xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 643d1f7f)
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 
29 #include <xen/interface/xen.h>
30 #include <xen/interface/physdev.h>
31 #include <xen/interface/vcpu.h>
32 #include <xen/interface/sched.h>
33 #include <xen/features.h>
34 #include <xen/page.h>
35 
36 #include <asm/paravirt.h>
37 #include <asm/page.h>
38 #include <asm/xen/hypercall.h>
39 #include <asm/xen/hypervisor.h>
40 #include <asm/fixmap.h>
41 #include <asm/processor.h>
42 #include <asm/setup.h>
43 #include <asm/desc.h>
44 #include <asm/pgtable.h>
45 #include <asm/tlbflush.h>
46 #include <asm/reboot.h>
47 
48 #include "xen-ops.h"
49 #include "mmu.h"
50 #include "multicalls.h"
51 
52 EXPORT_SYMBOL_GPL(hypercall_page);
53 
54 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
55 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
56 
57 /*
58  * Note about cr3 (pagetable base) values:
59  *
60  * xen_cr3 contains the current logical cr3 value; it contains the
61  * last set cr3.  This may not be the current effective cr3, because
62  * its update may be being lazily deferred.  However, a vcpu looking
63  * at its own cr3 can use this value knowing that it everything will
64  * be self-consistent.
65  *
66  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
67  * hypercall to set the vcpu cr3 is complete (so it may be a little
68  * out of date, but it will never be set early).  If one vcpu is
69  * looking at another vcpu's cr3 value, it should use this variable.
70  */
71 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
72 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
73 
74 struct start_info *xen_start_info;
75 EXPORT_SYMBOL_GPL(xen_start_info);
76 
77 static /* __initdata */ struct shared_info dummy_shared_info;
78 
79 /*
80  * Point at some empty memory to start with. We map the real shared_info
81  * page as soon as fixmap is up and running.
82  */
83 struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info;
84 
85 /*
86  * Flag to determine whether vcpu info placement is available on all
87  * VCPUs.  We assume it is to start with, and then set it to zero on
88  * the first failure.  This is because it can succeed on some VCPUs
89  * and not others, since it can involve hypervisor memory allocation,
90  * or because the guest failed to guarantee all the appropriate
91  * constraints on all VCPUs (ie buffer can't cross a page boundary).
92  *
93  * Note that any particular CPU may be using a placed vcpu structure,
94  * but we can only optimise if the all are.
95  *
96  * 0: not available, 1: available
97  */
98 static int have_vcpu_info_placement = 0;
99 
100 static void __init xen_vcpu_setup(int cpu)
101 {
102 	struct vcpu_register_vcpu_info info;
103 	int err;
104 	struct vcpu_info *vcpup;
105 
106 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
107 
108 	if (!have_vcpu_info_placement)
109 		return;		/* already tested, not available */
110 
111 	vcpup = &per_cpu(xen_vcpu_info, cpu);
112 
113 	info.mfn = virt_to_mfn(vcpup);
114 	info.offset = offset_in_page(vcpup);
115 
116 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
117 	       cpu, vcpup, info.mfn, info.offset);
118 
119 	/* Check to see if the hypervisor will put the vcpu_info
120 	   structure where we want it, which allows direct access via
121 	   a percpu-variable. */
122 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
123 
124 	if (err) {
125 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
126 		have_vcpu_info_placement = 0;
127 	} else {
128 		/* This cpu is using the registered vcpu info, even if
129 		   later ones fail to. */
130 		per_cpu(xen_vcpu, cpu) = vcpup;
131 
132 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
133 		       cpu, vcpup);
134 	}
135 }
136 
137 static void __init xen_banner(void)
138 {
139 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
140 	       pv_info.name);
141 	printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
142 }
143 
144 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
145 		      unsigned int *cx, unsigned int *dx)
146 {
147 	unsigned maskedx = ~0;
148 
149 	/*
150 	 * Mask out inconvenient features, to try and disable as many
151 	 * unsupported kernel subsystems as possible.
152 	 */
153 	if (*ax == 1)
154 		maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
155 			    (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
156 			    (1 << X86_FEATURE_ACC));   /* thermal monitoring */
157 
158 	asm(XEN_EMULATE_PREFIX "cpuid"
159 		: "=a" (*ax),
160 		  "=b" (*bx),
161 		  "=c" (*cx),
162 		  "=d" (*dx)
163 		: "0" (*ax), "2" (*cx));
164 	*dx &= maskedx;
165 }
166 
167 static void xen_set_debugreg(int reg, unsigned long val)
168 {
169 	HYPERVISOR_set_debugreg(reg, val);
170 }
171 
172 static unsigned long xen_get_debugreg(int reg)
173 {
174 	return HYPERVISOR_get_debugreg(reg);
175 }
176 
177 static unsigned long xen_save_fl(void)
178 {
179 	struct vcpu_info *vcpu;
180 	unsigned long flags;
181 
182 	vcpu = x86_read_percpu(xen_vcpu);
183 
184 	/* flag has opposite sense of mask */
185 	flags = !vcpu->evtchn_upcall_mask;
186 
187 	/* convert to IF type flag
188 	   -0 -> 0x00000000
189 	   -1 -> 0xffffffff
190 	*/
191 	return (-flags) & X86_EFLAGS_IF;
192 }
193 
194 static void xen_restore_fl(unsigned long flags)
195 {
196 	struct vcpu_info *vcpu;
197 
198 	/* convert from IF type flag */
199 	flags = !(flags & X86_EFLAGS_IF);
200 
201 	/* There's a one instruction preempt window here.  We need to
202 	   make sure we're don't switch CPUs between getting the vcpu
203 	   pointer and updating the mask. */
204 	preempt_disable();
205 	vcpu = x86_read_percpu(xen_vcpu);
206 	vcpu->evtchn_upcall_mask = flags;
207 	preempt_enable_no_resched();
208 
209 	/* Doesn't matter if we get preempted here, because any
210 	   pending event will get dealt with anyway. */
211 
212 	if (flags == 0) {
213 		preempt_check_resched();
214 		barrier(); /* unmask then check (avoid races) */
215 		if (unlikely(vcpu->evtchn_upcall_pending))
216 			force_evtchn_callback();
217 	}
218 }
219 
220 static void xen_irq_disable(void)
221 {
222 	/* There's a one instruction preempt window here.  We need to
223 	   make sure we're don't switch CPUs between getting the vcpu
224 	   pointer and updating the mask. */
225 	preempt_disable();
226 	x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
227 	preempt_enable_no_resched();
228 }
229 
230 static void xen_irq_enable(void)
231 {
232 	struct vcpu_info *vcpu;
233 
234 	/* There's a one instruction preempt window here.  We need to
235 	   make sure we're don't switch CPUs between getting the vcpu
236 	   pointer and updating the mask. */
237 	preempt_disable();
238 	vcpu = x86_read_percpu(xen_vcpu);
239 	vcpu->evtchn_upcall_mask = 0;
240 	preempt_enable_no_resched();
241 
242 	/* Doesn't matter if we get preempted here, because any
243 	   pending event will get dealt with anyway. */
244 
245 	barrier(); /* unmask then check (avoid races) */
246 	if (unlikely(vcpu->evtchn_upcall_pending))
247 		force_evtchn_callback();
248 }
249 
250 static void xen_safe_halt(void)
251 {
252 	/* Blocking includes an implicit local_irq_enable(). */
253 	if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
254 		BUG();
255 }
256 
257 static void xen_halt(void)
258 {
259 	if (irqs_disabled())
260 		HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
261 	else
262 		xen_safe_halt();
263 }
264 
265 static void xen_leave_lazy(void)
266 {
267 	paravirt_leave_lazy(paravirt_get_lazy_mode());
268 	xen_mc_flush();
269 }
270 
271 static unsigned long xen_store_tr(void)
272 {
273 	return 0;
274 }
275 
276 static void xen_set_ldt(const void *addr, unsigned entries)
277 {
278 	struct mmuext_op *op;
279 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
280 
281 	op = mcs.args;
282 	op->cmd = MMUEXT_SET_LDT;
283 	op->arg1.linear_addr = (unsigned long)addr;
284 	op->arg2.nr_ents = entries;
285 
286 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
287 
288 	xen_mc_issue(PARAVIRT_LAZY_CPU);
289 }
290 
291 static void xen_load_gdt(const struct desc_ptr *dtr)
292 {
293 	unsigned long *frames;
294 	unsigned long va = dtr->address;
295 	unsigned int size = dtr->size + 1;
296 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
297 	int f;
298 	struct multicall_space mcs;
299 
300 	/* A GDT can be up to 64k in size, which corresponds to 8192
301 	   8-byte entries, or 16 4k pages.. */
302 
303 	BUG_ON(size > 65536);
304 	BUG_ON(va & ~PAGE_MASK);
305 
306 	mcs = xen_mc_entry(sizeof(*frames) * pages);
307 	frames = mcs.args;
308 
309 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
310 		frames[f] = virt_to_mfn(va);
311 		make_lowmem_page_readonly((void *)va);
312 	}
313 
314 	MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
315 
316 	xen_mc_issue(PARAVIRT_LAZY_CPU);
317 }
318 
319 static void load_TLS_descriptor(struct thread_struct *t,
320 				unsigned int cpu, unsigned int i)
321 {
322 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
323 	xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
324 	struct multicall_space mc = __xen_mc_entry(0);
325 
326 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
327 }
328 
329 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
330 {
331 	xen_mc_batch();
332 
333 	load_TLS_descriptor(t, cpu, 0);
334 	load_TLS_descriptor(t, cpu, 1);
335 	load_TLS_descriptor(t, cpu, 2);
336 
337 	xen_mc_issue(PARAVIRT_LAZY_CPU);
338 
339 	/*
340 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
341 	 * it means we're in a context switch, and %gs has just been
342 	 * saved.  This means we can zero it out to prevent faults on
343 	 * exit from the hypervisor if the next process has no %gs.
344 	 * Either way, it has been saved, and the new value will get
345 	 * loaded properly.  This will go away as soon as Xen has been
346 	 * modified to not save/restore %gs for normal hypercalls.
347 	 */
348 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
349 		loadsegment(gs, 0);
350 }
351 
352 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
353 				const void *ptr)
354 {
355 	unsigned long lp = (unsigned long)&dt[entrynum];
356 	xmaddr_t mach_lp = virt_to_machine(lp);
357 	u64 entry = *(u64 *)ptr;
358 
359 	preempt_disable();
360 
361 	xen_mc_flush();
362 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
363 		BUG();
364 
365 	preempt_enable();
366 }
367 
368 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
369 			    struct trap_info *info)
370 {
371 	u8 type, dpl;
372 
373 	type = (high >> 8) & 0x1f;
374 	dpl = (high >> 13) & 3;
375 
376 	if (type != 0xf && type != 0xe)
377 		return 0;
378 
379 	info->vector = vector;
380 	info->address = (high & 0xffff0000) | (low & 0x0000ffff);
381 	info->cs = low >> 16;
382 	info->flags = dpl;
383 	/* interrupt gates clear IF */
384 	if (type == 0xe)
385 		info->flags |= 4;
386 
387 	return 1;
388 }
389 
390 /* Locations of each CPU's IDT */
391 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
392 
393 /* Set an IDT entry.  If the entry is part of the current IDT, then
394    also update Xen. */
395 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
396 {
397 	unsigned long p = (unsigned long)&dt[entrynum];
398 	unsigned long start, end;
399 
400 	preempt_disable();
401 
402 	start = __get_cpu_var(idt_desc).address;
403 	end = start + __get_cpu_var(idt_desc).size + 1;
404 
405 	xen_mc_flush();
406 
407 	native_write_idt_entry(dt, entrynum, g);
408 
409 	if (p >= start && (p + 8) <= end) {
410 		struct trap_info info[2];
411 		u32 *desc = (u32 *)g;
412 
413 		info[1].address = 0;
414 
415 		if (cvt_gate_to_trap(entrynum, desc[0], desc[1], &info[0]))
416 			if (HYPERVISOR_set_trap_table(info))
417 				BUG();
418 	}
419 
420 	preempt_enable();
421 }
422 
423 static void xen_convert_trap_info(const struct desc_ptr *desc,
424 				  struct trap_info *traps)
425 {
426 	unsigned in, out, count;
427 
428 	count = (desc->size+1) / 8;
429 	BUG_ON(count > 256);
430 
431 	for (in = out = 0; in < count; in++) {
432 		const u32 *entry = (u32 *)(desc->address + in * 8);
433 
434 		if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
435 			out++;
436 	}
437 	traps[out].address = 0;
438 }
439 
440 void xen_copy_trap_info(struct trap_info *traps)
441 {
442 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
443 
444 	xen_convert_trap_info(desc, traps);
445 }
446 
447 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
448    hold a spinlock to protect the static traps[] array (static because
449    it avoids allocation, and saves stack space). */
450 static void xen_load_idt(const struct desc_ptr *desc)
451 {
452 	static DEFINE_SPINLOCK(lock);
453 	static struct trap_info traps[257];
454 
455 	spin_lock(&lock);
456 
457 	__get_cpu_var(idt_desc) = *desc;
458 
459 	xen_convert_trap_info(desc, traps);
460 
461 	xen_mc_flush();
462 	if (HYPERVISOR_set_trap_table(traps))
463 		BUG();
464 
465 	spin_unlock(&lock);
466 }
467 
468 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
469    they're handled differently. */
470 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
471 				const void *desc, int type)
472 {
473 	preempt_disable();
474 
475 	switch (type) {
476 	case DESC_LDT:
477 	case DESC_TSS:
478 		/* ignore */
479 		break;
480 
481 	default: {
482 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
483 
484 		xen_mc_flush();
485 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
486 			BUG();
487 	}
488 
489 	}
490 
491 	preempt_enable();
492 }
493 
494 static void xen_load_sp0(struct tss_struct *tss,
495 			  struct thread_struct *thread)
496 {
497 	struct multicall_space mcs = xen_mc_entry(0);
498 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
499 	xen_mc_issue(PARAVIRT_LAZY_CPU);
500 }
501 
502 static void xen_set_iopl_mask(unsigned mask)
503 {
504 	struct physdev_set_iopl set_iopl;
505 
506 	/* Force the change at ring 0. */
507 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
508 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
509 }
510 
511 static void xen_io_delay(void)
512 {
513 }
514 
515 #ifdef CONFIG_X86_LOCAL_APIC
516 static u32 xen_apic_read(unsigned long reg)
517 {
518 	return 0;
519 }
520 
521 static void xen_apic_write(unsigned long reg, u32 val)
522 {
523 	/* Warn to see if there's any stray references */
524 	WARN_ON(1);
525 }
526 #endif
527 
528 static void xen_flush_tlb(void)
529 {
530 	struct mmuext_op *op;
531 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
532 
533 	op = mcs.args;
534 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
535 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
536 
537 	xen_mc_issue(PARAVIRT_LAZY_MMU);
538 }
539 
540 static void xen_flush_tlb_single(unsigned long addr)
541 {
542 	struct mmuext_op *op;
543 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
544 
545 	op = mcs.args;
546 	op->cmd = MMUEXT_INVLPG_LOCAL;
547 	op->arg1.linear_addr = addr & PAGE_MASK;
548 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
549 
550 	xen_mc_issue(PARAVIRT_LAZY_MMU);
551 }
552 
553 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
554 				 unsigned long va)
555 {
556 	struct {
557 		struct mmuext_op op;
558 		cpumask_t mask;
559 	} *args;
560 	cpumask_t cpumask = *cpus;
561 	struct multicall_space mcs;
562 
563 	/*
564 	 * A couple of (to be removed) sanity checks:
565 	 *
566 	 * - current CPU must not be in mask
567 	 * - mask must exist :)
568 	 */
569 	BUG_ON(cpus_empty(cpumask));
570 	BUG_ON(cpu_isset(smp_processor_id(), cpumask));
571 	BUG_ON(!mm);
572 
573 	/* If a CPU which we ran on has gone down, OK. */
574 	cpus_and(cpumask, cpumask, cpu_online_map);
575 	if (cpus_empty(cpumask))
576 		return;
577 
578 	mcs = xen_mc_entry(sizeof(*args));
579 	args = mcs.args;
580 	args->mask = cpumask;
581 	args->op.arg2.vcpumask = &args->mask;
582 
583 	if (va == TLB_FLUSH_ALL) {
584 		args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
585 	} else {
586 		args->op.cmd = MMUEXT_INVLPG_MULTI;
587 		args->op.arg1.linear_addr = va;
588 	}
589 
590 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
591 
592 	xen_mc_issue(PARAVIRT_LAZY_MMU);
593 }
594 
595 static void xen_write_cr2(unsigned long cr2)
596 {
597 	x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
598 }
599 
600 static unsigned long xen_read_cr2(void)
601 {
602 	return x86_read_percpu(xen_vcpu)->arch.cr2;
603 }
604 
605 static unsigned long xen_read_cr2_direct(void)
606 {
607 	return x86_read_percpu(xen_vcpu_info.arch.cr2);
608 }
609 
610 static void xen_write_cr4(unsigned long cr4)
611 {
612 	/* Just ignore cr4 changes; Xen doesn't allow us to do
613 	   anything anyway. */
614 }
615 
616 static unsigned long xen_read_cr3(void)
617 {
618 	return x86_read_percpu(xen_cr3);
619 }
620 
621 static void set_current_cr3(void *v)
622 {
623 	x86_write_percpu(xen_current_cr3, (unsigned long)v);
624 }
625 
626 static void xen_write_cr3(unsigned long cr3)
627 {
628 	struct mmuext_op *op;
629 	struct multicall_space mcs;
630 	unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
631 
632 	BUG_ON(preemptible());
633 
634 	mcs = xen_mc_entry(sizeof(*op));  /* disables interrupts */
635 
636 	/* Update while interrupts are disabled, so its atomic with
637 	   respect to ipis */
638 	x86_write_percpu(xen_cr3, cr3);
639 
640 	op = mcs.args;
641 	op->cmd = MMUEXT_NEW_BASEPTR;
642 	op->arg1.mfn = mfn;
643 
644 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
645 
646 	/* Update xen_update_cr3 once the batch has actually
647 	   been submitted. */
648 	xen_mc_callback(set_current_cr3, (void *)cr3);
649 
650 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
651 }
652 
653 /* Early in boot, while setting up the initial pagetable, assume
654    everything is pinned. */
655 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
656 {
657 	BUG_ON(mem_map);	/* should only be used early */
658 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
659 }
660 
661 /* Early release_pt assumes that all pts are pinned, since there's
662    only init_mm and anything attached to that is pinned. */
663 static void xen_release_pt_init(u32 pfn)
664 {
665 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
666 }
667 
668 static void pin_pagetable_pfn(unsigned level, unsigned long pfn)
669 {
670 	struct mmuext_op op;
671 	op.cmd = level;
672 	op.arg1.mfn = pfn_to_mfn(pfn);
673 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
674 		BUG();
675 }
676 
677 /* This needs to make sure the new pte page is pinned iff its being
678    attached to a pinned pagetable. */
679 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
680 {
681 	struct page *page = pfn_to_page(pfn);
682 
683 	if (PagePinned(virt_to_page(mm->pgd))) {
684 		SetPagePinned(page);
685 
686 		if (!PageHighMem(page)) {
687 			make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
688 			pin_pagetable_pfn(level, pfn);
689 		} else
690 			/* make sure there are no stray mappings of
691 			   this page */
692 			kmap_flush_unused();
693 	}
694 }
695 
696 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
697 {
698 	xen_alloc_ptpage(mm, pfn, MMUEXT_PIN_L1_TABLE);
699 }
700 
701 static void xen_alloc_pd(struct mm_struct *mm, u32 pfn)
702 {
703 	xen_alloc_ptpage(mm, pfn, MMUEXT_PIN_L2_TABLE);
704 }
705 
706 /* This should never happen until we're OK to use struct page */
707 static void xen_release_pt(u32 pfn)
708 {
709 	struct page *page = pfn_to_page(pfn);
710 
711 	if (PagePinned(page)) {
712 		if (!PageHighMem(page)) {
713 			pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
714 			make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
715 		}
716 	}
717 }
718 
719 #ifdef CONFIG_HIGHPTE
720 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
721 {
722 	pgprot_t prot = PAGE_KERNEL;
723 
724 	if (PagePinned(page))
725 		prot = PAGE_KERNEL_RO;
726 
727 	if (0 && PageHighMem(page))
728 		printk("mapping highpte %lx type %d prot %s\n",
729 		       page_to_pfn(page), type,
730 		       (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
731 
732 	return kmap_atomic_prot(page, type, prot);
733 }
734 #endif
735 
736 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
737 {
738 	/* If there's an existing pte, then don't allow _PAGE_RW to be set */
739 	if (pte_val_ma(*ptep) & _PAGE_PRESENT)
740 		pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
741 			       pte_val_ma(pte));
742 
743 	return pte;
744 }
745 
746 /* Init-time set_pte while constructing initial pagetables, which
747    doesn't allow RO pagetable pages to be remapped RW */
748 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
749 {
750 	pte = mask_rw_pte(ptep, pte);
751 
752 	xen_set_pte(ptep, pte);
753 }
754 
755 static __init void xen_pagetable_setup_start(pgd_t *base)
756 {
757 	pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
758 
759 	/* special set_pte for pagetable initialization */
760 	pv_mmu_ops.set_pte = xen_set_pte_init;
761 
762 	init_mm.pgd = base;
763 	/*
764 	 * copy top-level of Xen-supplied pagetable into place.	 For
765 	 * !PAE we can use this as-is, but for PAE it is a stand-in
766 	 * while we copy the pmd pages.
767 	 */
768 	memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
769 
770 	if (PTRS_PER_PMD > 1) {
771 		int i;
772 		/*
773 		 * For PAE, need to allocate new pmds, rather than
774 		 * share Xen's, since Xen doesn't like pmd's being
775 		 * shared between address spaces.
776 		 */
777 		for (i = 0; i < PTRS_PER_PGD; i++) {
778 			if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
779 				pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
780 
781 				memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
782 				       PAGE_SIZE);
783 
784 				make_lowmem_page_readonly(pmd);
785 
786 				set_pgd(&base[i], __pgd(1 + __pa(pmd)));
787 			} else
788 				pgd_clear(&base[i]);
789 		}
790 	}
791 
792 	/* make sure zero_page is mapped RO so we can use it in pagetables */
793 	make_lowmem_page_readonly(empty_zero_page);
794 	make_lowmem_page_readonly(base);
795 	/*
796 	 * Switch to new pagetable.  This is done before
797 	 * pagetable_init has done anything so that the new pages
798 	 * added to the table can be prepared properly for Xen.
799 	 */
800 	xen_write_cr3(__pa(base));
801 }
802 
803 static __init void xen_pagetable_setup_done(pgd_t *base)
804 {
805 	/* This will work as long as patching hasn't happened yet
806 	   (which it hasn't) */
807 	pv_mmu_ops.alloc_pt = xen_alloc_pt;
808 	pv_mmu_ops.alloc_pd = xen_alloc_pd;
809 	pv_mmu_ops.release_pt = xen_release_pt;
810 	pv_mmu_ops.release_pd = xen_release_pt;
811 	pv_mmu_ops.set_pte = xen_set_pte;
812 
813 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
814 		/*
815 		 * Create a mapping for the shared info page.
816 		 * Should be set_fixmap(), but shared_info is a machine
817 		 * address with no corresponding pseudo-phys address.
818 		 */
819 		set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
820 			    PFN_DOWN(xen_start_info->shared_info),
821 			    PAGE_KERNEL);
822 
823 		HYPERVISOR_shared_info =
824 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
825 
826 	} else
827 		HYPERVISOR_shared_info =
828 			(struct shared_info *)__va(xen_start_info->shared_info);
829 
830 	/* Actually pin the pagetable down, but we can't set PG_pinned
831 	   yet because the page structures don't exist yet. */
832 	{
833 		unsigned level;
834 
835 #ifdef CONFIG_X86_PAE
836 		level = MMUEXT_PIN_L3_TABLE;
837 #else
838 		level = MMUEXT_PIN_L2_TABLE;
839 #endif
840 
841 		pin_pagetable_pfn(level, PFN_DOWN(__pa(base)));
842 	}
843 }
844 
845 /* This is called once we have the cpu_possible_map */
846 void __init xen_setup_vcpu_info_placement(void)
847 {
848 	int cpu;
849 
850 	for_each_possible_cpu(cpu)
851 		xen_vcpu_setup(cpu);
852 
853 	/* xen_vcpu_setup managed to place the vcpu_info within the
854 	   percpu area for all cpus, so make use of it */
855 	if (have_vcpu_info_placement) {
856 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
857 
858 		pv_irq_ops.save_fl = xen_save_fl_direct;
859 		pv_irq_ops.restore_fl = xen_restore_fl_direct;
860 		pv_irq_ops.irq_disable = xen_irq_disable_direct;
861 		pv_irq_ops.irq_enable = xen_irq_enable_direct;
862 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
863 		pv_cpu_ops.iret = xen_iret_direct;
864 	}
865 }
866 
867 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
868 			  unsigned long addr, unsigned len)
869 {
870 	char *start, *end, *reloc;
871 	unsigned ret;
872 
873 	start = end = reloc = NULL;
874 
875 #define SITE(op, x)							\
876 	case PARAVIRT_PATCH(op.x):					\
877 	if (have_vcpu_info_placement) {					\
878 		start = (char *)xen_##x##_direct;			\
879 		end = xen_##x##_direct_end;				\
880 		reloc = xen_##x##_direct_reloc;				\
881 	}								\
882 	goto patch_site
883 
884 	switch (type) {
885 		SITE(pv_irq_ops, irq_enable);
886 		SITE(pv_irq_ops, irq_disable);
887 		SITE(pv_irq_ops, save_fl);
888 		SITE(pv_irq_ops, restore_fl);
889 #undef SITE
890 
891 	patch_site:
892 		if (start == NULL || (end-start) > len)
893 			goto default_patch;
894 
895 		ret = paravirt_patch_insns(insnbuf, len, start, end);
896 
897 		/* Note: because reloc is assigned from something that
898 		   appears to be an array, gcc assumes it's non-null,
899 		   but doesn't know its relationship with start and
900 		   end. */
901 		if (reloc > start && reloc < end) {
902 			int reloc_off = reloc - start;
903 			long *relocp = (long *)(insnbuf + reloc_off);
904 			long delta = start - (char *)addr;
905 
906 			*relocp += delta;
907 		}
908 		break;
909 
910 	default_patch:
911 	default:
912 		ret = paravirt_patch_default(type, clobbers, insnbuf,
913 					     addr, len);
914 		break;
915 	}
916 
917 	return ret;
918 }
919 
920 static const struct pv_info xen_info __initdata = {
921 	.paravirt_enabled = 1,
922 	.shared_kernel_pmd = 0,
923 
924 	.name = "Xen",
925 };
926 
927 static const struct pv_init_ops xen_init_ops __initdata = {
928 	.patch = xen_patch,
929 
930 	.banner = xen_banner,
931 	.memory_setup = xen_memory_setup,
932 	.arch_setup = xen_arch_setup,
933 	.post_allocator_init = xen_mark_init_mm_pinned,
934 };
935 
936 static const struct pv_time_ops xen_time_ops __initdata = {
937 	.time_init = xen_time_init,
938 
939 	.set_wallclock = xen_set_wallclock,
940 	.get_wallclock = xen_get_wallclock,
941 	.get_cpu_khz = xen_cpu_khz,
942 	.sched_clock = xen_sched_clock,
943 };
944 
945 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
946 	.cpuid = xen_cpuid,
947 
948 	.set_debugreg = xen_set_debugreg,
949 	.get_debugreg = xen_get_debugreg,
950 
951 	.clts = native_clts,
952 
953 	.read_cr0 = native_read_cr0,
954 	.write_cr0 = native_write_cr0,
955 
956 	.read_cr4 = native_read_cr4,
957 	.read_cr4_safe = native_read_cr4_safe,
958 	.write_cr4 = xen_write_cr4,
959 
960 	.wbinvd = native_wbinvd,
961 
962 	.read_msr = native_read_msr_safe,
963 	.write_msr = native_write_msr_safe,
964 	.read_tsc = native_read_tsc,
965 	.read_pmc = native_read_pmc,
966 
967 	.iret = (void *)&hypercall_page[__HYPERVISOR_iret],
968 	.irq_enable_syscall_ret = NULL,  /* never called */
969 
970 	.load_tr_desc = paravirt_nop,
971 	.set_ldt = xen_set_ldt,
972 	.load_gdt = xen_load_gdt,
973 	.load_idt = xen_load_idt,
974 	.load_tls = xen_load_tls,
975 
976 	.store_gdt = native_store_gdt,
977 	.store_idt = native_store_idt,
978 	.store_tr = xen_store_tr,
979 
980 	.write_ldt_entry = xen_write_ldt_entry,
981 	.write_gdt_entry = xen_write_gdt_entry,
982 	.write_idt_entry = xen_write_idt_entry,
983 	.load_sp0 = xen_load_sp0,
984 
985 	.set_iopl_mask = xen_set_iopl_mask,
986 	.io_delay = xen_io_delay,
987 
988 	.lazy_mode = {
989 		.enter = paravirt_enter_lazy_cpu,
990 		.leave = xen_leave_lazy,
991 	},
992 };
993 
994 static const struct pv_irq_ops xen_irq_ops __initdata = {
995 	.init_IRQ = xen_init_IRQ,
996 	.save_fl = xen_save_fl,
997 	.restore_fl = xen_restore_fl,
998 	.irq_disable = xen_irq_disable,
999 	.irq_enable = xen_irq_enable,
1000 	.safe_halt = xen_safe_halt,
1001 	.halt = xen_halt,
1002 };
1003 
1004 static const struct pv_apic_ops xen_apic_ops __initdata = {
1005 #ifdef CONFIG_X86_LOCAL_APIC
1006 	.apic_write = xen_apic_write,
1007 	.apic_write_atomic = xen_apic_write,
1008 	.apic_read = xen_apic_read,
1009 	.setup_boot_clock = paravirt_nop,
1010 	.setup_secondary_clock = paravirt_nop,
1011 	.startup_ipi_hook = paravirt_nop,
1012 #endif
1013 };
1014 
1015 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1016 	.pagetable_setup_start = xen_pagetable_setup_start,
1017 	.pagetable_setup_done = xen_pagetable_setup_done,
1018 
1019 	.read_cr2 = xen_read_cr2,
1020 	.write_cr2 = xen_write_cr2,
1021 
1022 	.read_cr3 = xen_read_cr3,
1023 	.write_cr3 = xen_write_cr3,
1024 
1025 	.flush_tlb_user = xen_flush_tlb,
1026 	.flush_tlb_kernel = xen_flush_tlb,
1027 	.flush_tlb_single = xen_flush_tlb_single,
1028 	.flush_tlb_others = xen_flush_tlb_others,
1029 
1030 	.pte_update = paravirt_nop,
1031 	.pte_update_defer = paravirt_nop,
1032 
1033 	.alloc_pt = xen_alloc_pt_init,
1034 	.release_pt = xen_release_pt_init,
1035 	.alloc_pd = xen_alloc_pt_init,
1036 	.alloc_pd_clone = paravirt_nop,
1037 	.release_pd = xen_release_pt_init,
1038 
1039 #ifdef CONFIG_HIGHPTE
1040 	.kmap_atomic_pte = xen_kmap_atomic_pte,
1041 #endif
1042 
1043 	.set_pte = NULL,	/* see xen_pagetable_setup_* */
1044 	.set_pte_at = xen_set_pte_at,
1045 	.set_pmd = xen_set_pmd,
1046 
1047 	.pte_val = xen_pte_val,
1048 	.pgd_val = xen_pgd_val,
1049 
1050 	.make_pte = xen_make_pte,
1051 	.make_pgd = xen_make_pgd,
1052 
1053 #ifdef CONFIG_X86_PAE
1054 	.set_pte_atomic = xen_set_pte_atomic,
1055 	.set_pte_present = xen_set_pte_at,
1056 	.set_pud = xen_set_pud,
1057 	.pte_clear = xen_pte_clear,
1058 	.pmd_clear = xen_pmd_clear,
1059 
1060 	.make_pmd = xen_make_pmd,
1061 	.pmd_val = xen_pmd_val,
1062 #endif	/* PAE */
1063 
1064 	.activate_mm = xen_activate_mm,
1065 	.dup_mmap = xen_dup_mmap,
1066 	.exit_mmap = xen_exit_mmap,
1067 
1068 	.lazy_mode = {
1069 		.enter = paravirt_enter_lazy_mmu,
1070 		.leave = xen_leave_lazy,
1071 	},
1072 };
1073 
1074 #ifdef CONFIG_SMP
1075 static const struct smp_ops xen_smp_ops __initdata = {
1076 	.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
1077 	.smp_prepare_cpus = xen_smp_prepare_cpus,
1078 	.cpu_up = xen_cpu_up,
1079 	.smp_cpus_done = xen_smp_cpus_done,
1080 
1081 	.smp_send_stop = xen_smp_send_stop,
1082 	.smp_send_reschedule = xen_smp_send_reschedule,
1083 	.smp_call_function_mask = xen_smp_call_function_mask,
1084 };
1085 #endif	/* CONFIG_SMP */
1086 
1087 static void xen_reboot(int reason)
1088 {
1089 #ifdef CONFIG_SMP
1090 	smp_send_stop();
1091 #endif
1092 
1093 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason))
1094 		BUG();
1095 }
1096 
1097 static void xen_restart(char *msg)
1098 {
1099 	xen_reboot(SHUTDOWN_reboot);
1100 }
1101 
1102 static void xen_emergency_restart(void)
1103 {
1104 	xen_reboot(SHUTDOWN_reboot);
1105 }
1106 
1107 static void xen_machine_halt(void)
1108 {
1109 	xen_reboot(SHUTDOWN_poweroff);
1110 }
1111 
1112 static void xen_crash_shutdown(struct pt_regs *regs)
1113 {
1114 	xen_reboot(SHUTDOWN_crash);
1115 }
1116 
1117 static const struct machine_ops __initdata xen_machine_ops = {
1118 	.restart = xen_restart,
1119 	.halt = xen_machine_halt,
1120 	.power_off = xen_machine_halt,
1121 	.shutdown = xen_machine_halt,
1122 	.crash_shutdown = xen_crash_shutdown,
1123 	.emergency_restart = xen_emergency_restart,
1124 };
1125 
1126 
1127 static void __init xen_reserve_top(void)
1128 {
1129 	unsigned long top = HYPERVISOR_VIRT_START;
1130 	struct xen_platform_parameters pp;
1131 
1132 	if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1133 		top = pp.virt_start;
1134 
1135 	reserve_top_address(-top + 2 * PAGE_SIZE);
1136 }
1137 
1138 /* First C function to be called on Xen boot */
1139 asmlinkage void __init xen_start_kernel(void)
1140 {
1141 	pgd_t *pgd;
1142 
1143 	if (!xen_start_info)
1144 		return;
1145 
1146 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1147 
1148 	/* Install Xen paravirt ops */
1149 	pv_info = xen_info;
1150 	pv_init_ops = xen_init_ops;
1151 	pv_time_ops = xen_time_ops;
1152 	pv_cpu_ops = xen_cpu_ops;
1153 	pv_irq_ops = xen_irq_ops;
1154 	pv_apic_ops = xen_apic_ops;
1155 	pv_mmu_ops = xen_mmu_ops;
1156 
1157 	machine_ops = xen_machine_ops;
1158 
1159 #ifdef CONFIG_SMP
1160 	smp_ops = xen_smp_ops;
1161 #endif
1162 
1163 	xen_setup_features();
1164 
1165 	/* Get mfn list */
1166 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1167 		phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
1168 
1169 	pgd = (pgd_t *)xen_start_info->pt_base;
1170 
1171 	init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1172 
1173 	init_mm.pgd = pgd; /* use the Xen pagetables to start */
1174 
1175 	/* keep using Xen gdt for now; no urgent need to change it */
1176 
1177 	x86_write_percpu(xen_cr3, __pa(pgd));
1178 	x86_write_percpu(xen_current_cr3, __pa(pgd));
1179 
1180 #ifdef CONFIG_SMP
1181 	/* Don't do the full vcpu_info placement stuff until we have a
1182 	   possible map. */
1183 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1184 #else
1185 	/* May as well do it now, since there's no good time to call
1186 	   it later on UP. */
1187 	xen_setup_vcpu_info_placement();
1188 #endif
1189 
1190 	pv_info.kernel_rpl = 1;
1191 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1192 		pv_info.kernel_rpl = 0;
1193 
1194 	/* set the limit of our address space */
1195 	xen_reserve_top();
1196 
1197 	/* set up basic CPUID stuff */
1198 	cpu_detect(&new_cpu_data);
1199 	new_cpu_data.hard_math = 1;
1200 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1201 
1202 	/* Poke various useful things into boot_params */
1203 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1204 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1205 		? __pa(xen_start_info->mod_start) : 0;
1206 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1207 
1208 	/* Start the world */
1209 	start_kernel();
1210 }
1211