xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 78c99ba1)
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/kprobes.h>
24 #include <linux/bootmem.h>
25 #include <linux/module.h>
26 #include <linux/mm.h>
27 #include <linux/page-flags.h>
28 #include <linux/highmem.h>
29 #include <linux/console.h>
30 
31 #include <xen/interface/xen.h>
32 #include <xen/interface/version.h>
33 #include <xen/interface/physdev.h>
34 #include <xen/interface/vcpu.h>
35 #include <xen/features.h>
36 #include <xen/page.h>
37 #include <xen/hvc-console.h>
38 
39 #include <asm/paravirt.h>
40 #include <asm/apic.h>
41 #include <asm/page.h>
42 #include <asm/xen/hypercall.h>
43 #include <asm/xen/hypervisor.h>
44 #include <asm/fixmap.h>
45 #include <asm/processor.h>
46 #include <asm/proto.h>
47 #include <asm/msr-index.h>
48 #include <asm/traps.h>
49 #include <asm/setup.h>
50 #include <asm/desc.h>
51 #include <asm/pgtable.h>
52 #include <asm/tlbflush.h>
53 #include <asm/reboot.h>
54 
55 #include "xen-ops.h"
56 #include "mmu.h"
57 #include "multicalls.h"
58 
59 EXPORT_SYMBOL_GPL(hypercall_page);
60 
61 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
62 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
63 
64 enum xen_domain_type xen_domain_type = XEN_NATIVE;
65 EXPORT_SYMBOL_GPL(xen_domain_type);
66 
67 struct start_info *xen_start_info;
68 EXPORT_SYMBOL_GPL(xen_start_info);
69 
70 struct shared_info xen_dummy_shared_info;
71 
72 void *xen_initial_gdt;
73 
74 /*
75  * Point at some empty memory to start with. We map the real shared_info
76  * page as soon as fixmap is up and running.
77  */
78 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
79 
80 /*
81  * Flag to determine whether vcpu info placement is available on all
82  * VCPUs.  We assume it is to start with, and then set it to zero on
83  * the first failure.  This is because it can succeed on some VCPUs
84  * and not others, since it can involve hypervisor memory allocation,
85  * or because the guest failed to guarantee all the appropriate
86  * constraints on all VCPUs (ie buffer can't cross a page boundary).
87  *
88  * Note that any particular CPU may be using a placed vcpu structure,
89  * but we can only optimise if the all are.
90  *
91  * 0: not available, 1: available
92  */
93 static int have_vcpu_info_placement = 1;
94 
95 static void xen_vcpu_setup(int cpu)
96 {
97 	struct vcpu_register_vcpu_info info;
98 	int err;
99 	struct vcpu_info *vcpup;
100 
101 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
102 	per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
103 
104 	if (!have_vcpu_info_placement)
105 		return;		/* already tested, not available */
106 
107 	vcpup = &per_cpu(xen_vcpu_info, cpu);
108 
109 	info.mfn = arbitrary_virt_to_mfn(vcpup);
110 	info.offset = offset_in_page(vcpup);
111 
112 	printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
113 	       cpu, vcpup, info.mfn, info.offset);
114 
115 	/* Check to see if the hypervisor will put the vcpu_info
116 	   structure where we want it, which allows direct access via
117 	   a percpu-variable. */
118 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
119 
120 	if (err) {
121 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
122 		have_vcpu_info_placement = 0;
123 	} else {
124 		/* This cpu is using the registered vcpu info, even if
125 		   later ones fail to. */
126 		per_cpu(xen_vcpu, cpu) = vcpup;
127 
128 		printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
129 		       cpu, vcpup);
130 	}
131 }
132 
133 /*
134  * On restore, set the vcpu placement up again.
135  * If it fails, then we're in a bad state, since
136  * we can't back out from using it...
137  */
138 void xen_vcpu_restore(void)
139 {
140 	if (have_vcpu_info_placement) {
141 		int cpu;
142 
143 		for_each_online_cpu(cpu) {
144 			bool other_cpu = (cpu != smp_processor_id());
145 
146 			if (other_cpu &&
147 			    HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
148 				BUG();
149 
150 			xen_vcpu_setup(cpu);
151 
152 			if (other_cpu &&
153 			    HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
154 				BUG();
155 		}
156 
157 		BUG_ON(!have_vcpu_info_placement);
158 	}
159 }
160 
161 static void __init xen_banner(void)
162 {
163 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
164 	struct xen_extraversion extra;
165 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
166 
167 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
168 	       pv_info.name);
169 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
170 	       version >> 16, version & 0xffff, extra.extraversion,
171 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
172 }
173 
174 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
175 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
176 
177 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
178 		      unsigned int *cx, unsigned int *dx)
179 {
180 	unsigned maskecx = ~0;
181 	unsigned maskedx = ~0;
182 
183 	/*
184 	 * Mask out inconvenient features, to try and disable as many
185 	 * unsupported kernel subsystems as possible.
186 	 */
187 	if (*ax == 1) {
188 		maskecx = cpuid_leaf1_ecx_mask;
189 		maskedx = cpuid_leaf1_edx_mask;
190 	}
191 
192 	asm(XEN_EMULATE_PREFIX "cpuid"
193 		: "=a" (*ax),
194 		  "=b" (*bx),
195 		  "=c" (*cx),
196 		  "=d" (*dx)
197 		: "0" (*ax), "2" (*cx));
198 
199 	*cx &= maskecx;
200 	*dx &= maskedx;
201 }
202 
203 static __init void xen_init_cpuid_mask(void)
204 {
205 	unsigned int ax, bx, cx, dx;
206 
207 	cpuid_leaf1_edx_mask =
208 		~((1 << X86_FEATURE_MCE)  |  /* disable MCE */
209 		  (1 << X86_FEATURE_MCA)  |  /* disable MCA */
210 		  (1 << X86_FEATURE_ACC));   /* thermal monitoring */
211 
212 	if (!xen_initial_domain())
213 		cpuid_leaf1_edx_mask &=
214 			~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
215 			  (1 << X86_FEATURE_ACPI));  /* disable ACPI */
216 
217 	ax = 1;
218 	xen_cpuid(&ax, &bx, &cx, &dx);
219 
220 	/* cpuid claims we support xsave; try enabling it to see what happens */
221 	if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
222 		unsigned long cr4;
223 
224 		set_in_cr4(X86_CR4_OSXSAVE);
225 
226 		cr4 = read_cr4();
227 
228 		if ((cr4 & X86_CR4_OSXSAVE) == 0)
229 			cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
230 
231 		clear_in_cr4(X86_CR4_OSXSAVE);
232 	}
233 }
234 
235 static void xen_set_debugreg(int reg, unsigned long val)
236 {
237 	HYPERVISOR_set_debugreg(reg, val);
238 }
239 
240 static unsigned long xen_get_debugreg(int reg)
241 {
242 	return HYPERVISOR_get_debugreg(reg);
243 }
244 
245 static void xen_end_context_switch(struct task_struct *next)
246 {
247 	xen_mc_flush();
248 	paravirt_end_context_switch(next);
249 }
250 
251 static unsigned long xen_store_tr(void)
252 {
253 	return 0;
254 }
255 
256 /*
257  * Set the page permissions for a particular virtual address.  If the
258  * address is a vmalloc mapping (or other non-linear mapping), then
259  * find the linear mapping of the page and also set its protections to
260  * match.
261  */
262 static void set_aliased_prot(void *v, pgprot_t prot)
263 {
264 	int level;
265 	pte_t *ptep;
266 	pte_t pte;
267 	unsigned long pfn;
268 	struct page *page;
269 
270 	ptep = lookup_address((unsigned long)v, &level);
271 	BUG_ON(ptep == NULL);
272 
273 	pfn = pte_pfn(*ptep);
274 	page = pfn_to_page(pfn);
275 
276 	pte = pfn_pte(pfn, prot);
277 
278 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
279 		BUG();
280 
281 	if (!PageHighMem(page)) {
282 		void *av = __va(PFN_PHYS(pfn));
283 
284 		if (av != v)
285 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
286 				BUG();
287 	} else
288 		kmap_flush_unused();
289 }
290 
291 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
292 {
293 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
294 	int i;
295 
296 	for(i = 0; i < entries; i += entries_per_page)
297 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
298 }
299 
300 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
301 {
302 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
303 	int i;
304 
305 	for(i = 0; i < entries; i += entries_per_page)
306 		set_aliased_prot(ldt + i, PAGE_KERNEL);
307 }
308 
309 static void xen_set_ldt(const void *addr, unsigned entries)
310 {
311 	struct mmuext_op *op;
312 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
313 
314 	op = mcs.args;
315 	op->cmd = MMUEXT_SET_LDT;
316 	op->arg1.linear_addr = (unsigned long)addr;
317 	op->arg2.nr_ents = entries;
318 
319 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
320 
321 	xen_mc_issue(PARAVIRT_LAZY_CPU);
322 }
323 
324 static void xen_load_gdt(const struct desc_ptr *dtr)
325 {
326 	unsigned long va = dtr->address;
327 	unsigned int size = dtr->size + 1;
328 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
329 	unsigned long frames[pages];
330 	int f;
331 
332 	/* A GDT can be up to 64k in size, which corresponds to 8192
333 	   8-byte entries, or 16 4k pages.. */
334 
335 	BUG_ON(size > 65536);
336 	BUG_ON(va & ~PAGE_MASK);
337 
338 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
339 		int level;
340 		pte_t *ptep = lookup_address(va, &level);
341 		unsigned long pfn, mfn;
342 		void *virt;
343 
344 		BUG_ON(ptep == NULL);
345 
346 		pfn = pte_pfn(*ptep);
347 		mfn = pfn_to_mfn(pfn);
348 		virt = __va(PFN_PHYS(pfn));
349 
350 		frames[f] = mfn;
351 
352 		make_lowmem_page_readonly((void *)va);
353 		make_lowmem_page_readonly(virt);
354 	}
355 
356 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
357 		BUG();
358 }
359 
360 static void load_TLS_descriptor(struct thread_struct *t,
361 				unsigned int cpu, unsigned int i)
362 {
363 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
364 	xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
365 	struct multicall_space mc = __xen_mc_entry(0);
366 
367 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
368 }
369 
370 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
371 {
372 	/*
373 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
374 	 * and lazy gs handling is enabled, it means we're in a
375 	 * context switch, and %gs has just been saved.  This means we
376 	 * can zero it out to prevent faults on exit from the
377 	 * hypervisor if the next process has no %gs.  Either way, it
378 	 * has been saved, and the new value will get loaded properly.
379 	 * This will go away as soon as Xen has been modified to not
380 	 * save/restore %gs for normal hypercalls.
381 	 *
382 	 * On x86_64, this hack is not used for %gs, because gs points
383 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
384 	 * must not zero %gs on x86_64
385 	 *
386 	 * For x86_64, we need to zero %fs, otherwise we may get an
387 	 * exception between the new %fs descriptor being loaded and
388 	 * %fs being effectively cleared at __switch_to().
389 	 */
390 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
391 #ifdef CONFIG_X86_32
392 		lazy_load_gs(0);
393 #else
394 		loadsegment(fs, 0);
395 #endif
396 	}
397 
398 	xen_mc_batch();
399 
400 	load_TLS_descriptor(t, cpu, 0);
401 	load_TLS_descriptor(t, cpu, 1);
402 	load_TLS_descriptor(t, cpu, 2);
403 
404 	xen_mc_issue(PARAVIRT_LAZY_CPU);
405 }
406 
407 #ifdef CONFIG_X86_64
408 static void xen_load_gs_index(unsigned int idx)
409 {
410 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
411 		BUG();
412 }
413 #endif
414 
415 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
416 				const void *ptr)
417 {
418 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
419 	u64 entry = *(u64 *)ptr;
420 
421 	preempt_disable();
422 
423 	xen_mc_flush();
424 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
425 		BUG();
426 
427 	preempt_enable();
428 }
429 
430 static int cvt_gate_to_trap(int vector, const gate_desc *val,
431 			    struct trap_info *info)
432 {
433 	unsigned long addr;
434 
435 	if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
436 		return 0;
437 
438 	info->vector = vector;
439 
440 	addr = gate_offset(*val);
441 #ifdef CONFIG_X86_64
442 	/*
443 	 * Look for known traps using IST, and substitute them
444 	 * appropriately.  The debugger ones are the only ones we care
445 	 * about.  Xen will handle faults like double_fault and
446 	 * machine_check, so we should never see them.  Warn if
447 	 * there's an unexpected IST-using fault handler.
448 	 */
449 	if (addr == (unsigned long)debug)
450 		addr = (unsigned long)xen_debug;
451 	else if (addr == (unsigned long)int3)
452 		addr = (unsigned long)xen_int3;
453 	else if (addr == (unsigned long)stack_segment)
454 		addr = (unsigned long)xen_stack_segment;
455 	else if (addr == (unsigned long)double_fault ||
456 		 addr == (unsigned long)nmi) {
457 		/* Don't need to handle these */
458 		return 0;
459 #ifdef CONFIG_X86_MCE
460 	} else if (addr == (unsigned long)machine_check) {
461 		return 0;
462 #endif
463 	} else {
464 		/* Some other trap using IST? */
465 		if (WARN_ON(val->ist != 0))
466 			return 0;
467 	}
468 #endif	/* CONFIG_X86_64 */
469 	info->address = addr;
470 
471 	info->cs = gate_segment(*val);
472 	info->flags = val->dpl;
473 	/* interrupt gates clear IF */
474 	if (val->type == GATE_INTERRUPT)
475 		info->flags |= 1 << 2;
476 
477 	return 1;
478 }
479 
480 /* Locations of each CPU's IDT */
481 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
482 
483 /* Set an IDT entry.  If the entry is part of the current IDT, then
484    also update Xen. */
485 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
486 {
487 	unsigned long p = (unsigned long)&dt[entrynum];
488 	unsigned long start, end;
489 
490 	preempt_disable();
491 
492 	start = __get_cpu_var(idt_desc).address;
493 	end = start + __get_cpu_var(idt_desc).size + 1;
494 
495 	xen_mc_flush();
496 
497 	native_write_idt_entry(dt, entrynum, g);
498 
499 	if (p >= start && (p + 8) <= end) {
500 		struct trap_info info[2];
501 
502 		info[1].address = 0;
503 
504 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
505 			if (HYPERVISOR_set_trap_table(info))
506 				BUG();
507 	}
508 
509 	preempt_enable();
510 }
511 
512 static void xen_convert_trap_info(const struct desc_ptr *desc,
513 				  struct trap_info *traps)
514 {
515 	unsigned in, out, count;
516 
517 	count = (desc->size+1) / sizeof(gate_desc);
518 	BUG_ON(count > 256);
519 
520 	for (in = out = 0; in < count; in++) {
521 		gate_desc *entry = (gate_desc*)(desc->address) + in;
522 
523 		if (cvt_gate_to_trap(in, entry, &traps[out]))
524 			out++;
525 	}
526 	traps[out].address = 0;
527 }
528 
529 void xen_copy_trap_info(struct trap_info *traps)
530 {
531 	const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
532 
533 	xen_convert_trap_info(desc, traps);
534 }
535 
536 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
537    hold a spinlock to protect the static traps[] array (static because
538    it avoids allocation, and saves stack space). */
539 static void xen_load_idt(const struct desc_ptr *desc)
540 {
541 	static DEFINE_SPINLOCK(lock);
542 	static struct trap_info traps[257];
543 
544 	spin_lock(&lock);
545 
546 	__get_cpu_var(idt_desc) = *desc;
547 
548 	xen_convert_trap_info(desc, traps);
549 
550 	xen_mc_flush();
551 	if (HYPERVISOR_set_trap_table(traps))
552 		BUG();
553 
554 	spin_unlock(&lock);
555 }
556 
557 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
558    they're handled differently. */
559 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
560 				const void *desc, int type)
561 {
562 	preempt_disable();
563 
564 	switch (type) {
565 	case DESC_LDT:
566 	case DESC_TSS:
567 		/* ignore */
568 		break;
569 
570 	default: {
571 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
572 
573 		xen_mc_flush();
574 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
575 			BUG();
576 	}
577 
578 	}
579 
580 	preempt_enable();
581 }
582 
583 static void xen_load_sp0(struct tss_struct *tss,
584 			 struct thread_struct *thread)
585 {
586 	struct multicall_space mcs = xen_mc_entry(0);
587 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
588 	xen_mc_issue(PARAVIRT_LAZY_CPU);
589 }
590 
591 static void xen_set_iopl_mask(unsigned mask)
592 {
593 	struct physdev_set_iopl set_iopl;
594 
595 	/* Force the change at ring 0. */
596 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
597 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
598 }
599 
600 static void xen_io_delay(void)
601 {
602 }
603 
604 #ifdef CONFIG_X86_LOCAL_APIC
605 static u32 xen_apic_read(u32 reg)
606 {
607 	return 0;
608 }
609 
610 static void xen_apic_write(u32 reg, u32 val)
611 {
612 	/* Warn to see if there's any stray references */
613 	WARN_ON(1);
614 }
615 
616 static u64 xen_apic_icr_read(void)
617 {
618 	return 0;
619 }
620 
621 static void xen_apic_icr_write(u32 low, u32 id)
622 {
623 	/* Warn to see if there's any stray references */
624 	WARN_ON(1);
625 }
626 
627 static void xen_apic_wait_icr_idle(void)
628 {
629         return;
630 }
631 
632 static u32 xen_safe_apic_wait_icr_idle(void)
633 {
634         return 0;
635 }
636 
637 static void set_xen_basic_apic_ops(void)
638 {
639 	apic->read = xen_apic_read;
640 	apic->write = xen_apic_write;
641 	apic->icr_read = xen_apic_icr_read;
642 	apic->icr_write = xen_apic_icr_write;
643 	apic->wait_icr_idle = xen_apic_wait_icr_idle;
644 	apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
645 }
646 
647 #endif
648 
649 
650 static void xen_clts(void)
651 {
652 	struct multicall_space mcs;
653 
654 	mcs = xen_mc_entry(0);
655 
656 	MULTI_fpu_taskswitch(mcs.mc, 0);
657 
658 	xen_mc_issue(PARAVIRT_LAZY_CPU);
659 }
660 
661 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
662 
663 static unsigned long xen_read_cr0(void)
664 {
665 	unsigned long cr0 = percpu_read(xen_cr0_value);
666 
667 	if (unlikely(cr0 == 0)) {
668 		cr0 = native_read_cr0();
669 		percpu_write(xen_cr0_value, cr0);
670 	}
671 
672 	return cr0;
673 }
674 
675 static void xen_write_cr0(unsigned long cr0)
676 {
677 	struct multicall_space mcs;
678 
679 	percpu_write(xen_cr0_value, cr0);
680 
681 	/* Only pay attention to cr0.TS; everything else is
682 	   ignored. */
683 	mcs = xen_mc_entry(0);
684 
685 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
686 
687 	xen_mc_issue(PARAVIRT_LAZY_CPU);
688 }
689 
690 static void xen_write_cr4(unsigned long cr4)
691 {
692 	cr4 &= ~X86_CR4_PGE;
693 	cr4 &= ~X86_CR4_PSE;
694 
695 	native_write_cr4(cr4);
696 }
697 
698 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
699 {
700 	int ret;
701 
702 	ret = 0;
703 
704 	switch (msr) {
705 #ifdef CONFIG_X86_64
706 		unsigned which;
707 		u64 base;
708 
709 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
710 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
711 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
712 
713 	set:
714 		base = ((u64)high << 32) | low;
715 		if (HYPERVISOR_set_segment_base(which, base) != 0)
716 			ret = -EFAULT;
717 		break;
718 #endif
719 
720 	case MSR_STAR:
721 	case MSR_CSTAR:
722 	case MSR_LSTAR:
723 	case MSR_SYSCALL_MASK:
724 	case MSR_IA32_SYSENTER_CS:
725 	case MSR_IA32_SYSENTER_ESP:
726 	case MSR_IA32_SYSENTER_EIP:
727 		/* Fast syscall setup is all done in hypercalls, so
728 		   these are all ignored.  Stub them out here to stop
729 		   Xen console noise. */
730 		break;
731 
732 	default:
733 		ret = native_write_msr_safe(msr, low, high);
734 	}
735 
736 	return ret;
737 }
738 
739 void xen_setup_shared_info(void)
740 {
741 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
742 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
743 			   xen_start_info->shared_info);
744 
745 		HYPERVISOR_shared_info =
746 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
747 	} else
748 		HYPERVISOR_shared_info =
749 			(struct shared_info *)__va(xen_start_info->shared_info);
750 
751 #ifndef CONFIG_SMP
752 	/* In UP this is as good a place as any to set up shared info */
753 	xen_setup_vcpu_info_placement();
754 #endif
755 
756 	xen_setup_mfn_list_list();
757 }
758 
759 /* This is called once we have the cpu_possible_map */
760 void xen_setup_vcpu_info_placement(void)
761 {
762 	int cpu;
763 
764 	for_each_possible_cpu(cpu)
765 		xen_vcpu_setup(cpu);
766 
767 	/* xen_vcpu_setup managed to place the vcpu_info within the
768 	   percpu area for all cpus, so make use of it */
769 	if (have_vcpu_info_placement) {
770 		printk(KERN_INFO "Xen: using vcpu_info placement\n");
771 
772 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
773 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
774 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
775 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
776 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
777 	}
778 }
779 
780 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
781 			  unsigned long addr, unsigned len)
782 {
783 	char *start, *end, *reloc;
784 	unsigned ret;
785 
786 	start = end = reloc = NULL;
787 
788 #define SITE(op, x)							\
789 	case PARAVIRT_PATCH(op.x):					\
790 	if (have_vcpu_info_placement) {					\
791 		start = (char *)xen_##x##_direct;			\
792 		end = xen_##x##_direct_end;				\
793 		reloc = xen_##x##_direct_reloc;				\
794 	}								\
795 	goto patch_site
796 
797 	switch (type) {
798 		SITE(pv_irq_ops, irq_enable);
799 		SITE(pv_irq_ops, irq_disable);
800 		SITE(pv_irq_ops, save_fl);
801 		SITE(pv_irq_ops, restore_fl);
802 #undef SITE
803 
804 	patch_site:
805 		if (start == NULL || (end-start) > len)
806 			goto default_patch;
807 
808 		ret = paravirt_patch_insns(insnbuf, len, start, end);
809 
810 		/* Note: because reloc is assigned from something that
811 		   appears to be an array, gcc assumes it's non-null,
812 		   but doesn't know its relationship with start and
813 		   end. */
814 		if (reloc > start && reloc < end) {
815 			int reloc_off = reloc - start;
816 			long *relocp = (long *)(insnbuf + reloc_off);
817 			long delta = start - (char *)addr;
818 
819 			*relocp += delta;
820 		}
821 		break;
822 
823 	default_patch:
824 	default:
825 		ret = paravirt_patch_default(type, clobbers, insnbuf,
826 					     addr, len);
827 		break;
828 	}
829 
830 	return ret;
831 }
832 
833 static const struct pv_info xen_info __initdata = {
834 	.paravirt_enabled = 1,
835 	.shared_kernel_pmd = 0,
836 
837 	.name = "Xen",
838 };
839 
840 static const struct pv_init_ops xen_init_ops __initdata = {
841 	.patch = xen_patch,
842 
843 	.banner = xen_banner,
844 	.memory_setup = xen_memory_setup,
845 	.arch_setup = xen_arch_setup,
846 	.post_allocator_init = xen_post_allocator_init,
847 };
848 
849 static const struct pv_time_ops xen_time_ops __initdata = {
850 	.time_init = xen_time_init,
851 
852 	.set_wallclock = xen_set_wallclock,
853 	.get_wallclock = xen_get_wallclock,
854 	.get_tsc_khz = xen_tsc_khz,
855 	.sched_clock = xen_sched_clock,
856 };
857 
858 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
859 	.cpuid = xen_cpuid,
860 
861 	.set_debugreg = xen_set_debugreg,
862 	.get_debugreg = xen_get_debugreg,
863 
864 	.clts = xen_clts,
865 
866 	.read_cr0 = xen_read_cr0,
867 	.write_cr0 = xen_write_cr0,
868 
869 	.read_cr4 = native_read_cr4,
870 	.read_cr4_safe = native_read_cr4_safe,
871 	.write_cr4 = xen_write_cr4,
872 
873 	.wbinvd = native_wbinvd,
874 
875 	.read_msr = native_read_msr_safe,
876 	.write_msr = xen_write_msr_safe,
877 	.read_tsc = native_read_tsc,
878 	.read_pmc = native_read_pmc,
879 
880 	.iret = xen_iret,
881 	.irq_enable_sysexit = xen_sysexit,
882 #ifdef CONFIG_X86_64
883 	.usergs_sysret32 = xen_sysret32,
884 	.usergs_sysret64 = xen_sysret64,
885 #endif
886 
887 	.load_tr_desc = paravirt_nop,
888 	.set_ldt = xen_set_ldt,
889 	.load_gdt = xen_load_gdt,
890 	.load_idt = xen_load_idt,
891 	.load_tls = xen_load_tls,
892 #ifdef CONFIG_X86_64
893 	.load_gs_index = xen_load_gs_index,
894 #endif
895 
896 	.alloc_ldt = xen_alloc_ldt,
897 	.free_ldt = xen_free_ldt,
898 
899 	.store_gdt = native_store_gdt,
900 	.store_idt = native_store_idt,
901 	.store_tr = xen_store_tr,
902 
903 	.write_ldt_entry = xen_write_ldt_entry,
904 	.write_gdt_entry = xen_write_gdt_entry,
905 	.write_idt_entry = xen_write_idt_entry,
906 	.load_sp0 = xen_load_sp0,
907 
908 	.set_iopl_mask = xen_set_iopl_mask,
909 	.io_delay = xen_io_delay,
910 
911 	/* Xen takes care of %gs when switching to usermode for us */
912 	.swapgs = paravirt_nop,
913 
914 	.start_context_switch = paravirt_start_context_switch,
915 	.end_context_switch = xen_end_context_switch,
916 };
917 
918 static const struct pv_apic_ops xen_apic_ops __initdata = {
919 #ifdef CONFIG_X86_LOCAL_APIC
920 	.setup_boot_clock = paravirt_nop,
921 	.setup_secondary_clock = paravirt_nop,
922 	.startup_ipi_hook = paravirt_nop,
923 #endif
924 };
925 
926 static void xen_reboot(int reason)
927 {
928 	struct sched_shutdown r = { .reason = reason };
929 
930 #ifdef CONFIG_SMP
931 	smp_send_stop();
932 #endif
933 
934 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
935 		BUG();
936 }
937 
938 static void xen_restart(char *msg)
939 {
940 	xen_reboot(SHUTDOWN_reboot);
941 }
942 
943 static void xen_emergency_restart(void)
944 {
945 	xen_reboot(SHUTDOWN_reboot);
946 }
947 
948 static void xen_machine_halt(void)
949 {
950 	xen_reboot(SHUTDOWN_poweroff);
951 }
952 
953 static void xen_crash_shutdown(struct pt_regs *regs)
954 {
955 	xen_reboot(SHUTDOWN_crash);
956 }
957 
958 static const struct machine_ops __initdata xen_machine_ops = {
959 	.restart = xen_restart,
960 	.halt = xen_machine_halt,
961 	.power_off = xen_machine_halt,
962 	.shutdown = xen_machine_halt,
963 	.crash_shutdown = xen_crash_shutdown,
964 	.emergency_restart = xen_emergency_restart,
965 };
966 
967 /* First C function to be called on Xen boot */
968 asmlinkage void __init xen_start_kernel(void)
969 {
970 	pgd_t *pgd;
971 
972 	if (!xen_start_info)
973 		return;
974 
975 	xen_domain_type = XEN_PV_DOMAIN;
976 
977 	BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
978 
979 	xen_setup_features();
980 
981 	/* Install Xen paravirt ops */
982 	pv_info = xen_info;
983 	pv_init_ops = xen_init_ops;
984 	pv_time_ops = xen_time_ops;
985 	pv_cpu_ops = xen_cpu_ops;
986 	pv_apic_ops = xen_apic_ops;
987 	pv_mmu_ops = xen_mmu_ops;
988 
989 	xen_init_irq_ops();
990 
991 	xen_init_cpuid_mask();
992 
993 #ifdef CONFIG_X86_LOCAL_APIC
994 	/*
995 	 * set up the basic apic ops.
996 	 */
997 	set_xen_basic_apic_ops();
998 #endif
999 
1000 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1001 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1002 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1003 	}
1004 
1005 	machine_ops = xen_machine_ops;
1006 
1007 #ifdef CONFIG_X86_64
1008 	/*
1009 	 * Setup percpu state.  We only need to do this for 64-bit
1010 	 * because 32-bit already has %fs set properly.
1011 	 */
1012 	load_percpu_segment(0);
1013 #endif
1014 	/*
1015 	 * The only reliable way to retain the initial address of the
1016 	 * percpu gdt_page is to remember it here, so we can go and
1017 	 * mark it RW later, when the initial percpu area is freed.
1018 	 */
1019 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1020 
1021 	xen_smp_init();
1022 
1023 	/* Get mfn list */
1024 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1025 		xen_build_dynamic_phys_to_machine();
1026 
1027 	pgd = (pgd_t *)xen_start_info->pt_base;
1028 
1029 	/* Prevent unwanted bits from being set in PTEs. */
1030 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1031 	if (!xen_initial_domain())
1032 		__supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1033 
1034 #ifdef CONFIG_X86_64
1035 	/* Work out if we support NX */
1036 	check_efer();
1037 #endif
1038 
1039 	/* Don't do the full vcpu_info placement stuff until we have a
1040 	   possible map and a non-dummy shared_info. */
1041 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1042 
1043 	local_irq_disable();
1044 	early_boot_irqs_off();
1045 
1046 	xen_raw_console_write("mapping kernel into physical memory\n");
1047 	pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1048 
1049 	init_mm.pgd = pgd;
1050 
1051 	/* keep using Xen gdt for now; no urgent need to change it */
1052 
1053 	pv_info.kernel_rpl = 1;
1054 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1055 		pv_info.kernel_rpl = 0;
1056 
1057 	/* set the limit of our address space */
1058 	xen_reserve_top();
1059 
1060 #ifdef CONFIG_X86_32
1061 	/* set up basic CPUID stuff */
1062 	cpu_detect(&new_cpu_data);
1063 	new_cpu_data.hard_math = 1;
1064 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1065 #endif
1066 
1067 	/* Poke various useful things into boot_params */
1068 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1069 	boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1070 		? __pa(xen_start_info->mod_start) : 0;
1071 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1072 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1073 
1074 	if (!xen_initial_domain()) {
1075 		add_preferred_console("xenboot", 0, NULL);
1076 		add_preferred_console("tty", 0, NULL);
1077 		add_preferred_console("hvc", 0, NULL);
1078 	}
1079 
1080 	xen_raw_console_write("about to get started...\n");
1081 
1082 	/* Start the world */
1083 #ifdef CONFIG_X86_32
1084 	i386_start_kernel();
1085 #else
1086 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1087 #endif
1088 }
1089