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
12 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
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>
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/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>
74 #ifdef CONFIG_X86_IOPL_IOPERM
75 #include <asm/io_bitmap.h>
79 #include <linux/acpi.h>
81 #include <acpi/pdc_intel.h>
82 #include <acpi/processor.h>
83 #include <xen/interface/platform.h>
89 #include "multicalls.h"
92 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
94 void *xen_initial_gdt;
96 static int xen_cpu_up_prepare_pv(unsigned int cpu);
97 static int xen_cpu_dead_pv(unsigned int cpu);
100 struct desc_struct desc[3];
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
110 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
112 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
114 static int __init parse_xen_msr_safe(char *str)
117 return kstrtobool(str, &xen_msr_safe);
120 early_param("xen_msr_safe", parse_xen_msr_safe);
122 static void __init xen_pv_init_platform(void)
124 /* PV guests can't operate virtio devices without grants. */
125 if (IS_ENABLED(CONFIG_XEN_VIRTIO))
126 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
128 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
130 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
131 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
133 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
134 xen_vcpu_info_reset(0);
136 /* pvclock is in shared info area */
140 static void __init xen_pv_guest_late_init(void)
143 /* Setup shared vcpu info for non-smp configurations */
144 xen_setup_vcpu_info_placement();
148 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
149 static __read_mostly unsigned int cpuid_leaf5_edx_val;
151 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
152 unsigned int *cx, unsigned int *dx)
154 unsigned maskebx = ~0;
157 * Mask out inconvenient features, to try and disable as many
158 * unsupported kernel subsystems as possible.
161 case CPUID_MWAIT_LEAF:
162 /* Synthesize the values.. */
165 *cx = cpuid_leaf5_ecx_val;
166 *dx = cpuid_leaf5_edx_val;
170 /* Suppress extended topology stuff */
175 asm(XEN_EMULATE_PREFIX "cpuid"
180 : "0" (*ax), "2" (*cx));
185 static bool __init xen_check_mwait(void)
188 struct xen_platform_op op = {
189 .cmd = XENPF_set_processor_pminfo,
190 .u.set_pminfo.id = -1,
191 .u.set_pminfo.type = XEN_PM_PDC,
194 unsigned int ax, bx, cx, dx;
195 unsigned int mwait_mask;
197 /* We need to determine whether it is OK to expose the MWAIT
198 * capability to the kernel to harvest deeper than C3 states from ACPI
199 * _CST using the processor_harvest_xen.c module. For this to work, we
200 * need to gather the MWAIT_LEAF values (which the cstate.c code
201 * checks against). The hypervisor won't expose the MWAIT flag because
202 * it would break backwards compatibility; so we will find out directly
203 * from the hardware and hypercall.
205 if (!xen_initial_domain())
209 * When running under platform earlier than Xen4.2, do not expose
210 * mwait, to avoid the risk of loading native acpi pad driver
212 if (!xen_running_on_version_or_later(4, 2))
218 native_cpuid(&ax, &bx, &cx, &dx);
220 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
221 (1 << (X86_FEATURE_MWAIT % 32));
223 if ((cx & mwait_mask) != mwait_mask)
226 /* We need to emulate the MWAIT_LEAF and for that we need both
227 * ecx and edx. The hypercall provides only partial information.
230 ax = CPUID_MWAIT_LEAF;
235 native_cpuid(&ax, &bx, &cx, &dx);
237 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
238 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
240 buf[0] = ACPI_PDC_REVISION_ID;
242 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
244 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
246 if ((HYPERVISOR_platform_op(&op) == 0) &&
247 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
248 cpuid_leaf5_ecx_val = cx;
249 cpuid_leaf5_edx_val = dx;
257 static bool __init xen_check_xsave(void)
259 unsigned int cx, xsave_mask;
263 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
264 (1 << (X86_FEATURE_OSXSAVE % 32));
266 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
267 return (cx & xsave_mask) == xsave_mask;
270 static void __init xen_init_capabilities(void)
272 setup_force_cpu_cap(X86_FEATURE_XENPV);
273 setup_clear_cpu_cap(X86_FEATURE_DCA);
274 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
275 setup_clear_cpu_cap(X86_FEATURE_MTRR);
276 setup_clear_cpu_cap(X86_FEATURE_ACC);
277 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
278 setup_clear_cpu_cap(X86_FEATURE_SME);
281 * Xen PV would need some work to support PCID: CR3 handling as well
282 * as xen_flush_tlb_others() would need updating.
284 setup_clear_cpu_cap(X86_FEATURE_PCID);
286 if (!xen_initial_domain())
287 setup_clear_cpu_cap(X86_FEATURE_ACPI);
289 if (xen_check_mwait())
290 setup_force_cpu_cap(X86_FEATURE_MWAIT);
292 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
294 if (!xen_check_xsave()) {
295 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
296 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
300 static noinstr void xen_set_debugreg(int reg, unsigned long val)
302 HYPERVISOR_set_debugreg(reg, val);
305 static noinstr unsigned long xen_get_debugreg(int reg)
307 return HYPERVISOR_get_debugreg(reg);
310 static void xen_end_context_switch(struct task_struct *next)
313 paravirt_end_context_switch(next);
316 static unsigned long xen_store_tr(void)
322 * Set the page permissions for a particular virtual address. If the
323 * address is a vmalloc mapping (or other non-linear mapping), then
324 * find the linear mapping of the page and also set its protections to
327 static void set_aliased_prot(void *v, pgprot_t prot)
336 ptep = lookup_address((unsigned long)v, &level);
337 BUG_ON(ptep == NULL);
339 pfn = pte_pfn(*ptep);
340 pte = pfn_pte(pfn, prot);
343 * Careful: update_va_mapping() will fail if the virtual address
344 * we're poking isn't populated in the page tables. We don't
345 * need to worry about the direct map (that's always in the page
346 * tables), but we need to be careful about vmap space. In
347 * particular, the top level page table can lazily propagate
348 * entries between processes, so if we've switched mms since we
349 * vmapped the target in the first place, we might not have the
350 * top-level page table entry populated.
352 * We disable preemption because we want the same mm active when
353 * we probe the target and when we issue the hypercall. We'll
354 * have the same nominal mm, but if we're a kernel thread, lazy
355 * mm dropping could change our pgd.
357 * Out of an abundance of caution, this uses __get_user() to fault
358 * in the target address just in case there's some obscure case
359 * in which the target address isn't readable.
364 copy_from_kernel_nofault(&dummy, v, 1);
366 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
369 va = __va(PFN_PHYS(pfn));
371 if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
377 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
379 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
383 * We need to mark the all aliases of the LDT pages RO. We
384 * don't need to call vm_flush_aliases(), though, since that's
385 * only responsible for flushing aliases out the TLBs, not the
386 * page tables, and Xen will flush the TLB for us if needed.
388 * To avoid confusing future readers: none of this is necessary
389 * to load the LDT. The hypervisor only checks this when the
390 * LDT is faulted in due to subsequent descriptor access.
393 for (i = 0; i < entries; i += entries_per_page)
394 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
397 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
399 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
402 for (i = 0; i < entries; i += entries_per_page)
403 set_aliased_prot(ldt + i, PAGE_KERNEL);
406 static void xen_set_ldt(const void *addr, unsigned entries)
408 struct mmuext_op *op;
409 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
411 trace_xen_cpu_set_ldt(addr, entries);
414 op->cmd = MMUEXT_SET_LDT;
415 op->arg1.linear_addr = (unsigned long)addr;
416 op->arg2.nr_ents = entries;
418 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
420 xen_mc_issue(PARAVIRT_LAZY_CPU);
423 static void xen_load_gdt(const struct desc_ptr *dtr)
425 unsigned long va = dtr->address;
426 unsigned int size = dtr->size + 1;
427 unsigned long pfn, mfn;
432 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
433 BUG_ON(size > PAGE_SIZE);
434 BUG_ON(va & ~PAGE_MASK);
437 * The GDT is per-cpu and is in the percpu data area.
438 * That can be virtually mapped, so we need to do a
439 * page-walk to get the underlying MFN for the
440 * hypercall. The page can also be in the kernel's
441 * linear range, so we need to RO that mapping too.
443 ptep = lookup_address(va, &level);
444 BUG_ON(ptep == NULL);
446 pfn = pte_pfn(*ptep);
447 mfn = pfn_to_mfn(pfn);
448 virt = __va(PFN_PHYS(pfn));
450 make_lowmem_page_readonly((void *)va);
451 make_lowmem_page_readonly(virt);
453 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
458 * load_gdt for early boot, when the gdt is only mapped once
460 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
462 unsigned long va = dtr->address;
463 unsigned int size = dtr->size + 1;
464 unsigned long pfn, mfn;
467 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
468 BUG_ON(size > PAGE_SIZE);
469 BUG_ON(va & ~PAGE_MASK);
471 pfn = virt_to_pfn(va);
472 mfn = pfn_to_mfn(pfn);
474 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
476 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
479 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
483 static inline bool desc_equal(const struct desc_struct *d1,
484 const struct desc_struct *d2)
486 return !memcmp(d1, d2, sizeof(*d1));
489 static void load_TLS_descriptor(struct thread_struct *t,
490 unsigned int cpu, unsigned int i)
492 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
493 struct desc_struct *gdt;
495 struct multicall_space mc;
497 if (desc_equal(shadow, &t->tls_array[i]))
500 *shadow = t->tls_array[i];
502 gdt = get_cpu_gdt_rw(cpu);
503 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
504 mc = __xen_mc_entry(0);
506 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
509 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
512 * In lazy mode we need to zero %fs, otherwise we may get an
513 * exception between the new %fs descriptor being loaded and
514 * %fs being effectively cleared at __switch_to().
516 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
521 load_TLS_descriptor(t, cpu, 0);
522 load_TLS_descriptor(t, cpu, 1);
523 load_TLS_descriptor(t, cpu, 2);
525 xen_mc_issue(PARAVIRT_LAZY_CPU);
528 static void xen_load_gs_index(unsigned int idx)
530 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
534 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
537 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
538 u64 entry = *(u64 *)ptr;
540 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
545 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
551 void noist_exc_debug(struct pt_regs *regs);
553 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
555 /* On Xen PV, NMI doesn't use IST. The C part is the same as native. */
559 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
561 /* On Xen PV, DF doesn't use IST. The C part is the same as native. */
562 exc_double_fault(regs, error_code);
565 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
568 * There's no IST on Xen PV, but we still need to dispatch
569 * to the correct handler.
572 noist_exc_debug(regs);
577 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
579 /* This should never happen and there is no way to handle it. */
580 instrumentation_begin();
581 pr_err("Unknown trap in Xen PV mode.");
583 instrumentation_end();
586 #ifdef CONFIG_X86_MCE
587 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
590 * There's no IST on Xen PV, but we still need to dispatch
591 * to the correct handler.
594 noist_exc_machine_check(regs);
596 exc_machine_check(regs);
600 struct trap_array_entry {
606 #define TRAP_ENTRY(func, ist_ok) { \
607 .orig = asm_##func, \
608 .xen = xen_asm_##func, \
611 #define TRAP_ENTRY_REDIR(func, ist_ok) { \
612 .orig = asm_##func, \
613 .xen = xen_asm_xenpv_##func, \
616 static struct trap_array_entry trap_array[] = {
617 TRAP_ENTRY_REDIR(exc_debug, true ),
618 TRAP_ENTRY_REDIR(exc_double_fault, true ),
619 #ifdef CONFIG_X86_MCE
620 TRAP_ENTRY_REDIR(exc_machine_check, true ),
622 TRAP_ENTRY_REDIR(exc_nmi, true ),
623 TRAP_ENTRY(exc_int3, false ),
624 TRAP_ENTRY(exc_overflow, false ),
625 #ifdef CONFIG_IA32_EMULATION
626 TRAP_ENTRY(int80_emulation, false ),
628 TRAP_ENTRY(exc_page_fault, false ),
629 TRAP_ENTRY(exc_divide_error, false ),
630 TRAP_ENTRY(exc_bounds, false ),
631 TRAP_ENTRY(exc_invalid_op, false ),
632 TRAP_ENTRY(exc_device_not_available, false ),
633 TRAP_ENTRY(exc_coproc_segment_overrun, false ),
634 TRAP_ENTRY(exc_invalid_tss, false ),
635 TRAP_ENTRY(exc_segment_not_present, false ),
636 TRAP_ENTRY(exc_stack_segment, false ),
637 TRAP_ENTRY(exc_general_protection, false ),
638 TRAP_ENTRY(exc_spurious_interrupt_bug, false ),
639 TRAP_ENTRY(exc_coprocessor_error, false ),
640 TRAP_ENTRY(exc_alignment_check, false ),
641 TRAP_ENTRY(exc_simd_coprocessor_error, false ),
642 #ifdef CONFIG_X86_KERNEL_IBT
643 TRAP_ENTRY(exc_control_protection, false ),
647 static bool __ref get_trap_addr(void **addr, unsigned int ist)
650 bool ist_okay = false;
654 * Replace trap handler addresses by Xen specific ones.
655 * Check for known traps using IST and whitelist them.
656 * The debugger ones are the only ones we care about.
657 * Xen will handle faults like double_fault, so we should never see
658 * them. Warn if there's an unexpected IST-using fault handler.
660 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
661 struct trap_array_entry *entry = trap_array + nr;
663 if (*addr == entry->orig) {
665 ist_okay = entry->ist_okay;
671 if (nr == ARRAY_SIZE(trap_array) &&
672 *addr >= (void *)early_idt_handler_array[0] &&
673 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
674 nr = (*addr - (void *)early_idt_handler_array[0]) /
675 EARLY_IDT_HANDLER_SIZE;
676 *addr = (void *)xen_early_idt_handler_array[nr];
681 *addr = (void *)xen_asm_exc_xen_unknown_trap;
683 if (WARN_ON(found && ist != 0 && !ist_okay))
689 static int cvt_gate_to_trap(int vector, const gate_desc *val,
690 struct trap_info *info)
694 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
697 info->vector = vector;
699 addr = gate_offset(val);
700 if (!get_trap_addr((void **)&addr, val->bits.ist))
702 info->address = addr;
704 info->cs = gate_segment(val);
705 info->flags = val->bits.dpl;
706 /* interrupt gates clear IF */
707 if (val->bits.type == GATE_INTERRUPT)
708 info->flags |= 1 << 2;
713 /* Locations of each CPU's IDT */
714 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
716 /* Set an IDT entry. If the entry is part of the current IDT, then
718 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
720 unsigned long p = (unsigned long)&dt[entrynum];
721 unsigned long start, end;
723 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
727 start = __this_cpu_read(idt_desc.address);
728 end = start + __this_cpu_read(idt_desc.size) + 1;
732 native_write_idt_entry(dt, entrynum, g);
734 if (p >= start && (p + 8) <= end) {
735 struct trap_info info[2];
739 if (cvt_gate_to_trap(entrynum, g, &info[0]))
740 if (HYPERVISOR_set_trap_table(info))
747 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
748 struct trap_info *traps, bool full)
750 unsigned in, out, count;
752 count = (desc->size+1) / sizeof(gate_desc);
755 for (in = out = 0; in < count; in++) {
756 gate_desc *entry = (gate_desc *)(desc->address) + in;
758 if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
765 void xen_copy_trap_info(struct trap_info *traps)
767 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
769 xen_convert_trap_info(desc, traps, true);
772 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
773 hold a spinlock to protect the static traps[] array (static because
774 it avoids allocation, and saves stack space). */
775 static void xen_load_idt(const struct desc_ptr *desc)
777 static DEFINE_SPINLOCK(lock);
778 static struct trap_info traps[257];
779 static const struct trap_info zero = { };
782 trace_xen_cpu_load_idt(desc);
786 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
788 out = xen_convert_trap_info(desc, traps, false);
792 if (HYPERVISOR_set_trap_table(traps))
798 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
799 they're handled differently. */
800 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
801 const void *desc, int type)
803 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
814 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
817 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
827 * Version of write_gdt_entry for use at early boot-time needed to
828 * update an entry as simply as possible.
830 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
831 const void *desc, int type)
833 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
842 xmaddr_t maddr = virt_to_machine(&dt[entry]);
844 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
845 dt[entry] = *(struct desc_struct *)desc;
851 static void xen_load_sp0(unsigned long sp0)
853 struct multicall_space mcs;
855 mcs = xen_mc_entry(0);
856 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
857 xen_mc_issue(PARAVIRT_LAZY_CPU);
858 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
861 #ifdef CONFIG_X86_IOPL_IOPERM
862 static void xen_invalidate_io_bitmap(void)
864 struct physdev_set_iobitmap iobitmap = {
869 native_tss_invalidate_io_bitmap();
870 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
873 static void xen_update_io_bitmap(void)
875 struct physdev_set_iobitmap iobitmap;
876 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
878 native_tss_update_io_bitmap();
880 iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
881 tss->x86_tss.io_bitmap_base;
882 if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
883 iobitmap.nr_ports = 0;
885 iobitmap.nr_ports = IO_BITMAP_BITS;
887 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
891 static void xen_io_delay(void)
895 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
897 static unsigned long xen_read_cr0(void)
899 unsigned long cr0 = this_cpu_read(xen_cr0_value);
901 if (unlikely(cr0 == 0)) {
902 cr0 = native_read_cr0();
903 this_cpu_write(xen_cr0_value, cr0);
909 static void xen_write_cr0(unsigned long cr0)
911 struct multicall_space mcs;
913 this_cpu_write(xen_cr0_value, cr0);
915 /* Only pay attention to cr0.TS; everything else is
917 mcs = xen_mc_entry(0);
919 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
921 xen_mc_issue(PARAVIRT_LAZY_CPU);
924 static void xen_write_cr4(unsigned long cr4)
926 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
928 native_write_cr4(cr4);
931 static u64 xen_do_read_msr(unsigned int msr, int *err)
933 u64 val = 0; /* Avoid uninitialized value for safe variant. */
935 if (pmu_msr_read(msr, &val, err))
939 val = native_read_msr_safe(msr, err);
941 val = native_read_msr(msr);
944 case MSR_IA32_APICBASE:
945 val &= ~X2APIC_ENABLE;
951 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
954 u64 base = ((u64)high << 32) | low;
956 if (HYPERVISOR_set_segment_base(which, base) == 0)
962 WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
966 * Support write_msr_safe() and write_msr() semantics.
967 * With err == NULL write_msr() semantics are selected.
968 * Supplying an err pointer requires err to be pre-initialized with 0.
970 static void xen_do_write_msr(unsigned int msr, unsigned int low,
971 unsigned int high, int *err)
975 set_seg(SEGBASE_FS, low, high, err);
978 case MSR_KERNEL_GS_BASE:
979 set_seg(SEGBASE_GS_USER, low, high, err);
983 set_seg(SEGBASE_GS_KERNEL, low, high, err);
989 case MSR_SYSCALL_MASK:
990 case MSR_IA32_SYSENTER_CS:
991 case MSR_IA32_SYSENTER_ESP:
992 case MSR_IA32_SYSENTER_EIP:
993 /* Fast syscall setup is all done in hypercalls, so
994 these are all ignored. Stub them out here to stop
995 Xen console noise. */
999 if (!pmu_msr_write(msr, low, high, err)) {
1001 *err = native_write_msr_safe(msr, low, high);
1003 native_write_msr(msr, low, high);
1008 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1010 return xen_do_read_msr(msr, err);
1013 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1018 xen_do_write_msr(msr, low, high, &err);
1023 static u64 xen_read_msr(unsigned int msr)
1027 return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1030 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1034 xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1037 /* This is called once we have the cpu_possible_mask */
1038 void __init xen_setup_vcpu_info_placement(void)
1042 for_each_possible_cpu(cpu) {
1043 /* Set up direct vCPU id mapping for PV guests. */
1044 per_cpu(xen_vcpu_id, cpu) = cpu;
1045 xen_vcpu_setup(cpu);
1048 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1049 pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1050 pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1051 pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1054 static const struct pv_info xen_info __initconst = {
1055 .extra_user_64bit_cs = FLAT_USER_CS64,
1059 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1063 .set_debugreg = xen_set_debugreg,
1064 .get_debugreg = xen_get_debugreg,
1066 .read_cr0 = xen_read_cr0,
1067 .write_cr0 = xen_write_cr0,
1069 .write_cr4 = xen_write_cr4,
1071 .wbinvd = native_wbinvd,
1073 .read_msr = xen_read_msr,
1074 .write_msr = xen_write_msr,
1076 .read_msr_safe = xen_read_msr_safe,
1077 .write_msr_safe = xen_write_msr_safe,
1079 .read_pmc = xen_read_pmc,
1081 .load_tr_desc = paravirt_nop,
1082 .set_ldt = xen_set_ldt,
1083 .load_gdt = xen_load_gdt,
1084 .load_idt = xen_load_idt,
1085 .load_tls = xen_load_tls,
1086 .load_gs_index = xen_load_gs_index,
1088 .alloc_ldt = xen_alloc_ldt,
1089 .free_ldt = xen_free_ldt,
1091 .store_tr = xen_store_tr,
1093 .write_ldt_entry = xen_write_ldt_entry,
1094 .write_gdt_entry = xen_write_gdt_entry,
1095 .write_idt_entry = xen_write_idt_entry,
1096 .load_sp0 = xen_load_sp0,
1098 #ifdef CONFIG_X86_IOPL_IOPERM
1099 .invalidate_io_bitmap = xen_invalidate_io_bitmap,
1100 .update_io_bitmap = xen_update_io_bitmap,
1102 .io_delay = xen_io_delay,
1104 .start_context_switch = paravirt_start_context_switch,
1105 .end_context_switch = xen_end_context_switch,
1109 static void xen_restart(char *msg)
1111 xen_reboot(SHUTDOWN_reboot);
1114 static void xen_machine_halt(void)
1116 xen_reboot(SHUTDOWN_poweroff);
1119 static void xen_machine_power_off(void)
1121 do_kernel_power_off();
1122 xen_reboot(SHUTDOWN_poweroff);
1125 static void xen_crash_shutdown(struct pt_regs *regs)
1127 xen_reboot(SHUTDOWN_crash);
1130 static const struct machine_ops xen_machine_ops __initconst = {
1131 .restart = xen_restart,
1132 .halt = xen_machine_halt,
1133 .power_off = xen_machine_power_off,
1134 .shutdown = xen_machine_halt,
1135 .crash_shutdown = xen_crash_shutdown,
1136 .emergency_restart = xen_emergency_restart,
1139 static unsigned char xen_get_nmi_reason(void)
1141 unsigned char reason = 0;
1143 /* Construct a value which looks like it came from port 0x61. */
1144 if (test_bit(_XEN_NMIREASON_io_error,
1145 &HYPERVISOR_shared_info->arch.nmi_reason))
1146 reason |= NMI_REASON_IOCHK;
1147 if (test_bit(_XEN_NMIREASON_pci_serr,
1148 &HYPERVISOR_shared_info->arch.nmi_reason))
1149 reason |= NMI_REASON_SERR;
1154 static void __init xen_boot_params_init_edd(void)
1156 #if IS_ENABLED(CONFIG_EDD)
1157 struct xen_platform_op op;
1158 struct edd_info *edd_info;
1163 edd_info = boot_params.eddbuf;
1164 mbr_signature = boot_params.edd_mbr_sig_buffer;
1166 op.cmd = XENPF_firmware_info;
1168 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1169 for (nr = 0; nr < EDDMAXNR; nr++) {
1170 struct edd_info *info = edd_info + nr;
1172 op.u.firmware_info.index = nr;
1173 info->params.length = sizeof(info->params);
1174 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1176 ret = HYPERVISOR_platform_op(&op);
1180 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1183 C(interface_support);
1184 C(legacy_max_cylinder);
1186 C(legacy_sectors_per_track);
1189 boot_params.eddbuf_entries = nr;
1191 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1192 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1193 op.u.firmware_info.index = nr;
1194 ret = HYPERVISOR_platform_op(&op);
1197 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1199 boot_params.edd_mbr_sig_buf_entries = nr;
1204 * Set up the GDT and segment registers for -fstack-protector. Until
1205 * we do this, we have to be careful not to call any stack-protected
1206 * function, which is most of the kernel.
1208 static void __init xen_setup_gdt(int cpu)
1210 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1211 pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1213 switch_to_new_gdt(cpu);
1215 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1216 pv_ops.cpu.load_gdt = xen_load_gdt;
1219 static void __init xen_dom0_set_legacy_features(void)
1221 x86_platform.legacy.rtc = 1;
1224 static void __init xen_domu_set_legacy_features(void)
1226 x86_platform.legacy.rtc = 0;
1229 extern void early_xen_iret_patch(void);
1231 /* First C function to be called on Xen boot */
1232 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1234 struct physdev_set_iopl set_iopl;
1235 unsigned long initrd_start = 0;
1243 xen_start_info = si;
1245 __text_gen_insn(&early_xen_iret_patch,
1246 JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1249 xen_domain_type = XEN_PV_DOMAIN;
1250 xen_start_flags = xen_start_info->flags;
1252 xen_setup_features();
1254 /* Install Xen paravirt ops */
1256 pv_ops.cpu = xen_cpu_ops.cpu;
1260 * Setup xen_vcpu early because it is needed for
1261 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1263 * Don't do the full vcpu_info placement stuff until we have
1264 * the cpu_possible_mask and a non-dummy shared_info.
1266 xen_vcpu_info_reset(0);
1268 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1269 x86_platform.realmode_reserve = x86_init_noop;
1270 x86_platform.realmode_init = x86_init_noop;
1272 x86_init.resources.memory_setup = xen_memory_setup;
1273 x86_init.irqs.intr_mode_select = x86_init_noop;
1274 x86_init.irqs.intr_mode_init = x86_init_noop;
1275 x86_init.oem.arch_setup = xen_arch_setup;
1276 x86_init.oem.banner = xen_banner;
1277 x86_init.hyper.init_platform = xen_pv_init_platform;
1278 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1281 * Set up some pagetable state before starting to set any ptes.
1284 xen_setup_machphys_mapping();
1287 /* Prevent unwanted bits from being set in PTEs. */
1288 __supported_pte_mask &= ~_PAGE_GLOBAL;
1289 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1292 xen_build_dynamic_phys_to_machine();
1294 /* Work out if we support NX */
1295 get_cpu_cap(&boot_cpu_data);
1299 * Set up kernel GDT and segment registers, mainly so that
1300 * -fstack-protector code can be executed.
1304 /* Determine virtual and physical address sizes */
1305 get_cpu_address_sizes(&boot_cpu_data);
1307 /* Let's presume PV guests always boot on vCPU with id 0. */
1308 per_cpu(xen_vcpu_id, 0) = 0;
1310 idt_setup_early_handler();
1312 xen_init_capabilities();
1314 #ifdef CONFIG_X86_LOCAL_APIC
1316 * set up the basic apic ops.
1321 machine_ops = xen_machine_ops;
1324 * The only reliable way to retain the initial address of the
1325 * percpu gdt_page is to remember it here, so we can go and
1326 * mark it RW later, when the initial percpu area is freed.
1328 xen_initial_gdt = &per_cpu(gdt_page, 0);
1332 #ifdef CONFIG_ACPI_NUMA
1334 * The pages we from Xen are not related to machine pages, so
1335 * any NUMA information the kernel tries to get from ACPI will
1336 * be meaningless. Prevent it from trying.
1340 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1342 local_irq_disable();
1343 early_boot_irqs_disabled = true;
1345 xen_raw_console_write("mapping kernel into physical memory\n");
1346 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1347 xen_start_info->nr_pages);
1348 xen_reserve_special_pages();
1351 * We used to do this in xen_arch_setup, but that is too late
1352 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1353 * early_amd_init which pokes 0xcf8 port.
1356 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1358 xen_raw_printk("physdev_op failed %d\n", rc);
1361 if (xen_start_info->mod_start) {
1362 if (xen_start_info->flags & SIF_MOD_START_PFN)
1363 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1365 initrd_start = __pa(xen_start_info->mod_start);
1368 /* Poke various useful things into boot_params */
1369 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1370 boot_params.hdr.ramdisk_image = initrd_start;
1371 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1372 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1373 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1375 if (!xen_initial_domain()) {
1377 x86_init.pci.arch_init = pci_xen_init;
1378 x86_platform.set_legacy_features =
1379 xen_domu_set_legacy_features;
1381 const struct dom0_vga_console_info *info =
1382 (void *)((char *)xen_start_info +
1383 xen_start_info->console.dom0.info_off);
1384 struct xen_platform_op op = {
1385 .cmd = XENPF_firmware_info,
1386 .interface_version = XENPF_INTERFACE_VERSION,
1387 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1390 x86_platform.set_legacy_features =
1391 xen_dom0_set_legacy_features;
1392 xen_init_vga(info, xen_start_info->console.dom0.info_size,
1393 &boot_params.screen_info);
1394 xen_start_info->console.domU.mfn = 0;
1395 xen_start_info->console.domU.evtchn = 0;
1397 if (HYPERVISOR_platform_op(&op) == 0)
1398 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1400 /* Make sure ACS will be enabled */
1403 xen_acpi_sleep_register();
1405 xen_boot_params_init_edd();
1409 * Disable selecting "Firmware First mode" for correctable
1410 * memory errors, as this is the duty of the hypervisor to
1413 acpi_disable_cmcff = 1;
1417 xen_add_preferred_consoles();
1420 /* PCI BIOS service won't work from a PV guest. */
1421 pci_probe &= ~PCI_PROBE_BIOS;
1423 xen_raw_console_write("about to get started...\n");
1425 /* We need this for printk timestamps */
1426 xen_setup_runstate_info(0);
1428 xen_efi_init(&boot_params);
1430 /* Start the world */
1431 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1432 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1435 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1439 if (per_cpu(xen_vcpu, cpu) == NULL)
1442 xen_setup_timer(cpu);
1444 rc = xen_smp_intr_init(cpu);
1446 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1451 rc = xen_smp_intr_init_pv(cpu);
1453 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1461 static int xen_cpu_dead_pv(unsigned int cpu)
1463 xen_smp_intr_free(cpu);
1464 xen_smp_intr_free_pv(cpu);
1466 xen_teardown_timer(cpu);
1471 static uint32_t __init xen_platform_pv(void)
1473 if (xen_pv_domain())
1474 return xen_cpuid_base();
1479 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1481 .detect = xen_platform_pv,
1482 .type = X86_HYPER_XEN_PV,
1483 .runtime.pin_vcpu = xen_pin_vcpu,
1484 .ignore_nopv = true,