xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision a1e58bbd)
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68  * out of date, but it will never be set early).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 = 1;
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 	BUG_ON(HYPERVISOR_shared_info == &dummy_shared_info);
107 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
108 
109 	if (!have_vcpu_info_placement)
110 		return;		/* already tested, not available */
111 
112 	vcpup = &per_cpu(xen_vcpu_info, cpu);
113 
114 	info.mfn = virt_to_mfn(vcpup);
115 	info.offset = offset_in_page(vcpup);
116 
117 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
118 	       cpu, vcpup, info.mfn, info.offset);
119 
120 	/* Check to see if the hypervisor will put the vcpu_info
121 	   structure where we want it, which allows direct access via
122 	   a percpu-variable. */
123 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
124 
125 	if (err) {
126 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
127 		have_vcpu_info_placement = 0;
128 	} else {
129 		/* This cpu is using the registered vcpu info, even if
130 		   later ones fail to. */
131 		per_cpu(xen_vcpu, cpu) = vcpup;
132 
133 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
134 		       cpu, vcpup);
135 	}
136 }
137 
138 static void __init xen_banner(void)
139 {
140 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
141 	       pv_info.name);
142 	printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
143 }
144 
145 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
146 		      unsigned int *cx, unsigned int *dx)
147 {
148 	unsigned maskedx = ~0;
149 
150 	/*
151 	 * Mask out inconvenient features, to try and disable as many
152 	 * unsupported kernel subsystems as possible.
153 	 */
154 	if (*ax == 1)
155 		maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
156 			    (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
157 			    (1 << X86_FEATURE_SEP)  |  /* disable SEP */
158 			    (1 << X86_FEATURE_ACC));   /* thermal monitoring */
159 
160 	asm(XEN_EMULATE_PREFIX "cpuid"
161 		: "=a" (*ax),
162 		  "=b" (*bx),
163 		  "=c" (*cx),
164 		  "=d" (*dx)
165 		: "0" (*ax), "2" (*cx));
166 	*dx &= maskedx;
167 }
168 
169 static void xen_set_debugreg(int reg, unsigned long val)
170 {
171 	HYPERVISOR_set_debugreg(reg, val);
172 }
173 
174 static unsigned long xen_get_debugreg(int reg)
175 {
176 	return HYPERVISOR_get_debugreg(reg);
177 }
178 
179 static unsigned long xen_save_fl(void)
180 {
181 	struct vcpu_info *vcpu;
182 	unsigned long flags;
183 
184 	vcpu = x86_read_percpu(xen_vcpu);
185 
186 	/* flag has opposite sense of mask */
187 	flags = !vcpu->evtchn_upcall_mask;
188 
189 	/* convert to IF type flag
190 	   -0 -> 0x00000000
191 	   -1 -> 0xffffffff
192 	*/
193 	return (-flags) & X86_EFLAGS_IF;
194 }
195 
196 static void xen_restore_fl(unsigned long flags)
197 {
198 	struct vcpu_info *vcpu;
199 
200 	/* convert from IF type flag */
201 	flags = !(flags & X86_EFLAGS_IF);
202 
203 	/* There's a one instruction preempt window here.  We need to
204 	   make sure we're don't switch CPUs between getting the vcpu
205 	   pointer and updating the mask. */
206 	preempt_disable();
207 	vcpu = x86_read_percpu(xen_vcpu);
208 	vcpu->evtchn_upcall_mask = flags;
209 	preempt_enable_no_resched();
210 
211 	/* Doesn't matter if we get preempted here, because any
212 	   pending event will get dealt with anyway. */
213 
214 	if (flags == 0) {
215 		preempt_check_resched();
216 		barrier(); /* unmask then check (avoid races) */
217 		if (unlikely(vcpu->evtchn_upcall_pending))
218 			force_evtchn_callback();
219 	}
220 }
221 
222 static void xen_irq_disable(void)
223 {
224 	/* There's a one instruction preempt window here.  We need to
225 	   make sure we're don't switch CPUs between getting the vcpu
226 	   pointer and updating the mask. */
227 	preempt_disable();
228 	x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
229 	preempt_enable_no_resched();
230 }
231 
232 static void xen_irq_enable(void)
233 {
234 	struct vcpu_info *vcpu;
235 
236 	/* There's a one instruction preempt window here.  We need to
237 	   make sure we're don't switch CPUs between getting the vcpu
238 	   pointer and updating the mask. */
239 	preempt_disable();
240 	vcpu = x86_read_percpu(xen_vcpu);
241 	vcpu->evtchn_upcall_mask = 0;
242 	preempt_enable_no_resched();
243 
244 	/* Doesn't matter if we get preempted here, because any
245 	   pending event will get dealt with anyway. */
246 
247 	barrier(); /* unmask then check (avoid races) */
248 	if (unlikely(vcpu->evtchn_upcall_pending))
249 		force_evtchn_callback();
250 }
251 
252 static void xen_safe_halt(void)
253 {
254 	/* Blocking includes an implicit local_irq_enable(). */
255 	if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
256 		BUG();
257 }
258 
259 static void xen_halt(void)
260 {
261 	if (irqs_disabled())
262 		HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
263 	else
264 		xen_safe_halt();
265 }
266 
267 static void xen_leave_lazy(void)
268 {
269 	paravirt_leave_lazy(paravirt_get_lazy_mode());
270 	xen_mc_flush();
271 }
272 
273 static unsigned long xen_store_tr(void)
274 {
275 	return 0;
276 }
277 
278 static void xen_set_ldt(const void *addr, unsigned entries)
279 {
280 	struct mmuext_op *op;
281 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
282 
283 	op = mcs.args;
284 	op->cmd = MMUEXT_SET_LDT;
285 	op->arg1.linear_addr = (unsigned long)addr;
286 	op->arg2.nr_ents = entries;
287 
288 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
289 
290 	xen_mc_issue(PARAVIRT_LAZY_CPU);
291 }
292 
293 static void xen_load_gdt(const struct desc_ptr *dtr)
294 {
295 	unsigned long *frames;
296 	unsigned long va = dtr->address;
297 	unsigned int size = dtr->size + 1;
298 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
299 	int f;
300 	struct multicall_space mcs;
301 
302 	/* A GDT can be up to 64k in size, which corresponds to 8192
303 	   8-byte entries, or 16 4k pages.. */
304 
305 	BUG_ON(size > 65536);
306 	BUG_ON(va & ~PAGE_MASK);
307 
308 	mcs = xen_mc_entry(sizeof(*frames) * pages);
309 	frames = mcs.args;
310 
311 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
312 		frames[f] = virt_to_mfn(va);
313 		make_lowmem_page_readonly((void *)va);
314 	}
315 
316 	MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
317 
318 	xen_mc_issue(PARAVIRT_LAZY_CPU);
319 }
320 
321 static void load_TLS_descriptor(struct thread_struct *t,
322 				unsigned int cpu, unsigned int i)
323 {
324 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
325 	xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
326 	struct multicall_space mc = __xen_mc_entry(0);
327 
328 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
329 }
330 
331 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
332 {
333 	xen_mc_batch();
334 
335 	load_TLS_descriptor(t, cpu, 0);
336 	load_TLS_descriptor(t, cpu, 1);
337 	load_TLS_descriptor(t, cpu, 2);
338 
339 	xen_mc_issue(PARAVIRT_LAZY_CPU);
340 
341 	/*
342 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
343 	 * it means we're in a context switch, and %gs has just been
344 	 * saved.  This means we can zero it out to prevent faults on
345 	 * exit from the hypervisor if the next process has no %gs.
346 	 * Either way, it has been saved, and the new value will get
347 	 * loaded properly.  This will go away as soon as Xen has been
348 	 * modified to not save/restore %gs for normal hypercalls.
349 	 */
350 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
351 		loadsegment(gs, 0);
352 }
353 
354 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
355 				const void *ptr)
356 {
357 	unsigned long lp = (unsigned long)&dt[entrynum];
358 	xmaddr_t mach_lp = virt_to_machine(lp);
359 	u64 entry = *(u64 *)ptr;
360 
361 	preempt_disable();
362 
363 	xen_mc_flush();
364 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
365 		BUG();
366 
367 	preempt_enable();
368 }
369 
370 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
371 			    struct trap_info *info)
372 {
373 	u8 type, dpl;
374 
375 	type = (high >> 8) & 0x1f;
376 	dpl = (high >> 13) & 3;
377 
378 	if (type != 0xf && type != 0xe)
379 		return 0;
380 
381 	info->vector = vector;
382 	info->address = (high & 0xffff0000) | (low & 0x0000ffff);
383 	info->cs = low >> 16;
384 	info->flags = dpl;
385 	/* interrupt gates clear IF */
386 	if (type == 0xe)
387 		info->flags |= 4;
388 
389 	return 1;
390 }
391 
392 /* Locations of each CPU's IDT */
393 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
394 
395 /* Set an IDT entry.  If the entry is part of the current IDT, then
396    also update Xen. */
397 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
398 {
399 	unsigned long p = (unsigned long)&dt[entrynum];
400 	unsigned long start, end;
401 
402 	preempt_disable();
403 
404 	start = __get_cpu_var(idt_desc).address;
405 	end = start + __get_cpu_var(idt_desc).size + 1;
406 
407 	xen_mc_flush();
408 
409 	native_write_idt_entry(dt, entrynum, g);
410 
411 	if (p >= start && (p + 8) <= end) {
412 		struct trap_info info[2];
413 		u32 *desc = (u32 *)g;
414 
415 		info[1].address = 0;
416 
417 		if (cvt_gate_to_trap(entrynum, desc[0], desc[1], &info[0]))
418 			if (HYPERVISOR_set_trap_table(info))
419 				BUG();
420 	}
421 
422 	preempt_enable();
423 }
424 
425 static void xen_convert_trap_info(const struct desc_ptr *desc,
426 				  struct trap_info *traps)
427 {
428 	unsigned in, out, count;
429 
430 	count = (desc->size+1) / 8;
431 	BUG_ON(count > 256);
432 
433 	for (in = out = 0; in < count; in++) {
434 		const u32 *entry = (u32 *)(desc->address + in * 8);
435 
436 		if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
437 			out++;
438 	}
439 	traps[out].address = 0;
440 }
441 
442 void xen_copy_trap_info(struct trap_info *traps)
443 {
444 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
445 
446 	xen_convert_trap_info(desc, traps);
447 }
448 
449 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
450    hold a spinlock to protect the static traps[] array (static because
451    it avoids allocation, and saves stack space). */
452 static void xen_load_idt(const struct desc_ptr *desc)
453 {
454 	static DEFINE_SPINLOCK(lock);
455 	static struct trap_info traps[257];
456 
457 	spin_lock(&lock);
458 
459 	__get_cpu_var(idt_desc) = *desc;
460 
461 	xen_convert_trap_info(desc, traps);
462 
463 	xen_mc_flush();
464 	if (HYPERVISOR_set_trap_table(traps))
465 		BUG();
466 
467 	spin_unlock(&lock);
468 }
469 
470 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
471    they're handled differently. */
472 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
473 				const void *desc, int type)
474 {
475 	preempt_disable();
476 
477 	switch (type) {
478 	case DESC_LDT:
479 	case DESC_TSS:
480 		/* ignore */
481 		break;
482 
483 	default: {
484 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
485 
486 		xen_mc_flush();
487 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
488 			BUG();
489 	}
490 
491 	}
492 
493 	preempt_enable();
494 }
495 
496 static void xen_load_sp0(struct tss_struct *tss,
497 			  struct thread_struct *thread)
498 {
499 	struct multicall_space mcs = xen_mc_entry(0);
500 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
501 	xen_mc_issue(PARAVIRT_LAZY_CPU);
502 }
503 
504 static void xen_set_iopl_mask(unsigned mask)
505 {
506 	struct physdev_set_iopl set_iopl;
507 
508 	/* Force the change at ring 0. */
509 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
510 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
511 }
512 
513 static void xen_io_delay(void)
514 {
515 }
516 
517 #ifdef CONFIG_X86_LOCAL_APIC
518 static u32 xen_apic_read(unsigned long reg)
519 {
520 	return 0;
521 }
522 
523 static void xen_apic_write(unsigned long reg, u32 val)
524 {
525 	/* Warn to see if there's any stray references */
526 	WARN_ON(1);
527 }
528 #endif
529 
530 static void xen_flush_tlb(void)
531 {
532 	struct mmuext_op *op;
533 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
534 
535 	op = mcs.args;
536 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
537 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
538 
539 	xen_mc_issue(PARAVIRT_LAZY_MMU);
540 }
541 
542 static void xen_flush_tlb_single(unsigned long addr)
543 {
544 	struct mmuext_op *op;
545 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
546 
547 	op = mcs.args;
548 	op->cmd = MMUEXT_INVLPG_LOCAL;
549 	op->arg1.linear_addr = addr & PAGE_MASK;
550 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
551 
552 	xen_mc_issue(PARAVIRT_LAZY_MMU);
553 }
554 
555 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
556 				 unsigned long va)
557 {
558 	struct {
559 		struct mmuext_op op;
560 		cpumask_t mask;
561 	} *args;
562 	cpumask_t cpumask = *cpus;
563 	struct multicall_space mcs;
564 
565 	/*
566 	 * A couple of (to be removed) sanity checks:
567 	 *
568 	 * - current CPU must not be in mask
569 	 * - mask must exist :)
570 	 */
571 	BUG_ON(cpus_empty(cpumask));
572 	BUG_ON(cpu_isset(smp_processor_id(), cpumask));
573 	BUG_ON(!mm);
574 
575 	/* If a CPU which we ran on has gone down, OK. */
576 	cpus_and(cpumask, cpumask, cpu_online_map);
577 	if (cpus_empty(cpumask))
578 		return;
579 
580 	mcs = xen_mc_entry(sizeof(*args));
581 	args = mcs.args;
582 	args->mask = cpumask;
583 	args->op.arg2.vcpumask = &args->mask;
584 
585 	if (va == TLB_FLUSH_ALL) {
586 		args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
587 	} else {
588 		args->op.cmd = MMUEXT_INVLPG_MULTI;
589 		args->op.arg1.linear_addr = va;
590 	}
591 
592 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
593 
594 	xen_mc_issue(PARAVIRT_LAZY_MMU);
595 }
596 
597 static void xen_write_cr2(unsigned long cr2)
598 {
599 	x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
600 }
601 
602 static unsigned long xen_read_cr2(void)
603 {
604 	return x86_read_percpu(xen_vcpu)->arch.cr2;
605 }
606 
607 static unsigned long xen_read_cr2_direct(void)
608 {
609 	return x86_read_percpu(xen_vcpu_info.arch.cr2);
610 }
611 
612 static void xen_write_cr4(unsigned long cr4)
613 {
614 	/* Just ignore cr4 changes; Xen doesn't allow us to do
615 	   anything anyway. */
616 }
617 
618 static unsigned long xen_read_cr3(void)
619 {
620 	return x86_read_percpu(xen_cr3);
621 }
622 
623 static void set_current_cr3(void *v)
624 {
625 	x86_write_percpu(xen_current_cr3, (unsigned long)v);
626 }
627 
628 static void xen_write_cr3(unsigned long cr3)
629 {
630 	struct mmuext_op *op;
631 	struct multicall_space mcs;
632 	unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
633 
634 	BUG_ON(preemptible());
635 
636 	mcs = xen_mc_entry(sizeof(*op));  /* disables interrupts */
637 
638 	/* Update while interrupts are disabled, so its atomic with
639 	   respect to ipis */
640 	x86_write_percpu(xen_cr3, cr3);
641 
642 	op = mcs.args;
643 	op->cmd = MMUEXT_NEW_BASEPTR;
644 	op->arg1.mfn = mfn;
645 
646 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
647 
648 	/* Update xen_update_cr3 once the batch has actually
649 	   been submitted. */
650 	xen_mc_callback(set_current_cr3, (void *)cr3);
651 
652 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
653 }
654 
655 /* Early in boot, while setting up the initial pagetable, assume
656    everything is pinned. */
657 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
658 {
659 	BUG_ON(mem_map);	/* should only be used early */
660 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
661 }
662 
663 /* Early release_pt assumes that all pts are pinned, since there's
664    only init_mm and anything attached to that is pinned. */
665 static void xen_release_pt_init(u32 pfn)
666 {
667 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
668 }
669 
670 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
671 {
672 	struct mmuext_op op;
673 	op.cmd = cmd;
674 	op.arg1.mfn = pfn_to_mfn(pfn);
675 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
676 		BUG();
677 }
678 
679 /* This needs to make sure the new pte page is pinned iff its being
680    attached to a pinned pagetable. */
681 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
682 {
683 	struct page *page = pfn_to_page(pfn);
684 
685 	if (PagePinned(virt_to_page(mm->pgd))) {
686 		SetPagePinned(page);
687 
688 		if (!PageHighMem(page)) {
689 			make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
690 			if (level == PT_PTE)
691 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
692 		} else
693 			/* make sure there are no stray mappings of
694 			   this page */
695 			kmap_flush_unused();
696 	}
697 }
698 
699 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
700 {
701 	xen_alloc_ptpage(mm, pfn, PT_PTE);
702 }
703 
704 static void xen_alloc_pd(struct mm_struct *mm, u32 pfn)
705 {
706 	xen_alloc_ptpage(mm, pfn, PT_PMD);
707 }
708 
709 /* This should never happen until we're OK to use struct page */
710 static void xen_release_ptpage(u32 pfn, unsigned level)
711 {
712 	struct page *page = pfn_to_page(pfn);
713 
714 	if (PagePinned(page)) {
715 		if (!PageHighMem(page)) {
716 			if (level == PT_PTE)
717 				pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
718 			make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
719 		}
720 		ClearPagePinned(page);
721 	}
722 }
723 
724 static void xen_release_pt(u32 pfn)
725 {
726 	xen_release_ptpage(pfn, PT_PTE);
727 }
728 
729 static void xen_release_pd(u32 pfn)
730 {
731 	xen_release_ptpage(pfn, PT_PMD);
732 }
733 
734 #ifdef CONFIG_HIGHPTE
735 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
736 {
737 	pgprot_t prot = PAGE_KERNEL;
738 
739 	if (PagePinned(page))
740 		prot = PAGE_KERNEL_RO;
741 
742 	if (0 && PageHighMem(page))
743 		printk("mapping highpte %lx type %d prot %s\n",
744 		       page_to_pfn(page), type,
745 		       (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
746 
747 	return kmap_atomic_prot(page, type, prot);
748 }
749 #endif
750 
751 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
752 {
753 	/* If there's an existing pte, then don't allow _PAGE_RW to be set */
754 	if (pte_val_ma(*ptep) & _PAGE_PRESENT)
755 		pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
756 			       pte_val_ma(pte));
757 
758 	return pte;
759 }
760 
761 /* Init-time set_pte while constructing initial pagetables, which
762    doesn't allow RO pagetable pages to be remapped RW */
763 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
764 {
765 	pte = mask_rw_pte(ptep, pte);
766 
767 	xen_set_pte(ptep, pte);
768 }
769 
770 static __init void xen_pagetable_setup_start(pgd_t *base)
771 {
772 	pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
773 
774 	/* special set_pte for pagetable initialization */
775 	pv_mmu_ops.set_pte = xen_set_pte_init;
776 
777 	init_mm.pgd = base;
778 	/*
779 	 * copy top-level of Xen-supplied pagetable into place.	 For
780 	 * !PAE we can use this as-is, but for PAE it is a stand-in
781 	 * while we copy the pmd pages.
782 	 */
783 	memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
784 
785 	if (PTRS_PER_PMD > 1) {
786 		int i;
787 		/*
788 		 * For PAE, need to allocate new pmds, rather than
789 		 * share Xen's, since Xen doesn't like pmd's being
790 		 * shared between address spaces.
791 		 */
792 		for (i = 0; i < PTRS_PER_PGD; i++) {
793 			if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
794 				pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
795 
796 				memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
797 				       PAGE_SIZE);
798 
799 				make_lowmem_page_readonly(pmd);
800 
801 				set_pgd(&base[i], __pgd(1 + __pa(pmd)));
802 			} else
803 				pgd_clear(&base[i]);
804 		}
805 	}
806 
807 	/* make sure zero_page is mapped RO so we can use it in pagetables */
808 	make_lowmem_page_readonly(empty_zero_page);
809 	make_lowmem_page_readonly(base);
810 	/*
811 	 * Switch to new pagetable.  This is done before
812 	 * pagetable_init has done anything so that the new pages
813 	 * added to the table can be prepared properly for Xen.
814 	 */
815 	xen_write_cr3(__pa(base));
816 
817 	/* Unpin initial Xen pagetable */
818 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
819 			  PFN_DOWN(__pa(xen_start_info->pt_base)));
820 }
821 
822 static __init void setup_shared_info(void)
823 {
824 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
825 		unsigned long addr = fix_to_virt(FIX_PARAVIRT_BOOTMAP);
826 
827 		/*
828 		 * Create a mapping for the shared info page.
829 		 * Should be set_fixmap(), but shared_info is a machine
830 		 * address with no corresponding pseudo-phys address.
831 		 */
832 		set_pte_mfn(addr,
833 			    PFN_DOWN(xen_start_info->shared_info),
834 			    PAGE_KERNEL);
835 
836 		HYPERVISOR_shared_info = (struct shared_info *)addr;
837 	} else
838 		HYPERVISOR_shared_info =
839 			(struct shared_info *)__va(xen_start_info->shared_info);
840 
841 #ifndef CONFIG_SMP
842 	/* In UP this is as good a place as any to set up shared info */
843 	xen_setup_vcpu_info_placement();
844 #endif
845 }
846 
847 static __init void xen_pagetable_setup_done(pgd_t *base)
848 {
849 	/* This will work as long as patching hasn't happened yet
850 	   (which it hasn't) */
851 	pv_mmu_ops.alloc_pt = xen_alloc_pt;
852 	pv_mmu_ops.alloc_pd = xen_alloc_pd;
853 	pv_mmu_ops.release_pt = xen_release_pt;
854 	pv_mmu_ops.release_pd = xen_release_pd;
855 	pv_mmu_ops.set_pte = xen_set_pte;
856 
857 	setup_shared_info();
858 
859 	/* Actually pin the pagetable down, but we can't set PG_pinned
860 	   yet because the page structures don't exist yet. */
861 	{
862 		unsigned level;
863 
864 #ifdef CONFIG_X86_PAE
865 		level = MMUEXT_PIN_L3_TABLE;
866 #else
867 		level = MMUEXT_PIN_L2_TABLE;
868 #endif
869 
870 		pin_pagetable_pfn(level, PFN_DOWN(__pa(base)));
871 	}
872 }
873 
874 /* This is called once we have the cpu_possible_map */
875 void __init xen_setup_vcpu_info_placement(void)
876 {
877 	int cpu;
878 
879 	for_each_possible_cpu(cpu)
880 		xen_vcpu_setup(cpu);
881 
882 	/* xen_vcpu_setup managed to place the vcpu_info within the
883 	   percpu area for all cpus, so make use of it */
884 	if (have_vcpu_info_placement) {
885 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
886 
887 		pv_irq_ops.save_fl = xen_save_fl_direct;
888 		pv_irq_ops.restore_fl = xen_restore_fl_direct;
889 		pv_irq_ops.irq_disable = xen_irq_disable_direct;
890 		pv_irq_ops.irq_enable = xen_irq_enable_direct;
891 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
892 		pv_cpu_ops.iret = xen_iret_direct;
893 	}
894 }
895 
896 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
897 			  unsigned long addr, unsigned len)
898 {
899 	char *start, *end, *reloc;
900 	unsigned ret;
901 
902 	start = end = reloc = NULL;
903 
904 #define SITE(op, x)							\
905 	case PARAVIRT_PATCH(op.x):					\
906 	if (have_vcpu_info_placement) {					\
907 		start = (char *)xen_##x##_direct;			\
908 		end = xen_##x##_direct_end;				\
909 		reloc = xen_##x##_direct_reloc;				\
910 	}								\
911 	goto patch_site
912 
913 	switch (type) {
914 		SITE(pv_irq_ops, irq_enable);
915 		SITE(pv_irq_ops, irq_disable);
916 		SITE(pv_irq_ops, save_fl);
917 		SITE(pv_irq_ops, restore_fl);
918 #undef SITE
919 
920 	patch_site:
921 		if (start == NULL || (end-start) > len)
922 			goto default_patch;
923 
924 		ret = paravirt_patch_insns(insnbuf, len, start, end);
925 
926 		/* Note: because reloc is assigned from something that
927 		   appears to be an array, gcc assumes it's non-null,
928 		   but doesn't know its relationship with start and
929 		   end. */
930 		if (reloc > start && reloc < end) {
931 			int reloc_off = reloc - start;
932 			long *relocp = (long *)(insnbuf + reloc_off);
933 			long delta = start - (char *)addr;
934 
935 			*relocp += delta;
936 		}
937 		break;
938 
939 	default_patch:
940 	default:
941 		ret = paravirt_patch_default(type, clobbers, insnbuf,
942 					     addr, len);
943 		break;
944 	}
945 
946 	return ret;
947 }
948 
949 static const struct pv_info xen_info __initdata = {
950 	.paravirt_enabled = 1,
951 	.shared_kernel_pmd = 0,
952 
953 	.name = "Xen",
954 };
955 
956 static const struct pv_init_ops xen_init_ops __initdata = {
957 	.patch = xen_patch,
958 
959 	.banner = xen_banner,
960 	.memory_setup = xen_memory_setup,
961 	.arch_setup = xen_arch_setup,
962 	.post_allocator_init = xen_mark_init_mm_pinned,
963 };
964 
965 static const struct pv_time_ops xen_time_ops __initdata = {
966 	.time_init = xen_time_init,
967 
968 	.set_wallclock = xen_set_wallclock,
969 	.get_wallclock = xen_get_wallclock,
970 	.get_cpu_khz = xen_cpu_khz,
971 	.sched_clock = xen_sched_clock,
972 };
973 
974 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
975 	.cpuid = xen_cpuid,
976 
977 	.set_debugreg = xen_set_debugreg,
978 	.get_debugreg = xen_get_debugreg,
979 
980 	.clts = native_clts,
981 
982 	.read_cr0 = native_read_cr0,
983 	.write_cr0 = native_write_cr0,
984 
985 	.read_cr4 = native_read_cr4,
986 	.read_cr4_safe = native_read_cr4_safe,
987 	.write_cr4 = xen_write_cr4,
988 
989 	.wbinvd = native_wbinvd,
990 
991 	.read_msr = native_read_msr_safe,
992 	.write_msr = native_write_msr_safe,
993 	.read_tsc = native_read_tsc,
994 	.read_pmc = native_read_pmc,
995 
996 	.iret = (void *)&hypercall_page[__HYPERVISOR_iret],
997 	.irq_enable_syscall_ret = NULL,  /* never called */
998 
999 	.load_tr_desc = paravirt_nop,
1000 	.set_ldt = xen_set_ldt,
1001 	.load_gdt = xen_load_gdt,
1002 	.load_idt = xen_load_idt,
1003 	.load_tls = xen_load_tls,
1004 
1005 	.store_gdt = native_store_gdt,
1006 	.store_idt = native_store_idt,
1007 	.store_tr = xen_store_tr,
1008 
1009 	.write_ldt_entry = xen_write_ldt_entry,
1010 	.write_gdt_entry = xen_write_gdt_entry,
1011 	.write_idt_entry = xen_write_idt_entry,
1012 	.load_sp0 = xen_load_sp0,
1013 
1014 	.set_iopl_mask = xen_set_iopl_mask,
1015 	.io_delay = xen_io_delay,
1016 
1017 	.lazy_mode = {
1018 		.enter = paravirt_enter_lazy_cpu,
1019 		.leave = xen_leave_lazy,
1020 	},
1021 };
1022 
1023 static const struct pv_irq_ops xen_irq_ops __initdata = {
1024 	.init_IRQ = xen_init_IRQ,
1025 	.save_fl = xen_save_fl,
1026 	.restore_fl = xen_restore_fl,
1027 	.irq_disable = xen_irq_disable,
1028 	.irq_enable = xen_irq_enable,
1029 	.safe_halt = xen_safe_halt,
1030 	.halt = xen_halt,
1031 };
1032 
1033 static const struct pv_apic_ops xen_apic_ops __initdata = {
1034 #ifdef CONFIG_X86_LOCAL_APIC
1035 	.apic_write = xen_apic_write,
1036 	.apic_write_atomic = xen_apic_write,
1037 	.apic_read = xen_apic_read,
1038 	.setup_boot_clock = paravirt_nop,
1039 	.setup_secondary_clock = paravirt_nop,
1040 	.startup_ipi_hook = paravirt_nop,
1041 #endif
1042 };
1043 
1044 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1045 	.pagetable_setup_start = xen_pagetable_setup_start,
1046 	.pagetable_setup_done = xen_pagetable_setup_done,
1047 
1048 	.read_cr2 = xen_read_cr2,
1049 	.write_cr2 = xen_write_cr2,
1050 
1051 	.read_cr3 = xen_read_cr3,
1052 	.write_cr3 = xen_write_cr3,
1053 
1054 	.flush_tlb_user = xen_flush_tlb,
1055 	.flush_tlb_kernel = xen_flush_tlb,
1056 	.flush_tlb_single = xen_flush_tlb_single,
1057 	.flush_tlb_others = xen_flush_tlb_others,
1058 
1059 	.pte_update = paravirt_nop,
1060 	.pte_update_defer = paravirt_nop,
1061 
1062 	.alloc_pt = xen_alloc_pt_init,
1063 	.release_pt = xen_release_pt_init,
1064 	.alloc_pd = xen_alloc_pt_init,
1065 	.alloc_pd_clone = paravirt_nop,
1066 	.release_pd = xen_release_pt_init,
1067 
1068 #ifdef CONFIG_HIGHPTE
1069 	.kmap_atomic_pte = xen_kmap_atomic_pte,
1070 #endif
1071 
1072 	.set_pte = NULL,	/* see xen_pagetable_setup_* */
1073 	.set_pte_at = xen_set_pte_at,
1074 	.set_pmd = xen_set_pmd,
1075 
1076 	.pte_val = xen_pte_val,
1077 	.pgd_val = xen_pgd_val,
1078 
1079 	.make_pte = xen_make_pte,
1080 	.make_pgd = xen_make_pgd,
1081 
1082 #ifdef CONFIG_X86_PAE
1083 	.set_pte_atomic = xen_set_pte_atomic,
1084 	.set_pte_present = xen_set_pte_at,
1085 	.set_pud = xen_set_pud,
1086 	.pte_clear = xen_pte_clear,
1087 	.pmd_clear = xen_pmd_clear,
1088 
1089 	.make_pmd = xen_make_pmd,
1090 	.pmd_val = xen_pmd_val,
1091 #endif	/* PAE */
1092 
1093 	.activate_mm = xen_activate_mm,
1094 	.dup_mmap = xen_dup_mmap,
1095 	.exit_mmap = xen_exit_mmap,
1096 
1097 	.lazy_mode = {
1098 		.enter = paravirt_enter_lazy_mmu,
1099 		.leave = xen_leave_lazy,
1100 	},
1101 };
1102 
1103 #ifdef CONFIG_SMP
1104 static const struct smp_ops xen_smp_ops __initdata = {
1105 	.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
1106 	.smp_prepare_cpus = xen_smp_prepare_cpus,
1107 	.cpu_up = xen_cpu_up,
1108 	.smp_cpus_done = xen_smp_cpus_done,
1109 
1110 	.smp_send_stop = xen_smp_send_stop,
1111 	.smp_send_reschedule = xen_smp_send_reschedule,
1112 	.smp_call_function_mask = xen_smp_call_function_mask,
1113 };
1114 #endif	/* CONFIG_SMP */
1115 
1116 static void xen_reboot(int reason)
1117 {
1118 #ifdef CONFIG_SMP
1119 	smp_send_stop();
1120 #endif
1121 
1122 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason))
1123 		BUG();
1124 }
1125 
1126 static void xen_restart(char *msg)
1127 {
1128 	xen_reboot(SHUTDOWN_reboot);
1129 }
1130 
1131 static void xen_emergency_restart(void)
1132 {
1133 	xen_reboot(SHUTDOWN_reboot);
1134 }
1135 
1136 static void xen_machine_halt(void)
1137 {
1138 	xen_reboot(SHUTDOWN_poweroff);
1139 }
1140 
1141 static void xen_crash_shutdown(struct pt_regs *regs)
1142 {
1143 	xen_reboot(SHUTDOWN_crash);
1144 }
1145 
1146 static const struct machine_ops __initdata xen_machine_ops = {
1147 	.restart = xen_restart,
1148 	.halt = xen_machine_halt,
1149 	.power_off = xen_machine_halt,
1150 	.shutdown = xen_machine_halt,
1151 	.crash_shutdown = xen_crash_shutdown,
1152 	.emergency_restart = xen_emergency_restart,
1153 };
1154 
1155 
1156 static void __init xen_reserve_top(void)
1157 {
1158 	unsigned long top = HYPERVISOR_VIRT_START;
1159 	struct xen_platform_parameters pp;
1160 
1161 	if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1162 		top = pp.virt_start;
1163 
1164 	reserve_top_address(-top + 2 * PAGE_SIZE);
1165 }
1166 
1167 /* First C function to be called on Xen boot */
1168 asmlinkage void __init xen_start_kernel(void)
1169 {
1170 	pgd_t *pgd;
1171 
1172 	if (!xen_start_info)
1173 		return;
1174 
1175 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1176 
1177 	/* Install Xen paravirt ops */
1178 	pv_info = xen_info;
1179 	pv_init_ops = xen_init_ops;
1180 	pv_time_ops = xen_time_ops;
1181 	pv_cpu_ops = xen_cpu_ops;
1182 	pv_irq_ops = xen_irq_ops;
1183 	pv_apic_ops = xen_apic_ops;
1184 	pv_mmu_ops = xen_mmu_ops;
1185 
1186 	machine_ops = xen_machine_ops;
1187 
1188 #ifdef CONFIG_SMP
1189 	smp_ops = xen_smp_ops;
1190 #endif
1191 
1192 	xen_setup_features();
1193 
1194 	/* Get mfn list */
1195 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1196 		phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
1197 
1198 	pgd = (pgd_t *)xen_start_info->pt_base;
1199 
1200 	init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1201 
1202 	init_mm.pgd = pgd; /* use the Xen pagetables to start */
1203 
1204 	/* keep using Xen gdt for now; no urgent need to change it */
1205 
1206 	x86_write_percpu(xen_cr3, __pa(pgd));
1207 	x86_write_percpu(xen_current_cr3, __pa(pgd));
1208 
1209 	/* Don't do the full vcpu_info placement stuff until we have a
1210 	   possible map and a non-dummy shared_info. */
1211 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1212 
1213 	pv_info.kernel_rpl = 1;
1214 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1215 		pv_info.kernel_rpl = 0;
1216 
1217 	/* set the limit of our address space */
1218 	xen_reserve_top();
1219 
1220 	/* set up basic CPUID stuff */
1221 	cpu_detect(&new_cpu_data);
1222 	new_cpu_data.hard_math = 1;
1223 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1224 
1225 	/* Poke various useful things into boot_params */
1226 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1227 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1228 		? __pa(xen_start_info->mod_start) : 0;
1229 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1230 
1231 	/* Start the world */
1232 	start_kernel();
1233 }
1234