xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision b04b4f78)
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/version.h>
32 #include <xen/interface/physdev.h>
33 #include <xen/interface/vcpu.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/apic.h>
40 #include <asm/page.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
43 #include <asm/fixmap.h>
44 #include <asm/processor.h>
45 #include <asm/proto.h>
46 #include <asm/msr-index.h>
47 #include <asm/setup.h>
48 #include <asm/desc.h>
49 #include <asm/pgtable.h>
50 #include <asm/tlbflush.h>
51 #include <asm/reboot.h>
52 
53 #include "xen-ops.h"
54 #include "mmu.h"
55 #include "multicalls.h"
56 
57 EXPORT_SYMBOL_GPL(hypercall_page);
58 
59 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
60 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
61 
62 enum xen_domain_type xen_domain_type = XEN_NATIVE;
63 EXPORT_SYMBOL_GPL(xen_domain_type);
64 
65 struct start_info *xen_start_info;
66 EXPORT_SYMBOL_GPL(xen_start_info);
67 
68 struct shared_info xen_dummy_shared_info;
69 
70 void *xen_initial_gdt;
71 
72 /*
73  * Point at some empty memory to start with. We map the real shared_info
74  * page as soon as fixmap is up and running.
75  */
76 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
77 
78 /*
79  * Flag to determine whether vcpu info placement is available on all
80  * VCPUs.  We assume it is to start with, and then set it to zero on
81  * the first failure.  This is because it can succeed on some VCPUs
82  * and not others, since it can involve hypervisor memory allocation,
83  * or because the guest failed to guarantee all the appropriate
84  * constraints on all VCPUs (ie buffer can't cross a page boundary).
85  *
86  * Note that any particular CPU may be using a placed vcpu structure,
87  * but we can only optimise if the all are.
88  *
89  * 0: not available, 1: available
90  */
91 static int have_vcpu_info_placement = 1;
92 
93 static void xen_vcpu_setup(int cpu)
94 {
95 	struct vcpu_register_vcpu_info info;
96 	int err;
97 	struct vcpu_info *vcpup;
98 
99 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
100 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
101 
102 	if (!have_vcpu_info_placement)
103 		return;		/* already tested, not available */
104 
105 	vcpup = &per_cpu(xen_vcpu_info, cpu);
106 
107 	info.mfn = arbitrary_virt_to_mfn(vcpup);
108 	info.offset = offset_in_page(vcpup);
109 
110 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
111 	       cpu, vcpup, info.mfn, info.offset);
112 
113 	/* Check to see if the hypervisor will put the vcpu_info
114 	   structure where we want it, which allows direct access via
115 	   a percpu-variable. */
116 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
117 
118 	if (err) {
119 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
120 		have_vcpu_info_placement = 0;
121 	} else {
122 		/* This cpu is using the registered vcpu info, even if
123 		   later ones fail to. */
124 		per_cpu(xen_vcpu, cpu) = vcpup;
125 
126 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
127 		       cpu, vcpup);
128 	}
129 }
130 
131 /*
132  * On restore, set the vcpu placement up again.
133  * If it fails, then we're in a bad state, since
134  * we can't back out from using it...
135  */
136 void xen_vcpu_restore(void)
137 {
138 	if (have_vcpu_info_placement) {
139 		int cpu;
140 
141 		for_each_online_cpu(cpu) {
142 			bool other_cpu = (cpu != smp_processor_id());
143 
144 			if (other_cpu &&
145 			    HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
146 				BUG();
147 
148 			xen_vcpu_setup(cpu);
149 
150 			if (other_cpu &&
151 			    HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
152 				BUG();
153 		}
154 
155 		BUG_ON(!have_vcpu_info_placement);
156 	}
157 }
158 
159 static void __init xen_banner(void)
160 {
161 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
162 	struct xen_extraversion extra;
163 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
164 
165 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
166 	       pv_info.name);
167 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
168 	       version >> 16, version & 0xffff, extra.extraversion,
169 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
170 }
171 
172 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
173 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
174 
175 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
176 		      unsigned int *cx, unsigned int *dx)
177 {
178 	unsigned maskecx = ~0;
179 	unsigned maskedx = ~0;
180 
181 	/*
182 	 * Mask out inconvenient features, to try and disable as many
183 	 * unsupported kernel subsystems as possible.
184 	 */
185 	if (*ax == 1) {
186 		maskecx = cpuid_leaf1_ecx_mask;
187 		maskedx = cpuid_leaf1_edx_mask;
188 	}
189 
190 	asm(XEN_EMULATE_PREFIX "cpuid"
191 		: "=a" (*ax),
192 		  "=b" (*bx),
193 		  "=c" (*cx),
194 		  "=d" (*dx)
195 		: "0" (*ax), "2" (*cx));
196 
197 	*cx &= maskecx;
198 	*dx &= maskedx;
199 }
200 
201 static __init void xen_init_cpuid_mask(void)
202 {
203 	unsigned int ax, bx, cx, dx;
204 
205 	cpuid_leaf1_edx_mask =
206 		~((1 << X86_FEATURE_MCE)  |  /* disable MCE */
207 		  (1 << X86_FEATURE_MCA)  |  /* disable MCA */
208 		  (1 << X86_FEATURE_ACC));   /* thermal monitoring */
209 
210 	if (!xen_initial_domain())
211 		cpuid_leaf1_edx_mask &=
212 			~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
213 			  (1 << X86_FEATURE_ACPI));  /* disable ACPI */
214 
215 	ax = 1;
216 	xen_cpuid(&ax, &bx, &cx, &dx);
217 
218 	/* cpuid claims we support xsave; try enabling it to see what happens */
219 	if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
220 		unsigned long cr4;
221 
222 		set_in_cr4(X86_CR4_OSXSAVE);
223 
224 		cr4 = read_cr4();
225 
226 		if ((cr4 & X86_CR4_OSXSAVE) == 0)
227 			cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
228 
229 		clear_in_cr4(X86_CR4_OSXSAVE);
230 	}
231 }
232 
233 static void xen_set_debugreg(int reg, unsigned long val)
234 {
235 	HYPERVISOR_set_debugreg(reg, val);
236 }
237 
238 static unsigned long xen_get_debugreg(int reg)
239 {
240 	return HYPERVISOR_get_debugreg(reg);
241 }
242 
243 void xen_leave_lazy(void)
244 {
245 	paravirt_leave_lazy(paravirt_get_lazy_mode());
246 	xen_mc_flush();
247 }
248 
249 static unsigned long xen_store_tr(void)
250 {
251 	return 0;
252 }
253 
254 /*
255  * Set the page permissions for a particular virtual address.  If the
256  * address is a vmalloc mapping (or other non-linear mapping), then
257  * find the linear mapping of the page and also set its protections to
258  * match.
259  */
260 static void set_aliased_prot(void *v, pgprot_t prot)
261 {
262 	int level;
263 	pte_t *ptep;
264 	pte_t pte;
265 	unsigned long pfn;
266 	struct page *page;
267 
268 	ptep = lookup_address((unsigned long)v, &level);
269 	BUG_ON(ptep == NULL);
270 
271 	pfn = pte_pfn(*ptep);
272 	page = pfn_to_page(pfn);
273 
274 	pte = pfn_pte(pfn, prot);
275 
276 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
277 		BUG();
278 
279 	if (!PageHighMem(page)) {
280 		void *av = __va(PFN_PHYS(pfn));
281 
282 		if (av != v)
283 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
284 				BUG();
285 	} else
286 		kmap_flush_unused();
287 }
288 
289 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
290 {
291 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
292 	int i;
293 
294 	for(i = 0; i < entries; i += entries_per_page)
295 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
296 }
297 
298 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
299 {
300 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
301 	int i;
302 
303 	for(i = 0; i < entries; i += entries_per_page)
304 		set_aliased_prot(ldt + i, PAGE_KERNEL);
305 }
306 
307 static void xen_set_ldt(const void *addr, unsigned entries)
308 {
309 	struct mmuext_op *op;
310 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
311 
312 	op = mcs.args;
313 	op->cmd = MMUEXT_SET_LDT;
314 	op->arg1.linear_addr = (unsigned long)addr;
315 	op->arg2.nr_ents = entries;
316 
317 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
318 
319 	xen_mc_issue(PARAVIRT_LAZY_CPU);
320 }
321 
322 static void xen_load_gdt(const struct desc_ptr *dtr)
323 {
324 	unsigned long va = dtr->address;
325 	unsigned int size = dtr->size + 1;
326 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
327 	unsigned long frames[pages];
328 	int f;
329 
330 	/* A GDT can be up to 64k in size, which corresponds to 8192
331 	   8-byte entries, or 16 4k pages.. */
332 
333 	BUG_ON(size > 65536);
334 	BUG_ON(va & ~PAGE_MASK);
335 
336 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
337 		int level;
338 		pte_t *ptep = lookup_address(va, &level);
339 		unsigned long pfn, mfn;
340 		void *virt;
341 
342 		BUG_ON(ptep == NULL);
343 
344 		pfn = pte_pfn(*ptep);
345 		mfn = pfn_to_mfn(pfn);
346 		virt = __va(PFN_PHYS(pfn));
347 
348 		frames[f] = mfn;
349 
350 		make_lowmem_page_readonly((void *)va);
351 		make_lowmem_page_readonly(virt);
352 	}
353 
354 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
355 		BUG();
356 }
357 
358 static void load_TLS_descriptor(struct thread_struct *t,
359 				unsigned int cpu, unsigned int i)
360 {
361 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
362 	xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
363 	struct multicall_space mc = __xen_mc_entry(0);
364 
365 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
366 }
367 
368 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
369 {
370 	/*
371 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
372 	 * and lazy gs handling is enabled, it means we're in a
373 	 * context switch, and %gs has just been saved.  This means we
374 	 * can zero it out to prevent faults on exit from the
375 	 * hypervisor if the next process has no %gs.  Either way, it
376 	 * has been saved, and the new value will get loaded properly.
377 	 * This will go away as soon as Xen has been modified to not
378 	 * save/restore %gs for normal hypercalls.
379 	 *
380 	 * On x86_64, this hack is not used for %gs, because gs points
381 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
382 	 * must not zero %gs on x86_64
383 	 *
384 	 * For x86_64, we need to zero %fs, otherwise we may get an
385 	 * exception between the new %fs descriptor being loaded and
386 	 * %fs being effectively cleared at __switch_to().
387 	 */
388 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
389 #ifdef CONFIG_X86_32
390 		lazy_load_gs(0);
391 #else
392 		loadsegment(fs, 0);
393 #endif
394 	}
395 
396 	xen_mc_batch();
397 
398 	load_TLS_descriptor(t, cpu, 0);
399 	load_TLS_descriptor(t, cpu, 1);
400 	load_TLS_descriptor(t, cpu, 2);
401 
402 	xen_mc_issue(PARAVIRT_LAZY_CPU);
403 }
404 
405 #ifdef CONFIG_X86_64
406 static void xen_load_gs_index(unsigned int idx)
407 {
408 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
409 		BUG();
410 }
411 #endif
412 
413 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
414 				const void *ptr)
415 {
416 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
417 	u64 entry = *(u64 *)ptr;
418 
419 	preempt_disable();
420 
421 	xen_mc_flush();
422 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
423 		BUG();
424 
425 	preempt_enable();
426 }
427 
428 static int cvt_gate_to_trap(int vector, const gate_desc *val,
429 			    struct trap_info *info)
430 {
431 	if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
432 		return 0;
433 
434 	info->vector = vector;
435 	info->address = gate_offset(*val);
436 	info->cs = gate_segment(*val);
437 	info->flags = val->dpl;
438 	/* interrupt gates clear IF */
439 	if (val->type == GATE_INTERRUPT)
440 		info->flags |= 1 << 2;
441 
442 	return 1;
443 }
444 
445 /* Locations of each CPU's IDT */
446 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
447 
448 /* Set an IDT entry.  If the entry is part of the current IDT, then
449    also update Xen. */
450 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
451 {
452 	unsigned long p = (unsigned long)&dt[entrynum];
453 	unsigned long start, end;
454 
455 	preempt_disable();
456 
457 	start = __get_cpu_var(idt_desc).address;
458 	end = start + __get_cpu_var(idt_desc).size + 1;
459 
460 	xen_mc_flush();
461 
462 	native_write_idt_entry(dt, entrynum, g);
463 
464 	if (p >= start && (p + 8) <= end) {
465 		struct trap_info info[2];
466 
467 		info[1].address = 0;
468 
469 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
470 			if (HYPERVISOR_set_trap_table(info))
471 				BUG();
472 	}
473 
474 	preempt_enable();
475 }
476 
477 static void xen_convert_trap_info(const struct desc_ptr *desc,
478 				  struct trap_info *traps)
479 {
480 	unsigned in, out, count;
481 
482 	count = (desc->size+1) / sizeof(gate_desc);
483 	BUG_ON(count > 256);
484 
485 	for (in = out = 0; in < count; in++) {
486 		gate_desc *entry = (gate_desc*)(desc->address) + in;
487 
488 		if (cvt_gate_to_trap(in, entry, &traps[out]))
489 			out++;
490 	}
491 	traps[out].address = 0;
492 }
493 
494 void xen_copy_trap_info(struct trap_info *traps)
495 {
496 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
497 
498 	xen_convert_trap_info(desc, traps);
499 }
500 
501 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
502    hold a spinlock to protect the static traps[] array (static because
503    it avoids allocation, and saves stack space). */
504 static void xen_load_idt(const struct desc_ptr *desc)
505 {
506 	static DEFINE_SPINLOCK(lock);
507 	static struct trap_info traps[257];
508 
509 	spin_lock(&lock);
510 
511 	__get_cpu_var(idt_desc) = *desc;
512 
513 	xen_convert_trap_info(desc, traps);
514 
515 	xen_mc_flush();
516 	if (HYPERVISOR_set_trap_table(traps))
517 		BUG();
518 
519 	spin_unlock(&lock);
520 }
521 
522 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
523    they're handled differently. */
524 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
525 				const void *desc, int type)
526 {
527 	preempt_disable();
528 
529 	switch (type) {
530 	case DESC_LDT:
531 	case DESC_TSS:
532 		/* ignore */
533 		break;
534 
535 	default: {
536 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
537 
538 		xen_mc_flush();
539 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
540 			BUG();
541 	}
542 
543 	}
544 
545 	preempt_enable();
546 }
547 
548 static void xen_load_sp0(struct tss_struct *tss,
549 			 struct thread_struct *thread)
550 {
551 	struct multicall_space mcs = xen_mc_entry(0);
552 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
553 	xen_mc_issue(PARAVIRT_LAZY_CPU);
554 }
555 
556 static void xen_set_iopl_mask(unsigned mask)
557 {
558 	struct physdev_set_iopl set_iopl;
559 
560 	/* Force the change at ring 0. */
561 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
562 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
563 }
564 
565 static void xen_io_delay(void)
566 {
567 }
568 
569 #ifdef CONFIG_X86_LOCAL_APIC
570 static u32 xen_apic_read(u32 reg)
571 {
572 	return 0;
573 }
574 
575 static void xen_apic_write(u32 reg, u32 val)
576 {
577 	/* Warn to see if there's any stray references */
578 	WARN_ON(1);
579 }
580 
581 static u64 xen_apic_icr_read(void)
582 {
583 	return 0;
584 }
585 
586 static void xen_apic_icr_write(u32 low, u32 id)
587 {
588 	/* Warn to see if there's any stray references */
589 	WARN_ON(1);
590 }
591 
592 static void xen_apic_wait_icr_idle(void)
593 {
594         return;
595 }
596 
597 static u32 xen_safe_apic_wait_icr_idle(void)
598 {
599         return 0;
600 }
601 
602 static void set_xen_basic_apic_ops(void)
603 {
604 	apic->read = xen_apic_read;
605 	apic->write = xen_apic_write;
606 	apic->icr_read = xen_apic_icr_read;
607 	apic->icr_write = xen_apic_icr_write;
608 	apic->wait_icr_idle = xen_apic_wait_icr_idle;
609 	apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
610 }
611 
612 #endif
613 
614 
615 static void xen_clts(void)
616 {
617 	struct multicall_space mcs;
618 
619 	mcs = xen_mc_entry(0);
620 
621 	MULTI_fpu_taskswitch(mcs.mc, 0);
622 
623 	xen_mc_issue(PARAVIRT_LAZY_CPU);
624 }
625 
626 static void xen_write_cr0(unsigned long cr0)
627 {
628 	struct multicall_space mcs;
629 
630 	/* Only pay attention to cr0.TS; everything else is
631 	   ignored. */
632 	mcs = xen_mc_entry(0);
633 
634 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
635 
636 	xen_mc_issue(PARAVIRT_LAZY_CPU);
637 }
638 
639 static void xen_write_cr4(unsigned long cr4)
640 {
641 	cr4 &= ~X86_CR4_PGE;
642 	cr4 &= ~X86_CR4_PSE;
643 
644 	native_write_cr4(cr4);
645 }
646 
647 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
648 {
649 	int ret;
650 
651 	ret = 0;
652 
653 	switch (msr) {
654 #ifdef CONFIG_X86_64
655 		unsigned which;
656 		u64 base;
657 
658 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
659 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
660 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
661 
662 	set:
663 		base = ((u64)high << 32) | low;
664 		if (HYPERVISOR_set_segment_base(which, base) != 0)
665 			ret = -EFAULT;
666 		break;
667 #endif
668 
669 	case MSR_STAR:
670 	case MSR_CSTAR:
671 	case MSR_LSTAR:
672 	case MSR_SYSCALL_MASK:
673 	case MSR_IA32_SYSENTER_CS:
674 	case MSR_IA32_SYSENTER_ESP:
675 	case MSR_IA32_SYSENTER_EIP:
676 		/* Fast syscall setup is all done in hypercalls, so
677 		   these are all ignored.  Stub them out here to stop
678 		   Xen console noise. */
679 		break;
680 
681 	default:
682 		ret = native_write_msr_safe(msr, low, high);
683 	}
684 
685 	return ret;
686 }
687 
688 void xen_setup_shared_info(void)
689 {
690 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
691 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
692 			   xen_start_info->shared_info);
693 
694 		HYPERVISOR_shared_info =
695 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
696 	} else
697 		HYPERVISOR_shared_info =
698 			(struct shared_info *)__va(xen_start_info->shared_info);
699 
700 #ifndef CONFIG_SMP
701 	/* In UP this is as good a place as any to set up shared info */
702 	xen_setup_vcpu_info_placement();
703 #endif
704 
705 	xen_setup_mfn_list_list();
706 }
707 
708 /* This is called once we have the cpu_possible_map */
709 void xen_setup_vcpu_info_placement(void)
710 {
711 	int cpu;
712 
713 	for_each_possible_cpu(cpu)
714 		xen_vcpu_setup(cpu);
715 
716 	/* xen_vcpu_setup managed to place the vcpu_info within the
717 	   percpu area for all cpus, so make use of it */
718 	if (have_vcpu_info_placement) {
719 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
720 
721 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
722 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
723 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
724 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
725 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
726 	}
727 }
728 
729 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
730 			  unsigned long addr, unsigned len)
731 {
732 	char *start, *end, *reloc;
733 	unsigned ret;
734 
735 	start = end = reloc = NULL;
736 
737 #define SITE(op, x)							\
738 	case PARAVIRT_PATCH(op.x):					\
739 	if (have_vcpu_info_placement) {					\
740 		start = (char *)xen_##x##_direct;			\
741 		end = xen_##x##_direct_end;				\
742 		reloc = xen_##x##_direct_reloc;				\
743 	}								\
744 	goto patch_site
745 
746 	switch (type) {
747 		SITE(pv_irq_ops, irq_enable);
748 		SITE(pv_irq_ops, irq_disable);
749 		SITE(pv_irq_ops, save_fl);
750 		SITE(pv_irq_ops, restore_fl);
751 #undef SITE
752 
753 	patch_site:
754 		if (start == NULL || (end-start) > len)
755 			goto default_patch;
756 
757 		ret = paravirt_patch_insns(insnbuf, len, start, end);
758 
759 		/* Note: because reloc is assigned from something that
760 		   appears to be an array, gcc assumes it's non-null,
761 		   but doesn't know its relationship with start and
762 		   end. */
763 		if (reloc > start && reloc < end) {
764 			int reloc_off = reloc - start;
765 			long *relocp = (long *)(insnbuf + reloc_off);
766 			long delta = start - (char *)addr;
767 
768 			*relocp += delta;
769 		}
770 		break;
771 
772 	default_patch:
773 	default:
774 		ret = paravirt_patch_default(type, clobbers, insnbuf,
775 					     addr, len);
776 		break;
777 	}
778 
779 	return ret;
780 }
781 
782 static const struct pv_info xen_info __initdata = {
783 	.paravirt_enabled = 1,
784 	.shared_kernel_pmd = 0,
785 
786 	.name = "Xen",
787 };
788 
789 static const struct pv_init_ops xen_init_ops __initdata = {
790 	.patch = xen_patch,
791 
792 	.banner = xen_banner,
793 	.memory_setup = xen_memory_setup,
794 	.arch_setup = xen_arch_setup,
795 	.post_allocator_init = xen_post_allocator_init,
796 };
797 
798 static const struct pv_time_ops xen_time_ops __initdata = {
799 	.time_init = xen_time_init,
800 
801 	.set_wallclock = xen_set_wallclock,
802 	.get_wallclock = xen_get_wallclock,
803 	.get_tsc_khz = xen_tsc_khz,
804 	.sched_clock = xen_sched_clock,
805 };
806 
807 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
808 	.cpuid = xen_cpuid,
809 
810 	.set_debugreg = xen_set_debugreg,
811 	.get_debugreg = xen_get_debugreg,
812 
813 	.clts = xen_clts,
814 
815 	.read_cr0 = native_read_cr0,
816 	.write_cr0 = xen_write_cr0,
817 
818 	.read_cr4 = native_read_cr4,
819 	.read_cr4_safe = native_read_cr4_safe,
820 	.write_cr4 = xen_write_cr4,
821 
822 	.wbinvd = native_wbinvd,
823 
824 	.read_msr = native_read_msr_safe,
825 	.write_msr = xen_write_msr_safe,
826 	.read_tsc = native_read_tsc,
827 	.read_pmc = native_read_pmc,
828 
829 	.iret = xen_iret,
830 	.irq_enable_sysexit = xen_sysexit,
831 #ifdef CONFIG_X86_64
832 	.usergs_sysret32 = xen_sysret32,
833 	.usergs_sysret64 = xen_sysret64,
834 #endif
835 
836 	.load_tr_desc = paravirt_nop,
837 	.set_ldt = xen_set_ldt,
838 	.load_gdt = xen_load_gdt,
839 	.load_idt = xen_load_idt,
840 	.load_tls = xen_load_tls,
841 #ifdef CONFIG_X86_64
842 	.load_gs_index = xen_load_gs_index,
843 #endif
844 
845 	.alloc_ldt = xen_alloc_ldt,
846 	.free_ldt = xen_free_ldt,
847 
848 	.store_gdt = native_store_gdt,
849 	.store_idt = native_store_idt,
850 	.store_tr = xen_store_tr,
851 
852 	.write_ldt_entry = xen_write_ldt_entry,
853 	.write_gdt_entry = xen_write_gdt_entry,
854 	.write_idt_entry = xen_write_idt_entry,
855 	.load_sp0 = xen_load_sp0,
856 
857 	.set_iopl_mask = xen_set_iopl_mask,
858 	.io_delay = xen_io_delay,
859 
860 	/* Xen takes care of %gs when switching to usermode for us */
861 	.swapgs = paravirt_nop,
862 
863 	.lazy_mode = {
864 		.enter = paravirt_enter_lazy_cpu,
865 		.leave = xen_leave_lazy,
866 	},
867 };
868 
869 static const struct pv_apic_ops xen_apic_ops __initdata = {
870 #ifdef CONFIG_X86_LOCAL_APIC
871 	.setup_boot_clock = paravirt_nop,
872 	.setup_secondary_clock = paravirt_nop,
873 	.startup_ipi_hook = paravirt_nop,
874 #endif
875 };
876 
877 static void xen_reboot(int reason)
878 {
879 	struct sched_shutdown r = { .reason = reason };
880 
881 #ifdef CONFIG_SMP
882 	smp_send_stop();
883 #endif
884 
885 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
886 		BUG();
887 }
888 
889 static void xen_restart(char *msg)
890 {
891 	xen_reboot(SHUTDOWN_reboot);
892 }
893 
894 static void xen_emergency_restart(void)
895 {
896 	xen_reboot(SHUTDOWN_reboot);
897 }
898 
899 static void xen_machine_halt(void)
900 {
901 	xen_reboot(SHUTDOWN_poweroff);
902 }
903 
904 static void xen_crash_shutdown(struct pt_regs *regs)
905 {
906 	xen_reboot(SHUTDOWN_crash);
907 }
908 
909 static const struct machine_ops __initdata xen_machine_ops = {
910 	.restart = xen_restart,
911 	.halt = xen_machine_halt,
912 	.power_off = xen_machine_halt,
913 	.shutdown = xen_machine_halt,
914 	.crash_shutdown = xen_crash_shutdown,
915 	.emergency_restart = xen_emergency_restart,
916 };
917 
918 /* First C function to be called on Xen boot */
919 asmlinkage void __init xen_start_kernel(void)
920 {
921 	pgd_t *pgd;
922 
923 	if (!xen_start_info)
924 		return;
925 
926 	xen_domain_type = XEN_PV_DOMAIN;
927 
928 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
929 
930 	xen_setup_features();
931 
932 	/* Install Xen paravirt ops */
933 	pv_info = xen_info;
934 	pv_init_ops = xen_init_ops;
935 	pv_time_ops = xen_time_ops;
936 	pv_cpu_ops = xen_cpu_ops;
937 	pv_apic_ops = xen_apic_ops;
938 	pv_mmu_ops = xen_mmu_ops;
939 
940 	xen_init_irq_ops();
941 
942 	xen_init_cpuid_mask();
943 
944 #ifdef CONFIG_X86_LOCAL_APIC
945 	/*
946 	 * set up the basic apic ops.
947 	 */
948 	set_xen_basic_apic_ops();
949 #endif
950 
951 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
952 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
953 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
954 	}
955 
956 	machine_ops = xen_machine_ops;
957 
958 #ifdef CONFIG_X86_64
959 	/*
960 	 * Setup percpu state.  We only need to do this for 64-bit
961 	 * because 32-bit already has %fs set properly.
962 	 */
963 	load_percpu_segment(0);
964 #endif
965 	/*
966 	 * The only reliable way to retain the initial address of the
967 	 * percpu gdt_page is to remember it here, so we can go and
968 	 * mark it RW later, when the initial percpu area is freed.
969 	 */
970 	xen_initial_gdt = &per_cpu(gdt_page, 0);
971 
972 	xen_smp_init();
973 
974 	/* Get mfn list */
975 	if (!xen_feature(XENFEAT_auto_translated_physmap))
976 		xen_build_dynamic_phys_to_machine();
977 
978 	pgd = (pgd_t *)xen_start_info->pt_base;
979 
980 	/* Prevent unwanted bits from being set in PTEs. */
981 	__supported_pte_mask &= ~_PAGE_GLOBAL;
982 	if (!xen_initial_domain())
983 		__supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
984 
985 #ifdef CONFIG_X86_64
986 	/* Work out if we support NX */
987 	check_efer();
988 #endif
989 
990 	/* Don't do the full vcpu_info placement stuff until we have a
991 	   possible map and a non-dummy shared_info. */
992 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
993 
994 	local_irq_disable();
995 	early_boot_irqs_off();
996 
997 	xen_raw_console_write("mapping kernel into physical memory\n");
998 	pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
999 
1000 	init_mm.pgd = pgd;
1001 
1002 	/* keep using Xen gdt for now; no urgent need to change it */
1003 
1004 	pv_info.kernel_rpl = 1;
1005 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1006 		pv_info.kernel_rpl = 0;
1007 
1008 	/* set the limit of our address space */
1009 	xen_reserve_top();
1010 
1011 #ifdef CONFIG_X86_32
1012 	/* set up basic CPUID stuff */
1013 	cpu_detect(&new_cpu_data);
1014 	new_cpu_data.hard_math = 1;
1015 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1016 #endif
1017 
1018 	/* Poke various useful things into boot_params */
1019 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1020 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1021 		? __pa(xen_start_info->mod_start) : 0;
1022 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1023 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1024 
1025 	if (!xen_initial_domain()) {
1026 		add_preferred_console("xenboot", 0, NULL);
1027 		add_preferred_console("tty", 0, NULL);
1028 		add_preferred_console("hvc", 0, NULL);
1029 	}
1030 
1031 	xen_raw_console_write("about to get started...\n");
1032 
1033 	/* Start the world */
1034 #ifdef CONFIG_X86_32
1035 	i386_start_kernel();
1036 #else
1037 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1038 #endif
1039 }
1040