } 1197