1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Core of Xen paravirt_ops implementation. 4 * 5 * This file contains the xen_paravirt_ops structure itself, and the 6 * implementations for: 7 * - privileged instructions 8 * - interrupt flags 9 * - segment operations 10 * - booting and setup 11 * 12 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 13 */ 14 15 #include <linux/cpu.h> 16 #include <linux/kernel.h> 17 #include <linux/init.h> 18 #include <linux/smp.h> 19 #include <linux/preempt.h> 20 #include <linux/hardirq.h> 21 #include <linux/percpu.h> 22 #include <linux/delay.h> 23 #include <linux/start_kernel.h> 24 #include <linux/sched.h> 25 #include <linux/kprobes.h> 26 #include <linux/memblock.h> 27 #include <linux/export.h> 28 #include <linux/mm.h> 29 #include <linux/page-flags.h> 30 #include <linux/highmem.h> 31 #include <linux/console.h> 32 #include <linux/pci.h> 33 #include <linux/gfp.h> 34 #include <linux/edd.h> 35 #include <linux/objtool.h> 36 37 #include <xen/xen.h> 38 #include <xen/events.h> 39 #include <xen/interface/xen.h> 40 #include <xen/interface/version.h> 41 #include <xen/interface/physdev.h> 42 #include <xen/interface/vcpu.h> 43 #include <xen/interface/memory.h> 44 #include <xen/interface/nmi.h> 45 #include <xen/interface/xen-mca.h> 46 #include <xen/features.h> 47 #include <xen/page.h> 48 #include <xen/hvc-console.h> 49 #include <xen/acpi.h> 50 51 #include <asm/paravirt.h> 52 #include <asm/apic.h> 53 #include <asm/page.h> 54 #include <asm/xen/pci.h> 55 #include <asm/xen/hypercall.h> 56 #include <asm/xen/hypervisor.h> 57 #include <asm/xen/cpuid.h> 58 #include <asm/fixmap.h> 59 #include <asm/processor.h> 60 #include <asm/proto.h> 61 #include <asm/msr-index.h> 62 #include <asm/traps.h> 63 #include <asm/setup.h> 64 #include <asm/desc.h> 65 #include <asm/pgalloc.h> 66 #include <asm/tlbflush.h> 67 #include <asm/reboot.h> 68 #include <asm/stackprotector.h> 69 #include <asm/hypervisor.h> 70 #include <asm/mach_traps.h> 71 #include <asm/mwait.h> 72 #include <asm/pci_x86.h> 73 #include <asm/cpu.h> 74 #ifdef CONFIG_X86_IOPL_IOPERM 75 #include <asm/io_bitmap.h> 76 #endif 77 78 #ifdef CONFIG_ACPI 79 #include <linux/acpi.h> 80 #include <asm/acpi.h> 81 #include <acpi/pdc_intel.h> 82 #include <acpi/processor.h> 83 #include <xen/interface/platform.h> 84 #endif 85 86 #include "xen-ops.h" 87 #include "mmu.h" 88 #include "smp.h" 89 #include "multicalls.h" 90 #include "pmu.h" 91 92 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */ 93 94 void *xen_initial_gdt; 95 96 static int xen_cpu_up_prepare_pv(unsigned int cpu); 97 static int xen_cpu_dead_pv(unsigned int cpu); 98 99 struct tls_descs { 100 struct desc_struct desc[3]; 101 }; 102 103 /* 104 * Updating the 3 TLS descriptors in the GDT on every task switch is 105 * surprisingly expensive so we avoid updating them if they haven't 106 * changed. Since Xen writes different descriptors than the one 107 * passed in the update_descriptor hypercall we keep shadow copies to 108 * compare against. 109 */ 110 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc); 111 112 static void __init xen_banner(void) 113 { 114 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); 115 struct xen_extraversion extra; 116 HYPERVISOR_xen_version(XENVER_extraversion, &extra); 117 118 pr_info("Booting paravirtualized kernel on %s\n", pv_info.name); 119 pr_info("Xen version: %d.%d%s (preserve-AD)\n", 120 version >> 16, version & 0xffff, extra.extraversion); 121 } 122 123 static void __init xen_pv_init_platform(void) 124 { 125 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP)); 126 127 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info); 128 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); 129 130 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */ 131 xen_vcpu_info_reset(0); 132 133 /* pvclock is in shared info area */ 134 xen_init_time_ops(); 135 } 136 137 static void __init xen_pv_guest_late_init(void) 138 { 139 #ifndef CONFIG_SMP 140 /* Setup shared vcpu info for non-smp configurations */ 141 xen_setup_vcpu_info_placement(); 142 #endif 143 } 144 145 /* Check if running on Xen version (major, minor) or later */ 146 bool 147 xen_running_on_version_or_later(unsigned int major, unsigned int minor) 148 { 149 unsigned int version; 150 151 if (!xen_domain()) 152 return false; 153 154 version = HYPERVISOR_xen_version(XENVER_version, NULL); 155 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) || 156 ((version >> 16) > major)) 157 return true; 158 return false; 159 } 160 161 static __read_mostly unsigned int cpuid_leaf5_ecx_val; 162 static __read_mostly unsigned int cpuid_leaf5_edx_val; 163 164 static void xen_cpuid(unsigned int *ax, unsigned int *bx, 165 unsigned int *cx, unsigned int *dx) 166 { 167 unsigned maskebx = ~0; 168 169 /* 170 * Mask out inconvenient features, to try and disable as many 171 * unsupported kernel subsystems as possible. 172 */ 173 switch (*ax) { 174 case CPUID_MWAIT_LEAF: 175 /* Synthesize the values.. */ 176 *ax = 0; 177 *bx = 0; 178 *cx = cpuid_leaf5_ecx_val; 179 *dx = cpuid_leaf5_edx_val; 180 return; 181 182 case 0xb: 183 /* Suppress extended topology stuff */ 184 maskebx = 0; 185 break; 186 } 187 188 asm(XEN_EMULATE_PREFIX "cpuid" 189 : "=a" (*ax), 190 "=b" (*bx), 191 "=c" (*cx), 192 "=d" (*dx) 193 : "0" (*ax), "2" (*cx)); 194 195 *bx &= maskebx; 196 } 197 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */ 198 199 static bool __init xen_check_mwait(void) 200 { 201 #ifdef CONFIG_ACPI 202 struct xen_platform_op op = { 203 .cmd = XENPF_set_processor_pminfo, 204 .u.set_pminfo.id = -1, 205 .u.set_pminfo.type = XEN_PM_PDC, 206 }; 207 uint32_t buf[3]; 208 unsigned int ax, bx, cx, dx; 209 unsigned int mwait_mask; 210 211 /* We need to determine whether it is OK to expose the MWAIT 212 * capability to the kernel to harvest deeper than C3 states from ACPI 213 * _CST using the processor_harvest_xen.c module. For this to work, we 214 * need to gather the MWAIT_LEAF values (which the cstate.c code 215 * checks against). The hypervisor won't expose the MWAIT flag because 216 * it would break backwards compatibility; so we will find out directly 217 * from the hardware and hypercall. 218 */ 219 if (!xen_initial_domain()) 220 return false; 221 222 /* 223 * When running under platform earlier than Xen4.2, do not expose 224 * mwait, to avoid the risk of loading native acpi pad driver 225 */ 226 if (!xen_running_on_version_or_later(4, 2)) 227 return false; 228 229 ax = 1; 230 cx = 0; 231 232 native_cpuid(&ax, &bx, &cx, &dx); 233 234 mwait_mask = (1 << (X86_FEATURE_EST % 32)) | 235 (1 << (X86_FEATURE_MWAIT % 32)); 236 237 if ((cx & mwait_mask) != mwait_mask) 238 return false; 239 240 /* We need to emulate the MWAIT_LEAF and for that we need both 241 * ecx and edx. The hypercall provides only partial information. 242 */ 243 244 ax = CPUID_MWAIT_LEAF; 245 bx = 0; 246 cx = 0; 247 dx = 0; 248 249 native_cpuid(&ax, &bx, &cx, &dx); 250 251 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so, 252 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3. 253 */ 254 buf[0] = ACPI_PDC_REVISION_ID; 255 buf[1] = 1; 256 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP); 257 258 set_xen_guest_handle(op.u.set_pminfo.pdc, buf); 259 260 if ((HYPERVISOR_platform_op(&op) == 0) && 261 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) { 262 cpuid_leaf5_ecx_val = cx; 263 cpuid_leaf5_edx_val = dx; 264 } 265 return true; 266 #else 267 return false; 268 #endif 269 } 270 271 static bool __init xen_check_xsave(void) 272 { 273 unsigned int cx, xsave_mask; 274 275 cx = cpuid_ecx(1); 276 277 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) | 278 (1 << (X86_FEATURE_OSXSAVE % 32)); 279 280 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */ 281 return (cx & xsave_mask) == xsave_mask; 282 } 283 284 static void __init xen_init_capabilities(void) 285 { 286 setup_force_cpu_cap(X86_FEATURE_XENPV); 287 setup_clear_cpu_cap(X86_FEATURE_DCA); 288 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF); 289 setup_clear_cpu_cap(X86_FEATURE_MTRR); 290 setup_clear_cpu_cap(X86_FEATURE_ACC); 291 setup_clear_cpu_cap(X86_FEATURE_X2APIC); 292 setup_clear_cpu_cap(X86_FEATURE_SME); 293 294 /* 295 * Xen PV would need some work to support PCID: CR3 handling as well 296 * as xen_flush_tlb_others() would need updating. 297 */ 298 setup_clear_cpu_cap(X86_FEATURE_PCID); 299 300 if (!xen_initial_domain()) 301 setup_clear_cpu_cap(X86_FEATURE_ACPI); 302 303 if (xen_check_mwait()) 304 setup_force_cpu_cap(X86_FEATURE_MWAIT); 305 else 306 setup_clear_cpu_cap(X86_FEATURE_MWAIT); 307 308 if (!xen_check_xsave()) { 309 setup_clear_cpu_cap(X86_FEATURE_XSAVE); 310 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE); 311 } 312 } 313 314 static void xen_set_debugreg(int reg, unsigned long val) 315 { 316 HYPERVISOR_set_debugreg(reg, val); 317 } 318 319 static unsigned long xen_get_debugreg(int reg) 320 { 321 return HYPERVISOR_get_debugreg(reg); 322 } 323 324 static void xen_end_context_switch(struct task_struct *next) 325 { 326 xen_mc_flush(); 327 paravirt_end_context_switch(next); 328 } 329 330 static unsigned long xen_store_tr(void) 331 { 332 return 0; 333 } 334 335 /* 336 * Set the page permissions for a particular virtual address. If the 337 * address is a vmalloc mapping (or other non-linear mapping), then 338 * find the linear mapping of the page and also set its protections to 339 * match. 340 */ 341 static void set_aliased_prot(void *v, pgprot_t prot) 342 { 343 int level; 344 pte_t *ptep; 345 pte_t pte; 346 unsigned long pfn; 347 unsigned char dummy; 348 void *va; 349 350 ptep = lookup_address((unsigned long)v, &level); 351 BUG_ON(ptep == NULL); 352 353 pfn = pte_pfn(*ptep); 354 pte = pfn_pte(pfn, prot); 355 356 /* 357 * Careful: update_va_mapping() will fail if the virtual address 358 * we're poking isn't populated in the page tables. We don't 359 * need to worry about the direct map (that's always in the page 360 * tables), but we need to be careful about vmap space. In 361 * particular, the top level page table can lazily propagate 362 * entries between processes, so if we've switched mms since we 363 * vmapped the target in the first place, we might not have the 364 * top-level page table entry populated. 365 * 366 * We disable preemption because we want the same mm active when 367 * we probe the target and when we issue the hypercall. We'll 368 * have the same nominal mm, but if we're a kernel thread, lazy 369 * mm dropping could change our pgd. 370 * 371 * Out of an abundance of caution, this uses __get_user() to fault 372 * in the target address just in case there's some obscure case 373 * in which the target address isn't readable. 374 */ 375 376 preempt_disable(); 377 378 copy_from_kernel_nofault(&dummy, v, 1); 379 380 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) 381 BUG(); 382 383 va = __va(PFN_PHYS(pfn)); 384 385 if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 386 BUG(); 387 388 preempt_enable(); 389 } 390 391 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) 392 { 393 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 394 int i; 395 396 /* 397 * We need to mark the all aliases of the LDT pages RO. We 398 * don't need to call vm_flush_aliases(), though, since that's 399 * only responsible for flushing aliases out the TLBs, not the 400 * page tables, and Xen will flush the TLB for us if needed. 401 * 402 * To avoid confusing future readers: none of this is necessary 403 * to load the LDT. The hypervisor only checks this when the 404 * LDT is faulted in due to subsequent descriptor access. 405 */ 406 407 for (i = 0; i < entries; i += entries_per_page) 408 set_aliased_prot(ldt + i, PAGE_KERNEL_RO); 409 } 410 411 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) 412 { 413 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; 414 int i; 415 416 for (i = 0; i < entries; i += entries_per_page) 417 set_aliased_prot(ldt + i, PAGE_KERNEL); 418 } 419 420 static void xen_set_ldt(const void *addr, unsigned entries) 421 { 422 struct mmuext_op *op; 423 struct multicall_space mcs = xen_mc_entry(sizeof(*op)); 424 425 trace_xen_cpu_set_ldt(addr, entries); 426 427 op = mcs.args; 428 op->cmd = MMUEXT_SET_LDT; 429 op->arg1.linear_addr = (unsigned long)addr; 430 op->arg2.nr_ents = entries; 431 432 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 433 434 xen_mc_issue(PARAVIRT_LAZY_CPU); 435 } 436 437 static void xen_load_gdt(const struct desc_ptr *dtr) 438 { 439 unsigned long va = dtr->address; 440 unsigned int size = dtr->size + 1; 441 unsigned long pfn, mfn; 442 int level; 443 pte_t *ptep; 444 void *virt; 445 446 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */ 447 BUG_ON(size > PAGE_SIZE); 448 BUG_ON(va & ~PAGE_MASK); 449 450 /* 451 * The GDT is per-cpu and is in the percpu data area. 452 * That can be virtually mapped, so we need to do a 453 * page-walk to get the underlying MFN for the 454 * hypercall. The page can also be in the kernel's 455 * linear range, so we need to RO that mapping too. 456 */ 457 ptep = lookup_address(va, &level); 458 BUG_ON(ptep == NULL); 459 460 pfn = pte_pfn(*ptep); 461 mfn = pfn_to_mfn(pfn); 462 virt = __va(PFN_PHYS(pfn)); 463 464 make_lowmem_page_readonly((void *)va); 465 make_lowmem_page_readonly(virt); 466 467 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct))) 468 BUG(); 469 } 470 471 /* 472 * load_gdt for early boot, when the gdt is only mapped once 473 */ 474 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr) 475 { 476 unsigned long va = dtr->address; 477 unsigned int size = dtr->size + 1; 478 unsigned long pfn, mfn; 479 pte_t pte; 480 481 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */ 482 BUG_ON(size > PAGE_SIZE); 483 BUG_ON(va & ~PAGE_MASK); 484 485 pfn = virt_to_pfn(va); 486 mfn = pfn_to_mfn(pfn); 487 488 pte = pfn_pte(pfn, PAGE_KERNEL_RO); 489 490 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) 491 BUG(); 492 493 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct))) 494 BUG(); 495 } 496 497 static inline bool desc_equal(const struct desc_struct *d1, 498 const struct desc_struct *d2) 499 { 500 return !memcmp(d1, d2, sizeof(*d1)); 501 } 502 503 static void load_TLS_descriptor(struct thread_struct *t, 504 unsigned int cpu, unsigned int i) 505 { 506 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i]; 507 struct desc_struct *gdt; 508 xmaddr_t maddr; 509 struct multicall_space mc; 510 511 if (desc_equal(shadow, &t->tls_array[i])) 512 return; 513 514 *shadow = t->tls_array[i]; 515 516 gdt = get_cpu_gdt_rw(cpu); 517 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); 518 mc = __xen_mc_entry(0); 519 520 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); 521 } 522 523 static void xen_load_tls(struct thread_struct *t, unsigned int cpu) 524 { 525 /* 526 * In lazy mode we need to zero %fs, otherwise we may get an 527 * exception between the new %fs descriptor being loaded and 528 * %fs being effectively cleared at __switch_to(). 529 */ 530 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) 531 loadsegment(fs, 0); 532 533 xen_mc_batch(); 534 535 load_TLS_descriptor(t, cpu, 0); 536 load_TLS_descriptor(t, cpu, 1); 537 load_TLS_descriptor(t, cpu, 2); 538 539 xen_mc_issue(PARAVIRT_LAZY_CPU); 540 } 541 542 static void xen_load_gs_index(unsigned int idx) 543 { 544 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) 545 BUG(); 546 } 547 548 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 549 const void *ptr) 550 { 551 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); 552 u64 entry = *(u64 *)ptr; 553 554 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry); 555 556 preempt_disable(); 557 558 xen_mc_flush(); 559 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) 560 BUG(); 561 562 preempt_enable(); 563 } 564 565 void noist_exc_debug(struct pt_regs *regs); 566 567 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi) 568 { 569 /* On Xen PV, NMI doesn't use IST. The C part is the same as native. */ 570 exc_nmi(regs); 571 } 572 573 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault) 574 { 575 /* On Xen PV, DF doesn't use IST. The C part is the same as native. */ 576 exc_double_fault(regs, error_code); 577 } 578 579 DEFINE_IDTENTRY_RAW(xenpv_exc_debug) 580 { 581 /* 582 * There's no IST on Xen PV, but we still need to dispatch 583 * to the correct handler. 584 */ 585 if (user_mode(regs)) 586 noist_exc_debug(regs); 587 else 588 exc_debug(regs); 589 } 590 591 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap) 592 { 593 /* This should never happen and there is no way to handle it. */ 594 instrumentation_begin(); 595 pr_err("Unknown trap in Xen PV mode."); 596 BUG(); 597 instrumentation_end(); 598 } 599 600 #ifdef CONFIG_X86_MCE 601 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check) 602 { 603 /* 604 * There's no IST on Xen PV, but we still need to dispatch 605 * to the correct handler. 606 */ 607 if (user_mode(regs)) 608 noist_exc_machine_check(regs); 609 else 610 exc_machine_check(regs); 611 } 612 #endif 613 614 struct trap_array_entry { 615 void (*orig)(void); 616 void (*xen)(void); 617 bool ist_okay; 618 }; 619 620 #define TRAP_ENTRY(func, ist_ok) { \ 621 .orig = asm_##func, \ 622 .xen = xen_asm_##func, \ 623 .ist_okay = ist_ok } 624 625 #define TRAP_ENTRY_REDIR(func, ist_ok) { \ 626 .orig = asm_##func, \ 627 .xen = xen_asm_xenpv_##func, \ 628 .ist_okay = ist_ok } 629 630 static struct trap_array_entry trap_array[] = { 631 TRAP_ENTRY_REDIR(exc_debug, true ), 632 TRAP_ENTRY_REDIR(exc_double_fault, true ), 633 #ifdef CONFIG_X86_MCE 634 TRAP_ENTRY_REDIR(exc_machine_check, true ), 635 #endif 636 TRAP_ENTRY_REDIR(exc_nmi, true ), 637 TRAP_ENTRY(exc_int3, false ), 638 TRAP_ENTRY(exc_overflow, false ), 639 #ifdef CONFIG_IA32_EMULATION 640 { entry_INT80_compat, xen_entry_INT80_compat, false }, 641 #endif 642 TRAP_ENTRY(exc_page_fault, false ), 643 TRAP_ENTRY(exc_divide_error, false ), 644 TRAP_ENTRY(exc_bounds, false ), 645 TRAP_ENTRY(exc_invalid_op, false ), 646 TRAP_ENTRY(exc_device_not_available, false ), 647 TRAP_ENTRY(exc_coproc_segment_overrun, false ), 648 TRAP_ENTRY(exc_invalid_tss, false ), 649 TRAP_ENTRY(exc_segment_not_present, false ), 650 TRAP_ENTRY(exc_stack_segment, false ), 651 TRAP_ENTRY(exc_general_protection, false ), 652 TRAP_ENTRY(exc_spurious_interrupt_bug, false ), 653 TRAP_ENTRY(exc_coprocessor_error, false ), 654 TRAP_ENTRY(exc_alignment_check, false ), 655 TRAP_ENTRY(exc_simd_coprocessor_error, false ), 656 }; 657 658 static bool __ref get_trap_addr(void **addr, unsigned int ist) 659 { 660 unsigned int nr; 661 bool ist_okay = false; 662 bool found = false; 663 664 /* 665 * Replace trap handler addresses by Xen specific ones. 666 * Check for known traps using IST and whitelist them. 667 * The debugger ones are the only ones we care about. 668 * Xen will handle faults like double_fault, so we should never see 669 * them. Warn if there's an unexpected IST-using fault handler. 670 */ 671 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) { 672 struct trap_array_entry *entry = trap_array + nr; 673 674 if (*addr == entry->orig) { 675 *addr = entry->xen; 676 ist_okay = entry->ist_okay; 677 found = true; 678 break; 679 } 680 } 681 682 if (nr == ARRAY_SIZE(trap_array) && 683 *addr >= (void *)early_idt_handler_array[0] && 684 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) { 685 nr = (*addr - (void *)early_idt_handler_array[0]) / 686 EARLY_IDT_HANDLER_SIZE; 687 *addr = (void *)xen_early_idt_handler_array[nr]; 688 found = true; 689 } 690 691 if (!found) 692 *addr = (void *)xen_asm_exc_xen_unknown_trap; 693 694 if (WARN_ON(found && ist != 0 && !ist_okay)) 695 return false; 696 697 return true; 698 } 699 700 static int cvt_gate_to_trap(int vector, const gate_desc *val, 701 struct trap_info *info) 702 { 703 unsigned long addr; 704 705 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT) 706 return 0; 707 708 info->vector = vector; 709 710 addr = gate_offset(val); 711 if (!get_trap_addr((void **)&addr, val->bits.ist)) 712 return 0; 713 info->address = addr; 714 715 info->cs = gate_segment(val); 716 info->flags = val->bits.dpl; 717 /* interrupt gates clear IF */ 718 if (val->bits.type == GATE_INTERRUPT) 719 info->flags |= 1 << 2; 720 721 return 1; 722 } 723 724 /* Locations of each CPU's IDT */ 725 static DEFINE_PER_CPU(struct desc_ptr, idt_desc); 726 727 /* Set an IDT entry. If the entry is part of the current IDT, then 728 also update Xen. */ 729 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) 730 { 731 unsigned long p = (unsigned long)&dt[entrynum]; 732 unsigned long start, end; 733 734 trace_xen_cpu_write_idt_entry(dt, entrynum, g); 735 736 preempt_disable(); 737 738 start = __this_cpu_read(idt_desc.address); 739 end = start + __this_cpu_read(idt_desc.size) + 1; 740 741 xen_mc_flush(); 742 743 native_write_idt_entry(dt, entrynum, g); 744 745 if (p >= start && (p + 8) <= end) { 746 struct trap_info info[2]; 747 748 info[1].address = 0; 749 750 if (cvt_gate_to_trap(entrynum, g, &info[0])) 751 if (HYPERVISOR_set_trap_table(info)) 752 BUG(); 753 } 754 755 preempt_enable(); 756 } 757 758 static void xen_convert_trap_info(const struct desc_ptr *desc, 759 struct trap_info *traps) 760 { 761 unsigned in, out, count; 762 763 count = (desc->size+1) / sizeof(gate_desc); 764 BUG_ON(count > 256); 765 766 for (in = out = 0; in < count; in++) { 767 gate_desc *entry = (gate_desc *)(desc->address) + in; 768 769 if (cvt_gate_to_trap(in, entry, &traps[out])) 770 out++; 771 } 772 traps[out].address = 0; 773 } 774 775 void xen_copy_trap_info(struct trap_info *traps) 776 { 777 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc); 778 779 xen_convert_trap_info(desc, traps); 780 } 781 782 /* Load a new IDT into Xen. In principle this can be per-CPU, so we 783 hold a spinlock to protect the static traps[] array (static because 784 it avoids allocation, and saves stack space). */ 785 static void xen_load_idt(const struct desc_ptr *desc) 786 { 787 static DEFINE_SPINLOCK(lock); 788 static struct trap_info traps[257]; 789 790 trace_xen_cpu_load_idt(desc); 791 792 spin_lock(&lock); 793 794 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc)); 795 796 xen_convert_trap_info(desc, traps); 797 798 xen_mc_flush(); 799 if (HYPERVISOR_set_trap_table(traps)) 800 BUG(); 801 802 spin_unlock(&lock); 803 } 804 805 /* Write a GDT descriptor entry. Ignore LDT descriptors, since 806 they're handled differently. */ 807 static void xen_write_gdt_entry(struct desc_struct *dt, int entry, 808 const void *desc, int type) 809 { 810 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 811 812 preempt_disable(); 813 814 switch (type) { 815 case DESC_LDT: 816 case DESC_TSS: 817 /* ignore */ 818 break; 819 820 default: { 821 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); 822 823 xen_mc_flush(); 824 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 825 BUG(); 826 } 827 828 } 829 830 preempt_enable(); 831 } 832 833 /* 834 * Version of write_gdt_entry for use at early boot-time needed to 835 * update an entry as simply as possible. 836 */ 837 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, 838 const void *desc, int type) 839 { 840 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); 841 842 switch (type) { 843 case DESC_LDT: 844 case DESC_TSS: 845 /* ignore */ 846 break; 847 848 default: { 849 xmaddr_t maddr = virt_to_machine(&dt[entry]); 850 851 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) 852 dt[entry] = *(struct desc_struct *)desc; 853 } 854 855 } 856 } 857 858 static void xen_load_sp0(unsigned long sp0) 859 { 860 struct multicall_space mcs; 861 862 mcs = xen_mc_entry(0); 863 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0); 864 xen_mc_issue(PARAVIRT_LAZY_CPU); 865 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0); 866 } 867 868 #ifdef CONFIG_X86_IOPL_IOPERM 869 static void xen_invalidate_io_bitmap(void) 870 { 871 struct physdev_set_iobitmap iobitmap = { 872 .bitmap = NULL, 873 .nr_ports = 0, 874 }; 875 876 native_tss_invalidate_io_bitmap(); 877 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap); 878 } 879 880 static void xen_update_io_bitmap(void) 881 { 882 struct physdev_set_iobitmap iobitmap; 883 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); 884 885 native_tss_update_io_bitmap(); 886 887 iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) + 888 tss->x86_tss.io_bitmap_base; 889 if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID) 890 iobitmap.nr_ports = 0; 891 else 892 iobitmap.nr_ports = IO_BITMAP_BITS; 893 894 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap); 895 } 896 #endif 897 898 static void xen_io_delay(void) 899 { 900 } 901 902 static DEFINE_PER_CPU(unsigned long, xen_cr0_value); 903 904 static unsigned long xen_read_cr0(void) 905 { 906 unsigned long cr0 = this_cpu_read(xen_cr0_value); 907 908 if (unlikely(cr0 == 0)) { 909 cr0 = native_read_cr0(); 910 this_cpu_write(xen_cr0_value, cr0); 911 } 912 913 return cr0; 914 } 915 916 static void xen_write_cr0(unsigned long cr0) 917 { 918 struct multicall_space mcs; 919 920 this_cpu_write(xen_cr0_value, cr0); 921 922 /* Only pay attention to cr0.TS; everything else is 923 ignored. */ 924 mcs = xen_mc_entry(0); 925 926 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); 927 928 xen_mc_issue(PARAVIRT_LAZY_CPU); 929 } 930 931 static void xen_write_cr4(unsigned long cr4) 932 { 933 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE); 934 935 native_write_cr4(cr4); 936 } 937 938 static u64 xen_read_msr_safe(unsigned int msr, int *err) 939 { 940 u64 val; 941 942 if (pmu_msr_read(msr, &val, err)) 943 return val; 944 945 val = native_read_msr_safe(msr, err); 946 switch (msr) { 947 case MSR_IA32_APICBASE: 948 val &= ~X2APIC_ENABLE; 949 break; 950 } 951 return val; 952 } 953 954 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) 955 { 956 int ret; 957 unsigned int which; 958 u64 base; 959 960 ret = 0; 961 962 switch (msr) { 963 case MSR_FS_BASE: which = SEGBASE_FS; goto set; 964 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; 965 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; 966 967 set: 968 base = ((u64)high << 32) | low; 969 if (HYPERVISOR_set_segment_base(which, base) != 0) 970 ret = -EIO; 971 break; 972 973 case MSR_STAR: 974 case MSR_CSTAR: 975 case MSR_LSTAR: 976 case MSR_SYSCALL_MASK: 977 case MSR_IA32_SYSENTER_CS: 978 case MSR_IA32_SYSENTER_ESP: 979 case MSR_IA32_SYSENTER_EIP: 980 /* Fast syscall setup is all done in hypercalls, so 981 these are all ignored. Stub them out here to stop 982 Xen console noise. */ 983 break; 984 985 default: 986 if (!pmu_msr_write(msr, low, high, &ret)) 987 ret = native_write_msr_safe(msr, low, high); 988 } 989 990 return ret; 991 } 992 993 static u64 xen_read_msr(unsigned int msr) 994 { 995 /* 996 * This will silently swallow a #GP from RDMSR. It may be worth 997 * changing that. 998 */ 999 int err; 1000 1001 return xen_read_msr_safe(msr, &err); 1002 } 1003 1004 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high) 1005 { 1006 /* 1007 * This will silently swallow a #GP from WRMSR. It may be worth 1008 * changing that. 1009 */ 1010 xen_write_msr_safe(msr, low, high); 1011 } 1012 1013 /* This is called once we have the cpu_possible_mask */ 1014 void __init xen_setup_vcpu_info_placement(void) 1015 { 1016 int cpu; 1017 1018 for_each_possible_cpu(cpu) { 1019 /* Set up direct vCPU id mapping for PV guests. */ 1020 per_cpu(xen_vcpu_id, cpu) = cpu; 1021 1022 /* 1023 * xen_vcpu_setup(cpu) can fail -- in which case it 1024 * falls back to the shared_info version for cpus 1025 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS. 1026 * 1027 * xen_cpu_up_prepare_pv() handles the rest by failing 1028 * them in hotplug. 1029 */ 1030 (void) xen_vcpu_setup(cpu); 1031 } 1032 1033 /* 1034 * xen_vcpu_setup managed to place the vcpu_info within the 1035 * percpu area for all cpus, so make use of it. 1036 */ 1037 if (xen_have_vcpu_info_placement) { 1038 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); 1039 pv_ops.irq.irq_disable = 1040 __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); 1041 pv_ops.irq.irq_enable = 1042 __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); 1043 pv_ops.mmu.read_cr2 = 1044 __PV_IS_CALLEE_SAVE(xen_read_cr2_direct); 1045 } 1046 } 1047 1048 static const struct pv_info xen_info __initconst = { 1049 .extra_user_64bit_cs = FLAT_USER_CS64, 1050 .name = "Xen", 1051 }; 1052 1053 static const struct pv_cpu_ops xen_cpu_ops __initconst = { 1054 .cpuid = xen_cpuid, 1055 1056 .set_debugreg = xen_set_debugreg, 1057 .get_debugreg = xen_get_debugreg, 1058 1059 .read_cr0 = xen_read_cr0, 1060 .write_cr0 = xen_write_cr0, 1061 1062 .write_cr4 = xen_write_cr4, 1063 1064 .wbinvd = native_wbinvd, 1065 1066 .read_msr = xen_read_msr, 1067 .write_msr = xen_write_msr, 1068 1069 .read_msr_safe = xen_read_msr_safe, 1070 .write_msr_safe = xen_write_msr_safe, 1071 1072 .read_pmc = xen_read_pmc, 1073 1074 .load_tr_desc = paravirt_nop, 1075 .set_ldt = xen_set_ldt, 1076 .load_gdt = xen_load_gdt, 1077 .load_idt = xen_load_idt, 1078 .load_tls = xen_load_tls, 1079 .load_gs_index = xen_load_gs_index, 1080 1081 .alloc_ldt = xen_alloc_ldt, 1082 .free_ldt = xen_free_ldt, 1083 1084 .store_tr = xen_store_tr, 1085 1086 .write_ldt_entry = xen_write_ldt_entry, 1087 .write_gdt_entry = xen_write_gdt_entry, 1088 .write_idt_entry = xen_write_idt_entry, 1089 .load_sp0 = xen_load_sp0, 1090 1091 #ifdef CONFIG_X86_IOPL_IOPERM 1092 .invalidate_io_bitmap = xen_invalidate_io_bitmap, 1093 .update_io_bitmap = xen_update_io_bitmap, 1094 #endif 1095 .io_delay = xen_io_delay, 1096 1097 .start_context_switch = paravirt_start_context_switch, 1098 .end_context_switch = xen_end_context_switch, 1099 }; 1100 1101 static void xen_restart(char *msg) 1102 { 1103 xen_reboot(SHUTDOWN_reboot); 1104 } 1105 1106 static void xen_machine_halt(void) 1107 { 1108 xen_reboot(SHUTDOWN_poweroff); 1109 } 1110 1111 static void xen_machine_power_off(void) 1112 { 1113 if (pm_power_off) 1114 pm_power_off(); 1115 xen_reboot(SHUTDOWN_poweroff); 1116 } 1117 1118 static void xen_crash_shutdown(struct pt_regs *regs) 1119 { 1120 xen_reboot(SHUTDOWN_crash); 1121 } 1122 1123 static const struct machine_ops xen_machine_ops __initconst = { 1124 .restart = xen_restart, 1125 .halt = xen_machine_halt, 1126 .power_off = xen_machine_power_off, 1127 .shutdown = xen_machine_halt, 1128 .crash_shutdown = xen_crash_shutdown, 1129 .emergency_restart = xen_emergency_restart, 1130 }; 1131 1132 static unsigned char xen_get_nmi_reason(void) 1133 { 1134 unsigned char reason = 0; 1135 1136 /* Construct a value which looks like it came from port 0x61. */ 1137 if (test_bit(_XEN_NMIREASON_io_error, 1138 &HYPERVISOR_shared_info->arch.nmi_reason)) 1139 reason |= NMI_REASON_IOCHK; 1140 if (test_bit(_XEN_NMIREASON_pci_serr, 1141 &HYPERVISOR_shared_info->arch.nmi_reason)) 1142 reason |= NMI_REASON_SERR; 1143 1144 return reason; 1145 } 1146 1147 static void __init xen_boot_params_init_edd(void) 1148 { 1149 #if IS_ENABLED(CONFIG_EDD) 1150 struct xen_platform_op op; 1151 struct edd_info *edd_info; 1152 u32 *mbr_signature; 1153 unsigned nr; 1154 int ret; 1155 1156 edd_info = boot_params.eddbuf; 1157 mbr_signature = boot_params.edd_mbr_sig_buffer; 1158 1159 op.cmd = XENPF_firmware_info; 1160 1161 op.u.firmware_info.type = XEN_FW_DISK_INFO; 1162 for (nr = 0; nr < EDDMAXNR; nr++) { 1163 struct edd_info *info = edd_info + nr; 1164 1165 op.u.firmware_info.index = nr; 1166 info->params.length = sizeof(info->params); 1167 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params, 1168 &info->params); 1169 ret = HYPERVISOR_platform_op(&op); 1170 if (ret) 1171 break; 1172 1173 #define C(x) info->x = op.u.firmware_info.u.disk_info.x 1174 C(device); 1175 C(version); 1176 C(interface_support); 1177 C(legacy_max_cylinder); 1178 C(legacy_max_head); 1179 C(legacy_sectors_per_track); 1180 #undef C 1181 } 1182 boot_params.eddbuf_entries = nr; 1183 1184 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE; 1185 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) { 1186 op.u.firmware_info.index = nr; 1187 ret = HYPERVISOR_platform_op(&op); 1188 if (ret) 1189 break; 1190 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature; 1191 } 1192 boot_params.edd_mbr_sig_buf_entries = nr; 1193 #endif 1194 } 1195 1196 /* 1197 * Set up the GDT and segment registers for -fstack-protector. Until 1198 * we do this, we have to be careful not to call any stack-protected 1199 * function, which is most of the kernel. 1200 */ 1201 static void __init xen_setup_gdt(int cpu) 1202 { 1203 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot; 1204 pv_ops.cpu.load_gdt = xen_load_gdt_boot; 1205 1206 switch_to_new_gdt(cpu); 1207 1208 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry; 1209 pv_ops.cpu.load_gdt = xen_load_gdt; 1210 } 1211 1212 static void __init xen_dom0_set_legacy_features(void) 1213 { 1214 x86_platform.legacy.rtc = 1; 1215 } 1216 1217 static void __init xen_domu_set_legacy_features(void) 1218 { 1219 x86_platform.legacy.rtc = 0; 1220 } 1221 1222 /* First C function to be called on Xen boot */ 1223 asmlinkage __visible void __init xen_start_kernel(void) 1224 { 1225 struct physdev_set_iopl set_iopl; 1226 unsigned long initrd_start = 0; 1227 int rc; 1228 1229 if (!xen_start_info) 1230 return; 1231 1232 xen_domain_type = XEN_PV_DOMAIN; 1233 xen_start_flags = xen_start_info->flags; 1234 1235 xen_setup_features(); 1236 1237 /* Install Xen paravirt ops */ 1238 pv_info = xen_info; 1239 pv_ops.cpu = xen_cpu_ops; 1240 paravirt_iret = xen_iret; 1241 xen_init_irq_ops(); 1242 1243 /* 1244 * Setup xen_vcpu early because it is needed for 1245 * local_irq_disable(), irqs_disabled(), e.g. in printk(). 1246 * 1247 * Don't do the full vcpu_info placement stuff until we have 1248 * the cpu_possible_mask and a non-dummy shared_info. 1249 */ 1250 xen_vcpu_info_reset(0); 1251 1252 x86_platform.get_nmi_reason = xen_get_nmi_reason; 1253 1254 x86_init.resources.memory_setup = xen_memory_setup; 1255 x86_init.irqs.intr_mode_select = x86_init_noop; 1256 x86_init.irqs.intr_mode_init = x86_init_noop; 1257 x86_init.oem.arch_setup = xen_arch_setup; 1258 x86_init.oem.banner = xen_banner; 1259 x86_init.hyper.init_platform = xen_pv_init_platform; 1260 x86_init.hyper.guest_late_init = xen_pv_guest_late_init; 1261 1262 /* 1263 * Set up some pagetable state before starting to set any ptes. 1264 */ 1265 1266 xen_setup_machphys_mapping(); 1267 xen_init_mmu_ops(); 1268 1269 /* Prevent unwanted bits from being set in PTEs. */ 1270 __supported_pte_mask &= ~_PAGE_GLOBAL; 1271 __default_kernel_pte_mask &= ~_PAGE_GLOBAL; 1272 1273 /* 1274 * Prevent page tables from being allocated in highmem, even 1275 * if CONFIG_HIGHPTE is enabled. 1276 */ 1277 __userpte_alloc_gfp &= ~__GFP_HIGHMEM; 1278 1279 /* Get mfn list */ 1280 xen_build_dynamic_phys_to_machine(); 1281 1282 /* Work out if we support NX */ 1283 get_cpu_cap(&boot_cpu_data); 1284 x86_configure_nx(); 1285 1286 /* 1287 * Set up kernel GDT and segment registers, mainly so that 1288 * -fstack-protector code can be executed. 1289 */ 1290 xen_setup_gdt(0); 1291 1292 /* Determine virtual and physical address sizes */ 1293 get_cpu_address_sizes(&boot_cpu_data); 1294 1295 /* Let's presume PV guests always boot on vCPU with id 0. */ 1296 per_cpu(xen_vcpu_id, 0) = 0; 1297 1298 idt_setup_early_handler(); 1299 1300 xen_init_capabilities(); 1301 1302 #ifdef CONFIG_X86_LOCAL_APIC 1303 /* 1304 * set up the basic apic ops. 1305 */ 1306 xen_init_apic(); 1307 #endif 1308 1309 machine_ops = xen_machine_ops; 1310 1311 /* 1312 * The only reliable way to retain the initial address of the 1313 * percpu gdt_page is to remember it here, so we can go and 1314 * mark it RW later, when the initial percpu area is freed. 1315 */ 1316 xen_initial_gdt = &per_cpu(gdt_page, 0); 1317 1318 xen_smp_init(); 1319 1320 #ifdef CONFIG_ACPI_NUMA 1321 /* 1322 * The pages we from Xen are not related to machine pages, so 1323 * any NUMA information the kernel tries to get from ACPI will 1324 * be meaningless. Prevent it from trying. 1325 */ 1326 disable_srat(); 1327 #endif 1328 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv)); 1329 1330 local_irq_disable(); 1331 early_boot_irqs_disabled = true; 1332 1333 xen_raw_console_write("mapping kernel into physical memory\n"); 1334 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, 1335 xen_start_info->nr_pages); 1336 xen_reserve_special_pages(); 1337 1338 /* 1339 * We used to do this in xen_arch_setup, but that is too late 1340 * on AMD were early_cpu_init (run before ->arch_setup()) calls 1341 * early_amd_init which pokes 0xcf8 port. 1342 */ 1343 set_iopl.iopl = 1; 1344 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 1345 if (rc != 0) 1346 xen_raw_printk("physdev_op failed %d\n", rc); 1347 1348 1349 if (xen_start_info->mod_start) { 1350 if (xen_start_info->flags & SIF_MOD_START_PFN) 1351 initrd_start = PFN_PHYS(xen_start_info->mod_start); 1352 else 1353 initrd_start = __pa(xen_start_info->mod_start); 1354 } 1355 1356 /* Poke various useful things into boot_params */ 1357 boot_params.hdr.type_of_loader = (9 << 4) | 0; 1358 boot_params.hdr.ramdisk_image = initrd_start; 1359 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1360 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1361 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN; 1362 1363 if (!xen_initial_domain()) { 1364 add_preferred_console("xenboot", 0, NULL); 1365 if (pci_xen) 1366 x86_init.pci.arch_init = pci_xen_init; 1367 x86_platform.set_legacy_features = 1368 xen_domu_set_legacy_features; 1369 } else { 1370 const struct dom0_vga_console_info *info = 1371 (void *)((char *)xen_start_info + 1372 xen_start_info->console.dom0.info_off); 1373 struct xen_platform_op op = { 1374 .cmd = XENPF_firmware_info, 1375 .interface_version = XENPF_INTERFACE_VERSION, 1376 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS, 1377 }; 1378 1379 x86_platform.set_legacy_features = 1380 xen_dom0_set_legacy_features; 1381 xen_init_vga(info, xen_start_info->console.dom0.info_size); 1382 xen_start_info->console.domU.mfn = 0; 1383 xen_start_info->console.domU.evtchn = 0; 1384 1385 if (HYPERVISOR_platform_op(&op) == 0) 1386 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags; 1387 1388 /* Make sure ACS will be enabled */ 1389 pci_request_acs(); 1390 1391 xen_acpi_sleep_register(); 1392 1393 /* Avoid searching for BIOS MP tables */ 1394 x86_init.mpparse.find_smp_config = x86_init_noop; 1395 x86_init.mpparse.get_smp_config = x86_init_uint_noop; 1396 1397 xen_boot_params_init_edd(); 1398 1399 #ifdef CONFIG_ACPI 1400 /* 1401 * Disable selecting "Firmware First mode" for correctable 1402 * memory errors, as this is the duty of the hypervisor to 1403 * decide. 1404 */ 1405 acpi_disable_cmcff = 1; 1406 #endif 1407 } 1408 1409 if (!boot_params.screen_info.orig_video_isVGA) 1410 add_preferred_console("tty", 0, NULL); 1411 add_preferred_console("hvc", 0, NULL); 1412 if (boot_params.screen_info.orig_video_isVGA) 1413 add_preferred_console("tty", 0, NULL); 1414 1415 #ifdef CONFIG_PCI 1416 /* PCI BIOS service won't work from a PV guest. */ 1417 pci_probe &= ~PCI_PROBE_BIOS; 1418 #endif 1419 xen_raw_console_write("about to get started...\n"); 1420 1421 /* We need this for printk timestamps */ 1422 xen_setup_runstate_info(0); 1423 1424 xen_efi_init(&boot_params); 1425 1426 /* Start the world */ 1427 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */ 1428 x86_64_start_reservations((char *)__pa_symbol(&boot_params)); 1429 } 1430 1431 static int xen_cpu_up_prepare_pv(unsigned int cpu) 1432 { 1433 int rc; 1434 1435 if (per_cpu(xen_vcpu, cpu) == NULL) 1436 return -ENODEV; 1437 1438 xen_setup_timer(cpu); 1439 1440 rc = xen_smp_intr_init(cpu); 1441 if (rc) { 1442 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n", 1443 cpu, rc); 1444 return rc; 1445 } 1446 1447 rc = xen_smp_intr_init_pv(cpu); 1448 if (rc) { 1449 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n", 1450 cpu, rc); 1451 return rc; 1452 } 1453 1454 return 0; 1455 } 1456 1457 static int xen_cpu_dead_pv(unsigned int cpu) 1458 { 1459 xen_smp_intr_free(cpu); 1460 xen_smp_intr_free_pv(cpu); 1461 1462 xen_teardown_timer(cpu); 1463 1464 return 0; 1465 } 1466 1467 static uint32_t __init xen_platform_pv(void) 1468 { 1469 if (xen_pv_domain()) 1470 return xen_cpuid_base(); 1471 1472 return 0; 1473 } 1474 1475 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = { 1476 .name = "Xen PV", 1477 .detect = xen_platform_pv, 1478 .type = X86_HYPER_XEN_PV, 1479 .runtime.pin_vcpu = xen_pin_vcpu, 1480 .ignore_nopv = true, 1481 }; 1482