2 * Core of Xen paravirt_ops implementation.
4 * This file contains the xen_paravirt_ops structure itself, and the
6 * - privileged instructions
11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
14 #include <linux/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/export.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
48 #include <xen/hvc-console.h>
51 #include <asm/paravirt.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
65 #include <asm/pgalloc.h>
66 #include <asm/pgtable.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/stackprotector.h>
70 #include <asm/hypervisor.h>
71 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
77 #include <linux/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
87 #include "multicalls.h"
90 void *xen_initial_gdt;
92 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
94 static int xen_cpu_up_prepare_pv(unsigned int cpu);
95 static int xen_cpu_dead_pv(unsigned int cpu);
98 struct desc_struct desc[3];
102 * Updating the 3 TLS descriptors in the GDT on every task switch is
103 * surprisingly expensive so we avoid updating them if they haven't
104 * changed. Since Xen writes different descriptors than the one
105 * passed in the update_descriptor hypercall we keep shadow copies to
108 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
110 static void __init xen_banner(void)
112 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
113 struct xen_extraversion extra;
114 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
116 pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
117 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
118 version >> 16, version & 0xffff, extra.extraversion,
119 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
121 /* Check if running on Xen version (major, minor) or later */
123 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
125 unsigned int version;
130 version = HYPERVISOR_xen_version(XENVER_version, NULL);
131 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
132 ((version >> 16) > major))
137 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
138 static __read_mostly unsigned int cpuid_leaf5_edx_val;
140 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
141 unsigned int *cx, unsigned int *dx)
143 unsigned maskebx = ~0;
146 * Mask out inconvenient features, to try and disable as many
147 * unsupported kernel subsystems as possible.
150 case CPUID_MWAIT_LEAF:
151 /* Synthesize the values.. */
154 *cx = cpuid_leaf5_ecx_val;
155 *dx = cpuid_leaf5_edx_val;
159 /* Suppress extended topology stuff */
164 asm(XEN_EMULATE_PREFIX "cpuid"
169 : "0" (*ax), "2" (*cx));
173 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
175 static bool __init xen_check_mwait(void)
178 struct xen_platform_op op = {
179 .cmd = XENPF_set_processor_pminfo,
180 .u.set_pminfo.id = -1,
181 .u.set_pminfo.type = XEN_PM_PDC,
184 unsigned int ax, bx, cx, dx;
185 unsigned int mwait_mask;
187 /* We need to determine whether it is OK to expose the MWAIT
188 * capability to the kernel to harvest deeper than C3 states from ACPI
189 * _CST using the processor_harvest_xen.c module. For this to work, we
190 * need to gather the MWAIT_LEAF values (which the cstate.c code
191 * checks against). The hypervisor won't expose the MWAIT flag because
192 * it would break backwards compatibility; so we will find out directly
193 * from the hardware and hypercall.
195 if (!xen_initial_domain())
199 * When running under platform earlier than Xen4.2, do not expose
200 * mwait, to avoid the risk of loading native acpi pad driver
202 if (!xen_running_on_version_or_later(4, 2))
208 native_cpuid(&ax, &bx, &cx, &dx);
210 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
211 (1 << (X86_FEATURE_MWAIT % 32));
213 if ((cx & mwait_mask) != mwait_mask)
216 /* We need to emulate the MWAIT_LEAF and for that we need both
217 * ecx and edx. The hypercall provides only partial information.
220 ax = CPUID_MWAIT_LEAF;
225 native_cpuid(&ax, &bx, &cx, &dx);
227 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
228 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
230 buf[0] = ACPI_PDC_REVISION_ID;
232 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
234 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
236 if ((HYPERVISOR_platform_op(&op) == 0) &&
237 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
238 cpuid_leaf5_ecx_val = cx;
239 cpuid_leaf5_edx_val = dx;
247 static bool __init xen_check_xsave(void)
249 unsigned int cx, xsave_mask;
253 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
254 (1 << (X86_FEATURE_OSXSAVE % 32));
256 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
257 return (cx & xsave_mask) == xsave_mask;
260 static void __init xen_init_capabilities(void)
262 setup_force_cpu_cap(X86_FEATURE_XENPV);
263 setup_clear_cpu_cap(X86_FEATURE_DCA);
264 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
265 setup_clear_cpu_cap(X86_FEATURE_MTRR);
266 setup_clear_cpu_cap(X86_FEATURE_ACC);
267 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
269 if (!xen_initial_domain())
270 setup_clear_cpu_cap(X86_FEATURE_ACPI);
272 if (xen_check_mwait())
273 setup_force_cpu_cap(X86_FEATURE_MWAIT);
275 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
277 if (!xen_check_xsave()) {
278 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
279 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
283 static void xen_set_debugreg(int reg, unsigned long val)
285 HYPERVISOR_set_debugreg(reg, val);
288 static unsigned long xen_get_debugreg(int reg)
290 return HYPERVISOR_get_debugreg(reg);
293 static void xen_end_context_switch(struct task_struct *next)
296 paravirt_end_context_switch(next);
299 static unsigned long xen_store_tr(void)
305 * Set the page permissions for a particular virtual address. If the
306 * address is a vmalloc mapping (or other non-linear mapping), then
307 * find the linear mapping of the page and also set its protections to
310 static void set_aliased_prot(void *v, pgprot_t prot)
319 ptep = lookup_address((unsigned long)v, &level);
320 BUG_ON(ptep == NULL);
322 pfn = pte_pfn(*ptep);
323 page = pfn_to_page(pfn);
325 pte = pfn_pte(pfn, prot);
328 * Careful: update_va_mapping() will fail if the virtual address
329 * we're poking isn't populated in the page tables. We don't
330 * need to worry about the direct map (that's always in the page
331 * tables), but we need to be careful about vmap space. In
332 * particular, the top level page table can lazily propagate
333 * entries between processes, so if we've switched mms since we
334 * vmapped the target in the first place, we might not have the
335 * top-level page table entry populated.
337 * We disable preemption because we want the same mm active when
338 * we probe the target and when we issue the hypercall. We'll
339 * have the same nominal mm, but if we're a kernel thread, lazy
340 * mm dropping could change our pgd.
342 * Out of an abundance of caution, this uses __get_user() to fault
343 * in the target address just in case there's some obscure case
344 * in which the target address isn't readable.
349 probe_kernel_read(&dummy, v, 1);
351 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
354 if (!PageHighMem(page)) {
355 void *av = __va(PFN_PHYS(pfn));
358 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
366 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
368 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
372 * We need to mark the all aliases of the LDT pages RO. We
373 * don't need to call vm_flush_aliases(), though, since that's
374 * only responsible for flushing aliases out the TLBs, not the
375 * page tables, and Xen will flush the TLB for us if needed.
377 * To avoid confusing future readers: none of this is necessary
378 * to load the LDT. The hypervisor only checks this when the
379 * LDT is faulted in due to subsequent descriptor access.
382 for (i = 0; i < entries; i += entries_per_page)
383 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
386 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
388 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
391 for (i = 0; i < entries; i += entries_per_page)
392 set_aliased_prot(ldt + i, PAGE_KERNEL);
395 static void xen_set_ldt(const void *addr, unsigned entries)
397 struct mmuext_op *op;
398 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
400 trace_xen_cpu_set_ldt(addr, entries);
403 op->cmd = MMUEXT_SET_LDT;
404 op->arg1.linear_addr = (unsigned long)addr;
405 op->arg2.nr_ents = entries;
407 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
409 xen_mc_issue(PARAVIRT_LAZY_CPU);
412 static void xen_load_gdt(const struct desc_ptr *dtr)
414 unsigned long va = dtr->address;
415 unsigned int size = dtr->size + 1;
416 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
417 unsigned long frames[pages];
421 * A GDT can be up to 64k in size, which corresponds to 8192
422 * 8-byte entries, or 16 4k pages..
425 BUG_ON(size > 65536);
426 BUG_ON(va & ~PAGE_MASK);
428 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
431 unsigned long pfn, mfn;
435 * The GDT is per-cpu and is in the percpu data area.
436 * That can be virtually mapped, so we need to do a
437 * page-walk to get the underlying MFN for the
438 * hypercall. The page can also be in the kernel's
439 * linear range, so we need to RO that mapping too.
441 ptep = lookup_address(va, &level);
442 BUG_ON(ptep == NULL);
444 pfn = pte_pfn(*ptep);
445 mfn = pfn_to_mfn(pfn);
446 virt = __va(PFN_PHYS(pfn));
450 make_lowmem_page_readonly((void *)va);
451 make_lowmem_page_readonly(virt);
454 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
459 * load_gdt for early boot, when the gdt is only mapped once
461 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
463 unsigned long va = dtr->address;
464 unsigned int size = dtr->size + 1;
465 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
466 unsigned long frames[pages];
470 * A GDT can be up to 64k in size, which corresponds to 8192
471 * 8-byte entries, or 16 4k pages..
474 BUG_ON(size > 65536);
475 BUG_ON(va & ~PAGE_MASK);
477 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
479 unsigned long pfn, mfn;
481 pfn = virt_to_pfn(va);
482 mfn = pfn_to_mfn(pfn);
484 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
486 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
492 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
496 static inline bool desc_equal(const struct desc_struct *d1,
497 const struct desc_struct *d2)
499 return d1->a == d2->a && d1->b == d2->b;
502 static void load_TLS_descriptor(struct thread_struct *t,
503 unsigned int cpu, unsigned int i)
505 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
506 struct desc_struct *gdt;
508 struct multicall_space mc;
510 if (desc_equal(shadow, &t->tls_array[i]))
513 *shadow = t->tls_array[i];
515 gdt = get_cpu_gdt_rw(cpu);
516 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
517 mc = __xen_mc_entry(0);
519 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
522 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
525 * XXX sleazy hack: If we're being called in a lazy-cpu zone
526 * and lazy gs handling is enabled, it means we're in a
527 * context switch, and %gs has just been saved. This means we
528 * can zero it out to prevent faults on exit from the
529 * hypervisor if the next process has no %gs. Either way, it
530 * has been saved, and the new value will get loaded properly.
531 * This will go away as soon as Xen has been modified to not
532 * save/restore %gs for normal hypercalls.
534 * On x86_64, this hack is not used for %gs, because gs points
535 * to KERNEL_GS_BASE (and uses it for PDA references), so we
536 * must not zero %gs on x86_64
538 * For x86_64, we need to zero %fs, otherwise we may get an
539 * exception between the new %fs descriptor being loaded and
540 * %fs being effectively cleared at __switch_to().
542 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
552 load_TLS_descriptor(t, cpu, 0);
553 load_TLS_descriptor(t, cpu, 1);
554 load_TLS_descriptor(t, cpu, 2);
556 xen_mc_issue(PARAVIRT_LAZY_CPU);
560 static void xen_load_gs_index(unsigned int idx)
562 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
567 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
570 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
571 u64 entry = *(u64 *)ptr;
573 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
578 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
584 static int cvt_gate_to_trap(int vector, const gate_desc *val,
585 struct trap_info *info)
589 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
592 info->vector = vector;
594 addr = gate_offset(*val);
597 * Look for known traps using IST, and substitute them
598 * appropriately. The debugger ones are the only ones we care
599 * about. Xen will handle faults like double_fault,
600 * so we should never see them. Warn if
601 * there's an unexpected IST-using fault handler.
603 if (addr == (unsigned long)debug)
604 addr = (unsigned long)xen_debug;
605 else if (addr == (unsigned long)int3)
606 addr = (unsigned long)xen_int3;
607 else if (addr == (unsigned long)stack_segment)
608 addr = (unsigned long)xen_stack_segment;
609 else if (addr == (unsigned long)double_fault) {
610 /* Don't need to handle these */
612 #ifdef CONFIG_X86_MCE
613 } else if (addr == (unsigned long)machine_check) {
615 * when xen hypervisor inject vMCE to guest,
616 * use native mce handler to handle it
620 } else if (addr == (unsigned long)nmi)
622 * Use the native version as well.
626 /* Some other trap using IST? */
627 if (WARN_ON(val->ist != 0))
630 #endif /* CONFIG_X86_64 */
631 info->address = addr;
633 info->cs = gate_segment(*val);
634 info->flags = val->dpl;
635 /* interrupt gates clear IF */
636 if (val->type == GATE_INTERRUPT)
637 info->flags |= 1 << 2;
642 /* Locations of each CPU's IDT */
643 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
645 /* Set an IDT entry. If the entry is part of the current IDT, then
647 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
649 unsigned long p = (unsigned long)&dt[entrynum];
650 unsigned long start, end;
652 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
656 start = __this_cpu_read(idt_desc.address);
657 end = start + __this_cpu_read(idt_desc.size) + 1;
661 native_write_idt_entry(dt, entrynum, g);
663 if (p >= start && (p + 8) <= end) {
664 struct trap_info info[2];
668 if (cvt_gate_to_trap(entrynum, g, &info[0]))
669 if (HYPERVISOR_set_trap_table(info))
676 static void xen_convert_trap_info(const struct desc_ptr *desc,
677 struct trap_info *traps)
679 unsigned in, out, count;
681 count = (desc->size+1) / sizeof(gate_desc);
684 for (in = out = 0; in < count; in++) {
685 gate_desc *entry = (gate_desc *)(desc->address) + in;
687 if (cvt_gate_to_trap(in, entry, &traps[out]))
690 traps[out].address = 0;
693 void xen_copy_trap_info(struct trap_info *traps)
695 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
697 xen_convert_trap_info(desc, traps);
700 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
701 hold a spinlock to protect the static traps[] array (static because
702 it avoids allocation, and saves stack space). */
703 static void xen_load_idt(const struct desc_ptr *desc)
705 static DEFINE_SPINLOCK(lock);
706 static struct trap_info traps[257];
708 trace_xen_cpu_load_idt(desc);
712 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
714 xen_convert_trap_info(desc, traps);
717 if (HYPERVISOR_set_trap_table(traps))
723 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
724 they're handled differently. */
725 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
726 const void *desc, int type)
728 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
739 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
742 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
752 * Version of write_gdt_entry for use at early boot-time needed to
753 * update an entry as simply as possible.
755 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
756 const void *desc, int type)
758 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
767 xmaddr_t maddr = virt_to_machine(&dt[entry]);
769 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
770 dt[entry] = *(struct desc_struct *)desc;
776 static void xen_load_sp0(struct tss_struct *tss,
777 struct thread_struct *thread)
779 struct multicall_space mcs;
781 mcs = xen_mc_entry(0);
782 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
783 xen_mc_issue(PARAVIRT_LAZY_CPU);
784 tss->x86_tss.sp0 = thread->sp0;
787 void xen_set_iopl_mask(unsigned mask)
789 struct physdev_set_iopl set_iopl;
791 /* Force the change at ring 0. */
792 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
793 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
796 static void xen_io_delay(void)
800 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
802 static unsigned long xen_read_cr0(void)
804 unsigned long cr0 = this_cpu_read(xen_cr0_value);
806 if (unlikely(cr0 == 0)) {
807 cr0 = native_read_cr0();
808 this_cpu_write(xen_cr0_value, cr0);
814 static void xen_write_cr0(unsigned long cr0)
816 struct multicall_space mcs;
818 this_cpu_write(xen_cr0_value, cr0);
820 /* Only pay attention to cr0.TS; everything else is
822 mcs = xen_mc_entry(0);
824 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
826 xen_mc_issue(PARAVIRT_LAZY_CPU);
829 static void xen_write_cr4(unsigned long cr4)
831 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
833 native_write_cr4(cr4);
836 static inline unsigned long xen_read_cr8(void)
840 static inline void xen_write_cr8(unsigned long val)
846 static u64 xen_read_msr_safe(unsigned int msr, int *err)
850 if (pmu_msr_read(msr, &val, err))
853 val = native_read_msr_safe(msr, err);
855 case MSR_IA32_APICBASE:
856 #ifdef CONFIG_X86_X2APIC
857 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
859 val &= ~X2APIC_ENABLE;
865 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
876 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
877 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
878 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
881 base = ((u64)high << 32) | low;
882 if (HYPERVISOR_set_segment_base(which, base) != 0)
890 case MSR_SYSCALL_MASK:
891 case MSR_IA32_SYSENTER_CS:
892 case MSR_IA32_SYSENTER_ESP:
893 case MSR_IA32_SYSENTER_EIP:
894 /* Fast syscall setup is all done in hypercalls, so
895 these are all ignored. Stub them out here to stop
896 Xen console noise. */
900 if (!pmu_msr_write(msr, low, high, &ret))
901 ret = native_write_msr_safe(msr, low, high);
907 static u64 xen_read_msr(unsigned int msr)
910 * This will silently swallow a #GP from RDMSR. It may be worth
915 return xen_read_msr_safe(msr, &err);
918 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
921 * This will silently swallow a #GP from WRMSR. It may be worth
924 xen_write_msr_safe(msr, low, high);
927 void xen_setup_shared_info(void)
929 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
931 HYPERVISOR_shared_info =
932 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
935 /* In UP this is as good a place as any to set up shared info */
936 xen_setup_vcpu_info_placement();
939 xen_setup_mfn_list_list();
942 * Now that shared info is set up we can start using routines that
943 * point to pvclock area.
945 if (system_state == SYSTEM_BOOTING)
949 /* This is called once we have the cpu_possible_mask */
950 void xen_setup_vcpu_info_placement(void)
954 for_each_possible_cpu(cpu) {
955 /* Set up direct vCPU id mapping for PV guests. */
956 per_cpu(xen_vcpu_id, cpu) = cpu;
961 * xen_vcpu_setup managed to place the vcpu_info within the
962 * percpu area for all cpus, so make use of it.
964 if (xen_have_vcpu_info_placement) {
965 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
966 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
967 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
968 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
969 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
973 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
974 unsigned long addr, unsigned len)
976 char *start, *end, *reloc;
979 start = end = reloc = NULL;
981 #define SITE(op, x) \
982 case PARAVIRT_PATCH(op.x): \
983 if (xen_have_vcpu_info_placement) { \
984 start = (char *)xen_##x##_direct; \
985 end = xen_##x##_direct_end; \
986 reloc = xen_##x##_direct_reloc; \
991 SITE(pv_irq_ops, irq_enable);
992 SITE(pv_irq_ops, irq_disable);
993 SITE(pv_irq_ops, save_fl);
994 SITE(pv_irq_ops, restore_fl);
998 if (start == NULL || (end-start) > len)
1001 ret = paravirt_patch_insns(insnbuf, len, start, end);
1003 /* Note: because reloc is assigned from something that
1004 appears to be an array, gcc assumes it's non-null,
1005 but doesn't know its relationship with start and
1007 if (reloc > start && reloc < end) {
1008 int reloc_off = reloc - start;
1009 long *relocp = (long *)(insnbuf + reloc_off);
1010 long delta = start - (char *)addr;
1018 ret = paravirt_patch_default(type, clobbers, insnbuf,
1026 static const struct pv_info xen_info __initconst = {
1027 .shared_kernel_pmd = 0,
1029 #ifdef CONFIG_X86_64
1030 .extra_user_64bit_cs = FLAT_USER_CS64,
1035 static const struct pv_init_ops xen_init_ops __initconst = {
1039 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1042 .set_debugreg = xen_set_debugreg,
1043 .get_debugreg = xen_get_debugreg,
1045 .read_cr0 = xen_read_cr0,
1046 .write_cr0 = xen_write_cr0,
1048 .read_cr4 = native_read_cr4,
1049 .write_cr4 = xen_write_cr4,
1051 #ifdef CONFIG_X86_64
1052 .read_cr8 = xen_read_cr8,
1053 .write_cr8 = xen_write_cr8,
1056 .wbinvd = native_wbinvd,
1058 .read_msr = xen_read_msr,
1059 .write_msr = xen_write_msr,
1061 .read_msr_safe = xen_read_msr_safe,
1062 .write_msr_safe = xen_write_msr_safe,
1064 .read_pmc = xen_read_pmc,
1067 #ifdef CONFIG_X86_64
1068 .usergs_sysret64 = xen_sysret64,
1071 .load_tr_desc = paravirt_nop,
1072 .set_ldt = xen_set_ldt,
1073 .load_gdt = xen_load_gdt,
1074 .load_idt = xen_load_idt,
1075 .load_tls = xen_load_tls,
1076 #ifdef CONFIG_X86_64
1077 .load_gs_index = xen_load_gs_index,
1080 .alloc_ldt = xen_alloc_ldt,
1081 .free_ldt = xen_free_ldt,
1083 .store_idt = native_store_idt,
1084 .store_tr = xen_store_tr,
1086 .write_ldt_entry = xen_write_ldt_entry,
1087 .write_gdt_entry = xen_write_gdt_entry,
1088 .write_idt_entry = xen_write_idt_entry,
1089 .load_sp0 = xen_load_sp0,
1091 .set_iopl_mask = xen_set_iopl_mask,
1092 .io_delay = xen_io_delay,
1094 /* Xen takes care of %gs when switching to usermode for us */
1095 .swapgs = paravirt_nop,
1097 .start_context_switch = paravirt_start_context_switch,
1098 .end_context_switch = xen_end_context_switch,
1101 static void xen_restart(char *msg)
1103 xen_reboot(SHUTDOWN_reboot);
1106 static void xen_machine_halt(void)
1108 xen_reboot(SHUTDOWN_poweroff);
1111 static void xen_machine_power_off(void)
1115 xen_reboot(SHUTDOWN_poweroff);
1118 static void xen_crash_shutdown(struct pt_regs *regs)
1120 xen_reboot(SHUTDOWN_crash);
1123 static const struct machine_ops xen_machine_ops __initconst = {
1124 .restart = xen_restart,
1125 .halt = xen_machine_halt,
1126 .power_off = xen_machine_power_off,
1127 .shutdown = xen_machine_halt,
1128 .crash_shutdown = xen_crash_shutdown,
1129 .emergency_restart = xen_emergency_restart,
1132 static unsigned char xen_get_nmi_reason(void)
1134 unsigned char reason = 0;
1136 /* Construct a value which looks like it came from port 0x61. */
1137 if (test_bit(_XEN_NMIREASON_io_error,
1138 &HYPERVISOR_shared_info->arch.nmi_reason))
1139 reason |= NMI_REASON_IOCHK;
1140 if (test_bit(_XEN_NMIREASON_pci_serr,
1141 &HYPERVISOR_shared_info->arch.nmi_reason))
1142 reason |= NMI_REASON_SERR;
1147 static void __init xen_boot_params_init_edd(void)
1149 #if IS_ENABLED(CONFIG_EDD)
1150 struct xen_platform_op op;
1151 struct edd_info *edd_info;
1156 edd_info = boot_params.eddbuf;
1157 mbr_signature = boot_params.edd_mbr_sig_buffer;
1159 op.cmd = XENPF_firmware_info;
1161 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1162 for (nr = 0; nr < EDDMAXNR; nr++) {
1163 struct edd_info *info = edd_info + nr;
1165 op.u.firmware_info.index = nr;
1166 info->params.length = sizeof(info->params);
1167 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1169 ret = HYPERVISOR_platform_op(&op);
1173 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1176 C(interface_support);
1177 C(legacy_max_cylinder);
1179 C(legacy_sectors_per_track);
1182 boot_params.eddbuf_entries = nr;
1184 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1185 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1186 op.u.firmware_info.index = nr;
1187 ret = HYPERVISOR_platform_op(&op);
1190 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1192 boot_params.edd_mbr_sig_buf_entries = nr;
1197 * Set up the GDT and segment registers for -fstack-protector. Until
1198 * we do this, we have to be careful not to call any stack-protected
1199 * function, which is most of the kernel.
1201 static void xen_setup_gdt(int cpu)
1203 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1204 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1206 setup_stack_canary_segment(0);
1207 switch_to_new_gdt(0);
1209 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1210 pv_cpu_ops.load_gdt = xen_load_gdt;
1213 static void __init xen_dom0_set_legacy_features(void)
1215 x86_platform.legacy.rtc = 1;
1218 /* First C function to be called on Xen boot */
1219 asmlinkage __visible void __init xen_start_kernel(void)
1221 struct physdev_set_iopl set_iopl;
1222 unsigned long initrd_start = 0;
1225 if (!xen_start_info)
1228 xen_domain_type = XEN_PV_DOMAIN;
1230 xen_setup_features();
1232 xen_setup_machphys_mapping();
1234 /* Install Xen paravirt ops */
1236 pv_init_ops = xen_init_ops;
1237 pv_cpu_ops = xen_cpu_ops;
1239 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1241 x86_init.resources.memory_setup = xen_memory_setup;
1242 x86_init.oem.arch_setup = xen_arch_setup;
1243 x86_init.oem.banner = xen_banner;
1246 * Set up some pagetable state before starting to set any ptes.
1251 /* Prevent unwanted bits from being set in PTEs. */
1252 __supported_pte_mask &= ~_PAGE_GLOBAL;
1255 * Prevent page tables from being allocated in highmem, even
1256 * if CONFIG_HIGHPTE is enabled.
1258 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1260 /* Work out if we support NX */
1264 xen_build_dynamic_phys_to_machine();
1267 * Set up kernel GDT and segment registers, mainly so that
1268 * -fstack-protector code can be executed.
1273 xen_init_capabilities();
1275 #ifdef CONFIG_X86_LOCAL_APIC
1277 * set up the basic apic ops.
1282 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1283 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1284 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1287 machine_ops = xen_machine_ops;
1290 * The only reliable way to retain the initial address of the
1291 * percpu gdt_page is to remember it here, so we can go and
1292 * mark it RW later, when the initial percpu area is freed.
1294 xen_initial_gdt = &per_cpu(gdt_page, 0);
1298 #ifdef CONFIG_ACPI_NUMA
1300 * The pages we from Xen are not related to machine pages, so
1301 * any NUMA information the kernel tries to get from ACPI will
1302 * be meaningless. Prevent it from trying.
1306 /* Let's presume PV guests always boot on vCPU with id 0. */
1307 per_cpu(xen_vcpu_id, 0) = 0;
1310 * Setup xen_vcpu early because start_kernel needs it for
1311 * local_irq_disable(), irqs_disabled().
1313 * Don't do the full vcpu_info placement stuff until we have
1314 * the cpu_possible_mask and a non-dummy shared_info.
1316 xen_vcpu_info_reset(0);
1318 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1320 local_irq_disable();
1321 early_boot_irqs_disabled = true;
1323 xen_raw_console_write("mapping kernel into physical memory\n");
1324 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1325 xen_start_info->nr_pages);
1326 xen_reserve_special_pages();
1328 /* keep using Xen gdt for now; no urgent need to change it */
1330 #ifdef CONFIG_X86_32
1331 pv_info.kernel_rpl = 1;
1332 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1333 pv_info.kernel_rpl = 0;
1335 pv_info.kernel_rpl = 0;
1337 /* set the limit of our address space */
1341 * We used to do this in xen_arch_setup, but that is too late
1342 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1343 * early_amd_init which pokes 0xcf8 port.
1346 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1348 xen_raw_printk("physdev_op failed %d\n", rc);
1350 #ifdef CONFIG_X86_32
1351 /* set up basic CPUID stuff */
1352 cpu_detect(&new_cpu_data);
1353 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1354 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1357 if (xen_start_info->mod_start) {
1358 if (xen_start_info->flags & SIF_MOD_START_PFN)
1359 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1361 initrd_start = __pa(xen_start_info->mod_start);
1364 /* Poke various useful things into boot_params */
1365 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1366 boot_params.hdr.ramdisk_image = initrd_start;
1367 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1368 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1369 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1371 if (!xen_initial_domain()) {
1372 add_preferred_console("xenboot", 0, NULL);
1373 add_preferred_console("tty", 0, NULL);
1374 add_preferred_console("hvc", 0, NULL);
1376 x86_init.pci.arch_init = pci_xen_init;
1378 const struct dom0_vga_console_info *info =
1379 (void *)((char *)xen_start_info +
1380 xen_start_info->console.dom0.info_off);
1381 struct xen_platform_op op = {
1382 .cmd = XENPF_firmware_info,
1383 .interface_version = XENPF_INTERFACE_VERSION,
1384 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1387 x86_platform.set_legacy_features =
1388 xen_dom0_set_legacy_features;
1389 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1390 xen_start_info->console.domU.mfn = 0;
1391 xen_start_info->console.domU.evtchn = 0;
1393 if (HYPERVISOR_platform_op(&op) == 0)
1394 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1396 /* Make sure ACS will be enabled */
1399 xen_acpi_sleep_register();
1401 /* Avoid searching for BIOS MP tables */
1402 x86_init.mpparse.find_smp_config = x86_init_noop;
1403 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1405 xen_boot_params_init_edd();
1408 /* PCI BIOS service won't work from a PV guest. */
1409 pci_probe &= ~PCI_PROBE_BIOS;
1411 xen_raw_console_write("about to get started...\n");
1413 /* We need this for printk timestamps */
1414 xen_setup_runstate_info(0);
1418 /* Start the world */
1419 #ifdef CONFIG_X86_32
1420 i386_start_kernel();
1422 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1423 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1427 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1431 xen_setup_timer(cpu);
1433 rc = xen_smp_intr_init(cpu);
1435 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1440 rc = xen_smp_intr_init_pv(cpu);
1442 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1450 static int xen_cpu_dead_pv(unsigned int cpu)
1452 xen_smp_intr_free(cpu);
1453 xen_smp_intr_free_pv(cpu);
1455 xen_teardown_timer(cpu);
1460 static uint32_t __init xen_platform_pv(void)
1462 if (xen_pv_domain())
1463 return xen_cpuid_base();
1468 const struct hypervisor_x86 x86_hyper_xen_pv = {
1470 .detect = xen_platform_pv,
1471 .pin_vcpu = xen_pin_vcpu,
1473 EXPORT_SYMBOL(x86_hyper_xen_pv);