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/smp.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>
37 #include <asm/paravirt.h>
39 #include <asm/xen/hypercall.h>
40 #include <asm/xen/hypervisor.h>
41 #include <asm/fixmap.h>
42 #include <asm/processor.h>
43 #include <asm/setup.h>
45 #include <asm/pgtable.h>
46 #include <asm/tlbflush.h>
47 #include <asm/reboot.h>
51 #include "multicalls.h"
53 EXPORT_SYMBOL_GPL(hypercall_page);
55 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
56 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
57 DEFINE_PER_CPU(unsigned long, xen_cr3);
59 struct start_info *xen_start_info;
60 EXPORT_SYMBOL_GPL(xen_start_info);
62 static /* __initdata */ struct shared_info dummy_shared_info;
65 * Point at some empty memory to start with. We map the real shared_info
66 * page as soon as fixmap is up and running.
68 struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info;
71 * Flag to determine whether vcpu info placement is available on all
72 * VCPUs. We assume it is to start with, and then set it to zero on
73 * the first failure. This is because it can succeed on some VCPUs
74 * and not others, since it can involve hypervisor memory allocation,
75 * or because the guest failed to guarantee all the appropriate
76 * constraints on all VCPUs (ie buffer can't cross a page boundary).
78 * Note that any particular CPU may be using a placed vcpu structure,
79 * but we can only optimise if the all are.
81 * 0: not available, 1: available
83 static int have_vcpu_info_placement = 1;
85 static void __init xen_vcpu_setup(int cpu)
87 struct vcpu_register_vcpu_info info;
89 struct vcpu_info *vcpup;
91 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
93 if (!have_vcpu_info_placement)
94 return; /* already tested, not available */
96 vcpup = &per_cpu(xen_vcpu_info, cpu);
98 info.mfn = virt_to_mfn(vcpup);
99 info.offset = offset_in_page(vcpup);
101 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %x, offset %d\n",
102 cpu, vcpup, info.mfn, info.offset);
104 /* Check to see if the hypervisor will put the vcpu_info
105 structure where we want it, which allows direct access via
106 a percpu-variable. */
107 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
110 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
111 have_vcpu_info_placement = 0;
113 /* This cpu is using the registered vcpu info, even if
114 later ones fail to. */
115 per_cpu(xen_vcpu, cpu) = vcpup;
117 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
122 static void __init xen_banner(void)
124 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
126 printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
129 static void xen_cpuid(unsigned int *eax, unsigned int *ebx,
130 unsigned int *ecx, unsigned int *edx)
132 unsigned maskedx = ~0;
135 * Mask out inconvenient features, to try and disable as many
136 * unsupported kernel subsystems as possible.
139 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
140 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
141 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
143 asm(XEN_EMULATE_PREFIX "cpuid"
148 : "0" (*eax), "2" (*ecx));
152 static void xen_set_debugreg(int reg, unsigned long val)
154 HYPERVISOR_set_debugreg(reg, val);
157 static unsigned long xen_get_debugreg(int reg)
159 return HYPERVISOR_get_debugreg(reg);
162 static unsigned long xen_save_fl(void)
164 struct vcpu_info *vcpu;
167 vcpu = x86_read_percpu(xen_vcpu);
169 /* flag has opposite sense of mask */
170 flags = !vcpu->evtchn_upcall_mask;
172 /* convert to IF type flag
176 return (-flags) & X86_EFLAGS_IF;
179 static void xen_restore_fl(unsigned long flags)
181 struct vcpu_info *vcpu;
183 /* convert from IF type flag */
184 flags = !(flags & X86_EFLAGS_IF);
186 /* There's a one instruction preempt window here. We need to
187 make sure we're don't switch CPUs between getting the vcpu
188 pointer and updating the mask. */
190 vcpu = x86_read_percpu(xen_vcpu);
191 vcpu->evtchn_upcall_mask = flags;
192 preempt_enable_no_resched();
194 /* Doesn't matter if we get preempted here, because any
195 pending event will get dealt with anyway. */
198 preempt_check_resched();
199 barrier(); /* unmask then check (avoid races) */
200 if (unlikely(vcpu->evtchn_upcall_pending))
201 force_evtchn_callback();
205 static void xen_irq_disable(void)
207 /* There's a one instruction preempt window here. We need to
208 make sure we're don't switch CPUs between getting the vcpu
209 pointer and updating the mask. */
211 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
212 preempt_enable_no_resched();
215 static void xen_irq_enable(void)
217 struct vcpu_info *vcpu;
219 /* There's a one instruction preempt window here. We need to
220 make sure we're don't switch CPUs between getting the vcpu
221 pointer and updating the mask. */
223 vcpu = x86_read_percpu(xen_vcpu);
224 vcpu->evtchn_upcall_mask = 0;
225 preempt_enable_no_resched();
227 /* Doesn't matter if we get preempted here, because any
228 pending event will get dealt with anyway. */
230 barrier(); /* unmask then check (avoid races) */
231 if (unlikely(vcpu->evtchn_upcall_pending))
232 force_evtchn_callback();
235 static void xen_safe_halt(void)
237 /* Blocking includes an implicit local_irq_enable(). */
238 if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
242 static void xen_halt(void)
245 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
250 static void xen_leave_lazy(void)
252 paravirt_leave_lazy(paravirt_get_lazy_mode());
256 static unsigned long xen_store_tr(void)
261 static void xen_set_ldt(const void *addr, unsigned entries)
263 unsigned long linear_addr = (unsigned long)addr;
264 struct mmuext_op *op;
265 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
268 op->cmd = MMUEXT_SET_LDT;
270 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
272 maddr = arbitrary_virt_to_machine((unsigned long)addr);
273 linear_addr = (unsigned long)maddr.maddr;
275 op->arg1.linear_addr = linear_addr;
276 op->arg2.nr_ents = entries;
278 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
280 xen_mc_issue(PARAVIRT_LAZY_CPU);
283 static void xen_load_gdt(const struct Xgt_desc_struct *dtr)
285 unsigned long *frames;
286 unsigned long va = dtr->address;
287 unsigned int size = dtr->size + 1;
288 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
290 struct multicall_space mcs;
292 /* A GDT can be up to 64k in size, which corresponds to 8192
293 8-byte entries, or 16 4k pages.. */
295 BUG_ON(size > 65536);
296 BUG_ON(va & ~PAGE_MASK);
298 mcs = xen_mc_entry(sizeof(*frames) * pages);
301 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
302 frames[f] = virt_to_mfn(va);
303 make_lowmem_page_readonly((void *)va);
306 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
308 xen_mc_issue(PARAVIRT_LAZY_CPU);
311 static void load_TLS_descriptor(struct thread_struct *t,
312 unsigned int cpu, unsigned int i)
314 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
315 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
316 struct multicall_space mc = __xen_mc_entry(0);
318 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
321 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
325 load_TLS_descriptor(t, cpu, 0);
326 load_TLS_descriptor(t, cpu, 1);
327 load_TLS_descriptor(t, cpu, 2);
329 xen_mc_issue(PARAVIRT_LAZY_CPU);
332 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
333 * it means we're in a context switch, and %gs has just been
334 * saved. This means we can zero it out to prevent faults on
335 * exit from the hypervisor if the next process has no %gs.
336 * Either way, it has been saved, and the new value will get
337 * loaded properly. This will go away as soon as Xen has been
338 * modified to not save/restore %gs for normal hypercalls.
340 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
344 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
347 unsigned long lp = (unsigned long)&dt[entrynum];
348 xmaddr_t mach_lp = virt_to_machine(lp);
349 u64 entry = (u64)high << 32 | low;
354 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
360 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
361 struct trap_info *info)
365 type = (high >> 8) & 0x1f;
366 dpl = (high >> 13) & 3;
368 if (type != 0xf && type != 0xe)
371 info->vector = vector;
372 info->address = (high & 0xffff0000) | (low & 0x0000ffff);
373 info->cs = low >> 16;
375 /* interrupt gates clear IF */
382 /* Locations of each CPU's IDT */
383 static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc);
385 /* Set an IDT entry. If the entry is part of the current IDT, then
387 static void xen_write_idt_entry(struct desc_struct *dt, int entrynum,
390 unsigned long p = (unsigned long)&dt[entrynum];
391 unsigned long start, end;
395 start = __get_cpu_var(idt_desc).address;
396 end = start + __get_cpu_var(idt_desc).size + 1;
400 write_dt_entry(dt, entrynum, low, high);
402 if (p >= start && (p + 8) <= end) {
403 struct trap_info info[2];
407 if (cvt_gate_to_trap(entrynum, low, high, &info[0]))
408 if (HYPERVISOR_set_trap_table(info))
415 static void xen_convert_trap_info(const struct Xgt_desc_struct *desc,
416 struct trap_info *traps)
418 unsigned in, out, count;
420 count = (desc->size+1) / 8;
423 for (in = out = 0; in < count; in++) {
424 const u32 *entry = (u32 *)(desc->address + in * 8);
426 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
429 traps[out].address = 0;
432 void xen_copy_trap_info(struct trap_info *traps)
434 const struct Xgt_desc_struct *desc = &__get_cpu_var(idt_desc);
436 xen_convert_trap_info(desc, traps);
439 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
440 hold a spinlock to protect the static traps[] array (static because
441 it avoids allocation, and saves stack space). */
442 static void xen_load_idt(const struct Xgt_desc_struct *desc)
444 static DEFINE_SPINLOCK(lock);
445 static struct trap_info traps[257];
449 __get_cpu_var(idt_desc) = *desc;
451 xen_convert_trap_info(desc, traps);
454 if (HYPERVISOR_set_trap_table(traps))
460 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
461 they're handled differently. */
462 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
467 switch ((high >> 8) & 0xff) {
474 xmaddr_t maddr = virt_to_machine(&dt[entry]);
475 u64 desc = (u64)high << 32 | low;
478 if (HYPERVISOR_update_descriptor(maddr.maddr, desc))
487 static void xen_load_esp0(struct tss_struct *tss,
488 struct thread_struct *thread)
490 struct multicall_space mcs = xen_mc_entry(0);
491 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0);
492 xen_mc_issue(PARAVIRT_LAZY_CPU);
495 static void xen_set_iopl_mask(unsigned mask)
497 struct physdev_set_iopl set_iopl;
499 /* Force the change at ring 0. */
500 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
501 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
504 static void xen_io_delay(void)
508 #ifdef CONFIG_X86_LOCAL_APIC
509 static unsigned long xen_apic_read(unsigned long reg)
514 static void xen_apic_write(unsigned long reg, unsigned long val)
516 /* Warn to see if there's any stray references */
521 static void xen_flush_tlb(void)
523 struct mmuext_op *op;
524 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
527 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
528 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
530 xen_mc_issue(PARAVIRT_LAZY_MMU);
533 static void xen_flush_tlb_single(unsigned long addr)
535 struct mmuext_op *op;
536 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
539 op->cmd = MMUEXT_INVLPG_LOCAL;
540 op->arg1.linear_addr = addr & PAGE_MASK;
541 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
543 xen_mc_issue(PARAVIRT_LAZY_MMU);
546 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
553 cpumask_t cpumask = *cpus;
554 struct multicall_space mcs;
557 * A couple of (to be removed) sanity checks:
559 * - current CPU must not be in mask
560 * - mask must exist :)
562 BUG_ON(cpus_empty(cpumask));
563 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
566 /* If a CPU which we ran on has gone down, OK. */
567 cpus_and(cpumask, cpumask, cpu_online_map);
568 if (cpus_empty(cpumask))
571 mcs = xen_mc_entry(sizeof(*args));
573 args->mask = cpumask;
574 args->op.arg2.vcpumask = &args->mask;
576 if (va == TLB_FLUSH_ALL) {
577 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
579 args->op.cmd = MMUEXT_INVLPG_MULTI;
580 args->op.arg1.linear_addr = va;
583 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
585 xen_mc_issue(PARAVIRT_LAZY_MMU);
588 static void xen_write_cr2(unsigned long cr2)
590 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
593 static unsigned long xen_read_cr2(void)
595 return x86_read_percpu(xen_vcpu)->arch.cr2;
598 static unsigned long xen_read_cr2_direct(void)
600 return x86_read_percpu(xen_vcpu_info.arch.cr2);
603 static void xen_write_cr4(unsigned long cr4)
605 /* Just ignore cr4 changes; Xen doesn't allow us to do
609 static unsigned long xen_read_cr3(void)
611 return x86_read_percpu(xen_cr3);
614 static void xen_write_cr3(unsigned long cr3)
616 BUG_ON(preemptible());
618 if (cr3 == x86_read_percpu(xen_cr3)) {
619 /* just a simple tlb flush */
624 x86_write_percpu(xen_cr3, cr3);
628 struct mmuext_op *op;
629 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
630 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
633 op->cmd = MMUEXT_NEW_BASEPTR;
636 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
638 xen_mc_issue(PARAVIRT_LAZY_CPU);
642 /* Early in boot, while setting up the initial pagetable, assume
643 everything is pinned. */
644 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
646 BUG_ON(mem_map); /* should only be used early */
647 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
650 /* This needs to make sure the new pte page is pinned iff its being
651 attached to a pinned pagetable. */
652 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
654 struct page *page = pfn_to_page(pfn);
656 if (PagePinned(virt_to_page(mm->pgd))) {
659 if (!PageHighMem(page))
660 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
662 /* make sure there are no stray mappings of
668 /* This should never happen until we're OK to use struct page */
669 static void xen_release_pt(u32 pfn)
671 struct page *page = pfn_to_page(pfn);
673 if (PagePinned(page)) {
674 if (!PageHighMem(page))
675 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
679 #ifdef CONFIG_HIGHPTE
680 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
682 pgprot_t prot = PAGE_KERNEL;
684 if (PagePinned(page))
685 prot = PAGE_KERNEL_RO;
687 if (0 && PageHighMem(page))
688 printk("mapping highpte %lx type %d prot %s\n",
689 page_to_pfn(page), type,
690 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
692 return kmap_atomic_prot(page, type, prot);
696 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
698 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
699 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
700 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
706 /* Init-time set_pte while constructing initial pagetables, which
707 doesn't allow RO pagetable pages to be remapped RW */
708 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
710 pte = mask_rw_pte(ptep, pte);
712 xen_set_pte(ptep, pte);
715 static __init void xen_pagetable_setup_start(pgd_t *base)
717 pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
719 /* special set_pte for pagetable initialization */
720 pv_mmu_ops.set_pte = xen_set_pte_init;
724 * copy top-level of Xen-supplied pagetable into place. For
725 * !PAE we can use this as-is, but for PAE it is a stand-in
726 * while we copy the pmd pages.
728 memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
730 if (PTRS_PER_PMD > 1) {
733 * For PAE, need to allocate new pmds, rather than
734 * share Xen's, since Xen doesn't like pmd's being
735 * shared between address spaces.
737 for (i = 0; i < PTRS_PER_PGD; i++) {
738 if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
739 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
741 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
744 make_lowmem_page_readonly(pmd);
746 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
752 /* make sure zero_page is mapped RO so we can use it in pagetables */
753 make_lowmem_page_readonly(empty_zero_page);
754 make_lowmem_page_readonly(base);
756 * Switch to new pagetable. This is done before
757 * pagetable_init has done anything so that the new pages
758 * added to the table can be prepared properly for Xen.
760 xen_write_cr3(__pa(base));
763 static __init void xen_pagetable_setup_done(pgd_t *base)
765 /* This will work as long as patching hasn't happened yet
767 pv_mmu_ops.alloc_pt = xen_alloc_pt;
768 pv_mmu_ops.set_pte = xen_set_pte;
770 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
772 * Create a mapping for the shared info page.
773 * Should be set_fixmap(), but shared_info is a machine
774 * address with no corresponding pseudo-phys address.
776 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
777 PFN_DOWN(xen_start_info->shared_info),
780 HYPERVISOR_shared_info =
781 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
784 HYPERVISOR_shared_info =
785 (struct shared_info *)__va(xen_start_info->shared_info);
787 /* Actually pin the pagetable down, but we can't set PG_pinned
788 yet because the page structures don't exist yet. */
791 #ifdef CONFIG_X86_PAE
792 op.cmd = MMUEXT_PIN_L3_TABLE;
794 op.cmd = MMUEXT_PIN_L3_TABLE;
796 op.arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(base)));
797 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
802 /* This is called once we have the cpu_possible_map */
803 void __init xen_setup_vcpu_info_placement(void)
807 for_each_possible_cpu(cpu)
810 /* xen_vcpu_setup managed to place the vcpu_info within the
811 percpu area for all cpus, so make use of it */
812 if (have_vcpu_info_placement) {
813 printk(KERN_INFO "Xen: using vcpu_info placement\n");
815 pv_irq_ops.save_fl = xen_save_fl_direct;
816 pv_irq_ops.restore_fl = xen_restore_fl_direct;
817 pv_irq_ops.irq_disable = xen_irq_disable_direct;
818 pv_irq_ops.irq_enable = xen_irq_enable_direct;
819 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
820 pv_cpu_ops.iret = xen_iret_direct;
824 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
825 unsigned long addr, unsigned len)
827 char *start, *end, *reloc;
830 start = end = reloc = NULL;
832 #define SITE(op, x) \
833 case PARAVIRT_PATCH(op.x): \
834 if (have_vcpu_info_placement) { \
835 start = (char *)xen_##x##_direct; \
836 end = xen_##x##_direct_end; \
837 reloc = xen_##x##_direct_reloc; \
842 SITE(pv_irq_ops, irq_enable);
843 SITE(pv_irq_ops, irq_disable);
844 SITE(pv_irq_ops, save_fl);
845 SITE(pv_irq_ops, restore_fl);
849 if (start == NULL || (end-start) > len)
852 ret = paravirt_patch_insns(insnbuf, len, start, end);
854 /* Note: because reloc is assigned from something that
855 appears to be an array, gcc assumes it's non-null,
856 but doesn't know its relationship with start and
858 if (reloc > start && reloc < end) {
859 int reloc_off = reloc - start;
860 long *relocp = (long *)(insnbuf + reloc_off);
861 long delta = start - (char *)addr;
869 ret = paravirt_patch_default(type, clobbers, insnbuf,
877 static const struct pv_info xen_info __initdata = {
878 .paravirt_enabled = 1,
879 .shared_kernel_pmd = 0,
884 static const struct pv_init_ops xen_init_ops __initdata = {
887 .banner = xen_banner,
888 .memory_setup = xen_memory_setup,
889 .arch_setup = xen_arch_setup,
890 .post_allocator_init = xen_mark_init_mm_pinned,
893 static const struct pv_time_ops xen_time_ops __initdata = {
894 .time_init = xen_time_init,
896 .set_wallclock = xen_set_wallclock,
897 .get_wallclock = xen_get_wallclock,
898 .get_cpu_khz = xen_cpu_khz,
899 .sched_clock = xen_sched_clock,
902 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
905 .set_debugreg = xen_set_debugreg,
906 .get_debugreg = xen_get_debugreg,
910 .read_cr0 = native_read_cr0,
911 .write_cr0 = native_write_cr0,
913 .read_cr4 = native_read_cr4,
914 .read_cr4_safe = native_read_cr4_safe,
915 .write_cr4 = xen_write_cr4,
917 .wbinvd = native_wbinvd,
919 .read_msr = native_read_msr_safe,
920 .write_msr = native_write_msr_safe,
921 .read_tsc = native_read_tsc,
922 .read_pmc = native_read_pmc,
924 .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
925 .irq_enable_sysexit = NULL, /* never called */
927 .load_tr_desc = paravirt_nop,
928 .set_ldt = xen_set_ldt,
929 .load_gdt = xen_load_gdt,
930 .load_idt = xen_load_idt,
931 .load_tls = xen_load_tls,
933 .store_gdt = native_store_gdt,
934 .store_idt = native_store_idt,
935 .store_tr = xen_store_tr,
937 .write_ldt_entry = xen_write_ldt_entry,
938 .write_gdt_entry = xen_write_gdt_entry,
939 .write_idt_entry = xen_write_idt_entry,
940 .load_esp0 = xen_load_esp0,
942 .set_iopl_mask = xen_set_iopl_mask,
943 .io_delay = xen_io_delay,
946 .enter = paravirt_enter_lazy_cpu,
947 .leave = xen_leave_lazy,
951 static const struct pv_irq_ops xen_irq_ops __initdata = {
952 .init_IRQ = xen_init_IRQ,
953 .save_fl = xen_save_fl,
954 .restore_fl = xen_restore_fl,
955 .irq_disable = xen_irq_disable,
956 .irq_enable = xen_irq_enable,
957 .safe_halt = xen_safe_halt,
961 static const struct pv_apic_ops xen_apic_ops __initdata = {
962 #ifdef CONFIG_X86_LOCAL_APIC
963 .apic_write = xen_apic_write,
964 .apic_write_atomic = xen_apic_write,
965 .apic_read = xen_apic_read,
966 .setup_boot_clock = paravirt_nop,
967 .setup_secondary_clock = paravirt_nop,
968 .startup_ipi_hook = paravirt_nop,
972 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
973 .pagetable_setup_start = xen_pagetable_setup_start,
974 .pagetable_setup_done = xen_pagetable_setup_done,
976 .read_cr2 = xen_read_cr2,
977 .write_cr2 = xen_write_cr2,
979 .read_cr3 = xen_read_cr3,
980 .write_cr3 = xen_write_cr3,
982 .flush_tlb_user = xen_flush_tlb,
983 .flush_tlb_kernel = xen_flush_tlb,
984 .flush_tlb_single = xen_flush_tlb_single,
985 .flush_tlb_others = xen_flush_tlb_others,
987 .pte_update = paravirt_nop,
988 .pte_update_defer = paravirt_nop,
990 .alloc_pt = xen_alloc_pt_init,
991 .release_pt = xen_release_pt,
992 .alloc_pd = paravirt_nop,
993 .alloc_pd_clone = paravirt_nop,
994 .release_pd = paravirt_nop,
996 #ifdef CONFIG_HIGHPTE
997 .kmap_atomic_pte = xen_kmap_atomic_pte,
1000 .set_pte = NULL, /* see xen_pagetable_setup_* */
1001 .set_pte_at = xen_set_pte_at,
1002 .set_pmd = xen_set_pmd,
1004 .pte_val = xen_pte_val,
1005 .pgd_val = xen_pgd_val,
1007 .make_pte = xen_make_pte,
1008 .make_pgd = xen_make_pgd,
1010 #ifdef CONFIG_X86_PAE
1011 .set_pte_atomic = xen_set_pte_atomic,
1012 .set_pte_present = xen_set_pte_at,
1013 .set_pud = xen_set_pud,
1014 .pte_clear = xen_pte_clear,
1015 .pmd_clear = xen_pmd_clear,
1017 .make_pmd = xen_make_pmd,
1018 .pmd_val = xen_pmd_val,
1021 .activate_mm = xen_activate_mm,
1022 .dup_mmap = xen_dup_mmap,
1023 .exit_mmap = xen_exit_mmap,
1026 .enter = paravirt_enter_lazy_mmu,
1027 .leave = xen_leave_lazy,
1032 static const struct smp_ops xen_smp_ops __initdata = {
1033 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
1034 .smp_prepare_cpus = xen_smp_prepare_cpus,
1035 .cpu_up = xen_cpu_up,
1036 .smp_cpus_done = xen_smp_cpus_done,
1038 .smp_send_stop = xen_smp_send_stop,
1039 .smp_send_reschedule = xen_smp_send_reschedule,
1040 .smp_call_function_mask = xen_smp_call_function_mask,
1042 #endif /* CONFIG_SMP */
1044 static void xen_reboot(int reason)
1050 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason))
1054 static void xen_restart(char *msg)
1056 xen_reboot(SHUTDOWN_reboot);
1059 static void xen_emergency_restart(void)
1061 xen_reboot(SHUTDOWN_reboot);
1064 static void xen_machine_halt(void)
1066 xen_reboot(SHUTDOWN_poweroff);
1069 static void xen_crash_shutdown(struct pt_regs *regs)
1071 xen_reboot(SHUTDOWN_crash);
1074 static const struct machine_ops __initdata xen_machine_ops = {
1075 .restart = xen_restart,
1076 .halt = xen_machine_halt,
1077 .power_off = xen_machine_halt,
1078 .shutdown = xen_machine_halt,
1079 .crash_shutdown = xen_crash_shutdown,
1080 .emergency_restart = xen_emergency_restart,
1084 /* First C function to be called on Xen boot */
1085 asmlinkage void __init xen_start_kernel(void)
1089 if (!xen_start_info)
1092 BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0);
1094 /* Install Xen paravirt ops */
1096 pv_init_ops = xen_init_ops;
1097 pv_time_ops = xen_time_ops;
1098 pv_cpu_ops = xen_cpu_ops;
1099 pv_irq_ops = xen_irq_ops;
1100 pv_apic_ops = xen_apic_ops;
1101 pv_mmu_ops = xen_mmu_ops;
1103 machine_ops = xen_machine_ops;
1106 smp_ops = xen_smp_ops;
1109 xen_setup_features();
1112 if (!xen_feature(XENFEAT_auto_translated_physmap))
1113 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
1115 pgd = (pgd_t *)xen_start_info->pt_base;
1117 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1119 init_mm.pgd = pgd; /* use the Xen pagetables to start */
1121 /* keep using Xen gdt for now; no urgent need to change it */
1123 x86_write_percpu(xen_cr3, __pa(pgd));
1126 /* Don't do the full vcpu_info placement stuff until we have a
1128 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1130 /* May as well do it now, since there's no good time to call
1132 xen_setup_vcpu_info_placement();
1135 pv_info.kernel_rpl = 1;
1136 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1137 pv_info.kernel_rpl = 0;
1139 /* set the limit of our address space */
1140 reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE);
1142 /* set up basic CPUID stuff */
1143 cpu_detect(&new_cpu_data);
1144 new_cpu_data.hard_math = 1;
1145 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1147 /* Poke various useful things into boot_params */
1148 LOADER_TYPE = (9 << 4) | 0;
1149 INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0;
1150 INITRD_SIZE = xen_start_info->mod_len;
1152 /* Start the world */