xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 8bdc2a19)
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 
34 #include <xen/xen.h>
35 #include <xen/events.h>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/version.h>
38 #include <xen/interface/physdev.h>
39 #include <xen/interface/vcpu.h>
40 #include <xen/interface/memory.h>
41 #include <xen/interface/nmi.h>
42 #include <xen/interface/xen-mca.h>
43 #include <xen/features.h>
44 #include <xen/page.h>
45 #include <xen/hvc-console.h>
46 #include <xen/acpi.h>
47 
48 #include <asm/paravirt.h>
49 #include <asm/apic.h>
50 #include <asm/page.h>
51 #include <asm/xen/pci.h>
52 #include <asm/xen/hypercall.h>
53 #include <asm/xen/hypervisor.h>
54 #include <asm/xen/cpuid.h>
55 #include <asm/fixmap.h>
56 #include <asm/processor.h>
57 #include <asm/proto.h>
58 #include <asm/msr-index.h>
59 #include <asm/traps.h>
60 #include <asm/setup.h>
61 #include <asm/desc.h>
62 #include <asm/pgalloc.h>
63 #include <asm/tlbflush.h>
64 #include <asm/reboot.h>
65 #include <asm/stackprotector.h>
66 #include <asm/hypervisor.h>
67 #include <asm/mach_traps.h>
68 #include <asm/mwait.h>
69 #include <asm/pci_x86.h>
70 #include <asm/cpu.h>
71 #ifdef CONFIG_X86_IOPL_IOPERM
72 #include <asm/io_bitmap.h>
73 #endif
74 
75 #ifdef CONFIG_ACPI
76 #include <linux/acpi.h>
77 #include <asm/acpi.h>
78 #include <acpi/pdc_intel.h>
79 #include <acpi/processor.h>
80 #include <xen/interface/platform.h>
81 #endif
82 
83 #include "xen-ops.h"
84 #include "mmu.h"
85 #include "smp.h"
86 #include "multicalls.h"
87 #include "pmu.h"
88 
89 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
90 
91 void *xen_initial_gdt;
92 
93 static int xen_cpu_up_prepare_pv(unsigned int cpu);
94 static int xen_cpu_dead_pv(unsigned int cpu);
95 
96 struct tls_descs {
97 	struct desc_struct desc[3];
98 };
99 
100 /*
101  * Updating the 3 TLS descriptors in the GDT on every task switch is
102  * surprisingly expensive so we avoid updating them if they haven't
103  * changed.  Since Xen writes different descriptors than the one
104  * passed in the update_descriptor hypercall we keep shadow copies to
105  * compare against.
106  */
107 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
108 
109 static void __init xen_pv_init_platform(void)
110 {
111 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
112 
113 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
114 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
115 
116 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
117 	xen_vcpu_info_reset(0);
118 
119 	/* pvclock is in shared info area */
120 	xen_init_time_ops();
121 }
122 
123 static void __init xen_pv_guest_late_init(void)
124 {
125 #ifndef CONFIG_SMP
126 	/* Setup shared vcpu info for non-smp configurations */
127 	xen_setup_vcpu_info_placement();
128 #endif
129 }
130 
131 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
132 static __read_mostly unsigned int cpuid_leaf5_edx_val;
133 
134 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
135 		      unsigned int *cx, unsigned int *dx)
136 {
137 	unsigned maskebx = ~0;
138 
139 	/*
140 	 * Mask out inconvenient features, to try and disable as many
141 	 * unsupported kernel subsystems as possible.
142 	 */
143 	switch (*ax) {
144 	case CPUID_MWAIT_LEAF:
145 		/* Synthesize the values.. */
146 		*ax = 0;
147 		*bx = 0;
148 		*cx = cpuid_leaf5_ecx_val;
149 		*dx = cpuid_leaf5_edx_val;
150 		return;
151 
152 	case 0xb:
153 		/* Suppress extended topology stuff */
154 		maskebx = 0;
155 		break;
156 	}
157 
158 	asm(XEN_EMULATE_PREFIX "cpuid"
159 		: "=a" (*ax),
160 		  "=b" (*bx),
161 		  "=c" (*cx),
162 		  "=d" (*dx)
163 		: "0" (*ax), "2" (*cx));
164 
165 	*bx &= maskebx;
166 }
167 
168 static bool __init xen_check_mwait(void)
169 {
170 #ifdef CONFIG_ACPI
171 	struct xen_platform_op op = {
172 		.cmd			= XENPF_set_processor_pminfo,
173 		.u.set_pminfo.id	= -1,
174 		.u.set_pminfo.type	= XEN_PM_PDC,
175 	};
176 	uint32_t buf[3];
177 	unsigned int ax, bx, cx, dx;
178 	unsigned int mwait_mask;
179 
180 	/* We need to determine whether it is OK to expose the MWAIT
181 	 * capability to the kernel to harvest deeper than C3 states from ACPI
182 	 * _CST using the processor_harvest_xen.c module. For this to work, we
183 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
184 	 * checks against). The hypervisor won't expose the MWAIT flag because
185 	 * it would break backwards compatibility; so we will find out directly
186 	 * from the hardware and hypercall.
187 	 */
188 	if (!xen_initial_domain())
189 		return false;
190 
191 	/*
192 	 * When running under platform earlier than Xen4.2, do not expose
193 	 * mwait, to avoid the risk of loading native acpi pad driver
194 	 */
195 	if (!xen_running_on_version_or_later(4, 2))
196 		return false;
197 
198 	ax = 1;
199 	cx = 0;
200 
201 	native_cpuid(&ax, &bx, &cx, &dx);
202 
203 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
204 		     (1 << (X86_FEATURE_MWAIT % 32));
205 
206 	if ((cx & mwait_mask) != mwait_mask)
207 		return false;
208 
209 	/* We need to emulate the MWAIT_LEAF and for that we need both
210 	 * ecx and edx. The hypercall provides only partial information.
211 	 */
212 
213 	ax = CPUID_MWAIT_LEAF;
214 	bx = 0;
215 	cx = 0;
216 	dx = 0;
217 
218 	native_cpuid(&ax, &bx, &cx, &dx);
219 
220 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
221 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
222 	 */
223 	buf[0] = ACPI_PDC_REVISION_ID;
224 	buf[1] = 1;
225 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
226 
227 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
228 
229 	if ((HYPERVISOR_platform_op(&op) == 0) &&
230 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
231 		cpuid_leaf5_ecx_val = cx;
232 		cpuid_leaf5_edx_val = dx;
233 	}
234 	return true;
235 #else
236 	return false;
237 #endif
238 }
239 
240 static bool __init xen_check_xsave(void)
241 {
242 	unsigned int cx, xsave_mask;
243 
244 	cx = cpuid_ecx(1);
245 
246 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
247 		     (1 << (X86_FEATURE_OSXSAVE % 32));
248 
249 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
250 	return (cx & xsave_mask) == xsave_mask;
251 }
252 
253 static void __init xen_init_capabilities(void)
254 {
255 	setup_force_cpu_cap(X86_FEATURE_XENPV);
256 	setup_clear_cpu_cap(X86_FEATURE_DCA);
257 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
258 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
259 	setup_clear_cpu_cap(X86_FEATURE_ACC);
260 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
261 	setup_clear_cpu_cap(X86_FEATURE_SME);
262 
263 	/*
264 	 * Xen PV would need some work to support PCID: CR3 handling as well
265 	 * as xen_flush_tlb_others() would need updating.
266 	 */
267 	setup_clear_cpu_cap(X86_FEATURE_PCID);
268 
269 	if (!xen_initial_domain())
270 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
271 
272 	if (xen_check_mwait())
273 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
274 	else
275 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
276 
277 	if (!xen_check_xsave()) {
278 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
279 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
280 	}
281 }
282 
283 static noinstr void xen_set_debugreg(int reg, unsigned long val)
284 {
285 	HYPERVISOR_set_debugreg(reg, val);
286 }
287 
288 static noinstr unsigned long xen_get_debugreg(int reg)
289 {
290 	return HYPERVISOR_get_debugreg(reg);
291 }
292 
293 static void xen_end_context_switch(struct task_struct *next)
294 {
295 	xen_mc_flush();
296 	paravirt_end_context_switch(next);
297 }
298 
299 static unsigned long xen_store_tr(void)
300 {
301 	return 0;
302 }
303 
304 /*
305  * Set the page permissions for a particular virtual address.  If the
306  * address is a vmalloc mapping (or other non-linear mapping), then
307  * find the linear mapping of the page and also set its protections to
308  * match.
309  */
310 static void set_aliased_prot(void *v, pgprot_t prot)
311 {
312 	int level;
313 	pte_t *ptep;
314 	pte_t pte;
315 	unsigned long pfn;
316 	unsigned char dummy;
317 	void *va;
318 
319 	ptep = lookup_address((unsigned long)v, &level);
320 	BUG_ON(ptep == NULL);
321 
322 	pfn = pte_pfn(*ptep);
323 	pte = pfn_pte(pfn, prot);
324 
325 	/*
326 	 * Careful: update_va_mapping() will fail if the virtual address
327 	 * we're poking isn't populated in the page tables.  We don't
328 	 * need to worry about the direct map (that's always in the page
329 	 * tables), but we need to be careful about vmap space.  In
330 	 * particular, the top level page table can lazily propagate
331 	 * entries between processes, so if we've switched mms since we
332 	 * vmapped the target in the first place, we might not have the
333 	 * top-level page table entry populated.
334 	 *
335 	 * We disable preemption because we want the same mm active when
336 	 * we probe the target and when we issue the hypercall.  We'll
337 	 * have the same nominal mm, but if we're a kernel thread, lazy
338 	 * mm dropping could change our pgd.
339 	 *
340 	 * Out of an abundance of caution, this uses __get_user() to fault
341 	 * in the target address just in case there's some obscure case
342 	 * in which the target address isn't readable.
343 	 */
344 
345 	preempt_disable();
346 
347 	copy_from_kernel_nofault(&dummy, v, 1);
348 
349 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
350 		BUG();
351 
352 	va = __va(PFN_PHYS(pfn));
353 
354 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
355 		BUG();
356 
357 	preempt_enable();
358 }
359 
360 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
361 {
362 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
363 	int i;
364 
365 	/*
366 	 * We need to mark the all aliases of the LDT pages RO.  We
367 	 * don't need to call vm_flush_aliases(), though, since that's
368 	 * only responsible for flushing aliases out the TLBs, not the
369 	 * page tables, and Xen will flush the TLB for us if needed.
370 	 *
371 	 * To avoid confusing future readers: none of this is necessary
372 	 * to load the LDT.  The hypervisor only checks this when the
373 	 * LDT is faulted in due to subsequent descriptor access.
374 	 */
375 
376 	for (i = 0; i < entries; i += entries_per_page)
377 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
378 }
379 
380 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
381 {
382 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
383 	int i;
384 
385 	for (i = 0; i < entries; i += entries_per_page)
386 		set_aliased_prot(ldt + i, PAGE_KERNEL);
387 }
388 
389 static void xen_set_ldt(const void *addr, unsigned entries)
390 {
391 	struct mmuext_op *op;
392 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
393 
394 	trace_xen_cpu_set_ldt(addr, entries);
395 
396 	op = mcs.args;
397 	op->cmd = MMUEXT_SET_LDT;
398 	op->arg1.linear_addr = (unsigned long)addr;
399 	op->arg2.nr_ents = entries;
400 
401 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
402 
403 	xen_mc_issue(PARAVIRT_LAZY_CPU);
404 }
405 
406 static void xen_load_gdt(const struct desc_ptr *dtr)
407 {
408 	unsigned long va = dtr->address;
409 	unsigned int size = dtr->size + 1;
410 	unsigned long pfn, mfn;
411 	int level;
412 	pte_t *ptep;
413 	void *virt;
414 
415 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
416 	BUG_ON(size > PAGE_SIZE);
417 	BUG_ON(va & ~PAGE_MASK);
418 
419 	/*
420 	 * The GDT is per-cpu and is in the percpu data area.
421 	 * That can be virtually mapped, so we need to do a
422 	 * page-walk to get the underlying MFN for the
423 	 * hypercall.  The page can also be in the kernel's
424 	 * linear range, so we need to RO that mapping too.
425 	 */
426 	ptep = lookup_address(va, &level);
427 	BUG_ON(ptep == NULL);
428 
429 	pfn = pte_pfn(*ptep);
430 	mfn = pfn_to_mfn(pfn);
431 	virt = __va(PFN_PHYS(pfn));
432 
433 	make_lowmem_page_readonly((void *)va);
434 	make_lowmem_page_readonly(virt);
435 
436 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
437 		BUG();
438 }
439 
440 /*
441  * load_gdt for early boot, when the gdt is only mapped once
442  */
443 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
444 {
445 	unsigned long va = dtr->address;
446 	unsigned int size = dtr->size + 1;
447 	unsigned long pfn, mfn;
448 	pte_t pte;
449 
450 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
451 	BUG_ON(size > PAGE_SIZE);
452 	BUG_ON(va & ~PAGE_MASK);
453 
454 	pfn = virt_to_pfn(va);
455 	mfn = pfn_to_mfn(pfn);
456 
457 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
458 
459 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
460 		BUG();
461 
462 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
463 		BUG();
464 }
465 
466 static inline bool desc_equal(const struct desc_struct *d1,
467 			      const struct desc_struct *d2)
468 {
469 	return !memcmp(d1, d2, sizeof(*d1));
470 }
471 
472 static void load_TLS_descriptor(struct thread_struct *t,
473 				unsigned int cpu, unsigned int i)
474 {
475 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
476 	struct desc_struct *gdt;
477 	xmaddr_t maddr;
478 	struct multicall_space mc;
479 
480 	if (desc_equal(shadow, &t->tls_array[i]))
481 		return;
482 
483 	*shadow = t->tls_array[i];
484 
485 	gdt = get_cpu_gdt_rw(cpu);
486 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
487 	mc = __xen_mc_entry(0);
488 
489 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
490 }
491 
492 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
493 {
494 	/*
495 	 * In lazy mode we need to zero %fs, otherwise we may get an
496 	 * exception between the new %fs descriptor being loaded and
497 	 * %fs being effectively cleared at __switch_to().
498 	 */
499 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
500 		loadsegment(fs, 0);
501 
502 	xen_mc_batch();
503 
504 	load_TLS_descriptor(t, cpu, 0);
505 	load_TLS_descriptor(t, cpu, 1);
506 	load_TLS_descriptor(t, cpu, 2);
507 
508 	xen_mc_issue(PARAVIRT_LAZY_CPU);
509 }
510 
511 static void xen_load_gs_index(unsigned int idx)
512 {
513 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
514 		BUG();
515 }
516 
517 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
518 				const void *ptr)
519 {
520 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
521 	u64 entry = *(u64 *)ptr;
522 
523 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
524 
525 	preempt_disable();
526 
527 	xen_mc_flush();
528 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
529 		BUG();
530 
531 	preempt_enable();
532 }
533 
534 void noist_exc_debug(struct pt_regs *regs);
535 
536 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
537 {
538 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
539 	exc_nmi(regs);
540 }
541 
542 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
543 {
544 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
545 	exc_double_fault(regs, error_code);
546 }
547 
548 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
549 {
550 	/*
551 	 * There's no IST on Xen PV, but we still need to dispatch
552 	 * to the correct handler.
553 	 */
554 	if (user_mode(regs))
555 		noist_exc_debug(regs);
556 	else
557 		exc_debug(regs);
558 }
559 
560 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
561 {
562 	/* This should never happen and there is no way to handle it. */
563 	instrumentation_begin();
564 	pr_err("Unknown trap in Xen PV mode.");
565 	BUG();
566 	instrumentation_end();
567 }
568 
569 #ifdef CONFIG_X86_MCE
570 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
571 {
572 	/*
573 	 * There's no IST on Xen PV, but we still need to dispatch
574 	 * to the correct handler.
575 	 */
576 	if (user_mode(regs))
577 		noist_exc_machine_check(regs);
578 	else
579 		exc_machine_check(regs);
580 }
581 #endif
582 
583 struct trap_array_entry {
584 	void (*orig)(void);
585 	void (*xen)(void);
586 	bool ist_okay;
587 };
588 
589 #define TRAP_ENTRY(func, ist_ok) {			\
590 	.orig		= asm_##func,			\
591 	.xen		= xen_asm_##func,		\
592 	.ist_okay	= ist_ok }
593 
594 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
595 	.orig		= asm_##func,			\
596 	.xen		= xen_asm_xenpv_##func,		\
597 	.ist_okay	= ist_ok }
598 
599 static struct trap_array_entry trap_array[] = {
600 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
601 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
602 #ifdef CONFIG_X86_MCE
603 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
604 #endif
605 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
606 	TRAP_ENTRY(exc_int3,				false ),
607 	TRAP_ENTRY(exc_overflow,			false ),
608 #ifdef CONFIG_IA32_EMULATION
609 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
610 #endif
611 	TRAP_ENTRY(exc_page_fault,			false ),
612 	TRAP_ENTRY(exc_divide_error,			false ),
613 	TRAP_ENTRY(exc_bounds,				false ),
614 	TRAP_ENTRY(exc_invalid_op,			false ),
615 	TRAP_ENTRY(exc_device_not_available,		false ),
616 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
617 	TRAP_ENTRY(exc_invalid_tss,			false ),
618 	TRAP_ENTRY(exc_segment_not_present,		false ),
619 	TRAP_ENTRY(exc_stack_segment,			false ),
620 	TRAP_ENTRY(exc_general_protection,		false ),
621 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
622 	TRAP_ENTRY(exc_coprocessor_error,		false ),
623 	TRAP_ENTRY(exc_alignment_check,			false ),
624 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
625 #ifdef CONFIG_X86_KERNEL_IBT
626 	TRAP_ENTRY(exc_control_protection,		false ),
627 #endif
628 };
629 
630 static bool __ref get_trap_addr(void **addr, unsigned int ist)
631 {
632 	unsigned int nr;
633 	bool ist_okay = false;
634 	bool found = false;
635 
636 	/*
637 	 * Replace trap handler addresses by Xen specific ones.
638 	 * Check for known traps using IST and whitelist them.
639 	 * The debugger ones are the only ones we care about.
640 	 * Xen will handle faults like double_fault, so we should never see
641 	 * them.  Warn if there's an unexpected IST-using fault handler.
642 	 */
643 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
644 		struct trap_array_entry *entry = trap_array + nr;
645 
646 		if (*addr == entry->orig) {
647 			*addr = entry->xen;
648 			ist_okay = entry->ist_okay;
649 			found = true;
650 			break;
651 		}
652 	}
653 
654 	if (nr == ARRAY_SIZE(trap_array) &&
655 	    *addr >= (void *)early_idt_handler_array[0] &&
656 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
657 		nr = (*addr - (void *)early_idt_handler_array[0]) /
658 		     EARLY_IDT_HANDLER_SIZE;
659 		*addr = (void *)xen_early_idt_handler_array[nr];
660 		found = true;
661 	}
662 
663 	if (!found)
664 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
665 
666 	if (WARN_ON(found && ist != 0 && !ist_okay))
667 		return false;
668 
669 	return true;
670 }
671 
672 static int cvt_gate_to_trap(int vector, const gate_desc *val,
673 			    struct trap_info *info)
674 {
675 	unsigned long addr;
676 
677 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
678 		return 0;
679 
680 	info->vector = vector;
681 
682 	addr = gate_offset(val);
683 	if (!get_trap_addr((void **)&addr, val->bits.ist))
684 		return 0;
685 	info->address = addr;
686 
687 	info->cs = gate_segment(val);
688 	info->flags = val->bits.dpl;
689 	/* interrupt gates clear IF */
690 	if (val->bits.type == GATE_INTERRUPT)
691 		info->flags |= 1 << 2;
692 
693 	return 1;
694 }
695 
696 /* Locations of each CPU's IDT */
697 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
698 
699 /* Set an IDT entry.  If the entry is part of the current IDT, then
700    also update Xen. */
701 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
702 {
703 	unsigned long p = (unsigned long)&dt[entrynum];
704 	unsigned long start, end;
705 
706 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
707 
708 	preempt_disable();
709 
710 	start = __this_cpu_read(idt_desc.address);
711 	end = start + __this_cpu_read(idt_desc.size) + 1;
712 
713 	xen_mc_flush();
714 
715 	native_write_idt_entry(dt, entrynum, g);
716 
717 	if (p >= start && (p + 8) <= end) {
718 		struct trap_info info[2];
719 
720 		info[1].address = 0;
721 
722 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
723 			if (HYPERVISOR_set_trap_table(info))
724 				BUG();
725 	}
726 
727 	preempt_enable();
728 }
729 
730 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
731 				      struct trap_info *traps, bool full)
732 {
733 	unsigned in, out, count;
734 
735 	count = (desc->size+1) / sizeof(gate_desc);
736 	BUG_ON(count > 256);
737 
738 	for (in = out = 0; in < count; in++) {
739 		gate_desc *entry = (gate_desc *)(desc->address) + in;
740 
741 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
742 			out++;
743 	}
744 
745 	return out;
746 }
747 
748 void xen_copy_trap_info(struct trap_info *traps)
749 {
750 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
751 
752 	xen_convert_trap_info(desc, traps, true);
753 }
754 
755 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
756    hold a spinlock to protect the static traps[] array (static because
757    it avoids allocation, and saves stack space). */
758 static void xen_load_idt(const struct desc_ptr *desc)
759 {
760 	static DEFINE_SPINLOCK(lock);
761 	static struct trap_info traps[257];
762 	unsigned out;
763 
764 	trace_xen_cpu_load_idt(desc);
765 
766 	spin_lock(&lock);
767 
768 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
769 
770 	out = xen_convert_trap_info(desc, traps, false);
771 	memset(&traps[out], 0, sizeof(traps[0]));
772 
773 	xen_mc_flush();
774 	if (HYPERVISOR_set_trap_table(traps))
775 		BUG();
776 
777 	spin_unlock(&lock);
778 }
779 
780 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
781    they're handled differently. */
782 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
783 				const void *desc, int type)
784 {
785 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
786 
787 	preempt_disable();
788 
789 	switch (type) {
790 	case DESC_LDT:
791 	case DESC_TSS:
792 		/* ignore */
793 		break;
794 
795 	default: {
796 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
797 
798 		xen_mc_flush();
799 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
800 			BUG();
801 	}
802 
803 	}
804 
805 	preempt_enable();
806 }
807 
808 /*
809  * Version of write_gdt_entry for use at early boot-time needed to
810  * update an entry as simply as possible.
811  */
812 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
813 					    const void *desc, int type)
814 {
815 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
816 
817 	switch (type) {
818 	case DESC_LDT:
819 	case DESC_TSS:
820 		/* ignore */
821 		break;
822 
823 	default: {
824 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
825 
826 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
827 			dt[entry] = *(struct desc_struct *)desc;
828 	}
829 
830 	}
831 }
832 
833 static void xen_load_sp0(unsigned long sp0)
834 {
835 	struct multicall_space mcs;
836 
837 	mcs = xen_mc_entry(0);
838 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
839 	xen_mc_issue(PARAVIRT_LAZY_CPU);
840 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
841 }
842 
843 #ifdef CONFIG_X86_IOPL_IOPERM
844 static void xen_invalidate_io_bitmap(void)
845 {
846 	struct physdev_set_iobitmap iobitmap = {
847 		.bitmap = NULL,
848 		.nr_ports = 0,
849 	};
850 
851 	native_tss_invalidate_io_bitmap();
852 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
853 }
854 
855 static void xen_update_io_bitmap(void)
856 {
857 	struct physdev_set_iobitmap iobitmap;
858 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
859 
860 	native_tss_update_io_bitmap();
861 
862 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
863 			  tss->x86_tss.io_bitmap_base;
864 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
865 		iobitmap.nr_ports = 0;
866 	else
867 		iobitmap.nr_ports = IO_BITMAP_BITS;
868 
869 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
870 }
871 #endif
872 
873 static void xen_io_delay(void)
874 {
875 }
876 
877 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
878 
879 static unsigned long xen_read_cr0(void)
880 {
881 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
882 
883 	if (unlikely(cr0 == 0)) {
884 		cr0 = native_read_cr0();
885 		this_cpu_write(xen_cr0_value, cr0);
886 	}
887 
888 	return cr0;
889 }
890 
891 static void xen_write_cr0(unsigned long cr0)
892 {
893 	struct multicall_space mcs;
894 
895 	this_cpu_write(xen_cr0_value, cr0);
896 
897 	/* Only pay attention to cr0.TS; everything else is
898 	   ignored. */
899 	mcs = xen_mc_entry(0);
900 
901 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
902 
903 	xen_mc_issue(PARAVIRT_LAZY_CPU);
904 }
905 
906 static void xen_write_cr4(unsigned long cr4)
907 {
908 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
909 
910 	native_write_cr4(cr4);
911 }
912 
913 static u64 xen_read_msr_safe(unsigned int msr, int *err)
914 {
915 	u64 val;
916 
917 	if (pmu_msr_read(msr, &val, err))
918 		return val;
919 
920 	val = native_read_msr_safe(msr, err);
921 	switch (msr) {
922 	case MSR_IA32_APICBASE:
923 		val &= ~X2APIC_ENABLE;
924 		break;
925 	}
926 	return val;
927 }
928 
929 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
930 {
931 	int ret;
932 	unsigned int which;
933 	u64 base;
934 
935 	ret = 0;
936 
937 	switch (msr) {
938 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
939 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
940 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
941 
942 	set:
943 		base = ((u64)high << 32) | low;
944 		if (HYPERVISOR_set_segment_base(which, base) != 0)
945 			ret = -EIO;
946 		break;
947 
948 	case MSR_STAR:
949 	case MSR_CSTAR:
950 	case MSR_LSTAR:
951 	case MSR_SYSCALL_MASK:
952 	case MSR_IA32_SYSENTER_CS:
953 	case MSR_IA32_SYSENTER_ESP:
954 	case MSR_IA32_SYSENTER_EIP:
955 		/* Fast syscall setup is all done in hypercalls, so
956 		   these are all ignored.  Stub them out here to stop
957 		   Xen console noise. */
958 		break;
959 
960 	default:
961 		if (!pmu_msr_write(msr, low, high, &ret))
962 			ret = native_write_msr_safe(msr, low, high);
963 	}
964 
965 	return ret;
966 }
967 
968 static u64 xen_read_msr(unsigned int msr)
969 {
970 	/*
971 	 * This will silently swallow a #GP from RDMSR.  It may be worth
972 	 * changing that.
973 	 */
974 	int err;
975 
976 	return xen_read_msr_safe(msr, &err);
977 }
978 
979 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
980 {
981 	/*
982 	 * This will silently swallow a #GP from WRMSR.  It may be worth
983 	 * changing that.
984 	 */
985 	xen_write_msr_safe(msr, low, high);
986 }
987 
988 /* This is called once we have the cpu_possible_mask */
989 void __init xen_setup_vcpu_info_placement(void)
990 {
991 	int cpu;
992 
993 	for_each_possible_cpu(cpu) {
994 		/* Set up direct vCPU id mapping for PV guests. */
995 		per_cpu(xen_vcpu_id, cpu) = cpu;
996 		xen_vcpu_setup(cpu);
997 	}
998 
999 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1000 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1001 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1002 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1003 }
1004 
1005 static const struct pv_info xen_info __initconst = {
1006 	.extra_user_64bit_cs = FLAT_USER_CS64,
1007 	.name = "Xen",
1008 };
1009 
1010 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1011 	.cpu = {
1012 		.cpuid = xen_cpuid,
1013 
1014 		.set_debugreg = xen_set_debugreg,
1015 		.get_debugreg = xen_get_debugreg,
1016 
1017 		.read_cr0 = xen_read_cr0,
1018 		.write_cr0 = xen_write_cr0,
1019 
1020 		.write_cr4 = xen_write_cr4,
1021 
1022 		.wbinvd = native_wbinvd,
1023 
1024 		.read_msr = xen_read_msr,
1025 		.write_msr = xen_write_msr,
1026 
1027 		.read_msr_safe = xen_read_msr_safe,
1028 		.write_msr_safe = xen_write_msr_safe,
1029 
1030 		.read_pmc = xen_read_pmc,
1031 
1032 		.load_tr_desc = paravirt_nop,
1033 		.set_ldt = xen_set_ldt,
1034 		.load_gdt = xen_load_gdt,
1035 		.load_idt = xen_load_idt,
1036 		.load_tls = xen_load_tls,
1037 		.load_gs_index = xen_load_gs_index,
1038 
1039 		.alloc_ldt = xen_alloc_ldt,
1040 		.free_ldt = xen_free_ldt,
1041 
1042 		.store_tr = xen_store_tr,
1043 
1044 		.write_ldt_entry = xen_write_ldt_entry,
1045 		.write_gdt_entry = xen_write_gdt_entry,
1046 		.write_idt_entry = xen_write_idt_entry,
1047 		.load_sp0 = xen_load_sp0,
1048 
1049 #ifdef CONFIG_X86_IOPL_IOPERM
1050 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1051 		.update_io_bitmap = xen_update_io_bitmap,
1052 #endif
1053 		.io_delay = xen_io_delay,
1054 
1055 		.start_context_switch = paravirt_start_context_switch,
1056 		.end_context_switch = xen_end_context_switch,
1057 	},
1058 };
1059 
1060 static void xen_restart(char *msg)
1061 {
1062 	xen_reboot(SHUTDOWN_reboot);
1063 }
1064 
1065 static void xen_machine_halt(void)
1066 {
1067 	xen_reboot(SHUTDOWN_poweroff);
1068 }
1069 
1070 static void xen_machine_power_off(void)
1071 {
1072 	if (pm_power_off)
1073 		pm_power_off();
1074 	xen_reboot(SHUTDOWN_poweroff);
1075 }
1076 
1077 static void xen_crash_shutdown(struct pt_regs *regs)
1078 {
1079 	xen_reboot(SHUTDOWN_crash);
1080 }
1081 
1082 static const struct machine_ops xen_machine_ops __initconst = {
1083 	.restart = xen_restart,
1084 	.halt = xen_machine_halt,
1085 	.power_off = xen_machine_power_off,
1086 	.shutdown = xen_machine_halt,
1087 	.crash_shutdown = xen_crash_shutdown,
1088 	.emergency_restart = xen_emergency_restart,
1089 };
1090 
1091 static unsigned char xen_get_nmi_reason(void)
1092 {
1093 	unsigned char reason = 0;
1094 
1095 	/* Construct a value which looks like it came from port 0x61. */
1096 	if (test_bit(_XEN_NMIREASON_io_error,
1097 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1098 		reason |= NMI_REASON_IOCHK;
1099 	if (test_bit(_XEN_NMIREASON_pci_serr,
1100 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1101 		reason |= NMI_REASON_SERR;
1102 
1103 	return reason;
1104 }
1105 
1106 static void __init xen_boot_params_init_edd(void)
1107 {
1108 #if IS_ENABLED(CONFIG_EDD)
1109 	struct xen_platform_op op;
1110 	struct edd_info *edd_info;
1111 	u32 *mbr_signature;
1112 	unsigned nr;
1113 	int ret;
1114 
1115 	edd_info = boot_params.eddbuf;
1116 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1117 
1118 	op.cmd = XENPF_firmware_info;
1119 
1120 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1121 	for (nr = 0; nr < EDDMAXNR; nr++) {
1122 		struct edd_info *info = edd_info + nr;
1123 
1124 		op.u.firmware_info.index = nr;
1125 		info->params.length = sizeof(info->params);
1126 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1127 				     &info->params);
1128 		ret = HYPERVISOR_platform_op(&op);
1129 		if (ret)
1130 			break;
1131 
1132 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1133 		C(device);
1134 		C(version);
1135 		C(interface_support);
1136 		C(legacy_max_cylinder);
1137 		C(legacy_max_head);
1138 		C(legacy_sectors_per_track);
1139 #undef C
1140 	}
1141 	boot_params.eddbuf_entries = nr;
1142 
1143 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1144 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1145 		op.u.firmware_info.index = nr;
1146 		ret = HYPERVISOR_platform_op(&op);
1147 		if (ret)
1148 			break;
1149 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1150 	}
1151 	boot_params.edd_mbr_sig_buf_entries = nr;
1152 #endif
1153 }
1154 
1155 /*
1156  * Set up the GDT and segment registers for -fstack-protector.  Until
1157  * we do this, we have to be careful not to call any stack-protected
1158  * function, which is most of the kernel.
1159  */
1160 static void __init xen_setup_gdt(int cpu)
1161 {
1162 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1163 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1164 
1165 	switch_to_new_gdt(cpu);
1166 
1167 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1168 	pv_ops.cpu.load_gdt = xen_load_gdt;
1169 }
1170 
1171 static void __init xen_dom0_set_legacy_features(void)
1172 {
1173 	x86_platform.legacy.rtc = 1;
1174 }
1175 
1176 static void __init xen_domu_set_legacy_features(void)
1177 {
1178 	x86_platform.legacy.rtc = 0;
1179 }
1180 
1181 extern void early_xen_iret_patch(void);
1182 
1183 /* First C function to be called on Xen boot */
1184 asmlinkage __visible void __init xen_start_kernel(void)
1185 {
1186 	struct physdev_set_iopl set_iopl;
1187 	unsigned long initrd_start = 0;
1188 	int rc;
1189 
1190 	if (!xen_start_info)
1191 		return;
1192 
1193 	__text_gen_insn(&early_xen_iret_patch,
1194 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1195 			JMP32_INSN_SIZE);
1196 
1197 	xen_domain_type = XEN_PV_DOMAIN;
1198 	xen_start_flags = xen_start_info->flags;
1199 
1200 	xen_setup_features();
1201 
1202 	/* Install Xen paravirt ops */
1203 	pv_info = xen_info;
1204 	pv_ops.cpu = xen_cpu_ops.cpu;
1205 	xen_init_irq_ops();
1206 
1207 	/*
1208 	 * Setup xen_vcpu early because it is needed for
1209 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1210 	 *
1211 	 * Don't do the full vcpu_info placement stuff until we have
1212 	 * the cpu_possible_mask and a non-dummy shared_info.
1213 	 */
1214 	xen_vcpu_info_reset(0);
1215 
1216 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1217 
1218 	x86_init.resources.memory_setup = xen_memory_setup;
1219 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1220 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1221 	x86_init.oem.arch_setup = xen_arch_setup;
1222 	x86_init.oem.banner = xen_banner;
1223 	x86_init.hyper.init_platform = xen_pv_init_platform;
1224 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1225 
1226 	/*
1227 	 * Set up some pagetable state before starting to set any ptes.
1228 	 */
1229 
1230 	xen_setup_machphys_mapping();
1231 	xen_init_mmu_ops();
1232 
1233 	/* Prevent unwanted bits from being set in PTEs. */
1234 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1235 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1236 
1237 	/* Get mfn list */
1238 	xen_build_dynamic_phys_to_machine();
1239 
1240 	/* Work out if we support NX */
1241 	get_cpu_cap(&boot_cpu_data);
1242 	x86_configure_nx();
1243 
1244 	/*
1245 	 * Set up kernel GDT and segment registers, mainly so that
1246 	 * -fstack-protector code can be executed.
1247 	 */
1248 	xen_setup_gdt(0);
1249 
1250 	/* Determine virtual and physical address sizes */
1251 	get_cpu_address_sizes(&boot_cpu_data);
1252 
1253 	/* Let's presume PV guests always boot on vCPU with id 0. */
1254 	per_cpu(xen_vcpu_id, 0) = 0;
1255 
1256 	idt_setup_early_handler();
1257 
1258 	xen_init_capabilities();
1259 
1260 #ifdef CONFIG_X86_LOCAL_APIC
1261 	/*
1262 	 * set up the basic apic ops.
1263 	 */
1264 	xen_init_apic();
1265 #endif
1266 
1267 	machine_ops = xen_machine_ops;
1268 
1269 	/*
1270 	 * The only reliable way to retain the initial address of the
1271 	 * percpu gdt_page is to remember it here, so we can go and
1272 	 * mark it RW later, when the initial percpu area is freed.
1273 	 */
1274 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1275 
1276 	xen_smp_init();
1277 
1278 #ifdef CONFIG_ACPI_NUMA
1279 	/*
1280 	 * The pages we from Xen are not related to machine pages, so
1281 	 * any NUMA information the kernel tries to get from ACPI will
1282 	 * be meaningless.  Prevent it from trying.
1283 	 */
1284 	disable_srat();
1285 #endif
1286 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1287 
1288 	local_irq_disable();
1289 	early_boot_irqs_disabled = true;
1290 
1291 	xen_raw_console_write("mapping kernel into physical memory\n");
1292 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1293 				   xen_start_info->nr_pages);
1294 	xen_reserve_special_pages();
1295 
1296 	/*
1297 	 * We used to do this in xen_arch_setup, but that is too late
1298 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1299 	 * early_amd_init which pokes 0xcf8 port.
1300 	 */
1301 	set_iopl.iopl = 1;
1302 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1303 	if (rc != 0)
1304 		xen_raw_printk("physdev_op failed %d\n", rc);
1305 
1306 
1307 	if (xen_start_info->mod_start) {
1308 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1309 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1310 	    else
1311 		initrd_start = __pa(xen_start_info->mod_start);
1312 	}
1313 
1314 	/* Poke various useful things into boot_params */
1315 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1316 	boot_params.hdr.ramdisk_image = initrd_start;
1317 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1318 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1319 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1320 
1321 	if (!xen_initial_domain()) {
1322 		if (pci_xen)
1323 			x86_init.pci.arch_init = pci_xen_init;
1324 		x86_platform.set_legacy_features =
1325 				xen_domu_set_legacy_features;
1326 	} else {
1327 		const struct dom0_vga_console_info *info =
1328 			(void *)((char *)xen_start_info +
1329 				 xen_start_info->console.dom0.info_off);
1330 		struct xen_platform_op op = {
1331 			.cmd = XENPF_firmware_info,
1332 			.interface_version = XENPF_INTERFACE_VERSION,
1333 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1334 		};
1335 
1336 		x86_platform.set_legacy_features =
1337 				xen_dom0_set_legacy_features;
1338 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1339 		xen_start_info->console.domU.mfn = 0;
1340 		xen_start_info->console.domU.evtchn = 0;
1341 
1342 		if (HYPERVISOR_platform_op(&op) == 0)
1343 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1344 
1345 		/* Make sure ACS will be enabled */
1346 		pci_request_acs();
1347 
1348 		xen_acpi_sleep_register();
1349 
1350 		xen_boot_params_init_edd();
1351 
1352 #ifdef CONFIG_ACPI
1353 		/*
1354 		 * Disable selecting "Firmware First mode" for correctable
1355 		 * memory errors, as this is the duty of the hypervisor to
1356 		 * decide.
1357 		 */
1358 		acpi_disable_cmcff = 1;
1359 #endif
1360 	}
1361 
1362 	xen_add_preferred_consoles();
1363 
1364 #ifdef CONFIG_PCI
1365 	/* PCI BIOS service won't work from a PV guest. */
1366 	pci_probe &= ~PCI_PROBE_BIOS;
1367 #endif
1368 	xen_raw_console_write("about to get started...\n");
1369 
1370 	/* We need this for printk timestamps */
1371 	xen_setup_runstate_info(0);
1372 
1373 	xen_efi_init(&boot_params);
1374 
1375 	/* Start the world */
1376 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1377 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1378 }
1379 
1380 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1381 {
1382 	int rc;
1383 
1384 	if (per_cpu(xen_vcpu, cpu) == NULL)
1385 		return -ENODEV;
1386 
1387 	xen_setup_timer(cpu);
1388 
1389 	rc = xen_smp_intr_init(cpu);
1390 	if (rc) {
1391 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1392 		     cpu, rc);
1393 		return rc;
1394 	}
1395 
1396 	rc = xen_smp_intr_init_pv(cpu);
1397 	if (rc) {
1398 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1399 		     cpu, rc);
1400 		return rc;
1401 	}
1402 
1403 	return 0;
1404 }
1405 
1406 static int xen_cpu_dead_pv(unsigned int cpu)
1407 {
1408 	xen_smp_intr_free(cpu);
1409 	xen_smp_intr_free_pv(cpu);
1410 
1411 	xen_teardown_timer(cpu);
1412 
1413 	return 0;
1414 }
1415 
1416 static uint32_t __init xen_platform_pv(void)
1417 {
1418 	if (xen_pv_domain())
1419 		return xen_cpuid_base();
1420 
1421 	return 0;
1422 }
1423 
1424 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1425 	.name                   = "Xen PV",
1426 	.detect                 = xen_platform_pv,
1427 	.type			= X86_HYPER_XEN_PV,
1428 	.runtime.pin_vcpu       = xen_pin_vcpu,
1429 	.ignore_nopv		= true,
1430 };
1431