xref: /linux/arch/x86/xen/enlighten_pv.c (revision cf30f6a5f0c60ec98a637b836bef6915f602c6ab)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Core of Xen paravirt_ops implementation.
4  *
5  * This file contains the xen_paravirt_ops structure itself, and the
6  * implementations for:
7  * - privileged instructions
8  * - interrupt flags
9  * - segment operations
10  * - booting and setup
11  *
12  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
13  */
14 
15 #include <linux/cpu.h>
16 #include <linux/kernel.h>
17 #include <linux/init.h>
18 #include <linux/smp.h>
19 #include <linux/preempt.h>
20 #include <linux/hardirq.h>
21 #include <linux/percpu.h>
22 #include <linux/delay.h>
23 #include <linux/start_kernel.h>
24 #include <linux/sched.h>
25 #include <linux/kprobes.h>
26 #include <linux/memblock.h>
27 #include <linux/export.h>
28 #include <linux/mm.h>
29 #include <linux/page-flags.h>
30 #include <linux/highmem.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/edd.h>
34 #include <linux/objtool.h>
35 
36 #include <xen/xen.h>
37 #include <xen/events.h>
38 #include <xen/interface/xen.h>
39 #include <xen/interface/version.h>
40 #include <xen/interface/physdev.h>
41 #include <xen/interface/vcpu.h>
42 #include <xen/interface/memory.h>
43 #include <xen/interface/nmi.h>
44 #include <xen/interface/xen-mca.h>
45 #include <xen/features.h>
46 #include <xen/page.h>
47 #include <xen/hvc-console.h>
48 #include <xen/acpi.h>
49 
50 #include <asm/paravirt.h>
51 #include <asm/apic.h>
52 #include <asm/page.h>
53 #include <asm/xen/pci.h>
54 #include <asm/xen/hypercall.h>
55 #include <asm/xen/hypervisor.h>
56 #include <asm/xen/cpuid.h>
57 #include <asm/fixmap.h>
58 #include <asm/processor.h>
59 #include <asm/proto.h>
60 #include <asm/msr-index.h>
61 #include <asm/traps.h>
62 #include <asm/setup.h>
63 #include <asm/desc.h>
64 #include <asm/pgalloc.h>
65 #include <asm/tlbflush.h>
66 #include <asm/reboot.h>
67 #include <asm/stackprotector.h>
68 #include <asm/hypervisor.h>
69 #include <asm/mach_traps.h>
70 #include <asm/mwait.h>
71 #include <asm/pci_x86.h>
72 #include <asm/cpu.h>
73 #ifdef CONFIG_X86_IOPL_IOPERM
74 #include <asm/io_bitmap.h>
75 #endif
76 
77 #ifdef CONFIG_ACPI
78 #include <linux/acpi.h>
79 #include <asm/acpi.h>
80 #include <acpi/pdc_intel.h>
81 #include <acpi/processor.h>
82 #include <xen/interface/platform.h>
83 #endif
84 
85 #include "xen-ops.h"
86 #include "mmu.h"
87 #include "smp.h"
88 #include "multicalls.h"
89 #include "pmu.h"
90 
91 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
92 
93 void *xen_initial_gdt;
94 
95 static int xen_cpu_up_prepare_pv(unsigned int cpu);
96 static int xen_cpu_dead_pv(unsigned int cpu);
97 
98 struct tls_descs {
99 	struct desc_struct desc[3];
100 };
101 
102 /*
103  * Updating the 3 TLS descriptors in the GDT on every task switch is
104  * surprisingly expensive so we avoid updating them if they haven't
105  * changed.  Since Xen writes different descriptors than the one
106  * passed in the update_descriptor hypercall we keep shadow copies to
107  * compare against.
108  */
109 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
110 
111 static void __init xen_pv_init_platform(void)
112 {
113 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
114 
115 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
116 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
117 
118 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
119 	xen_vcpu_info_reset(0);
120 
121 	/* pvclock is in shared info area */
122 	xen_init_time_ops();
123 }
124 
125 static void __init xen_pv_guest_late_init(void)
126 {
127 #ifndef CONFIG_SMP
128 	/* Setup shared vcpu info for non-smp configurations */
129 	xen_setup_vcpu_info_placement();
130 #endif
131 }
132 
133 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
134 static __read_mostly unsigned int cpuid_leaf5_edx_val;
135 
136 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
137 		      unsigned int *cx, unsigned int *dx)
138 {
139 	unsigned maskebx = ~0;
140 
141 	/*
142 	 * Mask out inconvenient features, to try and disable as many
143 	 * unsupported kernel subsystems as possible.
144 	 */
145 	switch (*ax) {
146 	case CPUID_MWAIT_LEAF:
147 		/* Synthesize the values.. */
148 		*ax = 0;
149 		*bx = 0;
150 		*cx = cpuid_leaf5_ecx_val;
151 		*dx = cpuid_leaf5_edx_val;
152 		return;
153 
154 	case 0xb:
155 		/* Suppress extended topology stuff */
156 		maskebx = 0;
157 		break;
158 	}
159 
160 	asm(XEN_EMULATE_PREFIX "cpuid"
161 		: "=a" (*ax),
162 		  "=b" (*bx),
163 		  "=c" (*cx),
164 		  "=d" (*dx)
165 		: "0" (*ax), "2" (*cx));
166 
167 	*bx &= maskebx;
168 }
169 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
170 
171 static bool __init xen_check_mwait(void)
172 {
173 #ifdef CONFIG_ACPI
174 	struct xen_platform_op op = {
175 		.cmd			= XENPF_set_processor_pminfo,
176 		.u.set_pminfo.id	= -1,
177 		.u.set_pminfo.type	= XEN_PM_PDC,
178 	};
179 	uint32_t buf[3];
180 	unsigned int ax, bx, cx, dx;
181 	unsigned int mwait_mask;
182 
183 	/* We need to determine whether it is OK to expose the MWAIT
184 	 * capability to the kernel to harvest deeper than C3 states from ACPI
185 	 * _CST using the processor_harvest_xen.c module. For this to work, we
186 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
187 	 * checks against). The hypervisor won't expose the MWAIT flag because
188 	 * it would break backwards compatibility; so we will find out directly
189 	 * from the hardware and hypercall.
190 	 */
191 	if (!xen_initial_domain())
192 		return false;
193 
194 	/*
195 	 * When running under platform earlier than Xen4.2, do not expose
196 	 * mwait, to avoid the risk of loading native acpi pad driver
197 	 */
198 	if (!xen_running_on_version_or_later(4, 2))
199 		return false;
200 
201 	ax = 1;
202 	cx = 0;
203 
204 	native_cpuid(&ax, &bx, &cx, &dx);
205 
206 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
207 		     (1 << (X86_FEATURE_MWAIT % 32));
208 
209 	if ((cx & mwait_mask) != mwait_mask)
210 		return false;
211 
212 	/* We need to emulate the MWAIT_LEAF and for that we need both
213 	 * ecx and edx. The hypercall provides only partial information.
214 	 */
215 
216 	ax = CPUID_MWAIT_LEAF;
217 	bx = 0;
218 	cx = 0;
219 	dx = 0;
220 
221 	native_cpuid(&ax, &bx, &cx, &dx);
222 
223 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
224 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
225 	 */
226 	buf[0] = ACPI_PDC_REVISION_ID;
227 	buf[1] = 1;
228 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
229 
230 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
231 
232 	if ((HYPERVISOR_platform_op(&op) == 0) &&
233 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
234 		cpuid_leaf5_ecx_val = cx;
235 		cpuid_leaf5_edx_val = dx;
236 	}
237 	return true;
238 #else
239 	return false;
240 #endif
241 }
242 
243 static bool __init xen_check_xsave(void)
244 {
245 	unsigned int cx, xsave_mask;
246 
247 	cx = cpuid_ecx(1);
248 
249 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
250 		     (1 << (X86_FEATURE_OSXSAVE % 32));
251 
252 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
253 	return (cx & xsave_mask) == xsave_mask;
254 }
255 
256 static void __init xen_init_capabilities(void)
257 {
258 	setup_force_cpu_cap(X86_FEATURE_XENPV);
259 	setup_clear_cpu_cap(X86_FEATURE_DCA);
260 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
261 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
262 	setup_clear_cpu_cap(X86_FEATURE_ACC);
263 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
264 	setup_clear_cpu_cap(X86_FEATURE_SME);
265 
266 	/*
267 	 * Xen PV would need some work to support PCID: CR3 handling as well
268 	 * as xen_flush_tlb_others() would need updating.
269 	 */
270 	setup_clear_cpu_cap(X86_FEATURE_PCID);
271 
272 	if (!xen_initial_domain())
273 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
274 
275 	if (xen_check_mwait())
276 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
277 	else
278 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
279 
280 	if (!xen_check_xsave()) {
281 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
282 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
283 	}
284 }
285 
286 static noinstr void xen_set_debugreg(int reg, unsigned long val)
287 {
288 	HYPERVISOR_set_debugreg(reg, val);
289 }
290 
291 static noinstr unsigned long xen_get_debugreg(int reg)
292 {
293 	return HYPERVISOR_get_debugreg(reg);
294 }
295 
296 static void xen_end_context_switch(struct task_struct *next)
297 {
298 	xen_mc_flush();
299 	paravirt_end_context_switch(next);
300 }
301 
302 static unsigned long xen_store_tr(void)
303 {
304 	return 0;
305 }
306 
307 /*
308  * Set the page permissions for a particular virtual address.  If the
309  * address is a vmalloc mapping (or other non-linear mapping), then
310  * find the linear mapping of the page and also set its protections to
311  * match.
312  */
313 static void set_aliased_prot(void *v, pgprot_t prot)
314 {
315 	int level;
316 	pte_t *ptep;
317 	pte_t pte;
318 	unsigned long pfn;
319 	unsigned char dummy;
320 	void *va;
321 
322 	ptep = lookup_address((unsigned long)v, &level);
323 	BUG_ON(ptep == NULL);
324 
325 	pfn = pte_pfn(*ptep);
326 	pte = pfn_pte(pfn, prot);
327 
328 	/*
329 	 * Careful: update_va_mapping() will fail if the virtual address
330 	 * we're poking isn't populated in the page tables.  We don't
331 	 * need to worry about the direct map (that's always in the page
332 	 * tables), but we need to be careful about vmap space.  In
333 	 * particular, the top level page table can lazily propagate
334 	 * entries between processes, so if we've switched mms since we
335 	 * vmapped the target in the first place, we might not have the
336 	 * top-level page table entry populated.
337 	 *
338 	 * We disable preemption because we want the same mm active when
339 	 * we probe the target and when we issue the hypercall.  We'll
340 	 * have the same nominal mm, but if we're a kernel thread, lazy
341 	 * mm dropping could change our pgd.
342 	 *
343 	 * Out of an abundance of caution, this uses __get_user() to fault
344 	 * in the target address just in case there's some obscure case
345 	 * in which the target address isn't readable.
346 	 */
347 
348 	preempt_disable();
349 
350 	copy_from_kernel_nofault(&dummy, v, 1);
351 
352 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
353 		BUG();
354 
355 	va = __va(PFN_PHYS(pfn));
356 
357 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
358 		BUG();
359 
360 	preempt_enable();
361 }
362 
363 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
364 {
365 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
366 	int i;
367 
368 	/*
369 	 * We need to mark the all aliases of the LDT pages RO.  We
370 	 * don't need to call vm_flush_aliases(), though, since that's
371 	 * only responsible for flushing aliases out the TLBs, not the
372 	 * page tables, and Xen will flush the TLB for us if needed.
373 	 *
374 	 * To avoid confusing future readers: none of this is necessary
375 	 * to load the LDT.  The hypervisor only checks this when the
376 	 * LDT is faulted in due to subsequent descriptor access.
377 	 */
378 
379 	for (i = 0; i < entries; i += entries_per_page)
380 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
381 }
382 
383 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
384 {
385 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
386 	int i;
387 
388 	for (i = 0; i < entries; i += entries_per_page)
389 		set_aliased_prot(ldt + i, PAGE_KERNEL);
390 }
391 
392 static void xen_set_ldt(const void *addr, unsigned entries)
393 {
394 	struct mmuext_op *op;
395 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
396 
397 	trace_xen_cpu_set_ldt(addr, entries);
398 
399 	op = mcs.args;
400 	op->cmd = MMUEXT_SET_LDT;
401 	op->arg1.linear_addr = (unsigned long)addr;
402 	op->arg2.nr_ents = entries;
403 
404 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
405 
406 	xen_mc_issue(PARAVIRT_LAZY_CPU);
407 }
408 
409 static void xen_load_gdt(const struct desc_ptr *dtr)
410 {
411 	unsigned long va = dtr->address;
412 	unsigned int size = dtr->size + 1;
413 	unsigned long pfn, mfn;
414 	int level;
415 	pte_t *ptep;
416 	void *virt;
417 
418 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
419 	BUG_ON(size > PAGE_SIZE);
420 	BUG_ON(va & ~PAGE_MASK);
421 
422 	/*
423 	 * The GDT is per-cpu and is in the percpu data area.
424 	 * That can be virtually mapped, so we need to do a
425 	 * page-walk to get the underlying MFN for the
426 	 * hypercall.  The page can also be in the kernel's
427 	 * linear range, so we need to RO that mapping too.
428 	 */
429 	ptep = lookup_address(va, &level);
430 	BUG_ON(ptep == NULL);
431 
432 	pfn = pte_pfn(*ptep);
433 	mfn = pfn_to_mfn(pfn);
434 	virt = __va(PFN_PHYS(pfn));
435 
436 	make_lowmem_page_readonly((void *)va);
437 	make_lowmem_page_readonly(virt);
438 
439 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
440 		BUG();
441 }
442 
443 /*
444  * load_gdt for early boot, when the gdt is only mapped once
445  */
446 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
447 {
448 	unsigned long va = dtr->address;
449 	unsigned int size = dtr->size + 1;
450 	unsigned long pfn, mfn;
451 	pte_t pte;
452 
453 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
454 	BUG_ON(size > PAGE_SIZE);
455 	BUG_ON(va & ~PAGE_MASK);
456 
457 	pfn = virt_to_pfn(va);
458 	mfn = pfn_to_mfn(pfn);
459 
460 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
461 
462 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
463 		BUG();
464 
465 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
466 		BUG();
467 }
468 
469 static inline bool desc_equal(const struct desc_struct *d1,
470 			      const struct desc_struct *d2)
471 {
472 	return !memcmp(d1, d2, sizeof(*d1));
473 }
474 
475 static void load_TLS_descriptor(struct thread_struct *t,
476 				unsigned int cpu, unsigned int i)
477 {
478 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
479 	struct desc_struct *gdt;
480 	xmaddr_t maddr;
481 	struct multicall_space mc;
482 
483 	if (desc_equal(shadow, &t->tls_array[i]))
484 		return;
485 
486 	*shadow = t->tls_array[i];
487 
488 	gdt = get_cpu_gdt_rw(cpu);
489 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
490 	mc = __xen_mc_entry(0);
491 
492 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
493 }
494 
495 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
496 {
497 	/*
498 	 * In lazy mode we need to zero %fs, otherwise we may get an
499 	 * exception between the new %fs descriptor being loaded and
500 	 * %fs being effectively cleared at __switch_to().
501 	 */
502 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
503 		loadsegment(fs, 0);
504 
505 	xen_mc_batch();
506 
507 	load_TLS_descriptor(t, cpu, 0);
508 	load_TLS_descriptor(t, cpu, 1);
509 	load_TLS_descriptor(t, cpu, 2);
510 
511 	xen_mc_issue(PARAVIRT_LAZY_CPU);
512 }
513 
514 static void xen_load_gs_index(unsigned int idx)
515 {
516 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
517 		BUG();
518 }
519 
520 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
521 				const void *ptr)
522 {
523 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
524 	u64 entry = *(u64 *)ptr;
525 
526 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
527 
528 	preempt_disable();
529 
530 	xen_mc_flush();
531 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
532 		BUG();
533 
534 	preempt_enable();
535 }
536 
537 void noist_exc_debug(struct pt_regs *regs);
538 
539 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
540 {
541 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
542 	exc_nmi(regs);
543 }
544 
545 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
546 {
547 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
548 	exc_double_fault(regs, error_code);
549 }
550 
551 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
552 {
553 	/*
554 	 * There's no IST on Xen PV, but we still need to dispatch
555 	 * to the correct handler.
556 	 */
557 	if (user_mode(regs))
558 		noist_exc_debug(regs);
559 	else
560 		exc_debug(regs);
561 }
562 
563 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
564 {
565 	/* This should never happen and there is no way to handle it. */
566 	instrumentation_begin();
567 	pr_err("Unknown trap in Xen PV mode.");
568 	BUG();
569 	instrumentation_end();
570 }
571 
572 #ifdef CONFIG_X86_MCE
573 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
574 {
575 	/*
576 	 * There's no IST on Xen PV, but we still need to dispatch
577 	 * to the correct handler.
578 	 */
579 	if (user_mode(regs))
580 		noist_exc_machine_check(regs);
581 	else
582 		exc_machine_check(regs);
583 }
584 #endif
585 
586 struct trap_array_entry {
587 	void (*orig)(void);
588 	void (*xen)(void);
589 	bool ist_okay;
590 };
591 
592 #define TRAP_ENTRY(func, ist_ok) {			\
593 	.orig		= asm_##func,			\
594 	.xen		= xen_asm_##func,		\
595 	.ist_okay	= ist_ok }
596 
597 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
598 	.orig		= asm_##func,			\
599 	.xen		= xen_asm_xenpv_##func,		\
600 	.ist_okay	= ist_ok }
601 
602 static struct trap_array_entry trap_array[] = {
603 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
604 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
605 #ifdef CONFIG_X86_MCE
606 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
607 #endif
608 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
609 	TRAP_ENTRY(exc_int3,				false ),
610 	TRAP_ENTRY(exc_overflow,			false ),
611 #ifdef CONFIG_IA32_EMULATION
612 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
613 #endif
614 	TRAP_ENTRY(exc_page_fault,			false ),
615 	TRAP_ENTRY(exc_divide_error,			false ),
616 	TRAP_ENTRY(exc_bounds,				false ),
617 	TRAP_ENTRY(exc_invalid_op,			false ),
618 	TRAP_ENTRY(exc_device_not_available,		false ),
619 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
620 	TRAP_ENTRY(exc_invalid_tss,			false ),
621 	TRAP_ENTRY(exc_segment_not_present,		false ),
622 	TRAP_ENTRY(exc_stack_segment,			false ),
623 	TRAP_ENTRY(exc_general_protection,		false ),
624 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
625 	TRAP_ENTRY(exc_coprocessor_error,		false ),
626 	TRAP_ENTRY(exc_alignment_check,			false ),
627 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
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 
997 		/*
998 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
999 		 * falls back to the shared_info version for cpus
1000 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1001 		 *
1002 		 * xen_cpu_up_prepare_pv() handles the rest by failing
1003 		 * them in hotplug.
1004 		 */
1005 		(void) xen_vcpu_setup(cpu);
1006 	}
1007 
1008 	/*
1009 	 * xen_vcpu_setup managed to place the vcpu_info within the
1010 	 * percpu area for all cpus, so make use of it.
1011 	 */
1012 	if (xen_have_vcpu_info_placement) {
1013 		pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1014 		pv_ops.irq.irq_disable =
1015 			__PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1016 		pv_ops.irq.irq_enable =
1017 			__PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1018 		pv_ops.mmu.read_cr2 =
1019 			__PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1020 	}
1021 }
1022 
1023 static const struct pv_info xen_info __initconst = {
1024 	.extra_user_64bit_cs = FLAT_USER_CS64,
1025 	.name = "Xen",
1026 };
1027 
1028 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1029 	.cpu = {
1030 		.cpuid = xen_cpuid,
1031 
1032 		.set_debugreg = xen_set_debugreg,
1033 		.get_debugreg = xen_get_debugreg,
1034 
1035 		.read_cr0 = xen_read_cr0,
1036 		.write_cr0 = xen_write_cr0,
1037 
1038 		.write_cr4 = xen_write_cr4,
1039 
1040 		.wbinvd = native_wbinvd,
1041 
1042 		.read_msr = xen_read_msr,
1043 		.write_msr = xen_write_msr,
1044 
1045 		.read_msr_safe = xen_read_msr_safe,
1046 		.write_msr_safe = xen_write_msr_safe,
1047 
1048 		.read_pmc = xen_read_pmc,
1049 
1050 		.load_tr_desc = paravirt_nop,
1051 		.set_ldt = xen_set_ldt,
1052 		.load_gdt = xen_load_gdt,
1053 		.load_idt = xen_load_idt,
1054 		.load_tls = xen_load_tls,
1055 		.load_gs_index = xen_load_gs_index,
1056 
1057 		.alloc_ldt = xen_alloc_ldt,
1058 		.free_ldt = xen_free_ldt,
1059 
1060 		.store_tr = xen_store_tr,
1061 
1062 		.write_ldt_entry = xen_write_ldt_entry,
1063 		.write_gdt_entry = xen_write_gdt_entry,
1064 		.write_idt_entry = xen_write_idt_entry,
1065 		.load_sp0 = xen_load_sp0,
1066 
1067 #ifdef CONFIG_X86_IOPL_IOPERM
1068 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1069 		.update_io_bitmap = xen_update_io_bitmap,
1070 #endif
1071 		.io_delay = xen_io_delay,
1072 
1073 		.start_context_switch = paravirt_start_context_switch,
1074 		.end_context_switch = xen_end_context_switch,
1075 	},
1076 };
1077 
1078 static void xen_restart(char *msg)
1079 {
1080 	xen_reboot(SHUTDOWN_reboot);
1081 }
1082 
1083 static void xen_machine_halt(void)
1084 {
1085 	xen_reboot(SHUTDOWN_poweroff);
1086 }
1087 
1088 static void xen_machine_power_off(void)
1089 {
1090 	if (pm_power_off)
1091 		pm_power_off();
1092 	xen_reboot(SHUTDOWN_poweroff);
1093 }
1094 
1095 static void xen_crash_shutdown(struct pt_regs *regs)
1096 {
1097 	xen_reboot(SHUTDOWN_crash);
1098 }
1099 
1100 static const struct machine_ops xen_machine_ops __initconst = {
1101 	.restart = xen_restart,
1102 	.halt = xen_machine_halt,
1103 	.power_off = xen_machine_power_off,
1104 	.shutdown = xen_machine_halt,
1105 	.crash_shutdown = xen_crash_shutdown,
1106 	.emergency_restart = xen_emergency_restart,
1107 };
1108 
1109 static unsigned char xen_get_nmi_reason(void)
1110 {
1111 	unsigned char reason = 0;
1112 
1113 	/* Construct a value which looks like it came from port 0x61. */
1114 	if (test_bit(_XEN_NMIREASON_io_error,
1115 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1116 		reason |= NMI_REASON_IOCHK;
1117 	if (test_bit(_XEN_NMIREASON_pci_serr,
1118 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1119 		reason |= NMI_REASON_SERR;
1120 
1121 	return reason;
1122 }
1123 
1124 static void __init xen_boot_params_init_edd(void)
1125 {
1126 #if IS_ENABLED(CONFIG_EDD)
1127 	struct xen_platform_op op;
1128 	struct edd_info *edd_info;
1129 	u32 *mbr_signature;
1130 	unsigned nr;
1131 	int ret;
1132 
1133 	edd_info = boot_params.eddbuf;
1134 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1135 
1136 	op.cmd = XENPF_firmware_info;
1137 
1138 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1139 	for (nr = 0; nr < EDDMAXNR; nr++) {
1140 		struct edd_info *info = edd_info + nr;
1141 
1142 		op.u.firmware_info.index = nr;
1143 		info->params.length = sizeof(info->params);
1144 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1145 				     &info->params);
1146 		ret = HYPERVISOR_platform_op(&op);
1147 		if (ret)
1148 			break;
1149 
1150 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1151 		C(device);
1152 		C(version);
1153 		C(interface_support);
1154 		C(legacy_max_cylinder);
1155 		C(legacy_max_head);
1156 		C(legacy_sectors_per_track);
1157 #undef C
1158 	}
1159 	boot_params.eddbuf_entries = nr;
1160 
1161 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1162 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1163 		op.u.firmware_info.index = nr;
1164 		ret = HYPERVISOR_platform_op(&op);
1165 		if (ret)
1166 			break;
1167 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1168 	}
1169 	boot_params.edd_mbr_sig_buf_entries = nr;
1170 #endif
1171 }
1172 
1173 /*
1174  * Set up the GDT and segment registers for -fstack-protector.  Until
1175  * we do this, we have to be careful not to call any stack-protected
1176  * function, which is most of the kernel.
1177  */
1178 static void __init xen_setup_gdt(int cpu)
1179 {
1180 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1181 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1182 
1183 	switch_to_new_gdt(cpu);
1184 
1185 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1186 	pv_ops.cpu.load_gdt = xen_load_gdt;
1187 }
1188 
1189 static void __init xen_dom0_set_legacy_features(void)
1190 {
1191 	x86_platform.legacy.rtc = 1;
1192 }
1193 
1194 static void __init xen_domu_set_legacy_features(void)
1195 {
1196 	x86_platform.legacy.rtc = 0;
1197 }
1198 
1199 /* First C function to be called on Xen boot */
1200 asmlinkage __visible void __init xen_start_kernel(void)
1201 {
1202 	struct physdev_set_iopl set_iopl;
1203 	unsigned long initrd_start = 0;
1204 	int rc;
1205 
1206 	if (!xen_start_info)
1207 		return;
1208 
1209 	xen_domain_type = XEN_PV_DOMAIN;
1210 	xen_start_flags = xen_start_info->flags;
1211 
1212 	xen_setup_features();
1213 
1214 	/* Install Xen paravirt ops */
1215 	pv_info = xen_info;
1216 	pv_ops.cpu = xen_cpu_ops.cpu;
1217 	paravirt_iret = xen_iret;
1218 	xen_init_irq_ops();
1219 
1220 	/*
1221 	 * Setup xen_vcpu early because it is needed for
1222 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1223 	 *
1224 	 * Don't do the full vcpu_info placement stuff until we have
1225 	 * the cpu_possible_mask and a non-dummy shared_info.
1226 	 */
1227 	xen_vcpu_info_reset(0);
1228 
1229 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1230 
1231 	x86_init.resources.memory_setup = xen_memory_setup;
1232 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1233 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1234 	x86_init.oem.arch_setup = xen_arch_setup;
1235 	x86_init.oem.banner = xen_banner;
1236 	x86_init.hyper.init_platform = xen_pv_init_platform;
1237 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1238 
1239 	/*
1240 	 * Set up some pagetable state before starting to set any ptes.
1241 	 */
1242 
1243 	xen_setup_machphys_mapping();
1244 	xen_init_mmu_ops();
1245 
1246 	/* Prevent unwanted bits from being set in PTEs. */
1247 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1248 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1249 
1250 	/*
1251 	 * Prevent page tables from being allocated in highmem, even
1252 	 * if CONFIG_HIGHPTE is enabled.
1253 	 */
1254 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1255 
1256 	/* Get mfn list */
1257 	xen_build_dynamic_phys_to_machine();
1258 
1259 	/* Work out if we support NX */
1260 	get_cpu_cap(&boot_cpu_data);
1261 	x86_configure_nx();
1262 
1263 	/*
1264 	 * Set up kernel GDT and segment registers, mainly so that
1265 	 * -fstack-protector code can be executed.
1266 	 */
1267 	xen_setup_gdt(0);
1268 
1269 	/* Determine virtual and physical address sizes */
1270 	get_cpu_address_sizes(&boot_cpu_data);
1271 
1272 	/* Let's presume PV guests always boot on vCPU with id 0. */
1273 	per_cpu(xen_vcpu_id, 0) = 0;
1274 
1275 	idt_setup_early_handler();
1276 
1277 	xen_init_capabilities();
1278 
1279 #ifdef CONFIG_X86_LOCAL_APIC
1280 	/*
1281 	 * set up the basic apic ops.
1282 	 */
1283 	xen_init_apic();
1284 #endif
1285 
1286 	machine_ops = xen_machine_ops;
1287 
1288 	/*
1289 	 * The only reliable way to retain the initial address of the
1290 	 * percpu gdt_page is to remember it here, so we can go and
1291 	 * mark it RW later, when the initial percpu area is freed.
1292 	 */
1293 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1294 
1295 	xen_smp_init();
1296 
1297 #ifdef CONFIG_ACPI_NUMA
1298 	/*
1299 	 * The pages we from Xen are not related to machine pages, so
1300 	 * any NUMA information the kernel tries to get from ACPI will
1301 	 * be meaningless.  Prevent it from trying.
1302 	 */
1303 	disable_srat();
1304 #endif
1305 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1306 
1307 	local_irq_disable();
1308 	early_boot_irqs_disabled = true;
1309 
1310 	xen_raw_console_write("mapping kernel into physical memory\n");
1311 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1312 				   xen_start_info->nr_pages);
1313 	xen_reserve_special_pages();
1314 
1315 	/*
1316 	 * We used to do this in xen_arch_setup, but that is too late
1317 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1318 	 * early_amd_init which pokes 0xcf8 port.
1319 	 */
1320 	set_iopl.iopl = 1;
1321 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1322 	if (rc != 0)
1323 		xen_raw_printk("physdev_op failed %d\n", rc);
1324 
1325 
1326 	if (xen_start_info->mod_start) {
1327 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1328 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1329 	    else
1330 		initrd_start = __pa(xen_start_info->mod_start);
1331 	}
1332 
1333 	/* Poke various useful things into boot_params */
1334 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1335 	boot_params.hdr.ramdisk_image = initrd_start;
1336 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1337 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1338 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1339 
1340 	if (!xen_initial_domain()) {
1341 		if (pci_xen)
1342 			x86_init.pci.arch_init = pci_xen_init;
1343 		x86_platform.set_legacy_features =
1344 				xen_domu_set_legacy_features;
1345 	} else {
1346 		const struct dom0_vga_console_info *info =
1347 			(void *)((char *)xen_start_info +
1348 				 xen_start_info->console.dom0.info_off);
1349 		struct xen_platform_op op = {
1350 			.cmd = XENPF_firmware_info,
1351 			.interface_version = XENPF_INTERFACE_VERSION,
1352 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1353 		};
1354 
1355 		x86_platform.set_legacy_features =
1356 				xen_dom0_set_legacy_features;
1357 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1358 		xen_start_info->console.domU.mfn = 0;
1359 		xen_start_info->console.domU.evtchn = 0;
1360 
1361 		if (HYPERVISOR_platform_op(&op) == 0)
1362 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1363 
1364 		/* Make sure ACS will be enabled */
1365 		pci_request_acs();
1366 
1367 		xen_acpi_sleep_register();
1368 
1369 		/* Avoid searching for BIOS MP tables */
1370 		x86_init.mpparse.find_smp_config = x86_init_noop;
1371 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1372 
1373 		xen_boot_params_init_edd();
1374 
1375 #ifdef CONFIG_ACPI
1376 		/*
1377 		 * Disable selecting "Firmware First mode" for correctable
1378 		 * memory errors, as this is the duty of the hypervisor to
1379 		 * decide.
1380 		 */
1381 		acpi_disable_cmcff = 1;
1382 #endif
1383 	}
1384 
1385 	xen_add_preferred_consoles();
1386 
1387 #ifdef CONFIG_PCI
1388 	/* PCI BIOS service won't work from a PV guest. */
1389 	pci_probe &= ~PCI_PROBE_BIOS;
1390 #endif
1391 	xen_raw_console_write("about to get started...\n");
1392 
1393 	/* We need this for printk timestamps */
1394 	xen_setup_runstate_info(0);
1395 
1396 	xen_efi_init(&boot_params);
1397 
1398 	/* Start the world */
1399 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1400 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1401 }
1402 
1403 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1404 {
1405 	int rc;
1406 
1407 	if (per_cpu(xen_vcpu, cpu) == NULL)
1408 		return -ENODEV;
1409 
1410 	xen_setup_timer(cpu);
1411 
1412 	rc = xen_smp_intr_init(cpu);
1413 	if (rc) {
1414 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1415 		     cpu, rc);
1416 		return rc;
1417 	}
1418 
1419 	rc = xen_smp_intr_init_pv(cpu);
1420 	if (rc) {
1421 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1422 		     cpu, rc);
1423 		return rc;
1424 	}
1425 
1426 	return 0;
1427 }
1428 
1429 static int xen_cpu_dead_pv(unsigned int cpu)
1430 {
1431 	xen_smp_intr_free(cpu);
1432 	xen_smp_intr_free_pv(cpu);
1433 
1434 	xen_teardown_timer(cpu);
1435 
1436 	return 0;
1437 }
1438 
1439 static uint32_t __init xen_platform_pv(void)
1440 {
1441 	if (xen_pv_domain())
1442 		return xen_cpuid_base();
1443 
1444 	return 0;
1445 }
1446 
1447 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1448 	.name                   = "Xen PV",
1449 	.detect                 = xen_platform_pv,
1450 	.type			= X86_HYPER_XEN_PV,
1451 	.runtime.pin_vcpu       = xen_pin_vcpu,
1452 	.ignore_nopv		= true,
1453 };
1454