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)" : "");
123 static void __init xen_pv_init_platform(void)
125 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
126 HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
128 /* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
129 xen_vcpu_info_reset(0);
131 /* pvclock is in shared info area */
135 static void __init xen_pv_guest_late_init(void)
138 /* Setup shared vcpu info for non-smp configurations */
139 xen_setup_vcpu_info_placement();
143 /* Check if running on Xen version (major, minor) or later */
145 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
147 unsigned int version;
152 version = HYPERVISOR_xen_version(XENVER_version, NULL);
153 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
154 ((version >> 16) > major))
159 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
160 static __read_mostly unsigned int cpuid_leaf5_edx_val;
162 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
163 unsigned int *cx, unsigned int *dx)
165 unsigned maskebx = ~0;
168 * Mask out inconvenient features, to try and disable as many
169 * unsupported kernel subsystems as possible.
172 case CPUID_MWAIT_LEAF:
173 /* Synthesize the values.. */
176 *cx = cpuid_leaf5_ecx_val;
177 *dx = cpuid_leaf5_edx_val;
181 /* Suppress extended topology stuff */
186 asm(XEN_EMULATE_PREFIX "cpuid"
191 : "0" (*ax), "2" (*cx));
195 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
197 static bool __init xen_check_mwait(void)
200 struct xen_platform_op op = {
201 .cmd = XENPF_set_processor_pminfo,
202 .u.set_pminfo.id = -1,
203 .u.set_pminfo.type = XEN_PM_PDC,
206 unsigned int ax, bx, cx, dx;
207 unsigned int mwait_mask;
209 /* We need to determine whether it is OK to expose the MWAIT
210 * capability to the kernel to harvest deeper than C3 states from ACPI
211 * _CST using the processor_harvest_xen.c module. For this to work, we
212 * need to gather the MWAIT_LEAF values (which the cstate.c code
213 * checks against). The hypervisor won't expose the MWAIT flag because
214 * it would break backwards compatibility; so we will find out directly
215 * from the hardware and hypercall.
217 if (!xen_initial_domain())
221 * When running under platform earlier than Xen4.2, do not expose
222 * mwait, to avoid the risk of loading native acpi pad driver
224 if (!xen_running_on_version_or_later(4, 2))
230 native_cpuid(&ax, &bx, &cx, &dx);
232 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
233 (1 << (X86_FEATURE_MWAIT % 32));
235 if ((cx & mwait_mask) != mwait_mask)
238 /* We need to emulate the MWAIT_LEAF and for that we need both
239 * ecx and edx. The hypercall provides only partial information.
242 ax = CPUID_MWAIT_LEAF;
247 native_cpuid(&ax, &bx, &cx, &dx);
249 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
250 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
252 buf[0] = ACPI_PDC_REVISION_ID;
254 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
256 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
258 if ((HYPERVISOR_platform_op(&op) == 0) &&
259 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
260 cpuid_leaf5_ecx_val = cx;
261 cpuid_leaf5_edx_val = dx;
269 static bool __init xen_check_xsave(void)
271 unsigned int cx, xsave_mask;
275 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
276 (1 << (X86_FEATURE_OSXSAVE % 32));
278 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
279 return (cx & xsave_mask) == xsave_mask;
282 static void __init xen_init_capabilities(void)
284 setup_force_cpu_cap(X86_FEATURE_XENPV);
285 setup_clear_cpu_cap(X86_FEATURE_DCA);
286 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
287 setup_clear_cpu_cap(X86_FEATURE_MTRR);
288 setup_clear_cpu_cap(X86_FEATURE_ACC);
289 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
290 setup_clear_cpu_cap(X86_FEATURE_SME);
293 * Xen PV would need some work to support PCID: CR3 handling as well
294 * as xen_flush_tlb_others() would need updating.
296 setup_clear_cpu_cap(X86_FEATURE_PCID);
298 if (!xen_initial_domain())
299 setup_clear_cpu_cap(X86_FEATURE_ACPI);
301 if (xen_check_mwait())
302 setup_force_cpu_cap(X86_FEATURE_MWAIT);
304 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
306 if (!xen_check_xsave()) {
307 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
308 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
312 static void xen_set_debugreg(int reg, unsigned long val)
314 HYPERVISOR_set_debugreg(reg, val);
317 static unsigned long xen_get_debugreg(int reg)
319 return HYPERVISOR_get_debugreg(reg);
322 static void xen_end_context_switch(struct task_struct *next)
325 paravirt_end_context_switch(next);
328 static unsigned long xen_store_tr(void)
334 * Set the page permissions for a particular virtual address. If the
335 * address is a vmalloc mapping (or other non-linear mapping), then
336 * find the linear mapping of the page and also set its protections to
339 static void set_aliased_prot(void *v, pgprot_t prot)
348 ptep = lookup_address((unsigned long)v, &level);
349 BUG_ON(ptep == NULL);
351 pfn = pte_pfn(*ptep);
352 page = pfn_to_page(pfn);
354 pte = pfn_pte(pfn, prot);
357 * Careful: update_va_mapping() will fail if the virtual address
358 * we're poking isn't populated in the page tables. We don't
359 * need to worry about the direct map (that's always in the page
360 * tables), but we need to be careful about vmap space. In
361 * particular, the top level page table can lazily propagate
362 * entries between processes, so if we've switched mms since we
363 * vmapped the target in the first place, we might not have the
364 * top-level page table entry populated.
366 * We disable preemption because we want the same mm active when
367 * we probe the target and when we issue the hypercall. We'll
368 * have the same nominal mm, but if we're a kernel thread, lazy
369 * mm dropping could change our pgd.
371 * Out of an abundance of caution, this uses __get_user() to fault
372 * in the target address just in case there's some obscure case
373 * in which the target address isn't readable.
378 probe_kernel_read(&dummy, v, 1);
380 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
383 if (!PageHighMem(page)) {
384 void *av = __va(PFN_PHYS(pfn));
387 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
395 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
397 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
401 * We need to mark the all aliases of the LDT pages RO. We
402 * don't need to call vm_flush_aliases(), though, since that's
403 * only responsible for flushing aliases out the TLBs, not the
404 * page tables, and Xen will flush the TLB for us if needed.
406 * To avoid confusing future readers: none of this is necessary
407 * to load the LDT. The hypervisor only checks this when the
408 * LDT is faulted in due to subsequent descriptor access.
411 for (i = 0; i < entries; i += entries_per_page)
412 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
415 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
417 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
420 for (i = 0; i < entries; i += entries_per_page)
421 set_aliased_prot(ldt + i, PAGE_KERNEL);
424 static void xen_set_ldt(const void *addr, unsigned entries)
426 struct mmuext_op *op;
427 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
429 trace_xen_cpu_set_ldt(addr, entries);
432 op->cmd = MMUEXT_SET_LDT;
433 op->arg1.linear_addr = (unsigned long)addr;
434 op->arg2.nr_ents = entries;
436 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
438 xen_mc_issue(PARAVIRT_LAZY_CPU);
441 static void xen_load_gdt(const struct desc_ptr *dtr)
443 unsigned long va = dtr->address;
444 unsigned int size = dtr->size + 1;
445 unsigned long pfn, mfn;
450 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
451 BUG_ON(size > PAGE_SIZE);
452 BUG_ON(va & ~PAGE_MASK);
455 * The GDT is per-cpu and is in the percpu data area.
456 * That can be virtually mapped, so we need to do a
457 * page-walk to get the underlying MFN for the
458 * hypercall. The page can also be in the kernel's
459 * linear range, so we need to RO that mapping too.
461 ptep = lookup_address(va, &level);
462 BUG_ON(ptep == NULL);
464 pfn = pte_pfn(*ptep);
465 mfn = pfn_to_mfn(pfn);
466 virt = __va(PFN_PHYS(pfn));
468 make_lowmem_page_readonly((void *)va);
469 make_lowmem_page_readonly(virt);
471 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
476 * load_gdt for early boot, when the gdt is only mapped once
478 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
480 unsigned long va = dtr->address;
481 unsigned int size = dtr->size + 1;
482 unsigned long pfn, mfn;
485 /* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
486 BUG_ON(size > PAGE_SIZE);
487 BUG_ON(va & ~PAGE_MASK);
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))
497 if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
501 static inline bool desc_equal(const struct desc_struct *d1,
502 const struct desc_struct *d2)
504 return !memcmp(d1, d2, sizeof(*d1));
507 static void load_TLS_descriptor(struct thread_struct *t,
508 unsigned int cpu, unsigned int i)
510 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
511 struct desc_struct *gdt;
513 struct multicall_space mc;
515 if (desc_equal(shadow, &t->tls_array[i]))
518 *shadow = t->tls_array[i];
520 gdt = get_cpu_gdt_rw(cpu);
521 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
522 mc = __xen_mc_entry(0);
524 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
527 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
530 * XXX sleazy hack: If we're being called in a lazy-cpu zone
531 * and lazy gs handling is enabled, it means we're in a
532 * context switch, and %gs has just been saved. This means we
533 * can zero it out to prevent faults on exit from the
534 * hypervisor if the next process has no %gs. Either way, it
535 * has been saved, and the new value will get loaded properly.
536 * This will go away as soon as Xen has been modified to not
537 * save/restore %gs for normal hypercalls.
539 * On x86_64, this hack is not used for %gs, because gs points
540 * to KERNEL_GS_BASE (and uses it for PDA references), so we
541 * must not zero %gs on x86_64
543 * For x86_64, we need to zero %fs, otherwise we may get an
544 * exception between the new %fs descriptor being loaded and
545 * %fs being effectively cleared at __switch_to().
547 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
557 load_TLS_descriptor(t, cpu, 0);
558 load_TLS_descriptor(t, cpu, 1);
559 load_TLS_descriptor(t, cpu, 2);
561 xen_mc_issue(PARAVIRT_LAZY_CPU);
565 static void xen_load_gs_index(unsigned int idx)
567 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
572 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
575 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
576 u64 entry = *(u64 *)ptr;
578 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
583 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
590 struct trap_array_entry {
596 static struct trap_array_entry trap_array[] = {
597 { debug, xen_xendebug, true },
598 { int3, xen_xenint3, true },
599 { double_fault, xen_double_fault, true },
600 #ifdef CONFIG_X86_MCE
601 { machine_check, xen_machine_check, true },
603 { nmi, xen_xennmi, true },
604 { overflow, xen_overflow, false },
605 #ifdef CONFIG_IA32_EMULATION
606 { entry_INT80_compat, xen_entry_INT80_compat, false },
608 { page_fault, xen_page_fault, false },
609 { divide_error, xen_divide_error, false },
610 { bounds, xen_bounds, false },
611 { invalid_op, xen_invalid_op, false },
612 { device_not_available, xen_device_not_available, false },
613 { coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
614 { invalid_TSS, xen_invalid_TSS, false },
615 { segment_not_present, xen_segment_not_present, false },
616 { stack_segment, xen_stack_segment, false },
617 { general_protection, xen_general_protection, false },
618 { spurious_interrupt_bug, xen_spurious_interrupt_bug, false },
619 { coprocessor_error, xen_coprocessor_error, false },
620 { alignment_check, xen_alignment_check, false },
621 { simd_coprocessor_error, xen_simd_coprocessor_error, false },
624 static bool __ref get_trap_addr(void **addr, unsigned int ist)
627 bool ist_okay = false;
630 * Replace trap handler addresses by Xen specific ones.
631 * Check for known traps using IST and whitelist them.
632 * The debugger ones are the only ones we care about.
633 * Xen will handle faults like double_fault, * so we should never see
634 * them. Warn if there's an unexpected IST-using fault handler.
636 for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
637 struct trap_array_entry *entry = trap_array + nr;
639 if (*addr == entry->orig) {
641 ist_okay = entry->ist_okay;
646 if (nr == ARRAY_SIZE(trap_array) &&
647 *addr >= (void *)early_idt_handler_array[0] &&
648 *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
649 nr = (*addr - (void *)early_idt_handler_array[0]) /
650 EARLY_IDT_HANDLER_SIZE;
651 *addr = (void *)xen_early_idt_handler_array[nr];
654 if (WARN_ON(ist != 0 && !ist_okay))
661 static int cvt_gate_to_trap(int vector, const gate_desc *val,
662 struct trap_info *info)
666 if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
669 info->vector = vector;
671 addr = gate_offset(val);
673 if (!get_trap_addr((void **)&addr, val->bits.ist))
675 #endif /* CONFIG_X86_64 */
676 info->address = addr;
678 info->cs = gate_segment(val);
679 info->flags = val->bits.dpl;
680 /* interrupt gates clear IF */
681 if (val->bits.type == GATE_INTERRUPT)
682 info->flags |= 1 << 2;
687 /* Locations of each CPU's IDT */
688 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
690 /* Set an IDT entry. If the entry is part of the current IDT, then
692 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
694 unsigned long p = (unsigned long)&dt[entrynum];
695 unsigned long start, end;
697 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
701 start = __this_cpu_read(idt_desc.address);
702 end = start + __this_cpu_read(idt_desc.size) + 1;
706 native_write_idt_entry(dt, entrynum, g);
708 if (p >= start && (p + 8) <= end) {
709 struct trap_info info[2];
713 if (cvt_gate_to_trap(entrynum, g, &info[0]))
714 if (HYPERVISOR_set_trap_table(info))
721 static void xen_convert_trap_info(const struct desc_ptr *desc,
722 struct trap_info *traps)
724 unsigned in, out, count;
726 count = (desc->size+1) / sizeof(gate_desc);
729 for (in = out = 0; in < count; in++) {
730 gate_desc *entry = (gate_desc *)(desc->address) + in;
732 if (cvt_gate_to_trap(in, entry, &traps[out]))
735 traps[out].address = 0;
738 void xen_copy_trap_info(struct trap_info *traps)
740 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
742 xen_convert_trap_info(desc, traps);
745 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
746 hold a spinlock to protect the static traps[] array (static because
747 it avoids allocation, and saves stack space). */
748 static void xen_load_idt(const struct desc_ptr *desc)
750 static DEFINE_SPINLOCK(lock);
751 static struct trap_info traps[257];
753 trace_xen_cpu_load_idt(desc);
757 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
759 xen_convert_trap_info(desc, traps);
762 if (HYPERVISOR_set_trap_table(traps))
768 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
769 they're handled differently. */
770 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
771 const void *desc, int type)
773 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
784 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
787 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
797 * Version of write_gdt_entry for use at early boot-time needed to
798 * update an entry as simply as possible.
800 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
801 const void *desc, int type)
803 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
812 xmaddr_t maddr = virt_to_machine(&dt[entry]);
814 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
815 dt[entry] = *(struct desc_struct *)desc;
821 static void xen_load_sp0(unsigned long sp0)
823 struct multicall_space mcs;
825 mcs = xen_mc_entry(0);
826 MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
827 xen_mc_issue(PARAVIRT_LAZY_CPU);
828 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
831 void xen_set_iopl_mask(unsigned mask)
833 struct physdev_set_iopl set_iopl;
835 /* Force the change at ring 0. */
836 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
837 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
840 static void xen_io_delay(void)
844 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
846 static unsigned long xen_read_cr0(void)
848 unsigned long cr0 = this_cpu_read(xen_cr0_value);
850 if (unlikely(cr0 == 0)) {
851 cr0 = native_read_cr0();
852 this_cpu_write(xen_cr0_value, cr0);
858 static void xen_write_cr0(unsigned long cr0)
860 struct multicall_space mcs;
862 this_cpu_write(xen_cr0_value, cr0);
864 /* Only pay attention to cr0.TS; everything else is
866 mcs = xen_mc_entry(0);
868 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
870 xen_mc_issue(PARAVIRT_LAZY_CPU);
873 static void xen_write_cr4(unsigned long cr4)
875 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
877 native_write_cr4(cr4);
880 static inline unsigned long xen_read_cr8(void)
884 static inline void xen_write_cr8(unsigned long val)
890 static u64 xen_read_msr_safe(unsigned int msr, int *err)
894 if (pmu_msr_read(msr, &val, err))
897 val = native_read_msr_safe(msr, err);
899 case MSR_IA32_APICBASE:
900 #ifdef CONFIG_X86_X2APIC
901 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
903 val &= ~X2APIC_ENABLE;
909 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
920 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
921 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
922 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
925 base = ((u64)high << 32) | low;
926 if (HYPERVISOR_set_segment_base(which, base) != 0)
934 case MSR_SYSCALL_MASK:
935 case MSR_IA32_SYSENTER_CS:
936 case MSR_IA32_SYSENTER_ESP:
937 case MSR_IA32_SYSENTER_EIP:
938 /* Fast syscall setup is all done in hypercalls, so
939 these are all ignored. Stub them out here to stop
940 Xen console noise. */
944 if (!pmu_msr_write(msr, low, high, &ret))
945 ret = native_write_msr_safe(msr, low, high);
951 static u64 xen_read_msr(unsigned int msr)
954 * This will silently swallow a #GP from RDMSR. It may be worth
959 return xen_read_msr_safe(msr, &err);
962 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
965 * This will silently swallow a #GP from WRMSR. It may be worth
968 xen_write_msr_safe(msr, low, high);
971 /* This is called once we have the cpu_possible_mask */
972 void __init xen_setup_vcpu_info_placement(void)
976 for_each_possible_cpu(cpu) {
977 /* Set up direct vCPU id mapping for PV guests. */
978 per_cpu(xen_vcpu_id, cpu) = cpu;
981 * xen_vcpu_setup(cpu) can fail -- in which case it
982 * falls back to the shared_info version for cpus
983 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
985 * xen_cpu_up_prepare_pv() handles the rest by failing
988 (void) xen_vcpu_setup(cpu);
992 * xen_vcpu_setup managed to place the vcpu_info within the
993 * percpu area for all cpus, so make use of it.
995 if (xen_have_vcpu_info_placement) {
996 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
997 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
998 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
999 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1000 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1004 static const struct pv_info xen_info __initconst = {
1005 .shared_kernel_pmd = 0,
1007 #ifdef CONFIG_X86_64
1008 .extra_user_64bit_cs = FLAT_USER_CS64,
1013 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1016 .set_debugreg = xen_set_debugreg,
1017 .get_debugreg = xen_get_debugreg,
1019 .read_cr0 = xen_read_cr0,
1020 .write_cr0 = xen_write_cr0,
1022 .write_cr4 = xen_write_cr4,
1024 #ifdef CONFIG_X86_64
1025 .read_cr8 = xen_read_cr8,
1026 .write_cr8 = xen_write_cr8,
1029 .wbinvd = native_wbinvd,
1031 .read_msr = xen_read_msr,
1032 .write_msr = xen_write_msr,
1034 .read_msr_safe = xen_read_msr_safe,
1035 .write_msr_safe = xen_write_msr_safe,
1037 .read_pmc = xen_read_pmc,
1040 #ifdef CONFIG_X86_64
1041 .usergs_sysret64 = xen_sysret64,
1044 .load_tr_desc = paravirt_nop,
1045 .set_ldt = xen_set_ldt,
1046 .load_gdt = xen_load_gdt,
1047 .load_idt = xen_load_idt,
1048 .load_tls = xen_load_tls,
1049 #ifdef CONFIG_X86_64
1050 .load_gs_index = xen_load_gs_index,
1053 .alloc_ldt = xen_alloc_ldt,
1054 .free_ldt = xen_free_ldt,
1056 .store_tr = xen_store_tr,
1058 .write_ldt_entry = xen_write_ldt_entry,
1059 .write_gdt_entry = xen_write_gdt_entry,
1060 .write_idt_entry = xen_write_idt_entry,
1061 .load_sp0 = xen_load_sp0,
1063 .set_iopl_mask = xen_set_iopl_mask,
1064 .io_delay = xen_io_delay,
1066 /* Xen takes care of %gs when switching to usermode for us */
1067 .swapgs = paravirt_nop,
1069 .start_context_switch = paravirt_start_context_switch,
1070 .end_context_switch = xen_end_context_switch,
1073 static void xen_restart(char *msg)
1075 xen_reboot(SHUTDOWN_reboot);
1078 static void xen_machine_halt(void)
1080 xen_reboot(SHUTDOWN_poweroff);
1083 static void xen_machine_power_off(void)
1087 xen_reboot(SHUTDOWN_poweroff);
1090 static void xen_crash_shutdown(struct pt_regs *regs)
1092 xen_reboot(SHUTDOWN_crash);
1095 static const struct machine_ops xen_machine_ops __initconst = {
1096 .restart = xen_restart,
1097 .halt = xen_machine_halt,
1098 .power_off = xen_machine_power_off,
1099 .shutdown = xen_machine_halt,
1100 .crash_shutdown = xen_crash_shutdown,
1101 .emergency_restart = xen_emergency_restart,
1104 static unsigned char xen_get_nmi_reason(void)
1106 unsigned char reason = 0;
1108 /* Construct a value which looks like it came from port 0x61. */
1109 if (test_bit(_XEN_NMIREASON_io_error,
1110 &HYPERVISOR_shared_info->arch.nmi_reason))
1111 reason |= NMI_REASON_IOCHK;
1112 if (test_bit(_XEN_NMIREASON_pci_serr,
1113 &HYPERVISOR_shared_info->arch.nmi_reason))
1114 reason |= NMI_REASON_SERR;
1119 static void __init xen_boot_params_init_edd(void)
1121 #if IS_ENABLED(CONFIG_EDD)
1122 struct xen_platform_op op;
1123 struct edd_info *edd_info;
1128 edd_info = boot_params.eddbuf;
1129 mbr_signature = boot_params.edd_mbr_sig_buffer;
1131 op.cmd = XENPF_firmware_info;
1133 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1134 for (nr = 0; nr < EDDMAXNR; nr++) {
1135 struct edd_info *info = edd_info + nr;
1137 op.u.firmware_info.index = nr;
1138 info->params.length = sizeof(info->params);
1139 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1141 ret = HYPERVISOR_platform_op(&op);
1145 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1148 C(interface_support);
1149 C(legacy_max_cylinder);
1151 C(legacy_sectors_per_track);
1154 boot_params.eddbuf_entries = nr;
1156 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1157 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1158 op.u.firmware_info.index = nr;
1159 ret = HYPERVISOR_platform_op(&op);
1162 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1164 boot_params.edd_mbr_sig_buf_entries = nr;
1169 * Set up the GDT and segment registers for -fstack-protector. Until
1170 * we do this, we have to be careful not to call any stack-protected
1171 * function, which is most of the kernel.
1173 static void xen_setup_gdt(int cpu)
1175 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1176 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1178 setup_stack_canary_segment(0);
1179 switch_to_new_gdt(0);
1181 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1182 pv_cpu_ops.load_gdt = xen_load_gdt;
1185 static void __init xen_dom0_set_legacy_features(void)
1187 x86_platform.legacy.rtc = 1;
1190 /* First C function to be called on Xen boot */
1191 asmlinkage __visible void __init xen_start_kernel(void)
1193 struct physdev_set_iopl set_iopl;
1194 unsigned long initrd_start = 0;
1197 if (!xen_start_info)
1200 xen_domain_type = XEN_PV_DOMAIN;
1201 xen_start_flags = xen_start_info->flags;
1203 xen_setup_features();
1205 /* Install Xen paravirt ops */
1207 pv_init_ops.patch = paravirt_patch_default;
1208 pv_cpu_ops = xen_cpu_ops;
1212 * Setup xen_vcpu early because it is needed for
1213 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1215 * Don't do the full vcpu_info placement stuff until we have
1216 * the cpu_possible_mask and a non-dummy shared_info.
1218 xen_vcpu_info_reset(0);
1220 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1222 x86_init.resources.memory_setup = xen_memory_setup;
1223 x86_init.irqs.intr_mode_init = x86_init_noop;
1224 x86_init.oem.arch_setup = xen_arch_setup;
1225 x86_init.oem.banner = xen_banner;
1226 x86_init.hyper.init_platform = xen_pv_init_platform;
1227 x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1230 * Set up some pagetable state before starting to set any ptes.
1233 xen_setup_machphys_mapping();
1236 /* Prevent unwanted bits from being set in PTEs. */
1237 __supported_pte_mask &= ~_PAGE_GLOBAL;
1238 __default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1241 * Prevent page tables from being allocated in highmem, even
1242 * if CONFIG_HIGHPTE is enabled.
1244 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1247 xen_build_dynamic_phys_to_machine();
1250 * Set up kernel GDT and segment registers, mainly so that
1251 * -fstack-protector code can be executed.
1255 /* Work out if we support NX */
1256 get_cpu_cap(&boot_cpu_data);
1259 /* Determine virtual and physical address sizes */
1260 get_cpu_address_sizes(&boot_cpu_data);
1262 /* Let's presume PV guests always boot on vCPU with id 0. */
1263 per_cpu(xen_vcpu_id, 0) = 0;
1265 idt_setup_early_handler();
1267 xen_init_capabilities();
1269 #ifdef CONFIG_X86_LOCAL_APIC
1271 * set up the basic apic ops.
1276 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1277 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1278 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1281 machine_ops = xen_machine_ops;
1284 * The only reliable way to retain the initial address of the
1285 * percpu gdt_page is to remember it here, so we can go and
1286 * mark it RW later, when the initial percpu area is freed.
1288 xen_initial_gdt = &per_cpu(gdt_page, 0);
1292 #ifdef CONFIG_ACPI_NUMA
1294 * The pages we from Xen are not related to machine pages, so
1295 * any NUMA information the kernel tries to get from ACPI will
1296 * be meaningless. Prevent it from trying.
1300 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1302 local_irq_disable();
1303 early_boot_irqs_disabled = true;
1305 xen_raw_console_write("mapping kernel into physical memory\n");
1306 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1307 xen_start_info->nr_pages);
1308 xen_reserve_special_pages();
1310 /* keep using Xen gdt for now; no urgent need to change it */
1312 #ifdef CONFIG_X86_32
1313 pv_info.kernel_rpl = 1;
1314 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1315 pv_info.kernel_rpl = 0;
1317 pv_info.kernel_rpl = 0;
1319 /* set the limit of our address space */
1323 * We used to do this in xen_arch_setup, but that is too late
1324 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1325 * early_amd_init which pokes 0xcf8 port.
1328 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1330 xen_raw_printk("physdev_op failed %d\n", rc);
1332 #ifdef CONFIG_X86_32
1333 /* set up basic CPUID stuff */
1334 cpu_detect(&new_cpu_data);
1335 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1336 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1339 if (xen_start_info->mod_start) {
1340 if (xen_start_info->flags & SIF_MOD_START_PFN)
1341 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1343 initrd_start = __pa(xen_start_info->mod_start);
1346 /* Poke various useful things into boot_params */
1347 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1348 boot_params.hdr.ramdisk_image = initrd_start;
1349 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1350 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1351 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1353 if (!xen_initial_domain()) {
1354 add_preferred_console("xenboot", 0, NULL);
1356 x86_init.pci.arch_init = pci_xen_init;
1358 const struct dom0_vga_console_info *info =
1359 (void *)((char *)xen_start_info +
1360 xen_start_info->console.dom0.info_off);
1361 struct xen_platform_op op = {
1362 .cmd = XENPF_firmware_info,
1363 .interface_version = XENPF_INTERFACE_VERSION,
1364 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1367 x86_platform.set_legacy_features =
1368 xen_dom0_set_legacy_features;
1369 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1370 xen_start_info->console.domU.mfn = 0;
1371 xen_start_info->console.domU.evtchn = 0;
1373 if (HYPERVISOR_platform_op(&op) == 0)
1374 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1376 /* Make sure ACS will be enabled */
1379 xen_acpi_sleep_register();
1381 /* Avoid searching for BIOS MP tables */
1382 x86_init.mpparse.find_smp_config = x86_init_noop;
1383 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1385 xen_boot_params_init_edd();
1388 add_preferred_console("tty", 0, NULL);
1389 add_preferred_console("hvc", 0, NULL);
1392 /* PCI BIOS service won't work from a PV guest. */
1393 pci_probe &= ~PCI_PROBE_BIOS;
1395 xen_raw_console_write("about to get started...\n");
1397 /* We need this for printk timestamps */
1398 xen_setup_runstate_info(0);
1402 /* Start the world */
1403 #ifdef CONFIG_X86_32
1404 i386_start_kernel();
1406 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1407 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1411 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1415 if (per_cpu(xen_vcpu, cpu) == NULL)
1418 xen_setup_timer(cpu);
1420 rc = xen_smp_intr_init(cpu);
1422 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1427 rc = xen_smp_intr_init_pv(cpu);
1429 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1437 static int xen_cpu_dead_pv(unsigned int cpu)
1439 xen_smp_intr_free(cpu);
1440 xen_smp_intr_free_pv(cpu);
1442 xen_teardown_timer(cpu);
1447 static uint32_t __init xen_platform_pv(void)
1449 if (xen_pv_domain())
1450 return xen_cpuid_base();
1455 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1457 .detect = xen_platform_pv,
1458 .type = X86_HYPER_XEN_PV,
1459 .runtime.pin_vcpu = xen_pin_vcpu,