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>
36 #include <linux/stackprotector.h>
39 #include <xen/events.h>
40 #include <xen/interface/xen.h>
41 #include <xen/interface/version.h>
42 #include <xen/interface/physdev.h>
43 #include <xen/interface/vcpu.h>
44 #include <xen/interface/memory.h>
45 #include <xen/interface/nmi.h>
46 #include <xen/interface/xen-mca.h>
47 #include <xen/features.h>
49 #include <xen/hvc-console.h>
52 #include <asm/paravirt.h>
55 #include <asm/xen/pci.h>
56 #include <asm/xen/hypercall.h>
57 #include <asm/xen/hypervisor.h>
58 #include <asm/xen/cpuid.h>
59 #include <asm/fixmap.h>
60 #include <asm/processor.h>
61 #include <asm/proto.h>
62 #include <asm/msr-index.h>
63 #include <asm/traps.h>
64 #include <asm/setup.h>
66 #include <asm/pgalloc.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/hypervisor.h>
70 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
75 #ifdef CONFIG_X86_IOPL_IOPERM
76 #include <asm/io_bitmap.h>
80 #include <linux/acpi.h>
82 #include <acpi/proc_cap_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
90 #include "multicalls.h"
93 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
95 void *xen_initial_gdt;
97 static int xen_cpu_up_prepare_pv(unsigned int cpu);
98 static int xen_cpu_dead_pv(unsigned int cpu);
101 struct desc_struct desc[3];
104 DEFINE_PER_CPU(enum xen_lazy_mode, xen_lazy_mode) = XEN_LAZY_NONE;
105 DEFINE_PER_CPU(unsigned int, xen_lazy_nesting);
107 enum xen_lazy_mode xen_get_lazy_mode(void)
110 return XEN_LAZY_NONE;
112 return this_cpu_read(xen_lazy_mode);
116 * Updating the 3 TLS descriptors in the GDT on every task switch is
117 * surprisingly expensive so we avoid updating them if they haven't
118 * changed. Since Xen writes different descriptors than the one
119 * passed in the update_descriptor hypercall we keep shadow copies to
122 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
124 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
126 static int __init parse_xen_msr_safe(char *str)
129 return kstrtobool(str, &xen_msr_safe);
132 early_param("xen_msr_safe", parse_xen_msr_safe);
134 /* Get MTRR settings from Xen and put them into mtrr_state. */
135 static void __init xen_set_mtrr_data(void)
138 struct xen_platform_op op = {
139 .cmd = XENPF_read_memtype,
140 .interface_version = XENPF_INTERFACE_VERSION,
145 static struct mtrr_var_range var[MTRR_MAX_VAR_RANGES] __initdata;
147 /* Get physical address width (only 64-bit cpus supported). */
149 eax = cpuid_eax(0x80000000);
150 if ((eax >> 16) == 0x8000 && eax >= 0x80000008) {
151 eax = cpuid_eax(0x80000008);
155 for (reg = 0; reg < MTRR_MAX_VAR_RANGES; reg++) {
156 op.u.read_memtype.reg = reg;
157 if (HYPERVISOR_platform_op(&op))
161 * Only called in dom0, which has all RAM PFNs mapped at
162 * RAM MFNs, and all PCI space etc. is identity mapped.
163 * This means we can treat MFN == PFN regarding MTRR settings.
165 var[reg].base_lo = op.u.read_memtype.type;
166 var[reg].base_lo |= op.u.read_memtype.mfn << PAGE_SHIFT;
167 var[reg].base_hi = op.u.read_memtype.mfn >> (32 - PAGE_SHIFT);
168 mask = ~((op.u.read_memtype.nr_mfns << PAGE_SHIFT) - 1);
169 mask &= (1UL << width) - 1;
171 mask |= MTRR_PHYSMASK_V;
172 var[reg].mask_lo = mask;
173 var[reg].mask_hi = mask >> 32;
176 /* Only overwrite MTRR state if any MTRR could be got from Xen. */
178 mtrr_overwrite_state(var, reg, MTRR_TYPE_UNCACHABLE);
182 static void __init xen_pv_init_platform(void)
184 /* PV guests can't operate virtio devices without grants. */
185 if (IS_ENABLED(CONFIG_XEN_VIRTIO))
186 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
188 populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
190 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
191 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
193 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
194 xen_vcpu_info_reset(0);
196 /* pvclock is in shared info area */
199 if (xen_initial_domain())
202 mtrr_overwrite_state(NULL, 0, MTRR_TYPE_WRBACK);
205 static void __init xen_pv_guest_late_init(void)
208 /* Setup shared vcpu info for non-smp configurations */
209 xen_setup_vcpu_info_placement();
213 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
214 static __read_mostly unsigned int cpuid_leaf5_edx_val;
216 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
217 unsigned int *cx, unsigned int *dx)
219 unsigned maskebx = ~0;
222 * Mask out inconvenient features, to try and disable as many
223 * unsupported kernel subsystems as possible.
226 case CPUID_MWAIT_LEAF:
227 /* Synthesize the values.. */
230 *cx = cpuid_leaf5_ecx_val;
231 *dx = cpuid_leaf5_edx_val;
235 /* Suppress extended topology stuff */
240 asm(XEN_EMULATE_PREFIX "cpuid"
245 : "0" (*ax), "2" (*cx));
250 static bool __init xen_check_mwait(void)
253 struct xen_platform_op op = {
254 .cmd = XENPF_set_processor_pminfo,
255 .u.set_pminfo.id = -1,
256 .u.set_pminfo.type = XEN_PM_PDC,
259 unsigned int ax, bx, cx, dx;
260 unsigned int mwait_mask;
262 /* We need to determine whether it is OK to expose the MWAIT
263 * capability to the kernel to harvest deeper than C3 states from ACPI
264 * _CST using the processor_harvest_xen.c module. For this to work, we
265 * need to gather the MWAIT_LEAF values (which the cstate.c code
266 * checks against). The hypervisor won't expose the MWAIT flag because
267 * it would break backwards compatibility; so we will find out directly
268 * from the hardware and hypercall.
270 if (!xen_initial_domain())
274 * When running under platform earlier than Xen4.2, do not expose
275 * mwait, to avoid the risk of loading native acpi pad driver
277 if (!xen_running_on_version_or_later(4, 2))
283 native_cpuid(&ax, &bx, &cx, &dx);
285 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
286 (1 << (X86_FEATURE_MWAIT % 32));
288 if ((cx & mwait_mask) != mwait_mask)
291 /* We need to emulate the MWAIT_LEAF and for that we need both
292 * ecx and edx. The hypercall provides only partial information.
295 ax = CPUID_MWAIT_LEAF;
300 native_cpuid(&ax, &bx, &cx, &dx);
302 /* Ask the Hypervisor whether to clear ACPI_PROC_CAP_C_C2C3_FFH. If so,
303 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
305 buf[0] = ACPI_PDC_REVISION_ID;
307 buf[2] = (ACPI_PROC_CAP_C_CAPABILITY_SMP | ACPI_PROC_CAP_EST_CAPABILITY_SWSMP);
309 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
311 if ((HYPERVISOR_platform_op(&op) == 0) &&
312 (buf[2] & (ACPI_PROC_CAP_C_C1_FFH | ACPI_PROC_CAP_C_C2C3_FFH))) {
313 cpuid_leaf5_ecx_val = cx;
314 cpuid_leaf5_edx_val = dx;
322 static bool __init xen_check_xsave(void)
324 unsigned int cx, xsave_mask;
328 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
329 (1 << (X86_FEATURE_OSXSAVE % 32));
331 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
332 return (cx & xsave_mask) == xsave_mask;
335 static void __init xen_init_capabilities(void)
337 setup_force_cpu_cap(X86_FEATURE_XENPV);
338 setup_clear_cpu_cap(X86_FEATURE_DCA);
339 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
340 setup_clear_cpu_cap(X86_FEATURE_MTRR);
341 setup_clear_cpu_cap(X86_FEATURE_ACC);
342 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
343 setup_clear_cpu_cap(X86_FEATURE_SME);
344 setup_clear_cpu_cap(X86_FEATURE_LKGS);
347 * Xen PV would need some work to support PCID: CR3 handling as well
348 * as xen_flush_tlb_others() would need updating.
350 setup_clear_cpu_cap(X86_FEATURE_PCID);
352 if (!xen_initial_domain())
353 setup_clear_cpu_cap(X86_FEATURE_ACPI);
355 if (xen_check_mwait())
356 setup_force_cpu_cap(X86_FEATURE_MWAIT);
358 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
360 if (!xen_check_xsave()) {
361 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
362 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
366 static noinstr void xen_set_debugreg(int reg, unsigned long val)
368 HYPERVISOR_set_debugreg(reg, val);
371 static noinstr unsigned long xen_get_debugreg(int reg)
373 return HYPERVISOR_get_debugreg(reg);
376 static void xen_start_context_switch(struct task_struct *prev)
378 BUG_ON(preemptible());
380 if (this_cpu_read(xen_lazy_mode) == XEN_LAZY_MMU) {
381 arch_leave_lazy_mmu_mode();
382 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
384 enter_lazy(XEN_LAZY_CPU);
387 static void xen_end_context_switch(struct task_struct *next)
389 BUG_ON(preemptible());
392 leave_lazy(XEN_LAZY_CPU);
393 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
394 arch_enter_lazy_mmu_mode();
397 static unsigned long xen_store_tr(void)
403 * Set the page permissions for a particular virtual address. If the
404 * address is a vmalloc mapping (or other non-linear mapping), then
405 * find the linear mapping of the page and also set its protections to
408 static void set_aliased_prot(void *v, pgprot_t prot)
417 ptep = lookup_address((unsigned long)v, &level);
418 BUG_ON(ptep == NULL);
420 pfn = pte_pfn(*ptep);
421 pte = pfn_pte(pfn, prot);
424 * Careful: update_va_mapping() will fail if the virtual address
425 * we're poking isn't populated in the page tables. We don't
426 * need to worry about the direct map (that's always in the page
427 * tables), but we need to be careful about vmap space. In
428 * particular, the top level page table can lazily propagate
429 * entries between processes, so if we've switched mms since we
430 * vmapped the target in the first place, we might not have the
431 * top-level page table entry populated.
433 * We disable preemption because we want the same mm active when
434 * we probe the target and when we issue the hypercall. We'll
435 * have the same nominal mm, but if we're a kernel thread, lazy
436 * mm dropping could change our pgd.
438 * Out of an abundance of caution, this uses __get_user() to fault
439 * in the target address just in case there's some obscure case
440 * in which the target address isn't readable.
445 copy_from_kernel_nofault(&dummy, v, 1);
447 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
450 va = __va(PFN_PHYS(pfn));
452 if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
458 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
460 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
464 * We need to mark the all aliases of the LDT pages RO. We
465 * don't need to call vm_flush_aliases(), though, since that's
466 * only responsible for flushing aliases out the TLBs, not the
467 * page tables, and Xen will flush the TLB for us if needed.
469 * To avoid confusing future readers: none of this is necessary
470 * to load the LDT. The hypervisor only checks this when the
471 * LDT is faulted in due to subsequent descriptor access.
474 for (i = 0; i < entries; i += entries_per_page)
475 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
478 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
480 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
483 for (i = 0; i < entries; i += entries_per_page)
484 set_aliased_prot(ldt + i, PAGE_KERNEL);
487 static void xen_set_ldt(const void *addr, unsigned entries)
489 struct mmuext_op *op;
490 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
492 trace_xen_cpu_set_ldt(addr, entries);
495 op->cmd = MMUEXT_SET_LDT;
496 op->arg1.linear_addr = (unsigned long)addr;
497 op->arg2.nr_ents = entries;
499 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
501 xen_mc_issue(XEN_LAZY_CPU);
504 static void xen_load_gdt(const struct desc_ptr *dtr)
506 unsigned long va = dtr->address;
507 unsigned int size = dtr->size + 1;
508 unsigned long pfn, mfn;
513 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
514 BUG_ON(size > PAGE_SIZE);
515 BUG_ON(va & ~PAGE_MASK);
518 * The GDT is per-cpu and is in the percpu data area.
519 * That can be virtually mapped, so we need to do a
520 * page-walk to get the underlying MFN for the
521 * hypercall. The page can also be in the kernel's
522 * linear range, so we need to RO that mapping too.
524 ptep = lookup_address(va, &level);
525 BUG_ON(ptep == NULL);
527 pfn = pte_pfn(*ptep);
528 mfn = pfn_to_mfn(pfn);
529 virt = __va(PFN_PHYS(pfn));
531 make_lowmem_page_readonly((void *)va);
532 make_lowmem_page_readonly(virt);
534 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
539 * load_gdt for early boot, when the gdt is only mapped once
541 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
543 unsigned long va = dtr->address;
544 unsigned int size = dtr->size + 1;
545 unsigned long pfn, mfn;
548 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
549 BUG_ON(size > PAGE_SIZE);
550 BUG_ON(va & ~PAGE_MASK);
552 pfn = virt_to_pfn((void *)va);
553 mfn = pfn_to_mfn(pfn);
555 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
557 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
560 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
564 static inline bool desc_equal(const struct desc_struct *d1,
565 const struct desc_struct *d2)
567 return !memcmp(d1, d2, sizeof(*d1));
570 static void load_TLS_descriptor(struct thread_struct *t,
571 unsigned int cpu, unsigned int i)
573 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
574 struct desc_struct *gdt;
576 struct multicall_space mc;
578 if (desc_equal(shadow, &t->tls_array[i]))
581 *shadow = t->tls_array[i];
583 gdt = get_cpu_gdt_rw(cpu);
584 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
585 mc = __xen_mc_entry(0);
587 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
590 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
593 * In lazy mode we need to zero %fs, otherwise we may get an
594 * exception between the new %fs descriptor being loaded and
595 * %fs being effectively cleared at __switch_to().
597 if (xen_get_lazy_mode() == XEN_LAZY_CPU)
602 load_TLS_descriptor(t, cpu, 0);
603 load_TLS_descriptor(t, cpu, 1);
604 load_TLS_descriptor(t, cpu, 2);
606 xen_mc_issue(XEN_LAZY_CPU);
609 static void xen_load_gs_index(unsigned int idx)
611 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
615 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
618 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
619 u64 entry = *(u64 *)ptr;
621 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
626 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
632 void noist_exc_debug(struct pt_regs *regs);
634 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
636 /* On Xen PV, NMI doesn't use IST. The C part is the same as native. */
640 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
642 /* On Xen PV, DF doesn't use IST. The C part is the same as native. */
643 exc_double_fault(regs, error_code);
646 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
649 * There's no IST on Xen PV, but we still need to dispatch
650 * to the correct handler.
653 noist_exc_debug(regs);
658 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
660 /* This should never happen and there is no way to handle it. */
661 instrumentation_begin();
662 pr_err("Unknown trap in Xen PV mode.");
664 instrumentation_end();
667 #ifdef CONFIG_X86_MCE
668 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
671 * There's no IST on Xen PV, but we still need to dispatch
672 * to the correct handler.
675 noist_exc_machine_check(regs);
677 exc_machine_check(regs);
681 struct trap_array_entry {
687 #define TRAP_ENTRY(func, ist_ok) { \
688 .orig = asm_##func, \
689 .xen = xen_asm_##func, \
692 #define TRAP_ENTRY_REDIR(func, ist_ok) { \
693 .orig = asm_##func, \
694 .xen = xen_asm_xenpv_##func, \
697 static struct trap_array_entry trap_array[] = {
698 TRAP_ENTRY_REDIR(exc_debug, true ),
699 TRAP_ENTRY_REDIR(exc_double_fault, true ),
700 #ifdef CONFIG_X86_MCE
701 TRAP_ENTRY_REDIR(exc_machine_check, true ),
703 TRAP_ENTRY_REDIR(exc_nmi, true ),
704 TRAP_ENTRY(exc_int3, false ),
705 TRAP_ENTRY(exc_overflow, false ),
706 #ifdef CONFIG_IA32_EMULATION
707 { entry_INT80_compat, xen_entry_INT80_compat, false },
709 TRAP_ENTRY(exc_page_fault, false ),
710 TRAP_ENTRY(exc_divide_error, false ),
711 TRAP_ENTRY(exc_bounds, false ),
712 TRAP_ENTRY(exc_invalid_op, false ),
713 TRAP_ENTRY(exc_device_not_available, false ),
714 TRAP_ENTRY(exc_coproc_segment_overrun, false ),
715 TRAP_ENTRY(exc_invalid_tss, false ),
716 TRAP_ENTRY(exc_segment_not_present, false ),
717 TRAP_ENTRY(exc_stack_segment, false ),
718 TRAP_ENTRY(exc_general_protection, false ),
719 TRAP_ENTRY(exc_spurious_interrupt_bug, false ),
720 TRAP_ENTRY(exc_coprocessor_error, false ),
721 TRAP_ENTRY(exc_alignment_check, false ),
722 TRAP_ENTRY(exc_simd_coprocessor_error, false ),
723 #ifdef CONFIG_X86_CET
724 TRAP_ENTRY(exc_control_protection, false ),
728 static bool __ref get_trap_addr(void **addr, unsigned int ist)
731 bool ist_okay = false;
735 * Replace trap handler addresses by Xen specific ones.
736 * Check for known traps using IST and whitelist them.
737 * The debugger ones are the only ones we care about.
738 * Xen will handle faults like double_fault, so we should never see
739 * them. Warn if there's an unexpected IST-using fault handler.
741 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
742 struct trap_array_entry *entry = trap_array + nr;
744 if (*addr == entry->orig) {
746 ist_okay = entry->ist_okay;
752 if (nr == ARRAY_SIZE(trap_array) &&
753 *addr >= (void *)early_idt_handler_array[0] &&
754 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
755 nr = (*addr - (void *)early_idt_handler_array[0]) /
756 EARLY_IDT_HANDLER_SIZE;
757 *addr = (void *)xen_early_idt_handler_array[nr];
762 *addr = (void *)xen_asm_exc_xen_unknown_trap;
764 if (WARN_ON(found && ist != 0 && !ist_okay))
770 static int cvt_gate_to_trap(int vector, const gate_desc *val,
771 struct trap_info *info)
775 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
778 info->vector = vector;
780 addr = gate_offset(val);
781 if (!get_trap_addr((void **)&addr, val->bits.ist))
783 info->address = addr;
785 info->cs = gate_segment(val);
786 info->flags = val->bits.dpl;
787 /* interrupt gates clear IF */
788 if (val->bits.type == GATE_INTERRUPT)
789 info->flags |= 1 << 2;
794 /* Locations of each CPU's IDT */
795 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
797 /* Set an IDT entry. If the entry is part of the current IDT, then
799 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
801 unsigned long p = (unsigned long)&dt[entrynum];
802 unsigned long start, end;
804 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
808 start = __this_cpu_read(idt_desc.address);
809 end = start + __this_cpu_read(idt_desc.size) + 1;
813 native_write_idt_entry(dt, entrynum, g);
815 if (p >= start && (p + 8) <= end) {
816 struct trap_info info[2];
820 if (cvt_gate_to_trap(entrynum, g, &info[0]))
821 if (HYPERVISOR_set_trap_table(info))
828 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
829 struct trap_info *traps, bool full)
831 unsigned in, out, count;
833 count = (desc->size+1) / sizeof(gate_desc);
836 for (in = out = 0; in < count; in++) {
837 gate_desc *entry = (gate_desc *)(desc->address) + in;
839 if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
846 void xen_copy_trap_info(struct trap_info *traps)
848 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
850 xen_convert_trap_info(desc, traps, true);
853 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
854 hold a spinlock to protect the static traps[] array (static because
855 it avoids allocation, and saves stack space). */
856 static void xen_load_idt(const struct desc_ptr *desc)
858 static DEFINE_SPINLOCK(lock);
859 static struct trap_info traps[257];
860 static const struct trap_info zero = { };
863 trace_xen_cpu_load_idt(desc);
867 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
869 out = xen_convert_trap_info(desc, traps, false);
873 if (HYPERVISOR_set_trap_table(traps))
879 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
880 they're handled differently. */
881 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
882 const void *desc, int type)
884 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
895 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
898 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
908 * Version of write_gdt_entry for use at early boot-time needed to
909 * update an entry as simply as possible.
911 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
912 const void *desc, int type)
914 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
923 xmaddr_t maddr = virt_to_machine(&dt[entry]);
925 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
926 dt[entry] = *(struct desc_struct *)desc;
932 static void xen_load_sp0(unsigned long sp0)
934 struct multicall_space mcs;
936 mcs = xen_mc_entry(0);
937 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
938 xen_mc_issue(XEN_LAZY_CPU);
939 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
942 #ifdef CONFIG_X86_IOPL_IOPERM
943 static void xen_invalidate_io_bitmap(void)
945 struct physdev_set_iobitmap iobitmap = {
950 native_tss_invalidate_io_bitmap();
951 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
954 static void xen_update_io_bitmap(void)
956 struct physdev_set_iobitmap iobitmap;
957 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
959 native_tss_update_io_bitmap();
961 iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
962 tss->x86_tss.io_bitmap_base;
963 if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
964 iobitmap.nr_ports = 0;
966 iobitmap.nr_ports = IO_BITMAP_BITS;
968 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
972 static void xen_io_delay(void)
976 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
978 static unsigned long xen_read_cr0(void)
980 unsigned long cr0 = this_cpu_read(xen_cr0_value);
982 if (unlikely(cr0 == 0)) {
983 cr0 = native_read_cr0();
984 this_cpu_write(xen_cr0_value, cr0);
990 static void xen_write_cr0(unsigned long cr0)
992 struct multicall_space mcs;
994 this_cpu_write(xen_cr0_value, cr0);
996 /* Only pay attention to cr0.TS; everything else is
998 mcs = xen_mc_entry(0);
1000 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1002 xen_mc_issue(XEN_LAZY_CPU);
1005 static void xen_write_cr4(unsigned long cr4)
1007 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1009 native_write_cr4(cr4);
1012 static u64 xen_do_read_msr(unsigned int msr, int *err)
1014 u64 val = 0; /* Avoid uninitialized value for safe variant. */
1016 if (pmu_msr_read(msr, &val, err))
1020 val = native_read_msr_safe(msr, err);
1022 val = native_read_msr(msr);
1025 case MSR_IA32_APICBASE:
1026 val &= ~X2APIC_ENABLE;
1032 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
1035 u64 base = ((u64)high << 32) | low;
1037 if (HYPERVISOR_set_segment_base(which, base) == 0)
1043 WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
1047 * Support write_msr_safe() and write_msr() semantics.
1048 * With err == NULL write_msr() semantics are selected.
1049 * Supplying an err pointer requires err to be pre-initialized with 0.
1051 static void xen_do_write_msr(unsigned int msr, unsigned int low,
1052 unsigned int high, int *err)
1056 set_seg(SEGBASE_FS, low, high, err);
1059 case MSR_KERNEL_GS_BASE:
1060 set_seg(SEGBASE_GS_USER, low, high, err);
1064 set_seg(SEGBASE_GS_KERNEL, low, high, err);
1070 case MSR_SYSCALL_MASK:
1071 case MSR_IA32_SYSENTER_CS:
1072 case MSR_IA32_SYSENTER_ESP:
1073 case MSR_IA32_SYSENTER_EIP:
1074 /* Fast syscall setup is all done in hypercalls, so
1075 these are all ignored. Stub them out here to stop
1076 Xen console noise. */
1080 if (!pmu_msr_write(msr, low, high, err)) {
1082 *err = native_write_msr_safe(msr, low, high);
1084 native_write_msr(msr, low, high);
1089 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1091 return xen_do_read_msr(msr, err);
1094 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1099 xen_do_write_msr(msr, low, high, &err);
1104 static u64 xen_read_msr(unsigned int msr)
1108 return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1111 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1115 xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1118 /* This is called once we have the cpu_possible_mask */
1119 void __init xen_setup_vcpu_info_placement(void)
1123 for_each_possible_cpu(cpu) {
1124 /* Set up direct vCPU id mapping for PV guests. */
1125 per_cpu(xen_vcpu_id, cpu) = cpu;
1126 xen_vcpu_setup(cpu);
1129 pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1130 pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1131 pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1132 pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1135 static const struct pv_info xen_info __initconst = {
1136 .extra_user_64bit_cs = FLAT_USER_CS64,
1140 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1144 .set_debugreg = xen_set_debugreg,
1145 .get_debugreg = xen_get_debugreg,
1147 .read_cr0 = xen_read_cr0,
1148 .write_cr0 = xen_write_cr0,
1150 .write_cr4 = xen_write_cr4,
1152 .wbinvd = pv_native_wbinvd,
1154 .read_msr = xen_read_msr,
1155 .write_msr = xen_write_msr,
1157 .read_msr_safe = xen_read_msr_safe,
1158 .write_msr_safe = xen_write_msr_safe,
1160 .read_pmc = xen_read_pmc,
1162 .load_tr_desc = paravirt_nop,
1163 .set_ldt = xen_set_ldt,
1164 .load_gdt = xen_load_gdt,
1165 .load_idt = xen_load_idt,
1166 .load_tls = xen_load_tls,
1167 .load_gs_index = xen_load_gs_index,
1169 .alloc_ldt = xen_alloc_ldt,
1170 .free_ldt = xen_free_ldt,
1172 .store_tr = xen_store_tr,
1174 .write_ldt_entry = xen_write_ldt_entry,
1175 .write_gdt_entry = xen_write_gdt_entry,
1176 .write_idt_entry = xen_write_idt_entry,
1177 .load_sp0 = xen_load_sp0,
1179 #ifdef CONFIG_X86_IOPL_IOPERM
1180 .invalidate_io_bitmap = xen_invalidate_io_bitmap,
1181 .update_io_bitmap = xen_update_io_bitmap,
1183 .io_delay = xen_io_delay,
1185 .start_context_switch = xen_start_context_switch,
1186 .end_context_switch = xen_end_context_switch,
1190 static void xen_restart(char *msg)
1192 xen_reboot(SHUTDOWN_reboot);
1195 static void xen_machine_halt(void)
1197 xen_reboot(SHUTDOWN_poweroff);
1200 static void xen_machine_power_off(void)
1202 do_kernel_power_off();
1203 xen_reboot(SHUTDOWN_poweroff);
1206 static void xen_crash_shutdown(struct pt_regs *regs)
1208 xen_reboot(SHUTDOWN_crash);
1211 static const struct machine_ops xen_machine_ops __initconst = {
1212 .restart = xen_restart,
1213 .halt = xen_machine_halt,
1214 .power_off = xen_machine_power_off,
1215 .shutdown = xen_machine_halt,
1216 .crash_shutdown = xen_crash_shutdown,
1217 .emergency_restart = xen_emergency_restart,
1220 static unsigned char xen_get_nmi_reason(void)
1222 unsigned char reason = 0;
1224 /* Construct a value which looks like it came from port 0x61. */
1225 if (test_bit(_XEN_NMIREASON_io_error,
1226 &HYPERVISOR_shared_info->arch.nmi_reason))
1227 reason |= NMI_REASON_IOCHK;
1228 if (test_bit(_XEN_NMIREASON_pci_serr,
1229 &HYPERVISOR_shared_info->arch.nmi_reason))
1230 reason |= NMI_REASON_SERR;
1235 static void __init xen_boot_params_init_edd(void)
1237 #if IS_ENABLED(CONFIG_EDD)
1238 struct xen_platform_op op;
1239 struct edd_info *edd_info;
1244 edd_info = boot_params.eddbuf;
1245 mbr_signature = boot_params.edd_mbr_sig_buffer;
1247 op.cmd = XENPF_firmware_info;
1249 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1250 for (nr = 0; nr < EDDMAXNR; nr++) {
1251 struct edd_info *info = edd_info + nr;
1253 op.u.firmware_info.index = nr;
1254 info->params.length = sizeof(info->params);
1255 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1257 ret = HYPERVISOR_platform_op(&op);
1261 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1264 C(interface_support);
1265 C(legacy_max_cylinder);
1267 C(legacy_sectors_per_track);
1270 boot_params.eddbuf_entries = nr;
1272 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1273 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1274 op.u.firmware_info.index = nr;
1275 ret = HYPERVISOR_platform_op(&op);
1278 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1280 boot_params.edd_mbr_sig_buf_entries = nr;
1285 * Set up the GDT and segment registers for -fstack-protector. Until
1286 * we do this, we have to be careful not to call any stack-protected
1287 * function, which is most of the kernel.
1289 static void __init xen_setup_gdt(int cpu)
1291 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1292 pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1294 switch_gdt_and_percpu_base(cpu);
1296 pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1297 pv_ops.cpu.load_gdt = xen_load_gdt;
1300 static void __init xen_dom0_set_legacy_features(void)
1302 x86_platform.legacy.rtc = 1;
1305 static void __init xen_domu_set_legacy_features(void)
1307 x86_platform.legacy.rtc = 0;
1310 extern void early_xen_iret_patch(void);
1312 /* First C function to be called on Xen boot */
1313 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1315 struct physdev_set_iopl set_iopl;
1316 unsigned long initrd_start = 0;
1324 xen_start_info = si;
1326 __text_gen_insn(&early_xen_iret_patch,
1327 JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1330 xen_domain_type = XEN_PV_DOMAIN;
1331 xen_start_flags = xen_start_info->flags;
1333 xen_setup_features();
1335 /* Install Xen paravirt ops */
1337 pv_ops.cpu = xen_cpu_ops.cpu;
1341 * Setup xen_vcpu early because it is needed for
1342 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1344 * Don't do the full vcpu_info placement stuff until we have
1345 * the cpu_possible_mask and a non-dummy shared_info.
1347 xen_vcpu_info_reset(0);
1349 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1350 x86_platform.realmode_reserve = x86_init_noop;
1351 x86_platform.realmode_init = x86_init_noop;
1353 x86_init.resources.memory_setup = xen_memory_setup;
1354 x86_init.irqs.intr_mode_select = x86_init_noop;
1355 x86_init.irqs.intr_mode_init = x86_64_probe_apic;
1356 x86_init.oem.arch_setup = xen_arch_setup;
1357 x86_init.oem.banner = xen_banner;
1358 x86_init.hyper.init_platform = xen_pv_init_platform;
1359 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1362 * Set up some pagetable state before starting to set any ptes.
1365 xen_setup_machphys_mapping();
1368 /* Prevent unwanted bits from being set in PTEs. */
1369 __supported_pte_mask &= ~_PAGE_GLOBAL;
1370 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1373 xen_build_dynamic_phys_to_machine();
1375 /* Work out if we support NX */
1376 get_cpu_cap(&boot_cpu_data);
1380 * Set up kernel GDT and segment registers, mainly so that
1381 * -fstack-protector code can be executed.
1385 /* Determine virtual and physical address sizes */
1386 get_cpu_address_sizes(&boot_cpu_data);
1388 /* Let's presume PV guests always boot on vCPU with id 0. */
1389 per_cpu(xen_vcpu_id, 0) = 0;
1391 idt_setup_early_handler();
1393 xen_init_capabilities();
1396 * set up the basic apic ops.
1400 machine_ops = xen_machine_ops;
1403 * The only reliable way to retain the initial address of the
1404 * percpu gdt_page is to remember it here, so we can go and
1405 * mark it RW later, when the initial percpu area is freed.
1407 xen_initial_gdt = &per_cpu(gdt_page, 0);
1411 #ifdef CONFIG_ACPI_NUMA
1413 * The pages we from Xen are not related to machine pages, so
1414 * any NUMA information the kernel tries to get from ACPI will
1415 * be meaningless. Prevent it from trying.
1419 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1421 local_irq_disable();
1422 early_boot_irqs_disabled = true;
1424 xen_raw_console_write("mapping kernel into physical memory\n");
1425 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1426 xen_start_info->nr_pages);
1427 xen_reserve_special_pages();
1430 * We used to do this in xen_arch_setup, but that is too late
1431 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1432 * early_amd_init which pokes 0xcf8 port.
1435 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1437 xen_raw_printk("physdev_op failed %d\n", rc);
1440 if (xen_start_info->mod_start) {
1441 if (xen_start_info->flags & SIF_MOD_START_PFN)
1442 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1444 initrd_start = __pa(xen_start_info->mod_start);
1447 /* Poke various useful things into boot_params */
1448 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1449 boot_params.hdr.ramdisk_image = initrd_start;
1450 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1451 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1452 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1454 if (!xen_initial_domain()) {
1456 x86_init.pci.arch_init = pci_xen_init;
1457 x86_platform.set_legacy_features =
1458 xen_domu_set_legacy_features;
1460 const struct dom0_vga_console_info *info =
1461 (void *)((char *)xen_start_info +
1462 xen_start_info->console.dom0.info_off);
1463 struct xen_platform_op op = {
1464 .cmd = XENPF_firmware_info,
1465 .interface_version = XENPF_INTERFACE_VERSION,
1466 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1469 x86_platform.set_legacy_features =
1470 xen_dom0_set_legacy_features;
1471 xen_init_vga(info, xen_start_info->console.dom0.info_size,
1472 &boot_params.screen_info);
1473 xen_start_info->console.domU.mfn = 0;
1474 xen_start_info->console.domU.evtchn = 0;
1476 if (HYPERVISOR_platform_op(&op) == 0)
1477 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1479 /* Make sure ACS will be enabled */
1482 xen_acpi_sleep_register();
1484 xen_boot_params_init_edd();
1488 * Disable selecting "Firmware First mode" for correctable
1489 * memory errors, as this is the duty of the hypervisor to
1492 acpi_disable_cmcff = 1;
1496 xen_add_preferred_consoles();
1499 /* PCI BIOS service won't work from a PV guest. */
1500 pci_probe &= ~PCI_PROBE_BIOS;
1502 xen_raw_console_write("about to get started...\n");
1504 /* We need this for printk timestamps */
1505 xen_setup_runstate_info(0);
1507 xen_efi_init(&boot_params);
1509 /* Start the world */
1510 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1511 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1514 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1518 if (per_cpu(xen_vcpu, cpu) == NULL)
1521 xen_setup_timer(cpu);
1523 rc = xen_smp_intr_init(cpu);
1525 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1530 rc = xen_smp_intr_init_pv(cpu);
1532 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1540 static int xen_cpu_dead_pv(unsigned int cpu)
1542 xen_smp_intr_free(cpu);
1543 xen_smp_intr_free_pv(cpu);
1545 xen_teardown_timer(cpu);
1550 static uint32_t __init xen_platform_pv(void)
1552 if (xen_pv_domain())
1553 return xen_cpuid_base();
1558 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1560 .detect = xen_platform_pv,
1561 .type = X86_HYPER_XEN_PV,
1562 .runtime.pin_vcpu = xen_pin_vcpu,
1563 .ignore_nopv = true,