xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 82df5b73)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Core of Xen paravirt_ops implementation.
4  *
5  * This file contains the xen_paravirt_ops structure itself, and the
6  * implementations for:
7  * - privileged instructions
8  * - interrupt flags
9  * - segment operations
10  * - booting and setup
11  *
12  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
13  */
14 
15 #include <linux/cpu.h>
16 #include <linux/kernel.h>
17 #include <linux/init.h>
18 #include <linux/smp.h>
19 #include <linux/preempt.h>
20 #include <linux/hardirq.h>
21 #include <linux/percpu.h>
22 #include <linux/delay.h>
23 #include <linux/start_kernel.h>
24 #include <linux/sched.h>
25 #include <linux/kprobes.h>
26 #include <linux/memblock.h>
27 #include <linux/export.h>
28 #include <linux/mm.h>
29 #include <linux/page-flags.h>
30 #include <linux/highmem.h>
31 #include <linux/console.h>
32 #include <linux/pci.h>
33 #include <linux/gfp.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
36 
37 #include <xen/xen.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
47 #include <xen/page.h>
48 #include <xen/hvc-console.h>
49 #include <xen/acpi.h>
50 
51 #include <asm/paravirt.h>
52 #include <asm/apic.h>
53 #include <asm/page.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
64 #include <asm/desc.h>
65 #include <asm/pgalloc.h>
66 #include <asm/tlbflush.h>
67 #include <asm/reboot.h>
68 #include <asm/stackprotector.h>
69 #include <asm/hypervisor.h>
70 #include <asm/mach_traps.h>
71 #include <asm/mwait.h>
72 #include <asm/pci_x86.h>
73 #include <asm/cpu.h>
74 #ifdef CONFIG_X86_IOPL_IOPERM
75 #include <asm/io_bitmap.h>
76 #endif
77 
78 #ifdef CONFIG_ACPI
79 #include <linux/acpi.h>
80 #include <asm/acpi.h>
81 #include <acpi/pdc_intel.h>
82 #include <acpi/processor.h>
83 #include <xen/interface/platform.h>
84 #endif
85 
86 #include "xen-ops.h"
87 #include "mmu.h"
88 #include "smp.h"
89 #include "multicalls.h"
90 #include "pmu.h"
91 
92 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
93 
94 void *xen_initial_gdt;
95 
96 static int xen_cpu_up_prepare_pv(unsigned int cpu);
97 static int xen_cpu_dead_pv(unsigned int cpu);
98 
99 struct tls_descs {
100 	struct desc_struct desc[3];
101 };
102 
103 /*
104  * Updating the 3 TLS descriptors in the GDT on every task switch is
105  * surprisingly expensive so we avoid updating them if they haven't
106  * changed.  Since Xen writes different descriptors than the one
107  * passed in the update_descriptor hypercall we keep shadow copies to
108  * compare against.
109  */
110 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
111 
112 static void __init xen_banner(void)
113 {
114 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
115 	struct xen_extraversion extra;
116 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
117 
118 	pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
119 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
120 	       version >> 16, version & 0xffff, extra.extraversion,
121 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
122 
123 #ifdef CONFIG_X86_32
124 	pr_warn("WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! WARNING!\n"
125 		"Support for running as 32-bit PV-guest under Xen will soon be removed\n"
126 		"from the Linux kernel!\n"
127 		"Please use either a 64-bit kernel or switch to HVM or PVH mode!\n"
128 		"WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! WARNING!\n");
129 #endif
130 }
131 
132 static void __init xen_pv_init_platform(void)
133 {
134 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
135 
136 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
137 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
138 
139 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
140 	xen_vcpu_info_reset(0);
141 
142 	/* pvclock is in shared info area */
143 	xen_init_time_ops();
144 }
145 
146 static void __init xen_pv_guest_late_init(void)
147 {
148 #ifndef CONFIG_SMP
149 	/* Setup shared vcpu info for non-smp configurations */
150 	xen_setup_vcpu_info_placement();
151 #endif
152 }
153 
154 /* Check if running on Xen version (major, minor) or later */
155 bool
156 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
157 {
158 	unsigned int version;
159 
160 	if (!xen_domain())
161 		return false;
162 
163 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
164 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
165 		((version >> 16) > major))
166 		return true;
167 	return false;
168 }
169 
170 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
171 static __read_mostly unsigned int cpuid_leaf5_edx_val;
172 
173 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
174 		      unsigned int *cx, unsigned int *dx)
175 {
176 	unsigned maskebx = ~0;
177 
178 	/*
179 	 * Mask out inconvenient features, to try and disable as many
180 	 * unsupported kernel subsystems as possible.
181 	 */
182 	switch (*ax) {
183 	case CPUID_MWAIT_LEAF:
184 		/* Synthesize the values.. */
185 		*ax = 0;
186 		*bx = 0;
187 		*cx = cpuid_leaf5_ecx_val;
188 		*dx = cpuid_leaf5_edx_val;
189 		return;
190 
191 	case 0xb:
192 		/* Suppress extended topology stuff */
193 		maskebx = 0;
194 		break;
195 	}
196 
197 	asm(XEN_EMULATE_PREFIX "cpuid"
198 		: "=a" (*ax),
199 		  "=b" (*bx),
200 		  "=c" (*cx),
201 		  "=d" (*dx)
202 		: "0" (*ax), "2" (*cx));
203 
204 	*bx &= maskebx;
205 }
206 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
207 
208 static bool __init xen_check_mwait(void)
209 {
210 #ifdef CONFIG_ACPI
211 	struct xen_platform_op op = {
212 		.cmd			= XENPF_set_processor_pminfo,
213 		.u.set_pminfo.id	= -1,
214 		.u.set_pminfo.type	= XEN_PM_PDC,
215 	};
216 	uint32_t buf[3];
217 	unsigned int ax, bx, cx, dx;
218 	unsigned int mwait_mask;
219 
220 	/* We need to determine whether it is OK to expose the MWAIT
221 	 * capability to the kernel to harvest deeper than C3 states from ACPI
222 	 * _CST using the processor_harvest_xen.c module. For this to work, we
223 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
224 	 * checks against). The hypervisor won't expose the MWAIT flag because
225 	 * it would break backwards compatibility; so we will find out directly
226 	 * from the hardware and hypercall.
227 	 */
228 	if (!xen_initial_domain())
229 		return false;
230 
231 	/*
232 	 * When running under platform earlier than Xen4.2, do not expose
233 	 * mwait, to avoid the risk of loading native acpi pad driver
234 	 */
235 	if (!xen_running_on_version_or_later(4, 2))
236 		return false;
237 
238 	ax = 1;
239 	cx = 0;
240 
241 	native_cpuid(&ax, &bx, &cx, &dx);
242 
243 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
244 		     (1 << (X86_FEATURE_MWAIT % 32));
245 
246 	if ((cx & mwait_mask) != mwait_mask)
247 		return false;
248 
249 	/* We need to emulate the MWAIT_LEAF and for that we need both
250 	 * ecx and edx. The hypercall provides only partial information.
251 	 */
252 
253 	ax = CPUID_MWAIT_LEAF;
254 	bx = 0;
255 	cx = 0;
256 	dx = 0;
257 
258 	native_cpuid(&ax, &bx, &cx, &dx);
259 
260 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
261 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
262 	 */
263 	buf[0] = ACPI_PDC_REVISION_ID;
264 	buf[1] = 1;
265 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
266 
267 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
268 
269 	if ((HYPERVISOR_platform_op(&op) == 0) &&
270 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
271 		cpuid_leaf5_ecx_val = cx;
272 		cpuid_leaf5_edx_val = dx;
273 	}
274 	return true;
275 #else
276 	return false;
277 #endif
278 }
279 
280 static bool __init xen_check_xsave(void)
281 {
282 	unsigned int cx, xsave_mask;
283 
284 	cx = cpuid_ecx(1);
285 
286 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
287 		     (1 << (X86_FEATURE_OSXSAVE % 32));
288 
289 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
290 	return (cx & xsave_mask) == xsave_mask;
291 }
292 
293 static void __init xen_init_capabilities(void)
294 {
295 	setup_force_cpu_cap(X86_FEATURE_XENPV);
296 	setup_clear_cpu_cap(X86_FEATURE_DCA);
297 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
298 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
299 	setup_clear_cpu_cap(X86_FEATURE_ACC);
300 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
301 	setup_clear_cpu_cap(X86_FEATURE_SME);
302 
303 	/*
304 	 * Xen PV would need some work to support PCID: CR3 handling as well
305 	 * as xen_flush_tlb_others() would need updating.
306 	 */
307 	setup_clear_cpu_cap(X86_FEATURE_PCID);
308 
309 	if (!xen_initial_domain())
310 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
311 
312 	if (xen_check_mwait())
313 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
314 	else
315 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
316 
317 	if (!xen_check_xsave()) {
318 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
319 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
320 	}
321 }
322 
323 static void xen_set_debugreg(int reg, unsigned long val)
324 {
325 	HYPERVISOR_set_debugreg(reg, val);
326 }
327 
328 static unsigned long xen_get_debugreg(int reg)
329 {
330 	return HYPERVISOR_get_debugreg(reg);
331 }
332 
333 static void xen_end_context_switch(struct task_struct *next)
334 {
335 	xen_mc_flush();
336 	paravirt_end_context_switch(next);
337 }
338 
339 static unsigned long xen_store_tr(void)
340 {
341 	return 0;
342 }
343 
344 /*
345  * Set the page permissions for a particular virtual address.  If the
346  * address is a vmalloc mapping (or other non-linear mapping), then
347  * find the linear mapping of the page and also set its protections to
348  * match.
349  */
350 static void set_aliased_prot(void *v, pgprot_t prot)
351 {
352 	int level;
353 	pte_t *ptep;
354 	pte_t pte;
355 	unsigned long pfn;
356 	struct page *page;
357 	unsigned char dummy;
358 
359 	ptep = lookup_address((unsigned long)v, &level);
360 	BUG_ON(ptep == NULL);
361 
362 	pfn = pte_pfn(*ptep);
363 	page = pfn_to_page(pfn);
364 
365 	pte = pfn_pte(pfn, prot);
366 
367 	/*
368 	 * Careful: update_va_mapping() will fail if the virtual address
369 	 * we're poking isn't populated in the page tables.  We don't
370 	 * need to worry about the direct map (that's always in the page
371 	 * tables), but we need to be careful about vmap space.  In
372 	 * particular, the top level page table can lazily propagate
373 	 * entries between processes, so if we've switched mms since we
374 	 * vmapped the target in the first place, we might not have the
375 	 * top-level page table entry populated.
376 	 *
377 	 * We disable preemption because we want the same mm active when
378 	 * we probe the target and when we issue the hypercall.  We'll
379 	 * have the same nominal mm, but if we're a kernel thread, lazy
380 	 * mm dropping could change our pgd.
381 	 *
382 	 * Out of an abundance of caution, this uses __get_user() to fault
383 	 * in the target address just in case there's some obscure case
384 	 * in which the target address isn't readable.
385 	 */
386 
387 	preempt_disable();
388 
389 	probe_kernel_read(&dummy, v, 1);
390 
391 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
392 		BUG();
393 
394 	if (!PageHighMem(page)) {
395 		void *av = __va(PFN_PHYS(pfn));
396 
397 		if (av != v)
398 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
399 				BUG();
400 	} else
401 		kmap_flush_unused();
402 
403 	preempt_enable();
404 }
405 
406 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
407 {
408 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
409 	int i;
410 
411 	/*
412 	 * We need to mark the all aliases of the LDT pages RO.  We
413 	 * don't need to call vm_flush_aliases(), though, since that's
414 	 * only responsible for flushing aliases out the TLBs, not the
415 	 * page tables, and Xen will flush the TLB for us if needed.
416 	 *
417 	 * To avoid confusing future readers: none of this is necessary
418 	 * to load the LDT.  The hypervisor only checks this when the
419 	 * LDT is faulted in due to subsequent descriptor access.
420 	 */
421 
422 	for (i = 0; i < entries; i += entries_per_page)
423 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
424 }
425 
426 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
427 {
428 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
429 	int i;
430 
431 	for (i = 0; i < entries; i += entries_per_page)
432 		set_aliased_prot(ldt + i, PAGE_KERNEL);
433 }
434 
435 static void xen_set_ldt(const void *addr, unsigned entries)
436 {
437 	struct mmuext_op *op;
438 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
439 
440 	trace_xen_cpu_set_ldt(addr, entries);
441 
442 	op = mcs.args;
443 	op->cmd = MMUEXT_SET_LDT;
444 	op->arg1.linear_addr = (unsigned long)addr;
445 	op->arg2.nr_ents = entries;
446 
447 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
448 
449 	xen_mc_issue(PARAVIRT_LAZY_CPU);
450 }
451 
452 static void xen_load_gdt(const struct desc_ptr *dtr)
453 {
454 	unsigned long va = dtr->address;
455 	unsigned int size = dtr->size + 1;
456 	unsigned long pfn, mfn;
457 	int level;
458 	pte_t *ptep;
459 	void *virt;
460 
461 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
462 	BUG_ON(size > PAGE_SIZE);
463 	BUG_ON(va & ~PAGE_MASK);
464 
465 	/*
466 	 * The GDT is per-cpu and is in the percpu data area.
467 	 * That can be virtually mapped, so we need to do a
468 	 * page-walk to get the underlying MFN for the
469 	 * hypercall.  The page can also be in the kernel's
470 	 * linear range, so we need to RO that mapping too.
471 	 */
472 	ptep = lookup_address(va, &level);
473 	BUG_ON(ptep == NULL);
474 
475 	pfn = pte_pfn(*ptep);
476 	mfn = pfn_to_mfn(pfn);
477 	virt = __va(PFN_PHYS(pfn));
478 
479 	make_lowmem_page_readonly((void *)va);
480 	make_lowmem_page_readonly(virt);
481 
482 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
483 		BUG();
484 }
485 
486 /*
487  * load_gdt for early boot, when the gdt is only mapped once
488  */
489 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
490 {
491 	unsigned long va = dtr->address;
492 	unsigned int size = dtr->size + 1;
493 	unsigned long pfn, mfn;
494 	pte_t pte;
495 
496 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
497 	BUG_ON(size > PAGE_SIZE);
498 	BUG_ON(va & ~PAGE_MASK);
499 
500 	pfn = virt_to_pfn(va);
501 	mfn = pfn_to_mfn(pfn);
502 
503 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
504 
505 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
506 		BUG();
507 
508 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
509 		BUG();
510 }
511 
512 static inline bool desc_equal(const struct desc_struct *d1,
513 			      const struct desc_struct *d2)
514 {
515 	return !memcmp(d1, d2, sizeof(*d1));
516 }
517 
518 static void load_TLS_descriptor(struct thread_struct *t,
519 				unsigned int cpu, unsigned int i)
520 {
521 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
522 	struct desc_struct *gdt;
523 	xmaddr_t maddr;
524 	struct multicall_space mc;
525 
526 	if (desc_equal(shadow, &t->tls_array[i]))
527 		return;
528 
529 	*shadow = t->tls_array[i];
530 
531 	gdt = get_cpu_gdt_rw(cpu);
532 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
533 	mc = __xen_mc_entry(0);
534 
535 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
536 }
537 
538 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
539 {
540 	/*
541 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
542 	 * and lazy gs handling is enabled, it means we're in a
543 	 * context switch, and %gs has just been saved.  This means we
544 	 * can zero it out to prevent faults on exit from the
545 	 * hypervisor if the next process has no %gs.  Either way, it
546 	 * has been saved, and the new value will get loaded properly.
547 	 * This will go away as soon as Xen has been modified to not
548 	 * save/restore %gs for normal hypercalls.
549 	 *
550 	 * On x86_64, this hack is not used for %gs, because gs points
551 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
552 	 * must not zero %gs on x86_64
553 	 *
554 	 * For x86_64, we need to zero %fs, otherwise we may get an
555 	 * exception between the new %fs descriptor being loaded and
556 	 * %fs being effectively cleared at __switch_to().
557 	 */
558 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
559 #ifdef CONFIG_X86_32
560 		lazy_load_gs(0);
561 #else
562 		loadsegment(fs, 0);
563 #endif
564 	}
565 
566 	xen_mc_batch();
567 
568 	load_TLS_descriptor(t, cpu, 0);
569 	load_TLS_descriptor(t, cpu, 1);
570 	load_TLS_descriptor(t, cpu, 2);
571 
572 	xen_mc_issue(PARAVIRT_LAZY_CPU);
573 }
574 
575 #ifdef CONFIG_X86_64
576 static void xen_load_gs_index(unsigned int idx)
577 {
578 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
579 		BUG();
580 }
581 #endif
582 
583 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
584 				const void *ptr)
585 {
586 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
587 	u64 entry = *(u64 *)ptr;
588 
589 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
590 
591 	preempt_disable();
592 
593 	xen_mc_flush();
594 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
595 		BUG();
596 
597 	preempt_enable();
598 }
599 
600 #ifdef CONFIG_X86_64
601 struct trap_array_entry {
602 	void (*orig)(void);
603 	void (*xen)(void);
604 	bool ist_okay;
605 };
606 
607 #define TRAP_ENTRY(func, ist_ok) {			\
608 	.orig		= asm_##func,			\
609 	.xen		= xen_asm_##func,		\
610 	.ist_okay	= ist_ok }
611 
612 #define TRAP_ENTRY_REDIR(func, xenfunc, ist_ok) {	\
613 	.orig		= asm_##func,			\
614 	.xen		= xen_asm_##xenfunc,		\
615 	.ist_okay	= ist_ok }
616 
617 static struct trap_array_entry trap_array[] = {
618 	TRAP_ENTRY_REDIR(exc_debug, exc_xendebug,	true  ),
619 	TRAP_ENTRY(exc_double_fault,			true  ),
620 #ifdef CONFIG_X86_MCE
621 	TRAP_ENTRY(exc_machine_check,			true  ),
622 #endif
623 	TRAP_ENTRY_REDIR(exc_nmi, exc_xennmi,		true  ),
624 	TRAP_ENTRY(exc_int3,				false ),
625 	TRAP_ENTRY(exc_overflow,			false ),
626 #ifdef CONFIG_IA32_EMULATION
627 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
628 #endif
629 	TRAP_ENTRY(exc_page_fault,			false ),
630 	TRAP_ENTRY(exc_divide_error,			false ),
631 	TRAP_ENTRY(exc_bounds,				false ),
632 	TRAP_ENTRY(exc_invalid_op,			false ),
633 	TRAP_ENTRY(exc_device_not_available,		false ),
634 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
635 	TRAP_ENTRY(exc_invalid_tss,			false ),
636 	TRAP_ENTRY(exc_segment_not_present,		false ),
637 	TRAP_ENTRY(exc_stack_segment,			false ),
638 	TRAP_ENTRY(exc_general_protection,		false ),
639 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
640 	TRAP_ENTRY(exc_coprocessor_error,		false ),
641 	TRAP_ENTRY(exc_alignment_check,			false ),
642 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
643 };
644 
645 static bool __ref get_trap_addr(void **addr, unsigned int ist)
646 {
647 	unsigned int nr;
648 	bool ist_okay = false;
649 
650 	/*
651 	 * Replace trap handler addresses by Xen specific ones.
652 	 * Check for known traps using IST and whitelist them.
653 	 * The debugger ones are the only ones we care about.
654 	 * Xen will handle faults like double_fault, so we should never see
655 	 * them.  Warn if there's an unexpected IST-using fault handler.
656 	 */
657 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
658 		struct trap_array_entry *entry = trap_array + nr;
659 
660 		if (*addr == entry->orig) {
661 			*addr = entry->xen;
662 			ist_okay = entry->ist_okay;
663 			break;
664 		}
665 	}
666 
667 	if (nr == ARRAY_SIZE(trap_array) &&
668 	    *addr >= (void *)early_idt_handler_array[0] &&
669 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
670 		nr = (*addr - (void *)early_idt_handler_array[0]) /
671 		     EARLY_IDT_HANDLER_SIZE;
672 		*addr = (void *)xen_early_idt_handler_array[nr];
673 	}
674 
675 	if (WARN_ON(ist != 0 && !ist_okay))
676 		return false;
677 
678 	return true;
679 }
680 #endif
681 
682 static int cvt_gate_to_trap(int vector, const gate_desc *val,
683 			    struct trap_info *info)
684 {
685 	unsigned long addr;
686 
687 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
688 		return 0;
689 
690 	info->vector = vector;
691 
692 	addr = gate_offset(val);
693 #ifdef CONFIG_X86_64
694 	if (!get_trap_addr((void **)&addr, val->bits.ist))
695 		return 0;
696 #endif	/* CONFIG_X86_64 */
697 	info->address = addr;
698 
699 	info->cs = gate_segment(val);
700 	info->flags = val->bits.dpl;
701 	/* interrupt gates clear IF */
702 	if (val->bits.type == GATE_INTERRUPT)
703 		info->flags |= 1 << 2;
704 
705 	return 1;
706 }
707 
708 /* Locations of each CPU's IDT */
709 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
710 
711 /* Set an IDT entry.  If the entry is part of the current IDT, then
712    also update Xen. */
713 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
714 {
715 	unsigned long p = (unsigned long)&dt[entrynum];
716 	unsigned long start, end;
717 
718 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
719 
720 	preempt_disable();
721 
722 	start = __this_cpu_read(idt_desc.address);
723 	end = start + __this_cpu_read(idt_desc.size) + 1;
724 
725 	xen_mc_flush();
726 
727 	native_write_idt_entry(dt, entrynum, g);
728 
729 	if (p >= start && (p + 8) <= end) {
730 		struct trap_info info[2];
731 
732 		info[1].address = 0;
733 
734 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
735 			if (HYPERVISOR_set_trap_table(info))
736 				BUG();
737 	}
738 
739 	preempt_enable();
740 }
741 
742 static void xen_convert_trap_info(const struct desc_ptr *desc,
743 				  struct trap_info *traps)
744 {
745 	unsigned in, out, count;
746 
747 	count = (desc->size+1) / sizeof(gate_desc);
748 	BUG_ON(count > 256);
749 
750 	for (in = out = 0; in < count; in++) {
751 		gate_desc *entry = (gate_desc *)(desc->address) + in;
752 
753 		if (cvt_gate_to_trap(in, entry, &traps[out]))
754 			out++;
755 	}
756 	traps[out].address = 0;
757 }
758 
759 void xen_copy_trap_info(struct trap_info *traps)
760 {
761 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
762 
763 	xen_convert_trap_info(desc, traps);
764 }
765 
766 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
767    hold a spinlock to protect the static traps[] array (static because
768    it avoids allocation, and saves stack space). */
769 static void xen_load_idt(const struct desc_ptr *desc)
770 {
771 	static DEFINE_SPINLOCK(lock);
772 	static struct trap_info traps[257];
773 
774 	trace_xen_cpu_load_idt(desc);
775 
776 	spin_lock(&lock);
777 
778 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
779 
780 	xen_convert_trap_info(desc, traps);
781 
782 	xen_mc_flush();
783 	if (HYPERVISOR_set_trap_table(traps))
784 		BUG();
785 
786 	spin_unlock(&lock);
787 }
788 
789 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
790    they're handled differently. */
791 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
792 				const void *desc, int type)
793 {
794 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
795 
796 	preempt_disable();
797 
798 	switch (type) {
799 	case DESC_LDT:
800 	case DESC_TSS:
801 		/* ignore */
802 		break;
803 
804 	default: {
805 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
806 
807 		xen_mc_flush();
808 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
809 			BUG();
810 	}
811 
812 	}
813 
814 	preempt_enable();
815 }
816 
817 /*
818  * Version of write_gdt_entry for use at early boot-time needed to
819  * update an entry as simply as possible.
820  */
821 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
822 					    const void *desc, int type)
823 {
824 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
825 
826 	switch (type) {
827 	case DESC_LDT:
828 	case DESC_TSS:
829 		/* ignore */
830 		break;
831 
832 	default: {
833 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
834 
835 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
836 			dt[entry] = *(struct desc_struct *)desc;
837 	}
838 
839 	}
840 }
841 
842 static void xen_load_sp0(unsigned long sp0)
843 {
844 	struct multicall_space mcs;
845 
846 	mcs = xen_mc_entry(0);
847 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
848 	xen_mc_issue(PARAVIRT_LAZY_CPU);
849 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
850 }
851 
852 #ifdef CONFIG_X86_IOPL_IOPERM
853 static void xen_update_io_bitmap(void)
854 {
855 	struct physdev_set_iobitmap iobitmap;
856 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
857 
858 	native_tss_update_io_bitmap();
859 
860 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
861 			  tss->x86_tss.io_bitmap_base;
862 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
863 		iobitmap.nr_ports = 0;
864 	else
865 		iobitmap.nr_ports = IO_BITMAP_BITS;
866 
867 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
868 }
869 #endif
870 
871 static void xen_io_delay(void)
872 {
873 }
874 
875 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
876 
877 static unsigned long xen_read_cr0(void)
878 {
879 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
880 
881 	if (unlikely(cr0 == 0)) {
882 		cr0 = native_read_cr0();
883 		this_cpu_write(xen_cr0_value, cr0);
884 	}
885 
886 	return cr0;
887 }
888 
889 static void xen_write_cr0(unsigned long cr0)
890 {
891 	struct multicall_space mcs;
892 
893 	this_cpu_write(xen_cr0_value, cr0);
894 
895 	/* Only pay attention to cr0.TS; everything else is
896 	   ignored. */
897 	mcs = xen_mc_entry(0);
898 
899 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
900 
901 	xen_mc_issue(PARAVIRT_LAZY_CPU);
902 }
903 
904 static void xen_write_cr4(unsigned long cr4)
905 {
906 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
907 
908 	native_write_cr4(cr4);
909 }
910 
911 static u64 xen_read_msr_safe(unsigned int msr, int *err)
912 {
913 	u64 val;
914 
915 	if (pmu_msr_read(msr, &val, err))
916 		return val;
917 
918 	val = native_read_msr_safe(msr, err);
919 	switch (msr) {
920 	case MSR_IA32_APICBASE:
921 		val &= ~X2APIC_ENABLE;
922 		break;
923 	}
924 	return val;
925 }
926 
927 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
928 {
929 	int ret;
930 #ifdef CONFIG_X86_64
931 	unsigned int which;
932 	u64 base;
933 #endif
934 
935 	ret = 0;
936 
937 	switch (msr) {
938 #ifdef CONFIG_X86_64
939 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
940 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
941 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
942 
943 	set:
944 		base = ((u64)high << 32) | low;
945 		if (HYPERVISOR_set_segment_base(which, base) != 0)
946 			ret = -EIO;
947 		break;
948 #endif
949 
950 	case MSR_STAR:
951 	case MSR_CSTAR:
952 	case MSR_LSTAR:
953 	case MSR_SYSCALL_MASK:
954 	case MSR_IA32_SYSENTER_CS:
955 	case MSR_IA32_SYSENTER_ESP:
956 	case MSR_IA32_SYSENTER_EIP:
957 		/* Fast syscall setup is all done in hypercalls, so
958 		   these are all ignored.  Stub them out here to stop
959 		   Xen console noise. */
960 		break;
961 
962 	default:
963 		if (!pmu_msr_write(msr, low, high, &ret))
964 			ret = native_write_msr_safe(msr, low, high);
965 	}
966 
967 	return ret;
968 }
969 
970 static u64 xen_read_msr(unsigned int msr)
971 {
972 	/*
973 	 * This will silently swallow a #GP from RDMSR.  It may be worth
974 	 * changing that.
975 	 */
976 	int err;
977 
978 	return xen_read_msr_safe(msr, &err);
979 }
980 
981 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
982 {
983 	/*
984 	 * This will silently swallow a #GP from WRMSR.  It may be worth
985 	 * changing that.
986 	 */
987 	xen_write_msr_safe(msr, low, high);
988 }
989 
990 /* This is called once we have the cpu_possible_mask */
991 void __init xen_setup_vcpu_info_placement(void)
992 {
993 	int cpu;
994 
995 	for_each_possible_cpu(cpu) {
996 		/* Set up direct vCPU id mapping for PV guests. */
997 		per_cpu(xen_vcpu_id, cpu) = cpu;
998 
999 		/*
1000 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
1001 		 * falls back to the shared_info version for cpus
1002 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1003 		 *
1004 		 * xen_cpu_up_prepare_pv() handles the rest by failing
1005 		 * them in hotplug.
1006 		 */
1007 		(void) xen_vcpu_setup(cpu);
1008 	}
1009 
1010 	/*
1011 	 * xen_vcpu_setup managed to place the vcpu_info within the
1012 	 * percpu area for all cpus, so make use of it.
1013 	 */
1014 	if (xen_have_vcpu_info_placement) {
1015 		pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1016 		pv_ops.irq.restore_fl =
1017 			__PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1018 		pv_ops.irq.irq_disable =
1019 			__PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1020 		pv_ops.irq.irq_enable =
1021 			__PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1022 		pv_ops.mmu.read_cr2 =
1023 			__PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1024 	}
1025 }
1026 
1027 static const struct pv_info xen_info __initconst = {
1028 	.shared_kernel_pmd = 0,
1029 
1030 #ifdef CONFIG_X86_64
1031 	.extra_user_64bit_cs = FLAT_USER_CS64,
1032 #endif
1033 	.name = "Xen",
1034 };
1035 
1036 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1037 	.cpuid = xen_cpuid,
1038 
1039 	.set_debugreg = xen_set_debugreg,
1040 	.get_debugreg = xen_get_debugreg,
1041 
1042 	.read_cr0 = xen_read_cr0,
1043 	.write_cr0 = xen_write_cr0,
1044 
1045 	.write_cr4 = xen_write_cr4,
1046 
1047 	.wbinvd = native_wbinvd,
1048 
1049 	.read_msr = xen_read_msr,
1050 	.write_msr = xen_write_msr,
1051 
1052 	.read_msr_safe = xen_read_msr_safe,
1053 	.write_msr_safe = xen_write_msr_safe,
1054 
1055 	.read_pmc = xen_read_pmc,
1056 
1057 	.iret = xen_iret,
1058 #ifdef CONFIG_X86_64
1059 	.usergs_sysret64 = xen_sysret64,
1060 #endif
1061 
1062 	.load_tr_desc = paravirt_nop,
1063 	.set_ldt = xen_set_ldt,
1064 	.load_gdt = xen_load_gdt,
1065 	.load_idt = xen_load_idt,
1066 	.load_tls = xen_load_tls,
1067 #ifdef CONFIG_X86_64
1068 	.load_gs_index = xen_load_gs_index,
1069 #endif
1070 
1071 	.alloc_ldt = xen_alloc_ldt,
1072 	.free_ldt = xen_free_ldt,
1073 
1074 	.store_tr = xen_store_tr,
1075 
1076 	.write_ldt_entry = xen_write_ldt_entry,
1077 	.write_gdt_entry = xen_write_gdt_entry,
1078 	.write_idt_entry = xen_write_idt_entry,
1079 	.load_sp0 = xen_load_sp0,
1080 
1081 #ifdef CONFIG_X86_IOPL_IOPERM
1082 	.update_io_bitmap = xen_update_io_bitmap,
1083 #endif
1084 	.io_delay = xen_io_delay,
1085 
1086 	/* Xen takes care of %gs when switching to usermode for us */
1087 	.swapgs = paravirt_nop,
1088 
1089 	.start_context_switch = paravirt_start_context_switch,
1090 	.end_context_switch = xen_end_context_switch,
1091 };
1092 
1093 static void xen_restart(char *msg)
1094 {
1095 	xen_reboot(SHUTDOWN_reboot);
1096 }
1097 
1098 static void xen_machine_halt(void)
1099 {
1100 	xen_reboot(SHUTDOWN_poweroff);
1101 }
1102 
1103 static void xen_machine_power_off(void)
1104 {
1105 	if (pm_power_off)
1106 		pm_power_off();
1107 	xen_reboot(SHUTDOWN_poweroff);
1108 }
1109 
1110 static void xen_crash_shutdown(struct pt_regs *regs)
1111 {
1112 	xen_reboot(SHUTDOWN_crash);
1113 }
1114 
1115 static const struct machine_ops xen_machine_ops __initconst = {
1116 	.restart = xen_restart,
1117 	.halt = xen_machine_halt,
1118 	.power_off = xen_machine_power_off,
1119 	.shutdown = xen_machine_halt,
1120 	.crash_shutdown = xen_crash_shutdown,
1121 	.emergency_restart = xen_emergency_restart,
1122 };
1123 
1124 static unsigned char xen_get_nmi_reason(void)
1125 {
1126 	unsigned char reason = 0;
1127 
1128 	/* Construct a value which looks like it came from port 0x61. */
1129 	if (test_bit(_XEN_NMIREASON_io_error,
1130 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1131 		reason |= NMI_REASON_IOCHK;
1132 	if (test_bit(_XEN_NMIREASON_pci_serr,
1133 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1134 		reason |= NMI_REASON_SERR;
1135 
1136 	return reason;
1137 }
1138 
1139 static void __init xen_boot_params_init_edd(void)
1140 {
1141 #if IS_ENABLED(CONFIG_EDD)
1142 	struct xen_platform_op op;
1143 	struct edd_info *edd_info;
1144 	u32 *mbr_signature;
1145 	unsigned nr;
1146 	int ret;
1147 
1148 	edd_info = boot_params.eddbuf;
1149 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1150 
1151 	op.cmd = XENPF_firmware_info;
1152 
1153 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1154 	for (nr = 0; nr < EDDMAXNR; nr++) {
1155 		struct edd_info *info = edd_info + nr;
1156 
1157 		op.u.firmware_info.index = nr;
1158 		info->params.length = sizeof(info->params);
1159 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1160 				     &info->params);
1161 		ret = HYPERVISOR_platform_op(&op);
1162 		if (ret)
1163 			break;
1164 
1165 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1166 		C(device);
1167 		C(version);
1168 		C(interface_support);
1169 		C(legacy_max_cylinder);
1170 		C(legacy_max_head);
1171 		C(legacy_sectors_per_track);
1172 #undef C
1173 	}
1174 	boot_params.eddbuf_entries = nr;
1175 
1176 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1177 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1178 		op.u.firmware_info.index = nr;
1179 		ret = HYPERVISOR_platform_op(&op);
1180 		if (ret)
1181 			break;
1182 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1183 	}
1184 	boot_params.edd_mbr_sig_buf_entries = nr;
1185 #endif
1186 }
1187 
1188 /*
1189  * Set up the GDT and segment registers for -fstack-protector.  Until
1190  * we do this, we have to be careful not to call any stack-protected
1191  * function, which is most of the kernel.
1192  */
1193 static void __init xen_setup_gdt(int cpu)
1194 {
1195 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1196 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1197 
1198 	setup_stack_canary_segment(cpu);
1199 	switch_to_new_gdt(cpu);
1200 
1201 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1202 	pv_ops.cpu.load_gdt = xen_load_gdt;
1203 }
1204 
1205 static void __init xen_dom0_set_legacy_features(void)
1206 {
1207 	x86_platform.legacy.rtc = 1;
1208 }
1209 
1210 /* First C function to be called on Xen boot */
1211 asmlinkage __visible void __init xen_start_kernel(void)
1212 {
1213 	struct physdev_set_iopl set_iopl;
1214 	unsigned long initrd_start = 0;
1215 	int rc;
1216 
1217 	if (!xen_start_info)
1218 		return;
1219 
1220 	xen_domain_type = XEN_PV_DOMAIN;
1221 	xen_start_flags = xen_start_info->flags;
1222 
1223 	xen_setup_features();
1224 
1225 	/* Install Xen paravirt ops */
1226 	pv_info = xen_info;
1227 	pv_ops.init.patch = paravirt_patch_default;
1228 	pv_ops.cpu = xen_cpu_ops;
1229 	xen_init_irq_ops();
1230 
1231 	/*
1232 	 * Setup xen_vcpu early because it is needed for
1233 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1234 	 *
1235 	 * Don't do the full vcpu_info placement stuff until we have
1236 	 * the cpu_possible_mask and a non-dummy shared_info.
1237 	 */
1238 	xen_vcpu_info_reset(0);
1239 
1240 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1241 
1242 	x86_init.resources.memory_setup = xen_memory_setup;
1243 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1244 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1245 	x86_init.oem.arch_setup = xen_arch_setup;
1246 	x86_init.oem.banner = xen_banner;
1247 	x86_init.hyper.init_platform = xen_pv_init_platform;
1248 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1249 
1250 	/*
1251 	 * Set up some pagetable state before starting to set any ptes.
1252 	 */
1253 
1254 	xen_setup_machphys_mapping();
1255 	xen_init_mmu_ops();
1256 
1257 	/* Prevent unwanted bits from being set in PTEs. */
1258 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1259 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1260 
1261 	/*
1262 	 * Prevent page tables from being allocated in highmem, even
1263 	 * if CONFIG_HIGHPTE is enabled.
1264 	 */
1265 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1266 
1267 	/* Get mfn list */
1268 	xen_build_dynamic_phys_to_machine();
1269 
1270 	/*
1271 	 * Set up kernel GDT and segment registers, mainly so that
1272 	 * -fstack-protector code can be executed.
1273 	 */
1274 	xen_setup_gdt(0);
1275 
1276 	/* Work out if we support NX */
1277 	get_cpu_cap(&boot_cpu_data);
1278 	x86_configure_nx();
1279 
1280 	/* Determine virtual and physical address sizes */
1281 	get_cpu_address_sizes(&boot_cpu_data);
1282 
1283 	/* Let's presume PV guests always boot on vCPU with id 0. */
1284 	per_cpu(xen_vcpu_id, 0) = 0;
1285 
1286 	idt_setup_early_handler();
1287 
1288 	xen_init_capabilities();
1289 
1290 #ifdef CONFIG_X86_LOCAL_APIC
1291 	/*
1292 	 * set up the basic apic ops.
1293 	 */
1294 	xen_init_apic();
1295 #endif
1296 
1297 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1298 		pv_ops.mmu.ptep_modify_prot_start =
1299 			xen_ptep_modify_prot_start;
1300 		pv_ops.mmu.ptep_modify_prot_commit =
1301 			xen_ptep_modify_prot_commit;
1302 	}
1303 
1304 	machine_ops = xen_machine_ops;
1305 
1306 	/*
1307 	 * The only reliable way to retain the initial address of the
1308 	 * percpu gdt_page is to remember it here, so we can go and
1309 	 * mark it RW later, when the initial percpu area is freed.
1310 	 */
1311 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1312 
1313 	xen_smp_init();
1314 
1315 #ifdef CONFIG_ACPI_NUMA
1316 	/*
1317 	 * The pages we from Xen are not related to machine pages, so
1318 	 * any NUMA information the kernel tries to get from ACPI will
1319 	 * be meaningless.  Prevent it from trying.
1320 	 */
1321 	acpi_numa = -1;
1322 #endif
1323 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1324 
1325 	local_irq_disable();
1326 	early_boot_irqs_disabled = true;
1327 
1328 	xen_raw_console_write("mapping kernel into physical memory\n");
1329 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1330 				   xen_start_info->nr_pages);
1331 	xen_reserve_special_pages();
1332 
1333 	/* keep using Xen gdt for now; no urgent need to change it */
1334 
1335 #ifdef CONFIG_X86_32
1336 	pv_info.kernel_rpl = 1;
1337 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1338 		pv_info.kernel_rpl = 0;
1339 #else
1340 	pv_info.kernel_rpl = 0;
1341 #endif
1342 	/* set the limit of our address space */
1343 	xen_reserve_top();
1344 
1345 	/*
1346 	 * We used to do this in xen_arch_setup, but that is too late
1347 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1348 	 * early_amd_init which pokes 0xcf8 port.
1349 	 */
1350 	set_iopl.iopl = 1;
1351 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1352 	if (rc != 0)
1353 		xen_raw_printk("physdev_op failed %d\n", rc);
1354 
1355 #ifdef CONFIG_X86_32
1356 	/* set up basic CPUID stuff */
1357 	cpu_detect(&new_cpu_data);
1358 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1359 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1360 #endif
1361 
1362 	if (xen_start_info->mod_start) {
1363 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1364 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1365 	    else
1366 		initrd_start = __pa(xen_start_info->mod_start);
1367 	}
1368 
1369 	/* Poke various useful things into boot_params */
1370 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1371 	boot_params.hdr.ramdisk_image = initrd_start;
1372 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1373 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1374 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1375 
1376 	if (!xen_initial_domain()) {
1377 		add_preferred_console("xenboot", 0, NULL);
1378 		if (pci_xen)
1379 			x86_init.pci.arch_init = pci_xen_init;
1380 	} else {
1381 		const struct dom0_vga_console_info *info =
1382 			(void *)((char *)xen_start_info +
1383 				 xen_start_info->console.dom0.info_off);
1384 		struct xen_platform_op op = {
1385 			.cmd = XENPF_firmware_info,
1386 			.interface_version = XENPF_INTERFACE_VERSION,
1387 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1388 		};
1389 
1390 		x86_platform.set_legacy_features =
1391 				xen_dom0_set_legacy_features;
1392 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1393 		xen_start_info->console.domU.mfn = 0;
1394 		xen_start_info->console.domU.evtchn = 0;
1395 
1396 		if (HYPERVISOR_platform_op(&op) == 0)
1397 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1398 
1399 		/* Make sure ACS will be enabled */
1400 		pci_request_acs();
1401 
1402 		xen_acpi_sleep_register();
1403 
1404 		/* Avoid searching for BIOS MP tables */
1405 		x86_init.mpparse.find_smp_config = x86_init_noop;
1406 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1407 
1408 		xen_boot_params_init_edd();
1409 	}
1410 
1411 	if (!boot_params.screen_info.orig_video_isVGA)
1412 		add_preferred_console("tty", 0, NULL);
1413 	add_preferred_console("hvc", 0, NULL);
1414 	if (boot_params.screen_info.orig_video_isVGA)
1415 		add_preferred_console("tty", 0, NULL);
1416 
1417 #ifdef CONFIG_PCI
1418 	/* PCI BIOS service won't work from a PV guest. */
1419 	pci_probe &= ~PCI_PROBE_BIOS;
1420 #endif
1421 	xen_raw_console_write("about to get started...\n");
1422 
1423 	/* We need this for printk timestamps */
1424 	xen_setup_runstate_info(0);
1425 
1426 	xen_efi_init(&boot_params);
1427 
1428 	/* Start the world */
1429 #ifdef CONFIG_X86_32
1430 	i386_start_kernel();
1431 #else
1432 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1433 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1434 #endif
1435 }
1436 
1437 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1438 {
1439 	int rc;
1440 
1441 	if (per_cpu(xen_vcpu, cpu) == NULL)
1442 		return -ENODEV;
1443 
1444 	xen_setup_timer(cpu);
1445 
1446 	rc = xen_smp_intr_init(cpu);
1447 	if (rc) {
1448 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1449 		     cpu, rc);
1450 		return rc;
1451 	}
1452 
1453 	rc = xen_smp_intr_init_pv(cpu);
1454 	if (rc) {
1455 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1456 		     cpu, rc);
1457 		return rc;
1458 	}
1459 
1460 	return 0;
1461 }
1462 
1463 static int xen_cpu_dead_pv(unsigned int cpu)
1464 {
1465 	xen_smp_intr_free(cpu);
1466 	xen_smp_intr_free_pv(cpu);
1467 
1468 	xen_teardown_timer(cpu);
1469 
1470 	return 0;
1471 }
1472 
1473 static uint32_t __init xen_platform_pv(void)
1474 {
1475 	if (xen_pv_domain())
1476 		return xen_cpuid_base();
1477 
1478 	return 0;
1479 }
1480 
1481 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1482 	.name                   = "Xen PV",
1483 	.detect                 = xen_platform_pv,
1484 	.type			= X86_HYPER_XEN_PV,
1485 	.runtime.pin_vcpu       = xen_pin_vcpu,
1486 	.ignore_nopv		= true,
1487 };
1488