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/bootmem.h>
27 #include <linux/export.h>
29 #include <linux/page-flags.h>
30 #include <linux/highmem.h>
31 #include <linux/console.h>
32 #include <linux/pci.h>
33 #include <linux/gfp.h>
34 #include <linux/memblock.h>
35 #include <linux/edd.h>
36 #include <linux/frame.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/pgtable.h>
68 #include <asm/tlbflush.h>
69 #include <asm/reboot.h>
70 #include <asm/stackprotector.h>
71 #include <asm/hypervisor.h>
72 #include <asm/mach_traps.h>
73 #include <asm/mwait.h>
74 #include <asm/pci_x86.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 void __init xen_banner(void)
113 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
114 struct xen_extraversion extra;
115 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
117 pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
118 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
119 version >> 16, version & 0xffff, extra.extraversion,
120 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
122 /* Check if running on Xen version (major, minor) or later */
124 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
126 unsigned int version;
131 version = HYPERVISOR_xen_version(XENVER_version, NULL);
132 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
133 ((version >> 16) > major))
138 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
139 static __read_mostly unsigned int cpuid_leaf5_edx_val;
141 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
142 unsigned int *cx, unsigned int *dx)
144 unsigned maskebx = ~0;
147 * Mask out inconvenient features, to try and disable as many
148 * unsupported kernel subsystems as possible.
151 case CPUID_MWAIT_LEAF:
152 /* Synthesize the values.. */
155 *cx = cpuid_leaf5_ecx_val;
156 *dx = cpuid_leaf5_edx_val;
160 /* Suppress extended topology stuff */
165 asm(XEN_EMULATE_PREFIX "cpuid"
170 : "0" (*ax), "2" (*cx));
174 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
176 static bool __init xen_check_mwait(void)
179 struct xen_platform_op op = {
180 .cmd = XENPF_set_processor_pminfo,
181 .u.set_pminfo.id = -1,
182 .u.set_pminfo.type = XEN_PM_PDC,
185 unsigned int ax, bx, cx, dx;
186 unsigned int mwait_mask;
188 /* We need to determine whether it is OK to expose the MWAIT
189 * capability to the kernel to harvest deeper than C3 states from ACPI
190 * _CST using the processor_harvest_xen.c module. For this to work, we
191 * need to gather the MWAIT_LEAF values (which the cstate.c code
192 * checks against). The hypervisor won't expose the MWAIT flag because
193 * it would break backwards compatibility; so we will find out directly
194 * from the hardware and hypercall.
196 if (!xen_initial_domain())
200 * When running under platform earlier than Xen4.2, do not expose
201 * mwait, to avoid the risk of loading native acpi pad driver
203 if (!xen_running_on_version_or_later(4, 2))
209 native_cpuid(&ax, &bx, &cx, &dx);
211 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
212 (1 << (X86_FEATURE_MWAIT % 32));
214 if ((cx & mwait_mask) != mwait_mask)
217 /* We need to emulate the MWAIT_LEAF and for that we need both
218 * ecx and edx. The hypercall provides only partial information.
221 ax = CPUID_MWAIT_LEAF;
226 native_cpuid(&ax, &bx, &cx, &dx);
228 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
229 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
231 buf[0] = ACPI_PDC_REVISION_ID;
233 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
235 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
237 if ((HYPERVISOR_platform_op(&op) == 0) &&
238 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
239 cpuid_leaf5_ecx_val = cx;
240 cpuid_leaf5_edx_val = dx;
248 static bool __init xen_check_xsave(void)
250 unsigned int cx, xsave_mask;
254 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
255 (1 << (X86_FEATURE_OSXSAVE % 32));
257 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
258 return (cx & xsave_mask) == xsave_mask;
261 static void __init xen_init_capabilities(void)
263 setup_force_cpu_cap(X86_FEATURE_XENPV);
264 setup_clear_cpu_cap(X86_FEATURE_DCA);
265 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
266 setup_clear_cpu_cap(X86_FEATURE_MTRR);
267 setup_clear_cpu_cap(X86_FEATURE_ACC);
268 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
269 setup_clear_cpu_cap(X86_FEATURE_SME);
272 * Xen PV would need some work to support PCID: CR3 handling as well
273 * as xen_flush_tlb_others() would need updating.
275 setup_clear_cpu_cap(X86_FEATURE_PCID);
277 if (!xen_initial_domain())
278 setup_clear_cpu_cap(X86_FEATURE_ACPI);
280 if (xen_check_mwait())
281 setup_force_cpu_cap(X86_FEATURE_MWAIT);
283 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
285 if (!xen_check_xsave()) {
286 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
287 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
291 static void xen_set_debugreg(int reg, unsigned long val)
293 HYPERVISOR_set_debugreg(reg, val);
296 static unsigned long xen_get_debugreg(int reg)
298 return HYPERVISOR_get_debugreg(reg);
301 static void xen_end_context_switch(struct task_struct *next)
304 paravirt_end_context_switch(next);
307 static unsigned long xen_store_tr(void)
313 * Set the page permissions for a particular virtual address. If the
314 * address is a vmalloc mapping (or other non-linear mapping), then
315 * find the linear mapping of the page and also set its protections to
318 static void set_aliased_prot(void *v, pgprot_t prot)
327 ptep = lookup_address((unsigned long)v, &level);
328 BUG_ON(ptep == NULL);
330 pfn = pte_pfn(*ptep);
331 page = pfn_to_page(pfn);
333 pte = pfn_pte(pfn, prot);
336 * Careful: update_va_mapping() will fail if the virtual address
337 * we're poking isn't populated in the page tables. We don't
338 * need to worry about the direct map (that's always in the page
339 * tables), but we need to be careful about vmap space. In
340 * particular, the top level page table can lazily propagate
341 * entries between processes, so if we've switched mms since we
342 * vmapped the target in the first place, we might not have the
343 * top-level page table entry populated.
345 * We disable preemption because we want the same mm active when
346 * we probe the target and when we issue the hypercall. We'll
347 * have the same nominal mm, but if we're a kernel thread, lazy
348 * mm dropping could change our pgd.
350 * Out of an abundance of caution, this uses __get_user() to fault
351 * in the target address just in case there's some obscure case
352 * in which the target address isn't readable.
357 probe_kernel_read(&dummy, v, 1);
359 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
362 if (!PageHighMem(page)) {
363 void *av = __va(PFN_PHYS(pfn));
366 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
374 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
376 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
380 * We need to mark the all aliases of the LDT pages RO. We
381 * don't need to call vm_flush_aliases(), though, since that's
382 * only responsible for flushing aliases out the TLBs, not the
383 * page tables, and Xen will flush the TLB for us if needed.
385 * To avoid confusing future readers: none of this is necessary
386 * to load the LDT. The hypervisor only checks this when the
387 * LDT is faulted in due to subsequent descriptor access.
390 for (i = 0; i < entries; i += entries_per_page)
391 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
394 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
396 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
399 for (i = 0; i < entries; i += entries_per_page)
400 set_aliased_prot(ldt + i, PAGE_KERNEL);
403 static void xen_set_ldt(const void *addr, unsigned entries)
405 struct mmuext_op *op;
406 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
408 trace_xen_cpu_set_ldt(addr, entries);
411 op->cmd = MMUEXT_SET_LDT;
412 op->arg1.linear_addr = (unsigned long)addr;
413 op->arg2.nr_ents = entries;
415 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
417 xen_mc_issue(PARAVIRT_LAZY_CPU);
420 static void xen_load_gdt(const struct desc_ptr *dtr)
422 unsigned long va = dtr->address;
423 unsigned int size = dtr->size + 1;
424 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
425 unsigned long frames[pages];
429 * A GDT can be up to 64k in size, which corresponds to 8192
430 * 8-byte entries, or 16 4k pages..
433 BUG_ON(size > 65536);
434 BUG_ON(va & ~PAGE_MASK);
436 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
439 unsigned long pfn, mfn;
443 * The GDT is per-cpu and is in the percpu data area.
444 * That can be virtually mapped, so we need to do a
445 * page-walk to get the underlying MFN for the
446 * hypercall. The page can also be in the kernel's
447 * linear range, so we need to RO that mapping too.
449 ptep = lookup_address(va, &level);
450 BUG_ON(ptep == NULL);
452 pfn = pte_pfn(*ptep);
453 mfn = pfn_to_mfn(pfn);
454 virt = __va(PFN_PHYS(pfn));
458 make_lowmem_page_readonly((void *)va);
459 make_lowmem_page_readonly(virt);
462 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
467 * load_gdt for early boot, when the gdt is only mapped once
469 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
471 unsigned long va = dtr->address;
472 unsigned int size = dtr->size + 1;
473 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
474 unsigned long frames[pages];
478 * A GDT can be up to 64k in size, which corresponds to 8192
479 * 8-byte entries, or 16 4k pages..
482 BUG_ON(size > 65536);
483 BUG_ON(va & ~PAGE_MASK);
485 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
487 unsigned long pfn, mfn;
489 pfn = virt_to_pfn(va);
490 mfn = pfn_to_mfn(pfn);
492 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
494 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
500 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
504 static inline bool desc_equal(const struct desc_struct *d1,
505 const struct desc_struct *d2)
507 return !memcmp(d1, d2, sizeof(*d1));
510 static void load_TLS_descriptor(struct thread_struct *t,
511 unsigned int cpu, unsigned int i)
513 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
514 struct desc_struct *gdt;
516 struct multicall_space mc;
518 if (desc_equal(shadow, &t->tls_array[i]))
521 *shadow = t->tls_array[i];
523 gdt = get_cpu_gdt_rw(cpu);
524 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
525 mc = __xen_mc_entry(0);
527 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
530 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
533 * XXX sleazy hack: If we're being called in a lazy-cpu zone
534 * and lazy gs handling is enabled, it means we're in a
535 * context switch, and %gs has just been saved. This means we
536 * can zero it out to prevent faults on exit from the
537 * hypervisor if the next process has no %gs. Either way, it
538 * has been saved, and the new value will get loaded properly.
539 * This will go away as soon as Xen has been modified to not
540 * save/restore %gs for normal hypercalls.
542 * On x86_64, this hack is not used for %gs, because gs points
543 * to KERNEL_GS_BASE (and uses it for PDA references), so we
544 * must not zero %gs on x86_64
546 * For x86_64, we need to zero %fs, otherwise we may get an
547 * exception between the new %fs descriptor being loaded and
548 * %fs being effectively cleared at __switch_to().
550 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
560 load_TLS_descriptor(t, cpu, 0);
561 load_TLS_descriptor(t, cpu, 1);
562 load_TLS_descriptor(t, cpu, 2);
564 xen_mc_issue(PARAVIRT_LAZY_CPU);
568 static void xen_load_gs_index(unsigned int idx)
570 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
575 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
578 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
579 u64 entry = *(u64 *)ptr;
581 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
586 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
593 struct trap_array_entry {
599 static struct trap_array_entry trap_array[] = {
600 { debug, xen_xendebug, true },
601 { double_fault, xen_double_fault, true },
602 #ifdef CONFIG_X86_MCE
603 { machine_check, xen_machine_check, true },
605 { nmi, xen_xennmi, true },
606 { int3, xen_int3, false },
607 { overflow, xen_overflow, false },
608 #ifdef CONFIG_IA32_EMULATION
609 { entry_INT80_compat, xen_entry_INT80_compat, false },
611 { page_fault, xen_page_fault, false },
612 { divide_error, xen_divide_error, false },
613 { bounds, xen_bounds, false },
614 { invalid_op, xen_invalid_op, false },
615 { device_not_available, xen_device_not_available, false },
616 { coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
617 { invalid_TSS, xen_invalid_TSS, false },
618 { segment_not_present, xen_segment_not_present, false },
619 { stack_segment, xen_stack_segment, false },
620 { general_protection, xen_general_protection, false },
621 { spurious_interrupt_bug, xen_spurious_interrupt_bug, false },
622 { coprocessor_error, xen_coprocessor_error, false },
623 { alignment_check, xen_alignment_check, false },
624 { simd_coprocessor_error, xen_simd_coprocessor_error, false },
627 static bool __ref get_trap_addr(void **addr, unsigned int ist)
630 bool ist_okay = false;
633 * Replace trap handler addresses by Xen specific ones.
634 * Check for known traps using IST and whitelist them.
635 * The debugger ones are the only ones we care about.
636 * Xen will handle faults like double_fault, * so we should never see
637 * them. Warn if there's an unexpected IST-using fault handler.
639 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
640 struct trap_array_entry *entry = trap_array + nr;
642 if (*addr == entry->orig) {
644 ist_okay = entry->ist_okay;
649 if (nr == ARRAY_SIZE(trap_array) &&
650 *addr >= (void *)early_idt_handler_array[0] &&
651 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
652 nr = (*addr - (void *)early_idt_handler_array[0]) /
653 EARLY_IDT_HANDLER_SIZE;
654 *addr = (void *)xen_early_idt_handler_array[nr];
657 if (WARN_ON(ist != 0 && !ist_okay))
664 static int cvt_gate_to_trap(int vector, const gate_desc *val,
665 struct trap_info *info)
669 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
672 info->vector = vector;
674 addr = gate_offset(val);
676 if (!get_trap_addr((void **)&addr, val->bits.ist))
678 #endif /* CONFIG_X86_64 */
679 info->address = addr;
681 info->cs = gate_segment(val);
682 info->flags = val->bits.dpl;
683 /* interrupt gates clear IF */
684 if (val->bits.type == GATE_INTERRUPT)
685 info->flags |= 1 << 2;
690 /* Locations of each CPU's IDT */
691 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
693 /* Set an IDT entry. If the entry is part of the current IDT, then
695 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
697 unsigned long p = (unsigned long)&dt[entrynum];
698 unsigned long start, end;
700 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
704 start = __this_cpu_read(idt_desc.address);
705 end = start + __this_cpu_read(idt_desc.size) + 1;
709 native_write_idt_entry(dt, entrynum, g);
711 if (p >= start && (p + 8) <= end) {
712 struct trap_info info[2];
716 if (cvt_gate_to_trap(entrynum, g, &info[0]))
717 if (HYPERVISOR_set_trap_table(info))
724 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
725 struct trap_info *traps, bool full)
727 unsigned in, out, count;
729 count = (desc->size+1) / sizeof(gate_desc);
732 for (in = out = 0; in < count; in++) {
733 gate_desc *entry = (gate_desc *)(desc->address) + in;
735 if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
742 void xen_copy_trap_info(struct trap_info *traps)
744 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
746 xen_convert_trap_info(desc, traps, true);
749 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
750 hold a spinlock to protect the static traps[] array (static because
751 it avoids allocation, and saves stack space). */
752 static void xen_load_idt(const struct desc_ptr *desc)
754 static DEFINE_SPINLOCK(lock);
755 static struct trap_info traps[257];
758 trace_xen_cpu_load_idt(desc);
762 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
764 out = xen_convert_trap_info(desc, traps, false);
765 memset(&traps[out], 0, sizeof(traps[0]));
768 if (HYPERVISOR_set_trap_table(traps))
774 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
775 they're handled differently. */
776 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
777 const void *desc, int type)
779 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
790 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
793 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
803 * Version of write_gdt_entry for use at early boot-time needed to
804 * update an entry as simply as possible.
806 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
807 const void *desc, int type)
809 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
818 xmaddr_t maddr = virt_to_machine(&dt[entry]);
820 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
821 dt[entry] = *(struct desc_struct *)desc;
827 static void xen_load_sp0(unsigned long sp0)
829 struct multicall_space mcs;
831 mcs = xen_mc_entry(0);
832 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
833 xen_mc_issue(PARAVIRT_LAZY_CPU);
834 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
837 void xen_set_iopl_mask(unsigned mask)
839 struct physdev_set_iopl set_iopl;
841 /* Force the change at ring 0. */
842 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
843 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
846 static void xen_io_delay(void)
850 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
852 static unsigned long xen_read_cr0(void)
854 unsigned long cr0 = this_cpu_read(xen_cr0_value);
856 if (unlikely(cr0 == 0)) {
857 cr0 = native_read_cr0();
858 this_cpu_write(xen_cr0_value, cr0);
864 static void xen_write_cr0(unsigned long cr0)
866 struct multicall_space mcs;
868 this_cpu_write(xen_cr0_value, cr0);
870 /* Only pay attention to cr0.TS; everything else is
872 mcs = xen_mc_entry(0);
874 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
876 xen_mc_issue(PARAVIRT_LAZY_CPU);
879 static void xen_write_cr4(unsigned long cr4)
881 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
883 native_write_cr4(cr4);
886 static inline unsigned long xen_read_cr8(void)
890 static inline void xen_write_cr8(unsigned long val)
896 static u64 xen_read_msr_safe(unsigned int msr, int *err)
900 if (pmu_msr_read(msr, &val, err))
903 val = native_read_msr_safe(msr, err);
905 case MSR_IA32_APICBASE:
906 val &= ~X2APIC_ENABLE;
912 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
924 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
925 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
926 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
929 base = ((u64)high << 32) | low;
930 if (HYPERVISOR_set_segment_base(which, base) != 0)
938 case MSR_SYSCALL_MASK:
939 case MSR_IA32_SYSENTER_CS:
940 case MSR_IA32_SYSENTER_ESP:
941 case MSR_IA32_SYSENTER_EIP:
942 /* Fast syscall setup is all done in hypercalls, so
943 these are all ignored. Stub them out here to stop
944 Xen console noise. */
948 if (!pmu_msr_write(msr, low, high, &ret))
949 ret = native_write_msr_safe(msr, low, high);
955 static u64 xen_read_msr(unsigned int msr)
958 * This will silently swallow a #GP from RDMSR. It may be worth
963 return xen_read_msr_safe(msr, &err);
966 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
969 * This will silently swallow a #GP from WRMSR. It may be worth
972 xen_write_msr_safe(msr, low, high);
975 void xen_setup_shared_info(void)
977 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
979 HYPERVISOR_shared_info =
980 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
982 xen_setup_mfn_list_list();
984 if (system_state == SYSTEM_BOOTING) {
987 * In UP this is as good a place as any to set up shared info.
988 * Limit this to boot only, at restore vcpu setup is done via
989 * xen_vcpu_restore().
991 xen_setup_vcpu_info_placement();
994 * Now that shared info is set up we can start using routines
995 * that point to pvclock area.
1001 /* This is called once we have the cpu_possible_mask */
1002 void __ref xen_setup_vcpu_info_placement(void)
1006 for_each_possible_cpu(cpu) {
1007 /* Set up direct vCPU id mapping for PV guests. */
1008 per_cpu(xen_vcpu_id, cpu) = cpu;
1011 * xen_vcpu_setup(cpu) can fail -- in which case it
1012 * falls back to the shared_info version for cpus
1013 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1015 * xen_cpu_up_prepare_pv() handles the rest by failing
1018 (void) xen_vcpu_setup(cpu);
1022 * xen_vcpu_setup managed to place the vcpu_info within the
1023 * percpu area for all cpus, so make use of it.
1025 if (xen_have_vcpu_info_placement) {
1026 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1027 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1028 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1029 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1030 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1034 static const struct pv_info xen_info __initconst = {
1035 .shared_kernel_pmd = 0,
1037 #ifdef CONFIG_X86_64
1038 .extra_user_64bit_cs = FLAT_USER_CS64,
1043 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1046 .set_debugreg = xen_set_debugreg,
1047 .get_debugreg = xen_get_debugreg,
1049 .read_cr0 = xen_read_cr0,
1050 .write_cr0 = xen_write_cr0,
1052 .write_cr4 = xen_write_cr4,
1054 #ifdef CONFIG_X86_64
1055 .read_cr8 = xen_read_cr8,
1056 .write_cr8 = xen_write_cr8,
1059 .wbinvd = native_wbinvd,
1061 .read_msr = xen_read_msr,
1062 .write_msr = xen_write_msr,
1064 .read_msr_safe = xen_read_msr_safe,
1065 .write_msr_safe = xen_write_msr_safe,
1067 .read_pmc = xen_read_pmc,
1070 #ifdef CONFIG_X86_64
1071 .usergs_sysret64 = xen_sysret64,
1074 .load_tr_desc = paravirt_nop,
1075 .set_ldt = xen_set_ldt,
1076 .load_gdt = xen_load_gdt,
1077 .load_idt = xen_load_idt,
1078 .load_tls = xen_load_tls,
1079 #ifdef CONFIG_X86_64
1080 .load_gs_index = xen_load_gs_index,
1083 .alloc_ldt = xen_alloc_ldt,
1084 .free_ldt = xen_free_ldt,
1086 .store_tr = xen_store_tr,
1088 .write_ldt_entry = xen_write_ldt_entry,
1089 .write_gdt_entry = xen_write_gdt_entry,
1090 .write_idt_entry = xen_write_idt_entry,
1091 .load_sp0 = xen_load_sp0,
1093 .set_iopl_mask = xen_set_iopl_mask,
1094 .io_delay = xen_io_delay,
1096 /* Xen takes care of %gs when switching to usermode for us */
1097 .swapgs = paravirt_nop,
1099 .start_context_switch = paravirt_start_context_switch,
1100 .end_context_switch = xen_end_context_switch,
1103 static void xen_restart(char *msg)
1105 xen_reboot(SHUTDOWN_reboot);
1108 static void xen_machine_halt(void)
1110 xen_reboot(SHUTDOWN_poweroff);
1113 static void xen_machine_power_off(void)
1117 xen_reboot(SHUTDOWN_poweroff);
1120 static void xen_crash_shutdown(struct pt_regs *regs)
1122 xen_reboot(SHUTDOWN_crash);
1125 static const struct machine_ops xen_machine_ops __initconst = {
1126 .restart = xen_restart,
1127 .halt = xen_machine_halt,
1128 .power_off = xen_machine_power_off,
1129 .shutdown = xen_machine_halt,
1130 .crash_shutdown = xen_crash_shutdown,
1131 .emergency_restart = xen_emergency_restart,
1134 static unsigned char xen_get_nmi_reason(void)
1136 unsigned char reason = 0;
1138 /* Construct a value which looks like it came from port 0x61. */
1139 if (test_bit(_XEN_NMIREASON_io_error,
1140 &HYPERVISOR_shared_info->arch.nmi_reason))
1141 reason |= NMI_REASON_IOCHK;
1142 if (test_bit(_XEN_NMIREASON_pci_serr,
1143 &HYPERVISOR_shared_info->arch.nmi_reason))
1144 reason |= NMI_REASON_SERR;
1149 static void __init xen_boot_params_init_edd(void)
1151 #if IS_ENABLED(CONFIG_EDD)
1152 struct xen_platform_op op;
1153 struct edd_info *edd_info;
1158 edd_info = boot_params.eddbuf;
1159 mbr_signature = boot_params.edd_mbr_sig_buffer;
1161 op.cmd = XENPF_firmware_info;
1163 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1164 for (nr = 0; nr < EDDMAXNR; nr++) {
1165 struct edd_info *info = edd_info + nr;
1167 op.u.firmware_info.index = nr;
1168 info->params.length = sizeof(info->params);
1169 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1171 ret = HYPERVISOR_platform_op(&op);
1175 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1178 C(interface_support);
1179 C(legacy_max_cylinder);
1181 C(legacy_sectors_per_track);
1184 boot_params.eddbuf_entries = nr;
1186 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1187 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1188 op.u.firmware_info.index = nr;
1189 ret = HYPERVISOR_platform_op(&op);
1192 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1194 boot_params.edd_mbr_sig_buf_entries = nr;
1199 * Set up the GDT and segment registers for -fstack-protector. Until
1200 * we do this, we have to be careful not to call any stack-protected
1201 * function, which is most of the kernel.
1203 static void xen_setup_gdt(int cpu)
1205 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1206 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1208 setup_stack_canary_segment(0);
1209 switch_to_new_gdt(0);
1211 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1212 pv_cpu_ops.load_gdt = xen_load_gdt;
1215 static void __init xen_dom0_set_legacy_features(void)
1217 x86_platform.legacy.rtc = 1;
1220 static void __init xen_domu_set_legacy_features(void)
1222 x86_platform.legacy.rtc = 0;
1225 /* First C function to be called on Xen boot */
1226 asmlinkage __visible void __init xen_start_kernel(void)
1228 struct physdev_set_iopl set_iopl;
1229 unsigned long initrd_start = 0;
1232 if (!xen_start_info)
1235 xen_domain_type = XEN_PV_DOMAIN;
1237 xen_setup_features();
1239 /* Install Xen paravirt ops */
1241 pv_init_ops.patch = paravirt_patch_default;
1242 pv_cpu_ops = xen_cpu_ops;
1246 * Setup xen_vcpu early because it is needed for
1247 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1249 * Don't do the full vcpu_info placement stuff until we have
1250 * the cpu_possible_mask and a non-dummy shared_info.
1252 xen_vcpu_info_reset(0);
1254 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1256 x86_init.resources.memory_setup = xen_memory_setup;
1257 x86_init.oem.arch_setup = xen_arch_setup;
1258 x86_init.oem.banner = xen_banner;
1261 * Set up some pagetable state before starting to set any ptes.
1264 xen_setup_machphys_mapping();
1267 /* Prevent unwanted bits from being set in PTEs. */
1268 __supported_pte_mask &= ~_PAGE_GLOBAL;
1271 * Prevent page tables from being allocated in highmem, even
1272 * if CONFIG_HIGHPTE is enabled.
1274 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1277 xen_build_dynamic_phys_to_machine();
1280 * Set up kernel GDT and segment registers, mainly so that
1281 * -fstack-protector code can be executed.
1285 /* Work out if we support NX */
1286 get_cpu_cap(&boot_cpu_data);
1289 /* Let's presume PV guests always boot on vCPU with id 0. */
1290 per_cpu(xen_vcpu_id, 0) = 0;
1292 idt_setup_early_handler();
1294 xen_init_capabilities();
1296 #ifdef CONFIG_X86_LOCAL_APIC
1298 * set up the basic apic ops.
1303 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1304 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1305 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1308 machine_ops = xen_machine_ops;
1311 * The only reliable way to retain the initial address of the
1312 * percpu gdt_page is to remember it here, so we can go and
1313 * mark it RW later, when the initial percpu area is freed.
1315 xen_initial_gdt = &per_cpu(gdt_page, 0);
1319 #ifdef CONFIG_ACPI_NUMA
1321 * The pages we from Xen are not related to machine pages, so
1322 * any NUMA information the kernel tries to get from ACPI will
1323 * be meaningless. Prevent it from trying.
1327 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1329 local_irq_disable();
1330 early_boot_irqs_disabled = true;
1332 xen_raw_console_write("mapping kernel into physical memory\n");
1333 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1334 xen_start_info->nr_pages);
1335 xen_reserve_special_pages();
1337 /* keep using Xen gdt for now; no urgent need to change it */
1339 #ifdef CONFIG_X86_32
1340 pv_info.kernel_rpl = 1;
1341 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1342 pv_info.kernel_rpl = 0;
1344 pv_info.kernel_rpl = 0;
1346 /* set the limit of our address space */
1350 * We used to do this in xen_arch_setup, but that is too late
1351 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1352 * early_amd_init which pokes 0xcf8 port.
1355 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1357 xen_raw_printk("physdev_op failed %d\n", rc);
1359 #ifdef CONFIG_X86_32
1360 /* set up basic CPUID stuff */
1361 cpu_detect(&new_cpu_data);
1362 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1363 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1366 if (xen_start_info->mod_start) {
1367 if (xen_start_info->flags & SIF_MOD_START_PFN)
1368 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1370 initrd_start = __pa(xen_start_info->mod_start);
1373 /* Poke various useful things into boot_params */
1374 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1375 boot_params.hdr.ramdisk_image = initrd_start;
1376 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1377 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1378 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1380 if (!xen_initial_domain()) {
1381 add_preferred_console("xenboot", 0, NULL);
1382 add_preferred_console("tty", 0, NULL);
1383 add_preferred_console("hvc", 0, NULL);
1385 x86_init.pci.arch_init = pci_xen_init;
1386 x86_platform.set_legacy_features =
1387 xen_domu_set_legacy_features;
1389 const struct dom0_vga_console_info *info =
1390 (void *)((char *)xen_start_info +
1391 xen_start_info->console.dom0.info_off);
1392 struct xen_platform_op op = {
1393 .cmd = XENPF_firmware_info,
1394 .interface_version = XENPF_INTERFACE_VERSION,
1395 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1398 x86_platform.set_legacy_features =
1399 xen_dom0_set_legacy_features;
1400 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1401 xen_start_info->console.domU.mfn = 0;
1402 xen_start_info->console.domU.evtchn = 0;
1404 if (HYPERVISOR_platform_op(&op) == 0)
1405 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1407 /* Make sure ACS will be enabled */
1410 xen_acpi_sleep_register();
1412 /* Avoid searching for BIOS MP tables */
1413 x86_init.mpparse.find_smp_config = x86_init_noop;
1414 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1416 xen_boot_params_init_edd();
1420 * Disable selecting "Firmware First mode" for correctable
1421 * memory errors, as this is the duty of the hypervisor to
1424 acpi_disable_cmcff = 1;
1428 /* PCI BIOS service won't work from a PV guest. */
1429 pci_probe &= ~PCI_PROBE_BIOS;
1431 xen_raw_console_write("about to get started...\n");
1433 /* We need this for printk timestamps */
1434 xen_setup_runstate_info(0);
1438 /* Start the world */
1439 #ifdef CONFIG_X86_32
1440 i386_start_kernel();
1442 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1443 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1447 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1451 if (per_cpu(xen_vcpu, cpu) == NULL)
1454 xen_setup_timer(cpu);
1456 rc = xen_smp_intr_init(cpu);
1458 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1463 rc = xen_smp_intr_init_pv(cpu);
1465 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1473 static int xen_cpu_dead_pv(unsigned int cpu)
1475 xen_smp_intr_free(cpu);
1476 xen_smp_intr_free_pv(cpu);
1478 xen_teardown_timer(cpu);
1483 static uint32_t __init xen_platform_pv(void)
1485 if (xen_pv_domain())
1486 return xen_cpuid_base();
1491 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1493 .detect = xen_platform_pv,
1494 .type = X86_HYPER_XEN_PV,
1495 .runtime.pin_vcpu = xen_pin_vcpu,