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>
29 #include <xen/interface/xen.h>
30 #include <xen/interface/physdev.h>
31 #include <xen/interface/vcpu.h>
32 #include <xen/interface/sched.h>
33 #include <xen/features.h>
36 #include <asm/paravirt.h>
38 #include <asm/xen/hypercall.h>
39 #include <asm/xen/hypervisor.h>
40 #include <asm/fixmap.h>
41 #include <asm/processor.h>
42 #include <asm/setup.h>
44 #include <asm/pgtable.h>
45 #include <asm/tlbflush.h>
46 #include <asm/reboot.h>
50 #include "multicalls.h"
52 EXPORT_SYMBOL_GPL(hypercall_page);
54 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
55 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
58 * Note about cr3 (pagetable base) values:
60 * xen_cr3 contains the current logical cr3 value; it contains the
61 * last set cr3. This may not be the current effective cr3, because
62 * its update may be being lazily deferred. However, a vcpu looking
63 * at its own cr3 can use this value knowing that it everything will
66 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
67 * hypercall to set the vcpu cr3 is complete (so it may be a little
68 * out of date, but it will never be set early). If one vcpu is
69 * looking at another vcpu's cr3 value, it should use this variable.
71 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
72 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
74 struct start_info *xen_start_info;
75 EXPORT_SYMBOL_GPL(xen_start_info);
77 static /* __initdata */ struct shared_info dummy_shared_info;
80 * Point at some empty memory to start with. We map the real shared_info
81 * page as soon as fixmap is up and running.
83 struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info;
86 * Flag to determine whether vcpu info placement is available on all
87 * VCPUs. We assume it is to start with, and then set it to zero on
88 * the first failure. This is because it can succeed on some VCPUs
89 * and not others, since it can involve hypervisor memory allocation,
90 * or because the guest failed to guarantee all the appropriate
91 * constraints on all VCPUs (ie buffer can't cross a page boundary).
93 * Note that any particular CPU may be using a placed vcpu structure,
94 * but we can only optimise if the all are.
96 * 0: not available, 1: available
98 static int have_vcpu_info_placement = 0;
100 static void __init xen_vcpu_setup(int cpu)
102 struct vcpu_register_vcpu_info info;
104 struct vcpu_info *vcpup;
106 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
108 if (!have_vcpu_info_placement)
109 return; /* already tested, not available */
111 vcpup = &per_cpu(xen_vcpu_info, cpu);
113 info.mfn = virt_to_mfn(vcpup);
114 info.offset = offset_in_page(vcpup);
116 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
117 cpu, vcpup, info.mfn, info.offset);
119 /* Check to see if the hypervisor will put the vcpu_info
120 structure where we want it, which allows direct access via
121 a percpu-variable. */
122 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
125 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
126 have_vcpu_info_placement = 0;
128 /* This cpu is using the registered vcpu info, even if
129 later ones fail to. */
130 per_cpu(xen_vcpu, cpu) = vcpup;
132 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
137 static void __init xen_banner(void)
139 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
141 printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
144 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
145 unsigned int *cx, unsigned int *dx)
147 unsigned maskedx = ~0;
150 * Mask out inconvenient features, to try and disable as many
151 * unsupported kernel subsystems as possible.
154 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
155 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
156 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
158 asm(XEN_EMULATE_PREFIX "cpuid"
163 : "0" (*ax), "2" (*cx));
167 static void xen_set_debugreg(int reg, unsigned long val)
169 HYPERVISOR_set_debugreg(reg, val);
172 static unsigned long xen_get_debugreg(int reg)
174 return HYPERVISOR_get_debugreg(reg);
177 static unsigned long xen_save_fl(void)
179 struct vcpu_info *vcpu;
182 vcpu = x86_read_percpu(xen_vcpu);
184 /* flag has opposite sense of mask */
185 flags = !vcpu->evtchn_upcall_mask;
187 /* convert to IF type flag
191 return (-flags) & X86_EFLAGS_IF;
194 static void xen_restore_fl(unsigned long flags)
196 struct vcpu_info *vcpu;
198 /* convert from IF type flag */
199 flags = !(flags & X86_EFLAGS_IF);
201 /* There's a one instruction preempt window here. We need to
202 make sure we're don't switch CPUs between getting the vcpu
203 pointer and updating the mask. */
205 vcpu = x86_read_percpu(xen_vcpu);
206 vcpu->evtchn_upcall_mask = flags;
207 preempt_enable_no_resched();
209 /* Doesn't matter if we get preempted here, because any
210 pending event will get dealt with anyway. */
213 preempt_check_resched();
214 barrier(); /* unmask then check (avoid races) */
215 if (unlikely(vcpu->evtchn_upcall_pending))
216 force_evtchn_callback();
220 static void xen_irq_disable(void)
222 /* There's a one instruction preempt window here. We need to
223 make sure we're don't switch CPUs between getting the vcpu
224 pointer and updating the mask. */
226 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
227 preempt_enable_no_resched();
230 static void xen_irq_enable(void)
232 struct vcpu_info *vcpu;
234 /* There's a one instruction preempt window here. We need to
235 make sure we're don't switch CPUs between getting the vcpu
236 pointer and updating the mask. */
238 vcpu = x86_read_percpu(xen_vcpu);
239 vcpu->evtchn_upcall_mask = 0;
240 preempt_enable_no_resched();
242 /* Doesn't matter if we get preempted here, because any
243 pending event will get dealt with anyway. */
245 barrier(); /* unmask then check (avoid races) */
246 if (unlikely(vcpu->evtchn_upcall_pending))
247 force_evtchn_callback();
250 static void xen_safe_halt(void)
252 /* Blocking includes an implicit local_irq_enable(). */
253 if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
257 static void xen_halt(void)
260 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
265 static void xen_leave_lazy(void)
267 paravirt_leave_lazy(paravirt_get_lazy_mode());
271 static unsigned long xen_store_tr(void)
276 static void xen_set_ldt(const void *addr, unsigned entries)
278 struct mmuext_op *op;
279 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
282 op->cmd = MMUEXT_SET_LDT;
283 op->arg1.linear_addr = (unsigned long)addr;
284 op->arg2.nr_ents = entries;
286 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
288 xen_mc_issue(PARAVIRT_LAZY_CPU);
291 static void xen_load_gdt(const struct desc_ptr *dtr)
293 unsigned long *frames;
294 unsigned long va = dtr->address;
295 unsigned int size = dtr->size + 1;
296 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
298 struct multicall_space mcs;
300 /* A GDT can be up to 64k in size, which corresponds to 8192
301 8-byte entries, or 16 4k pages.. */
303 BUG_ON(size > 65536);
304 BUG_ON(va & ~PAGE_MASK);
306 mcs = xen_mc_entry(sizeof(*frames) * pages);
309 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
310 frames[f] = virt_to_mfn(va);
311 make_lowmem_page_readonly((void *)va);
314 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
316 xen_mc_issue(PARAVIRT_LAZY_CPU);
319 static void load_TLS_descriptor(struct thread_struct *t,
320 unsigned int cpu, unsigned int i)
322 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
323 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
324 struct multicall_space mc = __xen_mc_entry(0);
326 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
329 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
333 load_TLS_descriptor(t, cpu, 0);
334 load_TLS_descriptor(t, cpu, 1);
335 load_TLS_descriptor(t, cpu, 2);
337 xen_mc_issue(PARAVIRT_LAZY_CPU);
340 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
341 * it means we're in a context switch, and %gs has just been
342 * saved. This means we can zero it out to prevent faults on
343 * exit from the hypervisor if the next process has no %gs.
344 * Either way, it has been saved, and the new value will get
345 * loaded properly. This will go away as soon as Xen has been
346 * modified to not save/restore %gs for normal hypercalls.
348 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
352 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
355 unsigned long lp = (unsigned long)&dt[entrynum];
356 xmaddr_t mach_lp = virt_to_machine(lp);
357 u64 entry = *(u64 *)ptr;
362 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
368 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
369 struct trap_info *info)
373 type = (high >> 8) & 0x1f;
374 dpl = (high >> 13) & 3;
376 if (type != 0xf && type != 0xe)
379 info->vector = vector;
380 info->address = (high & 0xffff0000) | (low & 0x0000ffff);
381 info->cs = low >> 16;
383 /* interrupt gates clear IF */
390 /* Locations of each CPU's IDT */
391 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
393 /* Set an IDT entry. If the entry is part of the current IDT, then
395 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
397 unsigned long p = (unsigned long)&dt[entrynum];
398 unsigned long start, end;
402 start = __get_cpu_var(idt_desc).address;
403 end = start + __get_cpu_var(idt_desc).size + 1;
407 native_write_idt_entry(dt, entrynum, g);
409 if (p >= start && (p + 8) <= end) {
410 struct trap_info info[2];
411 u32 *desc = (u32 *)g;
415 if (cvt_gate_to_trap(entrynum, desc[0], desc[1], &info[0]))
416 if (HYPERVISOR_set_trap_table(info))
423 static void xen_convert_trap_info(const struct desc_ptr *desc,
424 struct trap_info *traps)
426 unsigned in, out, count;
428 count = (desc->size+1) / 8;
431 for (in = out = 0; in < count; in++) {
432 const u32 *entry = (u32 *)(desc->address + in * 8);
434 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
437 traps[out].address = 0;
440 void xen_copy_trap_info(struct trap_info *traps)
442 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
444 xen_convert_trap_info(desc, traps);
447 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
448 hold a spinlock to protect the static traps[] array (static because
449 it avoids allocation, and saves stack space). */
450 static void xen_load_idt(const struct desc_ptr *desc)
452 static DEFINE_SPINLOCK(lock);
453 static struct trap_info traps[257];
457 __get_cpu_var(idt_desc) = *desc;
459 xen_convert_trap_info(desc, traps);
462 if (HYPERVISOR_set_trap_table(traps))
468 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
469 they're handled differently. */
470 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
471 const void *desc, int type)
482 xmaddr_t maddr = virt_to_machine(&dt[entry]);
485 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
494 static void xen_load_sp0(struct tss_struct *tss,
495 struct thread_struct *thread)
497 struct multicall_space mcs = xen_mc_entry(0);
498 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
499 xen_mc_issue(PARAVIRT_LAZY_CPU);
502 static void xen_set_iopl_mask(unsigned mask)
504 struct physdev_set_iopl set_iopl;
506 /* Force the change at ring 0. */
507 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
508 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
511 static void xen_io_delay(void)
515 #ifdef CONFIG_X86_LOCAL_APIC
516 static u32 xen_apic_read(unsigned long reg)
521 static void xen_apic_write(unsigned long reg, u32 val)
523 /* Warn to see if there's any stray references */
528 static void xen_flush_tlb(void)
530 struct mmuext_op *op;
531 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
534 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
535 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
537 xen_mc_issue(PARAVIRT_LAZY_MMU);
540 static void xen_flush_tlb_single(unsigned long addr)
542 struct mmuext_op *op;
543 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
546 op->cmd = MMUEXT_INVLPG_LOCAL;
547 op->arg1.linear_addr = addr & PAGE_MASK;
548 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
550 xen_mc_issue(PARAVIRT_LAZY_MMU);
553 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
560 cpumask_t cpumask = *cpus;
561 struct multicall_space mcs;
564 * A couple of (to be removed) sanity checks:
566 * - current CPU must not be in mask
567 * - mask must exist :)
569 BUG_ON(cpus_empty(cpumask));
570 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
573 /* If a CPU which we ran on has gone down, OK. */
574 cpus_and(cpumask, cpumask, cpu_online_map);
575 if (cpus_empty(cpumask))
578 mcs = xen_mc_entry(sizeof(*args));
580 args->mask = cpumask;
581 args->op.arg2.vcpumask = &args->mask;
583 if (va == TLB_FLUSH_ALL) {
584 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
586 args->op.cmd = MMUEXT_INVLPG_MULTI;
587 args->op.arg1.linear_addr = va;
590 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
592 xen_mc_issue(PARAVIRT_LAZY_MMU);
595 static void xen_write_cr2(unsigned long cr2)
597 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
600 static unsigned long xen_read_cr2(void)
602 return x86_read_percpu(xen_vcpu)->arch.cr2;
605 static unsigned long xen_read_cr2_direct(void)
607 return x86_read_percpu(xen_vcpu_info.arch.cr2);
610 static void xen_write_cr4(unsigned long cr4)
612 /* Just ignore cr4 changes; Xen doesn't allow us to do
616 static unsigned long xen_read_cr3(void)
618 return x86_read_percpu(xen_cr3);
621 static void set_current_cr3(void *v)
623 x86_write_percpu(xen_current_cr3, (unsigned long)v);
626 static void xen_write_cr3(unsigned long cr3)
628 struct mmuext_op *op;
629 struct multicall_space mcs;
630 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
632 BUG_ON(preemptible());
634 mcs = xen_mc_entry(sizeof(*op)); /* disables interrupts */
636 /* Update while interrupts are disabled, so its atomic with
638 x86_write_percpu(xen_cr3, cr3);
641 op->cmd = MMUEXT_NEW_BASEPTR;
644 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
646 /* Update xen_update_cr3 once the batch has actually
648 xen_mc_callback(set_current_cr3, (void *)cr3);
650 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
653 /* Early in boot, while setting up the initial pagetable, assume
654 everything is pinned. */
655 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
657 BUG_ON(mem_map); /* should only be used early */
658 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
661 /* Early release_pt assumes that all pts are pinned, since there's
662 only init_mm and anything attached to that is pinned. */
663 static void xen_release_pt_init(u32 pfn)
665 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
668 static void pin_pagetable_pfn(unsigned level, unsigned long pfn)
672 op.arg1.mfn = pfn_to_mfn(pfn);
673 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
677 /* This needs to make sure the new pte page is pinned iff its being
678 attached to a pinned pagetable. */
679 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
681 struct page *page = pfn_to_page(pfn);
683 if (PagePinned(virt_to_page(mm->pgd))) {
686 if (!PageHighMem(page)) {
687 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
688 pin_pagetable_pfn(level, pfn);
690 /* make sure there are no stray mappings of
696 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
698 xen_alloc_ptpage(mm, pfn, MMUEXT_PIN_L1_TABLE);
701 static void xen_alloc_pd(struct mm_struct *mm, u32 pfn)
703 xen_alloc_ptpage(mm, pfn, MMUEXT_PIN_L2_TABLE);
706 /* This should never happen until we're OK to use struct page */
707 static void xen_release_pt(u32 pfn)
709 struct page *page = pfn_to_page(pfn);
711 if (PagePinned(page)) {
712 if (!PageHighMem(page)) {
713 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
714 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
719 #ifdef CONFIG_HIGHPTE
720 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
722 pgprot_t prot = PAGE_KERNEL;
724 if (PagePinned(page))
725 prot = PAGE_KERNEL_RO;
727 if (0 && PageHighMem(page))
728 printk("mapping highpte %lx type %d prot %s\n",
729 page_to_pfn(page), type,
730 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
732 return kmap_atomic_prot(page, type, prot);
736 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
738 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
739 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
740 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
746 /* Init-time set_pte while constructing initial pagetables, which
747 doesn't allow RO pagetable pages to be remapped RW */
748 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
750 pte = mask_rw_pte(ptep, pte);
752 xen_set_pte(ptep, pte);
755 static __init void xen_pagetable_setup_start(pgd_t *base)
757 pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
759 /* special set_pte for pagetable initialization */
760 pv_mmu_ops.set_pte = xen_set_pte_init;
764 * copy top-level of Xen-supplied pagetable into place. For
765 * !PAE we can use this as-is, but for PAE it is a stand-in
766 * while we copy the pmd pages.
768 memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
770 if (PTRS_PER_PMD > 1) {
773 * For PAE, need to allocate new pmds, rather than
774 * share Xen's, since Xen doesn't like pmd's being
775 * shared between address spaces.
777 for (i = 0; i < PTRS_PER_PGD; i++) {
778 if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
779 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
781 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
784 make_lowmem_page_readonly(pmd);
786 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
792 /* make sure zero_page is mapped RO so we can use it in pagetables */
793 make_lowmem_page_readonly(empty_zero_page);
794 make_lowmem_page_readonly(base);
796 * Switch to new pagetable. This is done before
797 * pagetable_init has done anything so that the new pages
798 * added to the table can be prepared properly for Xen.
800 xen_write_cr3(__pa(base));
802 /* Unpin initial Xen pagetable */
803 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
804 PFN_DOWN(__pa(xen_start_info->pt_base)));
807 static __init void xen_pagetable_setup_done(pgd_t *base)
809 /* This will work as long as patching hasn't happened yet
811 pv_mmu_ops.alloc_pt = xen_alloc_pt;
812 pv_mmu_ops.alloc_pd = xen_alloc_pd;
813 pv_mmu_ops.release_pt = xen_release_pt;
814 pv_mmu_ops.release_pd = xen_release_pt;
815 pv_mmu_ops.set_pte = xen_set_pte;
817 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
819 * Create a mapping for the shared info page.
820 * Should be set_fixmap(), but shared_info is a machine
821 * address with no corresponding pseudo-phys address.
823 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
824 PFN_DOWN(xen_start_info->shared_info),
827 HYPERVISOR_shared_info =
828 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
831 HYPERVISOR_shared_info =
832 (struct shared_info *)__va(xen_start_info->shared_info);
834 /* Actually pin the pagetable down, but we can't set PG_pinned
835 yet because the page structures don't exist yet. */
839 #ifdef CONFIG_X86_PAE
840 level = MMUEXT_PIN_L3_TABLE;
842 level = MMUEXT_PIN_L2_TABLE;
845 pin_pagetable_pfn(level, PFN_DOWN(__pa(base)));
849 /* This is called once we have the cpu_possible_map */
850 void __init xen_setup_vcpu_info_placement(void)
854 for_each_possible_cpu(cpu)
857 /* xen_vcpu_setup managed to place the vcpu_info within the
858 percpu area for all cpus, so make use of it */
859 if (have_vcpu_info_placement) {
860 printk(KERN_INFO "Xen: using vcpu_info placement\n");
862 pv_irq_ops.save_fl = xen_save_fl_direct;
863 pv_irq_ops.restore_fl = xen_restore_fl_direct;
864 pv_irq_ops.irq_disable = xen_irq_disable_direct;
865 pv_irq_ops.irq_enable = xen_irq_enable_direct;
866 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
867 pv_cpu_ops.iret = xen_iret_direct;
871 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
872 unsigned long addr, unsigned len)
874 char *start, *end, *reloc;
877 start = end = reloc = NULL;
879 #define SITE(op, x) \
880 case PARAVIRT_PATCH(op.x): \
881 if (have_vcpu_info_placement) { \
882 start = (char *)xen_##x##_direct; \
883 end = xen_##x##_direct_end; \
884 reloc = xen_##x##_direct_reloc; \
889 SITE(pv_irq_ops, irq_enable);
890 SITE(pv_irq_ops, irq_disable);
891 SITE(pv_irq_ops, save_fl);
892 SITE(pv_irq_ops, restore_fl);
896 if (start == NULL || (end-start) > len)
899 ret = paravirt_patch_insns(insnbuf, len, start, end);
901 /* Note: because reloc is assigned from something that
902 appears to be an array, gcc assumes it's non-null,
903 but doesn't know its relationship with start and
905 if (reloc > start && reloc < end) {
906 int reloc_off = reloc - start;
907 long *relocp = (long *)(insnbuf + reloc_off);
908 long delta = start - (char *)addr;
916 ret = paravirt_patch_default(type, clobbers, insnbuf,
924 static const struct pv_info xen_info __initdata = {
925 .paravirt_enabled = 1,
926 .shared_kernel_pmd = 0,
931 static const struct pv_init_ops xen_init_ops __initdata = {
934 .banner = xen_banner,
935 .memory_setup = xen_memory_setup,
936 .arch_setup = xen_arch_setup,
937 .post_allocator_init = xen_mark_init_mm_pinned,
940 static const struct pv_time_ops xen_time_ops __initdata = {
941 .time_init = xen_time_init,
943 .set_wallclock = xen_set_wallclock,
944 .get_wallclock = xen_get_wallclock,
945 .get_cpu_khz = xen_cpu_khz,
946 .sched_clock = xen_sched_clock,
949 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
952 .set_debugreg = xen_set_debugreg,
953 .get_debugreg = xen_get_debugreg,
957 .read_cr0 = native_read_cr0,
958 .write_cr0 = native_write_cr0,
960 .read_cr4 = native_read_cr4,
961 .read_cr4_safe = native_read_cr4_safe,
962 .write_cr4 = xen_write_cr4,
964 .wbinvd = native_wbinvd,
966 .read_msr = native_read_msr_safe,
967 .write_msr = native_write_msr_safe,
968 .read_tsc = native_read_tsc,
969 .read_pmc = native_read_pmc,
971 .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
972 .irq_enable_syscall_ret = NULL, /* never called */
974 .load_tr_desc = paravirt_nop,
975 .set_ldt = xen_set_ldt,
976 .load_gdt = xen_load_gdt,
977 .load_idt = xen_load_idt,
978 .load_tls = xen_load_tls,
980 .store_gdt = native_store_gdt,
981 .store_idt = native_store_idt,
982 .store_tr = xen_store_tr,
984 .write_ldt_entry = xen_write_ldt_entry,
985 .write_gdt_entry = xen_write_gdt_entry,
986 .write_idt_entry = xen_write_idt_entry,
987 .load_sp0 = xen_load_sp0,
989 .set_iopl_mask = xen_set_iopl_mask,
990 .io_delay = xen_io_delay,
993 .enter = paravirt_enter_lazy_cpu,
994 .leave = xen_leave_lazy,
998 static const struct pv_irq_ops xen_irq_ops __initdata = {
999 .init_IRQ = xen_init_IRQ,
1000 .save_fl = xen_save_fl,
1001 .restore_fl = xen_restore_fl,
1002 .irq_disable = xen_irq_disable,
1003 .irq_enable = xen_irq_enable,
1004 .safe_halt = xen_safe_halt,
1008 static const struct pv_apic_ops xen_apic_ops __initdata = {
1009 #ifdef CONFIG_X86_LOCAL_APIC
1010 .apic_write = xen_apic_write,
1011 .apic_write_atomic = xen_apic_write,
1012 .apic_read = xen_apic_read,
1013 .setup_boot_clock = paravirt_nop,
1014 .setup_secondary_clock = paravirt_nop,
1015 .startup_ipi_hook = paravirt_nop,
1019 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1020 .pagetable_setup_start = xen_pagetable_setup_start,
1021 .pagetable_setup_done = xen_pagetable_setup_done,
1023 .read_cr2 = xen_read_cr2,
1024 .write_cr2 = xen_write_cr2,
1026 .read_cr3 = xen_read_cr3,
1027 .write_cr3 = xen_write_cr3,
1029 .flush_tlb_user = xen_flush_tlb,
1030 .flush_tlb_kernel = xen_flush_tlb,
1031 .flush_tlb_single = xen_flush_tlb_single,
1032 .flush_tlb_others = xen_flush_tlb_others,
1034 .pte_update = paravirt_nop,
1035 .pte_update_defer = paravirt_nop,
1037 .alloc_pt = xen_alloc_pt_init,
1038 .release_pt = xen_release_pt_init,
1039 .alloc_pd = xen_alloc_pt_init,
1040 .alloc_pd_clone = paravirt_nop,
1041 .release_pd = xen_release_pt_init,
1043 #ifdef CONFIG_HIGHPTE
1044 .kmap_atomic_pte = xen_kmap_atomic_pte,
1047 .set_pte = NULL, /* see xen_pagetable_setup_* */
1048 .set_pte_at = xen_set_pte_at,
1049 .set_pmd = xen_set_pmd,
1051 .pte_val = xen_pte_val,
1052 .pgd_val = xen_pgd_val,
1054 .make_pte = xen_make_pte,
1055 .make_pgd = xen_make_pgd,
1057 #ifdef CONFIG_X86_PAE
1058 .set_pte_atomic = xen_set_pte_atomic,
1059 .set_pte_present = xen_set_pte_at,
1060 .set_pud = xen_set_pud,
1061 .pte_clear = xen_pte_clear,
1062 .pmd_clear = xen_pmd_clear,
1064 .make_pmd = xen_make_pmd,
1065 .pmd_val = xen_pmd_val,
1068 .activate_mm = xen_activate_mm,
1069 .dup_mmap = xen_dup_mmap,
1070 .exit_mmap = xen_exit_mmap,
1073 .enter = paravirt_enter_lazy_mmu,
1074 .leave = xen_leave_lazy,
1079 static const struct smp_ops xen_smp_ops __initdata = {
1080 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
1081 .smp_prepare_cpus = xen_smp_prepare_cpus,
1082 .cpu_up = xen_cpu_up,
1083 .smp_cpus_done = xen_smp_cpus_done,
1085 .smp_send_stop = xen_smp_send_stop,
1086 .smp_send_reschedule = xen_smp_send_reschedule,
1087 .smp_call_function_mask = xen_smp_call_function_mask,
1089 #endif /* CONFIG_SMP */
1091 static void xen_reboot(int reason)
1097 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason))
1101 static void xen_restart(char *msg)
1103 xen_reboot(SHUTDOWN_reboot);
1106 static void xen_emergency_restart(void)
1108 xen_reboot(SHUTDOWN_reboot);
1111 static void xen_machine_halt(void)
1113 xen_reboot(SHUTDOWN_poweroff);
1116 static void xen_crash_shutdown(struct pt_regs *regs)
1118 xen_reboot(SHUTDOWN_crash);
1121 static const struct machine_ops __initdata xen_machine_ops = {
1122 .restart = xen_restart,
1123 .halt = xen_machine_halt,
1124 .power_off = xen_machine_halt,
1125 .shutdown = xen_machine_halt,
1126 .crash_shutdown = xen_crash_shutdown,
1127 .emergency_restart = xen_emergency_restart,
1131 static void __init xen_reserve_top(void)
1133 unsigned long top = HYPERVISOR_VIRT_START;
1134 struct xen_platform_parameters pp;
1136 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1137 top = pp.virt_start;
1139 reserve_top_address(-top + 2 * PAGE_SIZE);
1142 /* First C function to be called on Xen boot */
1143 asmlinkage void __init xen_start_kernel(void)
1147 if (!xen_start_info)
1150 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1152 /* Install Xen paravirt ops */
1154 pv_init_ops = xen_init_ops;
1155 pv_time_ops = xen_time_ops;
1156 pv_cpu_ops = xen_cpu_ops;
1157 pv_irq_ops = xen_irq_ops;
1158 pv_apic_ops = xen_apic_ops;
1159 pv_mmu_ops = xen_mmu_ops;
1161 machine_ops = xen_machine_ops;
1164 smp_ops = xen_smp_ops;
1167 xen_setup_features();
1170 if (!xen_feature(XENFEAT_auto_translated_physmap))
1171 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
1173 pgd = (pgd_t *)xen_start_info->pt_base;
1175 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1177 init_mm.pgd = pgd; /* use the Xen pagetables to start */
1179 /* keep using Xen gdt for now; no urgent need to change it */
1181 x86_write_percpu(xen_cr3, __pa(pgd));
1182 x86_write_percpu(xen_current_cr3, __pa(pgd));
1185 /* Don't do the full vcpu_info placement stuff until we have a
1187 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1189 /* May as well do it now, since there's no good time to call
1191 xen_setup_vcpu_info_placement();
1194 pv_info.kernel_rpl = 1;
1195 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1196 pv_info.kernel_rpl = 0;
1198 /* set the limit of our address space */
1201 /* set up basic CPUID stuff */
1202 cpu_detect(&new_cpu_data);
1203 new_cpu_data.hard_math = 1;
1204 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1206 /* Poke various useful things into boot_params */
1207 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1208 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1209 ? __pa(xen_start_info->mod_start) : 0;
1210 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1212 /* Start the world */