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/bootmem.h> 24 #include <linux/module.h> 25 #include <linux/mm.h> 26 #include <linux/page-flags.h> 27 #include <linux/highmem.h> 28 #include <linux/console.h> 29 30 #include <xen/interface/xen.h> 31 #include <xen/interface/physdev.h> 32 #include <xen/interface/vcpu.h> 33 #include <xen/interface/sched.h> 34 #include <xen/features.h> 35 #include <xen/page.h> 36 37 #include <asm/paravirt.h> 38 #include <asm/page.h> 39 #include <asm/xen/hypercall.h> 40 #include <asm/xen/hypervisor.h> 41 #include <asm/fixmap.h> 42 #include <asm/processor.h> 43 #include <asm/setup.h> 44 #include <asm/desc.h> 45 #include <asm/pgtable.h> 46 #include <asm/tlbflush.h> 47 #include <asm/reboot.h> 48 49 #include "xen-ops.h" 50 #include "mmu.h" 51 #include "multicalls.h" 52 53 EXPORT_SYMBOL_GPL(hypercall_page); 54 55 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 56 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 57 58 /* 59 * Note about cr3 (pagetable base) values: 60 * 61 * xen_cr3 contains the current logical cr3 value; it contains the 62 * last set cr3. This may not be the current effective cr3, because 63 * its update may be being lazily deferred. However, a vcpu looking 64 * at its own cr3 can use this value knowing that it everything will 65 * be self-consistent. 66 * 67 * xen_current_cr3 contains the actual vcpu cr3; it is set once the 68 * hypercall to set the vcpu cr3 is complete (so it may be a little 69 * out of date, but it will never be set early). If one vcpu is 70 * looking at another vcpu's cr3 value, it should use this variable. 71 */ 72 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ 73 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ 74 75 struct start_info *xen_start_info; 76 EXPORT_SYMBOL_GPL(xen_start_info); 77 78 static /* __initdata */ struct shared_info dummy_shared_info; 79 80 /* 81 * Point at some empty memory to start with. We map the real shared_info 82 * page as soon as fixmap is up and running. 83 */ 84 struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info; 85 86 /* 87 * Flag to determine whether vcpu info placement is available on all 88 * VCPUs. We assume it is to start with, and then set it to zero on 89 * the first failure. This is because it can succeed on some VCPUs 90 * and not others, since it can involve hypervisor memory allocation, 91 * or because the guest failed to guarantee all the appropriate 92 * constraints on all VCPUs (ie buffer can't cross a page boundary). 93 * 94 * Note that any particular CPU may be using a placed vcpu structure, 95 * but we can only optimise if the all are. 96 * 97 * 0: not available, 1: available 98 */ 99 static int have_vcpu_info_placement = 1; 100 101 static void __init xen_vcpu_setup(int cpu) 102 { 103 struct vcpu_register_vcpu_info info; 104 int err; 105 struct vcpu_info *vcpup; 106 107 BUG_ON(HYPERVISOR_shared_info == &dummy_shared_info); 108 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 109 110 if (!have_vcpu_info_placement) 111 return; /* already tested, not available */ 112 113 vcpup = &per_cpu(xen_vcpu_info, cpu); 114 115 info.mfn = virt_to_mfn(vcpup); 116 info.offset = offset_in_page(vcpup); 117 118 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", 119 cpu, vcpup, info.mfn, info.offset); 120 121 /* Check to see if the hypervisor will put the vcpu_info 122 structure where we want it, which allows direct access via 123 a percpu-variable. */ 124 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 125 126 if (err) { 127 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); 128 have_vcpu_info_placement = 0; 129 } else { 130 /* This cpu is using the registered vcpu info, even if 131 later ones fail to. */ 132 per_cpu(xen_vcpu, cpu) = vcpup; 133 134 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", 135 cpu, vcpup); 136 } 137 } 138 139 static void __init xen_banner(void) 140 { 141 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 142 pv_info.name); 143 printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic); 144 } 145 146 static void xen_cpuid(unsigned int *ax, unsigned int *bx, 147 unsigned int *cx, unsigned int *dx) 148 { 149 unsigned maskedx = ~0; 150 151 /* 152 * Mask out inconvenient features, to try and disable as many 153 * unsupported kernel subsystems as possible. 154 */ 155 if (*ax == 1) 156 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ 157 (1 << X86_FEATURE_ACPI) | /* disable ACPI */ 158 (1 << X86_FEATURE_MCE) | /* disable MCE */ 159 (1 << X86_FEATURE_MCA) | /* disable MCA */ 160 (1 << X86_FEATURE_ACC)); /* thermal monitoring */ 161 162 asm(XEN_EMULATE_PREFIX "cpuid" 163 : "=a" (*ax), 164 "=b" (*bx), 165 "=c" (*cx), 166 "=d" (*dx) 167 : "0" (*ax), "2" (*cx)); 168 *dx &= maskedx; 169 } 170 171 static void xen_set_debugreg(int reg, unsigned long val) 172 { 173 HYPERVISOR_set_debugreg(reg, val); 174 } 175 176 static unsigned long xen_get_debugreg(int reg) 177 { 178 return HYPERVISOR_get_debugreg(reg); 179 } 180 181 static unsigned long xen_save_fl(void) 182 { 183 struct vcpu_info *vcpu; 184 unsigned long flags; 185 186 vcpu = x86_read_percpu(xen_vcpu); 187 188 /* flag has opposite sense of mask */ 189 flags = !vcpu->evtchn_upcall_mask; 190 191 /* convert to IF type flag 192 -0 -> 0x00000000 193 -1 -> 0xffffffff 194 */ 195 return (-flags) & X86_EFLAGS_IF; 196 } 197 198 static void xen_restore_fl(unsigned long flags) 199 { 200 struct vcpu_info *vcpu; 201 202 /* convert from IF type flag */ 203 flags = !(flags & X86_EFLAGS_IF); 204 205 /* There's a one instruction preempt window here. We need to 206 make sure we're don't switch CPUs between getting the vcpu 207 pointer and updating the mask. */ 208 preempt_disable(); 209 vcpu = x86_read_percpu(xen_vcpu); 210 vcpu->evtchn_upcall_mask = flags; 211 preempt_enable_no_resched(); 212 213 /* Doesn't matter if we get preempted here, because any 214 pending event will get dealt with anyway. */ 215 216 if (flags == 0) { 217 preempt_check_resched(); 218 barrier(); /* unmask then check (avoid races) */ 219 if (unlikely(vcpu->evtchn_upcall_pending)) 220 force_evtchn_callback(); 221 } 222 } 223 224 static void xen_irq_disable(void) 225 { 226 /* There's a one instruction preempt window here. We need to 227 make sure we're don't switch CPUs between getting the vcpu 228 pointer and updating the mask. */ 229 preempt_disable(); 230 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1; 231 preempt_enable_no_resched(); 232 } 233 234 static void xen_irq_enable(void) 235 { 236 struct vcpu_info *vcpu; 237 238 /* There's a one instruction preempt window here. We need to 239 make sure we're don't switch CPUs between getting the vcpu 240 pointer and updating the mask. */ 241 preempt_disable(); 242 vcpu = x86_read_percpu(xen_vcpu); 243 vcpu->evtchn_upcall_mask = 0; 244 preempt_enable_no_resched(); 245 246 /* Doesn't matter if we get preempted here, because any 247 pending event will get dealt with anyway. */ 248 249 barrier(); /* unmask then check (avoid races) */ 250 if (unlikely(vcpu->evtchn_upcall_pending)) 251 force_evtchn_callback(); 252 } 253 254 static void xen_safe_halt(void) 255 { 256 /* Blocking includes an implicit local_irq_enable(). */ 257 if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0) 258 BUG(); 259 } 260 261 static void xen_halt(void) 262 { 263 if (irqs_disabled()) 264 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); 265 else 266 xen_safe_halt(); 267 } 268 269 static void xen_leave_lazy(void) 270 { 271 paravirt_leave_lazy(paravirt_get_lazy_mode()); 272 xen_mc_flush(); 273 } 274 275 static unsigned long xen_store_tr(void) 276 { 277 return 0; 278 } 279 280 static void xen_set_ldt(const void *addr, unsigned entries) 281 { 282 struct mmuext_op *op; 283 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 284 285 op = mcs.args; 286 op->cmd = MMUEXT_SET_LDT; 287 op->arg1.linear_addr = (unsigned long)addr; 288 op->arg2.nr_ents = entries; 289 290 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 291 292 xen_mc_issue(PARAVIRT_LAZY_CPU); 293 } 294 295 static void xen_load_gdt(const struct desc_ptr *dtr) 296 { 297 unsigned long *frames; 298 unsigned long va = dtr->address; 299 unsigned int size = dtr->size + 1; 300 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 301 int f; 302 struct multicall_space mcs; 303 304 /* A GDT can be up to 64k in size, which corresponds to 8192 305 8-byte entries, or 16 4k pages.. */ 306 307 BUG_ON(size > 65536); 308 BUG_ON(va & ~PAGE_MASK); 309 310 mcs = xen_mc_entry(sizeof(*frames) * pages); 311 frames = mcs.args; 312 313 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { 314 frames[f] = virt_to_mfn(va); 315 make_lowmem_page_readonly((void *)va); 316 } 317 318 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); 319 320 xen_mc_issue(PARAVIRT_LAZY_CPU); 321 } 322 323 static void load_TLS_descriptor(struct thread_struct *t, 324 unsigned int cpu, unsigned int i) 325 { 326 struct desc_struct *gdt = get_cpu_gdt_table(cpu); 327 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 328 struct multicall_space mc = __xen_mc_entry(0); 329 330 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 331 } 332 333 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 334 { 335 xen_mc_batch(); 336 337 load_TLS_descriptor(t, cpu, 0); 338 load_TLS_descriptor(t, cpu, 1); 339 load_TLS_descriptor(t, cpu, 2); 340 341 xen_mc_issue(PARAVIRT_LAZY_CPU); 342 343 /* 344 * XXX sleazy hack: If we're being called in a lazy-cpu zone, 345 * it means we're in a context switch, and %gs has just been 346 * saved. This means we can zero it out to prevent faults on 347 * exit from the hypervisor if the next process has no %gs. 348 * Either way, it has been saved, and the new value will get 349 * loaded properly. This will go away as soon as Xen has been 350 * modified to not save/restore %gs for normal hypercalls. 351 */ 352 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) 353 loadsegment(gs, 0); 354 } 355 356 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 357 const void *ptr) 358 { 359 unsigned long lp = (unsigned long)&dt[entrynum]; 360 xmaddr_t mach_lp = virt_to_machine(lp); 361 u64 entry = *(u64 *)ptr; 362 363 preempt_disable(); 364 365 xen_mc_flush(); 366 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 367 BUG(); 368 369 preempt_enable(); 370 } 371 372 static int cvt_gate_to_trap(int vector, u32 low, u32 high, 373 struct trap_info *info) 374 { 375 u8 type, dpl; 376 377 type = (high >> 8) & 0x1f; 378 dpl = (high >> 13) & 3; 379 380 if (type != 0xf && type != 0xe) 381 return 0; 382 383 info->vector = vector; 384 info->address = (high & 0xffff0000) | (low & 0x0000ffff); 385 info->cs = low >> 16; 386 info->flags = dpl; 387 /* interrupt gates clear IF */ 388 if (type == 0xe) 389 info->flags |= 4; 390 391 return 1; 392 } 393 394 /* Locations of each CPU's IDT */ 395 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 396 397 /* Set an IDT entry. If the entry is part of the current IDT, then 398 also update Xen. */ 399 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 400 { 401 unsigned long p = (unsigned long)&dt[entrynum]; 402 unsigned long start, end; 403 404 preempt_disable(); 405 406 start = __get_cpu_var(idt_desc).address; 407 end = start + __get_cpu_var(idt_desc).size + 1; 408 409 xen_mc_flush(); 410 411 native_write_idt_entry(dt, entrynum, g); 412 413 if (p >= start && (p + 8) <= end) { 414 struct trap_info info[2]; 415 u32 *desc = (u32 *)g; 416 417 info[1].address = 0; 418 419 if (cvt_gate_to_trap(entrynum, desc[0], desc[1], &info[0])) 420 if (HYPERVISOR_set_trap_table(info)) 421 BUG(); 422 } 423 424 preempt_enable(); 425 } 426 427 static void xen_convert_trap_info(const struct desc_ptr *desc, 428 struct trap_info *traps) 429 { 430 unsigned in, out, count; 431 432 count = (desc->size+1) / 8; 433 BUG_ON(count > 256); 434 435 for (in = out = 0; in < count; in++) { 436 const u32 *entry = (u32 *)(desc->address + in * 8); 437 438 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out])) 439 out++; 440 } 441 traps[out].address = 0; 442 } 443 444 void xen_copy_trap_info(struct trap_info *traps) 445 { 446 const struct desc_ptr *desc = &__get_cpu_var(idt_desc); 447 448 xen_convert_trap_info(desc, traps); 449 } 450 451 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 452 hold a spinlock to protect the static traps[] array (static because 453 it avoids allocation, and saves stack space). */ 454 static void xen_load_idt(const struct desc_ptr *desc) 455 { 456 static DEFINE_SPINLOCK(lock); 457 static struct trap_info traps[257]; 458 459 spin_lock(&lock); 460 461 __get_cpu_var(idt_desc) = *desc; 462 463 xen_convert_trap_info(desc, traps); 464 465 xen_mc_flush(); 466 if (HYPERVISOR_set_trap_table(traps)) 467 BUG(); 468 469 spin_unlock(&lock); 470 } 471 472 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 473 they're handled differently. */ 474 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 475 const void *desc, int type) 476 { 477 preempt_disable(); 478 479 switch (type) { 480 case DESC_LDT: 481 case DESC_TSS: 482 /* ignore */ 483 break; 484 485 default: { 486 xmaddr_t maddr = virt_to_machine(&dt[entry]); 487 488 xen_mc_flush(); 489 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 490 BUG(); 491 } 492 493 } 494 495 preempt_enable(); 496 } 497 498 static void xen_load_sp0(struct tss_struct *tss, 499 struct thread_struct *thread) 500 { 501 struct multicall_space mcs = xen_mc_entry(0); 502 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 503 xen_mc_issue(PARAVIRT_LAZY_CPU); 504 } 505 506 static void xen_set_iopl_mask(unsigned mask) 507 { 508 struct physdev_set_iopl set_iopl; 509 510 /* Force the change at ring 0. */ 511 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; 512 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 513 } 514 515 static void xen_io_delay(void) 516 { 517 } 518 519 #ifdef CONFIG_X86_LOCAL_APIC 520 static u32 xen_apic_read(unsigned long reg) 521 { 522 return 0; 523 } 524 525 static void xen_apic_write(unsigned long reg, u32 val) 526 { 527 /* Warn to see if there's any stray references */ 528 WARN_ON(1); 529 } 530 #endif 531 532 static void xen_flush_tlb(void) 533 { 534 struct mmuext_op *op; 535 struct multicall_space mcs; 536 537 preempt_disable(); 538 539 mcs = xen_mc_entry(sizeof(*op)); 540 541 op = mcs.args; 542 op->cmd = MMUEXT_TLB_FLUSH_LOCAL; 543 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 544 545 xen_mc_issue(PARAVIRT_LAZY_MMU); 546 547 preempt_enable(); 548 } 549 550 static void xen_flush_tlb_single(unsigned long addr) 551 { 552 struct mmuext_op *op; 553 struct multicall_space mcs; 554 555 preempt_disable(); 556 557 mcs = xen_mc_entry(sizeof(*op)); 558 op = mcs.args; 559 op->cmd = MMUEXT_INVLPG_LOCAL; 560 op->arg1.linear_addr = addr & PAGE_MASK; 561 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 562 563 xen_mc_issue(PARAVIRT_LAZY_MMU); 564 565 preempt_enable(); 566 } 567 568 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm, 569 unsigned long va) 570 { 571 struct { 572 struct mmuext_op op; 573 cpumask_t mask; 574 } *args; 575 cpumask_t cpumask = *cpus; 576 struct multicall_space mcs; 577 578 /* 579 * A couple of (to be removed) sanity checks: 580 * 581 * - current CPU must not be in mask 582 * - mask must exist :) 583 */ 584 BUG_ON(cpus_empty(cpumask)); 585 BUG_ON(cpu_isset(smp_processor_id(), cpumask)); 586 BUG_ON(!mm); 587 588 /* If a CPU which we ran on has gone down, OK. */ 589 cpus_and(cpumask, cpumask, cpu_online_map); 590 if (cpus_empty(cpumask)) 591 return; 592 593 mcs = xen_mc_entry(sizeof(*args)); 594 args = mcs.args; 595 args->mask = cpumask; 596 args->op.arg2.vcpumask = &args->mask; 597 598 if (va == TLB_FLUSH_ALL) { 599 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; 600 } else { 601 args->op.cmd = MMUEXT_INVLPG_MULTI; 602 args->op.arg1.linear_addr = va; 603 } 604 605 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); 606 607 xen_mc_issue(PARAVIRT_LAZY_MMU); 608 } 609 610 static void xen_write_cr2(unsigned long cr2) 611 { 612 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2; 613 } 614 615 static unsigned long xen_read_cr2(void) 616 { 617 return x86_read_percpu(xen_vcpu)->arch.cr2; 618 } 619 620 static unsigned long xen_read_cr2_direct(void) 621 { 622 return x86_read_percpu(xen_vcpu_info.arch.cr2); 623 } 624 625 static void xen_write_cr4(unsigned long cr4) 626 { 627 /* Just ignore cr4 changes; Xen doesn't allow us to do 628 anything anyway. */ 629 } 630 631 static unsigned long xen_read_cr3(void) 632 { 633 return x86_read_percpu(xen_cr3); 634 } 635 636 static void set_current_cr3(void *v) 637 { 638 x86_write_percpu(xen_current_cr3, (unsigned long)v); 639 } 640 641 static void xen_write_cr3(unsigned long cr3) 642 { 643 struct mmuext_op *op; 644 struct multicall_space mcs; 645 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3)); 646 647 BUG_ON(preemptible()); 648 649 mcs = xen_mc_entry(sizeof(*op)); /* disables interrupts */ 650 651 /* Update while interrupts are disabled, so its atomic with 652 respect to ipis */ 653 x86_write_percpu(xen_cr3, cr3); 654 655 op = mcs.args; 656 op->cmd = MMUEXT_NEW_BASEPTR; 657 op->arg1.mfn = mfn; 658 659 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 660 661 /* Update xen_update_cr3 once the batch has actually 662 been submitted. */ 663 xen_mc_callback(set_current_cr3, (void *)cr3); 664 665 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ 666 } 667 668 /* Early in boot, while setting up the initial pagetable, assume 669 everything is pinned. */ 670 static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn) 671 { 672 #ifdef CONFIG_FLATMEM 673 BUG_ON(mem_map); /* should only be used early */ 674 #endif 675 make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); 676 } 677 678 /* Early release_pte assumes that all pts are pinned, since there's 679 only init_mm and anything attached to that is pinned. */ 680 static void xen_release_pte_init(u32 pfn) 681 { 682 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); 683 } 684 685 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) 686 { 687 struct mmuext_op op; 688 op.cmd = cmd; 689 op.arg1.mfn = pfn_to_mfn(pfn); 690 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) 691 BUG(); 692 } 693 694 /* This needs to make sure the new pte page is pinned iff its being 695 attached to a pinned pagetable. */ 696 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level) 697 { 698 struct page *page = pfn_to_page(pfn); 699 700 if (PagePinned(virt_to_page(mm->pgd))) { 701 SetPagePinned(page); 702 703 if (!PageHighMem(page)) { 704 make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); 705 if (level == PT_PTE) 706 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); 707 } else 708 /* make sure there are no stray mappings of 709 this page */ 710 kmap_flush_unused(); 711 } 712 } 713 714 static void xen_alloc_pte(struct mm_struct *mm, u32 pfn) 715 { 716 xen_alloc_ptpage(mm, pfn, PT_PTE); 717 } 718 719 static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn) 720 { 721 xen_alloc_ptpage(mm, pfn, PT_PMD); 722 } 723 724 /* This should never happen until we're OK to use struct page */ 725 static void xen_release_ptpage(u32 pfn, unsigned level) 726 { 727 struct page *page = pfn_to_page(pfn); 728 729 if (PagePinned(page)) { 730 if (!PageHighMem(page)) { 731 if (level == PT_PTE) 732 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); 733 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); 734 } 735 ClearPagePinned(page); 736 } 737 } 738 739 static void xen_release_pte(u32 pfn) 740 { 741 xen_release_ptpage(pfn, PT_PTE); 742 } 743 744 static void xen_release_pmd(u32 pfn) 745 { 746 xen_release_ptpage(pfn, PT_PMD); 747 } 748 749 #ifdef CONFIG_HIGHPTE 750 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) 751 { 752 pgprot_t prot = PAGE_KERNEL; 753 754 if (PagePinned(page)) 755 prot = PAGE_KERNEL_RO; 756 757 if (0 && PageHighMem(page)) 758 printk("mapping highpte %lx type %d prot %s\n", 759 page_to_pfn(page), type, 760 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); 761 762 return kmap_atomic_prot(page, type, prot); 763 } 764 #endif 765 766 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) 767 { 768 /* If there's an existing pte, then don't allow _PAGE_RW to be set */ 769 if (pte_val_ma(*ptep) & _PAGE_PRESENT) 770 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & 771 pte_val_ma(pte)); 772 773 return pte; 774 } 775 776 /* Init-time set_pte while constructing initial pagetables, which 777 doesn't allow RO pagetable pages to be remapped RW */ 778 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) 779 { 780 pte = mask_rw_pte(ptep, pte); 781 782 xen_set_pte(ptep, pte); 783 } 784 785 static __init void xen_pagetable_setup_start(pgd_t *base) 786 { 787 pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base; 788 789 /* special set_pte for pagetable initialization */ 790 pv_mmu_ops.set_pte = xen_set_pte_init; 791 792 init_mm.pgd = base; 793 /* 794 * copy top-level of Xen-supplied pagetable into place. For 795 * !PAE we can use this as-is, but for PAE it is a stand-in 796 * while we copy the pmd pages. 797 */ 798 memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t)); 799 800 if (PTRS_PER_PMD > 1) { 801 int i; 802 /* 803 * For PAE, need to allocate new pmds, rather than 804 * share Xen's, since Xen doesn't like pmd's being 805 * shared between address spaces. 806 */ 807 for (i = 0; i < PTRS_PER_PGD; i++) { 808 if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) { 809 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE); 810 811 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]), 812 PAGE_SIZE); 813 814 make_lowmem_page_readonly(pmd); 815 816 set_pgd(&base[i], __pgd(1 + __pa(pmd))); 817 } else 818 pgd_clear(&base[i]); 819 } 820 } 821 822 /* make sure zero_page is mapped RO so we can use it in pagetables */ 823 make_lowmem_page_readonly(empty_zero_page); 824 make_lowmem_page_readonly(base); 825 /* 826 * Switch to new pagetable. This is done before 827 * pagetable_init has done anything so that the new pages 828 * added to the table can be prepared properly for Xen. 829 */ 830 xen_write_cr3(__pa(base)); 831 832 /* Unpin initial Xen pagetable */ 833 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, 834 PFN_DOWN(__pa(xen_start_info->pt_base))); 835 } 836 837 static __init void setup_shared_info(void) 838 { 839 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 840 unsigned long addr = fix_to_virt(FIX_PARAVIRT_BOOTMAP); 841 842 /* 843 * Create a mapping for the shared info page. 844 * Should be set_fixmap(), but shared_info is a machine 845 * address with no corresponding pseudo-phys address. 846 */ 847 set_pte_mfn(addr, 848 PFN_DOWN(xen_start_info->shared_info), 849 PAGE_KERNEL); 850 851 HYPERVISOR_shared_info = (struct shared_info *)addr; 852 } else 853 HYPERVISOR_shared_info = 854 (struct shared_info *)__va(xen_start_info->shared_info); 855 856 #ifndef CONFIG_SMP 857 /* In UP this is as good a place as any to set up shared info */ 858 xen_setup_vcpu_info_placement(); 859 #endif 860 } 861 862 static __init void xen_pagetable_setup_done(pgd_t *base) 863 { 864 /* This will work as long as patching hasn't happened yet 865 (which it hasn't) */ 866 pv_mmu_ops.alloc_pte = xen_alloc_pte; 867 pv_mmu_ops.alloc_pmd = xen_alloc_pmd; 868 pv_mmu_ops.release_pte = xen_release_pte; 869 pv_mmu_ops.release_pmd = xen_release_pmd; 870 pv_mmu_ops.set_pte = xen_set_pte; 871 872 setup_shared_info(); 873 874 /* Actually pin the pagetable down, but we can't set PG_pinned 875 yet because the page structures don't exist yet. */ 876 { 877 unsigned level; 878 879 #ifdef CONFIG_X86_PAE 880 level = MMUEXT_PIN_L3_TABLE; 881 #else 882 level = MMUEXT_PIN_L2_TABLE; 883 #endif 884 885 pin_pagetable_pfn(level, PFN_DOWN(__pa(base))); 886 } 887 } 888 889 /* This is called once we have the cpu_possible_map */ 890 void __init xen_setup_vcpu_info_placement(void) 891 { 892 int cpu; 893 894 for_each_possible_cpu(cpu) 895 xen_vcpu_setup(cpu); 896 897 /* xen_vcpu_setup managed to place the vcpu_info within the 898 percpu area for all cpus, so make use of it */ 899 if (have_vcpu_info_placement) { 900 printk(KERN_INFO "Xen: using vcpu_info placement\n"); 901 902 pv_irq_ops.save_fl = xen_save_fl_direct; 903 pv_irq_ops.restore_fl = xen_restore_fl_direct; 904 pv_irq_ops.irq_disable = xen_irq_disable_direct; 905 pv_irq_ops.irq_enable = xen_irq_enable_direct; 906 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 907 } 908 } 909 910 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 911 unsigned long addr, unsigned len) 912 { 913 char *start, *end, *reloc; 914 unsigned ret; 915 916 start = end = reloc = NULL; 917 918 #define SITE(op, x) \ 919 case PARAVIRT_PATCH(op.x): \ 920 if (have_vcpu_info_placement) { \ 921 start = (char *)xen_##x##_direct; \ 922 end = xen_##x##_direct_end; \ 923 reloc = xen_##x##_direct_reloc; \ 924 } \ 925 goto patch_site 926 927 switch (type) { 928 SITE(pv_irq_ops, irq_enable); 929 SITE(pv_irq_ops, irq_disable); 930 SITE(pv_irq_ops, save_fl); 931 SITE(pv_irq_ops, restore_fl); 932 #undef SITE 933 934 patch_site: 935 if (start == NULL || (end-start) > len) 936 goto default_patch; 937 938 ret = paravirt_patch_insns(insnbuf, len, start, end); 939 940 /* Note: because reloc is assigned from something that 941 appears to be an array, gcc assumes it's non-null, 942 but doesn't know its relationship with start and 943 end. */ 944 if (reloc > start && reloc < end) { 945 int reloc_off = reloc - start; 946 long *relocp = (long *)(insnbuf + reloc_off); 947 long delta = start - (char *)addr; 948 949 *relocp += delta; 950 } 951 break; 952 953 default_patch: 954 default: 955 ret = paravirt_patch_default(type, clobbers, insnbuf, 956 addr, len); 957 break; 958 } 959 960 return ret; 961 } 962 963 static const struct pv_info xen_info __initdata = { 964 .paravirt_enabled = 1, 965 .shared_kernel_pmd = 0, 966 967 .name = "Xen", 968 }; 969 970 static const struct pv_init_ops xen_init_ops __initdata = { 971 .patch = xen_patch, 972 973 .banner = xen_banner, 974 .memory_setup = xen_memory_setup, 975 .arch_setup = xen_arch_setup, 976 .post_allocator_init = xen_mark_init_mm_pinned, 977 }; 978 979 static const struct pv_time_ops xen_time_ops __initdata = { 980 .time_init = xen_time_init, 981 982 .set_wallclock = xen_set_wallclock, 983 .get_wallclock = xen_get_wallclock, 984 .get_cpu_khz = xen_cpu_khz, 985 .sched_clock = xen_sched_clock, 986 }; 987 988 static const struct pv_cpu_ops xen_cpu_ops __initdata = { 989 .cpuid = xen_cpuid, 990 991 .set_debugreg = xen_set_debugreg, 992 .get_debugreg = xen_get_debugreg, 993 994 .clts = native_clts, 995 996 .read_cr0 = native_read_cr0, 997 .write_cr0 = native_write_cr0, 998 999 .read_cr4 = native_read_cr4, 1000 .read_cr4_safe = native_read_cr4_safe, 1001 .write_cr4 = xen_write_cr4, 1002 1003 .wbinvd = native_wbinvd, 1004 1005 .read_msr = native_read_msr_safe, 1006 .write_msr = native_write_msr_safe, 1007 .read_tsc = native_read_tsc, 1008 .read_pmc = native_read_pmc, 1009 1010 .iret = xen_iret, 1011 .irq_enable_syscall_ret = xen_sysexit, 1012 1013 .load_tr_desc = paravirt_nop, 1014 .set_ldt = xen_set_ldt, 1015 .load_gdt = xen_load_gdt, 1016 .load_idt = xen_load_idt, 1017 .load_tls = xen_load_tls, 1018 1019 .store_gdt = native_store_gdt, 1020 .store_idt = native_store_idt, 1021 .store_tr = xen_store_tr, 1022 1023 .write_ldt_entry = xen_write_ldt_entry, 1024 .write_gdt_entry = xen_write_gdt_entry, 1025 .write_idt_entry = xen_write_idt_entry, 1026 .load_sp0 = xen_load_sp0, 1027 1028 .set_iopl_mask = xen_set_iopl_mask, 1029 .io_delay = xen_io_delay, 1030 1031 .lazy_mode = { 1032 .enter = paravirt_enter_lazy_cpu, 1033 .leave = xen_leave_lazy, 1034 }, 1035 }; 1036 1037 static const struct pv_irq_ops xen_irq_ops __initdata = { 1038 .init_IRQ = xen_init_IRQ, 1039 .save_fl = xen_save_fl, 1040 .restore_fl = xen_restore_fl, 1041 .irq_disable = xen_irq_disable, 1042 .irq_enable = xen_irq_enable, 1043 .safe_halt = xen_safe_halt, 1044 .halt = xen_halt, 1045 }; 1046 1047 static const struct pv_apic_ops xen_apic_ops __initdata = { 1048 #ifdef CONFIG_X86_LOCAL_APIC 1049 .apic_write = xen_apic_write, 1050 .apic_write_atomic = xen_apic_write, 1051 .apic_read = xen_apic_read, 1052 .setup_boot_clock = paravirt_nop, 1053 .setup_secondary_clock = paravirt_nop, 1054 .startup_ipi_hook = paravirt_nop, 1055 #endif 1056 }; 1057 1058 static const struct pv_mmu_ops xen_mmu_ops __initdata = { 1059 .pagetable_setup_start = xen_pagetable_setup_start, 1060 .pagetable_setup_done = xen_pagetable_setup_done, 1061 1062 .read_cr2 = xen_read_cr2, 1063 .write_cr2 = xen_write_cr2, 1064 1065 .read_cr3 = xen_read_cr3, 1066 .write_cr3 = xen_write_cr3, 1067 1068 .flush_tlb_user = xen_flush_tlb, 1069 .flush_tlb_kernel = xen_flush_tlb, 1070 .flush_tlb_single = xen_flush_tlb_single, 1071 .flush_tlb_others = xen_flush_tlb_others, 1072 1073 .pte_update = paravirt_nop, 1074 .pte_update_defer = paravirt_nop, 1075 1076 .alloc_pte = xen_alloc_pte_init, 1077 .release_pte = xen_release_pte_init, 1078 .alloc_pmd = xen_alloc_pte_init, 1079 .alloc_pmd_clone = paravirt_nop, 1080 .release_pmd = xen_release_pte_init, 1081 1082 #ifdef CONFIG_HIGHPTE 1083 .kmap_atomic_pte = xen_kmap_atomic_pte, 1084 #endif 1085 1086 .set_pte = NULL, /* see xen_pagetable_setup_* */ 1087 .set_pte_at = xen_set_pte_at, 1088 .set_pmd = xen_set_pmd, 1089 1090 .pte_val = xen_pte_val, 1091 .pgd_val = xen_pgd_val, 1092 1093 .make_pte = xen_make_pte, 1094 .make_pgd = xen_make_pgd, 1095 1096 #ifdef CONFIG_X86_PAE 1097 .set_pte_atomic = xen_set_pte_atomic, 1098 .set_pte_present = xen_set_pte_at, 1099 .set_pud = xen_set_pud, 1100 .pte_clear = xen_pte_clear, 1101 .pmd_clear = xen_pmd_clear, 1102 1103 .make_pmd = xen_make_pmd, 1104 .pmd_val = xen_pmd_val, 1105 #endif /* PAE */ 1106 1107 .activate_mm = xen_activate_mm, 1108 .dup_mmap = xen_dup_mmap, 1109 .exit_mmap = xen_exit_mmap, 1110 1111 .lazy_mode = { 1112 .enter = paravirt_enter_lazy_mmu, 1113 .leave = xen_leave_lazy, 1114 }, 1115 }; 1116 1117 #ifdef CONFIG_SMP 1118 static const struct smp_ops xen_smp_ops __initdata = { 1119 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, 1120 .smp_prepare_cpus = xen_smp_prepare_cpus, 1121 .cpu_up = xen_cpu_up, 1122 .smp_cpus_done = xen_smp_cpus_done, 1123 1124 .smp_send_stop = xen_smp_send_stop, 1125 .smp_send_reschedule = xen_smp_send_reschedule, 1126 .smp_call_function_mask = xen_smp_call_function_mask, 1127 }; 1128 #endif /* CONFIG_SMP */ 1129 1130 static void xen_reboot(int reason) 1131 { 1132 #ifdef CONFIG_SMP 1133 smp_send_stop(); 1134 #endif 1135 1136 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason)) 1137 BUG(); 1138 } 1139 1140 static void xen_restart(char *msg) 1141 { 1142 xen_reboot(SHUTDOWN_reboot); 1143 } 1144 1145 static void xen_emergency_restart(void) 1146 { 1147 xen_reboot(SHUTDOWN_reboot); 1148 } 1149 1150 static void xen_machine_halt(void) 1151 { 1152 xen_reboot(SHUTDOWN_poweroff); 1153 } 1154 1155 static void xen_crash_shutdown(struct pt_regs *regs) 1156 { 1157 xen_reboot(SHUTDOWN_crash); 1158 } 1159 1160 static const struct machine_ops __initdata xen_machine_ops = { 1161 .restart = xen_restart, 1162 .halt = xen_machine_halt, 1163 .power_off = xen_machine_halt, 1164 .shutdown = xen_machine_halt, 1165 .crash_shutdown = xen_crash_shutdown, 1166 .emergency_restart = xen_emergency_restart, 1167 }; 1168 1169 1170 static void __init xen_reserve_top(void) 1171 { 1172 unsigned long top = HYPERVISOR_VIRT_START; 1173 struct xen_platform_parameters pp; 1174 1175 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) 1176 top = pp.virt_start; 1177 1178 reserve_top_address(-top + 2 * PAGE_SIZE); 1179 } 1180 1181 /* First C function to be called on Xen boot */ 1182 asmlinkage void __init xen_start_kernel(void) 1183 { 1184 pgd_t *pgd; 1185 1186 if (!xen_start_info) 1187 return; 1188 1189 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); 1190 1191 /* Install Xen paravirt ops */ 1192 pv_info = xen_info; 1193 pv_init_ops = xen_init_ops; 1194 pv_time_ops = xen_time_ops; 1195 pv_cpu_ops = xen_cpu_ops; 1196 pv_irq_ops = xen_irq_ops; 1197 pv_apic_ops = xen_apic_ops; 1198 pv_mmu_ops = xen_mmu_ops; 1199 1200 machine_ops = xen_machine_ops; 1201 1202 #ifdef CONFIG_SMP 1203 smp_ops = xen_smp_ops; 1204 #endif 1205 1206 xen_setup_features(); 1207 1208 /* Get mfn list */ 1209 if (!xen_feature(XENFEAT_auto_translated_physmap)) 1210 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list; 1211 1212 pgd = (pgd_t *)xen_start_info->pt_base; 1213 1214 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; 1215 1216 init_mm.pgd = pgd; /* use the Xen pagetables to start */ 1217 1218 /* keep using Xen gdt for now; no urgent need to change it */ 1219 1220 x86_write_percpu(xen_cr3, __pa(pgd)); 1221 x86_write_percpu(xen_current_cr3, __pa(pgd)); 1222 1223 /* Don't do the full vcpu_info placement stuff until we have a 1224 possible map and a non-dummy shared_info. */ 1225 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; 1226 1227 pv_info.kernel_rpl = 1; 1228 if (xen_feature(XENFEAT_supervisor_mode_kernel)) 1229 pv_info.kernel_rpl = 0; 1230 1231 /* set the limit of our address space */ 1232 xen_reserve_top(); 1233 1234 /* set up basic CPUID stuff */ 1235 cpu_detect(&new_cpu_data); 1236 new_cpu_data.hard_math = 1; 1237 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1238 1239 /* Poke various useful things into boot_params */ 1240 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1241 boot_params.hdr.ramdisk_image = xen_start_info->mod_start 1242 ? __pa(xen_start_info->mod_start) : 0; 1243 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1244 1245 if (!is_initial_xendomain()) 1246 add_preferred_console("hvc", 0, NULL); 1247 1248 /* Start the world */ 1249 start_kernel(); 1250 } 1251