1 // SPDX-License-Identifier: GPL-2.0
3 * Core of Xen paravirt_ops implementation.
5 * This file contains the xen_paravirt_ops structure itself, and the
7 * - privileged instructions
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/kstrtox.h>
27 #include <linux/memblock.h>
28 #include <linux/export.h>
30 #include <linux/page-flags.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/edd.h>
34 #include <linux/reboot.h>
35 #include <linux/virtio_anchor.h>
36 #include <linux/stackprotector.h>
39 #include <xen/events.h>
40 #include <xen/interface/xen.h>
41 #include <xen/interface/version.h>
42 #include <xen/interface/physdev.h>
43 #include <xen/interface/vcpu.h>
44 #include <xen/interface/memory.h>
45 #include <xen/interface/nmi.h>
46 #include <xen/interface/xen-mca.h>
47 #include <xen/features.h>
49 #include <xen/hvc-console.h>
52 #include <asm/paravirt.h>
55 #include <asm/xen/pci.h>
56 #include <asm/xen/hypercall.h>
57 #include <asm/xen/hypervisor.h>
58 #include <asm/xen/cpuid.h>
59 #include <asm/fixmap.h>
60 #include <asm/processor.h>
61 #include <asm/proto.h>
62 #include <asm/msr-index.h>
63 #include <asm/traps.h>
64 #include <asm/setup.h>
66 #include <asm/pgalloc.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/hypervisor.h>
70 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
75 #ifdef CONFIG_X86_IOPL_IOPERM
76 #include <asm/io_bitmap.h>
80 #include <linux/acpi.h>
82 #include <acpi/proc_cap_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
89 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
91 void *xen_initial_gdt;
93 static int xen_cpu_up_prepare_pv(unsigned int cpu);
94 static int xen_cpu_dead_pv(unsigned int cpu);
97 struct desc_struct desc[3];
100 DEFINE_PER_CPU(enum xen_lazy_mode, xen_lazy_mode) = XEN_LAZY_NONE;
101 DEFINE_PER_CPU(unsigned int, xen_lazy_nesting);
103 enum xen_lazy_mode xen_get_lazy_mode(void)
106 return XEN_LAZY_NONE;
108 return this_cpu_read(xen_lazy_mode);
112 * Updating the 3 TLS descriptors in the GDT on every task switch is
113 * surprisingly expensive so we avoid updating them if they haven't
114 * changed. Since Xen writes different descriptors than the one
115 * passed in the update_descriptor hypercall we keep shadow copies to
118 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
120 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
122 static int __init parse_xen_msr_safe(char *str)
125 return kstrtobool(str, &xen_msr_safe);
128 early_param("xen_msr_safe", parse_xen_msr_safe);
130 /* Get MTRR settings from Xen and put them into mtrr_state. */
131 static void __init xen_set_mtrr_data(void)
134 struct xen_platform_op op = {
135 .cmd = XENPF_read_memtype,
136 .interface_version = XENPF_INTERFACE_VERSION,
141 static struct mtrr_var_range var[MTRR_MAX_VAR_RANGES] __initdata;
143 /* Get physical address width (only 64-bit cpus supported). */
145 eax = cpuid_eax(0x80000000);
146 if ((eax >> 16) == 0x8000 && eax >= 0x80000008) {
147 eax = cpuid_eax(0x80000008);
151 for (reg = 0; reg < MTRR_MAX_VAR_RANGES; reg++) {
152 op.u.read_memtype.reg = reg;
153 if (HYPERVISOR_platform_op(&op))
157 * Only called in dom0, which has all RAM PFNs mapped at
158 * RAM MFNs, and all PCI space etc. is identity mapped.
159 * This means we can treat MFN == PFN regarding MTRR settings.
161 var[reg].base_lo = op.u.read_memtype.type;
162 var[reg].base_lo |= op.u.read_memtype.mfn << PAGE_SHIFT;
163 var[reg].base_hi = op.u.read_memtype.mfn >> (32 - PAGE_SHIFT);
164 mask = ~((op.u.read_memtype.nr_mfns << PAGE_SHIFT) - 1);
165 mask &= (1UL << width) - 1;
167 mask |= MTRR_PHYSMASK_V;
168 var[reg].mask_lo = mask;
169 var[reg].mask_hi = mask >> 32;
172 /* Only overwrite MTRR state if any MTRR could be got from Xen. */
174 mtrr_overwrite_state(var, reg, MTRR_TYPE_UNCACHABLE);
178 static void __init xen_pv_init_platform(void)
180 /* PV guests can't operate virtio devices without grants. */
181 if (IS_ENABLED(CONFIG_XEN_VIRTIO))
182 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
184 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
186 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
187 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
189 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
190 xen_vcpu_info_reset(0);
192 /* pvclock is in shared info area */
195 if (xen_initial_domain())
198 mtrr_overwrite_state(NULL, 0, MTRR_TYPE_WRBACK);
200 /* Adjust nr_cpu_ids before "enumeration" happens */
201 xen_smp_count_cpus();
204 static void __init xen_pv_guest_late_init(void)
207 /* Setup shared vcpu info for non-smp configurations */
208 xen_setup_vcpu_info_placement();
212 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
213 static __read_mostly unsigned int cpuid_leaf5_edx_val;
215 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
216 unsigned int *cx, unsigned int *dx)
218 unsigned int maskebx = ~0;
219 unsigned int or_ebx = 0;
222 * Mask out inconvenient features, to try and disable as many
223 * unsupported kernel subsystems as possible.
227 /* Replace initial APIC ID in bits 24-31 of EBX. */
228 /* See xen_pv_smp_config() for related topology preparations. */
229 maskebx = 0x00ffffff;
230 or_ebx = smp_processor_id() << 24;
233 case CPUID_MWAIT_LEAF:
234 /* Synthesize the values.. */
237 *cx = cpuid_leaf5_ecx_val;
238 *dx = cpuid_leaf5_edx_val;
242 /* Suppress extended topology stuff */
247 asm(XEN_EMULATE_PREFIX "cpuid"
252 : "0" (*ax), "2" (*cx));
258 static bool __init xen_check_mwait(void)
261 struct xen_platform_op op = {
262 .cmd = XENPF_set_processor_pminfo,
263 .u.set_pminfo.id = -1,
264 .u.set_pminfo.type = XEN_PM_PDC,
267 unsigned int ax, bx, cx, dx;
268 unsigned int mwait_mask;
270 /* We need to determine whether it is OK to expose the MWAIT
271 * capability to the kernel to harvest deeper than C3 states from ACPI
272 * _CST using the processor_harvest_xen.c module. For this to work, we
273 * need to gather the MWAIT_LEAF values (which the cstate.c code
274 * checks against). The hypervisor won't expose the MWAIT flag because
275 * it would break backwards compatibility; so we will find out directly
276 * from the hardware and hypercall.
278 if (!xen_initial_domain())
282 * When running under platform earlier than Xen4.2, do not expose
283 * mwait, to avoid the risk of loading native acpi pad driver
285 if (!xen_running_on_version_or_later(4, 2))
291 native_cpuid(&ax, &bx, &cx, &dx);
293 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
294 (1 << (X86_FEATURE_MWAIT % 32));
296 if ((cx & mwait_mask) != mwait_mask)
299 /* We need to emulate the MWAIT_LEAF and for that we need both
300 * ecx and edx. The hypercall provides only partial information.
303 ax = CPUID_MWAIT_LEAF;
308 native_cpuid(&ax, &bx, &cx, &dx);
310 /* Ask the Hypervisor whether to clear ACPI_PROC_CAP_C_C2C3_FFH. If so,
311 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
313 buf[0] = ACPI_PDC_REVISION_ID;
315 buf[2] = (ACPI_PROC_CAP_C_CAPABILITY_SMP | ACPI_PROC_CAP_EST_CAPABILITY_SWSMP);
317 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
319 if ((HYPERVISOR_platform_op(&op) == 0) &&
320 (buf[2] & (ACPI_PROC_CAP_C_C1_FFH | ACPI_PROC_CAP_C_C2C3_FFH))) {
321 cpuid_leaf5_ecx_val = cx;
322 cpuid_leaf5_edx_val = dx;
330 static bool __init xen_check_xsave(void)
332 unsigned int cx, xsave_mask;
336 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
337 (1 << (X86_FEATURE_OSXSAVE % 32));
339 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
340 return (cx & xsave_mask) == xsave_mask;
343 static void __init xen_init_capabilities(void)
345 setup_force_cpu_cap(X86_FEATURE_XENPV);
346 setup_clear_cpu_cap(X86_FEATURE_DCA);
347 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
348 setup_clear_cpu_cap(X86_FEATURE_MTRR);
349 setup_clear_cpu_cap(X86_FEATURE_ACC);
350 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
351 setup_clear_cpu_cap(X86_FEATURE_SME);
352 setup_clear_cpu_cap(X86_FEATURE_LKGS);
355 * Xen PV would need some work to support PCID: CR3 handling as well
356 * as xen_flush_tlb_others() would need updating.
358 setup_clear_cpu_cap(X86_FEATURE_PCID);
360 if (!xen_initial_domain())
361 setup_clear_cpu_cap(X86_FEATURE_ACPI);
363 if (xen_check_mwait())
364 setup_force_cpu_cap(X86_FEATURE_MWAIT);
366 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
368 if (!xen_check_xsave()) {
369 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
370 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
374 static noinstr void xen_set_debugreg(int reg, unsigned long val)
376 HYPERVISOR_set_debugreg(reg, val);
379 static noinstr unsigned long xen_get_debugreg(int reg)
381 return HYPERVISOR_get_debugreg(reg);
384 static void xen_start_context_switch(struct task_struct *prev)
386 BUG_ON(preemptible());
388 if (this_cpu_read(xen_lazy_mode) == XEN_LAZY_MMU) {
389 arch_leave_lazy_mmu_mode();
390 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
392 enter_lazy(XEN_LAZY_CPU);
395 static void xen_end_context_switch(struct task_struct *next)
397 BUG_ON(preemptible());
400 leave_lazy(XEN_LAZY_CPU);
401 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
402 arch_enter_lazy_mmu_mode();
405 static unsigned long xen_store_tr(void)
411 * Set the page permissions for a particular virtual address. If the
412 * address is a vmalloc mapping (or other non-linear mapping), then
413 * find the linear mapping of the page and also set its protections to
416 static void set_aliased_prot(void *v, pgprot_t prot)
425 ptep = lookup_address((unsigned long)v, &level);
426 BUG_ON(ptep == NULL);
428 pfn = pte_pfn(*ptep);
429 pte = pfn_pte(pfn, prot);
432 * Careful: update_va_mapping() will fail if the virtual address
433 * we're poking isn't populated in the page tables. We don't
434 * need to worry about the direct map (that's always in the page
435 * tables), but we need to be careful about vmap space. In
436 * particular, the top level page table can lazily propagate
437 * entries between processes, so if we've switched mms since we
438 * vmapped the target in the first place, we might not have the
439 * top-level page table entry populated.
441 * We disable preemption because we want the same mm active when
442 * we probe the target and when we issue the hypercall. We'll
443 * have the same nominal mm, but if we're a kernel thread, lazy
444 * mm dropping could change our pgd.
446 * Out of an abundance of caution, this uses __get_user() to fault
447 * in the target address just in case there's some obscure case
448 * in which the target address isn't readable.
453 copy_from_kernel_nofault(&dummy, v, 1);
455 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
458 va = __va(PFN_PHYS(pfn));
460 if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
466 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
468 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
472 * We need to mark the all aliases of the LDT pages RO. We
473 * don't need to call vm_flush_aliases(), though, since that's
474 * only responsible for flushing aliases out the TLBs, not the
475 * page tables, and Xen will flush the TLB for us if needed.
477 * To avoid confusing future readers: none of this is necessary
478 * to load the LDT. The hypervisor only checks this when the
479 * LDT is faulted in due to subsequent descriptor access.
482 for (i = 0; i < entries; i += entries_per_page)
483 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
486 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
488 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
491 for (i = 0; i < entries; i += entries_per_page)
492 set_aliased_prot(ldt + i, PAGE_KERNEL);
495 static void xen_set_ldt(const void *addr, unsigned entries)
497 struct mmuext_op *op;
498 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
500 trace_xen_cpu_set_ldt(addr, entries);
503 op->cmd = MMUEXT_SET_LDT;
504 op->arg1.linear_addr = (unsigned long)addr;
505 op->arg2.nr_ents = entries;
507 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
509 xen_mc_issue(XEN_LAZY_CPU);
512 static void xen_load_gdt(const struct desc_ptr *dtr)
514 unsigned long va = dtr->address;
515 unsigned int size = dtr->size + 1;
516 unsigned long pfn, mfn;
521 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
522 BUG_ON(size > PAGE_SIZE);
523 BUG_ON(va & ~PAGE_MASK);
526 * The GDT is per-cpu and is in the percpu data area.
527 * That can be virtually mapped, so we need to do a
528 * page-walk to get the underlying MFN for the
529 * hypercall. The page can also be in the kernel's
530 * linear range, so we need to RO that mapping too.
532 ptep = lookup_address(va, &level);
533 BUG_ON(ptep == NULL);
535 pfn = pte_pfn(*ptep);
536 mfn = pfn_to_mfn(pfn);
537 virt = __va(PFN_PHYS(pfn));
539 make_lowmem_page_readonly((void *)va);
540 make_lowmem_page_readonly(virt);
542 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
547 * load_gdt for early boot, when the gdt is only mapped once
549 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
551 unsigned long va = dtr->address;
552 unsigned int size = dtr->size + 1;
553 unsigned long pfn, mfn;
556 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
557 BUG_ON(size > PAGE_SIZE);
558 BUG_ON(va & ~PAGE_MASK);
560 pfn = virt_to_pfn((void *)va);
561 mfn = pfn_to_mfn(pfn);
563 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
565 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
568 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
572 static inline bool desc_equal(const struct desc_struct *d1,
573 const struct desc_struct *d2)
575 return !memcmp(d1, d2, sizeof(*d1));
578 static void load_TLS_descriptor(struct thread_struct *t,
579 unsigned int cpu, unsigned int i)
581 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
582 struct desc_struct *gdt;
584 struct multicall_space mc;
586 if (desc_equal(shadow, &t->tls_array[i]))
589 *shadow = t->tls_array[i];
591 gdt = get_cpu_gdt_rw(cpu);
592 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
593 mc = __xen_mc_entry(0);
595 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
598 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
601 * In lazy mode we need to zero %fs, otherwise we may get an
602 * exception between the new %fs descriptor being loaded and
603 * %fs being effectively cleared at __switch_to().
605 if (xen_get_lazy_mode() == XEN_LAZY_CPU)
610 load_TLS_descriptor(t, cpu, 0);
611 load_TLS_descriptor(t, cpu, 1);
612 load_TLS_descriptor(t, cpu, 2);
614 xen_mc_issue(XEN_LAZY_CPU);
617 static void xen_load_gs_index(unsigned int idx)
619 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
623 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
626 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
627 u64 entry = *(u64 *)ptr;
629 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
634 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
640 void noist_exc_debug(struct pt_regs *regs);
642 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
644 /* On Xen PV, NMI doesn't use IST. The C part is the same as native. */
648 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
650 /* On Xen PV, DF doesn't use IST. The C part is the same as native. */
651 exc_double_fault(regs, error_code);
654 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
657 * There's no IST on Xen PV, but we still need to dispatch
658 * to the correct handler.
661 noist_exc_debug(regs);
666 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
668 /* This should never happen and there is no way to handle it. */
669 instrumentation_begin();
670 pr_err("Unknown trap in Xen PV mode.");
672 instrumentation_end();
675 #ifdef CONFIG_X86_MCE
676 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
679 * There's no IST on Xen PV, but we still need to dispatch
680 * to the correct handler.
683 noist_exc_machine_check(regs);
685 exc_machine_check(regs);
689 struct trap_array_entry {
695 #define TRAP_ENTRY(func, ist_ok) { \
696 .orig = asm_##func, \
697 .xen = xen_asm_##func, \
700 #define TRAP_ENTRY_REDIR(func, ist_ok) { \
701 .orig = asm_##func, \
702 .xen = xen_asm_xenpv_##func, \
705 static struct trap_array_entry trap_array[] = {
706 TRAP_ENTRY_REDIR(exc_debug, true ),
707 TRAP_ENTRY_REDIR(exc_double_fault, true ),
708 #ifdef CONFIG_X86_MCE
709 TRAP_ENTRY_REDIR(exc_machine_check, true ),
711 TRAP_ENTRY_REDIR(exc_nmi, true ),
712 TRAP_ENTRY(exc_int3, false ),
713 TRAP_ENTRY(exc_overflow, false ),
714 #ifdef CONFIG_IA32_EMULATION
715 TRAP_ENTRY(int80_emulation, false ),
717 TRAP_ENTRY(exc_page_fault, false ),
718 TRAP_ENTRY(exc_divide_error, false ),
719 TRAP_ENTRY(exc_bounds, false ),
720 TRAP_ENTRY(exc_invalid_op, false ),
721 TRAP_ENTRY(exc_device_not_available, false ),
722 TRAP_ENTRY(exc_coproc_segment_overrun, false ),
723 TRAP_ENTRY(exc_invalid_tss, false ),
724 TRAP_ENTRY(exc_segment_not_present, false ),
725 TRAP_ENTRY(exc_stack_segment, false ),
726 TRAP_ENTRY(exc_general_protection, false ),
727 TRAP_ENTRY(exc_spurious_interrupt_bug, false ),
728 TRAP_ENTRY(exc_coprocessor_error, false ),
729 TRAP_ENTRY(exc_alignment_check, false ),
730 TRAP_ENTRY(exc_simd_coprocessor_error, false ),
731 #ifdef CONFIG_X86_CET
732 TRAP_ENTRY(exc_control_protection, false ),
736 static bool __ref get_trap_addr(void **addr, unsigned int ist)
739 bool ist_okay = false;
743 * Replace trap handler addresses by Xen specific ones.
744 * Check for known traps using IST and whitelist them.
745 * The debugger ones are the only ones we care about.
746 * Xen will handle faults like double_fault, so we should never see
747 * them. Warn if there's an unexpected IST-using fault handler.
749 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
750 struct trap_array_entry *entry = trap_array + nr;
752 if (*addr == entry->orig) {
754 ist_okay = entry->ist_okay;
760 if (nr == ARRAY_SIZE(trap_array) &&
761 *addr >= (void *)early_idt_handler_array[0] &&
762 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
763 nr = (*addr - (void *)early_idt_handler_array[0]) /
764 EARLY_IDT_HANDLER_SIZE;
765 *addr = (void *)xen_early_idt_handler_array[nr];
770 *addr = (void *)xen_asm_exc_xen_unknown_trap;
772 if (WARN_ON(found && ist != 0 && !ist_okay))
778 static int cvt_gate_to_trap(int vector, const gate_desc *val,
779 struct trap_info *info)
783 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
786 info->vector = vector;
788 addr = gate_offset(val);
789 if (!get_trap_addr((void **)&addr, val->bits.ist))
791 info->address = addr;
793 info->cs = gate_segment(val);
794 info->flags = val->bits.dpl;
795 /* interrupt gates clear IF */
796 if (val->bits.type == GATE_INTERRUPT)
797 info->flags |= 1 << 2;
802 /* Locations of each CPU's IDT */
803 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
805 /* Set an IDT entry. If the entry is part of the current IDT, then
807 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
809 unsigned long p = (unsigned long)&dt[entrynum];
810 unsigned long start, end;
812 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
816 start = __this_cpu_read(idt_desc.address);
817 end = start + __this_cpu_read(idt_desc.size) + 1;
821 native_write_idt_entry(dt, entrynum, g);
823 if (p >= start && (p + 8) <= end) {
824 struct trap_info info[2];
828 if (cvt_gate_to_trap(entrynum, g, &info[0]))
829 if (HYPERVISOR_set_trap_table(info))
836 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
837 struct trap_info *traps, bool full)
839 unsigned in, out, count;
841 count = (desc->size+1) / sizeof(gate_desc);
844 for (in = out = 0; in < count; in++) {
845 gate_desc *entry = (gate_desc *)(desc->address) + in;
847 if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
854 void xen_copy_trap_info(struct trap_info *traps)
856 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
858 xen_convert_trap_info(desc, traps, true);
861 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
862 hold a spinlock to protect the static traps[] array (static because
863 it avoids allocation, and saves stack space). */
864 static void xen_load_idt(const struct desc_ptr *desc)
866 static DEFINE_SPINLOCK(lock);
867 static struct trap_info traps[257];
868 static const struct trap_info zero = { };
871 trace_xen_cpu_load_idt(desc);
875 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
877 out = xen_convert_trap_info(desc, traps, false);
881 if (HYPERVISOR_set_trap_table(traps))
887 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
888 they're handled differently. */
889 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
890 const void *desc, int type)
892 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
903 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
906 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
916 * Version of write_gdt_entry for use at early boot-time needed to
917 * update an entry as simply as possible.
919 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
920 const void *desc, int type)
922 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
931 xmaddr_t maddr = virt_to_machine(&dt[entry]);
933 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
934 dt[entry] = *(struct desc_struct *)desc;
940 static void xen_load_sp0(unsigned long sp0)
942 struct multicall_space mcs;
944 mcs = xen_mc_entry(0);
945 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
946 xen_mc_issue(XEN_LAZY_CPU);
947 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
950 #ifdef CONFIG_X86_IOPL_IOPERM
951 static void xen_invalidate_io_bitmap(void)
953 struct physdev_set_iobitmap iobitmap = {
958 native_tss_invalidate_io_bitmap();
959 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
962 static void xen_update_io_bitmap(void)
964 struct physdev_set_iobitmap iobitmap;
965 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
967 native_tss_update_io_bitmap();
969 iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
970 tss->x86_tss.io_bitmap_base;
971 if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
972 iobitmap.nr_ports = 0;
974 iobitmap.nr_ports = IO_BITMAP_BITS;
976 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
980 static void xen_io_delay(void)
984 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
986 static unsigned long xen_read_cr0(void)
988 unsigned long cr0 = this_cpu_read(xen_cr0_value);
990 if (unlikely(cr0 == 0)) {
991 cr0 = native_read_cr0();
992 this_cpu_write(xen_cr0_value, cr0);
998 static void xen_write_cr0(unsigned long cr0)
1000 struct multicall_space mcs;
1002 this_cpu_write(xen_cr0_value, cr0);
1004 /* Only pay attention to cr0.TS; everything else is
1006 mcs = xen_mc_entry(0);
1008 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1010 xen_mc_issue(XEN_LAZY_CPU);
1013 static void xen_write_cr4(unsigned long cr4)
1015 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1017 native_write_cr4(cr4);
1020 static u64 xen_do_read_msr(unsigned int msr, int *err)
1022 u64 val = 0; /* Avoid uninitialized value for safe variant. */
1024 if (pmu_msr_read(msr, &val, err))
1028 val = native_read_msr_safe(msr, err);
1030 val = native_read_msr(msr);
1033 case MSR_IA32_APICBASE:
1034 val &= ~X2APIC_ENABLE;
1035 if (smp_processor_id() == 0)
1036 val |= MSR_IA32_APICBASE_BSP;
1038 val &= ~MSR_IA32_APICBASE_BSP;
1044 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
1047 u64 base = ((u64)high << 32) | low;
1049 if (HYPERVISOR_set_segment_base(which, base) == 0)
1055 WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
1059 * Support write_msr_safe() and write_msr() semantics.
1060 * With err == NULL write_msr() semantics are selected.
1061 * Supplying an err pointer requires err to be pre-initialized with 0.
1063 static void xen_do_write_msr(unsigned int msr, unsigned int low,
1064 unsigned int high, int *err)
1068 set_seg(SEGBASE_FS, low, high, err);
1071 case MSR_KERNEL_GS_BASE:
1072 set_seg(SEGBASE_GS_USER, low, high, err);
1076 set_seg(SEGBASE_GS_KERNEL, low, high, err);
1082 case MSR_SYSCALL_MASK:
1083 case MSR_IA32_SYSENTER_CS:
1084 case MSR_IA32_SYSENTER_ESP:
1085 case MSR_IA32_SYSENTER_EIP:
1086 /* Fast syscall setup is all done in hypercalls, so
1087 these are all ignored. Stub them out here to stop
1088 Xen console noise. */
1092 if (!pmu_msr_write(msr, low, high, err)) {
1094 *err = native_write_msr_safe(msr, low, high);
1096 native_write_msr(msr, low, high);
1101 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1103 return xen_do_read_msr(msr, err);
1106 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1111 xen_do_write_msr(msr, low, high, &err);
1116 static u64 xen_read_msr(unsigned int msr)
1120 return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1123 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1127 xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1130 /* This is called once we have the cpu_possible_mask */
1131 void __init xen_setup_vcpu_info_placement(void)
1135 for_each_possible_cpu(cpu) {
1136 /* Set up direct vCPU id mapping for PV guests. */
1137 per_cpu(xen_vcpu_id, cpu) = cpu;
1138 xen_vcpu_setup(cpu);
1141 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1142 pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1143 pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1144 pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1147 static const struct pv_info xen_info __initconst = {
1148 .extra_user_64bit_cs = FLAT_USER_CS64,
1152 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1156 .set_debugreg = xen_set_debugreg,
1157 .get_debugreg = xen_get_debugreg,
1159 .read_cr0 = xen_read_cr0,
1160 .write_cr0 = xen_write_cr0,
1162 .write_cr4 = xen_write_cr4,
1164 .wbinvd = pv_native_wbinvd,
1166 .read_msr = xen_read_msr,
1167 .write_msr = xen_write_msr,
1169 .read_msr_safe = xen_read_msr_safe,
1170 .write_msr_safe = xen_write_msr_safe,
1172 .read_pmc = xen_read_pmc,
1174 .load_tr_desc = paravirt_nop,
1175 .set_ldt = xen_set_ldt,
1176 .load_gdt = xen_load_gdt,
1177 .load_idt = xen_load_idt,
1178 .load_tls = xen_load_tls,
1179 .load_gs_index = xen_load_gs_index,
1181 .alloc_ldt = xen_alloc_ldt,
1182 .free_ldt = xen_free_ldt,
1184 .store_tr = xen_store_tr,
1186 .write_ldt_entry = xen_write_ldt_entry,
1187 .write_gdt_entry = xen_write_gdt_entry,
1188 .write_idt_entry = xen_write_idt_entry,
1189 .load_sp0 = xen_load_sp0,
1191 #ifdef CONFIG_X86_IOPL_IOPERM
1192 .invalidate_io_bitmap = xen_invalidate_io_bitmap,
1193 .update_io_bitmap = xen_update_io_bitmap,
1195 .io_delay = xen_io_delay,
1197 .start_context_switch = xen_start_context_switch,
1198 .end_context_switch = xen_end_context_switch,
1202 static void xen_restart(char *msg)
1204 xen_reboot(SHUTDOWN_reboot);
1207 static void xen_machine_halt(void)
1209 xen_reboot(SHUTDOWN_poweroff);
1212 static void xen_machine_power_off(void)
1214 do_kernel_power_off();
1215 xen_reboot(SHUTDOWN_poweroff);
1218 static void xen_crash_shutdown(struct pt_regs *regs)
1220 xen_reboot(SHUTDOWN_crash);
1223 static const struct machine_ops xen_machine_ops __initconst = {
1224 .restart = xen_restart,
1225 .halt = xen_machine_halt,
1226 .power_off = xen_machine_power_off,
1227 .shutdown = xen_machine_halt,
1228 .crash_shutdown = xen_crash_shutdown,
1229 .emergency_restart = xen_emergency_restart,
1232 static unsigned char xen_get_nmi_reason(void)
1234 unsigned char reason = 0;
1236 /* Construct a value which looks like it came from port 0x61. */
1237 if (test_bit(_XEN_NMIREASON_io_error,
1238 &HYPERVISOR_shared_info->arch.nmi_reason))
1239 reason |= NMI_REASON_IOCHK;
1240 if (test_bit(_XEN_NMIREASON_pci_serr,
1241 &HYPERVISOR_shared_info->arch.nmi_reason))
1242 reason |= NMI_REASON_SERR;
1247 static void __init xen_boot_params_init_edd(void)
1249 #if IS_ENABLED(CONFIG_EDD)
1250 struct xen_platform_op op;
1251 struct edd_info *edd_info;
1256 edd_info = boot_params.eddbuf;
1257 mbr_signature = boot_params.edd_mbr_sig_buffer;
1259 op.cmd = XENPF_firmware_info;
1261 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1262 for (nr = 0; nr < EDDMAXNR; nr++) {
1263 struct edd_info *info = edd_info + nr;
1265 op.u.firmware_info.index = nr;
1266 info->params.length = sizeof(info->params);
1267 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1269 ret = HYPERVISOR_platform_op(&op);
1273 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1276 C(interface_support);
1277 C(legacy_max_cylinder);
1279 C(legacy_sectors_per_track);
1282 boot_params.eddbuf_entries = nr;
1284 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1285 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1286 op.u.firmware_info.index = nr;
1287 ret = HYPERVISOR_platform_op(&op);
1290 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1292 boot_params.edd_mbr_sig_buf_entries = nr;
1297 * Set up the GDT and segment registers for -fstack-protector. Until
1298 * we do this, we have to be careful not to call any stack-protected
1299 * function, which is most of the kernel.
1301 static void __init xen_setup_gdt(int cpu)
1303 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1304 pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1306 switch_gdt_and_percpu_base(cpu);
1308 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1309 pv_ops.cpu.load_gdt = xen_load_gdt;
1312 static void __init xen_dom0_set_legacy_features(void)
1314 x86_platform.legacy.rtc = 1;
1317 static void __init xen_domu_set_legacy_features(void)
1319 x86_platform.legacy.rtc = 0;
1322 extern void early_xen_iret_patch(void);
1324 /* First C function to be called on Xen boot */
1325 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1327 struct physdev_set_iopl set_iopl;
1328 unsigned long initrd_start = 0;
1336 xen_start_info = si;
1338 __text_gen_insn(&early_xen_iret_patch,
1339 JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1342 xen_domain_type = XEN_PV_DOMAIN;
1343 xen_start_flags = xen_start_info->flags;
1344 /* Interrupts are guaranteed to be off initially. */
1345 early_boot_irqs_disabled = true;
1346 static_call_update_early(xen_hypercall, xen_hypercall_pv);
1348 xen_setup_features();
1350 /* Install Xen paravirt ops */
1352 pv_ops.cpu = xen_cpu_ops.cpu;
1356 * Setup xen_vcpu early because it is needed for
1357 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1359 * Don't do the full vcpu_info placement stuff until we have
1360 * the cpu_possible_mask and a non-dummy shared_info.
1362 xen_vcpu_info_reset(0);
1364 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1365 x86_platform.realmode_reserve = x86_init_noop;
1366 x86_platform.realmode_init = x86_init_noop;
1368 x86_init.resources.memory_setup = xen_memory_setup;
1369 x86_init.irqs.intr_mode_select = x86_init_noop;
1370 x86_init.irqs.intr_mode_init = x86_64_probe_apic;
1371 x86_init.oem.arch_setup = xen_arch_setup;
1372 x86_init.oem.banner = xen_banner;
1373 x86_init.hyper.init_platform = xen_pv_init_platform;
1374 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1377 * Set up some pagetable state before starting to set any ptes.
1380 xen_setup_machphys_mapping();
1383 /* Prevent unwanted bits from being set in PTEs. */
1384 __supported_pte_mask &= ~_PAGE_GLOBAL;
1385 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1388 xen_build_dynamic_phys_to_machine();
1390 /* Work out if we support NX */
1391 get_cpu_cap(&boot_cpu_data);
1395 * Set up kernel GDT and segment registers, mainly so that
1396 * -fstack-protector code can be executed.
1400 /* Determine virtual and physical address sizes */
1401 get_cpu_address_sizes(&boot_cpu_data);
1403 /* Let's presume PV guests always boot on vCPU with id 0. */
1404 per_cpu(xen_vcpu_id, 0) = 0;
1406 idt_setup_early_handler();
1408 xen_init_capabilities();
1411 * set up the basic apic ops.
1415 machine_ops = xen_machine_ops;
1418 * The only reliable way to retain the initial address of the
1419 * percpu gdt_page is to remember it here, so we can go and
1420 * mark it RW later, when the initial percpu area is freed.
1422 xen_initial_gdt = &per_cpu(gdt_page, 0);
1426 #ifdef CONFIG_ACPI_NUMA
1428 * The pages we from Xen are not related to machine pages, so
1429 * any NUMA information the kernel tries to get from ACPI will
1430 * be meaningless. Prevent it from trying.
1434 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1436 local_irq_disable();
1438 xen_raw_console_write("mapping kernel into physical memory\n");
1439 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1440 xen_start_info->nr_pages);
1441 xen_reserve_special_pages();
1444 * We used to do this in xen_arch_setup, but that is too late
1445 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1446 * early_amd_init which pokes 0xcf8 port.
1449 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1451 xen_raw_printk("physdev_op failed %d\n", rc);
1454 if (xen_start_info->mod_start) {
1455 if (xen_start_info->flags & SIF_MOD_START_PFN)
1456 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1458 initrd_start = __pa(xen_start_info->mod_start);
1461 /* Poke various useful things into boot_params */
1462 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1463 boot_params.hdr.ramdisk_image = initrd_start;
1464 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1465 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1466 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1468 if (!xen_initial_domain()) {
1470 x86_init.pci.arch_init = pci_xen_init;
1471 x86_platform.set_legacy_features =
1472 xen_domu_set_legacy_features;
1474 const struct dom0_vga_console_info *info =
1475 (void *)((char *)xen_start_info +
1476 xen_start_info->console.dom0.info_off);
1477 struct xen_platform_op op = {
1478 .cmd = XENPF_firmware_info,
1479 .interface_version = XENPF_INTERFACE_VERSION,
1480 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1483 x86_platform.set_legacy_features =
1484 xen_dom0_set_legacy_features;
1485 xen_init_vga(info, xen_start_info->console.dom0.info_size,
1486 &boot_params.screen_info);
1487 xen_start_info->console.domU.mfn = 0;
1488 xen_start_info->console.domU.evtchn = 0;
1490 if (HYPERVISOR_platform_op(&op) == 0)
1491 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1493 /* Make sure ACS will be enabled */
1496 xen_acpi_sleep_register();
1498 xen_boot_params_init_edd();
1502 * Disable selecting "Firmware First mode" for correctable
1503 * memory errors, as this is the duty of the hypervisor to
1506 acpi_disable_cmcff = 1;
1510 xen_add_preferred_consoles();
1513 /* PCI BIOS service won't work from a PV guest. */
1514 pci_probe &= ~PCI_PROBE_BIOS;
1516 xen_raw_console_write("about to get started...\n");
1518 /* We need this for printk timestamps */
1519 xen_setup_runstate_info(0);
1521 xen_efi_init(&boot_params);
1523 /* Start the world */
1524 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1525 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1528 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1532 if (per_cpu(xen_vcpu, cpu) == NULL)
1535 xen_setup_timer(cpu);
1537 rc = xen_smp_intr_init(cpu);
1539 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1544 rc = xen_smp_intr_init_pv(cpu);
1546 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1554 static int xen_cpu_dead_pv(unsigned int cpu)
1556 xen_smp_intr_free(cpu);
1557 xen_smp_intr_free_pv(cpu);
1559 xen_teardown_timer(cpu);
1564 static uint32_t __init xen_platform_pv(void)
1566 if (xen_pv_domain())
1567 return xen_cpuid_base();
1572 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1574 .detect = xen_platform_pv,
1575 .type = X86_HYPER_XEN_PV,
1576 .runtime.pin_vcpu = xen_pin_vcpu,
1577 .ignore_nopv = true,