xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision d9679d00)
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/pci.h>
31 #include <linux/gfp.h>
32 #include <linux/edd.h>
33 #include <linux/objtool.h>
34 
35 #include <xen/xen.h>
36 #include <xen/events.h>
37 #include <xen/interface/xen.h>
38 #include <xen/interface/version.h>
39 #include <xen/interface/physdev.h>
40 #include <xen/interface/vcpu.h>
41 #include <xen/interface/memory.h>
42 #include <xen/interface/nmi.h>
43 #include <xen/interface/xen-mca.h>
44 #include <xen/features.h>
45 #include <xen/page.h>
46 #include <xen/hvc-console.h>
47 #include <xen/acpi.h>
48 
49 #include <asm/paravirt.h>
50 #include <asm/apic.h>
51 #include <asm/page.h>
52 #include <asm/xen/pci.h>
53 #include <asm/xen/hypercall.h>
54 #include <asm/xen/hypervisor.h>
55 #include <asm/xen/cpuid.h>
56 #include <asm/fixmap.h>
57 #include <asm/processor.h>
58 #include <asm/proto.h>
59 #include <asm/msr-index.h>
60 #include <asm/traps.h>
61 #include <asm/setup.h>
62 #include <asm/desc.h>
63 #include <asm/pgalloc.h>
64 #include <asm/tlbflush.h>
65 #include <asm/reboot.h>
66 #include <asm/stackprotector.h>
67 #include <asm/hypervisor.h>
68 #include <asm/mach_traps.h>
69 #include <asm/mwait.h>
70 #include <asm/pci_x86.h>
71 #include <asm/cpu.h>
72 #ifdef CONFIG_X86_IOPL_IOPERM
73 #include <asm/io_bitmap.h>
74 #endif
75 
76 #ifdef CONFIG_ACPI
77 #include <linux/acpi.h>
78 #include <asm/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
82 #endif
83 
84 #include "xen-ops.h"
85 #include "mmu.h"
86 #include "smp.h"
87 #include "multicalls.h"
88 #include "pmu.h"
89 
90 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
91 
92 void *xen_initial_gdt;
93 
94 static int xen_cpu_up_prepare_pv(unsigned int cpu);
95 static int xen_cpu_dead_pv(unsigned int cpu);
96 
97 struct tls_descs {
98 	struct desc_struct desc[3];
99 };
100 
101 /*
102  * Updating the 3 TLS descriptors in the GDT on every task switch is
103  * surprisingly expensive so we avoid updating them if they haven't
104  * changed.  Since Xen writes different descriptors than the one
105  * passed in the update_descriptor hypercall we keep shadow copies to
106  * compare against.
107  */
108 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
109 
110 static void __init xen_pv_init_platform(void)
111 {
112 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
113 
114 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
115 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
116 
117 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
118 	xen_vcpu_info_reset(0);
119 
120 	/* pvclock is in shared info area */
121 	xen_init_time_ops();
122 }
123 
124 static void __init xen_pv_guest_late_init(void)
125 {
126 #ifndef CONFIG_SMP
127 	/* Setup shared vcpu info for non-smp configurations */
128 	xen_setup_vcpu_info_placement();
129 #endif
130 }
131 
132 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
133 static __read_mostly unsigned int cpuid_leaf5_edx_val;
134 
135 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
136 		      unsigned int *cx, unsigned int *dx)
137 {
138 	unsigned maskebx = ~0;
139 
140 	/*
141 	 * Mask out inconvenient features, to try and disable as many
142 	 * unsupported kernel subsystems as possible.
143 	 */
144 	switch (*ax) {
145 	case CPUID_MWAIT_LEAF:
146 		/* Synthesize the values.. */
147 		*ax = 0;
148 		*bx = 0;
149 		*cx = cpuid_leaf5_ecx_val;
150 		*dx = cpuid_leaf5_edx_val;
151 		return;
152 
153 	case 0xb:
154 		/* Suppress extended topology stuff */
155 		maskebx = 0;
156 		break;
157 	}
158 
159 	asm(XEN_EMULATE_PREFIX "cpuid"
160 		: "=a" (*ax),
161 		  "=b" (*bx),
162 		  "=c" (*cx),
163 		  "=d" (*dx)
164 		: "0" (*ax), "2" (*cx));
165 
166 	*bx &= maskebx;
167 }
168 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
169 
170 static bool __init xen_check_mwait(void)
171 {
172 #ifdef CONFIG_ACPI
173 	struct xen_platform_op op = {
174 		.cmd			= XENPF_set_processor_pminfo,
175 		.u.set_pminfo.id	= -1,
176 		.u.set_pminfo.type	= XEN_PM_PDC,
177 	};
178 	uint32_t buf[3];
179 	unsigned int ax, bx, cx, dx;
180 	unsigned int mwait_mask;
181 
182 	/* We need to determine whether it is OK to expose the MWAIT
183 	 * capability to the kernel to harvest deeper than C3 states from ACPI
184 	 * _CST using the processor_harvest_xen.c module. For this to work, we
185 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
186 	 * checks against). The hypervisor won't expose the MWAIT flag because
187 	 * it would break backwards compatibility; so we will find out directly
188 	 * from the hardware and hypercall.
189 	 */
190 	if (!xen_initial_domain())
191 		return false;
192 
193 	/*
194 	 * When running under platform earlier than Xen4.2, do not expose
195 	 * mwait, to avoid the risk of loading native acpi pad driver
196 	 */
197 	if (!xen_running_on_version_or_later(4, 2))
198 		return false;
199 
200 	ax = 1;
201 	cx = 0;
202 
203 	native_cpuid(&ax, &bx, &cx, &dx);
204 
205 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
206 		     (1 << (X86_FEATURE_MWAIT % 32));
207 
208 	if ((cx & mwait_mask) != mwait_mask)
209 		return false;
210 
211 	/* We need to emulate the MWAIT_LEAF and for that we need both
212 	 * ecx and edx. The hypercall provides only partial information.
213 	 */
214 
215 	ax = CPUID_MWAIT_LEAF;
216 	bx = 0;
217 	cx = 0;
218 	dx = 0;
219 
220 	native_cpuid(&ax, &bx, &cx, &dx);
221 
222 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
223 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
224 	 */
225 	buf[0] = ACPI_PDC_REVISION_ID;
226 	buf[1] = 1;
227 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
228 
229 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
230 
231 	if ((HYPERVISOR_platform_op(&op) == 0) &&
232 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
233 		cpuid_leaf5_ecx_val = cx;
234 		cpuid_leaf5_edx_val = dx;
235 	}
236 	return true;
237 #else
238 	return false;
239 #endif
240 }
241 
242 static bool __init xen_check_xsave(void)
243 {
244 	unsigned int cx, xsave_mask;
245 
246 	cx = cpuid_ecx(1);
247 
248 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
249 		     (1 << (X86_FEATURE_OSXSAVE % 32));
250 
251 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
252 	return (cx & xsave_mask) == xsave_mask;
253 }
254 
255 static void __init xen_init_capabilities(void)
256 {
257 	setup_force_cpu_cap(X86_FEATURE_XENPV);
258 	setup_clear_cpu_cap(X86_FEATURE_DCA);
259 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
260 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
261 	setup_clear_cpu_cap(X86_FEATURE_ACC);
262 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
263 	setup_clear_cpu_cap(X86_FEATURE_SME);
264 
265 	/*
266 	 * Xen PV would need some work to support PCID: CR3 handling as well
267 	 * as xen_flush_tlb_others() would need updating.
268 	 */
269 	setup_clear_cpu_cap(X86_FEATURE_PCID);
270 
271 	if (!xen_initial_domain())
272 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
273 
274 	if (xen_check_mwait())
275 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
276 	else
277 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
278 
279 	if (!xen_check_xsave()) {
280 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
281 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
282 	}
283 }
284 
285 static noinstr void xen_set_debugreg(int reg, unsigned long val)
286 {
287 	HYPERVISOR_set_debugreg(reg, val);
288 }
289 
290 static noinstr unsigned long xen_get_debugreg(int reg)
291 {
292 	return HYPERVISOR_get_debugreg(reg);
293 }
294 
295 static void xen_end_context_switch(struct task_struct *next)
296 {
297 	xen_mc_flush();
298 	paravirt_end_context_switch(next);
299 }
300 
301 static unsigned long xen_store_tr(void)
302 {
303 	return 0;
304 }
305 
306 /*
307  * Set the page permissions for a particular virtual address.  If the
308  * address is a vmalloc mapping (or other non-linear mapping), then
309  * find the linear mapping of the page and also set its protections to
310  * match.
311  */
312 static void set_aliased_prot(void *v, pgprot_t prot)
313 {
314 	int level;
315 	pte_t *ptep;
316 	pte_t pte;
317 	unsigned long pfn;
318 	unsigned char dummy;
319 	void *va;
320 
321 	ptep = lookup_address((unsigned long)v, &level);
322 	BUG_ON(ptep == NULL);
323 
324 	pfn = pte_pfn(*ptep);
325 	pte = pfn_pte(pfn, prot);
326 
327 	/*
328 	 * Careful: update_va_mapping() will fail if the virtual address
329 	 * we're poking isn't populated in the page tables.  We don't
330 	 * need to worry about the direct map (that's always in the page
331 	 * tables), but we need to be careful about vmap space.  In
332 	 * particular, the top level page table can lazily propagate
333 	 * entries between processes, so if we've switched mms since we
334 	 * vmapped the target in the first place, we might not have the
335 	 * top-level page table entry populated.
336 	 *
337 	 * We disable preemption because we want the same mm active when
338 	 * we probe the target and when we issue the hypercall.  We'll
339 	 * have the same nominal mm, but if we're a kernel thread, lazy
340 	 * mm dropping could change our pgd.
341 	 *
342 	 * Out of an abundance of caution, this uses __get_user() to fault
343 	 * in the target address just in case there's some obscure case
344 	 * in which the target address isn't readable.
345 	 */
346 
347 	preempt_disable();
348 
349 	copy_from_kernel_nofault(&dummy, v, 1);
350 
351 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
352 		BUG();
353 
354 	va = __va(PFN_PHYS(pfn));
355 
356 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
357 		BUG();
358 
359 	preempt_enable();
360 }
361 
362 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
363 {
364 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
365 	int i;
366 
367 	/*
368 	 * We need to mark the all aliases of the LDT pages RO.  We
369 	 * don't need to call vm_flush_aliases(), though, since that's
370 	 * only responsible for flushing aliases out the TLBs, not the
371 	 * page tables, and Xen will flush the TLB for us if needed.
372 	 *
373 	 * To avoid confusing future readers: none of this is necessary
374 	 * to load the LDT.  The hypervisor only checks this when the
375 	 * LDT is faulted in due to subsequent descriptor access.
376 	 */
377 
378 	for (i = 0; i < entries; i += entries_per_page)
379 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
380 }
381 
382 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
383 {
384 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
385 	int i;
386 
387 	for (i = 0; i < entries; i += entries_per_page)
388 		set_aliased_prot(ldt + i, PAGE_KERNEL);
389 }
390 
391 static void xen_set_ldt(const void *addr, unsigned entries)
392 {
393 	struct mmuext_op *op;
394 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
395 
396 	trace_xen_cpu_set_ldt(addr, entries);
397 
398 	op = mcs.args;
399 	op->cmd = MMUEXT_SET_LDT;
400 	op->arg1.linear_addr = (unsigned long)addr;
401 	op->arg2.nr_ents = entries;
402 
403 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
404 
405 	xen_mc_issue(PARAVIRT_LAZY_CPU);
406 }
407 
408 static void xen_load_gdt(const struct desc_ptr *dtr)
409 {
410 	unsigned long va = dtr->address;
411 	unsigned int size = dtr->size + 1;
412 	unsigned long pfn, mfn;
413 	int level;
414 	pte_t *ptep;
415 	void *virt;
416 
417 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
418 	BUG_ON(size > PAGE_SIZE);
419 	BUG_ON(va & ~PAGE_MASK);
420 
421 	/*
422 	 * The GDT is per-cpu and is in the percpu data area.
423 	 * That can be virtually mapped, so we need to do a
424 	 * page-walk to get the underlying MFN for the
425 	 * hypercall.  The page can also be in the kernel's
426 	 * linear range, so we need to RO that mapping too.
427 	 */
428 	ptep = lookup_address(va, &level);
429 	BUG_ON(ptep == NULL);
430 
431 	pfn = pte_pfn(*ptep);
432 	mfn = pfn_to_mfn(pfn);
433 	virt = __va(PFN_PHYS(pfn));
434 
435 	make_lowmem_page_readonly((void *)va);
436 	make_lowmem_page_readonly(virt);
437 
438 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
439 		BUG();
440 }
441 
442 /*
443  * load_gdt for early boot, when the gdt is only mapped once
444  */
445 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
446 {
447 	unsigned long va = dtr->address;
448 	unsigned int size = dtr->size + 1;
449 	unsigned long pfn, mfn;
450 	pte_t pte;
451 
452 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
453 	BUG_ON(size > PAGE_SIZE);
454 	BUG_ON(va & ~PAGE_MASK);
455 
456 	pfn = virt_to_pfn(va);
457 	mfn = pfn_to_mfn(pfn);
458 
459 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
460 
461 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
462 		BUG();
463 
464 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
465 		BUG();
466 }
467 
468 static inline bool desc_equal(const struct desc_struct *d1,
469 			      const struct desc_struct *d2)
470 {
471 	return !memcmp(d1, d2, sizeof(*d1));
472 }
473 
474 static void load_TLS_descriptor(struct thread_struct *t,
475 				unsigned int cpu, unsigned int i)
476 {
477 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
478 	struct desc_struct *gdt;
479 	xmaddr_t maddr;
480 	struct multicall_space mc;
481 
482 	if (desc_equal(shadow, &t->tls_array[i]))
483 		return;
484 
485 	*shadow = t->tls_array[i];
486 
487 	gdt = get_cpu_gdt_rw(cpu);
488 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
489 	mc = __xen_mc_entry(0);
490 
491 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
492 }
493 
494 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
495 {
496 	/*
497 	 * In lazy mode we need to zero %fs, otherwise we may get an
498 	 * exception between the new %fs descriptor being loaded and
499 	 * %fs being effectively cleared at __switch_to().
500 	 */
501 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
502 		loadsegment(fs, 0);
503 
504 	xen_mc_batch();
505 
506 	load_TLS_descriptor(t, cpu, 0);
507 	load_TLS_descriptor(t, cpu, 1);
508 	load_TLS_descriptor(t, cpu, 2);
509 
510 	xen_mc_issue(PARAVIRT_LAZY_CPU);
511 }
512 
513 static void xen_load_gs_index(unsigned int idx)
514 {
515 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
516 		BUG();
517 }
518 
519 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
520 				const void *ptr)
521 {
522 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
523 	u64 entry = *(u64 *)ptr;
524 
525 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
526 
527 	preempt_disable();
528 
529 	xen_mc_flush();
530 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
531 		BUG();
532 
533 	preempt_enable();
534 }
535 
536 void noist_exc_debug(struct pt_regs *regs);
537 
538 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
539 {
540 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
541 	exc_nmi(regs);
542 }
543 
544 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
545 {
546 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
547 	exc_double_fault(regs, error_code);
548 }
549 
550 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
551 {
552 	/*
553 	 * There's no IST on Xen PV, but we still need to dispatch
554 	 * to the correct handler.
555 	 */
556 	if (user_mode(regs))
557 		noist_exc_debug(regs);
558 	else
559 		exc_debug(regs);
560 }
561 
562 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
563 {
564 	/* This should never happen and there is no way to handle it. */
565 	instrumentation_begin();
566 	pr_err("Unknown trap in Xen PV mode.");
567 	BUG();
568 	instrumentation_end();
569 }
570 
571 #ifdef CONFIG_X86_MCE
572 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
573 {
574 	/*
575 	 * There's no IST on Xen PV, but we still need to dispatch
576 	 * to the correct handler.
577 	 */
578 	if (user_mode(regs))
579 		noist_exc_machine_check(regs);
580 	else
581 		exc_machine_check(regs);
582 }
583 #endif
584 
585 struct trap_array_entry {
586 	void (*orig)(void);
587 	void (*xen)(void);
588 	bool ist_okay;
589 };
590 
591 #define TRAP_ENTRY(func, ist_ok) {			\
592 	.orig		= asm_##func,			\
593 	.xen		= xen_asm_##func,		\
594 	.ist_okay	= ist_ok }
595 
596 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
597 	.orig		= asm_##func,			\
598 	.xen		= xen_asm_xenpv_##func,		\
599 	.ist_okay	= ist_ok }
600 
601 static struct trap_array_entry trap_array[] = {
602 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
603 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
604 #ifdef CONFIG_X86_MCE
605 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
606 #endif
607 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
608 	TRAP_ENTRY(exc_int3,				false ),
609 	TRAP_ENTRY(exc_overflow,			false ),
610 #ifdef CONFIG_IA32_EMULATION
611 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
612 #endif
613 	TRAP_ENTRY(exc_page_fault,			false ),
614 	TRAP_ENTRY(exc_divide_error,			false ),
615 	TRAP_ENTRY(exc_bounds,				false ),
616 	TRAP_ENTRY(exc_invalid_op,			false ),
617 	TRAP_ENTRY(exc_device_not_available,		false ),
618 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
619 	TRAP_ENTRY(exc_invalid_tss,			false ),
620 	TRAP_ENTRY(exc_segment_not_present,		false ),
621 	TRAP_ENTRY(exc_stack_segment,			false ),
622 	TRAP_ENTRY(exc_general_protection,		false ),
623 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
624 	TRAP_ENTRY(exc_coprocessor_error,		false ),
625 	TRAP_ENTRY(exc_alignment_check,			false ),
626 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
627 };
628 
629 static bool __ref get_trap_addr(void **addr, unsigned int ist)
630 {
631 	unsigned int nr;
632 	bool ist_okay = false;
633 	bool found = false;
634 
635 	/*
636 	 * Replace trap handler addresses by Xen specific ones.
637 	 * Check for known traps using IST and whitelist them.
638 	 * The debugger ones are the only ones we care about.
639 	 * Xen will handle faults like double_fault, so we should never see
640 	 * them.  Warn if there's an unexpected IST-using fault handler.
641 	 */
642 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
643 		struct trap_array_entry *entry = trap_array + nr;
644 
645 		if (*addr == entry->orig) {
646 			*addr = entry->xen;
647 			ist_okay = entry->ist_okay;
648 			found = true;
649 			break;
650 		}
651 	}
652 
653 	if (nr == ARRAY_SIZE(trap_array) &&
654 	    *addr >= (void *)early_idt_handler_array[0] &&
655 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
656 		nr = (*addr - (void *)early_idt_handler_array[0]) /
657 		     EARLY_IDT_HANDLER_SIZE;
658 		*addr = (void *)xen_early_idt_handler_array[nr];
659 		found = true;
660 	}
661 
662 	if (!found)
663 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
664 
665 	if (WARN_ON(found && ist != 0 && !ist_okay))
666 		return false;
667 
668 	return true;
669 }
670 
671 static int cvt_gate_to_trap(int vector, const gate_desc *val,
672 			    struct trap_info *info)
673 {
674 	unsigned long addr;
675 
676 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
677 		return 0;
678 
679 	info->vector = vector;
680 
681 	addr = gate_offset(val);
682 	if (!get_trap_addr((void **)&addr, val->bits.ist))
683 		return 0;
684 	info->address = addr;
685 
686 	info->cs = gate_segment(val);
687 	info->flags = val->bits.dpl;
688 	/* interrupt gates clear IF */
689 	if (val->bits.type == GATE_INTERRUPT)
690 		info->flags |= 1 << 2;
691 
692 	return 1;
693 }
694 
695 /* Locations of each CPU's IDT */
696 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
697 
698 /* Set an IDT entry.  If the entry is part of the current IDT, then
699    also update Xen. */
700 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
701 {
702 	unsigned long p = (unsigned long)&dt[entrynum];
703 	unsigned long start, end;
704 
705 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
706 
707 	preempt_disable();
708 
709 	start = __this_cpu_read(idt_desc.address);
710 	end = start + __this_cpu_read(idt_desc.size) + 1;
711 
712 	xen_mc_flush();
713 
714 	native_write_idt_entry(dt, entrynum, g);
715 
716 	if (p >= start && (p + 8) <= end) {
717 		struct trap_info info[2];
718 
719 		info[1].address = 0;
720 
721 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
722 			if (HYPERVISOR_set_trap_table(info))
723 				BUG();
724 	}
725 
726 	preempt_enable();
727 }
728 
729 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
730 				      struct trap_info *traps, bool full)
731 {
732 	unsigned in, out, count;
733 
734 	count = (desc->size+1) / sizeof(gate_desc);
735 	BUG_ON(count > 256);
736 
737 	for (in = out = 0; in < count; in++) {
738 		gate_desc *entry = (gate_desc *)(desc->address) + in;
739 
740 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
741 			out++;
742 	}
743 
744 	return out;
745 }
746 
747 void xen_copy_trap_info(struct trap_info *traps)
748 {
749 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
750 
751 	xen_convert_trap_info(desc, traps, true);
752 }
753 
754 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
755    hold a spinlock to protect the static traps[] array (static because
756    it avoids allocation, and saves stack space). */
757 static void xen_load_idt(const struct desc_ptr *desc)
758 {
759 	static DEFINE_SPINLOCK(lock);
760 	static struct trap_info traps[257];
761 	unsigned out;
762 
763 	trace_xen_cpu_load_idt(desc);
764 
765 	spin_lock(&lock);
766 
767 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
768 
769 	out = xen_convert_trap_info(desc, traps, false);
770 	memset(&traps[out], 0, sizeof(traps[0]));
771 
772 	xen_mc_flush();
773 	if (HYPERVISOR_set_trap_table(traps))
774 		BUG();
775 
776 	spin_unlock(&lock);
777 }
778 
779 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
780    they're handled differently. */
781 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
782 				const void *desc, int type)
783 {
784 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
785 
786 	preempt_disable();
787 
788 	switch (type) {
789 	case DESC_LDT:
790 	case DESC_TSS:
791 		/* ignore */
792 		break;
793 
794 	default: {
795 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
796 
797 		xen_mc_flush();
798 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
799 			BUG();
800 	}
801 
802 	}
803 
804 	preempt_enable();
805 }
806 
807 /*
808  * Version of write_gdt_entry for use at early boot-time needed to
809  * update an entry as simply as possible.
810  */
811 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
812 					    const void *desc, int type)
813 {
814 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
815 
816 	switch (type) {
817 	case DESC_LDT:
818 	case DESC_TSS:
819 		/* ignore */
820 		break;
821 
822 	default: {
823 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
824 
825 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
826 			dt[entry] = *(struct desc_struct *)desc;
827 	}
828 
829 	}
830 }
831 
832 static void xen_load_sp0(unsigned long sp0)
833 {
834 	struct multicall_space mcs;
835 
836 	mcs = xen_mc_entry(0);
837 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
838 	xen_mc_issue(PARAVIRT_LAZY_CPU);
839 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
840 }
841 
842 #ifdef CONFIG_X86_IOPL_IOPERM
843 static void xen_invalidate_io_bitmap(void)
844 {
845 	struct physdev_set_iobitmap iobitmap = {
846 		.bitmap = NULL,
847 		.nr_ports = 0,
848 	};
849 
850 	native_tss_invalidate_io_bitmap();
851 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
852 }
853 
854 static void xen_update_io_bitmap(void)
855 {
856 	struct physdev_set_iobitmap iobitmap;
857 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
858 
859 	native_tss_update_io_bitmap();
860 
861 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
862 			  tss->x86_tss.io_bitmap_base;
863 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
864 		iobitmap.nr_ports = 0;
865 	else
866 		iobitmap.nr_ports = IO_BITMAP_BITS;
867 
868 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
869 }
870 #endif
871 
872 static void xen_io_delay(void)
873 {
874 }
875 
876 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
877 
878 static unsigned long xen_read_cr0(void)
879 {
880 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
881 
882 	if (unlikely(cr0 == 0)) {
883 		cr0 = native_read_cr0();
884 		this_cpu_write(xen_cr0_value, cr0);
885 	}
886 
887 	return cr0;
888 }
889 
890 static void xen_write_cr0(unsigned long cr0)
891 {
892 	struct multicall_space mcs;
893 
894 	this_cpu_write(xen_cr0_value, cr0);
895 
896 	/* Only pay attention to cr0.TS; everything else is
897 	   ignored. */
898 	mcs = xen_mc_entry(0);
899 
900 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
901 
902 	xen_mc_issue(PARAVIRT_LAZY_CPU);
903 }
904 
905 static void xen_write_cr4(unsigned long cr4)
906 {
907 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
908 
909 	native_write_cr4(cr4);
910 }
911 
912 static u64 xen_read_msr_safe(unsigned int msr, int *err)
913 {
914 	u64 val;
915 
916 	if (pmu_msr_read(msr, &val, err))
917 		return val;
918 
919 	val = native_read_msr_safe(msr, err);
920 	switch (msr) {
921 	case MSR_IA32_APICBASE:
922 		val &= ~X2APIC_ENABLE;
923 		break;
924 	}
925 	return val;
926 }
927 
928 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
929 {
930 	int ret;
931 	unsigned int which;
932 	u64 base;
933 
934 	ret = 0;
935 
936 	switch (msr) {
937 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
938 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
939 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
940 
941 	set:
942 		base = ((u64)high << 32) | low;
943 		if (HYPERVISOR_set_segment_base(which, base) != 0)
944 			ret = -EIO;
945 		break;
946 
947 	case MSR_STAR:
948 	case MSR_CSTAR:
949 	case MSR_LSTAR:
950 	case MSR_SYSCALL_MASK:
951 	case MSR_IA32_SYSENTER_CS:
952 	case MSR_IA32_SYSENTER_ESP:
953 	case MSR_IA32_SYSENTER_EIP:
954 		/* Fast syscall setup is all done in hypercalls, so
955 		   these are all ignored.  Stub them out here to stop
956 		   Xen console noise. */
957 		break;
958 
959 	default:
960 		if (!pmu_msr_write(msr, low, high, &ret))
961 			ret = native_write_msr_safe(msr, low, high);
962 	}
963 
964 	return ret;
965 }
966 
967 static u64 xen_read_msr(unsigned int msr)
968 {
969 	/*
970 	 * This will silently swallow a #GP from RDMSR.  It may be worth
971 	 * changing that.
972 	 */
973 	int err;
974 
975 	return xen_read_msr_safe(msr, &err);
976 }
977 
978 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
979 {
980 	/*
981 	 * This will silently swallow a #GP from WRMSR.  It may be worth
982 	 * changing that.
983 	 */
984 	xen_write_msr_safe(msr, low, high);
985 }
986 
987 /* This is called once we have the cpu_possible_mask */
988 void __init xen_setup_vcpu_info_placement(void)
989 {
990 	int cpu;
991 
992 	for_each_possible_cpu(cpu) {
993 		/* Set up direct vCPU id mapping for PV guests. */
994 		per_cpu(xen_vcpu_id, cpu) = cpu;
995 		xen_vcpu_setup(cpu);
996 	}
997 
998 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
999 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1000 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1001 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1002 }
1003 
1004 static const struct pv_info xen_info __initconst = {
1005 	.extra_user_64bit_cs = FLAT_USER_CS64,
1006 	.name = "Xen",
1007 };
1008 
1009 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1010 	.cpu = {
1011 		.cpuid = xen_cpuid,
1012 
1013 		.set_debugreg = xen_set_debugreg,
1014 		.get_debugreg = xen_get_debugreg,
1015 
1016 		.read_cr0 = xen_read_cr0,
1017 		.write_cr0 = xen_write_cr0,
1018 
1019 		.write_cr4 = xen_write_cr4,
1020 
1021 		.wbinvd = native_wbinvd,
1022 
1023 		.read_msr = xen_read_msr,
1024 		.write_msr = xen_write_msr,
1025 
1026 		.read_msr_safe = xen_read_msr_safe,
1027 		.write_msr_safe = xen_write_msr_safe,
1028 
1029 		.read_pmc = xen_read_pmc,
1030 
1031 		.load_tr_desc = paravirt_nop,
1032 		.set_ldt = xen_set_ldt,
1033 		.load_gdt = xen_load_gdt,
1034 		.load_idt = xen_load_idt,
1035 		.load_tls = xen_load_tls,
1036 		.load_gs_index = xen_load_gs_index,
1037 
1038 		.alloc_ldt = xen_alloc_ldt,
1039 		.free_ldt = xen_free_ldt,
1040 
1041 		.store_tr = xen_store_tr,
1042 
1043 		.write_ldt_entry = xen_write_ldt_entry,
1044 		.write_gdt_entry = xen_write_gdt_entry,
1045 		.write_idt_entry = xen_write_idt_entry,
1046 		.load_sp0 = xen_load_sp0,
1047 
1048 #ifdef CONFIG_X86_IOPL_IOPERM
1049 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1050 		.update_io_bitmap = xen_update_io_bitmap,
1051 #endif
1052 		.io_delay = xen_io_delay,
1053 
1054 		.start_context_switch = paravirt_start_context_switch,
1055 		.end_context_switch = xen_end_context_switch,
1056 	},
1057 };
1058 
1059 static void xen_restart(char *msg)
1060 {
1061 	xen_reboot(SHUTDOWN_reboot);
1062 }
1063 
1064 static void xen_machine_halt(void)
1065 {
1066 	xen_reboot(SHUTDOWN_poweroff);
1067 }
1068 
1069 static void xen_machine_power_off(void)
1070 {
1071 	if (pm_power_off)
1072 		pm_power_off();
1073 	xen_reboot(SHUTDOWN_poweroff);
1074 }
1075 
1076 static void xen_crash_shutdown(struct pt_regs *regs)
1077 {
1078 	xen_reboot(SHUTDOWN_crash);
1079 }
1080 
1081 static const struct machine_ops xen_machine_ops __initconst = {
1082 	.restart = xen_restart,
1083 	.halt = xen_machine_halt,
1084 	.power_off = xen_machine_power_off,
1085 	.shutdown = xen_machine_halt,
1086 	.crash_shutdown = xen_crash_shutdown,
1087 	.emergency_restart = xen_emergency_restart,
1088 };
1089 
1090 static unsigned char xen_get_nmi_reason(void)
1091 {
1092 	unsigned char reason = 0;
1093 
1094 	/* Construct a value which looks like it came from port 0x61. */
1095 	if (test_bit(_XEN_NMIREASON_io_error,
1096 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1097 		reason |= NMI_REASON_IOCHK;
1098 	if (test_bit(_XEN_NMIREASON_pci_serr,
1099 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1100 		reason |= NMI_REASON_SERR;
1101 
1102 	return reason;
1103 }
1104 
1105 static void __init xen_boot_params_init_edd(void)
1106 {
1107 #if IS_ENABLED(CONFIG_EDD)
1108 	struct xen_platform_op op;
1109 	struct edd_info *edd_info;
1110 	u32 *mbr_signature;
1111 	unsigned nr;
1112 	int ret;
1113 
1114 	edd_info = boot_params.eddbuf;
1115 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1116 
1117 	op.cmd = XENPF_firmware_info;
1118 
1119 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1120 	for (nr = 0; nr < EDDMAXNR; nr++) {
1121 		struct edd_info *info = edd_info + nr;
1122 
1123 		op.u.firmware_info.index = nr;
1124 		info->params.length = sizeof(info->params);
1125 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1126 				     &info->params);
1127 		ret = HYPERVISOR_platform_op(&op);
1128 		if (ret)
1129 			break;
1130 
1131 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1132 		C(device);
1133 		C(version);
1134 		C(interface_support);
1135 		C(legacy_max_cylinder);
1136 		C(legacy_max_head);
1137 		C(legacy_sectors_per_track);
1138 #undef C
1139 	}
1140 	boot_params.eddbuf_entries = nr;
1141 
1142 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1143 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1144 		op.u.firmware_info.index = nr;
1145 		ret = HYPERVISOR_platform_op(&op);
1146 		if (ret)
1147 			break;
1148 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1149 	}
1150 	boot_params.edd_mbr_sig_buf_entries = nr;
1151 #endif
1152 }
1153 
1154 /*
1155  * Set up the GDT and segment registers for -fstack-protector.  Until
1156  * we do this, we have to be careful not to call any stack-protected
1157  * function, which is most of the kernel.
1158  */
1159 static void __init xen_setup_gdt(int cpu)
1160 {
1161 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1162 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1163 
1164 	switch_to_new_gdt(cpu);
1165 
1166 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1167 	pv_ops.cpu.load_gdt = xen_load_gdt;
1168 }
1169 
1170 static void __init xen_dom0_set_legacy_features(void)
1171 {
1172 	x86_platform.legacy.rtc = 1;
1173 }
1174 
1175 static void __init xen_domu_set_legacy_features(void)
1176 {
1177 	x86_platform.legacy.rtc = 0;
1178 }
1179 
1180 /* First C function to be called on Xen boot */
1181 asmlinkage __visible void __init xen_start_kernel(void)
1182 {
1183 	struct physdev_set_iopl set_iopl;
1184 	unsigned long initrd_start = 0;
1185 	int rc;
1186 
1187 	if (!xen_start_info)
1188 		return;
1189 
1190 	xen_domain_type = XEN_PV_DOMAIN;
1191 	xen_start_flags = xen_start_info->flags;
1192 
1193 	xen_setup_features();
1194 
1195 	/* Install Xen paravirt ops */
1196 	pv_info = xen_info;
1197 	pv_ops.cpu = xen_cpu_ops.cpu;
1198 	paravirt_iret = xen_iret;
1199 	xen_init_irq_ops();
1200 
1201 	/*
1202 	 * Setup xen_vcpu early because it is needed for
1203 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1204 	 *
1205 	 * Don't do the full vcpu_info placement stuff until we have
1206 	 * the cpu_possible_mask and a non-dummy shared_info.
1207 	 */
1208 	xen_vcpu_info_reset(0);
1209 
1210 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1211 
1212 	x86_init.resources.memory_setup = xen_memory_setup;
1213 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1214 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1215 	x86_init.oem.arch_setup = xen_arch_setup;
1216 	x86_init.oem.banner = xen_banner;
1217 	x86_init.hyper.init_platform = xen_pv_init_platform;
1218 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1219 
1220 	/*
1221 	 * Set up some pagetable state before starting to set any ptes.
1222 	 */
1223 
1224 	xen_setup_machphys_mapping();
1225 	xen_init_mmu_ops();
1226 
1227 	/* Prevent unwanted bits from being set in PTEs. */
1228 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1229 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1230 
1231 	/* Get mfn list */
1232 	xen_build_dynamic_phys_to_machine();
1233 
1234 	/* Work out if we support NX */
1235 	get_cpu_cap(&boot_cpu_data);
1236 	x86_configure_nx();
1237 
1238 	/*
1239 	 * Set up kernel GDT and segment registers, mainly so that
1240 	 * -fstack-protector code can be executed.
1241 	 */
1242 	xen_setup_gdt(0);
1243 
1244 	/* Determine virtual and physical address sizes */
1245 	get_cpu_address_sizes(&boot_cpu_data);
1246 
1247 	/* Let's presume PV guests always boot on vCPU with id 0. */
1248 	per_cpu(xen_vcpu_id, 0) = 0;
1249 
1250 	idt_setup_early_handler();
1251 
1252 	xen_init_capabilities();
1253 
1254 #ifdef CONFIG_X86_LOCAL_APIC
1255 	/*
1256 	 * set up the basic apic ops.
1257 	 */
1258 	xen_init_apic();
1259 #endif
1260 
1261 	machine_ops = xen_machine_ops;
1262 
1263 	/*
1264 	 * The only reliable way to retain the initial address of the
1265 	 * percpu gdt_page is to remember it here, so we can go and
1266 	 * mark it RW later, when the initial percpu area is freed.
1267 	 */
1268 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1269 
1270 	xen_smp_init();
1271 
1272 #ifdef CONFIG_ACPI_NUMA
1273 	/*
1274 	 * The pages we from Xen are not related to machine pages, so
1275 	 * any NUMA information the kernel tries to get from ACPI will
1276 	 * be meaningless.  Prevent it from trying.
1277 	 */
1278 	disable_srat();
1279 #endif
1280 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1281 
1282 	local_irq_disable();
1283 	early_boot_irqs_disabled = true;
1284 
1285 	xen_raw_console_write("mapping kernel into physical memory\n");
1286 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1287 				   xen_start_info->nr_pages);
1288 	xen_reserve_special_pages();
1289 
1290 	/*
1291 	 * We used to do this in xen_arch_setup, but that is too late
1292 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1293 	 * early_amd_init which pokes 0xcf8 port.
1294 	 */
1295 	set_iopl.iopl = 1;
1296 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1297 	if (rc != 0)
1298 		xen_raw_printk("physdev_op failed %d\n", rc);
1299 
1300 
1301 	if (xen_start_info->mod_start) {
1302 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1303 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1304 	    else
1305 		initrd_start = __pa(xen_start_info->mod_start);
1306 	}
1307 
1308 	/* Poke various useful things into boot_params */
1309 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1310 	boot_params.hdr.ramdisk_image = initrd_start;
1311 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1312 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1313 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1314 
1315 	if (!xen_initial_domain()) {
1316 		if (pci_xen)
1317 			x86_init.pci.arch_init = pci_xen_init;
1318 		x86_platform.set_legacy_features =
1319 				xen_domu_set_legacy_features;
1320 	} else {
1321 		const struct dom0_vga_console_info *info =
1322 			(void *)((char *)xen_start_info +
1323 				 xen_start_info->console.dom0.info_off);
1324 		struct xen_platform_op op = {
1325 			.cmd = XENPF_firmware_info,
1326 			.interface_version = XENPF_INTERFACE_VERSION,
1327 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1328 		};
1329 
1330 		x86_platform.set_legacy_features =
1331 				xen_dom0_set_legacy_features;
1332 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1333 		xen_start_info->console.domU.mfn = 0;
1334 		xen_start_info->console.domU.evtchn = 0;
1335 
1336 		if (HYPERVISOR_platform_op(&op) == 0)
1337 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1338 
1339 		/* Make sure ACS will be enabled */
1340 		pci_request_acs();
1341 
1342 		xen_acpi_sleep_register();
1343 
1344 		/* Avoid searching for BIOS MP tables */
1345 		x86_init.mpparse.find_smp_config = x86_init_noop;
1346 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1347 
1348 		xen_boot_params_init_edd();
1349 
1350 #ifdef CONFIG_ACPI
1351 		/*
1352 		 * Disable selecting "Firmware First mode" for correctable
1353 		 * memory errors, as this is the duty of the hypervisor to
1354 		 * decide.
1355 		 */
1356 		acpi_disable_cmcff = 1;
1357 #endif
1358 	}
1359 
1360 	xen_add_preferred_consoles();
1361 
1362 #ifdef CONFIG_PCI
1363 	/* PCI BIOS service won't work from a PV guest. */
1364 	pci_probe &= ~PCI_PROBE_BIOS;
1365 #endif
1366 	xen_raw_console_write("about to get started...\n");
1367 
1368 	/* We need this for printk timestamps */
1369 	xen_setup_runstate_info(0);
1370 
1371 	xen_efi_init(&boot_params);
1372 
1373 	/* Start the world */
1374 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1375 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1376 }
1377 
1378 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1379 {
1380 	int rc;
1381 
1382 	if (per_cpu(xen_vcpu, cpu) == NULL)
1383 		return -ENODEV;
1384 
1385 	xen_setup_timer(cpu);
1386 
1387 	rc = xen_smp_intr_init(cpu);
1388 	if (rc) {
1389 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1390 		     cpu, rc);
1391 		return rc;
1392 	}
1393 
1394 	rc = xen_smp_intr_init_pv(cpu);
1395 	if (rc) {
1396 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1397 		     cpu, rc);
1398 		return rc;
1399 	}
1400 
1401 	return 0;
1402 }
1403 
1404 static int xen_cpu_dead_pv(unsigned int cpu)
1405 {
1406 	xen_smp_intr_free(cpu);
1407 	xen_smp_intr_free_pv(cpu);
1408 
1409 	xen_teardown_timer(cpu);
1410 
1411 	return 0;
1412 }
1413 
1414 static uint32_t __init xen_platform_pv(void)
1415 {
1416 	if (xen_pv_domain())
1417 		return xen_cpuid_base();
1418 
1419 	return 0;
1420 }
1421 
1422 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1423 	.name                   = "Xen PV",
1424 	.detect                 = xen_platform_pv,
1425 	.type			= X86_HYPER_XEN_PV,
1426 	.runtime.pin_vcpu       = xen_pin_vcpu,
1427 	.ignore_nopv		= true,
1428 };
1429