xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 545e4006)
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/interface/sched.h>
34 #include <xen/features.h>
35 #include <xen/page.h>
36 #include <xen/hvc-console.h>
37 
38 #include <asm/paravirt.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 /*
61  * Identity map, in addition to plain kernel map.  This needs to be
62  * large enough to allocate page table pages to allocate the rest.
63  * Each page can map 2MB.
64  */
65 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
66 
67 #ifdef CONFIG_X86_64
68 /* l3 pud for userspace vsyscall mapping */
69 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
70 #endif /* CONFIG_X86_64 */
71 
72 /*
73  * Note about cr3 (pagetable base) values:
74  *
75  * xen_cr3 contains the current logical cr3 value; it contains the
76  * last set cr3.  This may not be the current effective cr3, because
77  * its update may be being lazily deferred.  However, a vcpu looking
78  * at its own cr3 can use this value knowing that it everything will
79  * be self-consistent.
80  *
81  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
82  * hypercall to set the vcpu cr3 is complete (so it may be a little
83  * out of date, but it will never be set early).  If one vcpu is
84  * looking at another vcpu's cr3 value, it should use this variable.
85  */
86 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
87 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
88 
89 struct start_info *xen_start_info;
90 EXPORT_SYMBOL_GPL(xen_start_info);
91 
92 struct shared_info xen_dummy_shared_info;
93 
94 /*
95  * Point at some empty memory to start with. We map the real shared_info
96  * page as soon as fixmap is up and running.
97  */
98 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
99 
100 /*
101  * Flag to determine whether vcpu info placement is available on all
102  * VCPUs.  We assume it is to start with, and then set it to zero on
103  * the first failure.  This is because it can succeed on some VCPUs
104  * and not others, since it can involve hypervisor memory allocation,
105  * or because the guest failed to guarantee all the appropriate
106  * constraints on all VCPUs (ie buffer can't cross a page boundary).
107  *
108  * Note that any particular CPU may be using a placed vcpu structure,
109  * but we can only optimise if the all are.
110  *
111  * 0: not available, 1: available
112  */
113 static int have_vcpu_info_placement = 1;
114 
115 static void xen_vcpu_setup(int cpu)
116 {
117 	struct vcpu_register_vcpu_info info;
118 	int err;
119 	struct vcpu_info *vcpup;
120 
121 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
122 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
123 
124 	if (!have_vcpu_info_placement)
125 		return;		/* already tested, not available */
126 
127 	vcpup = &per_cpu(xen_vcpu_info, cpu);
128 
129 	info.mfn = virt_to_mfn(vcpup);
130 	info.offset = offset_in_page(vcpup);
131 
132 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
133 	       cpu, vcpup, info.mfn, info.offset);
134 
135 	/* Check to see if the hypervisor will put the vcpu_info
136 	   structure where we want it, which allows direct access via
137 	   a percpu-variable. */
138 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
139 
140 	if (err) {
141 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
142 		have_vcpu_info_placement = 0;
143 	} else {
144 		/* This cpu is using the registered vcpu info, even if
145 		   later ones fail to. */
146 		per_cpu(xen_vcpu, cpu) = vcpup;
147 
148 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
149 		       cpu, vcpup);
150 	}
151 }
152 
153 /*
154  * On restore, set the vcpu placement up again.
155  * If it fails, then we're in a bad state, since
156  * we can't back out from using it...
157  */
158 void xen_vcpu_restore(void)
159 {
160 	if (have_vcpu_info_placement) {
161 		int cpu;
162 
163 		for_each_online_cpu(cpu) {
164 			bool other_cpu = (cpu != smp_processor_id());
165 
166 			if (other_cpu &&
167 			    HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
168 				BUG();
169 
170 			xen_vcpu_setup(cpu);
171 
172 			if (other_cpu &&
173 			    HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
174 				BUG();
175 		}
176 
177 		BUG_ON(!have_vcpu_info_placement);
178 	}
179 }
180 
181 static void __init xen_banner(void)
182 {
183 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
184 	struct xen_extraversion extra;
185 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
186 
187 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
188 	       pv_info.name);
189 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
190 	       version >> 16, version & 0xffff, extra.extraversion,
191 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
192 }
193 
194 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
195 		      unsigned int *cx, unsigned int *dx)
196 {
197 	unsigned maskedx = ~0;
198 
199 	/*
200 	 * Mask out inconvenient features, to try and disable as many
201 	 * unsupported kernel subsystems as possible.
202 	 */
203 	if (*ax == 1)
204 		maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
205 			    (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
206 			    (1 << X86_FEATURE_MCE)  |  /* disable MCE */
207 			    (1 << X86_FEATURE_MCA)  |  /* disable MCA */
208 			    (1 << X86_FEATURE_ACC));   /* thermal monitoring */
209 
210 	asm(XEN_EMULATE_PREFIX "cpuid"
211 		: "=a" (*ax),
212 		  "=b" (*bx),
213 		  "=c" (*cx),
214 		  "=d" (*dx)
215 		: "0" (*ax), "2" (*cx));
216 	*dx &= maskedx;
217 }
218 
219 static void xen_set_debugreg(int reg, unsigned long val)
220 {
221 	HYPERVISOR_set_debugreg(reg, val);
222 }
223 
224 static unsigned long xen_get_debugreg(int reg)
225 {
226 	return HYPERVISOR_get_debugreg(reg);
227 }
228 
229 static unsigned long xen_save_fl(void)
230 {
231 	struct vcpu_info *vcpu;
232 	unsigned long flags;
233 
234 	vcpu = x86_read_percpu(xen_vcpu);
235 
236 	/* flag has opposite sense of mask */
237 	flags = !vcpu->evtchn_upcall_mask;
238 
239 	/* convert to IF type flag
240 	   -0 -> 0x00000000
241 	   -1 -> 0xffffffff
242 	*/
243 	return (-flags) & X86_EFLAGS_IF;
244 }
245 
246 static void xen_restore_fl(unsigned long flags)
247 {
248 	struct vcpu_info *vcpu;
249 
250 	/* convert from IF type flag */
251 	flags = !(flags & X86_EFLAGS_IF);
252 
253 	/* There's a one instruction preempt window here.  We need to
254 	   make sure we're don't switch CPUs between getting the vcpu
255 	   pointer and updating the mask. */
256 	preempt_disable();
257 	vcpu = x86_read_percpu(xen_vcpu);
258 	vcpu->evtchn_upcall_mask = flags;
259 	preempt_enable_no_resched();
260 
261 	/* Doesn't matter if we get preempted here, because any
262 	   pending event will get dealt with anyway. */
263 
264 	if (flags == 0) {
265 		preempt_check_resched();
266 		barrier(); /* unmask then check (avoid races) */
267 		if (unlikely(vcpu->evtchn_upcall_pending))
268 			force_evtchn_callback();
269 	}
270 }
271 
272 static void xen_irq_disable(void)
273 {
274 	/* There's a one instruction preempt window here.  We need to
275 	   make sure we're don't switch CPUs between getting the vcpu
276 	   pointer and updating the mask. */
277 	preempt_disable();
278 	x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
279 	preempt_enable_no_resched();
280 }
281 
282 static void xen_irq_enable(void)
283 {
284 	struct vcpu_info *vcpu;
285 
286 	/* We don't need to worry about being preempted here, since
287 	   either a) interrupts are disabled, so no preemption, or b)
288 	   the caller is confused and is trying to re-enable interrupts
289 	   on an indeterminate processor. */
290 
291 	vcpu = x86_read_percpu(xen_vcpu);
292 	vcpu->evtchn_upcall_mask = 0;
293 
294 	/* Doesn't matter if we get preempted here, because any
295 	   pending event will get dealt with anyway. */
296 
297 	barrier(); /* unmask then check (avoid races) */
298 	if (unlikely(vcpu->evtchn_upcall_pending))
299 		force_evtchn_callback();
300 }
301 
302 static void xen_safe_halt(void)
303 {
304 	/* Blocking includes an implicit local_irq_enable(). */
305 	if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0)
306 		BUG();
307 }
308 
309 static void xen_halt(void)
310 {
311 	if (irqs_disabled())
312 		HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
313 	else
314 		xen_safe_halt();
315 }
316 
317 static void xen_leave_lazy(void)
318 {
319 	paravirt_leave_lazy(paravirt_get_lazy_mode());
320 	xen_mc_flush();
321 }
322 
323 static unsigned long xen_store_tr(void)
324 {
325 	return 0;
326 }
327 
328 static void xen_set_ldt(const void *addr, unsigned entries)
329 {
330 	struct mmuext_op *op;
331 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
332 
333 	op = mcs.args;
334 	op->cmd = MMUEXT_SET_LDT;
335 	op->arg1.linear_addr = (unsigned long)addr;
336 	op->arg2.nr_ents = entries;
337 
338 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
339 
340 	xen_mc_issue(PARAVIRT_LAZY_CPU);
341 }
342 
343 static void xen_load_gdt(const struct desc_ptr *dtr)
344 {
345 	unsigned long *frames;
346 	unsigned long va = dtr->address;
347 	unsigned int size = dtr->size + 1;
348 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
349 	int f;
350 	struct multicall_space mcs;
351 
352 	/* A GDT can be up to 64k in size, which corresponds to 8192
353 	   8-byte entries, or 16 4k pages.. */
354 
355 	BUG_ON(size > 65536);
356 	BUG_ON(va & ~PAGE_MASK);
357 
358 	mcs = xen_mc_entry(sizeof(*frames) * pages);
359 	frames = mcs.args;
360 
361 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
362 		frames[f] = virt_to_mfn(va);
363 		make_lowmem_page_readonly((void *)va);
364 	}
365 
366 	MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
367 
368 	xen_mc_issue(PARAVIRT_LAZY_CPU);
369 }
370 
371 static void load_TLS_descriptor(struct thread_struct *t,
372 				unsigned int cpu, unsigned int i)
373 {
374 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
375 	xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
376 	struct multicall_space mc = __xen_mc_entry(0);
377 
378 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
379 }
380 
381 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
382 {
383 	/*
384 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
385 	 * it means we're in a context switch, and %gs has just been
386 	 * saved.  This means we can zero it out to prevent faults on
387 	 * exit from the hypervisor if the next process has no %gs.
388 	 * Either way, it has been saved, and the new value will get
389 	 * loaded properly.  This will go away as soon as Xen has been
390 	 * modified to not save/restore %gs for normal hypercalls.
391 	 *
392 	 * On x86_64, this hack is not used for %gs, because gs points
393 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
394 	 * must not zero %gs on x86_64
395 	 *
396 	 * For x86_64, we need to zero %fs, otherwise we may get an
397 	 * exception between the new %fs descriptor being loaded and
398 	 * %fs being effectively cleared at __switch_to().
399 	 */
400 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
401 #ifdef CONFIG_X86_32
402 		loadsegment(gs, 0);
403 #else
404 		loadsegment(fs, 0);
405 #endif
406 	}
407 
408 	xen_mc_batch();
409 
410 	load_TLS_descriptor(t, cpu, 0);
411 	load_TLS_descriptor(t, cpu, 1);
412 	load_TLS_descriptor(t, cpu, 2);
413 
414 	xen_mc_issue(PARAVIRT_LAZY_CPU);
415 }
416 
417 #ifdef CONFIG_X86_64
418 static void xen_load_gs_index(unsigned int idx)
419 {
420 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
421 		BUG();
422 }
423 #endif
424 
425 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
426 				const void *ptr)
427 {
428 	unsigned long lp = (unsigned long)&dt[entrynum];
429 	xmaddr_t mach_lp = virt_to_machine(lp);
430 	u64 entry = *(u64 *)ptr;
431 
432 	preempt_disable();
433 
434 	xen_mc_flush();
435 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
436 		BUG();
437 
438 	preempt_enable();
439 }
440 
441 static int cvt_gate_to_trap(int vector, const gate_desc *val,
442 			    struct trap_info *info)
443 {
444 	if (val->type != 0xf && val->type != 0xe)
445 		return 0;
446 
447 	info->vector = vector;
448 	info->address = gate_offset(*val);
449 	info->cs = gate_segment(*val);
450 	info->flags = val->dpl;
451 	/* interrupt gates clear IF */
452 	if (val->type == 0xe)
453 		info->flags |= 4;
454 
455 	return 1;
456 }
457 
458 /* Locations of each CPU's IDT */
459 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
460 
461 /* Set an IDT entry.  If the entry is part of the current IDT, then
462    also update Xen. */
463 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
464 {
465 	unsigned long p = (unsigned long)&dt[entrynum];
466 	unsigned long start, end;
467 
468 	preempt_disable();
469 
470 	start = __get_cpu_var(idt_desc).address;
471 	end = start + __get_cpu_var(idt_desc).size + 1;
472 
473 	xen_mc_flush();
474 
475 	native_write_idt_entry(dt, entrynum, g);
476 
477 	if (p >= start && (p + 8) <= end) {
478 		struct trap_info info[2];
479 
480 		info[1].address = 0;
481 
482 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
483 			if (HYPERVISOR_set_trap_table(info))
484 				BUG();
485 	}
486 
487 	preempt_enable();
488 }
489 
490 static void xen_convert_trap_info(const struct desc_ptr *desc,
491 				  struct trap_info *traps)
492 {
493 	unsigned in, out, count;
494 
495 	count = (desc->size+1) / sizeof(gate_desc);
496 	BUG_ON(count > 256);
497 
498 	for (in = out = 0; in < count; in++) {
499 		gate_desc *entry = (gate_desc*)(desc->address) + in;
500 
501 		if (cvt_gate_to_trap(in, entry, &traps[out]))
502 			out++;
503 	}
504 	traps[out].address = 0;
505 }
506 
507 void xen_copy_trap_info(struct trap_info *traps)
508 {
509 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
510 
511 	xen_convert_trap_info(desc, traps);
512 }
513 
514 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
515    hold a spinlock to protect the static traps[] array (static because
516    it avoids allocation, and saves stack space). */
517 static void xen_load_idt(const struct desc_ptr *desc)
518 {
519 	static DEFINE_SPINLOCK(lock);
520 	static struct trap_info traps[257];
521 
522 	spin_lock(&lock);
523 
524 	__get_cpu_var(idt_desc) = *desc;
525 
526 	xen_convert_trap_info(desc, traps);
527 
528 	xen_mc_flush();
529 	if (HYPERVISOR_set_trap_table(traps))
530 		BUG();
531 
532 	spin_unlock(&lock);
533 }
534 
535 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
536    they're handled differently. */
537 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
538 				const void *desc, int type)
539 {
540 	preempt_disable();
541 
542 	switch (type) {
543 	case DESC_LDT:
544 	case DESC_TSS:
545 		/* ignore */
546 		break;
547 
548 	default: {
549 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
550 
551 		xen_mc_flush();
552 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
553 			BUG();
554 	}
555 
556 	}
557 
558 	preempt_enable();
559 }
560 
561 static void xen_load_sp0(struct tss_struct *tss,
562 			  struct thread_struct *thread)
563 {
564 	struct multicall_space mcs = xen_mc_entry(0);
565 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
566 	xen_mc_issue(PARAVIRT_LAZY_CPU);
567 }
568 
569 static void xen_set_iopl_mask(unsigned mask)
570 {
571 	struct physdev_set_iopl set_iopl;
572 
573 	/* Force the change at ring 0. */
574 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
575 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
576 }
577 
578 static void xen_io_delay(void)
579 {
580 }
581 
582 #ifdef CONFIG_X86_LOCAL_APIC
583 static u32 xen_apic_read(unsigned long reg)
584 {
585 	return 0;
586 }
587 
588 static void xen_apic_write(unsigned long reg, u32 val)
589 {
590 	/* Warn to see if there's any stray references */
591 	WARN_ON(1);
592 }
593 #endif
594 
595 static void xen_flush_tlb(void)
596 {
597 	struct mmuext_op *op;
598 	struct multicall_space mcs;
599 
600 	preempt_disable();
601 
602 	mcs = xen_mc_entry(sizeof(*op));
603 
604 	op = mcs.args;
605 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
606 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
607 
608 	xen_mc_issue(PARAVIRT_LAZY_MMU);
609 
610 	preempt_enable();
611 }
612 
613 static void xen_flush_tlb_single(unsigned long addr)
614 {
615 	struct mmuext_op *op;
616 	struct multicall_space mcs;
617 
618 	preempt_disable();
619 
620 	mcs = xen_mc_entry(sizeof(*op));
621 	op = mcs.args;
622 	op->cmd = MMUEXT_INVLPG_LOCAL;
623 	op->arg1.linear_addr = addr & PAGE_MASK;
624 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
625 
626 	xen_mc_issue(PARAVIRT_LAZY_MMU);
627 
628 	preempt_enable();
629 }
630 
631 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
632 				 unsigned long va)
633 {
634 	struct {
635 		struct mmuext_op op;
636 		cpumask_t mask;
637 	} *args;
638 	cpumask_t cpumask = *cpus;
639 	struct multicall_space mcs;
640 
641 	/*
642 	 * A couple of (to be removed) sanity checks:
643 	 *
644 	 * - current CPU must not be in mask
645 	 * - mask must exist :)
646 	 */
647 	BUG_ON(cpus_empty(cpumask));
648 	BUG_ON(cpu_isset(smp_processor_id(), cpumask));
649 	BUG_ON(!mm);
650 
651 	/* If a CPU which we ran on has gone down, OK. */
652 	cpus_and(cpumask, cpumask, cpu_online_map);
653 	if (cpus_empty(cpumask))
654 		return;
655 
656 	mcs = xen_mc_entry(sizeof(*args));
657 	args = mcs.args;
658 	args->mask = cpumask;
659 	args->op.arg2.vcpumask = &args->mask;
660 
661 	if (va == TLB_FLUSH_ALL) {
662 		args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
663 	} else {
664 		args->op.cmd = MMUEXT_INVLPG_MULTI;
665 		args->op.arg1.linear_addr = va;
666 	}
667 
668 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
669 
670 	xen_mc_issue(PARAVIRT_LAZY_MMU);
671 }
672 
673 static void xen_clts(void)
674 {
675 	struct multicall_space mcs;
676 
677 	mcs = xen_mc_entry(0);
678 
679 	MULTI_fpu_taskswitch(mcs.mc, 0);
680 
681 	xen_mc_issue(PARAVIRT_LAZY_CPU);
682 }
683 
684 static void xen_write_cr0(unsigned long cr0)
685 {
686 	struct multicall_space mcs;
687 
688 	/* Only pay attention to cr0.TS; everything else is
689 	   ignored. */
690 	mcs = xen_mc_entry(0);
691 
692 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
693 
694 	xen_mc_issue(PARAVIRT_LAZY_CPU);
695 }
696 
697 static void xen_write_cr2(unsigned long cr2)
698 {
699 	x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
700 }
701 
702 static unsigned long xen_read_cr2(void)
703 {
704 	return x86_read_percpu(xen_vcpu)->arch.cr2;
705 }
706 
707 static unsigned long xen_read_cr2_direct(void)
708 {
709 	return x86_read_percpu(xen_vcpu_info.arch.cr2);
710 }
711 
712 static void xen_write_cr4(unsigned long cr4)
713 {
714 	cr4 &= ~X86_CR4_PGE;
715 	cr4 &= ~X86_CR4_PSE;
716 
717 	native_write_cr4(cr4);
718 }
719 
720 static unsigned long xen_read_cr3(void)
721 {
722 	return x86_read_percpu(xen_cr3);
723 }
724 
725 static void set_current_cr3(void *v)
726 {
727 	x86_write_percpu(xen_current_cr3, (unsigned long)v);
728 }
729 
730 static void __xen_write_cr3(bool kernel, unsigned long cr3)
731 {
732 	struct mmuext_op *op;
733 	struct multicall_space mcs;
734 	unsigned long mfn;
735 
736 	if (cr3)
737 		mfn = pfn_to_mfn(PFN_DOWN(cr3));
738 	else
739 		mfn = 0;
740 
741 	WARN_ON(mfn == 0 && kernel);
742 
743 	mcs = __xen_mc_entry(sizeof(*op));
744 
745 	op = mcs.args;
746 	op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
747 	op->arg1.mfn = mfn;
748 
749 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
750 
751 	if (kernel) {
752 		x86_write_percpu(xen_cr3, cr3);
753 
754 		/* Update xen_current_cr3 once the batch has actually
755 		   been submitted. */
756 		xen_mc_callback(set_current_cr3, (void *)cr3);
757 	}
758 }
759 
760 static void xen_write_cr3(unsigned long cr3)
761 {
762 	BUG_ON(preemptible());
763 
764 	xen_mc_batch();  /* disables interrupts */
765 
766 	/* Update while interrupts are disabled, so its atomic with
767 	   respect to ipis */
768 	x86_write_percpu(xen_cr3, cr3);
769 
770 	__xen_write_cr3(true, cr3);
771 
772 #ifdef CONFIG_X86_64
773 	{
774 		pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
775 		if (user_pgd)
776 			__xen_write_cr3(false, __pa(user_pgd));
777 		else
778 			__xen_write_cr3(false, 0);
779 	}
780 #endif
781 
782 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
783 }
784 
785 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
786 {
787 	int ret;
788 
789 	ret = 0;
790 
791 	switch(msr) {
792 #ifdef CONFIG_X86_64
793 		unsigned which;
794 		u64 base;
795 
796 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
797 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
798 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
799 
800 	set:
801 		base = ((u64)high << 32) | low;
802 		if (HYPERVISOR_set_segment_base(which, base) != 0)
803 			ret = -EFAULT;
804 		break;
805 #endif
806 	default:
807 		ret = native_write_msr_safe(msr, low, high);
808 	}
809 
810 	return ret;
811 }
812 
813 /* Early in boot, while setting up the initial pagetable, assume
814    everything is pinned. */
815 static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
816 {
817 #ifdef CONFIG_FLATMEM
818 	BUG_ON(mem_map);	/* should only be used early */
819 #endif
820 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
821 }
822 
823 /* Early release_pte assumes that all pts are pinned, since there's
824    only init_mm and anything attached to that is pinned. */
825 static void xen_release_pte_init(u32 pfn)
826 {
827 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
828 }
829 
830 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
831 {
832 	struct mmuext_op op;
833 	op.cmd = cmd;
834 	op.arg1.mfn = pfn_to_mfn(pfn);
835 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
836 		BUG();
837 }
838 
839 /* This needs to make sure the new pte page is pinned iff its being
840    attached to a pinned pagetable. */
841 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
842 {
843 	struct page *page = pfn_to_page(pfn);
844 
845 	if (PagePinned(virt_to_page(mm->pgd))) {
846 		SetPagePinned(page);
847 
848 		if (!PageHighMem(page)) {
849 			make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
850 			if (level == PT_PTE)
851 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
852 		} else
853 			/* make sure there are no stray mappings of
854 			   this page */
855 			kmap_flush_unused();
856 	}
857 }
858 
859 static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
860 {
861 	xen_alloc_ptpage(mm, pfn, PT_PTE);
862 }
863 
864 static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
865 {
866 	xen_alloc_ptpage(mm, pfn, PT_PMD);
867 }
868 
869 static int xen_pgd_alloc(struct mm_struct *mm)
870 {
871 	pgd_t *pgd = mm->pgd;
872 	int ret = 0;
873 
874 	BUG_ON(PagePinned(virt_to_page(pgd)));
875 
876 #ifdef CONFIG_X86_64
877 	{
878 		struct page *page = virt_to_page(pgd);
879 		pgd_t *user_pgd;
880 
881 		BUG_ON(page->private != 0);
882 
883 		ret = -ENOMEM;
884 
885 		user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
886 		page->private = (unsigned long)user_pgd;
887 
888 		if (user_pgd != NULL) {
889 			user_pgd[pgd_index(VSYSCALL_START)] =
890 				__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
891 			ret = 0;
892 		}
893 
894 		BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
895 	}
896 #endif
897 
898 	return ret;
899 }
900 
901 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
902 {
903 #ifdef CONFIG_X86_64
904 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
905 
906 	if (user_pgd)
907 		free_page((unsigned long)user_pgd);
908 #endif
909 }
910 
911 /* This should never happen until we're OK to use struct page */
912 static void xen_release_ptpage(u32 pfn, unsigned level)
913 {
914 	struct page *page = pfn_to_page(pfn);
915 
916 	if (PagePinned(page)) {
917 		if (!PageHighMem(page)) {
918 			if (level == PT_PTE)
919 				pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
920 			make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
921 		}
922 		ClearPagePinned(page);
923 	}
924 }
925 
926 static void xen_release_pte(u32 pfn)
927 {
928 	xen_release_ptpage(pfn, PT_PTE);
929 }
930 
931 static void xen_release_pmd(u32 pfn)
932 {
933 	xen_release_ptpage(pfn, PT_PMD);
934 }
935 
936 #if PAGETABLE_LEVELS == 4
937 static void xen_alloc_pud(struct mm_struct *mm, u32 pfn)
938 {
939 	xen_alloc_ptpage(mm, pfn, PT_PUD);
940 }
941 
942 static void xen_release_pud(u32 pfn)
943 {
944 	xen_release_ptpage(pfn, PT_PUD);
945 }
946 #endif
947 
948 #ifdef CONFIG_HIGHPTE
949 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
950 {
951 	pgprot_t prot = PAGE_KERNEL;
952 
953 	if (PagePinned(page))
954 		prot = PAGE_KERNEL_RO;
955 
956 	if (0 && PageHighMem(page))
957 		printk("mapping highpte %lx type %d prot %s\n",
958 		       page_to_pfn(page), type,
959 		       (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
960 
961 	return kmap_atomic_prot(page, type, prot);
962 }
963 #endif
964 
965 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
966 {
967 	/* If there's an existing pte, then don't allow _PAGE_RW to be set */
968 	if (pte_val_ma(*ptep) & _PAGE_PRESENT)
969 		pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
970 			       pte_val_ma(pte));
971 
972 	return pte;
973 }
974 
975 /* Init-time set_pte while constructing initial pagetables, which
976    doesn't allow RO pagetable pages to be remapped RW */
977 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
978 {
979 	pte = mask_rw_pte(ptep, pte);
980 
981 	xen_set_pte(ptep, pte);
982 }
983 
984 static __init void xen_pagetable_setup_start(pgd_t *base)
985 {
986 }
987 
988 void xen_setup_shared_info(void)
989 {
990 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
991 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
992 			   xen_start_info->shared_info);
993 
994 		HYPERVISOR_shared_info =
995 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
996 	} else
997 		HYPERVISOR_shared_info =
998 			(struct shared_info *)__va(xen_start_info->shared_info);
999 
1000 #ifndef CONFIG_SMP
1001 	/* In UP this is as good a place as any to set up shared info */
1002 	xen_setup_vcpu_info_placement();
1003 #endif
1004 
1005 	xen_setup_mfn_list_list();
1006 }
1007 
1008 static __init void xen_pagetable_setup_done(pgd_t *base)
1009 {
1010 	xen_setup_shared_info();
1011 }
1012 
1013 static __init void xen_post_allocator_init(void)
1014 {
1015 	pv_mmu_ops.set_pte = xen_set_pte;
1016 	pv_mmu_ops.set_pmd = xen_set_pmd;
1017 	pv_mmu_ops.set_pud = xen_set_pud;
1018 #if PAGETABLE_LEVELS == 4
1019 	pv_mmu_ops.set_pgd = xen_set_pgd;
1020 #endif
1021 
1022 	/* This will work as long as patching hasn't happened yet
1023 	   (which it hasn't) */
1024 	pv_mmu_ops.alloc_pte = xen_alloc_pte;
1025 	pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1026 	pv_mmu_ops.release_pte = xen_release_pte;
1027 	pv_mmu_ops.release_pmd = xen_release_pmd;
1028 #if PAGETABLE_LEVELS == 4
1029 	pv_mmu_ops.alloc_pud = xen_alloc_pud;
1030 	pv_mmu_ops.release_pud = xen_release_pud;
1031 #endif
1032 
1033 #ifdef CONFIG_X86_64
1034 	SetPagePinned(virt_to_page(level3_user_vsyscall));
1035 #endif
1036 	xen_mark_init_mm_pinned();
1037 }
1038 
1039 /* This is called once we have the cpu_possible_map */
1040 void xen_setup_vcpu_info_placement(void)
1041 {
1042 	int cpu;
1043 
1044 	for_each_possible_cpu(cpu)
1045 		xen_vcpu_setup(cpu);
1046 
1047 	/* xen_vcpu_setup managed to place the vcpu_info within the
1048 	   percpu area for all cpus, so make use of it */
1049 #ifdef CONFIG_X86_32
1050 	if (have_vcpu_info_placement) {
1051 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
1052 
1053 		pv_irq_ops.save_fl = xen_save_fl_direct;
1054 		pv_irq_ops.restore_fl = xen_restore_fl_direct;
1055 		pv_irq_ops.irq_disable = xen_irq_disable_direct;
1056 		pv_irq_ops.irq_enable = xen_irq_enable_direct;
1057 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1058 	}
1059 #endif
1060 }
1061 
1062 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1063 			  unsigned long addr, unsigned len)
1064 {
1065 	char *start, *end, *reloc;
1066 	unsigned ret;
1067 
1068 	start = end = reloc = NULL;
1069 
1070 #define SITE(op, x)							\
1071 	case PARAVIRT_PATCH(op.x):					\
1072 	if (have_vcpu_info_placement) {					\
1073 		start = (char *)xen_##x##_direct;			\
1074 		end = xen_##x##_direct_end;				\
1075 		reloc = xen_##x##_direct_reloc;				\
1076 	}								\
1077 	goto patch_site
1078 
1079 	switch (type) {
1080 #ifdef CONFIG_X86_32
1081 		SITE(pv_irq_ops, irq_enable);
1082 		SITE(pv_irq_ops, irq_disable);
1083 		SITE(pv_irq_ops, save_fl);
1084 		SITE(pv_irq_ops, restore_fl);
1085 #endif /* CONFIG_X86_32 */
1086 #undef SITE
1087 
1088 	patch_site:
1089 		if (start == NULL || (end-start) > len)
1090 			goto default_patch;
1091 
1092 		ret = paravirt_patch_insns(insnbuf, len, start, end);
1093 
1094 		/* Note: because reloc is assigned from something that
1095 		   appears to be an array, gcc assumes it's non-null,
1096 		   but doesn't know its relationship with start and
1097 		   end. */
1098 		if (reloc > start && reloc < end) {
1099 			int reloc_off = reloc - start;
1100 			long *relocp = (long *)(insnbuf + reloc_off);
1101 			long delta = start - (char *)addr;
1102 
1103 			*relocp += delta;
1104 		}
1105 		break;
1106 
1107 	default_patch:
1108 	default:
1109 		ret = paravirt_patch_default(type, clobbers, insnbuf,
1110 					     addr, len);
1111 		break;
1112 	}
1113 
1114 	return ret;
1115 }
1116 
1117 static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
1118 {
1119 	pte_t pte;
1120 
1121 	phys >>= PAGE_SHIFT;
1122 
1123 	switch (idx) {
1124 	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1125 #ifdef CONFIG_X86_F00F_BUG
1126 	case FIX_F00F_IDT:
1127 #endif
1128 #ifdef CONFIG_X86_32
1129 	case FIX_WP_TEST:
1130 	case FIX_VDSO:
1131 # ifdef CONFIG_HIGHMEM
1132 	case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1133 # endif
1134 #else
1135 	case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1136 #endif
1137 #ifdef CONFIG_X86_LOCAL_APIC
1138 	case FIX_APIC_BASE:	/* maps dummy local APIC */
1139 #endif
1140 		pte = pfn_pte(phys, prot);
1141 		break;
1142 
1143 	default:
1144 		pte = mfn_pte(phys, prot);
1145 		break;
1146 	}
1147 
1148 	__native_set_fixmap(idx, pte);
1149 
1150 #ifdef CONFIG_X86_64
1151 	/* Replicate changes to map the vsyscall page into the user
1152 	   pagetable vsyscall mapping. */
1153 	if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1154 		unsigned long vaddr = __fix_to_virt(idx);
1155 		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1156 	}
1157 #endif
1158 }
1159 
1160 static const struct pv_info xen_info __initdata = {
1161 	.paravirt_enabled = 1,
1162 	.shared_kernel_pmd = 0,
1163 
1164 	.name = "Xen",
1165 };
1166 
1167 static const struct pv_init_ops xen_init_ops __initdata = {
1168 	.patch = xen_patch,
1169 
1170 	.banner = xen_banner,
1171 	.memory_setup = xen_memory_setup,
1172 	.arch_setup = xen_arch_setup,
1173 	.post_allocator_init = xen_post_allocator_init,
1174 };
1175 
1176 static const struct pv_time_ops xen_time_ops __initdata = {
1177 	.time_init = xen_time_init,
1178 
1179 	.set_wallclock = xen_set_wallclock,
1180 	.get_wallclock = xen_get_wallclock,
1181 	.get_tsc_khz = xen_tsc_khz,
1182 	.sched_clock = xen_sched_clock,
1183 };
1184 
1185 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1186 	.cpuid = xen_cpuid,
1187 
1188 	.set_debugreg = xen_set_debugreg,
1189 	.get_debugreg = xen_get_debugreg,
1190 
1191 	.clts = xen_clts,
1192 
1193 	.read_cr0 = native_read_cr0,
1194 	.write_cr0 = xen_write_cr0,
1195 
1196 	.read_cr4 = native_read_cr4,
1197 	.read_cr4_safe = native_read_cr4_safe,
1198 	.write_cr4 = xen_write_cr4,
1199 
1200 	.wbinvd = native_wbinvd,
1201 
1202 	.read_msr = native_read_msr_safe,
1203 	.write_msr = xen_write_msr_safe,
1204 	.read_tsc = native_read_tsc,
1205 	.read_pmc = native_read_pmc,
1206 
1207 	.iret = xen_iret,
1208 	.irq_enable_sysexit = xen_sysexit,
1209 #ifdef CONFIG_X86_64
1210 	.usergs_sysret32 = xen_sysret32,
1211 	.usergs_sysret64 = xen_sysret64,
1212 #endif
1213 
1214 	.load_tr_desc = paravirt_nop,
1215 	.set_ldt = xen_set_ldt,
1216 	.load_gdt = xen_load_gdt,
1217 	.load_idt = xen_load_idt,
1218 	.load_tls = xen_load_tls,
1219 #ifdef CONFIG_X86_64
1220 	.load_gs_index = xen_load_gs_index,
1221 #endif
1222 
1223 	.store_gdt = native_store_gdt,
1224 	.store_idt = native_store_idt,
1225 	.store_tr = xen_store_tr,
1226 
1227 	.write_ldt_entry = xen_write_ldt_entry,
1228 	.write_gdt_entry = xen_write_gdt_entry,
1229 	.write_idt_entry = xen_write_idt_entry,
1230 	.load_sp0 = xen_load_sp0,
1231 
1232 	.set_iopl_mask = xen_set_iopl_mask,
1233 	.io_delay = xen_io_delay,
1234 
1235 	/* Xen takes care of %gs when switching to usermode for us */
1236 	.swapgs = paravirt_nop,
1237 
1238 	.lazy_mode = {
1239 		.enter = paravirt_enter_lazy_cpu,
1240 		.leave = xen_leave_lazy,
1241 	},
1242 };
1243 
1244 static void __init __xen_init_IRQ(void)
1245 {
1246 #ifdef CONFIG_X86_64
1247 	int i;
1248 
1249 	/* Create identity vector->irq map */
1250 	for(i = 0; i < NR_VECTORS; i++) {
1251 		int cpu;
1252 
1253 		for_each_possible_cpu(cpu)
1254 			per_cpu(vector_irq, cpu)[i] = i;
1255 	}
1256 #endif	/* CONFIG_X86_64 */
1257 
1258 	xen_init_IRQ();
1259 }
1260 
1261 static const struct pv_irq_ops xen_irq_ops __initdata = {
1262 	.init_IRQ = __xen_init_IRQ,
1263 	.save_fl = xen_save_fl,
1264 	.restore_fl = xen_restore_fl,
1265 	.irq_disable = xen_irq_disable,
1266 	.irq_enable = xen_irq_enable,
1267 	.safe_halt = xen_safe_halt,
1268 	.halt = xen_halt,
1269 #ifdef CONFIG_X86_64
1270 	.adjust_exception_frame = xen_adjust_exception_frame,
1271 #endif
1272 };
1273 
1274 static const struct pv_apic_ops xen_apic_ops __initdata = {
1275 #ifdef CONFIG_X86_LOCAL_APIC
1276 	.apic_write = xen_apic_write,
1277 	.apic_read = xen_apic_read,
1278 	.setup_boot_clock = paravirt_nop,
1279 	.setup_secondary_clock = paravirt_nop,
1280 	.startup_ipi_hook = paravirt_nop,
1281 #endif
1282 };
1283 
1284 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1285 	.pagetable_setup_start = xen_pagetable_setup_start,
1286 	.pagetable_setup_done = xen_pagetable_setup_done,
1287 
1288 	.read_cr2 = xen_read_cr2,
1289 	.write_cr2 = xen_write_cr2,
1290 
1291 	.read_cr3 = xen_read_cr3,
1292 	.write_cr3 = xen_write_cr3,
1293 
1294 	.flush_tlb_user = xen_flush_tlb,
1295 	.flush_tlb_kernel = xen_flush_tlb,
1296 	.flush_tlb_single = xen_flush_tlb_single,
1297 	.flush_tlb_others = xen_flush_tlb_others,
1298 
1299 	.pte_update = paravirt_nop,
1300 	.pte_update_defer = paravirt_nop,
1301 
1302 	.pgd_alloc = xen_pgd_alloc,
1303 	.pgd_free = xen_pgd_free,
1304 
1305 	.alloc_pte = xen_alloc_pte_init,
1306 	.release_pte = xen_release_pte_init,
1307 	.alloc_pmd = xen_alloc_pte_init,
1308 	.alloc_pmd_clone = paravirt_nop,
1309 	.release_pmd = xen_release_pte_init,
1310 
1311 #ifdef CONFIG_HIGHPTE
1312 	.kmap_atomic_pte = xen_kmap_atomic_pte,
1313 #endif
1314 
1315 #ifdef CONFIG_X86_64
1316 	.set_pte = xen_set_pte,
1317 #else
1318 	.set_pte = xen_set_pte_init,
1319 #endif
1320 	.set_pte_at = xen_set_pte_at,
1321 	.set_pmd = xen_set_pmd_hyper,
1322 
1323 	.ptep_modify_prot_start = __ptep_modify_prot_start,
1324 	.ptep_modify_prot_commit = __ptep_modify_prot_commit,
1325 
1326 	.pte_val = xen_pte_val,
1327 	.pte_flags = native_pte_val,
1328 	.pgd_val = xen_pgd_val,
1329 
1330 	.make_pte = xen_make_pte,
1331 	.make_pgd = xen_make_pgd,
1332 
1333 #ifdef CONFIG_X86_PAE
1334 	.set_pte_atomic = xen_set_pte_atomic,
1335 	.set_pte_present = xen_set_pte_at,
1336 	.pte_clear = xen_pte_clear,
1337 	.pmd_clear = xen_pmd_clear,
1338 #endif	/* CONFIG_X86_PAE */
1339 	.set_pud = xen_set_pud_hyper,
1340 
1341 	.make_pmd = xen_make_pmd,
1342 	.pmd_val = xen_pmd_val,
1343 
1344 #if PAGETABLE_LEVELS == 4
1345 	.pud_val = xen_pud_val,
1346 	.make_pud = xen_make_pud,
1347 	.set_pgd = xen_set_pgd_hyper,
1348 
1349 	.alloc_pud = xen_alloc_pte_init,
1350 	.release_pud = xen_release_pte_init,
1351 #endif	/* PAGETABLE_LEVELS == 4 */
1352 
1353 	.activate_mm = xen_activate_mm,
1354 	.dup_mmap = xen_dup_mmap,
1355 	.exit_mmap = xen_exit_mmap,
1356 
1357 	.lazy_mode = {
1358 		.enter = paravirt_enter_lazy_mmu,
1359 		.leave = xen_leave_lazy,
1360 	},
1361 
1362 	.set_fixmap = xen_set_fixmap,
1363 };
1364 
1365 static void xen_reboot(int reason)
1366 {
1367 	struct sched_shutdown r = { .reason = reason };
1368 
1369 #ifdef CONFIG_SMP
1370 	smp_send_stop();
1371 #endif
1372 
1373 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1374 		BUG();
1375 }
1376 
1377 static void xen_restart(char *msg)
1378 {
1379 	xen_reboot(SHUTDOWN_reboot);
1380 }
1381 
1382 static void xen_emergency_restart(void)
1383 {
1384 	xen_reboot(SHUTDOWN_reboot);
1385 }
1386 
1387 static void xen_machine_halt(void)
1388 {
1389 	xen_reboot(SHUTDOWN_poweroff);
1390 }
1391 
1392 static void xen_crash_shutdown(struct pt_regs *regs)
1393 {
1394 	xen_reboot(SHUTDOWN_crash);
1395 }
1396 
1397 static const struct machine_ops __initdata xen_machine_ops = {
1398 	.restart = xen_restart,
1399 	.halt = xen_machine_halt,
1400 	.power_off = xen_machine_halt,
1401 	.shutdown = xen_machine_halt,
1402 	.crash_shutdown = xen_crash_shutdown,
1403 	.emergency_restart = xen_emergency_restart,
1404 };
1405 
1406 
1407 static void __init xen_reserve_top(void)
1408 {
1409 #ifdef CONFIG_X86_32
1410 	unsigned long top = HYPERVISOR_VIRT_START;
1411 	struct xen_platform_parameters pp;
1412 
1413 	if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1414 		top = pp.virt_start;
1415 
1416 	reserve_top_address(-top + 2 * PAGE_SIZE);
1417 #endif	/* CONFIG_X86_32 */
1418 }
1419 
1420 /*
1421  * Like __va(), but returns address in the kernel mapping (which is
1422  * all we have until the physical memory mapping has been set up.
1423  */
1424 static void *__ka(phys_addr_t paddr)
1425 {
1426 #ifdef CONFIG_X86_64
1427 	return (void *)(paddr + __START_KERNEL_map);
1428 #else
1429 	return __va(paddr);
1430 #endif
1431 }
1432 
1433 /* Convert a machine address to physical address */
1434 static unsigned long m2p(phys_addr_t maddr)
1435 {
1436 	phys_addr_t paddr;
1437 
1438 	maddr &= PTE_PFN_MASK;
1439 	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1440 
1441 	return paddr;
1442 }
1443 
1444 /* Convert a machine address to kernel virtual */
1445 static void *m2v(phys_addr_t maddr)
1446 {
1447 	return __ka(m2p(maddr));
1448 }
1449 
1450 #ifdef CONFIG_X86_64
1451 static void walk(pgd_t *pgd, unsigned long addr)
1452 {
1453 	unsigned l4idx = pgd_index(addr);
1454 	unsigned l3idx = pud_index(addr);
1455 	unsigned l2idx = pmd_index(addr);
1456 	unsigned l1idx = pte_index(addr);
1457 	pgd_t l4;
1458 	pud_t l3;
1459 	pmd_t l2;
1460 	pte_t l1;
1461 
1462 	xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1463 		       pgd, addr, l4idx, l3idx, l2idx, l1idx);
1464 
1465 	l4 = pgd[l4idx];
1466 	xen_raw_printk("  l4: %016lx\n", l4.pgd);
1467 	xen_raw_printk("      %016lx\n", pgd_val(l4));
1468 
1469 	l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1470 	xen_raw_printk("  l3: %016lx\n", l3.pud);
1471 	xen_raw_printk("      %016lx\n", pud_val(l3));
1472 
1473 	l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1474 	xen_raw_printk("  l2: %016lx\n", l2.pmd);
1475 	xen_raw_printk("      %016lx\n", pmd_val(l2));
1476 
1477 	l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1478 	xen_raw_printk("  l1: %016lx\n", l1.pte);
1479 	xen_raw_printk("      %016lx\n", pte_val(l1));
1480 }
1481 #endif
1482 
1483 static void set_page_prot(void *addr, pgprot_t prot)
1484 {
1485 	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1486 	pte_t pte = pfn_pte(pfn, prot);
1487 
1488 	xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1489 		       addr, pfn, get_phys_to_machine(pfn),
1490 		       pgprot_val(prot), pte.pte);
1491 
1492 	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1493 		BUG();
1494 }
1495 
1496 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1497 {
1498 	unsigned pmdidx, pteidx;
1499 	unsigned ident_pte;
1500 	unsigned long pfn;
1501 
1502 	ident_pte = 0;
1503 	pfn = 0;
1504 	for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1505 		pte_t *pte_page;
1506 
1507 		/* Reuse or allocate a page of ptes */
1508 		if (pmd_present(pmd[pmdidx]))
1509 			pte_page = m2v(pmd[pmdidx].pmd);
1510 		else {
1511 			/* Check for free pte pages */
1512 			if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1513 				break;
1514 
1515 			pte_page = &level1_ident_pgt[ident_pte];
1516 			ident_pte += PTRS_PER_PTE;
1517 
1518 			pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1519 		}
1520 
1521 		/* Install mappings */
1522 		for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1523 			pte_t pte;
1524 
1525 			if (pfn > max_pfn_mapped)
1526 				max_pfn_mapped = pfn;
1527 
1528 			if (!pte_none(pte_page[pteidx]))
1529 				continue;
1530 
1531 			pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1532 			pte_page[pteidx] = pte;
1533 		}
1534 	}
1535 
1536 	for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1537 		set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1538 
1539 	set_page_prot(pmd, PAGE_KERNEL_RO);
1540 }
1541 
1542 #ifdef CONFIG_X86_64
1543 static void convert_pfn_mfn(void *v)
1544 {
1545 	pte_t *pte = v;
1546 	int i;
1547 
1548 	/* All levels are converted the same way, so just treat them
1549 	   as ptes. */
1550 	for(i = 0; i < PTRS_PER_PTE; i++)
1551 		pte[i] = xen_make_pte(pte[i].pte);
1552 }
1553 
1554 /*
1555  * Set up the inital kernel pagetable.
1556  *
1557  * We can construct this by grafting the Xen provided pagetable into
1558  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1559  * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt.  This
1560  * means that only the kernel has a physical mapping to start with -
1561  * but that's enough to get __va working.  We need to fill in the rest
1562  * of the physical mapping once some sort of allocator has been set
1563  * up.
1564  */
1565 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1566 {
1567 	pud_t *l3;
1568 	pmd_t *l2;
1569 
1570 	/* Zap identity mapping */
1571 	init_level4_pgt[0] = __pgd(0);
1572 
1573 	/* Pre-constructed entries are in pfn, so convert to mfn */
1574 	convert_pfn_mfn(init_level4_pgt);
1575 	convert_pfn_mfn(level3_ident_pgt);
1576 	convert_pfn_mfn(level3_kernel_pgt);
1577 
1578 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1579 	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1580 
1581 	memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1582 	memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1583 
1584 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1585 	l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1586 	memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1587 
1588 	/* Set up identity map */
1589 	xen_map_identity_early(level2_ident_pgt, max_pfn);
1590 
1591 	/* Make pagetable pieces RO */
1592 	set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1593 	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1594 	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1595 	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1596 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1597 	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1598 
1599 	/* Pin down new L4 */
1600 	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1601 			  PFN_DOWN(__pa_symbol(init_level4_pgt)));
1602 
1603 	/* Unpin Xen-provided one */
1604 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1605 
1606 	/* Switch over */
1607 	pgd = init_level4_pgt;
1608 
1609 	/*
1610 	 * At this stage there can be no user pgd, and no page
1611 	 * structure to attach it to, so make sure we just set kernel
1612 	 * pgd.
1613 	 */
1614 	xen_mc_batch();
1615 	__xen_write_cr3(true, __pa(pgd));
1616 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1617 
1618 	reserve_early(__pa(xen_start_info->pt_base),
1619 		      __pa(xen_start_info->pt_base +
1620 			   xen_start_info->nr_pt_frames * PAGE_SIZE),
1621 		      "XEN PAGETABLES");
1622 
1623 	return pgd;
1624 }
1625 #else	/* !CONFIG_X86_64 */
1626 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1627 
1628 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1629 {
1630 	pmd_t *kernel_pmd;
1631 
1632 	init_pg_tables_start = __pa(pgd);
1633 	init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1634 	max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1635 
1636 	kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1637 	memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1638 
1639 	xen_map_identity_early(level2_kernel_pgt, max_pfn);
1640 
1641 	memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1642 	set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1643 			__pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1644 
1645 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1646 	set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1647 	set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1648 
1649 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1650 
1651 	xen_write_cr3(__pa(swapper_pg_dir));
1652 
1653 	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1654 
1655 	return swapper_pg_dir;
1656 }
1657 #endif	/* CONFIG_X86_64 */
1658 
1659 /* First C function to be called on Xen boot */
1660 asmlinkage void __init xen_start_kernel(void)
1661 {
1662 	pgd_t *pgd;
1663 
1664 	if (!xen_start_info)
1665 		return;
1666 
1667 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1668 
1669 	xen_setup_features();
1670 
1671 	/* Install Xen paravirt ops */
1672 	pv_info = xen_info;
1673 	pv_init_ops = xen_init_ops;
1674 	pv_time_ops = xen_time_ops;
1675 	pv_cpu_ops = xen_cpu_ops;
1676 	pv_irq_ops = xen_irq_ops;
1677 	pv_apic_ops = xen_apic_ops;
1678 	pv_mmu_ops = xen_mmu_ops;
1679 
1680 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1681 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1682 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1683 	}
1684 
1685 	machine_ops = xen_machine_ops;
1686 
1687 #ifdef CONFIG_X86_64
1688 	/* Disable until direct per-cpu data access. */
1689 	have_vcpu_info_placement = 0;
1690 	x86_64_init_pda();
1691 #endif
1692 
1693 	xen_smp_init();
1694 
1695 	/* Get mfn list */
1696 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1697 		xen_build_dynamic_phys_to_machine();
1698 
1699 	pgd = (pgd_t *)xen_start_info->pt_base;
1700 
1701 	/* Prevent unwanted bits from being set in PTEs. */
1702 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1703 	if (!is_initial_xendomain())
1704 		__supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1705 
1706 	/* Don't do the full vcpu_info placement stuff until we have a
1707 	   possible map and a non-dummy shared_info. */
1708 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1709 
1710 	xen_raw_console_write("mapping kernel into physical memory\n");
1711 	pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1712 
1713 	init_mm.pgd = pgd;
1714 
1715 	/* keep using Xen gdt for now; no urgent need to change it */
1716 
1717 	pv_info.kernel_rpl = 1;
1718 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1719 		pv_info.kernel_rpl = 0;
1720 
1721 	/* set the limit of our address space */
1722 	xen_reserve_top();
1723 
1724 #ifdef CONFIG_X86_32
1725 	/* set up basic CPUID stuff */
1726 	cpu_detect(&new_cpu_data);
1727 	new_cpu_data.hard_math = 1;
1728 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1729 #endif
1730 
1731 	/* Poke various useful things into boot_params */
1732 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1733 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1734 		? __pa(xen_start_info->mod_start) : 0;
1735 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1736 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1737 
1738 	if (!is_initial_xendomain()) {
1739 		add_preferred_console("xenboot", 0, NULL);
1740 		add_preferred_console("tty", 0, NULL);
1741 		add_preferred_console("hvc", 0, NULL);
1742 	}
1743 
1744 	xen_raw_console_write("about to get started...\n");
1745 
1746 #if 0
1747 	xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1748 		       &boot_params, __pa_symbol(&boot_params),
1749 		       __va(__pa_symbol(&boot_params)));
1750 
1751 	walk(pgd, &boot_params);
1752 	walk(pgd, __va(__pa(&boot_params)));
1753 #endif
1754 
1755 	/* Start the world */
1756 #ifdef CONFIG_X86_32
1757 	i386_start_kernel();
1758 #else
1759 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1760 #endif
1761 }
1762