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 on %s\n", pv_info.name);
146 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
147 version >> 16, version & 0xffff, extra.extraversion,
148 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
150 /* Check if running on Xen version (major, minor) or later */
152 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
154 unsigned int version;
159 version = HYPERVISOR_xen_version(XENVER_version, NULL);
160 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
161 ((version >> 16) > major))
166 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
167 static __read_mostly unsigned int cpuid_leaf5_edx_val;
169 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
170 unsigned int *cx, unsigned int *dx)
172 unsigned maskebx = ~0;
175 * Mask out inconvenient features, to try and disable as many
176 * unsupported kernel subsystems as possible.
179 case CPUID_MWAIT_LEAF:
180 /* Synthesize the values.. */
183 *cx = cpuid_leaf5_ecx_val;
184 *dx = cpuid_leaf5_edx_val;
188 /* Suppress extended topology stuff */
193 asm(XEN_EMULATE_PREFIX "cpuid"
198 : "0" (*ax), "2" (*cx));
202 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
204 static bool __init xen_check_mwait(void)
207 struct xen_platform_op op = {
208 .cmd = XENPF_set_processor_pminfo,
209 .u.set_pminfo.id = -1,
210 .u.set_pminfo.type = XEN_PM_PDC,
213 unsigned int ax, bx, cx, dx;
214 unsigned int mwait_mask;
216 /* We need to determine whether it is OK to expose the MWAIT
217 * capability to the kernel to harvest deeper than C3 states from ACPI
218 * _CST using the processor_harvest_xen.c module. For this to work, we
219 * need to gather the MWAIT_LEAF values (which the cstate.c code
220 * checks against). The hypervisor won't expose the MWAIT flag because
221 * it would break backwards compatibility; so we will find out directly
222 * from the hardware and hypercall.
224 if (!xen_initial_domain())
228 * When running under platform earlier than Xen4.2, do not expose
229 * mwait, to avoid the risk of loading native acpi pad driver
231 if (!xen_running_on_version_or_later(4, 2))
237 native_cpuid(&ax, &bx, &cx, &dx);
239 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
240 (1 << (X86_FEATURE_MWAIT % 32));
242 if ((cx & mwait_mask) != mwait_mask)
245 /* We need to emulate the MWAIT_LEAF and for that we need both
246 * ecx and edx. The hypercall provides only partial information.
249 ax = CPUID_MWAIT_LEAF;
254 native_cpuid(&ax, &bx, &cx, &dx);
256 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
257 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
259 buf[0] = ACPI_PDC_REVISION_ID;
261 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
263 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
265 if ((HYPERVISOR_platform_op(&op) == 0) &&
266 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
267 cpuid_leaf5_ecx_val = cx;
268 cpuid_leaf5_edx_val = dx;
276 static bool __init xen_check_xsave(void)
278 unsigned int cx, xsave_mask;
282 xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
283 (1 << (X86_FEATURE_OSXSAVE % 32));
285 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
286 return (cx & xsave_mask) == xsave_mask;
289 static void __init xen_init_capabilities(void)
291 setup_force_cpu_cap(X86_FEATURE_XENPV);
292 setup_clear_cpu_cap(X86_FEATURE_DCA);
293 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
294 setup_clear_cpu_cap(X86_FEATURE_MTRR);
295 setup_clear_cpu_cap(X86_FEATURE_ACC);
296 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
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 pages = DIV_ROUND_UP(size, PAGE_SIZE);
446 unsigned long frames[pages];
450 * A GDT can be up to 64k in size, which corresponds to 8192
451 * 8-byte entries, or 16 4k pages..
454 BUG_ON(size > 65536);
455 BUG_ON(va & ~PAGE_MASK);
457 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
460 unsigned long pfn, mfn;
464 * The GDT is per-cpu and is in the percpu data area.
465 * That can be virtually mapped, so we need to do a
466 * page-walk to get the underlying MFN for the
467 * hypercall. The page can also be in the kernel's
468 * linear range, so we need to RO that mapping too.
470 ptep = lookup_address(va, &level);
471 BUG_ON(ptep == NULL);
473 pfn = pte_pfn(*ptep);
474 mfn = pfn_to_mfn(pfn);
475 virt = __va(PFN_PHYS(pfn));
479 make_lowmem_page_readonly((void *)va);
480 make_lowmem_page_readonly(virt);
483 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
488 * load_gdt for early boot, when the gdt is only mapped once
490 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
492 unsigned long va = dtr->address;
493 unsigned int size = dtr->size + 1;
494 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
495 unsigned long frames[pages];
499 * A GDT can be up to 64k in size, which corresponds to 8192
500 * 8-byte entries, or 16 4k pages..
503 BUG_ON(size > 65536);
504 BUG_ON(va & ~PAGE_MASK);
506 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
508 unsigned long pfn, mfn;
510 pfn = virt_to_pfn(va);
511 mfn = pfn_to_mfn(pfn);
513 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
515 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
521 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
525 static inline bool desc_equal(const struct desc_struct *d1,
526 const struct desc_struct *d2)
528 return d1->a == d2->a && d1->b == d2->b;
531 static void load_TLS_descriptor(struct thread_struct *t,
532 unsigned int cpu, unsigned int i)
534 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
535 struct desc_struct *gdt;
537 struct multicall_space mc;
539 if (desc_equal(shadow, &t->tls_array[i]))
542 *shadow = t->tls_array[i];
544 gdt = get_cpu_gdt_rw(cpu);
545 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
546 mc = __xen_mc_entry(0);
548 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
551 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
554 * XXX sleazy hack: If we're being called in a lazy-cpu zone
555 * and lazy gs handling is enabled, it means we're in a
556 * context switch, and %gs has just been saved. This means we
557 * can zero it out to prevent faults on exit from the
558 * hypervisor if the next process has no %gs. Either way, it
559 * has been saved, and the new value will get loaded properly.
560 * This will go away as soon as Xen has been modified to not
561 * save/restore %gs for normal hypercalls.
563 * On x86_64, this hack is not used for %gs, because gs points
564 * to KERNEL_GS_BASE (and uses it for PDA references), so we
565 * must not zero %gs on x86_64
567 * For x86_64, we need to zero %fs, otherwise we may get an
568 * exception between the new %fs descriptor being loaded and
569 * %fs being effectively cleared at __switch_to().
571 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
581 load_TLS_descriptor(t, cpu, 0);
582 load_TLS_descriptor(t, cpu, 1);
583 load_TLS_descriptor(t, cpu, 2);
585 xen_mc_issue(PARAVIRT_LAZY_CPU);
589 static void xen_load_gs_index(unsigned int idx)
591 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
596 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
599 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
600 u64 entry = *(u64 *)ptr;
602 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
607 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
613 static int cvt_gate_to_trap(int vector, const gate_desc *val,
614 struct trap_info *info)
618 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
621 info->vector = vector;
623 addr = gate_offset(*val);
626 * Look for known traps using IST, and substitute them
627 * appropriately. The debugger ones are the only ones we care
628 * about. Xen will handle faults like double_fault,
629 * so we should never see them. Warn if
630 * there's an unexpected IST-using fault handler.
632 if (addr == (unsigned long)debug)
633 addr = (unsigned long)xen_debug;
634 else if (addr == (unsigned long)int3)
635 addr = (unsigned long)xen_int3;
636 else if (addr == (unsigned long)stack_segment)
637 addr = (unsigned long)xen_stack_segment;
638 else if (addr == (unsigned long)double_fault) {
639 /* Don't need to handle these */
641 #ifdef CONFIG_X86_MCE
642 } else if (addr == (unsigned long)machine_check) {
644 * when xen hypervisor inject vMCE to guest,
645 * use native mce handler to handle it
649 } else if (addr == (unsigned long)nmi)
651 * Use the native version as well.
655 /* Some other trap using IST? */
656 if (WARN_ON(val->ist != 0))
659 #endif /* CONFIG_X86_64 */
660 info->address = addr;
662 info->cs = gate_segment(*val);
663 info->flags = val->dpl;
664 /* interrupt gates clear IF */
665 if (val->type == GATE_INTERRUPT)
666 info->flags |= 1 << 2;
671 /* Locations of each CPU's IDT */
672 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
674 /* Set an IDT entry. If the entry is part of the current IDT, then
676 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
678 unsigned long p = (unsigned long)&dt[entrynum];
679 unsigned long start, end;
681 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
685 start = __this_cpu_read(idt_desc.address);
686 end = start + __this_cpu_read(idt_desc.size) + 1;
690 native_write_idt_entry(dt, entrynum, g);
692 if (p >= start && (p + 8) <= end) {
693 struct trap_info info[2];
697 if (cvt_gate_to_trap(entrynum, g, &info[0]))
698 if (HYPERVISOR_set_trap_table(info))
705 static void xen_convert_trap_info(const struct desc_ptr *desc,
706 struct trap_info *traps)
708 unsigned in, out, count;
710 count = (desc->size+1) / sizeof(gate_desc);
713 for (in = out = 0; in < count; in++) {
714 gate_desc *entry = (gate_desc *)(desc->address) + in;
716 if (cvt_gate_to_trap(in, entry, &traps[out]))
719 traps[out].address = 0;
722 void xen_copy_trap_info(struct trap_info *traps)
724 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
726 xen_convert_trap_info(desc, traps);
729 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
730 hold a spinlock to protect the static traps[] array (static because
731 it avoids allocation, and saves stack space). */
732 static void xen_load_idt(const struct desc_ptr *desc)
734 static DEFINE_SPINLOCK(lock);
735 static struct trap_info traps[257];
737 trace_xen_cpu_load_idt(desc);
741 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
743 xen_convert_trap_info(desc, traps);
746 if (HYPERVISOR_set_trap_table(traps))
752 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
753 they're handled differently. */
754 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
755 const void *desc, int type)
757 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
768 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
771 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
781 * Version of write_gdt_entry for use at early boot-time needed to
782 * update an entry as simply as possible.
784 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
785 const void *desc, int type)
787 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
796 xmaddr_t maddr = virt_to_machine(&dt[entry]);
798 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
799 dt[entry] = *(struct desc_struct *)desc;
805 static void xen_load_sp0(struct tss_struct *tss,
806 struct thread_struct *thread)
808 struct multicall_space mcs;
810 mcs = xen_mc_entry(0);
811 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
812 xen_mc_issue(PARAVIRT_LAZY_CPU);
813 tss->x86_tss.sp0 = thread->sp0;
816 void xen_set_iopl_mask(unsigned mask)
818 struct physdev_set_iopl set_iopl;
820 /* Force the change at ring 0. */
821 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
822 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
825 static void xen_io_delay(void)
829 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
831 static unsigned long xen_read_cr0(void)
833 unsigned long cr0 = this_cpu_read(xen_cr0_value);
835 if (unlikely(cr0 == 0)) {
836 cr0 = native_read_cr0();
837 this_cpu_write(xen_cr0_value, cr0);
843 static void xen_write_cr0(unsigned long cr0)
845 struct multicall_space mcs;
847 this_cpu_write(xen_cr0_value, cr0);
849 /* Only pay attention to cr0.TS; everything else is
851 mcs = xen_mc_entry(0);
853 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
855 xen_mc_issue(PARAVIRT_LAZY_CPU);
858 static void xen_write_cr4(unsigned long cr4)
860 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
862 native_write_cr4(cr4);
865 static inline unsigned long xen_read_cr8(void)
869 static inline void xen_write_cr8(unsigned long val)
875 static u64 xen_read_msr_safe(unsigned int msr, int *err)
879 if (pmu_msr_read(msr, &val, err))
882 val = native_read_msr_safe(msr, err);
884 case MSR_IA32_APICBASE:
885 #ifdef CONFIG_X86_X2APIC
886 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
888 val &= ~X2APIC_ENABLE;
894 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
905 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
906 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
907 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
910 base = ((u64)high << 32) | low;
911 if (HYPERVISOR_set_segment_base(which, base) != 0)
919 case MSR_SYSCALL_MASK:
920 case MSR_IA32_SYSENTER_CS:
921 case MSR_IA32_SYSENTER_ESP:
922 case MSR_IA32_SYSENTER_EIP:
923 /* Fast syscall setup is all done in hypercalls, so
924 these are all ignored. Stub them out here to stop
925 Xen console noise. */
929 if (!pmu_msr_write(msr, low, high, &ret))
930 ret = native_write_msr_safe(msr, low, high);
936 static u64 xen_read_msr(unsigned int msr)
939 * This will silently swallow a #GP from RDMSR. It may be worth
944 return xen_read_msr_safe(msr, &err);
947 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
950 * This will silently swallow a #GP from WRMSR. It may be worth
953 xen_write_msr_safe(msr, low, high);
956 void xen_setup_shared_info(void)
958 set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
960 HYPERVISOR_shared_info =
961 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
964 /* In UP this is as good a place as any to set up shared info */
965 xen_setup_vcpu_info_placement();
968 xen_setup_mfn_list_list();
971 * Now that shared info is set up we can start using routines that
972 * point to pvclock area.
974 if (system_state == SYSTEM_BOOTING)
978 /* This is called once we have the cpu_possible_mask */
979 void xen_setup_vcpu_info_placement(void)
983 for_each_possible_cpu(cpu) {
984 /* Set up direct vCPU id mapping for PV guests. */
985 per_cpu(xen_vcpu_id, cpu) = cpu;
990 * xen_vcpu_setup managed to place the vcpu_info within the
991 * percpu area for all cpus, so make use of it.
993 if (xen_have_vcpu_info_placement) {
994 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
995 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
996 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
997 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
998 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1002 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1003 unsigned long addr, unsigned len)
1005 char *start, *end, *reloc;
1008 start = end = reloc = NULL;
1010 #define SITE(op, x) \
1011 case PARAVIRT_PATCH(op.x): \
1012 if (xen_have_vcpu_info_placement) { \
1013 start = (char *)xen_##x##_direct; \
1014 end = xen_##x##_direct_end; \
1015 reloc = xen_##x##_direct_reloc; \
1020 SITE(pv_irq_ops, irq_enable);
1021 SITE(pv_irq_ops, irq_disable);
1022 SITE(pv_irq_ops, save_fl);
1023 SITE(pv_irq_ops, restore_fl);
1027 if (start == NULL || (end-start) > len)
1030 ret = paravirt_patch_insns(insnbuf, len, start, end);
1032 /* Note: because reloc is assigned from something that
1033 appears to be an array, gcc assumes it's non-null,
1034 but doesn't know its relationship with start and
1036 if (reloc > start && reloc < end) {
1037 int reloc_off = reloc - start;
1038 long *relocp = (long *)(insnbuf + reloc_off);
1039 long delta = start - (char *)addr;
1047 ret = paravirt_patch_default(type, clobbers, insnbuf,
1055 static const struct pv_info xen_info __initconst = {
1056 .shared_kernel_pmd = 0,
1058 #ifdef CONFIG_X86_64
1059 .extra_user_64bit_cs = FLAT_USER_CS64,
1064 static const struct pv_init_ops xen_init_ops __initconst = {
1068 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1071 .set_debugreg = xen_set_debugreg,
1072 .get_debugreg = xen_get_debugreg,
1074 .read_cr0 = xen_read_cr0,
1075 .write_cr0 = xen_write_cr0,
1077 .read_cr4 = native_read_cr4,
1078 .write_cr4 = xen_write_cr4,
1080 #ifdef CONFIG_X86_64
1081 .read_cr8 = xen_read_cr8,
1082 .write_cr8 = xen_write_cr8,
1085 .wbinvd = native_wbinvd,
1087 .read_msr = xen_read_msr,
1088 .write_msr = xen_write_msr,
1090 .read_msr_safe = xen_read_msr_safe,
1091 .write_msr_safe = xen_write_msr_safe,
1093 .read_pmc = xen_read_pmc,
1096 #ifdef CONFIG_X86_64
1097 .usergs_sysret64 = xen_sysret64,
1100 .load_tr_desc = paravirt_nop,
1101 .set_ldt = xen_set_ldt,
1102 .load_gdt = xen_load_gdt,
1103 .load_idt = xen_load_idt,
1104 .load_tls = xen_load_tls,
1105 #ifdef CONFIG_X86_64
1106 .load_gs_index = xen_load_gs_index,
1109 .alloc_ldt = xen_alloc_ldt,
1110 .free_ldt = xen_free_ldt,
1112 .store_idt = native_store_idt,
1113 .store_tr = xen_store_tr,
1115 .write_ldt_entry = xen_write_ldt_entry,
1116 .write_gdt_entry = xen_write_gdt_entry,
1117 .write_idt_entry = xen_write_idt_entry,
1118 .load_sp0 = xen_load_sp0,
1120 .set_iopl_mask = xen_set_iopl_mask,
1121 .io_delay = xen_io_delay,
1123 /* Xen takes care of %gs when switching to usermode for us */
1124 .swapgs = paravirt_nop,
1126 .start_context_switch = paravirt_start_context_switch,
1127 .end_context_switch = xen_end_context_switch,
1130 static void xen_restart(char *msg)
1132 xen_reboot(SHUTDOWN_reboot);
1135 static void xen_machine_halt(void)
1137 xen_reboot(SHUTDOWN_poweroff);
1140 static void xen_machine_power_off(void)
1144 xen_reboot(SHUTDOWN_poweroff);
1147 static void xen_crash_shutdown(struct pt_regs *regs)
1149 xen_reboot(SHUTDOWN_crash);
1152 static const struct machine_ops xen_machine_ops __initconst = {
1153 .restart = xen_restart,
1154 .halt = xen_machine_halt,
1155 .power_off = xen_machine_power_off,
1156 .shutdown = xen_machine_halt,
1157 .crash_shutdown = xen_crash_shutdown,
1158 .emergency_restart = xen_emergency_restart,
1161 static unsigned char xen_get_nmi_reason(void)
1163 unsigned char reason = 0;
1165 /* Construct a value which looks like it came from port 0x61. */
1166 if (test_bit(_XEN_NMIREASON_io_error,
1167 &HYPERVISOR_shared_info->arch.nmi_reason))
1168 reason |= NMI_REASON_IOCHK;
1169 if (test_bit(_XEN_NMIREASON_pci_serr,
1170 &HYPERVISOR_shared_info->arch.nmi_reason))
1171 reason |= NMI_REASON_SERR;
1176 static void __init xen_boot_params_init_edd(void)
1178 #if IS_ENABLED(CONFIG_EDD)
1179 struct xen_platform_op op;
1180 struct edd_info *edd_info;
1185 edd_info = boot_params.eddbuf;
1186 mbr_signature = boot_params.edd_mbr_sig_buffer;
1188 op.cmd = XENPF_firmware_info;
1190 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1191 for (nr = 0; nr < EDDMAXNR; nr++) {
1192 struct edd_info *info = edd_info + nr;
1194 op.u.firmware_info.index = nr;
1195 info->params.length = sizeof(info->params);
1196 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1198 ret = HYPERVISOR_platform_op(&op);
1202 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1205 C(interface_support);
1206 C(legacy_max_cylinder);
1208 C(legacy_sectors_per_track);
1211 boot_params.eddbuf_entries = nr;
1213 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1214 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1215 op.u.firmware_info.index = nr;
1216 ret = HYPERVISOR_platform_op(&op);
1219 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1221 boot_params.edd_mbr_sig_buf_entries = nr;
1226 * Set up the GDT and segment registers for -fstack-protector. Until
1227 * we do this, we have to be careful not to call any stack-protected
1228 * function, which is most of the kernel.
1230 static void xen_setup_gdt(int cpu)
1232 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1233 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1235 setup_stack_canary_segment(0);
1236 switch_to_new_gdt(0);
1238 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1239 pv_cpu_ops.load_gdt = xen_load_gdt;
1242 static void __init xen_dom0_set_legacy_features(void)
1244 x86_platform.legacy.rtc = 1;
1247 /* First C function to be called on Xen boot */
1248 asmlinkage __visible void __init xen_start_kernel(void)
1250 struct physdev_set_iopl set_iopl;
1251 unsigned long initrd_start = 0;
1254 if (!xen_start_info)
1257 xen_domain_type = XEN_PV_DOMAIN;
1259 xen_setup_features();
1261 xen_setup_machphys_mapping();
1263 /* Install Xen paravirt ops */
1265 pv_init_ops = xen_init_ops;
1266 pv_cpu_ops = xen_cpu_ops;
1268 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1270 x86_init.resources.memory_setup = xen_memory_setup;
1271 x86_init.oem.arch_setup = xen_arch_setup;
1272 x86_init.oem.banner = xen_banner;
1275 * Set up some pagetable state before starting to set any ptes.
1280 /* Prevent unwanted bits from being set in PTEs. */
1281 __supported_pte_mask &= ~_PAGE_GLOBAL;
1284 * Prevent page tables from being allocated in highmem, even
1285 * if CONFIG_HIGHPTE is enabled.
1287 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1289 /* Work out if we support NX */
1293 xen_build_dynamic_phys_to_machine();
1296 * Set up kernel GDT and segment registers, mainly so that
1297 * -fstack-protector code can be executed.
1302 xen_init_capabilities();
1304 #ifdef CONFIG_X86_LOCAL_APIC
1306 * set up the basic apic ops.
1311 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1312 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1313 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1316 machine_ops = xen_machine_ops;
1319 * The only reliable way to retain the initial address of the
1320 * percpu gdt_page is to remember it here, so we can go and
1321 * mark it RW later, when the initial percpu area is freed.
1323 xen_initial_gdt = &per_cpu(gdt_page, 0);
1327 #ifdef CONFIG_ACPI_NUMA
1329 * The pages we from Xen are not related to machine pages, so
1330 * any NUMA information the kernel tries to get from ACPI will
1331 * be meaningless. Prevent it from trying.
1335 /* Don't do the full vcpu_info placement stuff until we have a
1336 possible map and a non-dummy shared_info. */
1337 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1339 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1341 local_irq_disable();
1342 early_boot_irqs_disabled = true;
1344 xen_raw_console_write("mapping kernel into physical memory\n");
1345 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1346 xen_start_info->nr_pages);
1347 xen_reserve_special_pages();
1349 /* keep using Xen gdt for now; no urgent need to change it */
1351 #ifdef CONFIG_X86_32
1352 pv_info.kernel_rpl = 1;
1353 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1354 pv_info.kernel_rpl = 0;
1356 pv_info.kernel_rpl = 0;
1358 /* set the limit of our address space */
1362 * We used to do this in xen_arch_setup, but that is too late
1363 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1364 * early_amd_init which pokes 0xcf8 port.
1367 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1369 xen_raw_printk("physdev_op failed %d\n", rc);
1371 #ifdef CONFIG_X86_32
1372 /* set up basic CPUID stuff */
1373 cpu_detect(&new_cpu_data);
1374 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1375 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1378 if (xen_start_info->mod_start) {
1379 if (xen_start_info->flags & SIF_MOD_START_PFN)
1380 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1382 initrd_start = __pa(xen_start_info->mod_start);
1385 /* Poke various useful things into boot_params */
1386 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1387 boot_params.hdr.ramdisk_image = initrd_start;
1388 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1389 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1390 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1392 if (!xen_initial_domain()) {
1393 add_preferred_console("xenboot", 0, NULL);
1394 add_preferred_console("tty", 0, NULL);
1395 add_preferred_console("hvc", 0, NULL);
1397 x86_init.pci.arch_init = pci_xen_init;
1399 const struct dom0_vga_console_info *info =
1400 (void *)((char *)xen_start_info +
1401 xen_start_info->console.dom0.info_off);
1402 struct xen_platform_op op = {
1403 .cmd = XENPF_firmware_info,
1404 .interface_version = XENPF_INTERFACE_VERSION,
1405 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1408 x86_platform.set_legacy_features =
1409 xen_dom0_set_legacy_features;
1410 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1411 xen_start_info->console.domU.mfn = 0;
1412 xen_start_info->console.domU.evtchn = 0;
1414 if (HYPERVISOR_platform_op(&op) == 0)
1415 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1417 /* Make sure ACS will be enabled */
1420 xen_acpi_sleep_register();
1422 /* Avoid searching for BIOS MP tables */
1423 x86_init.mpparse.find_smp_config = x86_init_noop;
1424 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1426 xen_boot_params_init_edd();
1429 /* PCI BIOS service won't work from a PV guest. */
1430 pci_probe &= ~PCI_PROBE_BIOS;
1432 xen_raw_console_write("about to get started...\n");
1434 /* Let's presume PV guests always boot on vCPU with id 0. */
1435 per_cpu(xen_vcpu_id, 0) = 0;
1437 xen_setup_runstate_info(0);
1441 /* Start the world */
1442 #ifdef CONFIG_X86_32
1443 i386_start_kernel();
1445 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1446 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1450 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1454 xen_setup_timer(cpu);
1456 rc = xen_smp_intr_init(cpu);
1458 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1463 rc = xen_smp_intr_init_pv(cpu);
1465 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1473 static int xen_cpu_dead_pv(unsigned int cpu)
1475 xen_smp_intr_free(cpu);
1476 xen_smp_intr_free_pv(cpu);
1478 xen_teardown_timer(cpu);
1483 static uint32_t __init xen_platform_pv(void)
1485 if (xen_pv_domain())
1486 return xen_cpuid_base();
1491 const struct hypervisor_x86 x86_hyper_xen_pv = {
1493 .detect = xen_platform_pv,
1494 .pin_vcpu = xen_pin_vcpu,
1496 EXPORT_SYMBOL(x86_hyper_xen_pv);