xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
1 /*
2  * Core of Xen paravirt_ops implementation.
3  *
4  * This file contains the xen_paravirt_ops structure itself, and the
5  * implementations for:
6  * - privileged instructions
7  * - interrupt flags
8  * - segment operations
9  * - booting and setup
10  *
11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12  */
13 
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>
25 #include <linux/mm.h>
26 #include <linux/page-flags.h>
27 #include <linux/highmem.h>
28 #include <linux/console.h>
29 
30 #include <xen/interface/xen.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/features.h>
34 #include <xen/page.h>
35 #include <xen/hvc-console.h>
36 
37 #include <asm/paravirt.h>
38 #include <asm/apic.h>
39 #include <asm/page.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/msr-index.h>
45 #include <asm/setup.h>
46 #include <asm/desc.h>
47 #include <asm/pgtable.h>
48 #include <asm/tlbflush.h>
49 #include <asm/reboot.h>
50 
51 #include "xen-ops.h"
52 #include "mmu.h"
53 #include "multicalls.h"
54 
55 EXPORT_SYMBOL_GPL(hypercall_page);
56 
57 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
58 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
59 
60 enum xen_domain_type xen_domain_type = XEN_NATIVE;
61 EXPORT_SYMBOL_GPL(xen_domain_type);
62 
63 /*
64  * Identity map, in addition to plain kernel map.  This needs to be
65  * large enough to allocate page table pages to allocate the rest.
66  * Each page can map 2MB.
67  */
68 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
69 
70 #ifdef CONFIG_X86_64
71 /* l3 pud for userspace vsyscall mapping */
72 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
73 #endif /* CONFIG_X86_64 */
74 
75 /*
76  * Note about cr3 (pagetable base) values:
77  *
78  * xen_cr3 contains the current logical cr3 value; it contains the
79  * last set cr3.  This may not be the current effective cr3, because
80  * its update may be being lazily deferred.  However, a vcpu looking
81  * at its own cr3 can use this value knowing that it everything will
82  * be self-consistent.
83  *
84  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
85  * hypercall to set the vcpu cr3 is complete (so it may be a little
86  * out of date, but it will never be set early).  If one vcpu is
87  * looking at another vcpu's cr3 value, it should use this variable.
88  */
89 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
90 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
91 
92 struct start_info *xen_start_info;
93 EXPORT_SYMBOL_GPL(xen_start_info);
94 
95 struct shared_info xen_dummy_shared_info;
96 
97 /*
98  * Point at some empty memory to start with. We map the real shared_info
99  * page as soon as fixmap is up and running.
100  */
101 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
102 
103 /*
104  * Flag to determine whether vcpu info placement is available on all
105  * VCPUs.  We assume it is to start with, and then set it to zero on
106  * the first failure.  This is because it can succeed on some VCPUs
107  * and not others, since it can involve hypervisor memory allocation,
108  * or because the guest failed to guarantee all the appropriate
109  * constraints on all VCPUs (ie buffer can't cross a page boundary).
110  *
111  * Note that any particular CPU may be using a placed vcpu structure,
112  * but we can only optimise if the all are.
113  *
114  * 0: not available, 1: available
115  */
116 static int have_vcpu_info_placement =
117 #ifdef CONFIG_X86_32
118 	1
119 #else
120 	0
121 #endif
122 	;
123 
124 
125 static void xen_vcpu_setup(int cpu)
126 {
127 	struct vcpu_register_vcpu_info info;
128 	int err;
129 	struct vcpu_info *vcpup;
130 
131 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
132 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
133 
134 	if (!have_vcpu_info_placement)
135 		return;		/* already tested, not available */
136 
137 	vcpup = &per_cpu(xen_vcpu_info, cpu);
138 
139 	info.mfn = virt_to_mfn(vcpup);
140 	info.offset = offset_in_page(vcpup);
141 
142 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
143 	       cpu, vcpup, info.mfn, info.offset);
144 
145 	/* Check to see if the hypervisor will put the vcpu_info
146 	   structure where we want it, which allows direct access via
147 	   a percpu-variable. */
148 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
149 
150 	if (err) {
151 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
152 		have_vcpu_info_placement = 0;
153 	} else {
154 		/* This cpu is using the registered vcpu info, even if
155 		   later ones fail to. */
156 		per_cpu(xen_vcpu, cpu) = vcpup;
157 
158 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
159 		       cpu, vcpup);
160 	}
161 }
162 
163 /*
164  * On restore, set the vcpu placement up again.
165  * If it fails, then we're in a bad state, since
166  * we can't back out from using it...
167  */
168 void xen_vcpu_restore(void)
169 {
170 	if (have_vcpu_info_placement) {
171 		int cpu;
172 
173 		for_each_online_cpu(cpu) {
174 			bool other_cpu = (cpu != smp_processor_id());
175 
176 			if (other_cpu &&
177 			    HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
178 				BUG();
179 
180 			xen_vcpu_setup(cpu);
181 
182 			if (other_cpu &&
183 			    HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
184 				BUG();
185 		}
186 
187 		BUG_ON(!have_vcpu_info_placement);
188 	}
189 }
190 
191 static void __init xen_banner(void)
192 {
193 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
194 	struct xen_extraversion extra;
195 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
196 
197 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
198 	       pv_info.name);
199 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
200 	       version >> 16, version & 0xffff, extra.extraversion,
201 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
202 }
203 
204 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
205 		      unsigned int *cx, unsigned int *dx)
206 {
207 	unsigned maskedx = ~0;
208 
209 	/*
210 	 * Mask out inconvenient features, to try and disable as many
211 	 * unsupported kernel subsystems as possible.
212 	 */
213 	if (*ax == 1)
214 		maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
215 			    (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
216 			    (1 << X86_FEATURE_MCE)  |  /* disable MCE */
217 			    (1 << X86_FEATURE_MCA)  |  /* disable MCA */
218 			    (1 << X86_FEATURE_ACC));   /* thermal monitoring */
219 
220 	asm(XEN_EMULATE_PREFIX "cpuid"
221 		: "=a" (*ax),
222 		  "=b" (*bx),
223 		  "=c" (*cx),
224 		  "=d" (*dx)
225 		: "0" (*ax), "2" (*cx));
226 	*dx &= maskedx;
227 }
228 
229 static void xen_set_debugreg(int reg, unsigned long val)
230 {
231 	HYPERVISOR_set_debugreg(reg, val);
232 }
233 
234 static unsigned long xen_get_debugreg(int reg)
235 {
236 	return HYPERVISOR_get_debugreg(reg);
237 }
238 
239 static void xen_leave_lazy(void)
240 {
241 	paravirt_leave_lazy(paravirt_get_lazy_mode());
242 	xen_mc_flush();
243 }
244 
245 static unsigned long xen_store_tr(void)
246 {
247 	return 0;
248 }
249 
250 /*
251  * Set the page permissions for a particular virtual address.  If the
252  * address is a vmalloc mapping (or other non-linear mapping), then
253  * find the linear mapping of the page and also set its protections to
254  * match.
255  */
256 static void set_aliased_prot(void *v, pgprot_t prot)
257 {
258 	int level;
259 	pte_t *ptep;
260 	pte_t pte;
261 	unsigned long pfn;
262 	struct page *page;
263 
264 	ptep = lookup_address((unsigned long)v, &level);
265 	BUG_ON(ptep == NULL);
266 
267 	pfn = pte_pfn(*ptep);
268 	page = pfn_to_page(pfn);
269 
270 	pte = pfn_pte(pfn, prot);
271 
272 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
273 		BUG();
274 
275 	if (!PageHighMem(page)) {
276 		void *av = __va(PFN_PHYS(pfn));
277 
278 		if (av != v)
279 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
280 				BUG();
281 	} else
282 		kmap_flush_unused();
283 }
284 
285 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
286 {
287 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
288 	int i;
289 
290 	for(i = 0; i < entries; i += entries_per_page)
291 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
292 }
293 
294 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
295 {
296 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
297 	int i;
298 
299 	for(i = 0; i < entries; i += entries_per_page)
300 		set_aliased_prot(ldt + i, PAGE_KERNEL);
301 }
302 
303 static void xen_set_ldt(const void *addr, unsigned entries)
304 {
305 	struct mmuext_op *op;
306 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
307 
308 	op = mcs.args;
309 	op->cmd = MMUEXT_SET_LDT;
310 	op->arg1.linear_addr = (unsigned long)addr;
311 	op->arg2.nr_ents = entries;
312 
313 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
314 
315 	xen_mc_issue(PARAVIRT_LAZY_CPU);
316 }
317 
318 static void xen_load_gdt(const struct desc_ptr *dtr)
319 {
320 	unsigned long *frames;
321 	unsigned long va = dtr->address;
322 	unsigned int size = dtr->size + 1;
323 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
324 	int f;
325 	struct multicall_space mcs;
326 
327 	/* A GDT can be up to 64k in size, which corresponds to 8192
328 	   8-byte entries, or 16 4k pages.. */
329 
330 	BUG_ON(size > 65536);
331 	BUG_ON(va & ~PAGE_MASK);
332 
333 	mcs = xen_mc_entry(sizeof(*frames) * pages);
334 	frames = mcs.args;
335 
336 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
337 		frames[f] = virt_to_mfn(va);
338 		make_lowmem_page_readonly((void *)va);
339 	}
340 
341 	MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
342 
343 	xen_mc_issue(PARAVIRT_LAZY_CPU);
344 }
345 
346 static void load_TLS_descriptor(struct thread_struct *t,
347 				unsigned int cpu, unsigned int i)
348 {
349 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
350 	xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
351 	struct multicall_space mc = __xen_mc_entry(0);
352 
353 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
354 }
355 
356 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
357 {
358 	/*
359 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
360 	 * it means we're in a context switch, and %gs has just been
361 	 * saved.  This means we can zero it out to prevent faults on
362 	 * exit from the hypervisor if the next process has no %gs.
363 	 * Either way, it has been saved, and the new value will get
364 	 * loaded properly.  This will go away as soon as Xen has been
365 	 * modified to not save/restore %gs for normal hypercalls.
366 	 *
367 	 * On x86_64, this hack is not used for %gs, because gs points
368 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
369 	 * must not zero %gs on x86_64
370 	 *
371 	 * For x86_64, we need to zero %fs, otherwise we may get an
372 	 * exception between the new %fs descriptor being loaded and
373 	 * %fs being effectively cleared at __switch_to().
374 	 */
375 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
376 #ifdef CONFIG_X86_32
377 		loadsegment(gs, 0);
378 #else
379 		loadsegment(fs, 0);
380 #endif
381 	}
382 
383 	xen_mc_batch();
384 
385 	load_TLS_descriptor(t, cpu, 0);
386 	load_TLS_descriptor(t, cpu, 1);
387 	load_TLS_descriptor(t, cpu, 2);
388 
389 	xen_mc_issue(PARAVIRT_LAZY_CPU);
390 }
391 
392 #ifdef CONFIG_X86_64
393 static void xen_load_gs_index(unsigned int idx)
394 {
395 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
396 		BUG();
397 }
398 #endif
399 
400 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
401 				const void *ptr)
402 {
403 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
404 	u64 entry = *(u64 *)ptr;
405 
406 	preempt_disable();
407 
408 	xen_mc_flush();
409 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
410 		BUG();
411 
412 	preempt_enable();
413 }
414 
415 static int cvt_gate_to_trap(int vector, const gate_desc *val,
416 			    struct trap_info *info)
417 {
418 	if (val->type != 0xf && val->type != 0xe)
419 		return 0;
420 
421 	info->vector = vector;
422 	info->address = gate_offset(*val);
423 	info->cs = gate_segment(*val);
424 	info->flags = val->dpl;
425 	/* interrupt gates clear IF */
426 	if (val->type == 0xe)
427 		info->flags |= 4;
428 
429 	return 1;
430 }
431 
432 /* Locations of each CPU's IDT */
433 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
434 
435 /* Set an IDT entry.  If the entry is part of the current IDT, then
436    also update Xen. */
437 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
438 {
439 	unsigned long p = (unsigned long)&dt[entrynum];
440 	unsigned long start, end;
441 
442 	preempt_disable();
443 
444 	start = __get_cpu_var(idt_desc).address;
445 	end = start + __get_cpu_var(idt_desc).size + 1;
446 
447 	xen_mc_flush();
448 
449 	native_write_idt_entry(dt, entrynum, g);
450 
451 	if (p >= start && (p + 8) <= end) {
452 		struct trap_info info[2];
453 
454 		info[1].address = 0;
455 
456 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
457 			if (HYPERVISOR_set_trap_table(info))
458 				BUG();
459 	}
460 
461 	preempt_enable();
462 }
463 
464 static void xen_convert_trap_info(const struct desc_ptr *desc,
465 				  struct trap_info *traps)
466 {
467 	unsigned in, out, count;
468 
469 	count = (desc->size+1) / sizeof(gate_desc);
470 	BUG_ON(count > 256);
471 
472 	for (in = out = 0; in < count; in++) {
473 		gate_desc *entry = (gate_desc*)(desc->address) + in;
474 
475 		if (cvt_gate_to_trap(in, entry, &traps[out]))
476 			out++;
477 	}
478 	traps[out].address = 0;
479 }
480 
481 void xen_copy_trap_info(struct trap_info *traps)
482 {
483 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
484 
485 	xen_convert_trap_info(desc, traps);
486 }
487 
488 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
489    hold a spinlock to protect the static traps[] array (static because
490    it avoids allocation, and saves stack space). */
491 static void xen_load_idt(const struct desc_ptr *desc)
492 {
493 	static DEFINE_SPINLOCK(lock);
494 	static struct trap_info traps[257];
495 
496 	spin_lock(&lock);
497 
498 	__get_cpu_var(idt_desc) = *desc;
499 
500 	xen_convert_trap_info(desc, traps);
501 
502 	xen_mc_flush();
503 	if (HYPERVISOR_set_trap_table(traps))
504 		BUG();
505 
506 	spin_unlock(&lock);
507 }
508 
509 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
510    they're handled differently. */
511 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
512 				const void *desc, int type)
513 {
514 	preempt_disable();
515 
516 	switch (type) {
517 	case DESC_LDT:
518 	case DESC_TSS:
519 		/* ignore */
520 		break;
521 
522 	default: {
523 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
524 
525 		xen_mc_flush();
526 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
527 			BUG();
528 	}
529 
530 	}
531 
532 	preempt_enable();
533 }
534 
535 static void xen_load_sp0(struct tss_struct *tss,
536 			 struct thread_struct *thread)
537 {
538 	struct multicall_space mcs = xen_mc_entry(0);
539 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
540 	xen_mc_issue(PARAVIRT_LAZY_CPU);
541 }
542 
543 static void xen_set_iopl_mask(unsigned mask)
544 {
545 	struct physdev_set_iopl set_iopl;
546 
547 	/* Force the change at ring 0. */
548 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
549 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
550 }
551 
552 static void xen_io_delay(void)
553 {
554 }
555 
556 #ifdef CONFIG_X86_LOCAL_APIC
557 static u32 xen_apic_read(u32 reg)
558 {
559 	return 0;
560 }
561 
562 static void xen_apic_write(u32 reg, u32 val)
563 {
564 	/* Warn to see if there's any stray references */
565 	WARN_ON(1);
566 }
567 
568 static u64 xen_apic_icr_read(void)
569 {
570 	return 0;
571 }
572 
573 static void xen_apic_icr_write(u32 low, u32 id)
574 {
575 	/* Warn to see if there's any stray references */
576 	WARN_ON(1);
577 }
578 
579 static void xen_apic_wait_icr_idle(void)
580 {
581         return;
582 }
583 
584 static u32 xen_safe_apic_wait_icr_idle(void)
585 {
586         return 0;
587 }
588 
589 static struct apic_ops xen_basic_apic_ops = {
590 	.read = xen_apic_read,
591 	.write = xen_apic_write,
592 	.icr_read = xen_apic_icr_read,
593 	.icr_write = xen_apic_icr_write,
594 	.wait_icr_idle = xen_apic_wait_icr_idle,
595 	.safe_wait_icr_idle = xen_safe_apic_wait_icr_idle,
596 };
597 
598 #endif
599 
600 static void xen_flush_tlb(void)
601 {
602 	struct mmuext_op *op;
603 	struct multicall_space mcs;
604 
605 	preempt_disable();
606 
607 	mcs = xen_mc_entry(sizeof(*op));
608 
609 	op = mcs.args;
610 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
611 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
612 
613 	xen_mc_issue(PARAVIRT_LAZY_MMU);
614 
615 	preempt_enable();
616 }
617 
618 static void xen_flush_tlb_single(unsigned long addr)
619 {
620 	struct mmuext_op *op;
621 	struct multicall_space mcs;
622 
623 	preempt_disable();
624 
625 	mcs = xen_mc_entry(sizeof(*op));
626 	op = mcs.args;
627 	op->cmd = MMUEXT_INVLPG_LOCAL;
628 	op->arg1.linear_addr = addr & PAGE_MASK;
629 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
630 
631 	xen_mc_issue(PARAVIRT_LAZY_MMU);
632 
633 	preempt_enable();
634 }
635 
636 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
637 				 unsigned long va)
638 {
639 	struct {
640 		struct mmuext_op op;
641 		cpumask_t mask;
642 	} *args;
643 	cpumask_t cpumask = *cpus;
644 	struct multicall_space mcs;
645 
646 	/*
647 	 * A couple of (to be removed) sanity checks:
648 	 *
649 	 * - current CPU must not be in mask
650 	 * - mask must exist :)
651 	 */
652 	BUG_ON(cpus_empty(cpumask));
653 	BUG_ON(cpu_isset(smp_processor_id(), cpumask));
654 	BUG_ON(!mm);
655 
656 	/* If a CPU which we ran on has gone down, OK. */
657 	cpus_and(cpumask, cpumask, cpu_online_map);
658 	if (cpus_empty(cpumask))
659 		return;
660 
661 	mcs = xen_mc_entry(sizeof(*args));
662 	args = mcs.args;
663 	args->mask = cpumask;
664 	args->op.arg2.vcpumask = &args->mask;
665 
666 	if (va == TLB_FLUSH_ALL) {
667 		args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
668 	} else {
669 		args->op.cmd = MMUEXT_INVLPG_MULTI;
670 		args->op.arg1.linear_addr = va;
671 	}
672 
673 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
674 
675 	xen_mc_issue(PARAVIRT_LAZY_MMU);
676 }
677 
678 static void xen_clts(void)
679 {
680 	struct multicall_space mcs;
681 
682 	mcs = xen_mc_entry(0);
683 
684 	MULTI_fpu_taskswitch(mcs.mc, 0);
685 
686 	xen_mc_issue(PARAVIRT_LAZY_CPU);
687 }
688 
689 static void xen_write_cr0(unsigned long cr0)
690 {
691 	struct multicall_space mcs;
692 
693 	/* Only pay attention to cr0.TS; everything else is
694 	   ignored. */
695 	mcs = xen_mc_entry(0);
696 
697 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
698 
699 	xen_mc_issue(PARAVIRT_LAZY_CPU);
700 }
701 
702 static void xen_write_cr2(unsigned long cr2)
703 {
704 	x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
705 }
706 
707 static unsigned long xen_read_cr2(void)
708 {
709 	return x86_read_percpu(xen_vcpu)->arch.cr2;
710 }
711 
712 static unsigned long xen_read_cr2_direct(void)
713 {
714 	return x86_read_percpu(xen_vcpu_info.arch.cr2);
715 }
716 
717 static void xen_write_cr4(unsigned long cr4)
718 {
719 	cr4 &= ~X86_CR4_PGE;
720 	cr4 &= ~X86_CR4_PSE;
721 
722 	native_write_cr4(cr4);
723 }
724 
725 static unsigned long xen_read_cr3(void)
726 {
727 	return x86_read_percpu(xen_cr3);
728 }
729 
730 static void set_current_cr3(void *v)
731 {
732 	x86_write_percpu(xen_current_cr3, (unsigned long)v);
733 }
734 
735 static void __xen_write_cr3(bool kernel, unsigned long cr3)
736 {
737 	struct mmuext_op *op;
738 	struct multicall_space mcs;
739 	unsigned long mfn;
740 
741 	if (cr3)
742 		mfn = pfn_to_mfn(PFN_DOWN(cr3));
743 	else
744 		mfn = 0;
745 
746 	WARN_ON(mfn == 0 && kernel);
747 
748 	mcs = __xen_mc_entry(sizeof(*op));
749 
750 	op = mcs.args;
751 	op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
752 	op->arg1.mfn = mfn;
753 
754 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
755 
756 	if (kernel) {
757 		x86_write_percpu(xen_cr3, cr3);
758 
759 		/* Update xen_current_cr3 once the batch has actually
760 		   been submitted. */
761 		xen_mc_callback(set_current_cr3, (void *)cr3);
762 	}
763 }
764 
765 static void xen_write_cr3(unsigned long cr3)
766 {
767 	BUG_ON(preemptible());
768 
769 	xen_mc_batch();  /* disables interrupts */
770 
771 	/* Update while interrupts are disabled, so its atomic with
772 	   respect to ipis */
773 	x86_write_percpu(xen_cr3, cr3);
774 
775 	__xen_write_cr3(true, cr3);
776 
777 #ifdef CONFIG_X86_64
778 	{
779 		pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
780 		if (user_pgd)
781 			__xen_write_cr3(false, __pa(user_pgd));
782 		else
783 			__xen_write_cr3(false, 0);
784 	}
785 #endif
786 
787 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
788 }
789 
790 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
791 {
792 	int ret;
793 
794 	ret = 0;
795 
796 	switch(msr) {
797 #ifdef CONFIG_X86_64
798 		unsigned which;
799 		u64 base;
800 
801 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
802 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
803 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
804 
805 	set:
806 		base = ((u64)high << 32) | low;
807 		if (HYPERVISOR_set_segment_base(which, base) != 0)
808 			ret = -EFAULT;
809 		break;
810 #endif
811 
812 	case MSR_STAR:
813 	case MSR_CSTAR:
814 	case MSR_LSTAR:
815 	case MSR_SYSCALL_MASK:
816 	case MSR_IA32_SYSENTER_CS:
817 	case MSR_IA32_SYSENTER_ESP:
818 	case MSR_IA32_SYSENTER_EIP:
819 		/* Fast syscall setup is all done in hypercalls, so
820 		   these are all ignored.  Stub them out here to stop
821 		   Xen console noise. */
822 		break;
823 
824 	default:
825 		ret = native_write_msr_safe(msr, low, high);
826 	}
827 
828 	return ret;
829 }
830 
831 /* Early in boot, while setting up the initial pagetable, assume
832    everything is pinned. */
833 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
834 {
835 #ifdef CONFIG_FLATMEM
836 	BUG_ON(mem_map);	/* should only be used early */
837 #endif
838 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
839 }
840 
841 /* Early release_pte assumes that all pts are pinned, since there's
842    only init_mm and anything attached to that is pinned. */
843 static void xen_release_pte_init(unsigned long pfn)
844 {
845 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
846 }
847 
848 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
849 {
850 	struct mmuext_op op;
851 	op.cmd = cmd;
852 	op.arg1.mfn = pfn_to_mfn(pfn);
853 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
854 		BUG();
855 }
856 
857 /* This needs to make sure the new pte page is pinned iff its being
858    attached to a pinned pagetable. */
859 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
860 {
861 	struct page *page = pfn_to_page(pfn);
862 
863 	if (PagePinned(virt_to_page(mm->pgd))) {
864 		SetPagePinned(page);
865 
866 		if (!PageHighMem(page)) {
867 			make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
868 			if (level == PT_PTE && USE_SPLIT_PTLOCKS)
869 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
870 		} else
871 			/* make sure there are no stray mappings of
872 			   this page */
873 			kmap_flush_unused();
874 			vm_unmap_aliases();
875 	}
876 }
877 
878 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
879 {
880 	xen_alloc_ptpage(mm, pfn, PT_PTE);
881 }
882 
883 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
884 {
885 	xen_alloc_ptpage(mm, pfn, PT_PMD);
886 }
887 
888 static int xen_pgd_alloc(struct mm_struct *mm)
889 {
890 	pgd_t *pgd = mm->pgd;
891 	int ret = 0;
892 
893 	BUG_ON(PagePinned(virt_to_page(pgd)));
894 
895 #ifdef CONFIG_X86_64
896 	{
897 		struct page *page = virt_to_page(pgd);
898 		pgd_t *user_pgd;
899 
900 		BUG_ON(page->private != 0);
901 
902 		ret = -ENOMEM;
903 
904 		user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
905 		page->private = (unsigned long)user_pgd;
906 
907 		if (user_pgd != NULL) {
908 			user_pgd[pgd_index(VSYSCALL_START)] =
909 				__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
910 			ret = 0;
911 		}
912 
913 		BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
914 	}
915 #endif
916 
917 	return ret;
918 }
919 
920 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
921 {
922 #ifdef CONFIG_X86_64
923 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
924 
925 	if (user_pgd)
926 		free_page((unsigned long)user_pgd);
927 #endif
928 }
929 
930 /* This should never happen until we're OK to use struct page */
931 static void xen_release_ptpage(unsigned long pfn, unsigned level)
932 {
933 	struct page *page = pfn_to_page(pfn);
934 
935 	if (PagePinned(page)) {
936 		if (!PageHighMem(page)) {
937 			if (level == PT_PTE && USE_SPLIT_PTLOCKS)
938 				pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
939 			make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
940 		}
941 		ClearPagePinned(page);
942 	}
943 }
944 
945 static void xen_release_pte(unsigned long pfn)
946 {
947 	xen_release_ptpage(pfn, PT_PTE);
948 }
949 
950 static void xen_release_pmd(unsigned long pfn)
951 {
952 	xen_release_ptpage(pfn, PT_PMD);
953 }
954 
955 #if PAGETABLE_LEVELS == 4
956 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
957 {
958 	xen_alloc_ptpage(mm, pfn, PT_PUD);
959 }
960 
961 static void xen_release_pud(unsigned long pfn)
962 {
963 	xen_release_ptpage(pfn, PT_PUD);
964 }
965 #endif
966 
967 #ifdef CONFIG_HIGHPTE
968 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
969 {
970 	pgprot_t prot = PAGE_KERNEL;
971 
972 	if (PagePinned(page))
973 		prot = PAGE_KERNEL_RO;
974 
975 	if (0 && PageHighMem(page))
976 		printk("mapping highpte %lx type %d prot %s\n",
977 		       page_to_pfn(page), type,
978 		       (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
979 
980 	return kmap_atomic_prot(page, type, prot);
981 }
982 #endif
983 
984 #ifdef CONFIG_X86_32
985 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
986 {
987 	/* If there's an existing pte, then don't allow _PAGE_RW to be set */
988 	if (pte_val_ma(*ptep) & _PAGE_PRESENT)
989 		pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
990 			       pte_val_ma(pte));
991 
992 	return pte;
993 }
994 
995 /* Init-time set_pte while constructing initial pagetables, which
996    doesn't allow RO pagetable pages to be remapped RW */
997 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
998 {
999 	pte = mask_rw_pte(ptep, pte);
1000 
1001 	xen_set_pte(ptep, pte);
1002 }
1003 #endif
1004 
1005 static __init void xen_pagetable_setup_start(pgd_t *base)
1006 {
1007 }
1008 
1009 void xen_setup_shared_info(void)
1010 {
1011 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1012 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
1013 			   xen_start_info->shared_info);
1014 
1015 		HYPERVISOR_shared_info =
1016 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1017 	} else
1018 		HYPERVISOR_shared_info =
1019 			(struct shared_info *)__va(xen_start_info->shared_info);
1020 
1021 #ifndef CONFIG_SMP
1022 	/* In UP this is as good a place as any to set up shared info */
1023 	xen_setup_vcpu_info_placement();
1024 #endif
1025 
1026 	xen_setup_mfn_list_list();
1027 }
1028 
1029 static __init void xen_pagetable_setup_done(pgd_t *base)
1030 {
1031 	xen_setup_shared_info();
1032 }
1033 
1034 static __init void xen_post_allocator_init(void)
1035 {
1036 	pv_mmu_ops.set_pte = xen_set_pte;
1037 	pv_mmu_ops.set_pmd = xen_set_pmd;
1038 	pv_mmu_ops.set_pud = xen_set_pud;
1039 #if PAGETABLE_LEVELS == 4
1040 	pv_mmu_ops.set_pgd = xen_set_pgd;
1041 #endif
1042 
1043 	/* This will work as long as patching hasn't happened yet
1044 	   (which it hasn't) */
1045 	pv_mmu_ops.alloc_pte = xen_alloc_pte;
1046 	pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1047 	pv_mmu_ops.release_pte = xen_release_pte;
1048 	pv_mmu_ops.release_pmd = xen_release_pmd;
1049 #if PAGETABLE_LEVELS == 4
1050 	pv_mmu_ops.alloc_pud = xen_alloc_pud;
1051 	pv_mmu_ops.release_pud = xen_release_pud;
1052 #endif
1053 
1054 #ifdef CONFIG_X86_64
1055 	SetPagePinned(virt_to_page(level3_user_vsyscall));
1056 #endif
1057 	xen_mark_init_mm_pinned();
1058 }
1059 
1060 /* This is called once we have the cpu_possible_map */
1061 void xen_setup_vcpu_info_placement(void)
1062 {
1063 	int cpu;
1064 
1065 	for_each_possible_cpu(cpu)
1066 		xen_vcpu_setup(cpu);
1067 
1068 	/* xen_vcpu_setup managed to place the vcpu_info within the
1069 	   percpu area for all cpus, so make use of it */
1070 	if (have_vcpu_info_placement) {
1071 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
1072 
1073 		pv_irq_ops.save_fl = xen_save_fl_direct;
1074 		pv_irq_ops.restore_fl = xen_restore_fl_direct;
1075 		pv_irq_ops.irq_disable = xen_irq_disable_direct;
1076 		pv_irq_ops.irq_enable = xen_irq_enable_direct;
1077 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1078 	}
1079 }
1080 
1081 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1082 			  unsigned long addr, unsigned len)
1083 {
1084 	char *start, *end, *reloc;
1085 	unsigned ret;
1086 
1087 	start = end = reloc = NULL;
1088 
1089 #define SITE(op, x)							\
1090 	case PARAVIRT_PATCH(op.x):					\
1091 	if (have_vcpu_info_placement) {					\
1092 		start = (char *)xen_##x##_direct;			\
1093 		end = xen_##x##_direct_end;				\
1094 		reloc = xen_##x##_direct_reloc;				\
1095 	}								\
1096 	goto patch_site
1097 
1098 	switch (type) {
1099 		SITE(pv_irq_ops, irq_enable);
1100 		SITE(pv_irq_ops, irq_disable);
1101 		SITE(pv_irq_ops, save_fl);
1102 		SITE(pv_irq_ops, restore_fl);
1103 #undef SITE
1104 
1105 	patch_site:
1106 		if (start == NULL || (end-start) > len)
1107 			goto default_patch;
1108 
1109 		ret = paravirt_patch_insns(insnbuf, len, start, end);
1110 
1111 		/* Note: because reloc is assigned from something that
1112 		   appears to be an array, gcc assumes it's non-null,
1113 		   but doesn't know its relationship with start and
1114 		   end. */
1115 		if (reloc > start && reloc < end) {
1116 			int reloc_off = reloc - start;
1117 			long *relocp = (long *)(insnbuf + reloc_off);
1118 			long delta = start - (char *)addr;
1119 
1120 			*relocp += delta;
1121 		}
1122 		break;
1123 
1124 	default_patch:
1125 	default:
1126 		ret = paravirt_patch_default(type, clobbers, insnbuf,
1127 					     addr, len);
1128 		break;
1129 	}
1130 
1131 	return ret;
1132 }
1133 
1134 static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
1135 {
1136 	pte_t pte;
1137 
1138 	phys >>= PAGE_SHIFT;
1139 
1140 	switch (idx) {
1141 	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1142 #ifdef CONFIG_X86_F00F_BUG
1143 	case FIX_F00F_IDT:
1144 #endif
1145 #ifdef CONFIG_X86_32
1146 	case FIX_WP_TEST:
1147 	case FIX_VDSO:
1148 # ifdef CONFIG_HIGHMEM
1149 	case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1150 # endif
1151 #else
1152 	case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1153 #endif
1154 #ifdef CONFIG_X86_LOCAL_APIC
1155 	case FIX_APIC_BASE:	/* maps dummy local APIC */
1156 #endif
1157 		pte = pfn_pte(phys, prot);
1158 		break;
1159 
1160 	default:
1161 		pte = mfn_pte(phys, prot);
1162 		break;
1163 	}
1164 
1165 	__native_set_fixmap(idx, pte);
1166 
1167 #ifdef CONFIG_X86_64
1168 	/* Replicate changes to map the vsyscall page into the user
1169 	   pagetable vsyscall mapping. */
1170 	if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1171 		unsigned long vaddr = __fix_to_virt(idx);
1172 		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1173 	}
1174 #endif
1175 }
1176 
1177 static const struct pv_info xen_info __initdata = {
1178 	.paravirt_enabled = 1,
1179 	.shared_kernel_pmd = 0,
1180 
1181 	.name = "Xen",
1182 };
1183 
1184 static const struct pv_init_ops xen_init_ops __initdata = {
1185 	.patch = xen_patch,
1186 
1187 	.banner = xen_banner,
1188 	.memory_setup = xen_memory_setup,
1189 	.arch_setup = xen_arch_setup,
1190 	.post_allocator_init = xen_post_allocator_init,
1191 };
1192 
1193 static const struct pv_time_ops xen_time_ops __initdata = {
1194 	.time_init = xen_time_init,
1195 
1196 	.set_wallclock = xen_set_wallclock,
1197 	.get_wallclock = xen_get_wallclock,
1198 	.get_tsc_khz = xen_tsc_khz,
1199 	.sched_clock = xen_sched_clock,
1200 };
1201 
1202 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1203 	.cpuid = xen_cpuid,
1204 
1205 	.set_debugreg = xen_set_debugreg,
1206 	.get_debugreg = xen_get_debugreg,
1207 
1208 	.clts = xen_clts,
1209 
1210 	.read_cr0 = native_read_cr0,
1211 	.write_cr0 = xen_write_cr0,
1212 
1213 	.read_cr4 = native_read_cr4,
1214 	.read_cr4_safe = native_read_cr4_safe,
1215 	.write_cr4 = xen_write_cr4,
1216 
1217 	.wbinvd = native_wbinvd,
1218 
1219 	.read_msr = native_read_msr_safe,
1220 	.write_msr = xen_write_msr_safe,
1221 	.read_tsc = native_read_tsc,
1222 	.read_pmc = native_read_pmc,
1223 
1224 	.iret = xen_iret,
1225 	.irq_enable_sysexit = xen_sysexit,
1226 #ifdef CONFIG_X86_64
1227 	.usergs_sysret32 = xen_sysret32,
1228 	.usergs_sysret64 = xen_sysret64,
1229 #endif
1230 
1231 	.load_tr_desc = paravirt_nop,
1232 	.set_ldt = xen_set_ldt,
1233 	.load_gdt = xen_load_gdt,
1234 	.load_idt = xen_load_idt,
1235 	.load_tls = xen_load_tls,
1236 #ifdef CONFIG_X86_64
1237 	.load_gs_index = xen_load_gs_index,
1238 #endif
1239 
1240 	.alloc_ldt = xen_alloc_ldt,
1241 	.free_ldt = xen_free_ldt,
1242 
1243 	.store_gdt = native_store_gdt,
1244 	.store_idt = native_store_idt,
1245 	.store_tr = xen_store_tr,
1246 
1247 	.write_ldt_entry = xen_write_ldt_entry,
1248 	.write_gdt_entry = xen_write_gdt_entry,
1249 	.write_idt_entry = xen_write_idt_entry,
1250 	.load_sp0 = xen_load_sp0,
1251 
1252 	.set_iopl_mask = xen_set_iopl_mask,
1253 	.io_delay = xen_io_delay,
1254 
1255 	/* Xen takes care of %gs when switching to usermode for us */
1256 	.swapgs = paravirt_nop,
1257 
1258 	.lazy_mode = {
1259 		.enter = paravirt_enter_lazy_cpu,
1260 		.leave = xen_leave_lazy,
1261 	},
1262 };
1263 
1264 static const struct pv_apic_ops xen_apic_ops __initdata = {
1265 #ifdef CONFIG_X86_LOCAL_APIC
1266 	.setup_boot_clock = paravirt_nop,
1267 	.setup_secondary_clock = paravirt_nop,
1268 	.startup_ipi_hook = paravirt_nop,
1269 #endif
1270 };
1271 
1272 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1273 	.pagetable_setup_start = xen_pagetable_setup_start,
1274 	.pagetable_setup_done = xen_pagetable_setup_done,
1275 
1276 	.read_cr2 = xen_read_cr2,
1277 	.write_cr2 = xen_write_cr2,
1278 
1279 	.read_cr3 = xen_read_cr3,
1280 	.write_cr3 = xen_write_cr3,
1281 
1282 	.flush_tlb_user = xen_flush_tlb,
1283 	.flush_tlb_kernel = xen_flush_tlb,
1284 	.flush_tlb_single = xen_flush_tlb_single,
1285 	.flush_tlb_others = xen_flush_tlb_others,
1286 
1287 	.pte_update = paravirt_nop,
1288 	.pte_update_defer = paravirt_nop,
1289 
1290 	.pgd_alloc = xen_pgd_alloc,
1291 	.pgd_free = xen_pgd_free,
1292 
1293 	.alloc_pte = xen_alloc_pte_init,
1294 	.release_pte = xen_release_pte_init,
1295 	.alloc_pmd = xen_alloc_pte_init,
1296 	.alloc_pmd_clone = paravirt_nop,
1297 	.release_pmd = xen_release_pte_init,
1298 
1299 #ifdef CONFIG_HIGHPTE
1300 	.kmap_atomic_pte = xen_kmap_atomic_pte,
1301 #endif
1302 
1303 #ifdef CONFIG_X86_64
1304 	.set_pte = xen_set_pte,
1305 #else
1306 	.set_pte = xen_set_pte_init,
1307 #endif
1308 	.set_pte_at = xen_set_pte_at,
1309 	.set_pmd = xen_set_pmd_hyper,
1310 
1311 	.ptep_modify_prot_start = __ptep_modify_prot_start,
1312 	.ptep_modify_prot_commit = __ptep_modify_prot_commit,
1313 
1314 	.pte_val = xen_pte_val,
1315 	.pte_flags = native_pte_flags,
1316 	.pgd_val = xen_pgd_val,
1317 
1318 	.make_pte = xen_make_pte,
1319 	.make_pgd = xen_make_pgd,
1320 
1321 #ifdef CONFIG_X86_PAE
1322 	.set_pte_atomic = xen_set_pte_atomic,
1323 	.set_pte_present = xen_set_pte_at,
1324 	.pte_clear = xen_pte_clear,
1325 	.pmd_clear = xen_pmd_clear,
1326 #endif	/* CONFIG_X86_PAE */
1327 	.set_pud = xen_set_pud_hyper,
1328 
1329 	.make_pmd = xen_make_pmd,
1330 	.pmd_val = xen_pmd_val,
1331 
1332 #if PAGETABLE_LEVELS == 4
1333 	.pud_val = xen_pud_val,
1334 	.make_pud = xen_make_pud,
1335 	.set_pgd = xen_set_pgd_hyper,
1336 
1337 	.alloc_pud = xen_alloc_pte_init,
1338 	.release_pud = xen_release_pte_init,
1339 #endif	/* PAGETABLE_LEVELS == 4 */
1340 
1341 	.activate_mm = xen_activate_mm,
1342 	.dup_mmap = xen_dup_mmap,
1343 	.exit_mmap = xen_exit_mmap,
1344 
1345 	.lazy_mode = {
1346 		.enter = paravirt_enter_lazy_mmu,
1347 		.leave = xen_leave_lazy,
1348 	},
1349 
1350 	.set_fixmap = xen_set_fixmap,
1351 };
1352 
1353 static void xen_reboot(int reason)
1354 {
1355 	struct sched_shutdown r = { .reason = reason };
1356 
1357 #ifdef CONFIG_SMP
1358 	smp_send_stop();
1359 #endif
1360 
1361 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1362 		BUG();
1363 }
1364 
1365 static void xen_restart(char *msg)
1366 {
1367 	xen_reboot(SHUTDOWN_reboot);
1368 }
1369 
1370 static void xen_emergency_restart(void)
1371 {
1372 	xen_reboot(SHUTDOWN_reboot);
1373 }
1374 
1375 static void xen_machine_halt(void)
1376 {
1377 	xen_reboot(SHUTDOWN_poweroff);
1378 }
1379 
1380 static void xen_crash_shutdown(struct pt_regs *regs)
1381 {
1382 	xen_reboot(SHUTDOWN_crash);
1383 }
1384 
1385 static const struct machine_ops __initdata xen_machine_ops = {
1386 	.restart = xen_restart,
1387 	.halt = xen_machine_halt,
1388 	.power_off = xen_machine_halt,
1389 	.shutdown = xen_machine_halt,
1390 	.crash_shutdown = xen_crash_shutdown,
1391 	.emergency_restart = xen_emergency_restart,
1392 };
1393 
1394 
1395 static void __init xen_reserve_top(void)
1396 {
1397 #ifdef CONFIG_X86_32
1398 	unsigned long top = HYPERVISOR_VIRT_START;
1399 	struct xen_platform_parameters pp;
1400 
1401 	if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1402 		top = pp.virt_start;
1403 
1404 	reserve_top_address(-top);
1405 #endif	/* CONFIG_X86_32 */
1406 }
1407 
1408 /*
1409  * Like __va(), but returns address in the kernel mapping (which is
1410  * all we have until the physical memory mapping has been set up.
1411  */
1412 static void *__ka(phys_addr_t paddr)
1413 {
1414 #ifdef CONFIG_X86_64
1415 	return (void *)(paddr + __START_KERNEL_map);
1416 #else
1417 	return __va(paddr);
1418 #endif
1419 }
1420 
1421 /* Convert a machine address to physical address */
1422 static unsigned long m2p(phys_addr_t maddr)
1423 {
1424 	phys_addr_t paddr;
1425 
1426 	maddr &= PTE_PFN_MASK;
1427 	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1428 
1429 	return paddr;
1430 }
1431 
1432 /* Convert a machine address to kernel virtual */
1433 static void *m2v(phys_addr_t maddr)
1434 {
1435 	return __ka(m2p(maddr));
1436 }
1437 
1438 static void set_page_prot(void *addr, pgprot_t prot)
1439 {
1440 	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1441 	pte_t pte = pfn_pte(pfn, prot);
1442 
1443 	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1444 		BUG();
1445 }
1446 
1447 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1448 {
1449 	unsigned pmdidx, pteidx;
1450 	unsigned ident_pte;
1451 	unsigned long pfn;
1452 
1453 	ident_pte = 0;
1454 	pfn = 0;
1455 	for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1456 		pte_t *pte_page;
1457 
1458 		/* Reuse or allocate a page of ptes */
1459 		if (pmd_present(pmd[pmdidx]))
1460 			pte_page = m2v(pmd[pmdidx].pmd);
1461 		else {
1462 			/* Check for free pte pages */
1463 			if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1464 				break;
1465 
1466 			pte_page = &level1_ident_pgt[ident_pte];
1467 			ident_pte += PTRS_PER_PTE;
1468 
1469 			pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1470 		}
1471 
1472 		/* Install mappings */
1473 		for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1474 			pte_t pte;
1475 
1476 			if (pfn > max_pfn_mapped)
1477 				max_pfn_mapped = pfn;
1478 
1479 			if (!pte_none(pte_page[pteidx]))
1480 				continue;
1481 
1482 			pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1483 			pte_page[pteidx] = pte;
1484 		}
1485 	}
1486 
1487 	for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1488 		set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1489 
1490 	set_page_prot(pmd, PAGE_KERNEL_RO);
1491 }
1492 
1493 #ifdef CONFIG_X86_64
1494 static void convert_pfn_mfn(void *v)
1495 {
1496 	pte_t *pte = v;
1497 	int i;
1498 
1499 	/* All levels are converted the same way, so just treat them
1500 	   as ptes. */
1501 	for(i = 0; i < PTRS_PER_PTE; i++)
1502 		pte[i] = xen_make_pte(pte[i].pte);
1503 }
1504 
1505 /*
1506  * Set up the inital kernel pagetable.
1507  *
1508  * We can construct this by grafting the Xen provided pagetable into
1509  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1510  * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt.  This
1511  * means that only the kernel has a physical mapping to start with -
1512  * but that's enough to get __va working.  We need to fill in the rest
1513  * of the physical mapping once some sort of allocator has been set
1514  * up.
1515  */
1516 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1517 {
1518 	pud_t *l3;
1519 	pmd_t *l2;
1520 
1521 	/* Zap identity mapping */
1522 	init_level4_pgt[0] = __pgd(0);
1523 
1524 	/* Pre-constructed entries are in pfn, so convert to mfn */
1525 	convert_pfn_mfn(init_level4_pgt);
1526 	convert_pfn_mfn(level3_ident_pgt);
1527 	convert_pfn_mfn(level3_kernel_pgt);
1528 
1529 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1530 	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1531 
1532 	memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1533 	memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1534 
1535 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1536 	l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1537 	memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1538 
1539 	/* Set up identity map */
1540 	xen_map_identity_early(level2_ident_pgt, max_pfn);
1541 
1542 	/* Make pagetable pieces RO */
1543 	set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1544 	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1545 	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1546 	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1547 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1548 	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1549 
1550 	/* Pin down new L4 */
1551 	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1552 			  PFN_DOWN(__pa_symbol(init_level4_pgt)));
1553 
1554 	/* Unpin Xen-provided one */
1555 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1556 
1557 	/* Switch over */
1558 	pgd = init_level4_pgt;
1559 
1560 	/*
1561 	 * At this stage there can be no user pgd, and no page
1562 	 * structure to attach it to, so make sure we just set kernel
1563 	 * pgd.
1564 	 */
1565 	xen_mc_batch();
1566 	__xen_write_cr3(true, __pa(pgd));
1567 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1568 
1569 	reserve_early(__pa(xen_start_info->pt_base),
1570 		      __pa(xen_start_info->pt_base +
1571 			   xen_start_info->nr_pt_frames * PAGE_SIZE),
1572 		      "XEN PAGETABLES");
1573 
1574 	return pgd;
1575 }
1576 #else	/* !CONFIG_X86_64 */
1577 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1578 
1579 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1580 {
1581 	pmd_t *kernel_pmd;
1582 
1583 	init_pg_tables_start = __pa(pgd);
1584 	init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1585 	max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1586 
1587 	kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1588 	memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1589 
1590 	xen_map_identity_early(level2_kernel_pgt, max_pfn);
1591 
1592 	memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1593 	set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1594 			__pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1595 
1596 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1597 	set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1598 	set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1599 
1600 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1601 
1602 	xen_write_cr3(__pa(swapper_pg_dir));
1603 
1604 	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1605 
1606 	return swapper_pg_dir;
1607 }
1608 #endif	/* CONFIG_X86_64 */
1609 
1610 /* First C function to be called on Xen boot */
1611 asmlinkage void __init xen_start_kernel(void)
1612 {
1613 	pgd_t *pgd;
1614 
1615 	if (!xen_start_info)
1616 		return;
1617 
1618 	xen_domain_type = XEN_PV_DOMAIN;
1619 
1620 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1621 
1622 	xen_setup_features();
1623 
1624 	/* Install Xen paravirt ops */
1625 	pv_info = xen_info;
1626 	pv_init_ops = xen_init_ops;
1627 	pv_time_ops = xen_time_ops;
1628 	pv_cpu_ops = xen_cpu_ops;
1629 	pv_apic_ops = xen_apic_ops;
1630 	pv_mmu_ops = xen_mmu_ops;
1631 
1632 	xen_init_irq_ops();
1633 
1634 #ifdef CONFIG_X86_LOCAL_APIC
1635 	/*
1636 	 * set up the basic apic ops.
1637 	 */
1638 	apic_ops = &xen_basic_apic_ops;
1639 #endif
1640 
1641 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1642 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1643 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1644 	}
1645 
1646 	machine_ops = xen_machine_ops;
1647 
1648 #ifdef CONFIG_X86_64
1649 	/* Disable until direct per-cpu data access. */
1650 	have_vcpu_info_placement = 0;
1651 	x86_64_init_pda();
1652 #endif
1653 
1654 	xen_smp_init();
1655 
1656 	/* Get mfn list */
1657 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1658 		xen_build_dynamic_phys_to_machine();
1659 
1660 	pgd = (pgd_t *)xen_start_info->pt_base;
1661 
1662 	/* Prevent unwanted bits from being set in PTEs. */
1663 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1664 	if (!xen_initial_domain())
1665 		__supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1666 
1667 	/* Don't do the full vcpu_info placement stuff until we have a
1668 	   possible map and a non-dummy shared_info. */
1669 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1670 
1671 	xen_raw_console_write("mapping kernel into physical memory\n");
1672 	pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1673 
1674 	init_mm.pgd = pgd;
1675 
1676 	/* keep using Xen gdt for now; no urgent need to change it */
1677 
1678 	pv_info.kernel_rpl = 1;
1679 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1680 		pv_info.kernel_rpl = 0;
1681 
1682 	/* set the limit of our address space */
1683 	xen_reserve_top();
1684 
1685 #ifdef CONFIG_X86_32
1686 	/* set up basic CPUID stuff */
1687 	cpu_detect(&new_cpu_data);
1688 	new_cpu_data.hard_math = 1;
1689 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1690 #endif
1691 
1692 	/* Poke various useful things into boot_params */
1693 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1694 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1695 		? __pa(xen_start_info->mod_start) : 0;
1696 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1697 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1698 
1699 	if (!xen_initial_domain()) {
1700 		add_preferred_console("xenboot", 0, NULL);
1701 		add_preferred_console("tty", 0, NULL);
1702 		add_preferred_console("hvc", 0, NULL);
1703 	}
1704 
1705 	xen_raw_console_write("about to get started...\n");
1706 
1707 	/* Start the world */
1708 #ifdef CONFIG_X86_32
1709 	i386_start_kernel();
1710 #else
1711 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1712 #endif
1713 }
1714