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);
111 * On restore, set the vcpu placement up again.
112 * If it fails, then we're in a bad state, since
113 * we can't back out from using it...
115 void xen_vcpu_restore(void)
119 for_each_possible_cpu(cpu) {
120 bool other_cpu = (cpu != smp_processor_id());
121 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
124 if (other_cpu && is_up &&
125 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
128 xen_setup_runstate_info(cpu);
130 if (xen_have_vcpu_info_placement)
133 if (other_cpu && is_up &&
134 HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
139 static void __init xen_banner(void)
141 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
142 struct xen_extraversion extra;
143 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
145 pr_info("Booting paravirtualized kernel %son %s\n",
146 xen_feature(XENFEAT_auto_translated_physmap) ?
147 "with PVH extensions " : "", pv_info.name);
148 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
149 version >> 16, version & 0xffff, extra.extraversion,
150 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
152 /* Check if running on Xen version (major, minor) or later */
154 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
156 unsigned int version;
161 version = HYPERVISOR_xen_version(XENVER_version, NULL);
162 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
163 ((version >> 16) > major))
168 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
169 static __read_mostly unsigned int cpuid_leaf5_edx_val;
171 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
172 unsigned int *cx, unsigned int *dx)
174 unsigned maskebx = ~0;
177 * Mask out inconvenient features, to try and disable as many
178 * unsupported kernel subsystems as possible.
181 case CPUID_MWAIT_LEAF:
182 /* Synthesize the values.. */
185 *cx = cpuid_leaf5_ecx_val;
186 *dx = cpuid_leaf5_edx_val;
190 /* Suppress extended topology stuff */
195 asm(XEN_EMULATE_PREFIX "cpuid"
200 : "0" (*ax), "2" (*cx));
204 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
206 static bool __init xen_check_mwait(void)
209 struct xen_platform_op op = {
210 .cmd = XENPF_set_processor_pminfo,
211 .u.set_pminfo.id = -1,
212 .u.set_pminfo.type = XEN_PM_PDC,
215 unsigned int ax, bx, cx, dx;
216 unsigned int mwait_mask;
218 /* We need to determine whether it is OK to expose the MWAIT
219 * capability to the kernel to harvest deeper than C3 states from ACPI
220 * _CST using the processor_harvest_xen.c module. For this to work, we
221 * need to gather the MWAIT_LEAF values (which the cstate.c code
222 * checks against). The hypervisor won't expose the MWAIT flag because
223 * it would break backwards compatibility; so we will find out directly
224 * from the hardware and hypercall.
226 if (!xen_initial_domain())
230 * When running under platform earlier than Xen4.2, do not expose
231 * mwait, to avoid the risk of loading native acpi pad driver
233 if (!xen_running_on_version_or_later(4, 2))
239 native_cpuid(&ax, &bx, &cx, &dx);
241 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
242 (1 << (X86_FEATURE_MWAIT % 32));
244 if ((cx & mwait_mask) != mwait_mask)
247 /* We need to emulate the MWAIT_LEAF and for that we need both
248 * ecx and edx. The hypercall provides only partial information.
251 ax = CPUID_MWAIT_LEAF;
256 native_cpuid(&ax, &bx, &cx, &dx);
258 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
259 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
261 buf[0] = ACPI_PDC_REVISION_ID;
263 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
265 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
267 if ((HYPERVISOR_platform_op(&op) == 0) &&
268 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
269 cpuid_leaf5_ecx_val = cx;
270 cpuid_leaf5_edx_val = dx;
278 static bool __init xen_check_xsave(void)
280 unsigned int cx, xsave_mask;
284 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
285 (1 << (X86_FEATURE_OSXSAVE % 32));
287 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
288 return (cx & xsave_mask) == xsave_mask;
291 static void __init xen_init_capabilities(void)
293 setup_force_cpu_cap(X86_FEATURE_XENPV);
294 setup_clear_cpu_cap(X86_FEATURE_DCA);
295 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
296 setup_clear_cpu_cap(X86_FEATURE_MTRR);
297 setup_clear_cpu_cap(X86_FEATURE_ACC);
298 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
300 if (!xen_initial_domain())
301 setup_clear_cpu_cap(X86_FEATURE_ACPI);
303 if (xen_check_mwait())
304 setup_force_cpu_cap(X86_FEATURE_MWAIT);
306 setup_clear_cpu_cap(X86_FEATURE_MWAIT);
308 if (!xen_check_xsave()) {
309 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
310 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
314 static void xen_set_debugreg(int reg, unsigned long val)
316 HYPERVISOR_set_debugreg(reg, val);
319 static unsigned long xen_get_debugreg(int reg)
321 return HYPERVISOR_get_debugreg(reg);
324 static void xen_end_context_switch(struct task_struct *next)
327 paravirt_end_context_switch(next);
330 static unsigned long xen_store_tr(void)
336 * Set the page permissions for a particular virtual address. If the
337 * address is a vmalloc mapping (or other non-linear mapping), then
338 * find the linear mapping of the page and also set its protections to
341 static void set_aliased_prot(void *v, pgprot_t prot)
350 ptep = lookup_address((unsigned long)v, &level);
351 BUG_ON(ptep == NULL);
353 pfn = pte_pfn(*ptep);
354 page = pfn_to_page(pfn);
356 pte = pfn_pte(pfn, prot);
359 * Careful: update_va_mapping() will fail if the virtual address
360 * we're poking isn't populated in the page tables. We don't
361 * need to worry about the direct map (that's always in the page
362 * tables), but we need to be careful about vmap space. In
363 * particular, the top level page table can lazily propagate
364 * entries between processes, so if we've switched mms since we
365 * vmapped the target in the first place, we might not have the
366 * top-level page table entry populated.
368 * We disable preemption because we want the same mm active when
369 * we probe the target and when we issue the hypercall. We'll
370 * have the same nominal mm, but if we're a kernel thread, lazy
371 * mm dropping could change our pgd.
373 * Out of an abundance of caution, this uses __get_user() to fault
374 * in the target address just in case there's some obscure case
375 * in which the target address isn't readable.
380 probe_kernel_read(&dummy, v, 1);
382 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
385 if (!PageHighMem(page)) {
386 void *av = __va(PFN_PHYS(pfn));
389 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
397 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
399 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
403 * We need to mark the all aliases of the LDT pages RO. We
404 * don't need to call vm_flush_aliases(), though, since that's
405 * only responsible for flushing aliases out the TLBs, not the
406 * page tables, and Xen will flush the TLB for us if needed.
408 * To avoid confusing future readers: none of this is necessary
409 * to load the LDT. The hypervisor only checks this when the
410 * LDT is faulted in due to subsequent descriptor access.
413 for (i = 0; i < entries; i += entries_per_page)
414 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
417 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
419 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
422 for (i = 0; i < entries; i += entries_per_page)
423 set_aliased_prot(ldt + i, PAGE_KERNEL);
426 static void xen_set_ldt(const void *addr, unsigned entries)
428 struct mmuext_op *op;
429 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
431 trace_xen_cpu_set_ldt(addr, entries);
434 op->cmd = MMUEXT_SET_LDT;
435 op->arg1.linear_addr = (unsigned long)addr;
436 op->arg2.nr_ents = entries;
438 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
440 xen_mc_issue(PARAVIRT_LAZY_CPU);
443 static void xen_load_gdt(const struct desc_ptr *dtr)
445 unsigned long va = dtr->address;
446 unsigned int size = dtr->size + 1;
447 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
448 unsigned long frames[pages];
452 * A GDT can be up to 64k in size, which corresponds to 8192
453 * 8-byte entries, or 16 4k pages..
456 BUG_ON(size > 65536);
457 BUG_ON(va & ~PAGE_MASK);
459 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
462 unsigned long pfn, mfn;
466 * The GDT is per-cpu and is in the percpu data area.
467 * That can be virtually mapped, so we need to do a
468 * page-walk to get the underlying MFN for the
469 * hypercall. The page can also be in the kernel's
470 * linear range, so we need to RO that mapping too.
472 ptep = lookup_address(va, &level);
473 BUG_ON(ptep == NULL);
475 pfn = pte_pfn(*ptep);
476 mfn = pfn_to_mfn(pfn);
477 virt = __va(PFN_PHYS(pfn));
481 make_lowmem_page_readonly((void *)va);
482 make_lowmem_page_readonly(virt);
485 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
490 * load_gdt for early boot, when the gdt is only mapped once
492 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
494 unsigned long va = dtr->address;
495 unsigned int size = dtr->size + 1;
496 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
497 unsigned long frames[pages];
501 * A GDT can be up to 64k in size, which corresponds to 8192
502 * 8-byte entries, or 16 4k pages..
505 BUG_ON(size > 65536);
506 BUG_ON(va & ~PAGE_MASK);
508 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
510 unsigned long pfn, mfn;
512 pfn = virt_to_pfn(va);
513 mfn = pfn_to_mfn(pfn);
515 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
517 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
523 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
527 static inline bool desc_equal(const struct desc_struct *d1,
528 const struct desc_struct *d2)
530 return d1->a == d2->a && d1->b == d2->b;
533 static void load_TLS_descriptor(struct thread_struct *t,
534 unsigned int cpu, unsigned int i)
536 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
537 struct desc_struct *gdt;
539 struct multicall_space mc;
541 if (desc_equal(shadow, &t->tls_array[i]))
544 *shadow = t->tls_array[i];
546 gdt = get_cpu_gdt_rw(cpu);
547 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
548 mc = __xen_mc_entry(0);
550 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
553 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
556 * XXX sleazy hack: If we're being called in a lazy-cpu zone
557 * and lazy gs handling is enabled, it means we're in a
558 * context switch, and %gs has just been saved. This means we
559 * can zero it out to prevent faults on exit from the
560 * hypervisor if the next process has no %gs. Either way, it
561 * has been saved, and the new value will get loaded properly.
562 * This will go away as soon as Xen has been modified to not
563 * save/restore %gs for normal hypercalls.
565 * On x86_64, this hack is not used for %gs, because gs points
566 * to KERNEL_GS_BASE (and uses it for PDA references), so we
567 * must not zero %gs on x86_64
569 * For x86_64, we need to zero %fs, otherwise we may get an
570 * exception between the new %fs descriptor being loaded and
571 * %fs being effectively cleared at __switch_to().
573 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
583 load_TLS_descriptor(t, cpu, 0);
584 load_TLS_descriptor(t, cpu, 1);
585 load_TLS_descriptor(t, cpu, 2);
587 xen_mc_issue(PARAVIRT_LAZY_CPU);
591 static void xen_load_gs_index(unsigned int idx)
593 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
598 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
601 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
602 u64 entry = *(u64 *)ptr;
604 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
609 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
615 static int cvt_gate_to_trap(int vector, const gate_desc *val,
616 struct trap_info *info)
620 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
623 info->vector = vector;
625 addr = gate_offset(*val);
628 * Look for known traps using IST, and substitute them
629 * appropriately. The debugger ones are the only ones we care
630 * about. Xen will handle faults like double_fault,
631 * so we should never see them. Warn if
632 * there's an unexpected IST-using fault handler.
634 if (addr == (unsigned long)debug)
635 addr = (unsigned long)xen_debug;
636 else if (addr == (unsigned long)int3)
637 addr = (unsigned long)xen_int3;
638 else if (addr == (unsigned long)stack_segment)
639 addr = (unsigned long)xen_stack_segment;
640 else if (addr == (unsigned long)double_fault) {
641 /* Don't need to handle these */
643 #ifdef CONFIG_X86_MCE
644 } else if (addr == (unsigned long)machine_check) {
646 * when xen hypervisor inject vMCE to guest,
647 * use native mce handler to handle it
651 } else if (addr == (unsigned long)nmi)
653 * Use the native version as well.
657 /* Some other trap using IST? */
658 if (WARN_ON(val->ist != 0))
661 #endif /* CONFIG_X86_64 */
662 info->address = addr;
664 info->cs = gate_segment(*val);
665 info->flags = val->dpl;
666 /* interrupt gates clear IF */
667 if (val->type == GATE_INTERRUPT)
668 info->flags |= 1 << 2;
673 /* Locations of each CPU's IDT */
674 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
676 /* Set an IDT entry. If the entry is part of the current IDT, then
678 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
680 unsigned long p = (unsigned long)&dt[entrynum];
681 unsigned long start, end;
683 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
687 start = __this_cpu_read(idt_desc.address);
688 end = start + __this_cpu_read(idt_desc.size) + 1;
692 native_write_idt_entry(dt, entrynum, g);
694 if (p >= start && (p + 8) <= end) {
695 struct trap_info info[2];
699 if (cvt_gate_to_trap(entrynum, g, &info[0]))
700 if (HYPERVISOR_set_trap_table(info))
707 static void xen_convert_trap_info(const struct desc_ptr *desc,
708 struct trap_info *traps)
710 unsigned in, out, count;
712 count = (desc->size+1) / sizeof(gate_desc);
715 for (in = out = 0; in < count; in++) {
716 gate_desc *entry = (gate_desc *)(desc->address) + in;
718 if (cvt_gate_to_trap(in, entry, &traps[out]))
721 traps[out].address = 0;
724 void xen_copy_trap_info(struct trap_info *traps)
726 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
728 xen_convert_trap_info(desc, traps);
731 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
732 hold a spinlock to protect the static traps[] array (static because
733 it avoids allocation, and saves stack space). */
734 static void xen_load_idt(const struct desc_ptr *desc)
736 static DEFINE_SPINLOCK(lock);
737 static struct trap_info traps[257];
739 trace_xen_cpu_load_idt(desc);
743 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
745 xen_convert_trap_info(desc, traps);
748 if (HYPERVISOR_set_trap_table(traps))
754 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
755 they're handled differently. */
756 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
757 const void *desc, int type)
759 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
770 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
773 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
783 * Version of write_gdt_entry for use at early boot-time needed to
784 * update an entry as simply as possible.
786 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
787 const void *desc, int type)
789 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
798 xmaddr_t maddr = virt_to_machine(&dt[entry]);
800 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
801 dt[entry] = *(struct desc_struct *)desc;
807 static void xen_load_sp0(struct tss_struct *tss,
808 struct thread_struct *thread)
810 struct multicall_space mcs;
812 mcs = xen_mc_entry(0);
813 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
814 xen_mc_issue(PARAVIRT_LAZY_CPU);
815 tss->x86_tss.sp0 = thread->sp0;
818 void xen_set_iopl_mask(unsigned mask)
820 struct physdev_set_iopl set_iopl;
822 /* Force the change at ring 0. */
823 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
824 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
827 static void xen_io_delay(void)
831 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
833 static unsigned long xen_read_cr0(void)
835 unsigned long cr0 = this_cpu_read(xen_cr0_value);
837 if (unlikely(cr0 == 0)) {
838 cr0 = native_read_cr0();
839 this_cpu_write(xen_cr0_value, cr0);
845 static void xen_write_cr0(unsigned long cr0)
847 struct multicall_space mcs;
849 this_cpu_write(xen_cr0_value, cr0);
851 /* Only pay attention to cr0.TS; everything else is
853 mcs = xen_mc_entry(0);
855 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
857 xen_mc_issue(PARAVIRT_LAZY_CPU);
860 static void xen_write_cr4(unsigned long cr4)
862 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
864 native_write_cr4(cr4);
867 static inline unsigned long xen_read_cr8(void)
871 static inline void xen_write_cr8(unsigned long val)
877 static u64 xen_read_msr_safe(unsigned int msr, int *err)
881 if (pmu_msr_read(msr, &val, err))
884 val = native_read_msr_safe(msr, err);
886 case MSR_IA32_APICBASE:
887 #ifdef CONFIG_X86_X2APIC
888 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
890 val &= ~X2APIC_ENABLE;
896 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
907 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
908 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
909 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
912 base = ((u64)high << 32) | low;
913 if (HYPERVISOR_set_segment_base(which, base) != 0)
921 case MSR_SYSCALL_MASK:
922 case MSR_IA32_SYSENTER_CS:
923 case MSR_IA32_SYSENTER_ESP:
924 case MSR_IA32_SYSENTER_EIP:
925 /* Fast syscall setup is all done in hypercalls, so
926 these are all ignored. Stub them out here to stop
927 Xen console noise. */
931 if (!pmu_msr_write(msr, low, high, &ret))
932 ret = native_write_msr_safe(msr, low, high);
938 static u64 xen_read_msr(unsigned int msr)
941 * This will silently swallow a #GP from RDMSR. It may be worth
946 return xen_read_msr_safe(msr, &err);
949 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
952 * This will silently swallow a #GP from WRMSR. It may be worth
955 xen_write_msr_safe(msr, low, high);
958 void xen_setup_shared_info(void)
960 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
961 set_fixmap(FIX_PARAVIRT_BOOTMAP,
962 xen_start_info->shared_info);
964 HYPERVISOR_shared_info =
965 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
967 HYPERVISOR_shared_info =
968 (struct shared_info *)__va(xen_start_info->shared_info);
971 /* In UP this is as good a place as any to set up shared info */
972 xen_setup_vcpu_info_placement();
975 xen_setup_mfn_list_list();
978 * Now that shared info is set up we can start using routines that
979 * point to pvclock area.
981 if (system_state == SYSTEM_BOOTING)
985 /* This is called once we have the cpu_possible_mask */
986 void xen_setup_vcpu_info_placement(void)
990 for_each_possible_cpu(cpu) {
991 /* Set up direct vCPU id mapping for PV guests. */
992 per_cpu(xen_vcpu_id, cpu) = cpu;
997 * xen_vcpu_setup managed to place the vcpu_info within the
998 * percpu area for all cpus, so make use of it.
1000 if (xen_have_vcpu_info_placement) {
1001 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1002 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1003 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1004 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1005 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1009 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1010 unsigned long addr, unsigned len)
1012 char *start, *end, *reloc;
1015 start = end = reloc = NULL;
1017 #define SITE(op, x) \
1018 case PARAVIRT_PATCH(op.x): \
1019 if (xen_have_vcpu_info_placement) { \
1020 start = (char *)xen_##x##_direct; \
1021 end = xen_##x##_direct_end; \
1022 reloc = xen_##x##_direct_reloc; \
1027 SITE(pv_irq_ops, irq_enable);
1028 SITE(pv_irq_ops, irq_disable);
1029 SITE(pv_irq_ops, save_fl);
1030 SITE(pv_irq_ops, restore_fl);
1034 if (start == NULL || (end-start) > len)
1037 ret = paravirt_patch_insns(insnbuf, len, start, end);
1039 /* Note: because reloc is assigned from something that
1040 appears to be an array, gcc assumes it's non-null,
1041 but doesn't know its relationship with start and
1043 if (reloc > start && reloc < end) {
1044 int reloc_off = reloc - start;
1045 long *relocp = (long *)(insnbuf + reloc_off);
1046 long delta = start - (char *)addr;
1054 ret = paravirt_patch_default(type, clobbers, insnbuf,
1062 static const struct pv_info xen_info __initconst = {
1063 .shared_kernel_pmd = 0,
1065 #ifdef CONFIG_X86_64
1066 .extra_user_64bit_cs = FLAT_USER_CS64,
1071 static const struct pv_init_ops xen_init_ops __initconst = {
1075 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1078 .set_debugreg = xen_set_debugreg,
1079 .get_debugreg = xen_get_debugreg,
1081 .read_cr0 = xen_read_cr0,
1082 .write_cr0 = xen_write_cr0,
1084 .read_cr4 = native_read_cr4,
1085 .write_cr4 = xen_write_cr4,
1087 #ifdef CONFIG_X86_64
1088 .read_cr8 = xen_read_cr8,
1089 .write_cr8 = xen_write_cr8,
1092 .wbinvd = native_wbinvd,
1094 .read_msr = xen_read_msr,
1095 .write_msr = xen_write_msr,
1097 .read_msr_safe = xen_read_msr_safe,
1098 .write_msr_safe = xen_write_msr_safe,
1100 .read_pmc = xen_read_pmc,
1103 #ifdef CONFIG_X86_64
1104 .usergs_sysret64 = xen_sysret64,
1107 .load_tr_desc = paravirt_nop,
1108 .set_ldt = xen_set_ldt,
1109 .load_gdt = xen_load_gdt,
1110 .load_idt = xen_load_idt,
1111 .load_tls = xen_load_tls,
1112 #ifdef CONFIG_X86_64
1113 .load_gs_index = xen_load_gs_index,
1116 .alloc_ldt = xen_alloc_ldt,
1117 .free_ldt = xen_free_ldt,
1119 .store_idt = native_store_idt,
1120 .store_tr = xen_store_tr,
1122 .write_ldt_entry = xen_write_ldt_entry,
1123 .write_gdt_entry = xen_write_gdt_entry,
1124 .write_idt_entry = xen_write_idt_entry,
1125 .load_sp0 = xen_load_sp0,
1127 .set_iopl_mask = xen_set_iopl_mask,
1128 .io_delay = xen_io_delay,
1130 /* Xen takes care of %gs when switching to usermode for us */
1131 .swapgs = paravirt_nop,
1133 .start_context_switch = paravirt_start_context_switch,
1134 .end_context_switch = xen_end_context_switch,
1137 static void xen_restart(char *msg)
1139 xen_reboot(SHUTDOWN_reboot);
1142 static void xen_machine_halt(void)
1144 xen_reboot(SHUTDOWN_poweroff);
1147 static void xen_machine_power_off(void)
1151 xen_reboot(SHUTDOWN_poweroff);
1154 static void xen_crash_shutdown(struct pt_regs *regs)
1156 xen_reboot(SHUTDOWN_crash);
1159 static const struct machine_ops xen_machine_ops __initconst = {
1160 .restart = xen_restart,
1161 .halt = xen_machine_halt,
1162 .power_off = xen_machine_power_off,
1163 .shutdown = xen_machine_halt,
1164 .crash_shutdown = xen_crash_shutdown,
1165 .emergency_restart = xen_emergency_restart,
1168 static unsigned char xen_get_nmi_reason(void)
1170 unsigned char reason = 0;
1172 /* Construct a value which looks like it came from port 0x61. */
1173 if (test_bit(_XEN_NMIREASON_io_error,
1174 &HYPERVISOR_shared_info->arch.nmi_reason))
1175 reason |= NMI_REASON_IOCHK;
1176 if (test_bit(_XEN_NMIREASON_pci_serr,
1177 &HYPERVISOR_shared_info->arch.nmi_reason))
1178 reason |= NMI_REASON_SERR;
1183 static void __init xen_boot_params_init_edd(void)
1185 #if IS_ENABLED(CONFIG_EDD)
1186 struct xen_platform_op op;
1187 struct edd_info *edd_info;
1192 edd_info = boot_params.eddbuf;
1193 mbr_signature = boot_params.edd_mbr_sig_buffer;
1195 op.cmd = XENPF_firmware_info;
1197 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1198 for (nr = 0; nr < EDDMAXNR; nr++) {
1199 struct edd_info *info = edd_info + nr;
1201 op.u.firmware_info.index = nr;
1202 info->params.length = sizeof(info->params);
1203 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1205 ret = HYPERVISOR_platform_op(&op);
1209 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1212 C(interface_support);
1213 C(legacy_max_cylinder);
1215 C(legacy_sectors_per_track);
1218 boot_params.eddbuf_entries = nr;
1220 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1221 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1222 op.u.firmware_info.index = nr;
1223 ret = HYPERVISOR_platform_op(&op);
1226 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1228 boot_params.edd_mbr_sig_buf_entries = nr;
1233 * Set up the GDT and segment registers for -fstack-protector. Until
1234 * we do this, we have to be careful not to call any stack-protected
1235 * function, which is most of the kernel.
1237 static void xen_setup_gdt(int cpu)
1239 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1240 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1242 setup_stack_canary_segment(0);
1243 switch_to_new_gdt(0);
1245 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1246 pv_cpu_ops.load_gdt = xen_load_gdt;
1249 static void __init xen_dom0_set_legacy_features(void)
1251 x86_platform.legacy.rtc = 1;
1254 /* First C function to be called on Xen boot */
1255 asmlinkage __visible void __init xen_start_kernel(void)
1257 struct physdev_set_iopl set_iopl;
1258 unsigned long initrd_start = 0;
1261 if (!xen_start_info)
1264 xen_domain_type = XEN_PV_DOMAIN;
1266 xen_setup_features();
1268 xen_setup_machphys_mapping();
1270 /* Install Xen paravirt ops */
1272 pv_init_ops = xen_init_ops;
1273 pv_cpu_ops = xen_cpu_ops;
1275 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1277 x86_init.resources.memory_setup = xen_memory_setup;
1278 x86_init.oem.arch_setup = xen_arch_setup;
1279 x86_init.oem.banner = xen_banner;
1282 * Set up some pagetable state before starting to set any ptes.
1287 /* Prevent unwanted bits from being set in PTEs. */
1288 __supported_pte_mask &= ~_PAGE_GLOBAL;
1291 * Prevent page tables from being allocated in highmem, even
1292 * if CONFIG_HIGHPTE is enabled.
1294 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1296 /* Work out if we support NX */
1300 xen_build_dynamic_phys_to_machine();
1303 * Set up kernel GDT and segment registers, mainly so that
1304 * -fstack-protector code can be executed.
1309 xen_init_capabilities();
1311 #ifdef CONFIG_X86_LOCAL_APIC
1313 * set up the basic apic ops.
1318 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1319 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1320 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1323 machine_ops = xen_machine_ops;
1326 * The only reliable way to retain the initial address of the
1327 * percpu gdt_page is to remember it here, so we can go and
1328 * mark it RW later, when the initial percpu area is freed.
1330 xen_initial_gdt = &per_cpu(gdt_page, 0);
1334 #ifdef CONFIG_ACPI_NUMA
1336 * The pages we from Xen are not related to machine pages, so
1337 * any NUMA information the kernel tries to get from ACPI will
1338 * be meaningless. Prevent it from trying.
1342 /* Don't do the full vcpu_info placement stuff until we have a
1343 possible map and a non-dummy shared_info. */
1344 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1346 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1348 local_irq_disable();
1349 early_boot_irqs_disabled = true;
1351 xen_raw_console_write("mapping kernel into physical memory\n");
1352 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1353 xen_start_info->nr_pages);
1354 xen_reserve_special_pages();
1356 /* keep using Xen gdt for now; no urgent need to change it */
1358 #ifdef CONFIG_X86_32
1359 pv_info.kernel_rpl = 1;
1360 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1361 pv_info.kernel_rpl = 0;
1363 pv_info.kernel_rpl = 0;
1365 /* set the limit of our address space */
1369 * We used to do this in xen_arch_setup, but that is too late
1370 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1371 * early_amd_init which pokes 0xcf8 port.
1374 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1376 xen_raw_printk("physdev_op failed %d\n", rc);
1378 #ifdef CONFIG_X86_32
1379 /* set up basic CPUID stuff */
1380 cpu_detect(&new_cpu_data);
1381 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1382 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1385 if (xen_start_info->mod_start) {
1386 if (xen_start_info->flags & SIF_MOD_START_PFN)
1387 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1389 initrd_start = __pa(xen_start_info->mod_start);
1392 /* Poke various useful things into boot_params */
1393 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1394 boot_params.hdr.ramdisk_image = initrd_start;
1395 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1396 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1397 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1399 if (!xen_initial_domain()) {
1400 add_preferred_console("xenboot", 0, NULL);
1401 add_preferred_console("tty", 0, NULL);
1402 add_preferred_console("hvc", 0, NULL);
1404 x86_init.pci.arch_init = pci_xen_init;
1406 const struct dom0_vga_console_info *info =
1407 (void *)((char *)xen_start_info +
1408 xen_start_info->console.dom0.info_off);
1409 struct xen_platform_op op = {
1410 .cmd = XENPF_firmware_info,
1411 .interface_version = XENPF_INTERFACE_VERSION,
1412 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1415 x86_platform.set_legacy_features =
1416 xen_dom0_set_legacy_features;
1417 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1418 xen_start_info->console.domU.mfn = 0;
1419 xen_start_info->console.domU.evtchn = 0;
1421 if (HYPERVISOR_platform_op(&op) == 0)
1422 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1424 /* Make sure ACS will be enabled */
1427 xen_acpi_sleep_register();
1429 /* Avoid searching for BIOS MP tables */
1430 x86_init.mpparse.find_smp_config = x86_init_noop;
1431 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1433 xen_boot_params_init_edd();
1436 /* PCI BIOS service won't work from a PV guest. */
1437 pci_probe &= ~PCI_PROBE_BIOS;
1439 xen_raw_console_write("about to get started...\n");
1441 /* Let's presume PV guests always boot on vCPU with id 0. */
1442 per_cpu(xen_vcpu_id, 0) = 0;
1444 xen_setup_runstate_info(0);
1448 /* Start the world */
1449 #ifdef CONFIG_X86_32
1450 i386_start_kernel();
1452 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1453 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1457 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1461 xen_setup_timer(cpu);
1463 rc = xen_smp_intr_init(cpu);
1465 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1470 rc = xen_smp_intr_init_pv(cpu);
1472 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1480 static int xen_cpu_dead_pv(unsigned int cpu)
1482 xen_smp_intr_free(cpu);
1483 xen_smp_intr_free_pv(cpu);
1485 xen_teardown_timer(cpu);
1490 static uint32_t __init xen_platform_pv(void)
1492 if (xen_pv_domain())
1493 return xen_cpuid_base();
1498 const struct hypervisor_x86 x86_hyper_xen_pv = {
1500 .detect = xen_platform_pv,
1501 .pin_vcpu = xen_pin_vcpu,
1503 EXPORT_SYMBOL(x86_hyper_xen_pv);