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/memblock.h>
27 #include <linux/export.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/frame.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>
48 #include <xen/hvc-console.h>
51 #include <asm/paravirt.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>
65 #include <asm/pgalloc.h>
66 #include <asm/pgtable.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/stackprotector.h>
70 #include <asm/hypervisor.h>
71 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
77 #include <linux/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
87 #include "multicalls.h"
90 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
92 void *xen_initial_gdt;
94 static int xen_cpu_up_prepare_pv(unsigned int cpu);
95 static int xen_cpu_dead_pv(unsigned int cpu);
98 struct desc_struct desc[3];
102 * Updating the 3 TLS descriptors in the GDT on every task switch is
103 * surprisingly expensive so we avoid updating them if they haven't
104 * changed. Since Xen writes different descriptors than the one
105 * passed in the update_descriptor hypercall we keep shadow copies to
108 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
110 static void __init xen_banner(void)
112 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
113 struct xen_extraversion extra;
114 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
116 pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
117 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
118 version >> 16, version & 0xffff, extra.extraversion,
119 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
122 static void __init xen_pv_init_platform(void)
124 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
126 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
127 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
129 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
130 xen_vcpu_info_reset(0);
132 /* pvclock is in shared info area */
136 static void __init xen_pv_guest_late_init(void)
139 /* Setup shared vcpu info for non-smp configurations */
140 xen_setup_vcpu_info_placement();
144 /* Check if running on Xen version (major, minor) or later */
146 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
148 unsigned int version;
153 version = HYPERVISOR_xen_version(XENVER_version, NULL);
154 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
155 ((version >> 16) > major))
160 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
161 static __read_mostly unsigned int cpuid_leaf5_edx_val;
163 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
164 unsigned int *cx, unsigned int *dx)
166 unsigned maskebx = ~0;
169 * Mask out inconvenient features, to try and disable as many
170 * unsupported kernel subsystems as possible.
173 case CPUID_MWAIT_LEAF:
174 /* Synthesize the values.. */
177 *cx = cpuid_leaf5_ecx_val;
178 *dx = cpuid_leaf5_edx_val;
182 /* Suppress extended topology stuff */
187 asm(XEN_EMULATE_PREFIX "cpuid"
192 : "0" (*ax), "2" (*cx));
196 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
198 static bool __init xen_check_mwait(void)
201 struct xen_platform_op op = {
202 .cmd = XENPF_set_processor_pminfo,
203 .u.set_pminfo.id = -1,
204 .u.set_pminfo.type = XEN_PM_PDC,
207 unsigned int ax, bx, cx, dx;
208 unsigned int mwait_mask;
210 /* We need to determine whether it is OK to expose the MWAIT
211 * capability to the kernel to harvest deeper than C3 states from ACPI
212 * _CST using the processor_harvest_xen.c module. For this to work, we
213 * need to gather the MWAIT_LEAF values (which the cstate.c code
214 * checks against). The hypervisor won't expose the MWAIT flag because
215 * it would break backwards compatibility; so we will find out directly
216 * from the hardware and hypercall.
218 if (!xen_initial_domain())
222 * When running under platform earlier than Xen4.2, do not expose
223 * mwait, to avoid the risk of loading native acpi pad driver
225 if (!xen_running_on_version_or_later(4, 2))
231 native_cpuid(&ax, &bx, &cx, &dx);
233 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
234 (1 << (X86_FEATURE_MWAIT % 32));
236 if ((cx & mwait_mask) != mwait_mask)
239 /* We need to emulate the MWAIT_LEAF and for that we need both
240 * ecx and edx. The hypercall provides only partial information.
243 ax = CPUID_MWAIT_LEAF;
248 native_cpuid(&ax, &bx, &cx, &dx);
250 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
251 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
253 buf[0] = ACPI_PDC_REVISION_ID;
255 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
257 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
259 if ((HYPERVISOR_platform_op(&op) == 0) &&
260 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
261 cpuid_leaf5_ecx_val = cx;
262 cpuid_leaf5_edx_val = dx;
270 static bool __init xen_check_xsave(void)
272 unsigned int cx, xsave_mask;
276 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
277 (1 << (X86_FEATURE_OSXSAVE % 32));
279 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
280 return (cx & xsave_mask) == xsave_mask;
283 static void __init xen_init_capabilities(void)
285 setup_force_cpu_cap(X86_FEATURE_XENPV);
286 setup_clear_cpu_cap(X86_FEATURE_DCA);
287 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
288 setup_clear_cpu_cap(X86_FEATURE_MTRR);
289 setup_clear_cpu_cap(X86_FEATURE_ACC);
290 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
291 setup_clear_cpu_cap(X86_FEATURE_SME);
294 * Xen PV would need some work to support PCID: CR3 handling as well
295 * as xen_flush_tlb_others() would need updating.
297 setup_clear_cpu_cap(X86_FEATURE_PCID);
299 if (!xen_initial_domain())
300 setup_clear_cpu_cap(X86_FEATURE_ACPI);
302 if (xen_check_mwait())
303 setup_force_cpu_cap(X86_FEATURE_MWAIT);
305 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
307 if (!xen_check_xsave()) {
308 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
309 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
313 static void xen_set_debugreg(int reg, unsigned long val)
315 HYPERVISOR_set_debugreg(reg, val);
318 static unsigned long xen_get_debugreg(int reg)
320 return HYPERVISOR_get_debugreg(reg);
323 static void xen_end_context_switch(struct task_struct *next)
326 paravirt_end_context_switch(next);
329 static unsigned long xen_store_tr(void)
335 * Set the page permissions for a particular virtual address. If the
336 * address is a vmalloc mapping (or other non-linear mapping), then
337 * find the linear mapping of the page and also set its protections to
340 static void set_aliased_prot(void *v, pgprot_t prot)
349 ptep = lookup_address((unsigned long)v, &level);
350 BUG_ON(ptep == NULL);
352 pfn = pte_pfn(*ptep);
353 page = pfn_to_page(pfn);
355 pte = pfn_pte(pfn, prot);
358 * Careful: update_va_mapping() will fail if the virtual address
359 * we're poking isn't populated in the page tables. We don't
360 * need to worry about the direct map (that's always in the page
361 * tables), but we need to be careful about vmap space. In
362 * particular, the top level page table can lazily propagate
363 * entries between processes, so if we've switched mms since we
364 * vmapped the target in the first place, we might not have the
365 * top-level page table entry populated.
367 * We disable preemption because we want the same mm active when
368 * we probe the target and when we issue the hypercall. We'll
369 * have the same nominal mm, but if we're a kernel thread, lazy
370 * mm dropping could change our pgd.
372 * Out of an abundance of caution, this uses __get_user() to fault
373 * in the target address just in case there's some obscure case
374 * in which the target address isn't readable.
379 probe_kernel_read(&dummy, v, 1);
381 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
384 if (!PageHighMem(page)) {
385 void *av = __va(PFN_PHYS(pfn));
388 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
396 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
398 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
402 * We need to mark the all aliases of the LDT pages RO. We
403 * don't need to call vm_flush_aliases(), though, since that's
404 * only responsible for flushing aliases out the TLBs, not the
405 * page tables, and Xen will flush the TLB for us if needed.
407 * To avoid confusing future readers: none of this is necessary
408 * to load the LDT. The hypervisor only checks this when the
409 * LDT is faulted in due to subsequent descriptor access.
412 for (i = 0; i < entries; i += entries_per_page)
413 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
416 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
418 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
421 for (i = 0; i < entries; i += entries_per_page)
422 set_aliased_prot(ldt + i, PAGE_KERNEL);
425 static void xen_set_ldt(const void *addr, unsigned entries)
427 struct mmuext_op *op;
428 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
430 trace_xen_cpu_set_ldt(addr, entries);
433 op->cmd = MMUEXT_SET_LDT;
434 op->arg1.linear_addr = (unsigned long)addr;
435 op->arg2.nr_ents = entries;
437 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
439 xen_mc_issue(PARAVIRT_LAZY_CPU);
442 static void xen_load_gdt(const struct desc_ptr *dtr)
444 unsigned long va = dtr->address;
445 unsigned int size = dtr->size + 1;
446 unsigned long pfn, mfn;
451 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
452 BUG_ON(size > PAGE_SIZE);
453 BUG_ON(va & ~PAGE_MASK);
456 * The GDT is per-cpu and is in the percpu data area.
457 * That can be virtually mapped, so we need to do a
458 * page-walk to get the underlying MFN for the
459 * hypercall. The page can also be in the kernel's
460 * linear range, so we need to RO that mapping too.
462 ptep = lookup_address(va, &level);
463 BUG_ON(ptep == NULL);
465 pfn = pte_pfn(*ptep);
466 mfn = pfn_to_mfn(pfn);
467 virt = __va(PFN_PHYS(pfn));
469 make_lowmem_page_readonly((void *)va);
470 make_lowmem_page_readonly(virt);
472 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
477 * load_gdt for early boot, when the gdt is only mapped once
479 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
481 unsigned long va = dtr->address;
482 unsigned int size = dtr->size + 1;
483 unsigned long pfn, mfn;
486 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
487 BUG_ON(size > PAGE_SIZE);
488 BUG_ON(va & ~PAGE_MASK);
490 pfn = virt_to_pfn(va);
491 mfn = pfn_to_mfn(pfn);
493 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
495 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
498 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
502 static inline bool desc_equal(const struct desc_struct *d1,
503 const struct desc_struct *d2)
505 return !memcmp(d1, d2, sizeof(*d1));
508 static void load_TLS_descriptor(struct thread_struct *t,
509 unsigned int cpu, unsigned int i)
511 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
512 struct desc_struct *gdt;
514 struct multicall_space mc;
516 if (desc_equal(shadow, &t->tls_array[i]))
519 *shadow = t->tls_array[i];
521 gdt = get_cpu_gdt_rw(cpu);
522 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
523 mc = __xen_mc_entry(0);
525 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
528 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
531 * XXX sleazy hack: If we're being called in a lazy-cpu zone
532 * and lazy gs handling is enabled, it means we're in a
533 * context switch, and %gs has just been saved. This means we
534 * can zero it out to prevent faults on exit from the
535 * hypervisor if the next process has no %gs. Either way, it
536 * has been saved, and the new value will get loaded properly.
537 * This will go away as soon as Xen has been modified to not
538 * save/restore %gs for normal hypercalls.
540 * On x86_64, this hack is not used for %gs, because gs points
541 * to KERNEL_GS_BASE (and uses it for PDA references), so we
542 * must not zero %gs on x86_64
544 * For x86_64, we need to zero %fs, otherwise we may get an
545 * exception between the new %fs descriptor being loaded and
546 * %fs being effectively cleared at __switch_to().
548 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
558 load_TLS_descriptor(t, cpu, 0);
559 load_TLS_descriptor(t, cpu, 1);
560 load_TLS_descriptor(t, cpu, 2);
562 xen_mc_issue(PARAVIRT_LAZY_CPU);
566 static void xen_load_gs_index(unsigned int idx)
568 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
573 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
576 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
577 u64 entry = *(u64 *)ptr;
579 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
584 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
591 struct trap_array_entry {
597 static struct trap_array_entry trap_array[] = {
598 { debug, xen_xendebug, true },
599 { int3, xen_xenint3, true },
600 { double_fault, xen_double_fault, true },
601 #ifdef CONFIG_X86_MCE
602 { machine_check, xen_machine_check, true },
604 { nmi, xen_xennmi, true },
605 { overflow, xen_overflow, false },
606 #ifdef CONFIG_IA32_EMULATION
607 { entry_INT80_compat, xen_entry_INT80_compat, false },
609 { page_fault, xen_page_fault, false },
610 { divide_error, xen_divide_error, false },
611 { bounds, xen_bounds, false },
612 { invalid_op, xen_invalid_op, false },
613 { device_not_available, xen_device_not_available, false },
614 { coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
615 { invalid_TSS, xen_invalid_TSS, false },
616 { segment_not_present, xen_segment_not_present, false },
617 { stack_segment, xen_stack_segment, false },
618 { general_protection, xen_general_protection, false },
619 { spurious_interrupt_bug, xen_spurious_interrupt_bug, false },
620 { coprocessor_error, xen_coprocessor_error, false },
621 { alignment_check, xen_alignment_check, false },
622 { simd_coprocessor_error, xen_simd_coprocessor_error, false },
625 static bool __ref get_trap_addr(void **addr, unsigned int ist)
628 bool ist_okay = false;
631 * Replace trap handler addresses by Xen specific ones.
632 * Check for known traps using IST and whitelist them.
633 * The debugger ones are the only ones we care about.
634 * Xen will handle faults like double_fault, * so we should never see
635 * them. Warn if there's an unexpected IST-using fault handler.
637 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
638 struct trap_array_entry *entry = trap_array + nr;
640 if (*addr == entry->orig) {
642 ist_okay = entry->ist_okay;
647 if (nr == ARRAY_SIZE(trap_array) &&
648 *addr >= (void *)early_idt_handler_array[0] &&
649 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
650 nr = (*addr - (void *)early_idt_handler_array[0]) /
651 EARLY_IDT_HANDLER_SIZE;
652 *addr = (void *)xen_early_idt_handler_array[nr];
655 if (WARN_ON(ist != 0 && !ist_okay))
662 static int cvt_gate_to_trap(int vector, const gate_desc *val,
663 struct trap_info *info)
667 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
670 info->vector = vector;
672 addr = gate_offset(val);
674 if (!get_trap_addr((void **)&addr, val->bits.ist))
676 #endif /* CONFIG_X86_64 */
677 info->address = addr;
679 info->cs = gate_segment(val);
680 info->flags = val->bits.dpl;
681 /* interrupt gates clear IF */
682 if (val->bits.type == GATE_INTERRUPT)
683 info->flags |= 1 << 2;
688 /* Locations of each CPU's IDT */
689 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
691 /* Set an IDT entry. If the entry is part of the current IDT, then
693 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
695 unsigned long p = (unsigned long)&dt[entrynum];
696 unsigned long start, end;
698 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
702 start = __this_cpu_read(idt_desc.address);
703 end = start + __this_cpu_read(idt_desc.size) + 1;
707 native_write_idt_entry(dt, entrynum, g);
709 if (p >= start && (p + 8) <= end) {
710 struct trap_info info[2];
714 if (cvt_gate_to_trap(entrynum, g, &info[0]))
715 if (HYPERVISOR_set_trap_table(info))
722 static void xen_convert_trap_info(const struct desc_ptr *desc,
723 struct trap_info *traps)
725 unsigned in, out, count;
727 count = (desc->size+1) / sizeof(gate_desc);
730 for (in = out = 0; in < count; in++) {
731 gate_desc *entry = (gate_desc *)(desc->address) + in;
733 if (cvt_gate_to_trap(in, entry, &traps[out]))
736 traps[out].address = 0;
739 void xen_copy_trap_info(struct trap_info *traps)
741 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
743 xen_convert_trap_info(desc, traps);
746 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
747 hold a spinlock to protect the static traps[] array (static because
748 it avoids allocation, and saves stack space). */
749 static void xen_load_idt(const struct desc_ptr *desc)
751 static DEFINE_SPINLOCK(lock);
752 static struct trap_info traps[257];
754 trace_xen_cpu_load_idt(desc);
758 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
760 xen_convert_trap_info(desc, traps);
763 if (HYPERVISOR_set_trap_table(traps))
769 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
770 they're handled differently. */
771 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
772 const void *desc, int type)
774 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
785 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
788 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
798 * Version of write_gdt_entry for use at early boot-time needed to
799 * update an entry as simply as possible.
801 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
802 const void *desc, int type)
804 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
813 xmaddr_t maddr = virt_to_machine(&dt[entry]);
815 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
816 dt[entry] = *(struct desc_struct *)desc;
822 static void xen_load_sp0(unsigned long sp0)
824 struct multicall_space mcs;
826 mcs = xen_mc_entry(0);
827 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
828 xen_mc_issue(PARAVIRT_LAZY_CPU);
829 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
832 void xen_set_iopl_mask(unsigned mask)
834 struct physdev_set_iopl set_iopl;
836 /* Force the change at ring 0. */
837 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
838 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
841 static void xen_io_delay(void)
845 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
847 static unsigned long xen_read_cr0(void)
849 unsigned long cr0 = this_cpu_read(xen_cr0_value);
851 if (unlikely(cr0 == 0)) {
852 cr0 = native_read_cr0();
853 this_cpu_write(xen_cr0_value, cr0);
859 static void xen_write_cr0(unsigned long cr0)
861 struct multicall_space mcs;
863 this_cpu_write(xen_cr0_value, cr0);
865 /* Only pay attention to cr0.TS; everything else is
867 mcs = xen_mc_entry(0);
869 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
871 xen_mc_issue(PARAVIRT_LAZY_CPU);
874 static void xen_write_cr4(unsigned long cr4)
876 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
878 native_write_cr4(cr4);
881 static inline unsigned long xen_read_cr8(void)
885 static inline void xen_write_cr8(unsigned long val)
891 static u64 xen_read_msr_safe(unsigned int msr, int *err)
895 if (pmu_msr_read(msr, &val, err))
898 val = native_read_msr_safe(msr, err);
900 case MSR_IA32_APICBASE:
901 #ifdef CONFIG_X86_X2APIC
902 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
904 val &= ~X2APIC_ENABLE;
910 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
921 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
922 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
923 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
926 base = ((u64)high << 32) | low;
927 if (HYPERVISOR_set_segment_base(which, base) != 0)
935 case MSR_SYSCALL_MASK:
936 case MSR_IA32_SYSENTER_CS:
937 case MSR_IA32_SYSENTER_ESP:
938 case MSR_IA32_SYSENTER_EIP:
939 /* Fast syscall setup is all done in hypercalls, so
940 these are all ignored. Stub them out here to stop
941 Xen console noise. */
945 if (!pmu_msr_write(msr, low, high, &ret))
946 ret = native_write_msr_safe(msr, low, high);
952 static u64 xen_read_msr(unsigned int msr)
955 * This will silently swallow a #GP from RDMSR. It may be worth
960 return xen_read_msr_safe(msr, &err);
963 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
966 * This will silently swallow a #GP from WRMSR. It may be worth
969 xen_write_msr_safe(msr, low, high);
972 /* This is called once we have the cpu_possible_mask */
973 void __init xen_setup_vcpu_info_placement(void)
977 for_each_possible_cpu(cpu) {
978 /* Set up direct vCPU id mapping for PV guests. */
979 per_cpu(xen_vcpu_id, cpu) = cpu;
982 * xen_vcpu_setup(cpu) can fail -- in which case it
983 * falls back to the shared_info version for cpus
984 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
986 * xen_cpu_up_prepare_pv() handles the rest by failing
989 (void) xen_vcpu_setup(cpu);
993 * xen_vcpu_setup managed to place the vcpu_info within the
994 * percpu area for all cpus, so make use of it.
996 if (xen_have_vcpu_info_placement) {
997 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
998 pv_ops.irq.restore_fl =
999 __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1000 pv_ops.irq.irq_disable =
1001 __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1002 pv_ops.irq.irq_enable =
1003 __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1004 pv_ops.mmu.read_cr2 = xen_read_cr2_direct;
1008 static const struct pv_info xen_info __initconst = {
1009 .shared_kernel_pmd = 0,
1011 #ifdef CONFIG_X86_64
1012 .extra_user_64bit_cs = FLAT_USER_CS64,
1017 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1020 .set_debugreg = xen_set_debugreg,
1021 .get_debugreg = xen_get_debugreg,
1023 .read_cr0 = xen_read_cr0,
1024 .write_cr0 = xen_write_cr0,
1026 .write_cr4 = xen_write_cr4,
1028 #ifdef CONFIG_X86_64
1029 .read_cr8 = xen_read_cr8,
1030 .write_cr8 = xen_write_cr8,
1033 .wbinvd = native_wbinvd,
1035 .read_msr = xen_read_msr,
1036 .write_msr = xen_write_msr,
1038 .read_msr_safe = xen_read_msr_safe,
1039 .write_msr_safe = xen_write_msr_safe,
1041 .read_pmc = xen_read_pmc,
1044 #ifdef CONFIG_X86_64
1045 .usergs_sysret64 = xen_sysret64,
1048 .load_tr_desc = paravirt_nop,
1049 .set_ldt = xen_set_ldt,
1050 .load_gdt = xen_load_gdt,
1051 .load_idt = xen_load_idt,
1052 .load_tls = xen_load_tls,
1053 #ifdef CONFIG_X86_64
1054 .load_gs_index = xen_load_gs_index,
1057 .alloc_ldt = xen_alloc_ldt,
1058 .free_ldt = xen_free_ldt,
1060 .store_tr = xen_store_tr,
1062 .write_ldt_entry = xen_write_ldt_entry,
1063 .write_gdt_entry = xen_write_gdt_entry,
1064 .write_idt_entry = xen_write_idt_entry,
1065 .load_sp0 = xen_load_sp0,
1067 .set_iopl_mask = xen_set_iopl_mask,
1068 .io_delay = xen_io_delay,
1070 /* Xen takes care of %gs when switching to usermode for us */
1071 .swapgs = paravirt_nop,
1073 .start_context_switch = paravirt_start_context_switch,
1074 .end_context_switch = xen_end_context_switch,
1077 static void xen_restart(char *msg)
1079 xen_reboot(SHUTDOWN_reboot);
1082 static void xen_machine_halt(void)
1084 xen_reboot(SHUTDOWN_poweroff);
1087 static void xen_machine_power_off(void)
1091 xen_reboot(SHUTDOWN_poweroff);
1094 static void xen_crash_shutdown(struct pt_regs *regs)
1096 xen_reboot(SHUTDOWN_crash);
1099 static const struct machine_ops xen_machine_ops __initconst = {
1100 .restart = xen_restart,
1101 .halt = xen_machine_halt,
1102 .power_off = xen_machine_power_off,
1103 .shutdown = xen_machine_halt,
1104 .crash_shutdown = xen_crash_shutdown,
1105 .emergency_restart = xen_emergency_restart,
1108 static unsigned char xen_get_nmi_reason(void)
1110 unsigned char reason = 0;
1112 /* Construct a value which looks like it came from port 0x61. */
1113 if (test_bit(_XEN_NMIREASON_io_error,
1114 &HYPERVISOR_shared_info->arch.nmi_reason))
1115 reason |= NMI_REASON_IOCHK;
1116 if (test_bit(_XEN_NMIREASON_pci_serr,
1117 &HYPERVISOR_shared_info->arch.nmi_reason))
1118 reason |= NMI_REASON_SERR;
1123 static void __init xen_boot_params_init_edd(void)
1125 #if IS_ENABLED(CONFIG_EDD)
1126 struct xen_platform_op op;
1127 struct edd_info *edd_info;
1132 edd_info = boot_params.eddbuf;
1133 mbr_signature = boot_params.edd_mbr_sig_buffer;
1135 op.cmd = XENPF_firmware_info;
1137 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1138 for (nr = 0; nr < EDDMAXNR; nr++) {
1139 struct edd_info *info = edd_info + nr;
1141 op.u.firmware_info.index = nr;
1142 info->params.length = sizeof(info->params);
1143 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1145 ret = HYPERVISOR_platform_op(&op);
1149 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1152 C(interface_support);
1153 C(legacy_max_cylinder);
1155 C(legacy_sectors_per_track);
1158 boot_params.eddbuf_entries = nr;
1160 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1161 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1162 op.u.firmware_info.index = nr;
1163 ret = HYPERVISOR_platform_op(&op);
1166 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1168 boot_params.edd_mbr_sig_buf_entries = nr;
1173 * Set up the GDT and segment registers for -fstack-protector. Until
1174 * we do this, we have to be careful not to call any stack-protected
1175 * function, which is most of the kernel.
1177 static void __init xen_setup_gdt(int cpu)
1179 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1180 pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1182 setup_stack_canary_segment(cpu);
1183 switch_to_new_gdt(cpu);
1185 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1186 pv_ops.cpu.load_gdt = xen_load_gdt;
1189 static void __init xen_dom0_set_legacy_features(void)
1191 x86_platform.legacy.rtc = 1;
1194 /* First C function to be called on Xen boot */
1195 asmlinkage __visible void __init xen_start_kernel(void)
1197 struct physdev_set_iopl set_iopl;
1198 unsigned long initrd_start = 0;
1201 if (!xen_start_info)
1204 xen_domain_type = XEN_PV_DOMAIN;
1205 xen_start_flags = xen_start_info->flags;
1207 xen_setup_features();
1209 /* Install Xen paravirt ops */
1211 pv_ops.init.patch = paravirt_patch_default;
1212 pv_ops.cpu = xen_cpu_ops;
1216 * Setup xen_vcpu early because it is needed for
1217 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1219 * Don't do the full vcpu_info placement stuff until we have
1220 * the cpu_possible_mask and a non-dummy shared_info.
1222 xen_vcpu_info_reset(0);
1224 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1226 x86_init.resources.memory_setup = xen_memory_setup;
1227 x86_init.irqs.intr_mode_init = x86_init_noop;
1228 x86_init.oem.arch_setup = xen_arch_setup;
1229 x86_init.oem.banner = xen_banner;
1230 x86_init.hyper.init_platform = xen_pv_init_platform;
1231 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1234 * Set up some pagetable state before starting to set any ptes.
1237 xen_setup_machphys_mapping();
1240 /* Prevent unwanted bits from being set in PTEs. */
1241 __supported_pte_mask &= ~_PAGE_GLOBAL;
1242 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1245 * Prevent page tables from being allocated in highmem, even
1246 * if CONFIG_HIGHPTE is enabled.
1248 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1251 xen_build_dynamic_phys_to_machine();
1254 * Set up kernel GDT and segment registers, mainly so that
1255 * -fstack-protector code can be executed.
1259 /* Work out if we support NX */
1260 get_cpu_cap(&boot_cpu_data);
1263 /* Determine virtual and physical address sizes */
1264 get_cpu_address_sizes(&boot_cpu_data);
1266 /* Let's presume PV guests always boot on vCPU with id 0. */
1267 per_cpu(xen_vcpu_id, 0) = 0;
1269 idt_setup_early_handler();
1271 xen_init_capabilities();
1273 #ifdef CONFIG_X86_LOCAL_APIC
1275 * set up the basic apic ops.
1280 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1281 pv_ops.mmu.ptep_modify_prot_start =
1282 xen_ptep_modify_prot_start;
1283 pv_ops.mmu.ptep_modify_prot_commit =
1284 xen_ptep_modify_prot_commit;
1287 machine_ops = xen_machine_ops;
1290 * The only reliable way to retain the initial address of the
1291 * percpu gdt_page is to remember it here, so we can go and
1292 * mark it RW later, when the initial percpu area is freed.
1294 xen_initial_gdt = &per_cpu(gdt_page, 0);
1298 #ifdef CONFIG_ACPI_NUMA
1300 * The pages we from Xen are not related to machine pages, so
1301 * any NUMA information the kernel tries to get from ACPI will
1302 * be meaningless. Prevent it from trying.
1306 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1308 local_irq_disable();
1309 early_boot_irqs_disabled = true;
1311 xen_raw_console_write("mapping kernel into physical memory\n");
1312 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1313 xen_start_info->nr_pages);
1314 xen_reserve_special_pages();
1316 /* keep using Xen gdt for now; no urgent need to change it */
1318 #ifdef CONFIG_X86_32
1319 pv_info.kernel_rpl = 1;
1320 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1321 pv_info.kernel_rpl = 0;
1323 pv_info.kernel_rpl = 0;
1325 /* set the limit of our address space */
1329 * We used to do this in xen_arch_setup, but that is too late
1330 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1331 * early_amd_init which pokes 0xcf8 port.
1334 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1336 xen_raw_printk("physdev_op failed %d\n", rc);
1338 #ifdef CONFIG_X86_32
1339 /* set up basic CPUID stuff */
1340 cpu_detect(&new_cpu_data);
1341 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1342 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1345 if (xen_start_info->mod_start) {
1346 if (xen_start_info->flags & SIF_MOD_START_PFN)
1347 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1349 initrd_start = __pa(xen_start_info->mod_start);
1352 /* Poke various useful things into boot_params */
1353 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1354 boot_params.hdr.ramdisk_image = initrd_start;
1355 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1356 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1357 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1359 if (!xen_initial_domain()) {
1360 add_preferred_console("xenboot", 0, NULL);
1362 x86_init.pci.arch_init = pci_xen_init;
1364 const struct dom0_vga_console_info *info =
1365 (void *)((char *)xen_start_info +
1366 xen_start_info->console.dom0.info_off);
1367 struct xen_platform_op op = {
1368 .cmd = XENPF_firmware_info,
1369 .interface_version = XENPF_INTERFACE_VERSION,
1370 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1373 x86_platform.set_legacy_features =
1374 xen_dom0_set_legacy_features;
1375 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1376 xen_start_info->console.domU.mfn = 0;
1377 xen_start_info->console.domU.evtchn = 0;
1379 if (HYPERVISOR_platform_op(&op) == 0)
1380 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1382 /* Make sure ACS will be enabled */
1385 xen_acpi_sleep_register();
1387 /* Avoid searching for BIOS MP tables */
1388 x86_init.mpparse.find_smp_config = x86_init_noop;
1389 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1391 xen_boot_params_init_edd();
1394 if (!boot_params.screen_info.orig_video_isVGA)
1395 add_preferred_console("tty", 0, NULL);
1396 add_preferred_console("hvc", 0, NULL);
1397 if (boot_params.screen_info.orig_video_isVGA)
1398 add_preferred_console("tty", 0, NULL);
1401 /* PCI BIOS service won't work from a PV guest. */
1402 pci_probe &= ~PCI_PROBE_BIOS;
1404 xen_raw_console_write("about to get started...\n");
1406 /* We need this for printk timestamps */
1407 xen_setup_runstate_info(0);
1411 /* Start the world */
1412 #ifdef CONFIG_X86_32
1413 i386_start_kernel();
1415 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1416 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1420 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1424 if (per_cpu(xen_vcpu, cpu) == NULL)
1427 xen_setup_timer(cpu);
1429 rc = xen_smp_intr_init(cpu);
1431 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1436 rc = xen_smp_intr_init_pv(cpu);
1438 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1446 static int xen_cpu_dead_pv(unsigned int cpu)
1448 xen_smp_intr_free(cpu);
1449 xen_smp_intr_free_pv(cpu);
1451 xen_teardown_timer(cpu);
1456 static uint32_t __init xen_platform_pv(void)
1458 if (xen_pv_domain())
1459 return xen_cpuid_base();
1464 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1466 .detect = xen_platform_pv,
1467 .type = X86_HYPER_XEN_PV,
1468 .runtime.pin_vcpu = xen_pin_vcpu,