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/memblock.h>
27 #include <linux/export.h>
29 #include <linux/page-flags.h>
30 #include <linux/pci.h>
31 #include <linux/gfp.h>
32 #include <linux/edd.h>
33 #include <linux/reboot.h>
34 #include <linux/virtio_anchor.h>
37 #include <xen/events.h>
38 #include <xen/interface/xen.h>
39 #include <xen/interface/version.h>
40 #include <xen/interface/physdev.h>
41 #include <xen/interface/vcpu.h>
42 #include <xen/interface/memory.h>
43 #include <xen/interface/nmi.h>
44 #include <xen/interface/xen-mca.h>
45 #include <xen/features.h>
47 #include <xen/hvc-console.h>
50 #include <asm/paravirt.h>
53 #include <asm/xen/pci.h>
54 #include <asm/xen/hypercall.h>
55 #include <asm/xen/hypervisor.h>
56 #include <asm/xen/cpuid.h>
57 #include <asm/fixmap.h>
58 #include <asm/processor.h>
59 #include <asm/proto.h>
60 #include <asm/msr-index.h>
61 #include <asm/traps.h>
62 #include <asm/setup.h>
64 #include <asm/pgalloc.h>
65 #include <asm/tlbflush.h>
66 #include <asm/reboot.h>
67 #include <asm/stackprotector.h>
68 #include <asm/hypervisor.h>
69 #include <asm/mach_traps.h>
70 #include <asm/mwait.h>
71 #include <asm/pci_x86.h>
73 #ifdef CONFIG_X86_IOPL_IOPERM
74 #include <asm/io_bitmap.h>
78 #include <linux/acpi.h>
80 #include <acpi/pdc_intel.h>
81 #include <acpi/processor.h>
82 #include <xen/interface/platform.h>
88 #include "multicalls.h"
91 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
93 void *xen_initial_gdt;
95 static int xen_cpu_up_prepare_pv(unsigned int cpu);
96 static int xen_cpu_dead_pv(unsigned int cpu);
99 struct desc_struct desc[3];
103 * Updating the 3 TLS descriptors in the GDT on every task switch is
104 * surprisingly expensive so we avoid updating them if they haven't
105 * changed. Since Xen writes different descriptors than the one
106 * passed in the update_descriptor hypercall we keep shadow copies to
109 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
111 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
113 static int __init parse_xen_msr_safe(char *str)
116 return strtobool(str, &xen_msr_safe);
119 early_param("xen_msr_safe", parse_xen_msr_safe);
121 static void __init xen_pv_init_platform(void)
123 /* PV guests can't operate virtio devices without grants. */
124 if (IS_ENABLED(CONFIG_XEN_VIRTIO))
125 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
127 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
129 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
130 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
132 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
133 xen_vcpu_info_reset(0);
135 /* pvclock is in shared info area */
139 static void __init xen_pv_guest_late_init(void)
142 /* Setup shared vcpu info for non-smp configurations */
143 xen_setup_vcpu_info_placement();
147 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
148 static __read_mostly unsigned int cpuid_leaf5_edx_val;
150 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
151 unsigned int *cx, unsigned int *dx)
153 unsigned maskebx = ~0;
156 * Mask out inconvenient features, to try and disable as many
157 * unsupported kernel subsystems as possible.
160 case CPUID_MWAIT_LEAF:
161 /* Synthesize the values.. */
164 *cx = cpuid_leaf5_ecx_val;
165 *dx = cpuid_leaf5_edx_val;
169 /* Suppress extended topology stuff */
174 asm(XEN_EMULATE_PREFIX "cpuid"
179 : "0" (*ax), "2" (*cx));
184 static bool __init xen_check_mwait(void)
187 struct xen_platform_op op = {
188 .cmd = XENPF_set_processor_pminfo,
189 .u.set_pminfo.id = -1,
190 .u.set_pminfo.type = XEN_PM_PDC,
193 unsigned int ax, bx, cx, dx;
194 unsigned int mwait_mask;
196 /* We need to determine whether it is OK to expose the MWAIT
197 * capability to the kernel to harvest deeper than C3 states from ACPI
198 * _CST using the processor_harvest_xen.c module. For this to work, we
199 * need to gather the MWAIT_LEAF values (which the cstate.c code
200 * checks against). The hypervisor won't expose the MWAIT flag because
201 * it would break backwards compatibility; so we will find out directly
202 * from the hardware and hypercall.
204 if (!xen_initial_domain())
208 * When running under platform earlier than Xen4.2, do not expose
209 * mwait, to avoid the risk of loading native acpi pad driver
211 if (!xen_running_on_version_or_later(4, 2))
217 native_cpuid(&ax, &bx, &cx, &dx);
219 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
220 (1 << (X86_FEATURE_MWAIT % 32));
222 if ((cx & mwait_mask) != mwait_mask)
225 /* We need to emulate the MWAIT_LEAF and for that we need both
226 * ecx and edx. The hypercall provides only partial information.
229 ax = CPUID_MWAIT_LEAF;
234 native_cpuid(&ax, &bx, &cx, &dx);
236 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
237 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
239 buf[0] = ACPI_PDC_REVISION_ID;
241 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
243 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
245 if ((HYPERVISOR_platform_op(&op) == 0) &&
246 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
247 cpuid_leaf5_ecx_val = cx;
248 cpuid_leaf5_edx_val = dx;
256 static bool __init xen_check_xsave(void)
258 unsigned int cx, xsave_mask;
262 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
263 (1 << (X86_FEATURE_OSXSAVE % 32));
265 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
266 return (cx & xsave_mask) == xsave_mask;
269 static void __init xen_init_capabilities(void)
271 setup_force_cpu_cap(X86_FEATURE_XENPV);
272 setup_clear_cpu_cap(X86_FEATURE_DCA);
273 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
274 setup_clear_cpu_cap(X86_FEATURE_MTRR);
275 setup_clear_cpu_cap(X86_FEATURE_ACC);
276 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
277 setup_clear_cpu_cap(X86_FEATURE_SME);
280 * Xen PV would need some work to support PCID: CR3 handling as well
281 * as xen_flush_tlb_others() would need updating.
283 setup_clear_cpu_cap(X86_FEATURE_PCID);
285 if (!xen_initial_domain())
286 setup_clear_cpu_cap(X86_FEATURE_ACPI);
288 if (xen_check_mwait())
289 setup_force_cpu_cap(X86_FEATURE_MWAIT);
291 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
293 if (!xen_check_xsave()) {
294 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
295 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
299 static noinstr void xen_set_debugreg(int reg, unsigned long val)
301 HYPERVISOR_set_debugreg(reg, val);
304 static noinstr unsigned long xen_get_debugreg(int reg)
306 return HYPERVISOR_get_debugreg(reg);
309 static void xen_end_context_switch(struct task_struct *next)
312 paravirt_end_context_switch(next);
315 static unsigned long xen_store_tr(void)
321 * Set the page permissions for a particular virtual address. If the
322 * address is a vmalloc mapping (or other non-linear mapping), then
323 * find the linear mapping of the page and also set its protections to
326 static void set_aliased_prot(void *v, pgprot_t prot)
335 ptep = lookup_address((unsigned long)v, &level);
336 BUG_ON(ptep == NULL);
338 pfn = pte_pfn(*ptep);
339 pte = pfn_pte(pfn, prot);
342 * Careful: update_va_mapping() will fail if the virtual address
343 * we're poking isn't populated in the page tables. We don't
344 * need to worry about the direct map (that's always in the page
345 * tables), but we need to be careful about vmap space. In
346 * particular, the top level page table can lazily propagate
347 * entries between processes, so if we've switched mms since we
348 * vmapped the target in the first place, we might not have the
349 * top-level page table entry populated.
351 * We disable preemption because we want the same mm active when
352 * we probe the target and when we issue the hypercall. We'll
353 * have the same nominal mm, but if we're a kernel thread, lazy
354 * mm dropping could change our pgd.
356 * Out of an abundance of caution, this uses __get_user() to fault
357 * in the target address just in case there's some obscure case
358 * in which the target address isn't readable.
363 copy_from_kernel_nofault(&dummy, v, 1);
365 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
368 va = __va(PFN_PHYS(pfn));
370 if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
376 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
378 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
382 * We need to mark the all aliases of the LDT pages RO. We
383 * don't need to call vm_flush_aliases(), though, since that's
384 * only responsible for flushing aliases out the TLBs, not the
385 * page tables, and Xen will flush the TLB for us if needed.
387 * To avoid confusing future readers: none of this is necessary
388 * to load the LDT. The hypervisor only checks this when the
389 * LDT is faulted in due to subsequent descriptor access.
392 for (i = 0; i < entries; i += entries_per_page)
393 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
396 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
398 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
401 for (i = 0; i < entries; i += entries_per_page)
402 set_aliased_prot(ldt + i, PAGE_KERNEL);
405 static void xen_set_ldt(const void *addr, unsigned entries)
407 struct mmuext_op *op;
408 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
410 trace_xen_cpu_set_ldt(addr, entries);
413 op->cmd = MMUEXT_SET_LDT;
414 op->arg1.linear_addr = (unsigned long)addr;
415 op->arg2.nr_ents = entries;
417 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
419 xen_mc_issue(PARAVIRT_LAZY_CPU);
422 static void xen_load_gdt(const struct desc_ptr *dtr)
424 unsigned long va = dtr->address;
425 unsigned int size = dtr->size + 1;
426 unsigned long pfn, mfn;
431 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
432 BUG_ON(size > PAGE_SIZE);
433 BUG_ON(va & ~PAGE_MASK);
436 * The GDT is per-cpu and is in the percpu data area.
437 * That can be virtually mapped, so we need to do a
438 * page-walk to get the underlying MFN for the
439 * hypercall. The page can also be in the kernel's
440 * linear range, so we need to RO that mapping too.
442 ptep = lookup_address(va, &level);
443 BUG_ON(ptep == NULL);
445 pfn = pte_pfn(*ptep);
446 mfn = pfn_to_mfn(pfn);
447 virt = __va(PFN_PHYS(pfn));
449 make_lowmem_page_readonly((void *)va);
450 make_lowmem_page_readonly(virt);
452 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
457 * load_gdt for early boot, when the gdt is only mapped once
459 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
461 unsigned long va = dtr->address;
462 unsigned int size = dtr->size + 1;
463 unsigned long pfn, mfn;
466 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
467 BUG_ON(size > PAGE_SIZE);
468 BUG_ON(va & ~PAGE_MASK);
470 pfn = virt_to_pfn(va);
471 mfn = pfn_to_mfn(pfn);
473 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
475 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
478 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
482 static inline bool desc_equal(const struct desc_struct *d1,
483 const struct desc_struct *d2)
485 return !memcmp(d1, d2, sizeof(*d1));
488 static void load_TLS_descriptor(struct thread_struct *t,
489 unsigned int cpu, unsigned int i)
491 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
492 struct desc_struct *gdt;
494 struct multicall_space mc;
496 if (desc_equal(shadow, &t->tls_array[i]))
499 *shadow = t->tls_array[i];
501 gdt = get_cpu_gdt_rw(cpu);
502 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
503 mc = __xen_mc_entry(0);
505 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
508 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
511 * In lazy mode we need to zero %fs, otherwise we may get an
512 * exception between the new %fs descriptor being loaded and
513 * %fs being effectively cleared at __switch_to().
515 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
520 load_TLS_descriptor(t, cpu, 0);
521 load_TLS_descriptor(t, cpu, 1);
522 load_TLS_descriptor(t, cpu, 2);
524 xen_mc_issue(PARAVIRT_LAZY_CPU);
527 static void xen_load_gs_index(unsigned int idx)
529 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
533 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
536 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
537 u64 entry = *(u64 *)ptr;
539 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
544 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
550 void noist_exc_debug(struct pt_regs *regs);
552 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
554 /* On Xen PV, NMI doesn't use IST. The C part is the same as native. */
558 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
560 /* On Xen PV, DF doesn't use IST. The C part is the same as native. */
561 exc_double_fault(regs, error_code);
564 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
567 * There's no IST on Xen PV, but we still need to dispatch
568 * to the correct handler.
571 noist_exc_debug(regs);
576 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
578 /* This should never happen and there is no way to handle it. */
579 instrumentation_begin();
580 pr_err("Unknown trap in Xen PV mode.");
582 instrumentation_end();
585 #ifdef CONFIG_X86_MCE
586 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
589 * There's no IST on Xen PV, but we still need to dispatch
590 * to the correct handler.
593 noist_exc_machine_check(regs);
595 exc_machine_check(regs);
599 struct trap_array_entry {
605 #define TRAP_ENTRY(func, ist_ok) { \
606 .orig = asm_##func, \
607 .xen = xen_asm_##func, \
610 #define TRAP_ENTRY_REDIR(func, ist_ok) { \
611 .orig = asm_##func, \
612 .xen = xen_asm_xenpv_##func, \
615 static struct trap_array_entry trap_array[] = {
616 TRAP_ENTRY_REDIR(exc_debug, true ),
617 TRAP_ENTRY_REDIR(exc_double_fault, true ),
618 #ifdef CONFIG_X86_MCE
619 TRAP_ENTRY_REDIR(exc_machine_check, true ),
621 TRAP_ENTRY_REDIR(exc_nmi, true ),
622 TRAP_ENTRY(exc_int3, false ),
623 TRAP_ENTRY(exc_overflow, false ),
624 #ifdef CONFIG_IA32_EMULATION
625 { entry_INT80_compat, xen_entry_INT80_compat, false },
627 TRAP_ENTRY(exc_page_fault, false ),
628 TRAP_ENTRY(exc_divide_error, false ),
629 TRAP_ENTRY(exc_bounds, false ),
630 TRAP_ENTRY(exc_invalid_op, false ),
631 TRAP_ENTRY(exc_device_not_available, false ),
632 TRAP_ENTRY(exc_coproc_segment_overrun, false ),
633 TRAP_ENTRY(exc_invalid_tss, false ),
634 TRAP_ENTRY(exc_segment_not_present, false ),
635 TRAP_ENTRY(exc_stack_segment, false ),
636 TRAP_ENTRY(exc_general_protection, false ),
637 TRAP_ENTRY(exc_spurious_interrupt_bug, false ),
638 TRAP_ENTRY(exc_coprocessor_error, false ),
639 TRAP_ENTRY(exc_alignment_check, false ),
640 TRAP_ENTRY(exc_simd_coprocessor_error, false ),
641 #ifdef CONFIG_X86_KERNEL_IBT
642 TRAP_ENTRY(exc_control_protection, false ),
646 static bool __ref get_trap_addr(void **addr, unsigned int ist)
649 bool ist_okay = false;
653 * Replace trap handler addresses by Xen specific ones.
654 * Check for known traps using IST and whitelist them.
655 * The debugger ones are the only ones we care about.
656 * Xen will handle faults like double_fault, so we should never see
657 * them. Warn if there's an unexpected IST-using fault handler.
659 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
660 struct trap_array_entry *entry = trap_array + nr;
662 if (*addr == entry->orig) {
664 ist_okay = entry->ist_okay;
670 if (nr == ARRAY_SIZE(trap_array) &&
671 *addr >= (void *)early_idt_handler_array[0] &&
672 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
673 nr = (*addr - (void *)early_idt_handler_array[0]) /
674 EARLY_IDT_HANDLER_SIZE;
675 *addr = (void *)xen_early_idt_handler_array[nr];
680 *addr = (void *)xen_asm_exc_xen_unknown_trap;
682 if (WARN_ON(found && ist != 0 && !ist_okay))
688 static int cvt_gate_to_trap(int vector, const gate_desc *val,
689 struct trap_info *info)
693 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
696 info->vector = vector;
698 addr = gate_offset(val);
699 if (!get_trap_addr((void **)&addr, val->bits.ist))
701 info->address = addr;
703 info->cs = gate_segment(val);
704 info->flags = val->bits.dpl;
705 /* interrupt gates clear IF */
706 if (val->bits.type == GATE_INTERRUPT)
707 info->flags |= 1 << 2;
712 /* Locations of each CPU's IDT */
713 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
715 /* Set an IDT entry. If the entry is part of the current IDT, then
717 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
719 unsigned long p = (unsigned long)&dt[entrynum];
720 unsigned long start, end;
722 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
726 start = __this_cpu_read(idt_desc.address);
727 end = start + __this_cpu_read(idt_desc.size) + 1;
731 native_write_idt_entry(dt, entrynum, g);
733 if (p >= start && (p + 8) <= end) {
734 struct trap_info info[2];
738 if (cvt_gate_to_trap(entrynum, g, &info[0]))
739 if (HYPERVISOR_set_trap_table(info))
746 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
747 struct trap_info *traps, bool full)
749 unsigned in, out, count;
751 count = (desc->size+1) / sizeof(gate_desc);
754 for (in = out = 0; in < count; in++) {
755 gate_desc *entry = (gate_desc *)(desc->address) + in;
757 if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
764 void xen_copy_trap_info(struct trap_info *traps)
766 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
768 xen_convert_trap_info(desc, traps, true);
771 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
772 hold a spinlock to protect the static traps[] array (static because
773 it avoids allocation, and saves stack space). */
774 static void xen_load_idt(const struct desc_ptr *desc)
776 static DEFINE_SPINLOCK(lock);
777 static struct trap_info traps[257];
778 static const struct trap_info zero = { };
781 trace_xen_cpu_load_idt(desc);
785 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
787 out = xen_convert_trap_info(desc, traps, false);
791 if (HYPERVISOR_set_trap_table(traps))
797 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
798 they're handled differently. */
799 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
800 const void *desc, int type)
802 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
813 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
816 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
826 * Version of write_gdt_entry for use at early boot-time needed to
827 * update an entry as simply as possible.
829 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
830 const void *desc, int type)
832 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
841 xmaddr_t maddr = virt_to_machine(&dt[entry]);
843 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
844 dt[entry] = *(struct desc_struct *)desc;
850 static void xen_load_sp0(unsigned long sp0)
852 struct multicall_space mcs;
854 mcs = xen_mc_entry(0);
855 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
856 xen_mc_issue(PARAVIRT_LAZY_CPU);
857 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
860 #ifdef CONFIG_X86_IOPL_IOPERM
861 static void xen_invalidate_io_bitmap(void)
863 struct physdev_set_iobitmap iobitmap = {
868 native_tss_invalidate_io_bitmap();
869 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
872 static void xen_update_io_bitmap(void)
874 struct physdev_set_iobitmap iobitmap;
875 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
877 native_tss_update_io_bitmap();
879 iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
880 tss->x86_tss.io_bitmap_base;
881 if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
882 iobitmap.nr_ports = 0;
884 iobitmap.nr_ports = IO_BITMAP_BITS;
886 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
890 static void xen_io_delay(void)
894 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
896 static unsigned long xen_read_cr0(void)
898 unsigned long cr0 = this_cpu_read(xen_cr0_value);
900 if (unlikely(cr0 == 0)) {
901 cr0 = native_read_cr0();
902 this_cpu_write(xen_cr0_value, cr0);
908 static void xen_write_cr0(unsigned long cr0)
910 struct multicall_space mcs;
912 this_cpu_write(xen_cr0_value, cr0);
914 /* Only pay attention to cr0.TS; everything else is
916 mcs = xen_mc_entry(0);
918 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
920 xen_mc_issue(PARAVIRT_LAZY_CPU);
923 static void xen_write_cr4(unsigned long cr4)
925 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
927 native_write_cr4(cr4);
930 static u64 xen_do_read_msr(unsigned int msr, int *err)
932 u64 val = 0; /* Avoid uninitialized value for safe variant. */
934 if (pmu_msr_read(msr, &val, err))
938 val = native_read_msr_safe(msr, err);
940 val = native_read_msr(msr);
943 case MSR_IA32_APICBASE:
944 val &= ~X2APIC_ENABLE;
950 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
953 u64 base = ((u64)high << 32) | low;
955 if (HYPERVISOR_set_segment_base(which, base) == 0)
961 WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
965 * Support write_msr_safe() and write_msr() semantics.
966 * With err == NULL write_msr() semantics are selected.
967 * Supplying an err pointer requires err to be pre-initialized with 0.
969 static void xen_do_write_msr(unsigned int msr, unsigned int low,
970 unsigned int high, int *err)
974 set_seg(SEGBASE_FS, low, high, err);
977 case MSR_KERNEL_GS_BASE:
978 set_seg(SEGBASE_GS_USER, low, high, err);
982 set_seg(SEGBASE_GS_KERNEL, low, high, err);
988 case MSR_SYSCALL_MASK:
989 case MSR_IA32_SYSENTER_CS:
990 case MSR_IA32_SYSENTER_ESP:
991 case MSR_IA32_SYSENTER_EIP:
992 /* Fast syscall setup is all done in hypercalls, so
993 these are all ignored. Stub them out here to stop
994 Xen console noise. */
998 if (!pmu_msr_write(msr, low, high, err)) {
1000 *err = native_write_msr_safe(msr, low, high);
1002 native_write_msr(msr, low, high);
1007 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1009 return xen_do_read_msr(msr, err);
1012 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1017 xen_do_write_msr(msr, low, high, &err);
1022 static u64 xen_read_msr(unsigned int msr)
1026 return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1029 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1033 xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1036 /* This is called once we have the cpu_possible_mask */
1037 void __init xen_setup_vcpu_info_placement(void)
1041 for_each_possible_cpu(cpu) {
1042 /* Set up direct vCPU id mapping for PV guests. */
1043 per_cpu(xen_vcpu_id, cpu) = cpu;
1044 xen_vcpu_setup(cpu);
1047 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1048 pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1049 pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1050 pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1053 static const struct pv_info xen_info __initconst = {
1054 .extra_user_64bit_cs = FLAT_USER_CS64,
1058 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1062 .set_debugreg = xen_set_debugreg,
1063 .get_debugreg = xen_get_debugreg,
1065 .read_cr0 = xen_read_cr0,
1066 .write_cr0 = xen_write_cr0,
1068 .write_cr4 = xen_write_cr4,
1070 .wbinvd = native_wbinvd,
1072 .read_msr = xen_read_msr,
1073 .write_msr = xen_write_msr,
1075 .read_msr_safe = xen_read_msr_safe,
1076 .write_msr_safe = xen_write_msr_safe,
1078 .read_pmc = xen_read_pmc,
1080 .load_tr_desc = paravirt_nop,
1081 .set_ldt = xen_set_ldt,
1082 .load_gdt = xen_load_gdt,
1083 .load_idt = xen_load_idt,
1084 .load_tls = xen_load_tls,
1085 .load_gs_index = xen_load_gs_index,
1087 .alloc_ldt = xen_alloc_ldt,
1088 .free_ldt = xen_free_ldt,
1090 .store_tr = xen_store_tr,
1092 .write_ldt_entry = xen_write_ldt_entry,
1093 .write_gdt_entry = xen_write_gdt_entry,
1094 .write_idt_entry = xen_write_idt_entry,
1095 .load_sp0 = xen_load_sp0,
1097 #ifdef CONFIG_X86_IOPL_IOPERM
1098 .invalidate_io_bitmap = xen_invalidate_io_bitmap,
1099 .update_io_bitmap = xen_update_io_bitmap,
1101 .io_delay = xen_io_delay,
1103 .start_context_switch = paravirt_start_context_switch,
1104 .end_context_switch = xen_end_context_switch,
1108 static void xen_restart(char *msg)
1110 xen_reboot(SHUTDOWN_reboot);
1113 static void xen_machine_halt(void)
1115 xen_reboot(SHUTDOWN_poweroff);
1118 static void xen_machine_power_off(void)
1120 do_kernel_power_off();
1121 xen_reboot(SHUTDOWN_poweroff);
1124 static void xen_crash_shutdown(struct pt_regs *regs)
1126 xen_reboot(SHUTDOWN_crash);
1129 static const struct machine_ops xen_machine_ops __initconst = {
1130 .restart = xen_restart,
1131 .halt = xen_machine_halt,
1132 .power_off = xen_machine_power_off,
1133 .shutdown = xen_machine_halt,
1134 .crash_shutdown = xen_crash_shutdown,
1135 .emergency_restart = xen_emergency_restart,
1138 static unsigned char xen_get_nmi_reason(void)
1140 unsigned char reason = 0;
1142 /* Construct a value which looks like it came from port 0x61. */
1143 if (test_bit(_XEN_NMIREASON_io_error,
1144 &HYPERVISOR_shared_info->arch.nmi_reason))
1145 reason |= NMI_REASON_IOCHK;
1146 if (test_bit(_XEN_NMIREASON_pci_serr,
1147 &HYPERVISOR_shared_info->arch.nmi_reason))
1148 reason |= NMI_REASON_SERR;
1153 static void __init xen_boot_params_init_edd(void)
1155 #if IS_ENABLED(CONFIG_EDD)
1156 struct xen_platform_op op;
1157 struct edd_info *edd_info;
1162 edd_info = boot_params.eddbuf;
1163 mbr_signature = boot_params.edd_mbr_sig_buffer;
1165 op.cmd = XENPF_firmware_info;
1167 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1168 for (nr = 0; nr < EDDMAXNR; nr++) {
1169 struct edd_info *info = edd_info + nr;
1171 op.u.firmware_info.index = nr;
1172 info->params.length = sizeof(info->params);
1173 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1175 ret = HYPERVISOR_platform_op(&op);
1179 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1182 C(interface_support);
1183 C(legacy_max_cylinder);
1185 C(legacy_sectors_per_track);
1188 boot_params.eddbuf_entries = nr;
1190 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1191 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1192 op.u.firmware_info.index = nr;
1193 ret = HYPERVISOR_platform_op(&op);
1196 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1198 boot_params.edd_mbr_sig_buf_entries = nr;
1203 * Set up the GDT and segment registers for -fstack-protector. Until
1204 * we do this, we have to be careful not to call any stack-protected
1205 * function, which is most of the kernel.
1207 static void __init xen_setup_gdt(int cpu)
1209 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1210 pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1212 switch_to_new_gdt(cpu);
1214 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1215 pv_ops.cpu.load_gdt = xen_load_gdt;
1218 static void __init xen_dom0_set_legacy_features(void)
1220 x86_platform.legacy.rtc = 1;
1223 static void __init xen_domu_set_legacy_features(void)
1225 x86_platform.legacy.rtc = 0;
1228 extern void early_xen_iret_patch(void);
1230 /* First C function to be called on Xen boot */
1231 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1233 struct physdev_set_iopl set_iopl;
1234 unsigned long initrd_start = 0;
1242 xen_start_info = si;
1244 __text_gen_insn(&early_xen_iret_patch,
1245 JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1248 xen_domain_type = XEN_PV_DOMAIN;
1249 xen_start_flags = xen_start_info->flags;
1251 xen_setup_features();
1253 /* Install Xen paravirt ops */
1255 pv_ops.cpu = xen_cpu_ops.cpu;
1259 * Setup xen_vcpu early because it is needed for
1260 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1262 * Don't do the full vcpu_info placement stuff until we have
1263 * the cpu_possible_mask and a non-dummy shared_info.
1265 xen_vcpu_info_reset(0);
1267 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1269 x86_init.resources.memory_setup = xen_memory_setup;
1270 x86_init.irqs.intr_mode_select = x86_init_noop;
1271 x86_init.irqs.intr_mode_init = x86_init_noop;
1272 x86_init.oem.arch_setup = xen_arch_setup;
1273 x86_init.oem.banner = xen_banner;
1274 x86_init.hyper.init_platform = xen_pv_init_platform;
1275 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1278 * Set up some pagetable state before starting to set any ptes.
1281 xen_setup_machphys_mapping();
1284 /* Prevent unwanted bits from being set in PTEs. */
1285 __supported_pte_mask &= ~_PAGE_GLOBAL;
1286 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1289 xen_build_dynamic_phys_to_machine();
1291 /* Work out if we support NX */
1292 get_cpu_cap(&boot_cpu_data);
1296 * Set up kernel GDT and segment registers, mainly so that
1297 * -fstack-protector code can be executed.
1301 /* Determine virtual and physical address sizes */
1302 get_cpu_address_sizes(&boot_cpu_data);
1304 /* Let's presume PV guests always boot on vCPU with id 0. */
1305 per_cpu(xen_vcpu_id, 0) = 0;
1307 idt_setup_early_handler();
1309 xen_init_capabilities();
1311 #ifdef CONFIG_X86_LOCAL_APIC
1313 * set up the basic apic ops.
1318 machine_ops = xen_machine_ops;
1321 * The only reliable way to retain the initial address of the
1322 * percpu gdt_page is to remember it here, so we can go and
1323 * mark it RW later, when the initial percpu area is freed.
1325 xen_initial_gdt = &per_cpu(gdt_page, 0);
1329 #ifdef CONFIG_ACPI_NUMA
1331 * The pages we from Xen are not related to machine pages, so
1332 * any NUMA information the kernel tries to get from ACPI will
1333 * be meaningless. Prevent it from trying.
1337 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1339 local_irq_disable();
1340 early_boot_irqs_disabled = true;
1342 xen_raw_console_write("mapping kernel into physical memory\n");
1343 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1344 xen_start_info->nr_pages);
1345 xen_reserve_special_pages();
1348 * We used to do this in xen_arch_setup, but that is too late
1349 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1350 * early_amd_init which pokes 0xcf8 port.
1353 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1355 xen_raw_printk("physdev_op failed %d\n", rc);
1358 if (xen_start_info->mod_start) {
1359 if (xen_start_info->flags & SIF_MOD_START_PFN)
1360 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1362 initrd_start = __pa(xen_start_info->mod_start);
1365 /* Poke various useful things into boot_params */
1366 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1367 boot_params.hdr.ramdisk_image = initrd_start;
1368 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1369 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1370 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1372 if (!xen_initial_domain()) {
1374 x86_init.pci.arch_init = pci_xen_init;
1375 x86_platform.set_legacy_features =
1376 xen_domu_set_legacy_features;
1378 const struct dom0_vga_console_info *info =
1379 (void *)((char *)xen_start_info +
1380 xen_start_info->console.dom0.info_off);
1381 struct xen_platform_op op = {
1382 .cmd = XENPF_firmware_info,
1383 .interface_version = XENPF_INTERFACE_VERSION,
1384 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1387 x86_platform.set_legacy_features =
1388 xen_dom0_set_legacy_features;
1389 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1390 xen_start_info->console.domU.mfn = 0;
1391 xen_start_info->console.domU.evtchn = 0;
1393 if (HYPERVISOR_platform_op(&op) == 0)
1394 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1396 /* Make sure ACS will be enabled */
1399 xen_acpi_sleep_register();
1401 xen_boot_params_init_edd();
1405 * Disable selecting "Firmware First mode" for correctable
1406 * memory errors, as this is the duty of the hypervisor to
1409 acpi_disable_cmcff = 1;
1413 xen_add_preferred_consoles();
1416 /* PCI BIOS service won't work from a PV guest. */
1417 pci_probe &= ~PCI_PROBE_BIOS;
1419 xen_raw_console_write("about to get started...\n");
1421 /* We need this for printk timestamps */
1422 xen_setup_runstate_info(0);
1424 xen_efi_init(&boot_params);
1426 /* Start the world */
1427 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1428 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1431 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1435 if (per_cpu(xen_vcpu, cpu) == NULL)
1438 xen_setup_timer(cpu);
1440 rc = xen_smp_intr_init(cpu);
1442 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1447 rc = xen_smp_intr_init_pv(cpu);
1449 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1457 static int xen_cpu_dead_pv(unsigned int cpu)
1459 xen_smp_intr_free(cpu);
1460 xen_smp_intr_free_pv(cpu);
1462 xen_teardown_timer(cpu);
1467 static uint32_t __init xen_platform_pv(void)
1469 if (xen_pv_domain())
1470 return xen_cpuid_base();
1475 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1477 .detect = xen_platform_pv,
1478 .type = X86_HYPER_XEN_PV,
1479 .runtime.pin_vcpu = xen_pin_vcpu,
1480 .ignore_nopv = true,