xref: /linux/arch/x86/xen/enlighten.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
27 #include <linux/mm.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
35 
36 #ifdef CONFIG_KEXEC_CORE
37 #include <linux/kexec.h>
38 #endif
39 
40 #include <xen/xen.h>
41 #include <xen/events.h>
42 #include <xen/interface/xen.h>
43 #include <xen/interface/version.h>
44 #include <xen/interface/physdev.h>
45 #include <xen/interface/vcpu.h>
46 #include <xen/interface/memory.h>
47 #include <xen/interface/nmi.h>
48 #include <xen/interface/xen-mca.h>
49 #include <xen/features.h>
50 #include <xen/page.h>
51 #include <xen/hvm.h>
52 #include <xen/hvc-console.h>
53 #include <xen/acpi.h>
54 
55 #include <asm/paravirt.h>
56 #include <asm/apic.h>
57 #include <asm/page.h>
58 #include <asm/xen/pci.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
61 #include <asm/fixmap.h>
62 #include <asm/processor.h>
63 #include <asm/proto.h>
64 #include <asm/msr-index.h>
65 #include <asm/traps.h>
66 #include <asm/setup.h>
67 #include <asm/desc.h>
68 #include <asm/pgalloc.h>
69 #include <asm/pgtable.h>
70 #include <asm/tlbflush.h>
71 #include <asm/reboot.h>
72 #include <asm/stackprotector.h>
73 #include <asm/hypervisor.h>
74 #include <asm/mach_traps.h>
75 #include <asm/mwait.h>
76 #include <asm/pci_x86.h>
77 #include <asm/pat.h>
78 
79 #ifdef CONFIG_ACPI
80 #include <linux/acpi.h>
81 #include <asm/acpi.h>
82 #include <acpi/pdc_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
85 #endif
86 
87 #include "xen-ops.h"
88 #include "mmu.h"
89 #include "smp.h"
90 #include "multicalls.h"
91 #include "pmu.h"
92 
93 EXPORT_SYMBOL_GPL(hypercall_page);
94 
95 /*
96  * Pointer to the xen_vcpu_info structure or
97  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
98  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
99  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
100  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
101  * acknowledge pending events.
102  * Also more subtly it is used by the patched version of irq enable/disable
103  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
104  *
105  * The desire to be able to do those mask/unmask operations as a single
106  * instruction by using the per-cpu offset held in %gs is the real reason
107  * vcpu info is in a per-cpu pointer and the original reason for this
108  * hypercall.
109  *
110  */
111 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
112 
113 /*
114  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
115  * hypercall. This can be used both in PV and PVHVM mode. The structure
116  * overrides the default per_cpu(xen_vcpu, cpu) value.
117  */
118 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
119 
120 enum xen_domain_type xen_domain_type = XEN_NATIVE;
121 EXPORT_SYMBOL_GPL(xen_domain_type);
122 
123 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
124 EXPORT_SYMBOL(machine_to_phys_mapping);
125 unsigned long  machine_to_phys_nr;
126 EXPORT_SYMBOL(machine_to_phys_nr);
127 
128 struct start_info *xen_start_info;
129 EXPORT_SYMBOL_GPL(xen_start_info);
130 
131 struct shared_info xen_dummy_shared_info;
132 
133 void *xen_initial_gdt;
134 
135 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
136 __read_mostly int xen_have_vector_callback;
137 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
138 
139 /*
140  * Point at some empty memory to start with. We map the real shared_info
141  * page as soon as fixmap is up and running.
142  */
143 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
144 
145 /*
146  * Flag to determine whether vcpu info placement is available on all
147  * VCPUs.  We assume it is to start with, and then set it to zero on
148  * the first failure.  This is because it can succeed on some VCPUs
149  * and not others, since it can involve hypervisor memory allocation,
150  * or because the guest failed to guarantee all the appropriate
151  * constraints on all VCPUs (ie buffer can't cross a page boundary).
152  *
153  * Note that any particular CPU may be using a placed vcpu structure,
154  * but we can only optimise if the all are.
155  *
156  * 0: not available, 1: available
157  */
158 static int have_vcpu_info_placement = 1;
159 
160 struct tls_descs {
161 	struct desc_struct desc[3];
162 };
163 
164 /*
165  * Updating the 3 TLS descriptors in the GDT on every task switch is
166  * surprisingly expensive so we avoid updating them if they haven't
167  * changed.  Since Xen writes different descriptors than the one
168  * passed in the update_descriptor hypercall we keep shadow copies to
169  * compare against.
170  */
171 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
172 
173 static void clamp_max_cpus(void)
174 {
175 #ifdef CONFIG_SMP
176 	if (setup_max_cpus > MAX_VIRT_CPUS)
177 		setup_max_cpus = MAX_VIRT_CPUS;
178 #endif
179 }
180 
181 static void xen_vcpu_setup(int cpu)
182 {
183 	struct vcpu_register_vcpu_info info;
184 	int err;
185 	struct vcpu_info *vcpup;
186 
187 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
188 
189 	/*
190 	 * This path is called twice on PVHVM - first during bootup via
191 	 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
192 	 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
193 	 * As we can only do the VCPUOP_register_vcpu_info once lets
194 	 * not over-write its result.
195 	 *
196 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
197 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
198 	 * use this function.
199 	 */
200 	if (xen_hvm_domain()) {
201 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
202 			return;
203 	}
204 	if (cpu < MAX_VIRT_CPUS)
205 		per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
206 
207 	if (!have_vcpu_info_placement) {
208 		if (cpu >= MAX_VIRT_CPUS)
209 			clamp_max_cpus();
210 		return;
211 	}
212 
213 	vcpup = &per_cpu(xen_vcpu_info, cpu);
214 	info.mfn = arbitrary_virt_to_mfn(vcpup);
215 	info.offset = offset_in_page(vcpup);
216 
217 	/* Check to see if the hypervisor will put the vcpu_info
218 	   structure where we want it, which allows direct access via
219 	   a percpu-variable.
220 	   N.B. This hypercall can _only_ be called once per CPU. Subsequent
221 	   calls will error out with -EINVAL. This is due to the fact that
222 	   hypervisor has no unregister variant and this hypercall does not
223 	   allow to over-write info.mfn and info.offset.
224 	 */
225 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
226 
227 	if (err) {
228 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
229 		have_vcpu_info_placement = 0;
230 		clamp_max_cpus();
231 	} else {
232 		/* This cpu is using the registered vcpu info, even if
233 		   later ones fail to. */
234 		per_cpu(xen_vcpu, cpu) = vcpup;
235 	}
236 }
237 
238 /*
239  * On restore, set the vcpu placement up again.
240  * If it fails, then we're in a bad state, since
241  * we can't back out from using it...
242  */
243 void xen_vcpu_restore(void)
244 {
245 	int cpu;
246 
247 	for_each_possible_cpu(cpu) {
248 		bool other_cpu = (cpu != smp_processor_id());
249 		bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
250 
251 		if (other_cpu && is_up &&
252 		    HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
253 			BUG();
254 
255 		xen_setup_runstate_info(cpu);
256 
257 		if (have_vcpu_info_placement)
258 			xen_vcpu_setup(cpu);
259 
260 		if (other_cpu && is_up &&
261 		    HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
262 			BUG();
263 	}
264 }
265 
266 static void __init xen_banner(void)
267 {
268 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
269 	struct xen_extraversion extra;
270 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
271 
272 	pr_info("Booting paravirtualized kernel %son %s\n",
273 		xen_feature(XENFEAT_auto_translated_physmap) ?
274 			"with PVH extensions " : "", pv_info.name);
275 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
276 	       version >> 16, version & 0xffff, extra.extraversion,
277 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
278 }
279 /* Check if running on Xen version (major, minor) or later */
280 bool
281 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
282 {
283 	unsigned int version;
284 
285 	if (!xen_domain())
286 		return false;
287 
288 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
289 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
290 		((version >> 16) > major))
291 		return true;
292 	return false;
293 }
294 
295 #define CPUID_THERM_POWER_LEAF 6
296 #define APERFMPERF_PRESENT 0
297 
298 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
299 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
300 
301 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
302 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
303 static __read_mostly unsigned int cpuid_leaf5_edx_val;
304 
305 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
306 		      unsigned int *cx, unsigned int *dx)
307 {
308 	unsigned maskebx = ~0;
309 	unsigned maskecx = ~0;
310 	unsigned maskedx = ~0;
311 	unsigned setecx = 0;
312 	/*
313 	 * Mask out inconvenient features, to try and disable as many
314 	 * unsupported kernel subsystems as possible.
315 	 */
316 	switch (*ax) {
317 	case 1:
318 		maskecx = cpuid_leaf1_ecx_mask;
319 		setecx = cpuid_leaf1_ecx_set_mask;
320 		maskedx = cpuid_leaf1_edx_mask;
321 		break;
322 
323 	case CPUID_MWAIT_LEAF:
324 		/* Synthesize the values.. */
325 		*ax = 0;
326 		*bx = 0;
327 		*cx = cpuid_leaf5_ecx_val;
328 		*dx = cpuid_leaf5_edx_val;
329 		return;
330 
331 	case CPUID_THERM_POWER_LEAF:
332 		/* Disabling APERFMPERF for kernel usage */
333 		maskecx = ~(1 << APERFMPERF_PRESENT);
334 		break;
335 
336 	case 0xb:
337 		/* Suppress extended topology stuff */
338 		maskebx = 0;
339 		break;
340 	}
341 
342 	asm(XEN_EMULATE_PREFIX "cpuid"
343 		: "=a" (*ax),
344 		  "=b" (*bx),
345 		  "=c" (*cx),
346 		  "=d" (*dx)
347 		: "0" (*ax), "2" (*cx));
348 
349 	*bx &= maskebx;
350 	*cx &= maskecx;
351 	*cx |= setecx;
352 	*dx &= maskedx;
353 
354 }
355 
356 static bool __init xen_check_mwait(void)
357 {
358 #ifdef CONFIG_ACPI
359 	struct xen_platform_op op = {
360 		.cmd			= XENPF_set_processor_pminfo,
361 		.u.set_pminfo.id	= -1,
362 		.u.set_pminfo.type	= XEN_PM_PDC,
363 	};
364 	uint32_t buf[3];
365 	unsigned int ax, bx, cx, dx;
366 	unsigned int mwait_mask;
367 
368 	/* We need to determine whether it is OK to expose the MWAIT
369 	 * capability to the kernel to harvest deeper than C3 states from ACPI
370 	 * _CST using the processor_harvest_xen.c module. For this to work, we
371 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
372 	 * checks against). The hypervisor won't expose the MWAIT flag because
373 	 * it would break backwards compatibility; so we will find out directly
374 	 * from the hardware and hypercall.
375 	 */
376 	if (!xen_initial_domain())
377 		return false;
378 
379 	/*
380 	 * When running under platform earlier than Xen4.2, do not expose
381 	 * mwait, to avoid the risk of loading native acpi pad driver
382 	 */
383 	if (!xen_running_on_version_or_later(4, 2))
384 		return false;
385 
386 	ax = 1;
387 	cx = 0;
388 
389 	native_cpuid(&ax, &bx, &cx, &dx);
390 
391 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
392 		     (1 << (X86_FEATURE_MWAIT % 32));
393 
394 	if ((cx & mwait_mask) != mwait_mask)
395 		return false;
396 
397 	/* We need to emulate the MWAIT_LEAF and for that we need both
398 	 * ecx and edx. The hypercall provides only partial information.
399 	 */
400 
401 	ax = CPUID_MWAIT_LEAF;
402 	bx = 0;
403 	cx = 0;
404 	dx = 0;
405 
406 	native_cpuid(&ax, &bx, &cx, &dx);
407 
408 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
409 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
410 	 */
411 	buf[0] = ACPI_PDC_REVISION_ID;
412 	buf[1] = 1;
413 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
414 
415 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
416 
417 	if ((HYPERVISOR_dom0_op(&op) == 0) &&
418 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
419 		cpuid_leaf5_ecx_val = cx;
420 		cpuid_leaf5_edx_val = dx;
421 	}
422 	return true;
423 #else
424 	return false;
425 #endif
426 }
427 static void __init xen_init_cpuid_mask(void)
428 {
429 	unsigned int ax, bx, cx, dx;
430 	unsigned int xsave_mask;
431 
432 	cpuid_leaf1_edx_mask =
433 		~((1 << X86_FEATURE_MTRR) |  /* disable MTRR */
434 		  (1 << X86_FEATURE_ACC));   /* thermal monitoring */
435 
436 	if (!xen_initial_domain())
437 		cpuid_leaf1_edx_mask &=
438 			~((1 << X86_FEATURE_ACPI));  /* disable ACPI */
439 
440 	cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
441 
442 	ax = 1;
443 	cx = 0;
444 	cpuid(1, &ax, &bx, &cx, &dx);
445 
446 	xsave_mask =
447 		(1 << (X86_FEATURE_XSAVE % 32)) |
448 		(1 << (X86_FEATURE_OSXSAVE % 32));
449 
450 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
451 	if ((cx & xsave_mask) != xsave_mask)
452 		cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
453 	if (xen_check_mwait())
454 		cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
455 }
456 
457 static void xen_set_debugreg(int reg, unsigned long val)
458 {
459 	HYPERVISOR_set_debugreg(reg, val);
460 }
461 
462 static unsigned long xen_get_debugreg(int reg)
463 {
464 	return HYPERVISOR_get_debugreg(reg);
465 }
466 
467 static void xen_end_context_switch(struct task_struct *next)
468 {
469 	xen_mc_flush();
470 	paravirt_end_context_switch(next);
471 }
472 
473 static unsigned long xen_store_tr(void)
474 {
475 	return 0;
476 }
477 
478 /*
479  * Set the page permissions for a particular virtual address.  If the
480  * address is a vmalloc mapping (or other non-linear mapping), then
481  * find the linear mapping of the page and also set its protections to
482  * match.
483  */
484 static void set_aliased_prot(void *v, pgprot_t prot)
485 {
486 	int level;
487 	pte_t *ptep;
488 	pte_t pte;
489 	unsigned long pfn;
490 	struct page *page;
491 	unsigned char dummy;
492 
493 	ptep = lookup_address((unsigned long)v, &level);
494 	BUG_ON(ptep == NULL);
495 
496 	pfn = pte_pfn(*ptep);
497 	page = pfn_to_page(pfn);
498 
499 	pte = pfn_pte(pfn, prot);
500 
501 	/*
502 	 * Careful: update_va_mapping() will fail if the virtual address
503 	 * we're poking isn't populated in the page tables.  We don't
504 	 * need to worry about the direct map (that's always in the page
505 	 * tables), but we need to be careful about vmap space.  In
506 	 * particular, the top level page table can lazily propagate
507 	 * entries between processes, so if we've switched mms since we
508 	 * vmapped the target in the first place, we might not have the
509 	 * top-level page table entry populated.
510 	 *
511 	 * We disable preemption because we want the same mm active when
512 	 * we probe the target and when we issue the hypercall.  We'll
513 	 * have the same nominal mm, but if we're a kernel thread, lazy
514 	 * mm dropping could change our pgd.
515 	 *
516 	 * Out of an abundance of caution, this uses __get_user() to fault
517 	 * in the target address just in case there's some obscure case
518 	 * in which the target address isn't readable.
519 	 */
520 
521 	preempt_disable();
522 
523 	pagefault_disable();	/* Avoid warnings due to being atomic. */
524 	__get_user(dummy, (unsigned char __user __force *)v);
525 	pagefault_enable();
526 
527 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
528 		BUG();
529 
530 	if (!PageHighMem(page)) {
531 		void *av = __va(PFN_PHYS(pfn));
532 
533 		if (av != v)
534 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
535 				BUG();
536 	} else
537 		kmap_flush_unused();
538 
539 	preempt_enable();
540 }
541 
542 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
543 {
544 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
545 	int i;
546 
547 	/*
548 	 * We need to mark the all aliases of the LDT pages RO.  We
549 	 * don't need to call vm_flush_aliases(), though, since that's
550 	 * only responsible for flushing aliases out the TLBs, not the
551 	 * page tables, and Xen will flush the TLB for us if needed.
552 	 *
553 	 * To avoid confusing future readers: none of this is necessary
554 	 * to load the LDT.  The hypervisor only checks this when the
555 	 * LDT is faulted in due to subsequent descriptor access.
556 	 */
557 
558 	for(i = 0; i < entries; i += entries_per_page)
559 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
560 }
561 
562 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
563 {
564 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
565 	int i;
566 
567 	for(i = 0; i < entries; i += entries_per_page)
568 		set_aliased_prot(ldt + i, PAGE_KERNEL);
569 }
570 
571 static void xen_set_ldt(const void *addr, unsigned entries)
572 {
573 	struct mmuext_op *op;
574 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
575 
576 	trace_xen_cpu_set_ldt(addr, entries);
577 
578 	op = mcs.args;
579 	op->cmd = MMUEXT_SET_LDT;
580 	op->arg1.linear_addr = (unsigned long)addr;
581 	op->arg2.nr_ents = entries;
582 
583 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
584 
585 	xen_mc_issue(PARAVIRT_LAZY_CPU);
586 }
587 
588 static void xen_load_gdt(const struct desc_ptr *dtr)
589 {
590 	unsigned long va = dtr->address;
591 	unsigned int size = dtr->size + 1;
592 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
593 	unsigned long frames[pages];
594 	int f;
595 
596 	/*
597 	 * A GDT can be up to 64k in size, which corresponds to 8192
598 	 * 8-byte entries, or 16 4k pages..
599 	 */
600 
601 	BUG_ON(size > 65536);
602 	BUG_ON(va & ~PAGE_MASK);
603 
604 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
605 		int level;
606 		pte_t *ptep;
607 		unsigned long pfn, mfn;
608 		void *virt;
609 
610 		/*
611 		 * The GDT is per-cpu and is in the percpu data area.
612 		 * That can be virtually mapped, so we need to do a
613 		 * page-walk to get the underlying MFN for the
614 		 * hypercall.  The page can also be in the kernel's
615 		 * linear range, so we need to RO that mapping too.
616 		 */
617 		ptep = lookup_address(va, &level);
618 		BUG_ON(ptep == NULL);
619 
620 		pfn = pte_pfn(*ptep);
621 		mfn = pfn_to_mfn(pfn);
622 		virt = __va(PFN_PHYS(pfn));
623 
624 		frames[f] = mfn;
625 
626 		make_lowmem_page_readonly((void *)va);
627 		make_lowmem_page_readonly(virt);
628 	}
629 
630 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
631 		BUG();
632 }
633 
634 /*
635  * load_gdt for early boot, when the gdt is only mapped once
636  */
637 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
638 {
639 	unsigned long va = dtr->address;
640 	unsigned int size = dtr->size + 1;
641 	unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
642 	unsigned long frames[pages];
643 	int f;
644 
645 	/*
646 	 * A GDT can be up to 64k in size, which corresponds to 8192
647 	 * 8-byte entries, or 16 4k pages..
648 	 */
649 
650 	BUG_ON(size > 65536);
651 	BUG_ON(va & ~PAGE_MASK);
652 
653 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
654 		pte_t pte;
655 		unsigned long pfn, mfn;
656 
657 		pfn = virt_to_pfn(va);
658 		mfn = pfn_to_mfn(pfn);
659 
660 		pte = pfn_pte(pfn, PAGE_KERNEL_RO);
661 
662 		if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
663 			BUG();
664 
665 		frames[f] = mfn;
666 	}
667 
668 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
669 		BUG();
670 }
671 
672 static inline bool desc_equal(const struct desc_struct *d1,
673 			      const struct desc_struct *d2)
674 {
675 	return d1->a == d2->a && d1->b == d2->b;
676 }
677 
678 static void load_TLS_descriptor(struct thread_struct *t,
679 				unsigned int cpu, unsigned int i)
680 {
681 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
682 	struct desc_struct *gdt;
683 	xmaddr_t maddr;
684 	struct multicall_space mc;
685 
686 	if (desc_equal(shadow, &t->tls_array[i]))
687 		return;
688 
689 	*shadow = t->tls_array[i];
690 
691 	gdt = get_cpu_gdt_table(cpu);
692 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
693 	mc = __xen_mc_entry(0);
694 
695 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
696 }
697 
698 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
699 {
700 	/*
701 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
702 	 * and lazy gs handling is enabled, it means we're in a
703 	 * context switch, and %gs has just been saved.  This means we
704 	 * can zero it out to prevent faults on exit from the
705 	 * hypervisor if the next process has no %gs.  Either way, it
706 	 * has been saved, and the new value will get loaded properly.
707 	 * This will go away as soon as Xen has been modified to not
708 	 * save/restore %gs for normal hypercalls.
709 	 *
710 	 * On x86_64, this hack is not used for %gs, because gs points
711 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
712 	 * must not zero %gs on x86_64
713 	 *
714 	 * For x86_64, we need to zero %fs, otherwise we may get an
715 	 * exception between the new %fs descriptor being loaded and
716 	 * %fs being effectively cleared at __switch_to().
717 	 */
718 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
719 #ifdef CONFIG_X86_32
720 		lazy_load_gs(0);
721 #else
722 		loadsegment(fs, 0);
723 #endif
724 	}
725 
726 	xen_mc_batch();
727 
728 	load_TLS_descriptor(t, cpu, 0);
729 	load_TLS_descriptor(t, cpu, 1);
730 	load_TLS_descriptor(t, cpu, 2);
731 
732 	xen_mc_issue(PARAVIRT_LAZY_CPU);
733 }
734 
735 #ifdef CONFIG_X86_64
736 static void xen_load_gs_index(unsigned int idx)
737 {
738 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
739 		BUG();
740 }
741 #endif
742 
743 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
744 				const void *ptr)
745 {
746 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
747 	u64 entry = *(u64 *)ptr;
748 
749 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
750 
751 	preempt_disable();
752 
753 	xen_mc_flush();
754 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
755 		BUG();
756 
757 	preempt_enable();
758 }
759 
760 static int cvt_gate_to_trap(int vector, const gate_desc *val,
761 			    struct trap_info *info)
762 {
763 	unsigned long addr;
764 
765 	if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
766 		return 0;
767 
768 	info->vector = vector;
769 
770 	addr = gate_offset(*val);
771 #ifdef CONFIG_X86_64
772 	/*
773 	 * Look for known traps using IST, and substitute them
774 	 * appropriately.  The debugger ones are the only ones we care
775 	 * about.  Xen will handle faults like double_fault,
776 	 * so we should never see them.  Warn if
777 	 * there's an unexpected IST-using fault handler.
778 	 */
779 	if (addr == (unsigned long)debug)
780 		addr = (unsigned long)xen_debug;
781 	else if (addr == (unsigned long)int3)
782 		addr = (unsigned long)xen_int3;
783 	else if (addr == (unsigned long)stack_segment)
784 		addr = (unsigned long)xen_stack_segment;
785 	else if (addr == (unsigned long)double_fault) {
786 		/* Don't need to handle these */
787 		return 0;
788 #ifdef CONFIG_X86_MCE
789 	} else if (addr == (unsigned long)machine_check) {
790 		/*
791 		 * when xen hypervisor inject vMCE to guest,
792 		 * use native mce handler to handle it
793 		 */
794 		;
795 #endif
796 	} else if (addr == (unsigned long)nmi)
797 		/*
798 		 * Use the native version as well.
799 		 */
800 		;
801 	else {
802 		/* Some other trap using IST? */
803 		if (WARN_ON(val->ist != 0))
804 			return 0;
805 	}
806 #endif	/* CONFIG_X86_64 */
807 	info->address = addr;
808 
809 	info->cs = gate_segment(*val);
810 	info->flags = val->dpl;
811 	/* interrupt gates clear IF */
812 	if (val->type == GATE_INTERRUPT)
813 		info->flags |= 1 << 2;
814 
815 	return 1;
816 }
817 
818 /* Locations of each CPU's IDT */
819 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
820 
821 /* Set an IDT entry.  If the entry is part of the current IDT, then
822    also update Xen. */
823 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
824 {
825 	unsigned long p = (unsigned long)&dt[entrynum];
826 	unsigned long start, end;
827 
828 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
829 
830 	preempt_disable();
831 
832 	start = __this_cpu_read(idt_desc.address);
833 	end = start + __this_cpu_read(idt_desc.size) + 1;
834 
835 	xen_mc_flush();
836 
837 	native_write_idt_entry(dt, entrynum, g);
838 
839 	if (p >= start && (p + 8) <= end) {
840 		struct trap_info info[2];
841 
842 		info[1].address = 0;
843 
844 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
845 			if (HYPERVISOR_set_trap_table(info))
846 				BUG();
847 	}
848 
849 	preempt_enable();
850 }
851 
852 static void xen_convert_trap_info(const struct desc_ptr *desc,
853 				  struct trap_info *traps)
854 {
855 	unsigned in, out, count;
856 
857 	count = (desc->size+1) / sizeof(gate_desc);
858 	BUG_ON(count > 256);
859 
860 	for (in = out = 0; in < count; in++) {
861 		gate_desc *entry = (gate_desc*)(desc->address) + in;
862 
863 		if (cvt_gate_to_trap(in, entry, &traps[out]))
864 			out++;
865 	}
866 	traps[out].address = 0;
867 }
868 
869 void xen_copy_trap_info(struct trap_info *traps)
870 {
871 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
872 
873 	xen_convert_trap_info(desc, traps);
874 }
875 
876 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
877    hold a spinlock to protect the static traps[] array (static because
878    it avoids allocation, and saves stack space). */
879 static void xen_load_idt(const struct desc_ptr *desc)
880 {
881 	static DEFINE_SPINLOCK(lock);
882 	static struct trap_info traps[257];
883 
884 	trace_xen_cpu_load_idt(desc);
885 
886 	spin_lock(&lock);
887 
888 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
889 
890 	xen_convert_trap_info(desc, traps);
891 
892 	xen_mc_flush();
893 	if (HYPERVISOR_set_trap_table(traps))
894 		BUG();
895 
896 	spin_unlock(&lock);
897 }
898 
899 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
900    they're handled differently. */
901 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
902 				const void *desc, int type)
903 {
904 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
905 
906 	preempt_disable();
907 
908 	switch (type) {
909 	case DESC_LDT:
910 	case DESC_TSS:
911 		/* ignore */
912 		break;
913 
914 	default: {
915 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
916 
917 		xen_mc_flush();
918 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
919 			BUG();
920 	}
921 
922 	}
923 
924 	preempt_enable();
925 }
926 
927 /*
928  * Version of write_gdt_entry for use at early boot-time needed to
929  * update an entry as simply as possible.
930  */
931 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
932 					    const void *desc, int type)
933 {
934 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
935 
936 	switch (type) {
937 	case DESC_LDT:
938 	case DESC_TSS:
939 		/* ignore */
940 		break;
941 
942 	default: {
943 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
944 
945 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
946 			dt[entry] = *(struct desc_struct *)desc;
947 	}
948 
949 	}
950 }
951 
952 static void xen_load_sp0(struct tss_struct *tss,
953 			 struct thread_struct *thread)
954 {
955 	struct multicall_space mcs;
956 
957 	mcs = xen_mc_entry(0);
958 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
959 	xen_mc_issue(PARAVIRT_LAZY_CPU);
960 	tss->x86_tss.sp0 = thread->sp0;
961 }
962 
963 static void xen_set_iopl_mask(unsigned mask)
964 {
965 	struct physdev_set_iopl set_iopl;
966 
967 	/* Force the change at ring 0. */
968 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
969 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
970 }
971 
972 static void xen_io_delay(void)
973 {
974 }
975 
976 static void xen_clts(void)
977 {
978 	struct multicall_space mcs;
979 
980 	mcs = xen_mc_entry(0);
981 
982 	MULTI_fpu_taskswitch(mcs.mc, 0);
983 
984 	xen_mc_issue(PARAVIRT_LAZY_CPU);
985 }
986 
987 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
988 
989 static unsigned long xen_read_cr0(void)
990 {
991 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
992 
993 	if (unlikely(cr0 == 0)) {
994 		cr0 = native_read_cr0();
995 		this_cpu_write(xen_cr0_value, cr0);
996 	}
997 
998 	return cr0;
999 }
1000 
1001 static void xen_write_cr0(unsigned long cr0)
1002 {
1003 	struct multicall_space mcs;
1004 
1005 	this_cpu_write(xen_cr0_value, cr0);
1006 
1007 	/* Only pay attention to cr0.TS; everything else is
1008 	   ignored. */
1009 	mcs = xen_mc_entry(0);
1010 
1011 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1012 
1013 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1014 }
1015 
1016 static void xen_write_cr4(unsigned long cr4)
1017 {
1018 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1019 
1020 	native_write_cr4(cr4);
1021 }
1022 #ifdef CONFIG_X86_64
1023 static inline unsigned long xen_read_cr8(void)
1024 {
1025 	return 0;
1026 }
1027 static inline void xen_write_cr8(unsigned long val)
1028 {
1029 	BUG_ON(val);
1030 }
1031 #endif
1032 
1033 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1034 {
1035 	u64 val;
1036 
1037 	if (pmu_msr_read(msr, &val, err))
1038 		return val;
1039 
1040 	val = native_read_msr_safe(msr, err);
1041 	switch (msr) {
1042 	case MSR_IA32_APICBASE:
1043 #ifdef CONFIG_X86_X2APIC
1044 		if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1045 #endif
1046 			val &= ~X2APIC_ENABLE;
1047 		break;
1048 	}
1049 	return val;
1050 }
1051 
1052 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1053 {
1054 	int ret;
1055 
1056 	ret = 0;
1057 
1058 	switch (msr) {
1059 #ifdef CONFIG_X86_64
1060 		unsigned which;
1061 		u64 base;
1062 
1063 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
1064 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
1065 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
1066 
1067 	set:
1068 		base = ((u64)high << 32) | low;
1069 		if (HYPERVISOR_set_segment_base(which, base) != 0)
1070 			ret = -EIO;
1071 		break;
1072 #endif
1073 
1074 	case MSR_STAR:
1075 	case MSR_CSTAR:
1076 	case MSR_LSTAR:
1077 	case MSR_SYSCALL_MASK:
1078 	case MSR_IA32_SYSENTER_CS:
1079 	case MSR_IA32_SYSENTER_ESP:
1080 	case MSR_IA32_SYSENTER_EIP:
1081 		/* Fast syscall setup is all done in hypercalls, so
1082 		   these are all ignored.  Stub them out here to stop
1083 		   Xen console noise. */
1084 		break;
1085 
1086 	default:
1087 		if (!pmu_msr_write(msr, low, high, &ret))
1088 			ret = native_write_msr_safe(msr, low, high);
1089 	}
1090 
1091 	return ret;
1092 }
1093 
1094 void xen_setup_shared_info(void)
1095 {
1096 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1097 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
1098 			   xen_start_info->shared_info);
1099 
1100 		HYPERVISOR_shared_info =
1101 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1102 	} else
1103 		HYPERVISOR_shared_info =
1104 			(struct shared_info *)__va(xen_start_info->shared_info);
1105 
1106 #ifndef CONFIG_SMP
1107 	/* In UP this is as good a place as any to set up shared info */
1108 	xen_setup_vcpu_info_placement();
1109 #endif
1110 
1111 	xen_setup_mfn_list_list();
1112 }
1113 
1114 /* This is called once we have the cpu_possible_mask */
1115 void xen_setup_vcpu_info_placement(void)
1116 {
1117 	int cpu;
1118 
1119 	for_each_possible_cpu(cpu)
1120 		xen_vcpu_setup(cpu);
1121 
1122 	/* xen_vcpu_setup managed to place the vcpu_info within the
1123 	 * percpu area for all cpus, so make use of it. Note that for
1124 	 * PVH we want to use native IRQ mechanism. */
1125 	if (have_vcpu_info_placement && !xen_pvh_domain()) {
1126 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1127 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1128 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1129 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1130 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1131 	}
1132 }
1133 
1134 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1135 			  unsigned long addr, unsigned len)
1136 {
1137 	char *start, *end, *reloc;
1138 	unsigned ret;
1139 
1140 	start = end = reloc = NULL;
1141 
1142 #define SITE(op, x)							\
1143 	case PARAVIRT_PATCH(op.x):					\
1144 	if (have_vcpu_info_placement) {					\
1145 		start = (char *)xen_##x##_direct;			\
1146 		end = xen_##x##_direct_end;				\
1147 		reloc = xen_##x##_direct_reloc;				\
1148 	}								\
1149 	goto patch_site
1150 
1151 	switch (type) {
1152 		SITE(pv_irq_ops, irq_enable);
1153 		SITE(pv_irq_ops, irq_disable);
1154 		SITE(pv_irq_ops, save_fl);
1155 		SITE(pv_irq_ops, restore_fl);
1156 #undef SITE
1157 
1158 	patch_site:
1159 		if (start == NULL || (end-start) > len)
1160 			goto default_patch;
1161 
1162 		ret = paravirt_patch_insns(insnbuf, len, start, end);
1163 
1164 		/* Note: because reloc is assigned from something that
1165 		   appears to be an array, gcc assumes it's non-null,
1166 		   but doesn't know its relationship with start and
1167 		   end. */
1168 		if (reloc > start && reloc < end) {
1169 			int reloc_off = reloc - start;
1170 			long *relocp = (long *)(insnbuf + reloc_off);
1171 			long delta = start - (char *)addr;
1172 
1173 			*relocp += delta;
1174 		}
1175 		break;
1176 
1177 	default_patch:
1178 	default:
1179 		ret = paravirt_patch_default(type, clobbers, insnbuf,
1180 					     addr, len);
1181 		break;
1182 	}
1183 
1184 	return ret;
1185 }
1186 
1187 static const struct pv_info xen_info __initconst = {
1188 	.paravirt_enabled = 1,
1189 	.shared_kernel_pmd = 0,
1190 
1191 #ifdef CONFIG_X86_64
1192 	.extra_user_64bit_cs = FLAT_USER_CS64,
1193 #endif
1194 
1195 	.name = "Xen",
1196 };
1197 
1198 static const struct pv_init_ops xen_init_ops __initconst = {
1199 	.patch = xen_patch,
1200 };
1201 
1202 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1203 	.cpuid = xen_cpuid,
1204 
1205 	.set_debugreg = xen_set_debugreg,
1206 	.get_debugreg = xen_get_debugreg,
1207 
1208 	.clts = xen_clts,
1209 
1210 	.read_cr0 = xen_read_cr0,
1211 	.write_cr0 = xen_write_cr0,
1212 
1213 	.read_cr4 = native_read_cr4,
1214 	.read_cr4_safe = native_read_cr4_safe,
1215 	.write_cr4 = xen_write_cr4,
1216 
1217 #ifdef CONFIG_X86_64
1218 	.read_cr8 = xen_read_cr8,
1219 	.write_cr8 = xen_write_cr8,
1220 #endif
1221 
1222 	.wbinvd = native_wbinvd,
1223 
1224 	.read_msr = xen_read_msr_safe,
1225 	.write_msr = xen_write_msr_safe,
1226 
1227 	.read_pmc = xen_read_pmc,
1228 
1229 	.iret = xen_iret,
1230 #ifdef CONFIG_X86_64
1231 	.usergs_sysret32 = xen_sysret32,
1232 	.usergs_sysret64 = xen_sysret64,
1233 #else
1234 	.irq_enable_sysexit = xen_sysexit,
1235 #endif
1236 
1237 	.load_tr_desc = paravirt_nop,
1238 	.set_ldt = xen_set_ldt,
1239 	.load_gdt = xen_load_gdt,
1240 	.load_idt = xen_load_idt,
1241 	.load_tls = xen_load_tls,
1242 #ifdef CONFIG_X86_64
1243 	.load_gs_index = xen_load_gs_index,
1244 #endif
1245 
1246 	.alloc_ldt = xen_alloc_ldt,
1247 	.free_ldt = xen_free_ldt,
1248 
1249 	.store_idt = native_store_idt,
1250 	.store_tr = xen_store_tr,
1251 
1252 	.write_ldt_entry = xen_write_ldt_entry,
1253 	.write_gdt_entry = xen_write_gdt_entry,
1254 	.write_idt_entry = xen_write_idt_entry,
1255 	.load_sp0 = xen_load_sp0,
1256 
1257 	.set_iopl_mask = xen_set_iopl_mask,
1258 	.io_delay = xen_io_delay,
1259 
1260 	/* Xen takes care of %gs when switching to usermode for us */
1261 	.swapgs = paravirt_nop,
1262 
1263 	.start_context_switch = paravirt_start_context_switch,
1264 	.end_context_switch = xen_end_context_switch,
1265 };
1266 
1267 static const struct pv_apic_ops xen_apic_ops __initconst = {
1268 #ifdef CONFIG_X86_LOCAL_APIC
1269 	.startup_ipi_hook = paravirt_nop,
1270 #endif
1271 };
1272 
1273 static void xen_reboot(int reason)
1274 {
1275 	struct sched_shutdown r = { .reason = reason };
1276 	int cpu;
1277 
1278 	for_each_online_cpu(cpu)
1279 		xen_pmu_finish(cpu);
1280 
1281 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1282 		BUG();
1283 }
1284 
1285 static void xen_restart(char *msg)
1286 {
1287 	xen_reboot(SHUTDOWN_reboot);
1288 }
1289 
1290 static void xen_emergency_restart(void)
1291 {
1292 	xen_reboot(SHUTDOWN_reboot);
1293 }
1294 
1295 static void xen_machine_halt(void)
1296 {
1297 	xen_reboot(SHUTDOWN_poweroff);
1298 }
1299 
1300 static void xen_machine_power_off(void)
1301 {
1302 	if (pm_power_off)
1303 		pm_power_off();
1304 	xen_reboot(SHUTDOWN_poweroff);
1305 }
1306 
1307 static void xen_crash_shutdown(struct pt_regs *regs)
1308 {
1309 	xen_reboot(SHUTDOWN_crash);
1310 }
1311 
1312 static int
1313 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1314 {
1315 	xen_reboot(SHUTDOWN_crash);
1316 	return NOTIFY_DONE;
1317 }
1318 
1319 static struct notifier_block xen_panic_block = {
1320 	.notifier_call= xen_panic_event,
1321 	.priority = INT_MIN
1322 };
1323 
1324 int xen_panic_handler_init(void)
1325 {
1326 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1327 	return 0;
1328 }
1329 
1330 static const struct machine_ops xen_machine_ops __initconst = {
1331 	.restart = xen_restart,
1332 	.halt = xen_machine_halt,
1333 	.power_off = xen_machine_power_off,
1334 	.shutdown = xen_machine_halt,
1335 	.crash_shutdown = xen_crash_shutdown,
1336 	.emergency_restart = xen_emergency_restart,
1337 };
1338 
1339 static unsigned char xen_get_nmi_reason(void)
1340 {
1341 	unsigned char reason = 0;
1342 
1343 	/* Construct a value which looks like it came from port 0x61. */
1344 	if (test_bit(_XEN_NMIREASON_io_error,
1345 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1346 		reason |= NMI_REASON_IOCHK;
1347 	if (test_bit(_XEN_NMIREASON_pci_serr,
1348 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1349 		reason |= NMI_REASON_SERR;
1350 
1351 	return reason;
1352 }
1353 
1354 static void __init xen_boot_params_init_edd(void)
1355 {
1356 #if IS_ENABLED(CONFIG_EDD)
1357 	struct xen_platform_op op;
1358 	struct edd_info *edd_info;
1359 	u32 *mbr_signature;
1360 	unsigned nr;
1361 	int ret;
1362 
1363 	edd_info = boot_params.eddbuf;
1364 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1365 
1366 	op.cmd = XENPF_firmware_info;
1367 
1368 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1369 	for (nr = 0; nr < EDDMAXNR; nr++) {
1370 		struct edd_info *info = edd_info + nr;
1371 
1372 		op.u.firmware_info.index = nr;
1373 		info->params.length = sizeof(info->params);
1374 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1375 				     &info->params);
1376 		ret = HYPERVISOR_dom0_op(&op);
1377 		if (ret)
1378 			break;
1379 
1380 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1381 		C(device);
1382 		C(version);
1383 		C(interface_support);
1384 		C(legacy_max_cylinder);
1385 		C(legacy_max_head);
1386 		C(legacy_sectors_per_track);
1387 #undef C
1388 	}
1389 	boot_params.eddbuf_entries = nr;
1390 
1391 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1392 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1393 		op.u.firmware_info.index = nr;
1394 		ret = HYPERVISOR_dom0_op(&op);
1395 		if (ret)
1396 			break;
1397 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1398 	}
1399 	boot_params.edd_mbr_sig_buf_entries = nr;
1400 #endif
1401 }
1402 
1403 /*
1404  * Set up the GDT and segment registers for -fstack-protector.  Until
1405  * we do this, we have to be careful not to call any stack-protected
1406  * function, which is most of the kernel.
1407  *
1408  * Note, that it is __ref because the only caller of this after init
1409  * is PVH which is not going to use xen_load_gdt_boot or other
1410  * __init functions.
1411  */
1412 static void __ref xen_setup_gdt(int cpu)
1413 {
1414 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
1415 #ifdef CONFIG_X86_64
1416 		unsigned long dummy;
1417 
1418 		load_percpu_segment(cpu); /* We need to access per-cpu area */
1419 		switch_to_new_gdt(cpu); /* GDT and GS set */
1420 
1421 		/* We are switching of the Xen provided GDT to our HVM mode
1422 		 * GDT. The new GDT has  __KERNEL_CS with CS.L = 1
1423 		 * and we are jumping to reload it.
1424 		 */
1425 		asm volatile ("pushq %0\n"
1426 			      "leaq 1f(%%rip),%0\n"
1427 			      "pushq %0\n"
1428 			      "lretq\n"
1429 			      "1:\n"
1430 			      : "=&r" (dummy) : "0" (__KERNEL_CS));
1431 
1432 		/*
1433 		 * While not needed, we also set the %es, %ds, and %fs
1434 		 * to zero. We don't care about %ss as it is NULL.
1435 		 * Strictly speaking this is not needed as Xen zeros those
1436 		 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1437 		 *
1438 		 * Linux zeros them in cpu_init() and in secondary_startup_64
1439 		 * (for BSP).
1440 		 */
1441 		loadsegment(es, 0);
1442 		loadsegment(ds, 0);
1443 		loadsegment(fs, 0);
1444 #else
1445 		/* PVH: TODO Implement. */
1446 		BUG();
1447 #endif
1448 		return; /* PVH does not need any PV GDT ops. */
1449 	}
1450 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1451 	pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1452 
1453 	setup_stack_canary_segment(0);
1454 	switch_to_new_gdt(0);
1455 
1456 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1457 	pv_cpu_ops.load_gdt = xen_load_gdt;
1458 }
1459 
1460 #ifdef CONFIG_XEN_PVH
1461 /*
1462  * A PV guest starts with default flags that are not set for PVH, set them
1463  * here asap.
1464  */
1465 static void xen_pvh_set_cr_flags(int cpu)
1466 {
1467 
1468 	/* Some of these are setup in 'secondary_startup_64'. The others:
1469 	 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1470 	 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1471 	write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
1472 
1473 	if (!cpu)
1474 		return;
1475 	/*
1476 	 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1477 	 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
1478 	*/
1479 	if (cpu_has_pse)
1480 		cr4_set_bits_and_update_boot(X86_CR4_PSE);
1481 
1482 	if (cpu_has_pge)
1483 		cr4_set_bits_and_update_boot(X86_CR4_PGE);
1484 }
1485 
1486 /*
1487  * Note, that it is ref - because the only caller of this after init
1488  * is PVH which is not going to use xen_load_gdt_boot or other
1489  * __init functions.
1490  */
1491 void __ref xen_pvh_secondary_vcpu_init(int cpu)
1492 {
1493 	xen_setup_gdt(cpu);
1494 	xen_pvh_set_cr_flags(cpu);
1495 }
1496 
1497 static void __init xen_pvh_early_guest_init(void)
1498 {
1499 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1500 		return;
1501 
1502 	if (!xen_feature(XENFEAT_hvm_callback_vector))
1503 		return;
1504 
1505 	xen_have_vector_callback = 1;
1506 
1507 	xen_pvh_early_cpu_init(0, false);
1508 	xen_pvh_set_cr_flags(0);
1509 
1510 #ifdef CONFIG_X86_32
1511 	BUG(); /* PVH: Implement proper support. */
1512 #endif
1513 }
1514 #endif    /* CONFIG_XEN_PVH */
1515 
1516 /* First C function to be called on Xen boot */
1517 asmlinkage __visible void __init xen_start_kernel(void)
1518 {
1519 	struct physdev_set_iopl set_iopl;
1520 	unsigned long initrd_start = 0;
1521 	u64 pat;
1522 	int rc;
1523 
1524 	if (!xen_start_info)
1525 		return;
1526 
1527 	xen_domain_type = XEN_PV_DOMAIN;
1528 
1529 	xen_setup_features();
1530 #ifdef CONFIG_XEN_PVH
1531 	xen_pvh_early_guest_init();
1532 #endif
1533 	xen_setup_machphys_mapping();
1534 
1535 	/* Install Xen paravirt ops */
1536 	pv_info = xen_info;
1537 	pv_init_ops = xen_init_ops;
1538 	pv_apic_ops = xen_apic_ops;
1539 	if (!xen_pvh_domain()) {
1540 		pv_cpu_ops = xen_cpu_ops;
1541 
1542 		x86_platform.get_nmi_reason = xen_get_nmi_reason;
1543 	}
1544 
1545 	if (xen_feature(XENFEAT_auto_translated_physmap))
1546 		x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1547 	else
1548 		x86_init.resources.memory_setup = xen_memory_setup;
1549 	x86_init.oem.arch_setup = xen_arch_setup;
1550 	x86_init.oem.banner = xen_banner;
1551 
1552 	xen_init_time_ops();
1553 
1554 	/*
1555 	 * Set up some pagetable state before starting to set any ptes.
1556 	 */
1557 
1558 	xen_init_mmu_ops();
1559 
1560 	/* Prevent unwanted bits from being set in PTEs. */
1561 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1562 
1563 	/*
1564 	 * Prevent page tables from being allocated in highmem, even
1565 	 * if CONFIG_HIGHPTE is enabled.
1566 	 */
1567 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1568 
1569 	/* Work out if we support NX */
1570 	x86_configure_nx();
1571 
1572 	/* Get mfn list */
1573 	xen_build_dynamic_phys_to_machine();
1574 
1575 	/*
1576 	 * Set up kernel GDT and segment registers, mainly so that
1577 	 * -fstack-protector code can be executed.
1578 	 */
1579 	xen_setup_gdt(0);
1580 
1581 	xen_init_irq_ops();
1582 	xen_init_cpuid_mask();
1583 
1584 #ifdef CONFIG_X86_LOCAL_APIC
1585 	/*
1586 	 * set up the basic apic ops.
1587 	 */
1588 	xen_init_apic();
1589 #endif
1590 
1591 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1592 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1593 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1594 	}
1595 
1596 	machine_ops = xen_machine_ops;
1597 
1598 	/*
1599 	 * The only reliable way to retain the initial address of the
1600 	 * percpu gdt_page is to remember it here, so we can go and
1601 	 * mark it RW later, when the initial percpu area is freed.
1602 	 */
1603 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1604 
1605 	xen_smp_init();
1606 
1607 #ifdef CONFIG_ACPI_NUMA
1608 	/*
1609 	 * The pages we from Xen are not related to machine pages, so
1610 	 * any NUMA information the kernel tries to get from ACPI will
1611 	 * be meaningless.  Prevent it from trying.
1612 	 */
1613 	acpi_numa = -1;
1614 #endif
1615 	/* Don't do the full vcpu_info placement stuff until we have a
1616 	   possible map and a non-dummy shared_info. */
1617 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1618 
1619 	local_irq_disable();
1620 	early_boot_irqs_disabled = true;
1621 
1622 	xen_raw_console_write("mapping kernel into physical memory\n");
1623 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1624 				   xen_start_info->nr_pages);
1625 	xen_reserve_special_pages();
1626 
1627 	/*
1628 	 * Modify the cache mode translation tables to match Xen's PAT
1629 	 * configuration.
1630 	 */
1631 	rdmsrl(MSR_IA32_CR_PAT, pat);
1632 	pat_init_cache_modes(pat);
1633 
1634 	/* keep using Xen gdt for now; no urgent need to change it */
1635 
1636 #ifdef CONFIG_X86_32
1637 	pv_info.kernel_rpl = 1;
1638 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1639 		pv_info.kernel_rpl = 0;
1640 #else
1641 	pv_info.kernel_rpl = 0;
1642 #endif
1643 	/* set the limit of our address space */
1644 	xen_reserve_top();
1645 
1646 	/* PVH: runs at default kernel iopl of 0 */
1647 	if (!xen_pvh_domain()) {
1648 		/*
1649 		 * We used to do this in xen_arch_setup, but that is too late
1650 		 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1651 		 * early_amd_init which pokes 0xcf8 port.
1652 		 */
1653 		set_iopl.iopl = 1;
1654 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1655 		if (rc != 0)
1656 			xen_raw_printk("physdev_op failed %d\n", rc);
1657 	}
1658 
1659 #ifdef CONFIG_X86_32
1660 	/* set up basic CPUID stuff */
1661 	cpu_detect(&new_cpu_data);
1662 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1663 	new_cpu_data.wp_works_ok = 1;
1664 	new_cpu_data.x86_capability[0] = cpuid_edx(1);
1665 #endif
1666 
1667 	if (xen_start_info->mod_start) {
1668 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1669 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1670 	    else
1671 		initrd_start = __pa(xen_start_info->mod_start);
1672 	}
1673 
1674 	/* Poke various useful things into boot_params */
1675 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1676 	boot_params.hdr.ramdisk_image = initrd_start;
1677 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1678 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1679 
1680 	if (!xen_initial_domain()) {
1681 		add_preferred_console("xenboot", 0, NULL);
1682 		add_preferred_console("tty", 0, NULL);
1683 		add_preferred_console("hvc", 0, NULL);
1684 		if (pci_xen)
1685 			x86_init.pci.arch_init = pci_xen_init;
1686 	} else {
1687 		const struct dom0_vga_console_info *info =
1688 			(void *)((char *)xen_start_info +
1689 				 xen_start_info->console.dom0.info_off);
1690 		struct xen_platform_op op = {
1691 			.cmd = XENPF_firmware_info,
1692 			.interface_version = XENPF_INTERFACE_VERSION,
1693 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1694 		};
1695 
1696 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1697 		xen_start_info->console.domU.mfn = 0;
1698 		xen_start_info->console.domU.evtchn = 0;
1699 
1700 		if (HYPERVISOR_dom0_op(&op) == 0)
1701 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1702 
1703 		/* Make sure ACS will be enabled */
1704 		pci_request_acs();
1705 
1706 		xen_acpi_sleep_register();
1707 
1708 		/* Avoid searching for BIOS MP tables */
1709 		x86_init.mpparse.find_smp_config = x86_init_noop;
1710 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1711 
1712 		xen_boot_params_init_edd();
1713 	}
1714 #ifdef CONFIG_PCI
1715 	/* PCI BIOS service won't work from a PV guest. */
1716 	pci_probe &= ~PCI_PROBE_BIOS;
1717 #endif
1718 	xen_raw_console_write("about to get started...\n");
1719 
1720 	xen_setup_runstate_info(0);
1721 
1722 	xen_efi_init();
1723 
1724 	/* Start the world */
1725 #ifdef CONFIG_X86_32
1726 	i386_start_kernel();
1727 #else
1728 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1729 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1730 #endif
1731 }
1732 
1733 void __ref xen_hvm_init_shared_info(void)
1734 {
1735 	int cpu;
1736 	struct xen_add_to_physmap xatp;
1737 	static struct shared_info *shared_info_page = 0;
1738 
1739 	if (!shared_info_page)
1740 		shared_info_page = (struct shared_info *)
1741 			extend_brk(PAGE_SIZE, PAGE_SIZE);
1742 	xatp.domid = DOMID_SELF;
1743 	xatp.idx = 0;
1744 	xatp.space = XENMAPSPACE_shared_info;
1745 	xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1746 	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1747 		BUG();
1748 
1749 	HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1750 
1751 	/* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1752 	 * page, we use it in the event channel upcall and in some pvclock
1753 	 * related functions. We don't need the vcpu_info placement
1754 	 * optimizations because we don't use any pv_mmu or pv_irq op on
1755 	 * HVM.
1756 	 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1757 	 * online but xen_hvm_init_shared_info is run at resume time too and
1758 	 * in that case multiple vcpus might be online. */
1759 	for_each_online_cpu(cpu) {
1760 		/* Leave it to be NULL. */
1761 		if (cpu >= MAX_VIRT_CPUS)
1762 			continue;
1763 		per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1764 	}
1765 }
1766 
1767 #ifdef CONFIG_XEN_PVHVM
1768 static void __init init_hvm_pv_info(void)
1769 {
1770 	int major, minor;
1771 	uint32_t eax, ebx, ecx, edx, pages, msr, base;
1772 	u64 pfn;
1773 
1774 	base = xen_cpuid_base();
1775 	cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1776 
1777 	major = eax >> 16;
1778 	minor = eax & 0xffff;
1779 	printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1780 
1781 	cpuid(base + 2, &pages, &msr, &ecx, &edx);
1782 
1783 	pfn = __pa(hypercall_page);
1784 	wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1785 
1786 	xen_setup_features();
1787 
1788 	pv_info.name = "Xen HVM";
1789 
1790 	xen_domain_type = XEN_HVM_DOMAIN;
1791 }
1792 
1793 static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
1794 			      void *hcpu)
1795 {
1796 	int cpu = (long)hcpu;
1797 	switch (action) {
1798 	case CPU_UP_PREPARE:
1799 		xen_vcpu_setup(cpu);
1800 		if (xen_have_vector_callback) {
1801 			if (xen_feature(XENFEAT_hvm_safe_pvclock))
1802 				xen_setup_timer(cpu);
1803 		}
1804 		break;
1805 	default:
1806 		break;
1807 	}
1808 	return NOTIFY_OK;
1809 }
1810 
1811 static struct notifier_block xen_hvm_cpu_notifier = {
1812 	.notifier_call	= xen_hvm_cpu_notify,
1813 };
1814 
1815 #ifdef CONFIG_KEXEC_CORE
1816 static void xen_hvm_shutdown(void)
1817 {
1818 	native_machine_shutdown();
1819 	if (kexec_in_progress)
1820 		xen_reboot(SHUTDOWN_soft_reset);
1821 }
1822 
1823 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1824 {
1825 	native_machine_crash_shutdown(regs);
1826 	xen_reboot(SHUTDOWN_soft_reset);
1827 }
1828 #endif
1829 
1830 static void __init xen_hvm_guest_init(void)
1831 {
1832 	if (xen_pv_domain())
1833 		return;
1834 
1835 	init_hvm_pv_info();
1836 
1837 	xen_hvm_init_shared_info();
1838 
1839 	xen_panic_handler_init();
1840 
1841 	if (xen_feature(XENFEAT_hvm_callback_vector))
1842 		xen_have_vector_callback = 1;
1843 	xen_hvm_smp_init();
1844 	register_cpu_notifier(&xen_hvm_cpu_notifier);
1845 	xen_unplug_emulated_devices();
1846 	x86_init.irqs.intr_init = xen_init_IRQ;
1847 	xen_hvm_init_time_ops();
1848 	xen_hvm_init_mmu_ops();
1849 #ifdef CONFIG_KEXEC_CORE
1850 	machine_ops.shutdown = xen_hvm_shutdown;
1851 	machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1852 #endif
1853 }
1854 #endif
1855 
1856 static bool xen_nopv = false;
1857 static __init int xen_parse_nopv(char *arg)
1858 {
1859        xen_nopv = true;
1860        return 0;
1861 }
1862 early_param("xen_nopv", xen_parse_nopv);
1863 
1864 static uint32_t __init xen_platform(void)
1865 {
1866 	if (xen_nopv)
1867 		return 0;
1868 
1869 	return xen_cpuid_base();
1870 }
1871 
1872 bool xen_hvm_need_lapic(void)
1873 {
1874 	if (xen_nopv)
1875 		return false;
1876 	if (xen_pv_domain())
1877 		return false;
1878 	if (!xen_hvm_domain())
1879 		return false;
1880 	if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1881 		return false;
1882 	return true;
1883 }
1884 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1885 
1886 static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1887 {
1888 	if (xen_pv_domain())
1889 		clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1890 }
1891 
1892 const struct hypervisor_x86 x86_hyper_xen = {
1893 	.name			= "Xen",
1894 	.detect			= xen_platform,
1895 #ifdef CONFIG_XEN_PVHVM
1896 	.init_platform		= xen_hvm_guest_init,
1897 #endif
1898 	.x2apic_available	= xen_x2apic_para_available,
1899 	.set_cpu_features       = xen_set_cpu_features,
1900 };
1901 EXPORT_SYMBOL(x86_hyper_xen);
1902