xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 920b0442)
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/reboot.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 
169 static bool __init xen_check_mwait(void)
170 {
171 #ifdef CONFIG_ACPI
172 	struct xen_platform_op op = {
173 		.cmd			= XENPF_set_processor_pminfo,
174 		.u.set_pminfo.id	= -1,
175 		.u.set_pminfo.type	= XEN_PM_PDC,
176 	};
177 	uint32_t buf[3];
178 	unsigned int ax, bx, cx, dx;
179 	unsigned int mwait_mask;
180 
181 	/* We need to determine whether it is OK to expose the MWAIT
182 	 * capability to the kernel to harvest deeper than C3 states from ACPI
183 	 * _CST using the processor_harvest_xen.c module. For this to work, we
184 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
185 	 * checks against). The hypervisor won't expose the MWAIT flag because
186 	 * it would break backwards compatibility; so we will find out directly
187 	 * from the hardware and hypercall.
188 	 */
189 	if (!xen_initial_domain())
190 		return false;
191 
192 	/*
193 	 * When running under platform earlier than Xen4.2, do not expose
194 	 * mwait, to avoid the risk of loading native acpi pad driver
195 	 */
196 	if (!xen_running_on_version_or_later(4, 2))
197 		return false;
198 
199 	ax = 1;
200 	cx = 0;
201 
202 	native_cpuid(&ax, &bx, &cx, &dx);
203 
204 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
205 		     (1 << (X86_FEATURE_MWAIT % 32));
206 
207 	if ((cx & mwait_mask) != mwait_mask)
208 		return false;
209 
210 	/* We need to emulate the MWAIT_LEAF and for that we need both
211 	 * ecx and edx. The hypercall provides only partial information.
212 	 */
213 
214 	ax = CPUID_MWAIT_LEAF;
215 	bx = 0;
216 	cx = 0;
217 	dx = 0;
218 
219 	native_cpuid(&ax, &bx, &cx, &dx);
220 
221 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
222 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
223 	 */
224 	buf[0] = ACPI_PDC_REVISION_ID;
225 	buf[1] = 1;
226 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
227 
228 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
229 
230 	if ((HYPERVISOR_platform_op(&op) == 0) &&
231 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
232 		cpuid_leaf5_ecx_val = cx;
233 		cpuid_leaf5_edx_val = dx;
234 	}
235 	return true;
236 #else
237 	return false;
238 #endif
239 }
240 
241 static bool __init xen_check_xsave(void)
242 {
243 	unsigned int cx, xsave_mask;
244 
245 	cx = cpuid_ecx(1);
246 
247 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
248 		     (1 << (X86_FEATURE_OSXSAVE % 32));
249 
250 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
251 	return (cx & xsave_mask) == xsave_mask;
252 }
253 
254 static void __init xen_init_capabilities(void)
255 {
256 	setup_force_cpu_cap(X86_FEATURE_XENPV);
257 	setup_clear_cpu_cap(X86_FEATURE_DCA);
258 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
259 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
260 	setup_clear_cpu_cap(X86_FEATURE_ACC);
261 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
262 	setup_clear_cpu_cap(X86_FEATURE_SME);
263 
264 	/*
265 	 * Xen PV would need some work to support PCID: CR3 handling as well
266 	 * as xen_flush_tlb_others() would need updating.
267 	 */
268 	setup_clear_cpu_cap(X86_FEATURE_PCID);
269 
270 	if (!xen_initial_domain())
271 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
272 
273 	if (xen_check_mwait())
274 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
275 	else
276 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
277 
278 	if (!xen_check_xsave()) {
279 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
280 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
281 	}
282 }
283 
284 static noinstr void xen_set_debugreg(int reg, unsigned long val)
285 {
286 	HYPERVISOR_set_debugreg(reg, val);
287 }
288 
289 static noinstr unsigned long xen_get_debugreg(int reg)
290 {
291 	return HYPERVISOR_get_debugreg(reg);
292 }
293 
294 static void xen_end_context_switch(struct task_struct *next)
295 {
296 	xen_mc_flush();
297 	paravirt_end_context_switch(next);
298 }
299 
300 static unsigned long xen_store_tr(void)
301 {
302 	return 0;
303 }
304 
305 /*
306  * Set the page permissions for a particular virtual address.  If the
307  * address is a vmalloc mapping (or other non-linear mapping), then
308  * find the linear mapping of the page and also set its protections to
309  * match.
310  */
311 static void set_aliased_prot(void *v, pgprot_t prot)
312 {
313 	int level;
314 	pte_t *ptep;
315 	pte_t pte;
316 	unsigned long pfn;
317 	unsigned char dummy;
318 	void *va;
319 
320 	ptep = lookup_address((unsigned long)v, &level);
321 	BUG_ON(ptep == NULL);
322 
323 	pfn = pte_pfn(*ptep);
324 	pte = pfn_pte(pfn, prot);
325 
326 	/*
327 	 * Careful: update_va_mapping() will fail if the virtual address
328 	 * we're poking isn't populated in the page tables.  We don't
329 	 * need to worry about the direct map (that's always in the page
330 	 * tables), but we need to be careful about vmap space.  In
331 	 * particular, the top level page table can lazily propagate
332 	 * entries between processes, so if we've switched mms since we
333 	 * vmapped the target in the first place, we might not have the
334 	 * top-level page table entry populated.
335 	 *
336 	 * We disable preemption because we want the same mm active when
337 	 * we probe the target and when we issue the hypercall.  We'll
338 	 * have the same nominal mm, but if we're a kernel thread, lazy
339 	 * mm dropping could change our pgd.
340 	 *
341 	 * Out of an abundance of caution, this uses __get_user() to fault
342 	 * in the target address just in case there's some obscure case
343 	 * in which the target address isn't readable.
344 	 */
345 
346 	preempt_disable();
347 
348 	copy_from_kernel_nofault(&dummy, v, 1);
349 
350 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
351 		BUG();
352 
353 	va = __va(PFN_PHYS(pfn));
354 
355 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
356 		BUG();
357 
358 	preempt_enable();
359 }
360 
361 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
362 {
363 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
364 	int i;
365 
366 	/*
367 	 * We need to mark the all aliases of the LDT pages RO.  We
368 	 * don't need to call vm_flush_aliases(), though, since that's
369 	 * only responsible for flushing aliases out the TLBs, not the
370 	 * page tables, and Xen will flush the TLB for us if needed.
371 	 *
372 	 * To avoid confusing future readers: none of this is necessary
373 	 * to load the LDT.  The hypervisor only checks this when the
374 	 * LDT is faulted in due to subsequent descriptor access.
375 	 */
376 
377 	for (i = 0; i < entries; i += entries_per_page)
378 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
379 }
380 
381 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
382 {
383 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
384 	int i;
385 
386 	for (i = 0; i < entries; i += entries_per_page)
387 		set_aliased_prot(ldt + i, PAGE_KERNEL);
388 }
389 
390 static void xen_set_ldt(const void *addr, unsigned entries)
391 {
392 	struct mmuext_op *op;
393 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
394 
395 	trace_xen_cpu_set_ldt(addr, entries);
396 
397 	op = mcs.args;
398 	op->cmd = MMUEXT_SET_LDT;
399 	op->arg1.linear_addr = (unsigned long)addr;
400 	op->arg2.nr_ents = entries;
401 
402 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
403 
404 	xen_mc_issue(PARAVIRT_LAZY_CPU);
405 }
406 
407 static void xen_load_gdt(const struct desc_ptr *dtr)
408 {
409 	unsigned long va = dtr->address;
410 	unsigned int size = dtr->size + 1;
411 	unsigned long pfn, mfn;
412 	int level;
413 	pte_t *ptep;
414 	void *virt;
415 
416 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
417 	BUG_ON(size > PAGE_SIZE);
418 	BUG_ON(va & ~PAGE_MASK);
419 
420 	/*
421 	 * The GDT is per-cpu and is in the percpu data area.
422 	 * That can be virtually mapped, so we need to do a
423 	 * page-walk to get the underlying MFN for the
424 	 * hypercall.  The page can also be in the kernel's
425 	 * linear range, so we need to RO that mapping too.
426 	 */
427 	ptep = lookup_address(va, &level);
428 	BUG_ON(ptep == NULL);
429 
430 	pfn = pte_pfn(*ptep);
431 	mfn = pfn_to_mfn(pfn);
432 	virt = __va(PFN_PHYS(pfn));
433 
434 	make_lowmem_page_readonly((void *)va);
435 	make_lowmem_page_readonly(virt);
436 
437 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
438 		BUG();
439 }
440 
441 /*
442  * load_gdt for early boot, when the gdt is only mapped once
443  */
444 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
445 {
446 	unsigned long va = dtr->address;
447 	unsigned int size = dtr->size + 1;
448 	unsigned long pfn, mfn;
449 	pte_t pte;
450 
451 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
452 	BUG_ON(size > PAGE_SIZE);
453 	BUG_ON(va & ~PAGE_MASK);
454 
455 	pfn = virt_to_pfn(va);
456 	mfn = pfn_to_mfn(pfn);
457 
458 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
459 
460 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
461 		BUG();
462 
463 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
464 		BUG();
465 }
466 
467 static inline bool desc_equal(const struct desc_struct *d1,
468 			      const struct desc_struct *d2)
469 {
470 	return !memcmp(d1, d2, sizeof(*d1));
471 }
472 
473 static void load_TLS_descriptor(struct thread_struct *t,
474 				unsigned int cpu, unsigned int i)
475 {
476 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
477 	struct desc_struct *gdt;
478 	xmaddr_t maddr;
479 	struct multicall_space mc;
480 
481 	if (desc_equal(shadow, &t->tls_array[i]))
482 		return;
483 
484 	*shadow = t->tls_array[i];
485 
486 	gdt = get_cpu_gdt_rw(cpu);
487 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
488 	mc = __xen_mc_entry(0);
489 
490 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
491 }
492 
493 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
494 {
495 	/*
496 	 * In lazy mode we need to zero %fs, otherwise we may get an
497 	 * exception between the new %fs descriptor being loaded and
498 	 * %fs being effectively cleared at __switch_to().
499 	 */
500 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
501 		loadsegment(fs, 0);
502 
503 	xen_mc_batch();
504 
505 	load_TLS_descriptor(t, cpu, 0);
506 	load_TLS_descriptor(t, cpu, 1);
507 	load_TLS_descriptor(t, cpu, 2);
508 
509 	xen_mc_issue(PARAVIRT_LAZY_CPU);
510 }
511 
512 static void xen_load_gs_index(unsigned int idx)
513 {
514 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
515 		BUG();
516 }
517 
518 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
519 				const void *ptr)
520 {
521 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
522 	u64 entry = *(u64 *)ptr;
523 
524 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
525 
526 	preempt_disable();
527 
528 	xen_mc_flush();
529 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
530 		BUG();
531 
532 	preempt_enable();
533 }
534 
535 void noist_exc_debug(struct pt_regs *regs);
536 
537 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
538 {
539 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
540 	exc_nmi(regs);
541 }
542 
543 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
544 {
545 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
546 	exc_double_fault(regs, error_code);
547 }
548 
549 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
550 {
551 	/*
552 	 * There's no IST on Xen PV, but we still need to dispatch
553 	 * to the correct handler.
554 	 */
555 	if (user_mode(regs))
556 		noist_exc_debug(regs);
557 	else
558 		exc_debug(regs);
559 }
560 
561 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
562 {
563 	/* This should never happen and there is no way to handle it. */
564 	instrumentation_begin();
565 	pr_err("Unknown trap in Xen PV mode.");
566 	BUG();
567 	instrumentation_end();
568 }
569 
570 #ifdef CONFIG_X86_MCE
571 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
572 {
573 	/*
574 	 * There's no IST on Xen PV, but we still need to dispatch
575 	 * to the correct handler.
576 	 */
577 	if (user_mode(regs))
578 		noist_exc_machine_check(regs);
579 	else
580 		exc_machine_check(regs);
581 }
582 #endif
583 
584 struct trap_array_entry {
585 	void (*orig)(void);
586 	void (*xen)(void);
587 	bool ist_okay;
588 };
589 
590 #define TRAP_ENTRY(func, ist_ok) {			\
591 	.orig		= asm_##func,			\
592 	.xen		= xen_asm_##func,		\
593 	.ist_okay	= ist_ok }
594 
595 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
596 	.orig		= asm_##func,			\
597 	.xen		= xen_asm_xenpv_##func,		\
598 	.ist_okay	= ist_ok }
599 
600 static struct trap_array_entry trap_array[] = {
601 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
602 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
603 #ifdef CONFIG_X86_MCE
604 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
605 #endif
606 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
607 	TRAP_ENTRY(exc_int3,				false ),
608 	TRAP_ENTRY(exc_overflow,			false ),
609 #ifdef CONFIG_IA32_EMULATION
610 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
611 #endif
612 	TRAP_ENTRY(exc_page_fault,			false ),
613 	TRAP_ENTRY(exc_divide_error,			false ),
614 	TRAP_ENTRY(exc_bounds,				false ),
615 	TRAP_ENTRY(exc_invalid_op,			false ),
616 	TRAP_ENTRY(exc_device_not_available,		false ),
617 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
618 	TRAP_ENTRY(exc_invalid_tss,			false ),
619 	TRAP_ENTRY(exc_segment_not_present,		false ),
620 	TRAP_ENTRY(exc_stack_segment,			false ),
621 	TRAP_ENTRY(exc_general_protection,		false ),
622 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
623 	TRAP_ENTRY(exc_coprocessor_error,		false ),
624 	TRAP_ENTRY(exc_alignment_check,			false ),
625 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
626 #ifdef CONFIG_X86_KERNEL_IBT
627 	TRAP_ENTRY(exc_control_protection,		false ),
628 #endif
629 };
630 
631 static bool __ref get_trap_addr(void **addr, unsigned int ist)
632 {
633 	unsigned int nr;
634 	bool ist_okay = false;
635 	bool found = false;
636 
637 	/*
638 	 * Replace trap handler addresses by Xen specific ones.
639 	 * Check for known traps using IST and whitelist them.
640 	 * The debugger ones are the only ones we care about.
641 	 * Xen will handle faults like double_fault, so we should never see
642 	 * them.  Warn if there's an unexpected IST-using fault handler.
643 	 */
644 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
645 		struct trap_array_entry *entry = trap_array + nr;
646 
647 		if (*addr == entry->orig) {
648 			*addr = entry->xen;
649 			ist_okay = entry->ist_okay;
650 			found = true;
651 			break;
652 		}
653 	}
654 
655 	if (nr == ARRAY_SIZE(trap_array) &&
656 	    *addr >= (void *)early_idt_handler_array[0] &&
657 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
658 		nr = (*addr - (void *)early_idt_handler_array[0]) /
659 		     EARLY_IDT_HANDLER_SIZE;
660 		*addr = (void *)xen_early_idt_handler_array[nr];
661 		found = true;
662 	}
663 
664 	if (!found)
665 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
666 
667 	if (WARN_ON(found && ist != 0 && !ist_okay))
668 		return false;
669 
670 	return true;
671 }
672 
673 static int cvt_gate_to_trap(int vector, const gate_desc *val,
674 			    struct trap_info *info)
675 {
676 	unsigned long addr;
677 
678 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
679 		return 0;
680 
681 	info->vector = vector;
682 
683 	addr = gate_offset(val);
684 	if (!get_trap_addr((void **)&addr, val->bits.ist))
685 		return 0;
686 	info->address = addr;
687 
688 	info->cs = gate_segment(val);
689 	info->flags = val->bits.dpl;
690 	/* interrupt gates clear IF */
691 	if (val->bits.type == GATE_INTERRUPT)
692 		info->flags |= 1 << 2;
693 
694 	return 1;
695 }
696 
697 /* Locations of each CPU's IDT */
698 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
699 
700 /* Set an IDT entry.  If the entry is part of the current IDT, then
701    also update Xen. */
702 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
703 {
704 	unsigned long p = (unsigned long)&dt[entrynum];
705 	unsigned long start, end;
706 
707 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
708 
709 	preempt_disable();
710 
711 	start = __this_cpu_read(idt_desc.address);
712 	end = start + __this_cpu_read(idt_desc.size) + 1;
713 
714 	xen_mc_flush();
715 
716 	native_write_idt_entry(dt, entrynum, g);
717 
718 	if (p >= start && (p + 8) <= end) {
719 		struct trap_info info[2];
720 
721 		info[1].address = 0;
722 
723 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
724 			if (HYPERVISOR_set_trap_table(info))
725 				BUG();
726 	}
727 
728 	preempt_enable();
729 }
730 
731 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
732 				      struct trap_info *traps, bool full)
733 {
734 	unsigned in, out, count;
735 
736 	count = (desc->size+1) / sizeof(gate_desc);
737 	BUG_ON(count > 256);
738 
739 	for (in = out = 0; in < count; in++) {
740 		gate_desc *entry = (gate_desc *)(desc->address) + in;
741 
742 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
743 			out++;
744 	}
745 
746 	return out;
747 }
748 
749 void xen_copy_trap_info(struct trap_info *traps)
750 {
751 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
752 
753 	xen_convert_trap_info(desc, traps, true);
754 }
755 
756 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
757    hold a spinlock to protect the static traps[] array (static because
758    it avoids allocation, and saves stack space). */
759 static void xen_load_idt(const struct desc_ptr *desc)
760 {
761 	static DEFINE_SPINLOCK(lock);
762 	static struct trap_info traps[257];
763 	unsigned out;
764 
765 	trace_xen_cpu_load_idt(desc);
766 
767 	spin_lock(&lock);
768 
769 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
770 
771 	out = xen_convert_trap_info(desc, traps, false);
772 	memset(&traps[out], 0, sizeof(traps[0]));
773 
774 	xen_mc_flush();
775 	if (HYPERVISOR_set_trap_table(traps))
776 		BUG();
777 
778 	spin_unlock(&lock);
779 }
780 
781 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
782    they're handled differently. */
783 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
784 				const void *desc, int type)
785 {
786 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
787 
788 	preempt_disable();
789 
790 	switch (type) {
791 	case DESC_LDT:
792 	case DESC_TSS:
793 		/* ignore */
794 		break;
795 
796 	default: {
797 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
798 
799 		xen_mc_flush();
800 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
801 			BUG();
802 	}
803 
804 	}
805 
806 	preempt_enable();
807 }
808 
809 /*
810  * Version of write_gdt_entry for use at early boot-time needed to
811  * update an entry as simply as possible.
812  */
813 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
814 					    const void *desc, int type)
815 {
816 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
817 
818 	switch (type) {
819 	case DESC_LDT:
820 	case DESC_TSS:
821 		/* ignore */
822 		break;
823 
824 	default: {
825 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
826 
827 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
828 			dt[entry] = *(struct desc_struct *)desc;
829 	}
830 
831 	}
832 }
833 
834 static void xen_load_sp0(unsigned long sp0)
835 {
836 	struct multicall_space mcs;
837 
838 	mcs = xen_mc_entry(0);
839 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
840 	xen_mc_issue(PARAVIRT_LAZY_CPU);
841 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
842 }
843 
844 #ifdef CONFIG_X86_IOPL_IOPERM
845 static void xen_invalidate_io_bitmap(void)
846 {
847 	struct physdev_set_iobitmap iobitmap = {
848 		.bitmap = NULL,
849 		.nr_ports = 0,
850 	};
851 
852 	native_tss_invalidate_io_bitmap();
853 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
854 }
855 
856 static void xen_update_io_bitmap(void)
857 {
858 	struct physdev_set_iobitmap iobitmap;
859 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
860 
861 	native_tss_update_io_bitmap();
862 
863 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
864 			  tss->x86_tss.io_bitmap_base;
865 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
866 		iobitmap.nr_ports = 0;
867 	else
868 		iobitmap.nr_ports = IO_BITMAP_BITS;
869 
870 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
871 }
872 #endif
873 
874 static void xen_io_delay(void)
875 {
876 }
877 
878 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
879 
880 static unsigned long xen_read_cr0(void)
881 {
882 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
883 
884 	if (unlikely(cr0 == 0)) {
885 		cr0 = native_read_cr0();
886 		this_cpu_write(xen_cr0_value, cr0);
887 	}
888 
889 	return cr0;
890 }
891 
892 static void xen_write_cr0(unsigned long cr0)
893 {
894 	struct multicall_space mcs;
895 
896 	this_cpu_write(xen_cr0_value, cr0);
897 
898 	/* Only pay attention to cr0.TS; everything else is
899 	   ignored. */
900 	mcs = xen_mc_entry(0);
901 
902 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
903 
904 	xen_mc_issue(PARAVIRT_LAZY_CPU);
905 }
906 
907 static void xen_write_cr4(unsigned long cr4)
908 {
909 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
910 
911 	native_write_cr4(cr4);
912 }
913 
914 static u64 xen_read_msr_safe(unsigned int msr, int *err)
915 {
916 	u64 val;
917 
918 	if (pmu_msr_read(msr, &val, err))
919 		return val;
920 
921 	val = native_read_msr_safe(msr, err);
922 	switch (msr) {
923 	case MSR_IA32_APICBASE:
924 		val &= ~X2APIC_ENABLE;
925 		break;
926 	}
927 	return val;
928 }
929 
930 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
931 {
932 	int ret;
933 	unsigned int which;
934 	u64 base;
935 
936 	ret = 0;
937 
938 	switch (msr) {
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 
949 	case MSR_STAR:
950 	case MSR_CSTAR:
951 	case MSR_LSTAR:
952 	case MSR_SYSCALL_MASK:
953 	case MSR_IA32_SYSENTER_CS:
954 	case MSR_IA32_SYSENTER_ESP:
955 	case MSR_IA32_SYSENTER_EIP:
956 		/* Fast syscall setup is all done in hypercalls, so
957 		   these are all ignored.  Stub them out here to stop
958 		   Xen console noise. */
959 		break;
960 
961 	default:
962 		if (!pmu_msr_write(msr, low, high, &ret))
963 			ret = native_write_msr_safe(msr, low, high);
964 	}
965 
966 	return ret;
967 }
968 
969 static u64 xen_read_msr(unsigned int msr)
970 {
971 	/*
972 	 * This will silently swallow a #GP from RDMSR.  It may be worth
973 	 * changing that.
974 	 */
975 	int err;
976 
977 	return xen_read_msr_safe(msr, &err);
978 }
979 
980 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
981 {
982 	/*
983 	 * This will silently swallow a #GP from WRMSR.  It may be worth
984 	 * changing that.
985 	 */
986 	xen_write_msr_safe(msr, low, high);
987 }
988 
989 /* This is called once we have the cpu_possible_mask */
990 void __init xen_setup_vcpu_info_placement(void)
991 {
992 	int cpu;
993 
994 	for_each_possible_cpu(cpu) {
995 		/* Set up direct vCPU id mapping for PV guests. */
996 		per_cpu(xen_vcpu_id, cpu) = cpu;
997 		xen_vcpu_setup(cpu);
998 	}
999 
1000 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1001 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1002 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1003 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1004 }
1005 
1006 static const struct pv_info xen_info __initconst = {
1007 	.extra_user_64bit_cs = FLAT_USER_CS64,
1008 	.name = "Xen",
1009 };
1010 
1011 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1012 	.cpu = {
1013 		.cpuid = xen_cpuid,
1014 
1015 		.set_debugreg = xen_set_debugreg,
1016 		.get_debugreg = xen_get_debugreg,
1017 
1018 		.read_cr0 = xen_read_cr0,
1019 		.write_cr0 = xen_write_cr0,
1020 
1021 		.write_cr4 = xen_write_cr4,
1022 
1023 		.wbinvd = native_wbinvd,
1024 
1025 		.read_msr = xen_read_msr,
1026 		.write_msr = xen_write_msr,
1027 
1028 		.read_msr_safe = xen_read_msr_safe,
1029 		.write_msr_safe = xen_write_msr_safe,
1030 
1031 		.read_pmc = xen_read_pmc,
1032 
1033 		.load_tr_desc = paravirt_nop,
1034 		.set_ldt = xen_set_ldt,
1035 		.load_gdt = xen_load_gdt,
1036 		.load_idt = xen_load_idt,
1037 		.load_tls = xen_load_tls,
1038 		.load_gs_index = xen_load_gs_index,
1039 
1040 		.alloc_ldt = xen_alloc_ldt,
1041 		.free_ldt = xen_free_ldt,
1042 
1043 		.store_tr = xen_store_tr,
1044 
1045 		.write_ldt_entry = xen_write_ldt_entry,
1046 		.write_gdt_entry = xen_write_gdt_entry,
1047 		.write_idt_entry = xen_write_idt_entry,
1048 		.load_sp0 = xen_load_sp0,
1049 
1050 #ifdef CONFIG_X86_IOPL_IOPERM
1051 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1052 		.update_io_bitmap = xen_update_io_bitmap,
1053 #endif
1054 		.io_delay = xen_io_delay,
1055 
1056 		.start_context_switch = paravirt_start_context_switch,
1057 		.end_context_switch = xen_end_context_switch,
1058 	},
1059 };
1060 
1061 static void xen_restart(char *msg)
1062 {
1063 	xen_reboot(SHUTDOWN_reboot);
1064 }
1065 
1066 static void xen_machine_halt(void)
1067 {
1068 	xen_reboot(SHUTDOWN_poweroff);
1069 }
1070 
1071 static void xen_machine_power_off(void)
1072 {
1073 	do_kernel_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