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/kernel.h>
15 #include <linux/init.h>
16 #include <linux/smp.h>
17 #include <linux/preempt.h>
18 #include <linux/hardirq.h>
19 #include <linux/percpu.h>
20 #include <linux/delay.h>
21 #include <linux/start_kernel.h>
22 #include <linux/sched.h>
23 #include <linux/bootmem.h>
24 #include <linux/module.h>
26 #include <linux/page-flags.h>
27 #include <linux/highmem.h>
28 #include <linux/console.h>
30 #include <xen/interface/xen.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/interface/sched.h>
34 #include <xen/features.h>
36 #include <xen/hvc-console.h>
38 #include <asm/paravirt.h>
40 #include <asm/xen/hypercall.h>
41 #include <asm/xen/hypervisor.h>
42 #include <asm/fixmap.h>
43 #include <asm/processor.h>
44 #include <asm/setup.h>
46 #include <asm/pgtable.h>
47 #include <asm/tlbflush.h>
48 #include <asm/reboot.h>
49 #include <asm/pgalloc.h>
53 #include "multicalls.h"
55 EXPORT_SYMBOL_GPL(hypercall_page);
57 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
58 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
61 * Note about cr3 (pagetable base) values:
63 * xen_cr3 contains the current logical cr3 value; it contains the
64 * last set cr3. This may not be the current effective cr3, because
65 * its update may be being lazily deferred. However, a vcpu looking
66 * at its own cr3 can use this value knowing that it everything will
69 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
70 * hypercall to set the vcpu cr3 is complete (so it may be a little
71 * out of date, but it will never be set early). If one vcpu is
72 * looking at another vcpu's cr3 value, it should use this variable.
74 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
75 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
77 struct start_info *xen_start_info;
78 EXPORT_SYMBOL_GPL(xen_start_info);
80 struct shared_info xen_dummy_shared_info;
83 * Point at some empty memory to start with. We map the real shared_info
84 * page as soon as fixmap is up and running.
86 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
89 * Flag to determine whether vcpu info placement is available on all
90 * VCPUs. We assume it is to start with, and then set it to zero on
91 * the first failure. This is because it can succeed on some VCPUs
92 * and not others, since it can involve hypervisor memory allocation,
93 * or because the guest failed to guarantee all the appropriate
94 * constraints on all VCPUs (ie buffer can't cross a page boundary).
96 * Note that any particular CPU may be using a placed vcpu structure,
97 * but we can only optimise if the all are.
99 * 0: not available, 1: available
101 static int have_vcpu_info_placement = 1;
103 static void xen_vcpu_setup(int cpu)
105 struct vcpu_register_vcpu_info info;
107 struct vcpu_info *vcpup;
109 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
110 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
112 if (!have_vcpu_info_placement)
113 return; /* already tested, not available */
115 vcpup = &per_cpu(xen_vcpu_info, cpu);
117 info.mfn = virt_to_mfn(vcpup);
118 info.offset = offset_in_page(vcpup);
120 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
121 cpu, vcpup, info.mfn, info.offset);
123 /* Check to see if the hypervisor will put the vcpu_info
124 structure where we want it, which allows direct access via
125 a percpu-variable. */
126 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
129 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
130 have_vcpu_info_placement = 0;
132 /* This cpu is using the registered vcpu info, even if
133 later ones fail to. */
134 per_cpu(xen_vcpu, cpu) = vcpup;
136 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
142 * On restore, set the vcpu placement up again.
143 * If it fails, then we're in a bad state, since
144 * we can't back out from using it...
146 void xen_vcpu_restore(void)
148 if (have_vcpu_info_placement) {
151 for_each_online_cpu(cpu) {
152 bool other_cpu = (cpu != smp_processor_id());
155 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
161 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
165 BUG_ON(!have_vcpu_info_placement);
169 static void __init xen_banner(void)
171 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
173 printk(KERN_INFO "Hypervisor signature: %s%s\n",
174 xen_start_info->magic,
175 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
178 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
179 unsigned int *cx, unsigned int *dx)
181 unsigned maskedx = ~0;
184 * Mask out inconvenient features, to try and disable as many
185 * unsupported kernel subsystems as possible.
188 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
189 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
190 (1 << X86_FEATURE_MCE) | /* disable MCE */
191 (1 << X86_FEATURE_MCA) | /* disable MCA */
192 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
194 asm(XEN_EMULATE_PREFIX "cpuid"
199 : "0" (*ax), "2" (*cx));
203 static void xen_set_debugreg(int reg, unsigned long val)
205 HYPERVISOR_set_debugreg(reg, val);
208 static unsigned long xen_get_debugreg(int reg)
210 return HYPERVISOR_get_debugreg(reg);
213 static unsigned long xen_save_fl(void)
215 struct vcpu_info *vcpu;
218 vcpu = x86_read_percpu(xen_vcpu);
220 /* flag has opposite sense of mask */
221 flags = !vcpu->evtchn_upcall_mask;
223 /* convert to IF type flag
227 return (-flags) & X86_EFLAGS_IF;
230 static void xen_restore_fl(unsigned long flags)
232 struct vcpu_info *vcpu;
234 /* convert from IF type flag */
235 flags = !(flags & X86_EFLAGS_IF);
237 /* There's a one instruction preempt window here. We need to
238 make sure we're don't switch CPUs between getting the vcpu
239 pointer and updating the mask. */
241 vcpu = x86_read_percpu(xen_vcpu);
242 vcpu->evtchn_upcall_mask = flags;
243 preempt_enable_no_resched();
245 /* Doesn't matter if we get preempted here, because any
246 pending event will get dealt with anyway. */
249 preempt_check_resched();
250 barrier(); /* unmask then check (avoid races) */
251 if (unlikely(vcpu->evtchn_upcall_pending))
252 force_evtchn_callback();
256 static void xen_irq_disable(void)
258 /* There's a one instruction preempt window here. We need to
259 make sure we're don't switch CPUs between getting the vcpu
260 pointer and updating the mask. */
262 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
263 preempt_enable_no_resched();
266 static void xen_irq_enable(void)
268 struct vcpu_info *vcpu;
270 /* We don't need to worry about being preempted here, since
271 either a) interrupts are disabled, so no preemption, or b)
272 the caller is confused and is trying to re-enable interrupts
273 on an indeterminate processor. */
275 vcpu = x86_read_percpu(xen_vcpu);
276 vcpu->evtchn_upcall_mask = 0;
278 /* Doesn't matter if we get preempted here, because any
279 pending event will get dealt with anyway. */
281 barrier(); /* unmask then check (avoid races) */
282 if (unlikely(vcpu->evtchn_upcall_pending))
283 force_evtchn_callback();
286 static void xen_safe_halt(void)
288 /* Blocking includes an implicit local_irq_enable(). */
289 if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0)
293 static void xen_halt(void)
296 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
301 static void xen_leave_lazy(void)
303 paravirt_leave_lazy(paravirt_get_lazy_mode());
307 static unsigned long xen_store_tr(void)
312 static void xen_set_ldt(const void *addr, unsigned entries)
314 struct mmuext_op *op;
315 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
318 op->cmd = MMUEXT_SET_LDT;
319 op->arg1.linear_addr = (unsigned long)addr;
320 op->arg2.nr_ents = entries;
322 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
324 xen_mc_issue(PARAVIRT_LAZY_CPU);
327 static void xen_load_gdt(const struct desc_ptr *dtr)
329 unsigned long *frames;
330 unsigned long va = dtr->address;
331 unsigned int size = dtr->size + 1;
332 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
334 struct multicall_space mcs;
336 /* A GDT can be up to 64k in size, which corresponds to 8192
337 8-byte entries, or 16 4k pages.. */
339 BUG_ON(size > 65536);
340 BUG_ON(va & ~PAGE_MASK);
342 mcs = xen_mc_entry(sizeof(*frames) * pages);
345 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
346 frames[f] = virt_to_mfn(va);
347 make_lowmem_page_readonly((void *)va);
350 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
352 xen_mc_issue(PARAVIRT_LAZY_CPU);
355 static void load_TLS_descriptor(struct thread_struct *t,
356 unsigned int cpu, unsigned int i)
358 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
359 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
360 struct multicall_space mc = __xen_mc_entry(0);
362 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
365 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
369 load_TLS_descriptor(t, cpu, 0);
370 load_TLS_descriptor(t, cpu, 1);
371 load_TLS_descriptor(t, cpu, 2);
373 xen_mc_issue(PARAVIRT_LAZY_CPU);
376 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
377 * it means we're in a context switch, and %gs has just been
378 * saved. This means we can zero it out to prevent faults on
379 * exit from the hypervisor if the next process has no %gs.
380 * Either way, it has been saved, and the new value will get
381 * loaded properly. This will go away as soon as Xen has been
382 * modified to not save/restore %gs for normal hypercalls.
384 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
388 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
391 unsigned long lp = (unsigned long)&dt[entrynum];
392 xmaddr_t mach_lp = virt_to_machine(lp);
393 u64 entry = *(u64 *)ptr;
398 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
404 static int cvt_gate_to_trap(int vector, const gate_desc *val,
405 struct trap_info *info)
407 if (val->type != 0xf && val->type != 0xe)
410 info->vector = vector;
411 info->address = gate_offset(*val);
412 info->cs = gate_segment(*val);
413 info->flags = val->dpl;
414 /* interrupt gates clear IF */
415 if (val->type == 0xe)
421 /* Locations of each CPU's IDT */
422 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
424 /* Set an IDT entry. If the entry is part of the current IDT, then
426 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
428 unsigned long p = (unsigned long)&dt[entrynum];
429 unsigned long start, end;
433 start = __get_cpu_var(idt_desc).address;
434 end = start + __get_cpu_var(idt_desc).size + 1;
438 native_write_idt_entry(dt, entrynum, g);
440 if (p >= start && (p + 8) <= end) {
441 struct trap_info info[2];
445 if (cvt_gate_to_trap(entrynum, g, &info[0]))
446 if (HYPERVISOR_set_trap_table(info))
453 static void xen_convert_trap_info(const struct desc_ptr *desc,
454 struct trap_info *traps)
456 unsigned in, out, count;
458 count = (desc->size+1) / sizeof(gate_desc);
461 for (in = out = 0; in < count; in++) {
462 gate_desc *entry = (gate_desc*)(desc->address) + in;
464 if (cvt_gate_to_trap(in, entry, &traps[out]))
467 traps[out].address = 0;
470 void xen_copy_trap_info(struct trap_info *traps)
472 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
474 xen_convert_trap_info(desc, traps);
477 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
478 hold a spinlock to protect the static traps[] array (static because
479 it avoids allocation, and saves stack space). */
480 static void xen_load_idt(const struct desc_ptr *desc)
482 static DEFINE_SPINLOCK(lock);
483 static struct trap_info traps[257];
487 __get_cpu_var(idt_desc) = *desc;
489 xen_convert_trap_info(desc, traps);
492 if (HYPERVISOR_set_trap_table(traps))
498 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
499 they're handled differently. */
500 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
501 const void *desc, int type)
512 xmaddr_t maddr = virt_to_machine(&dt[entry]);
515 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
524 static void xen_load_sp0(struct tss_struct *tss,
525 struct thread_struct *thread)
527 struct multicall_space mcs = xen_mc_entry(0);
528 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
529 xen_mc_issue(PARAVIRT_LAZY_CPU);
532 static void xen_set_iopl_mask(unsigned mask)
534 struct physdev_set_iopl set_iopl;
536 /* Force the change at ring 0. */
537 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
538 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
541 static void xen_io_delay(void)
545 #ifdef CONFIG_X86_LOCAL_APIC
546 static u32 xen_apic_read(unsigned long reg)
551 static void xen_apic_write(unsigned long reg, u32 val)
553 /* Warn to see if there's any stray references */
558 static void xen_flush_tlb(void)
560 struct mmuext_op *op;
561 struct multicall_space mcs;
565 mcs = xen_mc_entry(sizeof(*op));
568 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
569 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
571 xen_mc_issue(PARAVIRT_LAZY_MMU);
576 static void xen_flush_tlb_single(unsigned long addr)
578 struct mmuext_op *op;
579 struct multicall_space mcs;
583 mcs = xen_mc_entry(sizeof(*op));
585 op->cmd = MMUEXT_INVLPG_LOCAL;
586 op->arg1.linear_addr = addr & PAGE_MASK;
587 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
589 xen_mc_issue(PARAVIRT_LAZY_MMU);
594 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
601 cpumask_t cpumask = *cpus;
602 struct multicall_space mcs;
605 * A couple of (to be removed) sanity checks:
607 * - current CPU must not be in mask
608 * - mask must exist :)
610 BUG_ON(cpus_empty(cpumask));
611 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
614 /* If a CPU which we ran on has gone down, OK. */
615 cpus_and(cpumask, cpumask, cpu_online_map);
616 if (cpus_empty(cpumask))
619 mcs = xen_mc_entry(sizeof(*args));
621 args->mask = cpumask;
622 args->op.arg2.vcpumask = &args->mask;
624 if (va == TLB_FLUSH_ALL) {
625 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
627 args->op.cmd = MMUEXT_INVLPG_MULTI;
628 args->op.arg1.linear_addr = va;
631 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
633 xen_mc_issue(PARAVIRT_LAZY_MMU);
636 static void xen_clts(void)
638 struct multicall_space mcs;
640 mcs = xen_mc_entry(0);
642 MULTI_fpu_taskswitch(mcs.mc, 0);
644 xen_mc_issue(PARAVIRT_LAZY_CPU);
647 static void xen_write_cr0(unsigned long cr0)
649 struct multicall_space mcs;
651 /* Only pay attention to cr0.TS; everything else is
653 mcs = xen_mc_entry(0);
655 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
657 xen_mc_issue(PARAVIRT_LAZY_CPU);
660 static void xen_write_cr2(unsigned long cr2)
662 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
665 static unsigned long xen_read_cr2(void)
667 return x86_read_percpu(xen_vcpu)->arch.cr2;
670 static unsigned long xen_read_cr2_direct(void)
672 return x86_read_percpu(xen_vcpu_info.arch.cr2);
675 static void xen_write_cr4(unsigned long cr4)
680 native_write_cr4(cr4);
683 static unsigned long xen_read_cr3(void)
685 return x86_read_percpu(xen_cr3);
688 static void set_current_cr3(void *v)
690 x86_write_percpu(xen_current_cr3, (unsigned long)v);
693 static void xen_write_cr3(unsigned long cr3)
695 struct mmuext_op *op;
696 struct multicall_space mcs;
697 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
699 BUG_ON(preemptible());
701 mcs = xen_mc_entry(sizeof(*op)); /* disables interrupts */
703 /* Update while interrupts are disabled, so its atomic with
705 x86_write_percpu(xen_cr3, cr3);
708 op->cmd = MMUEXT_NEW_BASEPTR;
711 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
713 /* Update xen_update_cr3 once the batch has actually
715 xen_mc_callback(set_current_cr3, (void *)cr3);
717 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
720 /* Early in boot, while setting up the initial pagetable, assume
721 everything is pinned. */
722 static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
724 #ifdef CONFIG_FLATMEM
725 BUG_ON(mem_map); /* should only be used early */
727 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
730 /* Early release_pte assumes that all pts are pinned, since there's
731 only init_mm and anything attached to that is pinned. */
732 static void xen_release_pte_init(u32 pfn)
734 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
737 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
741 op.arg1.mfn = pfn_to_mfn(pfn);
742 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
746 /* This needs to make sure the new pte page is pinned iff its being
747 attached to a pinned pagetable. */
748 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
750 struct page *page = pfn_to_page(pfn);
752 if (PagePinned(virt_to_page(mm->pgd))) {
755 if (!PageHighMem(page)) {
756 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
758 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
760 /* make sure there are no stray mappings of
766 static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
768 xen_alloc_ptpage(mm, pfn, PT_PTE);
771 static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
773 xen_alloc_ptpage(mm, pfn, PT_PMD);
776 /* This should never happen until we're OK to use struct page */
777 static void xen_release_ptpage(u32 pfn, unsigned level)
779 struct page *page = pfn_to_page(pfn);
781 if (PagePinned(page)) {
782 if (!PageHighMem(page)) {
784 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
785 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
787 ClearPagePinned(page);
791 static void xen_release_pte(u32 pfn)
793 xen_release_ptpage(pfn, PT_PTE);
796 static void xen_release_pmd(u32 pfn)
798 xen_release_ptpage(pfn, PT_PMD);
801 #if PAGETABLE_LEVELS == 4
802 static void xen_alloc_pud(struct mm_struct *mm, u32 pfn)
804 xen_alloc_ptpage(mm, pfn, PT_PUD);
807 static void xen_release_pud(u32 pfn)
809 xen_release_ptpage(pfn, PT_PUD);
813 #ifdef CONFIG_HIGHPTE
814 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
816 pgprot_t prot = PAGE_KERNEL;
818 if (PagePinned(page))
819 prot = PAGE_KERNEL_RO;
821 if (0 && PageHighMem(page))
822 printk("mapping highpte %lx type %d prot %s\n",
823 page_to_pfn(page), type,
824 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
826 return kmap_atomic_prot(page, type, prot);
830 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
832 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
833 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
834 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
840 /* Init-time set_pte while constructing initial pagetables, which
841 doesn't allow RO pagetable pages to be remapped RW */
842 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
844 pte = mask_rw_pte(ptep, pte);
846 xen_set_pte(ptep, pte);
849 static __init void xen_pagetable_setup_start(pgd_t *base)
853 void xen_setup_shared_info(void)
855 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
856 set_fixmap(FIX_PARAVIRT_BOOTMAP,
857 xen_start_info->shared_info);
859 HYPERVISOR_shared_info =
860 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
862 HYPERVISOR_shared_info =
863 (struct shared_info *)__va(xen_start_info->shared_info);
866 /* In UP this is as good a place as any to set up shared info */
867 xen_setup_vcpu_info_placement();
870 xen_setup_mfn_list_list();
873 static __init void xen_pagetable_setup_done(pgd_t *base)
875 xen_setup_shared_info();
878 static __init void xen_post_allocator_init(void)
880 pv_mmu_ops.set_pte = xen_set_pte;
881 pv_mmu_ops.set_pmd = xen_set_pmd;
882 pv_mmu_ops.set_pud = xen_set_pud;
883 #if PAGETABLE_LEVELS == 4
884 pv_mmu_ops.set_pgd = xen_set_pgd;
887 /* This will work as long as patching hasn't happened yet
889 pv_mmu_ops.alloc_pte = xen_alloc_pte;
890 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
891 pv_mmu_ops.release_pte = xen_release_pte;
892 pv_mmu_ops.release_pmd = xen_release_pmd;
893 #if PAGETABLE_LEVELS == 4
894 pv_mmu_ops.alloc_pud = xen_alloc_pud;
895 pv_mmu_ops.release_pud = xen_release_pud;
898 xen_mark_init_mm_pinned();
901 /* This is called once we have the cpu_possible_map */
902 void xen_setup_vcpu_info_placement(void)
906 for_each_possible_cpu(cpu)
909 /* xen_vcpu_setup managed to place the vcpu_info within the
910 percpu area for all cpus, so make use of it */
912 if (have_vcpu_info_placement) {
913 printk(KERN_INFO "Xen: using vcpu_info placement\n");
915 pv_irq_ops.save_fl = xen_save_fl_direct;
916 pv_irq_ops.restore_fl = xen_restore_fl_direct;
917 pv_irq_ops.irq_disable = xen_irq_disable_direct;
918 pv_irq_ops.irq_enable = xen_irq_enable_direct;
919 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
924 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
925 unsigned long addr, unsigned len)
927 char *start, *end, *reloc;
930 start = end = reloc = NULL;
932 #define SITE(op, x) \
933 case PARAVIRT_PATCH(op.x): \
934 if (have_vcpu_info_placement) { \
935 start = (char *)xen_##x##_direct; \
936 end = xen_##x##_direct_end; \
937 reloc = xen_##x##_direct_reloc; \
943 SITE(pv_irq_ops, irq_enable);
944 SITE(pv_irq_ops, irq_disable);
945 SITE(pv_irq_ops, save_fl);
946 SITE(pv_irq_ops, restore_fl);
947 #endif /* CONFIG_X86_32 */
951 if (start == NULL || (end-start) > len)
954 ret = paravirt_patch_insns(insnbuf, len, start, end);
956 /* Note: because reloc is assigned from something that
957 appears to be an array, gcc assumes it's non-null,
958 but doesn't know its relationship with start and
960 if (reloc > start && reloc < end) {
961 int reloc_off = reloc - start;
962 long *relocp = (long *)(insnbuf + reloc_off);
963 long delta = start - (char *)addr;
971 ret = paravirt_patch_default(type, clobbers, insnbuf,
979 static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
986 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
987 #ifdef CONFIG_X86_F00F_BUG
993 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
995 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
997 #ifdef CONFIG_X86_LOCAL_APIC
998 case FIX_APIC_BASE: /* maps dummy local APIC */
1000 pte = pfn_pte(phys, prot);
1004 pte = mfn_pte(phys, prot);
1008 __native_set_fixmap(idx, pte);
1011 static const struct pv_info xen_info __initdata = {
1012 .paravirt_enabled = 1,
1013 .shared_kernel_pmd = 0,
1018 static const struct pv_init_ops xen_init_ops __initdata = {
1021 .banner = xen_banner,
1022 .memory_setup = xen_memory_setup,
1023 .arch_setup = xen_arch_setup,
1024 .post_allocator_init = xen_post_allocator_init,
1027 static const struct pv_time_ops xen_time_ops __initdata = {
1028 .time_init = xen_time_init,
1030 .set_wallclock = xen_set_wallclock,
1031 .get_wallclock = xen_get_wallclock,
1032 .get_tsc_khz = xen_tsc_khz,
1033 .sched_clock = xen_sched_clock,
1036 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1039 .set_debugreg = xen_set_debugreg,
1040 .get_debugreg = xen_get_debugreg,
1044 .read_cr0 = native_read_cr0,
1045 .write_cr0 = xen_write_cr0,
1047 .read_cr4 = native_read_cr4,
1048 .read_cr4_safe = native_read_cr4_safe,
1049 .write_cr4 = xen_write_cr4,
1051 .wbinvd = native_wbinvd,
1053 .read_msr = native_read_msr_safe,
1054 .write_msr = native_write_msr_safe,
1055 .read_tsc = native_read_tsc,
1056 .read_pmc = native_read_pmc,
1059 .irq_enable_sysexit = xen_sysexit,
1061 .load_tr_desc = paravirt_nop,
1062 .set_ldt = xen_set_ldt,
1063 .load_gdt = xen_load_gdt,
1064 .load_idt = xen_load_idt,
1065 .load_tls = xen_load_tls,
1067 .store_gdt = native_store_gdt,
1068 .store_idt = native_store_idt,
1069 .store_tr = xen_store_tr,
1071 .write_ldt_entry = xen_write_ldt_entry,
1072 .write_gdt_entry = xen_write_gdt_entry,
1073 .write_idt_entry = xen_write_idt_entry,
1074 .load_sp0 = xen_load_sp0,
1076 .set_iopl_mask = xen_set_iopl_mask,
1077 .io_delay = xen_io_delay,
1080 .enter = paravirt_enter_lazy_cpu,
1081 .leave = xen_leave_lazy,
1085 static const struct pv_irq_ops xen_irq_ops __initdata = {
1086 .init_IRQ = xen_init_IRQ,
1087 .save_fl = xen_save_fl,
1088 .restore_fl = xen_restore_fl,
1089 .irq_disable = xen_irq_disable,
1090 .irq_enable = xen_irq_enable,
1091 .safe_halt = xen_safe_halt,
1093 #ifdef CONFIG_X86_64
1094 .adjust_exception_frame = paravirt_nop,
1098 static const struct pv_apic_ops xen_apic_ops __initdata = {
1099 #ifdef CONFIG_X86_LOCAL_APIC
1100 .apic_write = xen_apic_write,
1101 .apic_write_atomic = xen_apic_write,
1102 .apic_read = xen_apic_read,
1103 .setup_boot_clock = paravirt_nop,
1104 .setup_secondary_clock = paravirt_nop,
1105 .startup_ipi_hook = paravirt_nop,
1109 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1110 .pagetable_setup_start = xen_pagetable_setup_start,
1111 .pagetable_setup_done = xen_pagetable_setup_done,
1113 .read_cr2 = xen_read_cr2,
1114 .write_cr2 = xen_write_cr2,
1116 .read_cr3 = xen_read_cr3,
1117 .write_cr3 = xen_write_cr3,
1119 .flush_tlb_user = xen_flush_tlb,
1120 .flush_tlb_kernel = xen_flush_tlb,
1121 .flush_tlb_single = xen_flush_tlb_single,
1122 .flush_tlb_others = xen_flush_tlb_others,
1124 .pte_update = paravirt_nop,
1125 .pte_update_defer = paravirt_nop,
1127 .pgd_alloc = __paravirt_pgd_alloc,
1128 .pgd_free = paravirt_nop,
1130 .alloc_pte = xen_alloc_pte_init,
1131 .release_pte = xen_release_pte_init,
1132 .alloc_pmd = xen_alloc_pte_init,
1133 .alloc_pmd_clone = paravirt_nop,
1134 .release_pmd = xen_release_pte_init,
1136 #ifdef CONFIG_HIGHPTE
1137 .kmap_atomic_pte = xen_kmap_atomic_pte,
1140 #ifdef CONFIG_X86_64
1141 .set_pte = xen_set_pte,
1143 .set_pte = xen_set_pte_init,
1145 .set_pte_at = xen_set_pte_at,
1146 .set_pmd = xen_set_pmd_hyper,
1148 .ptep_modify_prot_start = __ptep_modify_prot_start,
1149 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1151 .pte_val = xen_pte_val,
1152 .pte_flags = native_pte_val,
1153 .pgd_val = xen_pgd_val,
1155 .make_pte = xen_make_pte,
1156 .make_pgd = xen_make_pgd,
1158 #ifdef CONFIG_X86_PAE
1159 .set_pte_atomic = xen_set_pte_atomic,
1160 .set_pte_present = xen_set_pte_at,
1161 .pte_clear = xen_pte_clear,
1162 .pmd_clear = xen_pmd_clear,
1163 #endif /* CONFIG_X86_PAE */
1164 .set_pud = xen_set_pud_hyper,
1166 .make_pmd = xen_make_pmd,
1167 .pmd_val = xen_pmd_val,
1169 #if PAGETABLE_LEVELS == 4
1170 .pud_val = xen_pud_val,
1171 .make_pud = xen_make_pud,
1172 .set_pgd = xen_set_pgd_hyper,
1174 .alloc_pud = xen_alloc_pte_init,
1175 .release_pud = xen_release_pte_init,
1176 #endif /* PAGETABLE_LEVELS == 4 */
1178 .activate_mm = xen_activate_mm,
1179 .dup_mmap = xen_dup_mmap,
1180 .exit_mmap = xen_exit_mmap,
1183 .enter = paravirt_enter_lazy_mmu,
1184 .leave = xen_leave_lazy,
1187 .set_fixmap = xen_set_fixmap,
1190 static void xen_reboot(int reason)
1192 struct sched_shutdown r = { .reason = reason };
1198 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1202 static void xen_restart(char *msg)
1204 xen_reboot(SHUTDOWN_reboot);
1207 static void xen_emergency_restart(void)
1209 xen_reboot(SHUTDOWN_reboot);
1212 static void xen_machine_halt(void)
1214 xen_reboot(SHUTDOWN_poweroff);
1217 static void xen_crash_shutdown(struct pt_regs *regs)
1219 xen_reboot(SHUTDOWN_crash);
1222 static const struct machine_ops __initdata xen_machine_ops = {
1223 .restart = xen_restart,
1224 .halt = xen_machine_halt,
1225 .power_off = xen_machine_halt,
1226 .shutdown = xen_machine_halt,
1227 .crash_shutdown = xen_crash_shutdown,
1228 .emergency_restart = xen_emergency_restart,
1232 static void __init xen_reserve_top(void)
1234 #ifdef CONFIG_X86_32
1235 unsigned long top = HYPERVISOR_VIRT_START;
1236 struct xen_platform_parameters pp;
1238 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1239 top = pp.virt_start;
1241 reserve_top_address(-top + 2 * PAGE_SIZE);
1242 #endif /* CONFIG_X86_32 */
1246 * Like __va(), but returns address in the kernel mapping (which is
1247 * all we have until the physical memory mapping has been set up.
1249 static void *__ka(phys_addr_t paddr)
1251 #ifdef CONFIG_X86_64
1252 return (void *)(paddr + __START_KERNEL_map);
1258 /* Convert a machine address to physical address */
1259 static unsigned long m2p(phys_addr_t maddr)
1264 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1269 /* Convert a machine address to kernel virtual */
1270 static void *m2v(phys_addr_t maddr)
1272 return __ka(m2p(maddr));
1275 #ifdef CONFIG_X86_64
1276 static void walk(pgd_t *pgd, unsigned long addr)
1278 unsigned l4idx = pgd_index(addr);
1279 unsigned l3idx = pud_index(addr);
1280 unsigned l2idx = pmd_index(addr);
1281 unsigned l1idx = pte_index(addr);
1287 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1288 pgd, addr, l4idx, l3idx, l2idx, l1idx);
1291 xen_raw_printk(" l4: %016lx\n", l4.pgd);
1292 xen_raw_printk(" %016lx\n", pgd_val(l4));
1294 l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1295 xen_raw_printk(" l3: %016lx\n", l3.pud);
1296 xen_raw_printk(" %016lx\n", pud_val(l3));
1298 l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1299 xen_raw_printk(" l2: %016lx\n", l2.pmd);
1300 xen_raw_printk(" %016lx\n", pmd_val(l2));
1302 l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1303 xen_raw_printk(" l1: %016lx\n", l1.pte);
1304 xen_raw_printk(" %016lx\n", pte_val(l1));
1308 static void set_page_prot(void *addr, pgprot_t prot)
1310 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1311 pte_t pte = pfn_pte(pfn, prot);
1313 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1314 addr, pfn, get_phys_to_machine(pfn),
1315 pgprot_val(prot), pte.pte);
1317 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1322 * Identity map, in addition to plain kernel map. This needs to be
1323 * large enough to allocate page table pages to allocate the rest.
1324 * Each page can map 2MB.
1326 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
1328 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1330 unsigned pmdidx, pteidx;
1336 for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1339 /* Reuse or allocate a page of ptes */
1340 if (pmd_present(pmd[pmdidx]))
1341 pte_page = m2v(pmd[pmdidx].pmd);
1343 /* Check for free pte pages */
1344 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1347 pte_page = &level1_ident_pgt[ident_pte];
1348 ident_pte += PTRS_PER_PTE;
1350 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1353 /* Install mappings */
1354 for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1357 if (pfn > max_pfn_mapped)
1358 max_pfn_mapped = pfn;
1360 if (!pte_none(pte_page[pteidx]))
1363 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1364 pte_page[pteidx] = pte;
1368 for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1369 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1371 set_page_prot(pmd, PAGE_KERNEL_RO);
1374 #ifdef CONFIG_X86_64
1375 static void convert_pfn_mfn(void *v)
1380 /* All levels are converted the same way, so just treat them
1382 for(i = 0; i < PTRS_PER_PTE; i++)
1383 pte[i] = xen_make_pte(pte[i].pte);
1387 * Set up the inital kernel pagetable.
1389 * We can construct this by grafting the Xen provided pagetable into
1390 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1391 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1392 * means that only the kernel has a physical mapping to start with -
1393 * but that's enough to get __va working. We need to fill in the rest
1394 * of the physical mapping once some sort of allocator has been set
1397 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1402 /* Zap identity mapping */
1403 init_level4_pgt[0] = __pgd(0);
1405 /* Pre-constructed entries are in pfn, so convert to mfn */
1406 convert_pfn_mfn(init_level4_pgt);
1407 convert_pfn_mfn(level3_ident_pgt);
1408 convert_pfn_mfn(level3_kernel_pgt);
1410 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1411 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1413 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1414 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1416 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1417 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1418 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1420 /* Set up identity map */
1421 xen_map_identity_early(level2_ident_pgt, max_pfn);
1423 /* Make pagetable pieces RO */
1424 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1425 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1426 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1427 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1428 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1430 /* Pin down new L4 */
1431 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1432 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1434 /* Unpin Xen-provided one */
1435 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1438 pgd = init_level4_pgt;
1439 xen_write_cr3(__pa(pgd));
1441 reserve_early(__pa(xen_start_info->pt_base),
1442 __pa(xen_start_info->pt_base +
1443 xen_start_info->nr_pt_frames * PAGE_SIZE),
1448 #else /* !CONFIG_X86_64 */
1449 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1451 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1455 init_pg_tables_start = __pa(pgd);
1456 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1457 max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1459 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1460 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1462 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1464 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1465 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1466 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1468 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1469 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1470 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1472 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1474 xen_write_cr3(__pa(swapper_pg_dir));
1476 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1478 return swapper_pg_dir;
1480 #endif /* CONFIG_X86_64 */
1482 /* First C function to be called on Xen boot */
1483 asmlinkage void __init xen_start_kernel(void)
1487 if (!xen_start_info)
1490 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1492 xen_setup_features();
1494 /* Install Xen paravirt ops */
1496 pv_init_ops = xen_init_ops;
1497 pv_time_ops = xen_time_ops;
1498 pv_cpu_ops = xen_cpu_ops;
1499 pv_irq_ops = xen_irq_ops;
1500 pv_apic_ops = xen_apic_ops;
1501 pv_mmu_ops = xen_mmu_ops;
1503 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1504 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1505 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1508 machine_ops = xen_machine_ops;
1510 #ifdef CONFIG_X86_64
1511 /* Disable until direct per-cpu data access. */
1512 have_vcpu_info_placement = 0;
1519 if (!xen_feature(XENFEAT_auto_translated_physmap))
1520 xen_build_dynamic_phys_to_machine();
1522 pgd = (pgd_t *)xen_start_info->pt_base;
1524 /* Prevent unwanted bits from being set in PTEs. */
1525 __supported_pte_mask &= ~_PAGE_GLOBAL;
1526 if (!is_initial_xendomain())
1527 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1529 /* Don't do the full vcpu_info placement stuff until we have a
1530 possible map and a non-dummy shared_info. */
1531 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1533 xen_raw_console_write("mapping kernel into physical memory\n");
1534 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1538 /* keep using Xen gdt for now; no urgent need to change it */
1540 pv_info.kernel_rpl = 1;
1541 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1542 pv_info.kernel_rpl = 0;
1544 /* set the limit of our address space */
1547 #ifdef CONFIG_X86_32
1548 /* set up basic CPUID stuff */
1549 cpu_detect(&new_cpu_data);
1550 new_cpu_data.hard_math = 1;
1551 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1554 /* Poke various useful things into boot_params */
1555 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1556 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1557 ? __pa(xen_start_info->mod_start) : 0;
1558 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1560 if (!is_initial_xendomain()) {
1561 add_preferred_console("xenboot", 0, NULL);
1562 add_preferred_console("tty", 0, NULL);
1563 add_preferred_console("hvc", 0, NULL);
1566 xen_raw_console_write("about to get started...\n");
1569 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1570 &boot_params, __pa_symbol(&boot_params),
1571 __va(__pa_symbol(&boot_params)));
1573 walk(pgd, &boot_params);
1574 walk(pgd, __va(__pa(&boot_params)));
1577 /* Start the world */
1578 #ifdef CONFIG_X86_32
1579 i386_start_kernel();
1581 x86_64_start_reservations((char *)__pa_symbol(&boot_params));