xref: /linux/arch/x86/xen/enlighten_pv.c (revision d8e473182ab9e85708067be81d20424045d939fa)
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/kstrtox.h>
27 #include <linux/memblock.h>
28 #include <linux/export.h>
29 #include <linux/mm.h>
30 #include <linux/page-flags.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/edd.h>
34 #include <linux/reboot.h>
35 #include <linux/virtio_anchor.h>
36 #include <linux/stackprotector.h>
37 
38 #include <xen/xen.h>
39 #include <xen/events.h>
40 #include <xen/interface/xen.h>
41 #include <xen/interface/version.h>
42 #include <xen/interface/physdev.h>
43 #include <xen/interface/vcpu.h>
44 #include <xen/interface/memory.h>
45 #include <xen/interface/nmi.h>
46 #include <xen/interface/xen-mca.h>
47 #include <xen/features.h>
48 #include <xen/page.h>
49 #include <xen/hvc-console.h>
50 #include <xen/acpi.h>
51 
52 #include <asm/paravirt.h>
53 #include <asm/apic.h>
54 #include <asm/page.h>
55 #include <asm/xen/pci.h>
56 #include <asm/xen/hypercall.h>
57 #include <asm/xen/hypervisor.h>
58 #include <asm/xen/cpuid.h>
59 #include <asm/fixmap.h>
60 #include <asm/processor.h>
61 #include <asm/proto.h>
62 #include <asm/msr-index.h>
63 #include <asm/traps.h>
64 #include <asm/setup.h>
65 #include <asm/desc.h>
66 #include <asm/pgalloc.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/hypervisor.h>
70 #include <asm/mach_traps.h>
71 #include <asm/mwait.h>
72 #include <asm/pci_x86.h>
73 #include <asm/cpu.h>
74 #ifdef CONFIG_X86_IOPL_IOPERM
75 #include <asm/io_bitmap.h>
76 #endif
77 
78 #ifdef CONFIG_ACPI
79 #include <linux/acpi.h>
80 #include <asm/acpi.h>
81 #include <acpi/pdc_intel.h>
82 #include <acpi/processor.h>
83 #include <xen/interface/platform.h>
84 #endif
85 
86 #include "xen-ops.h"
87 #include "mmu.h"
88 #include "smp.h"
89 #include "multicalls.h"
90 #include "pmu.h"
91 
92 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
93 
94 void *xen_initial_gdt;
95 
96 static int xen_cpu_up_prepare_pv(unsigned int cpu);
97 static int xen_cpu_dead_pv(unsigned int cpu);
98 
99 struct tls_descs {
100 	struct desc_struct desc[3];
101 };
102 
103 /*
104  * Updating the 3 TLS descriptors in the GDT on every task switch is
105  * surprisingly expensive so we avoid updating them if they haven't
106  * changed.  Since Xen writes different descriptors than the one
107  * passed in the update_descriptor hypercall we keep shadow copies to
108  * compare against.
109  */
110 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
111 
112 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
113 
114 static int __init parse_xen_msr_safe(char *str)
115 {
116 	if (str)
117 		return kstrtobool(str, &xen_msr_safe);
118 	return -EINVAL;
119 }
120 early_param("xen_msr_safe", parse_xen_msr_safe);
121 
122 static void __init xen_pv_init_platform(void)
123 {
124 	/* PV guests can't operate virtio devices without grants. */
125 	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
126 		virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
127 
128 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
129 
130 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
131 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
132 
133 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
134 	xen_vcpu_info_reset(0);
135 
136 	/* pvclock is in shared info area */
137 	xen_init_time_ops();
138 }
139 
140 static void __init xen_pv_guest_late_init(void)
141 {
142 #ifndef CONFIG_SMP
143 	/* Setup shared vcpu info for non-smp configurations */
144 	xen_setup_vcpu_info_placement();
145 #endif
146 }
147 
148 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
149 static __read_mostly unsigned int cpuid_leaf5_edx_val;
150 
151 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
152 		      unsigned int *cx, unsigned int *dx)
153 {
154 	unsigned maskebx = ~0;
155 
156 	/*
157 	 * Mask out inconvenient features, to try and disable as many
158 	 * unsupported kernel subsystems as possible.
159 	 */
160 	switch (*ax) {
161 	case CPUID_MWAIT_LEAF:
162 		/* Synthesize the values.. */
163 		*ax = 0;
164 		*bx = 0;
165 		*cx = cpuid_leaf5_ecx_val;
166 		*dx = cpuid_leaf5_edx_val;
167 		return;
168 
169 	case 0xb:
170 		/* Suppress extended topology stuff */
171 		maskebx = 0;
172 		break;
173 	}
174 
175 	asm(XEN_EMULATE_PREFIX "cpuid"
176 		: "=a" (*ax),
177 		  "=b" (*bx),
178 		  "=c" (*cx),
179 		  "=d" (*dx)
180 		: "0" (*ax), "2" (*cx));
181 
182 	*bx &= maskebx;
183 }
184 
185 static bool __init xen_check_mwait(void)
186 {
187 #ifdef CONFIG_ACPI
188 	struct xen_platform_op op = {
189 		.cmd			= XENPF_set_processor_pminfo,
190 		.u.set_pminfo.id	= -1,
191 		.u.set_pminfo.type	= XEN_PM_PDC,
192 	};
193 	uint32_t buf[3];
194 	unsigned int ax, bx, cx, dx;
195 	unsigned int mwait_mask;
196 
197 	/* We need to determine whether it is OK to expose the MWAIT
198 	 * capability to the kernel to harvest deeper than C3 states from ACPI
199 	 * _CST using the processor_harvest_xen.c module. For this to work, we
200 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
201 	 * checks against). The hypervisor won't expose the MWAIT flag because
202 	 * it would break backwards compatibility; so we will find out directly
203 	 * from the hardware and hypercall.
204 	 */
205 	if (!xen_initial_domain())
206 		return false;
207 
208 	/*
209 	 * When running under platform earlier than Xen4.2, do not expose
210 	 * mwait, to avoid the risk of loading native acpi pad driver
211 	 */
212 	if (!xen_running_on_version_or_later(4, 2))
213 		return false;
214 
215 	ax = 1;
216 	cx = 0;
217 
218 	native_cpuid(&ax, &bx, &cx, &dx);
219 
220 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
221 		     (1 << (X86_FEATURE_MWAIT % 32));
222 
223 	if ((cx & mwait_mask) != mwait_mask)
224 		return false;
225 
226 	/* We need to emulate the MWAIT_LEAF and for that we need both
227 	 * ecx and edx. The hypercall provides only partial information.
228 	 */
229 
230 	ax = CPUID_MWAIT_LEAF;
231 	bx = 0;
232 	cx = 0;
233 	dx = 0;
234 
235 	native_cpuid(&ax, &bx, &cx, &dx);
236 
237 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
238 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
239 	 */
240 	buf[0] = ACPI_PDC_REVISION_ID;
241 	buf[1] = 1;
242 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
243 
244 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
245 
246 	if ((HYPERVISOR_platform_op(&op) == 0) &&
247 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
248 		cpuid_leaf5_ecx_val = cx;
249 		cpuid_leaf5_edx_val = dx;
250 	}
251 	return true;
252 #else
253 	return false;
254 #endif
255 }
256 
257 static bool __init xen_check_xsave(void)
258 {
259 	unsigned int cx, xsave_mask;
260 
261 	cx = cpuid_ecx(1);
262 
263 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
264 		     (1 << (X86_FEATURE_OSXSAVE % 32));
265 
266 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
267 	return (cx & xsave_mask) == xsave_mask;
268 }
269 
270 static void __init xen_init_capabilities(void)
271 {
272 	setup_force_cpu_cap(X86_FEATURE_XENPV);
273 	setup_clear_cpu_cap(X86_FEATURE_DCA);
274 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
275 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
276 	setup_clear_cpu_cap(X86_FEATURE_ACC);
277 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
278 	setup_clear_cpu_cap(X86_FEATURE_SME);
279 	setup_clear_cpu_cap(X86_FEATURE_LKGS);
280 
281 	/*
282 	 * Xen PV would need some work to support PCID: CR3 handling as well
283 	 * as xen_flush_tlb_others() would need updating.
284 	 */
285 	setup_clear_cpu_cap(X86_FEATURE_PCID);
286 
287 	if (!xen_initial_domain())
288 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
289 
290 	if (xen_check_mwait())
291 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
292 	else
293 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
294 
295 	if (!xen_check_xsave()) {
296 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
297 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
298 	}
299 }
300 
301 static noinstr void xen_set_debugreg(int reg, unsigned long val)
302 {
303 	HYPERVISOR_set_debugreg(reg, val);
304 }
305 
306 static noinstr unsigned long xen_get_debugreg(int reg)
307 {
308 	return HYPERVISOR_get_debugreg(reg);
309 }
310 
311 static void xen_end_context_switch(struct task_struct *next)
312 {
313 	xen_mc_flush();
314 	paravirt_end_context_switch(next);
315 }
316 
317 static unsigned long xen_store_tr(void)
318 {
319 	return 0;
320 }
321 
322 /*
323  * Set the page permissions for a particular virtual address.  If the
324  * address is a vmalloc mapping (or other non-linear mapping), then
325  * find the linear mapping of the page and also set its protections to
326  * match.
327  */
328 static void set_aliased_prot(void *v, pgprot_t prot)
329 {
330 	int level;
331 	pte_t *ptep;
332 	pte_t pte;
333 	unsigned long pfn;
334 	unsigned char dummy;
335 	void *va;
336 
337 	ptep = lookup_address((unsigned long)v, &level);
338 	BUG_ON(ptep == NULL);
339 
340 	pfn = pte_pfn(*ptep);
341 	pte = pfn_pte(pfn, prot);
342 
343 	/*
344 	 * Careful: update_va_mapping() will fail if the virtual address
345 	 * we're poking isn't populated in the page tables.  We don't
346 	 * need to worry about the direct map (that's always in the page
347 	 * tables), but we need to be careful about vmap space.  In
348 	 * particular, the top level page table can lazily propagate
349 	 * entries between processes, so if we've switched mms since we
350 	 * vmapped the target in the first place, we might not have the
351 	 * top-level page table entry populated.
352 	 *
353 	 * We disable preemption because we want the same mm active when
354 	 * we probe the target and when we issue the hypercall.  We'll
355 	 * have the same nominal mm, but if we're a kernel thread, lazy
356 	 * mm dropping could change our pgd.
357 	 *
358 	 * Out of an abundance of caution, this uses __get_user() to fault
359 	 * in the target address just in case there's some obscure case
360 	 * in which the target address isn't readable.
361 	 */
362 
363 	preempt_disable();
364 
365 	copy_from_kernel_nofault(&dummy, v, 1);
366 
367 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
368 		BUG();
369 
370 	va = __va(PFN_PHYS(pfn));
371 
372 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
373 		BUG();
374 
375 	preempt_enable();
376 }
377 
378 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
379 {
380 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
381 	int i;
382 
383 	/*
384 	 * We need to mark the all aliases of the LDT pages RO.  We
385 	 * don't need to call vm_flush_aliases(), though, since that's
386 	 * only responsible for flushing aliases out the TLBs, not the
387 	 * page tables, and Xen will flush the TLB for us if needed.
388 	 *
389 	 * To avoid confusing future readers: none of this is necessary
390 	 * to load the LDT.  The hypervisor only checks this when the
391 	 * LDT is faulted in due to subsequent descriptor access.
392 	 */
393 
394 	for (i = 0; i < entries; i += entries_per_page)
395 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
396 }
397 
398 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
399 {
400 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
401 	int i;
402 
403 	for (i = 0; i < entries; i += entries_per_page)
404 		set_aliased_prot(ldt + i, PAGE_KERNEL);
405 }
406 
407 static void xen_set_ldt(const void *addr, unsigned entries)
408 {
409 	struct mmuext_op *op;
410 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
411 
412 	trace_xen_cpu_set_ldt(addr, entries);
413 
414 	op = mcs.args;
415 	op->cmd = MMUEXT_SET_LDT;
416 	op->arg1.linear_addr = (unsigned long)addr;
417 	op->arg2.nr_ents = entries;
418 
419 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
420 
421 	xen_mc_issue(PARAVIRT_LAZY_CPU);
422 }
423 
424 static void xen_load_gdt(const struct desc_ptr *dtr)
425 {
426 	unsigned long va = dtr->address;
427 	unsigned int size = dtr->size + 1;
428 	unsigned long pfn, mfn;
429 	int level;
430 	pte_t *ptep;
431 	void *virt;
432 
433 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
434 	BUG_ON(size > PAGE_SIZE);
435 	BUG_ON(va & ~PAGE_MASK);
436 
437 	/*
438 	 * The GDT is per-cpu and is in the percpu data area.
439 	 * That can be virtually mapped, so we need to do a
440 	 * page-walk to get the underlying MFN for the
441 	 * hypercall.  The page can also be in the kernel's
442 	 * linear range, so we need to RO that mapping too.
443 	 */
444 	ptep = lookup_address(va, &level);
445 	BUG_ON(ptep == NULL);
446 
447 	pfn = pte_pfn(*ptep);
448 	mfn = pfn_to_mfn(pfn);
449 	virt = __va(PFN_PHYS(pfn));
450 
451 	make_lowmem_page_readonly((void *)va);
452 	make_lowmem_page_readonly(virt);
453 
454 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
455 		BUG();
456 }
457 
458 /*
459  * load_gdt for early boot, when the gdt is only mapped once
460  */
461 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
462 {
463 	unsigned long va = dtr->address;
464 	unsigned int size = dtr->size + 1;
465 	unsigned long pfn, mfn;
466 	pte_t pte;
467 
468 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
469 	BUG_ON(size > PAGE_SIZE);
470 	BUG_ON(va & ~PAGE_MASK);
471 
472 	pfn = virt_to_pfn(va);
473 	mfn = pfn_to_mfn(pfn);
474 
475 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
476 
477 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
478 		BUG();
479 
480 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
481 		BUG();
482 }
483 
484 static inline bool desc_equal(const struct desc_struct *d1,
485 			      const struct desc_struct *d2)
486 {
487 	return !memcmp(d1, d2, sizeof(*d1));
488 }
489 
490 static void load_TLS_descriptor(struct thread_struct *t,
491 				unsigned int cpu, unsigned int i)
492 {
493 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
494 	struct desc_struct *gdt;
495 	xmaddr_t maddr;
496 	struct multicall_space mc;
497 
498 	if (desc_equal(shadow, &t->tls_array[i]))
499 		return;
500 
501 	*shadow = t->tls_array[i];
502 
503 	gdt = get_cpu_gdt_rw(cpu);
504 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
505 	mc = __xen_mc_entry(0);
506 
507 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
508 }
509 
510 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
511 {
512 	/*
513 	 * In lazy mode we need to zero %fs, otherwise we may get an
514 	 * exception between the new %fs descriptor being loaded and
515 	 * %fs being effectively cleared at __switch_to().
516 	 */
517 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
518 		loadsegment(fs, 0);
519 
520 	xen_mc_batch();
521 
522 	load_TLS_descriptor(t, cpu, 0);
523 	load_TLS_descriptor(t, cpu, 1);
524 	load_TLS_descriptor(t, cpu, 2);
525 
526 	xen_mc_issue(PARAVIRT_LAZY_CPU);
527 }
528 
529 static void xen_load_gs_index(unsigned int idx)
530 {
531 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
532 		BUG();
533 }
534 
535 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
536 				const void *ptr)
537 {
538 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
539 	u64 entry = *(u64 *)ptr;
540 
541 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
542 
543 	preempt_disable();
544 
545 	xen_mc_flush();
546 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
547 		BUG();
548 
549 	preempt_enable();
550 }
551 
552 void noist_exc_debug(struct pt_regs *regs);
553 
554 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
555 {
556 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
557 	exc_nmi(regs);
558 }
559 
560 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
561 {
562 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
563 	exc_double_fault(regs, error_code);
564 }
565 
566 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
567 {
568 	/*
569 	 * There's no IST on Xen PV, but we still need to dispatch
570 	 * to the correct handler.
571 	 */
572 	if (user_mode(regs))
573 		noist_exc_debug(regs);
574 	else
575 		exc_debug(regs);
576 }
577 
578 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
579 {
580 	/* This should never happen and there is no way to handle it. */
581 	instrumentation_begin();
582 	pr_err("Unknown trap in Xen PV mode.");
583 	BUG();
584 	instrumentation_end();
585 }
586 
587 #ifdef CONFIG_X86_MCE
588 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
589 {
590 	/*
591 	 * There's no IST on Xen PV, but we still need to dispatch
592 	 * to the correct handler.
593 	 */
594 	if (user_mode(regs))
595 		noist_exc_machine_check(regs);
596 	else
597 		exc_machine_check(regs);
598 }
599 #endif
600 
601 struct trap_array_entry {
602 	void (*orig)(void);
603 	void (*xen)(void);
604 	bool ist_okay;
605 };
606 
607 #define TRAP_ENTRY(func, ist_ok) {			\
608 	.orig		= asm_##func,			\
609 	.xen		= xen_asm_##func,		\
610 	.ist_okay	= ist_ok }
611 
612 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
613 	.orig		= asm_##func,			\
614 	.xen		= xen_asm_xenpv_##func,		\
615 	.ist_okay	= ist_ok }
616 
617 static struct trap_array_entry trap_array[] = {
618 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
619 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
620 #ifdef CONFIG_X86_MCE
621 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
622 #endif
623 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
624 	TRAP_ENTRY(exc_int3,				false ),
625 	TRAP_ENTRY(exc_overflow,			false ),
626 #ifdef CONFIG_IA32_EMULATION
627 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
628 #endif
629 	TRAP_ENTRY(exc_page_fault,			false ),
630 	TRAP_ENTRY(exc_divide_error,			false ),
631 	TRAP_ENTRY(exc_bounds,				false ),
632 	TRAP_ENTRY(exc_invalid_op,			false ),
633 	TRAP_ENTRY(exc_device_not_available,		false ),
634 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
635 	TRAP_ENTRY(exc_invalid_tss,			false ),
636 	TRAP_ENTRY(exc_segment_not_present,		false ),
637 	TRAP_ENTRY(exc_stack_segment,			false ),
638 	TRAP_ENTRY(exc_general_protection,		false ),
639 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
640 	TRAP_ENTRY(exc_coprocessor_error,		false ),
641 	TRAP_ENTRY(exc_alignment_check,			false ),
642 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
643 #ifdef CONFIG_X86_KERNEL_IBT
644 	TRAP_ENTRY(exc_control_protection,		false ),
645 #endif
646 };
647 
648 static bool __ref get_trap_addr(void **addr, unsigned int ist)
649 {
650 	unsigned int nr;
651 	bool ist_okay = false;
652 	bool found = false;
653 
654 	/*
655 	 * Replace trap handler addresses by Xen specific ones.
656 	 * Check for known traps using IST and whitelist them.
657 	 * The debugger ones are the only ones we care about.
658 	 * Xen will handle faults like double_fault, so we should never see
659 	 * them.  Warn if there's an unexpected IST-using fault handler.
660 	 */
661 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
662 		struct trap_array_entry *entry = trap_array + nr;
663 
664 		if (*addr == entry->orig) {
665 			*addr = entry->xen;
666 			ist_okay = entry->ist_okay;
667 			found = true;
668 			break;
669 		}
670 	}
671 
672 	if (nr == ARRAY_SIZE(trap_array) &&
673 	    *addr >= (void *)early_idt_handler_array[0] &&
674 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
675 		nr = (*addr - (void *)early_idt_handler_array[0]) /
676 		     EARLY_IDT_HANDLER_SIZE;
677 		*addr = (void *)xen_early_idt_handler_array[nr];
678 		found = true;
679 	}
680 
681 	if (!found)
682 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
683 
684 	if (WARN_ON(found && ist != 0 && !ist_okay))
685 		return false;
686 
687 	return true;
688 }
689 
690 static int cvt_gate_to_trap(int vector, const gate_desc *val,
691 			    struct trap_info *info)
692 {
693 	unsigned long addr;
694 
695 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
696 		return 0;
697 
698 	info->vector = vector;
699 
700 	addr = gate_offset(val);
701 	if (!get_trap_addr((void **)&addr, val->bits.ist))
702 		return 0;
703 	info->address = addr;
704 
705 	info->cs = gate_segment(val);
706 	info->flags = val->bits.dpl;
707 	/* interrupt gates clear IF */
708 	if (val->bits.type == GATE_INTERRUPT)
709 		info->flags |= 1 << 2;
710 
711 	return 1;
712 }
713 
714 /* Locations of each CPU's IDT */
715 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
716 
717 /* Set an IDT entry.  If the entry is part of the current IDT, then
718    also update Xen. */
719 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
720 {
721 	unsigned long p = (unsigned long)&dt[entrynum];
722 	unsigned long start, end;
723 
724 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
725 
726 	preempt_disable();
727 
728 	start = __this_cpu_read(idt_desc.address);
729 	end = start + __this_cpu_read(idt_desc.size) + 1;
730 
731 	xen_mc_flush();
732 
733 	native_write_idt_entry(dt, entrynum, g);
734 
735 	if (p >= start && (p + 8) <= end) {
736 		struct trap_info info[2];
737 
738 		info[1].address = 0;
739 
740 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
741 			if (HYPERVISOR_set_trap_table(info))
742 				BUG();
743 	}
744 
745 	preempt_enable();
746 }
747 
748 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
749 				      struct trap_info *traps, bool full)
750 {
751 	unsigned in, out, count;
752 
753 	count = (desc->size+1) / sizeof(gate_desc);
754 	BUG_ON(count > 256);
755 
756 	for (in = out = 0; in < count; in++) {
757 		gate_desc *entry = (gate_desc *)(desc->address) + in;
758 
759 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
760 			out++;
761 	}
762 
763 	return out;
764 }
765 
766 void xen_copy_trap_info(struct trap_info *traps)
767 {
768 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
769 
770 	xen_convert_trap_info(desc, traps, true);
771 }
772 
773 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
774    hold a spinlock to protect the static traps[] array (static because
775    it avoids allocation, and saves stack space). */
776 static void xen_load_idt(const struct desc_ptr *desc)
777 {
778 	static DEFINE_SPINLOCK(lock);
779 	static struct trap_info traps[257];
780 	static const struct trap_info zero = { };
781 	unsigned out;
782 
783 	trace_xen_cpu_load_idt(desc);
784 
785 	spin_lock(&lock);
786 
787 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
788 
789 	out = xen_convert_trap_info(desc, traps, false);
790 	traps[out] = zero;
791 
792 	xen_mc_flush();
793 	if (HYPERVISOR_set_trap_table(traps))
794 		BUG();
795 
796 	spin_unlock(&lock);
797 }
798 
799 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
800    they're handled differently. */
801 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
802 				const void *desc, int type)
803 {
804 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
805 
806 	preempt_disable();
807 
808 	switch (type) {
809 	case DESC_LDT:
810 	case DESC_TSS:
811 		/* ignore */
812 		break;
813 
814 	default: {
815 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
816 
817 		xen_mc_flush();
818 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
819 			BUG();
820 	}
821 
822 	}
823 
824 	preempt_enable();
825 }
826 
827 /*
828  * Version of write_gdt_entry for use at early boot-time needed to
829  * update an entry as simply as possible.
830  */
831 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
832 					    const void *desc, int type)
833 {
834 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
835 
836 	switch (type) {
837 	case DESC_LDT:
838 	case DESC_TSS:
839 		/* ignore */
840 		break;
841 
842 	default: {
843 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
844 
845 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
846 			dt[entry] = *(struct desc_struct *)desc;
847 	}
848 
849 	}
850 }
851 
852 static void xen_load_sp0(unsigned long sp0)
853 {
854 	struct multicall_space mcs;
855 
856 	mcs = xen_mc_entry(0);
857 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
858 	xen_mc_issue(PARAVIRT_LAZY_CPU);
859 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
860 }
861 
862 #ifdef CONFIG_X86_IOPL_IOPERM
863 static void xen_invalidate_io_bitmap(void)
864 {
865 	struct physdev_set_iobitmap iobitmap = {
866 		.bitmap = NULL,
867 		.nr_ports = 0,
868 	};
869 
870 	native_tss_invalidate_io_bitmap();
871 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
872 }
873 
874 static void xen_update_io_bitmap(void)
875 {
876 	struct physdev_set_iobitmap iobitmap;
877 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
878 
879 	native_tss_update_io_bitmap();
880 
881 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
882 			  tss->x86_tss.io_bitmap_base;
883 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
884 		iobitmap.nr_ports = 0;
885 	else
886 		iobitmap.nr_ports = IO_BITMAP_BITS;
887 
888 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
889 }
890 #endif
891 
892 static void xen_io_delay(void)
893 {
894 }
895 
896 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
897 
898 static unsigned long xen_read_cr0(void)
899 {
900 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
901 
902 	if (unlikely(cr0 == 0)) {
903 		cr0 = native_read_cr0();
904 		this_cpu_write(xen_cr0_value, cr0);
905 	}
906 
907 	return cr0;
908 }
909 
910 static void xen_write_cr0(unsigned long cr0)
911 {
912 	struct multicall_space mcs;
913 
914 	this_cpu_write(xen_cr0_value, cr0);
915 
916 	/* Only pay attention to cr0.TS; everything else is
917 	   ignored. */
918 	mcs = xen_mc_entry(0);
919 
920 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
921 
922 	xen_mc_issue(PARAVIRT_LAZY_CPU);
923 }
924 
925 static void xen_write_cr4(unsigned long cr4)
926 {
927 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
928 
929 	native_write_cr4(cr4);
930 }
931 
932 static u64 xen_do_read_msr(unsigned int msr, int *err)
933 {
934 	u64 val = 0;	/* Avoid uninitialized value for safe variant. */
935 
936 	if (pmu_msr_read(msr, &val, err))
937 		return val;
938 
939 	if (err)
940 		val = native_read_msr_safe(msr, err);
941 	else
942 		val = native_read_msr(msr);
943 
944 	switch (msr) {
945 	case MSR_IA32_APICBASE:
946 		val &= ~X2APIC_ENABLE;
947 		break;
948 	}
949 	return val;
950 }
951 
952 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
953 		    int *err)
954 {
955 	u64 base = ((u64)high << 32) | low;
956 
957 	if (HYPERVISOR_set_segment_base(which, base) == 0)
958 		return;
959 
960 	if (err)
961 		*err = -EIO;
962 	else
963 		WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
964 }
965 
966 /*
967  * Support write_msr_safe() and write_msr() semantics.
968  * With err == NULL write_msr() semantics are selected.
969  * Supplying an err pointer requires err to be pre-initialized with 0.
970  */
971 static void xen_do_write_msr(unsigned int msr, unsigned int low,
972 			     unsigned int high, int *err)
973 {
974 	switch (msr) {
975 	case MSR_FS_BASE:
976 		set_seg(SEGBASE_FS, low, high, err);
977 		break;
978 
979 	case MSR_KERNEL_GS_BASE:
980 		set_seg(SEGBASE_GS_USER, low, high, err);
981 		break;
982 
983 	case MSR_GS_BASE:
984 		set_seg(SEGBASE_GS_KERNEL, low, high, err);
985 		break;
986 
987 	case MSR_STAR:
988 	case MSR_CSTAR:
989 	case MSR_LSTAR:
990 	case MSR_SYSCALL_MASK:
991 	case MSR_IA32_SYSENTER_CS:
992 	case MSR_IA32_SYSENTER_ESP:
993 	case MSR_IA32_SYSENTER_EIP:
994 		/* Fast syscall setup is all done in hypercalls, so
995 		   these are all ignored.  Stub them out here to stop
996 		   Xen console noise. */
997 		break;
998 
999 	default:
1000 		if (!pmu_msr_write(msr, low, high, err)) {
1001 			if (err)
1002 				*err = native_write_msr_safe(msr, low, high);
1003 			else
1004 				native_write_msr(msr, low, high);
1005 		}
1006 	}
1007 }
1008 
1009 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1010 {
1011 	return xen_do_read_msr(msr, err);
1012 }
1013 
1014 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1015 			      unsigned int high)
1016 {
1017 	int err = 0;
1018 
1019 	xen_do_write_msr(msr, low, high, &err);
1020 
1021 	return err;
1022 }
1023 
1024 static u64 xen_read_msr(unsigned int msr)
1025 {
1026 	int err;
1027 
1028 	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1029 }
1030 
1031 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1032 {
1033 	int err;
1034 
1035 	xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1036 }
1037 
1038 /* This is called once we have the cpu_possible_mask */
1039 void __init xen_setup_vcpu_info_placement(void)
1040 {
1041 	int cpu;
1042 
1043 	for_each_possible_cpu(cpu) {
1044 		/* Set up direct vCPU id mapping for PV guests. */
1045 		per_cpu(xen_vcpu_id, cpu) = cpu;
1046 		xen_vcpu_setup(cpu);
1047 	}
1048 
1049 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1050 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1051 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1052 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1053 }
1054 
1055 static const struct pv_info xen_info __initconst = {
1056 	.extra_user_64bit_cs = FLAT_USER_CS64,
1057 	.name = "Xen",
1058 };
1059 
1060 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1061 	.cpu = {
1062 		.cpuid = xen_cpuid,
1063 
1064 		.set_debugreg = xen_set_debugreg,
1065 		.get_debugreg = xen_get_debugreg,
1066 
1067 		.read_cr0 = xen_read_cr0,
1068 		.write_cr0 = xen_write_cr0,
1069 
1070 		.write_cr4 = xen_write_cr4,
1071 
1072 		.wbinvd = pv_native_wbinvd,
1073 
1074 		.read_msr = xen_read_msr,
1075 		.write_msr = xen_write_msr,
1076 
1077 		.read_msr_safe = xen_read_msr_safe,
1078 		.write_msr_safe = xen_write_msr_safe,
1079 
1080 		.read_pmc = xen_read_pmc,
1081 
1082 		.load_tr_desc = paravirt_nop,
1083 		.set_ldt = xen_set_ldt,
1084 		.load_gdt = xen_load_gdt,
1085 		.load_idt = xen_load_idt,
1086 		.load_tls = xen_load_tls,
1087 		.load_gs_index = xen_load_gs_index,
1088 
1089 		.alloc_ldt = xen_alloc_ldt,
1090 		.free_ldt = xen_free_ldt,
1091 
1092 		.store_tr = xen_store_tr,
1093 
1094 		.write_ldt_entry = xen_write_ldt_entry,
1095 		.write_gdt_entry = xen_write_gdt_entry,
1096 		.write_idt_entry = xen_write_idt_entry,
1097 		.load_sp0 = xen_load_sp0,
1098 
1099 #ifdef CONFIG_X86_IOPL_IOPERM
1100 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1101 		.update_io_bitmap = xen_update_io_bitmap,
1102 #endif
1103 		.io_delay = xen_io_delay,
1104 
1105 		.start_context_switch = paravirt_start_context_switch,
1106 		.end_context_switch = xen_end_context_switch,
1107 	},
1108 };
1109 
1110 static void xen_restart(char *msg)
1111 {
1112 	xen_reboot(SHUTDOWN_reboot);
1113 }
1114 
1115 static void xen_machine_halt(void)
1116 {
1117 	xen_reboot(SHUTDOWN_poweroff);
1118 }
1119 
1120 static void xen_machine_power_off(void)
1121 {
1122 	do_kernel_power_off();
1123 	xen_reboot(SHUTDOWN_poweroff);
1124 }
1125 
1126 static void xen_crash_shutdown(struct pt_regs *regs)
1127 {
1128 	xen_reboot(SHUTDOWN_crash);
1129 }
1130 
1131 static const struct machine_ops xen_machine_ops __initconst = {
1132 	.restart = xen_restart,
1133 	.halt = xen_machine_halt,
1134 	.power_off = xen_machine_power_off,
1135 	.shutdown = xen_machine_halt,
1136 	.crash_shutdown = xen_crash_shutdown,
1137 	.emergency_restart = xen_emergency_restart,
1138 };
1139 
1140 static unsigned char xen_get_nmi_reason(void)
1141 {
1142 	unsigned char reason = 0;
1143 
1144 	/* Construct a value which looks like it came from port 0x61. */
1145 	if (test_bit(_XEN_NMIREASON_io_error,
1146 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1147 		reason |= NMI_REASON_IOCHK;
1148 	if (test_bit(_XEN_NMIREASON_pci_serr,
1149 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1150 		reason |= NMI_REASON_SERR;
1151 
1152 	return reason;
1153 }
1154 
1155 static void __init xen_boot_params_init_edd(void)
1156 {
1157 #if IS_ENABLED(CONFIG_EDD)
1158 	struct xen_platform_op op;
1159 	struct edd_info *edd_info;
1160 	u32 *mbr_signature;
1161 	unsigned nr;
1162 	int ret;
1163 
1164 	edd_info = boot_params.eddbuf;
1165 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1166 
1167 	op.cmd = XENPF_firmware_info;
1168 
1169 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1170 	for (nr = 0; nr < EDDMAXNR; nr++) {
1171 		struct edd_info *info = edd_info + nr;
1172 
1173 		op.u.firmware_info.index = nr;
1174 		info->params.length = sizeof(info->params);
1175 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1176 				     &info->params);
1177 		ret = HYPERVISOR_platform_op(&op);
1178 		if (ret)
1179 			break;
1180 
1181 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1182 		C(device);
1183 		C(version);
1184 		C(interface_support);
1185 		C(legacy_max_cylinder);
1186 		C(legacy_max_head);
1187 		C(legacy_sectors_per_track);
1188 #undef C
1189 	}
1190 	boot_params.eddbuf_entries = nr;
1191 
1192 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1193 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1194 		op.u.firmware_info.index = nr;
1195 		ret = HYPERVISOR_platform_op(&op);
1196 		if (ret)
1197 			break;
1198 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1199 	}
1200 	boot_params.edd_mbr_sig_buf_entries = nr;
1201 #endif
1202 }
1203 
1204 /*
1205  * Set up the GDT and segment registers for -fstack-protector.  Until
1206  * we do this, we have to be careful not to call any stack-protected
1207  * function, which is most of the kernel.
1208  */
1209 static void __init xen_setup_gdt(int cpu)
1210 {
1211 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1212 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1213 
1214 	switch_gdt_and_percpu_base(cpu);
1215 
1216 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1217 	pv_ops.cpu.load_gdt = xen_load_gdt;
1218 }
1219 
1220 static void __init xen_dom0_set_legacy_features(void)
1221 {
1222 	x86_platform.legacy.rtc = 1;
1223 }
1224 
1225 static void __init xen_domu_set_legacy_features(void)
1226 {
1227 	x86_platform.legacy.rtc = 0;
1228 }
1229 
1230 extern void early_xen_iret_patch(void);
1231 
1232 /* First C function to be called on Xen boot */
1233 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1234 {
1235 	struct physdev_set_iopl set_iopl;
1236 	unsigned long initrd_start = 0;
1237 	int rc;
1238 
1239 	if (!si)
1240 		return;
1241 
1242 	clear_bss();
1243 
1244 	xen_start_info = si;
1245 
1246 	__text_gen_insn(&early_xen_iret_patch,
1247 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1248 			JMP32_INSN_SIZE);
1249 
1250 	xen_domain_type = XEN_PV_DOMAIN;
1251 	xen_start_flags = xen_start_info->flags;
1252 
1253 	xen_setup_features();
1254 
1255 	/* Install Xen paravirt ops */
1256 	pv_info = xen_info;
1257 	pv_ops.cpu = xen_cpu_ops.cpu;
1258 	xen_init_irq_ops();
1259 
1260 	/*
1261 	 * Setup xen_vcpu early because it is needed for
1262 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1263 	 *
1264 	 * Don't do the full vcpu_info placement stuff until we have
1265 	 * the cpu_possible_mask and a non-dummy shared_info.
1266 	 */
1267 	xen_vcpu_info_reset(0);
1268 
1269 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1270 	x86_platform.realmode_reserve = x86_init_noop;
1271 	x86_platform.realmode_init = x86_init_noop;
1272 
1273 	x86_init.resources.memory_setup = xen_memory_setup;
1274 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1275 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1276 	x86_init.oem.arch_setup = xen_arch_setup;
1277 	x86_init.oem.banner = xen_banner;
1278 	x86_init.hyper.init_platform = xen_pv_init_platform;
1279 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1280 
1281 	/*
1282 	 * Set up some pagetable state before starting to set any ptes.
1283 	 */
1284 
1285 	xen_setup_machphys_mapping();
1286 	xen_init_mmu_ops();
1287 
1288 	/* Prevent unwanted bits from being set in PTEs. */
1289 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1290 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1291 
1292 	/* Get mfn list */
1293 	xen_build_dynamic_phys_to_machine();
1294 
1295 	/* Work out if we support NX */
1296 	get_cpu_cap(&boot_cpu_data);
1297 	x86_configure_nx();
1298 
1299 	/*
1300 	 * Set up kernel GDT and segment registers, mainly so that
1301 	 * -fstack-protector code can be executed.
1302 	 */
1303 	xen_setup_gdt(0);
1304 
1305 	/* Determine virtual and physical address sizes */
1306 	get_cpu_address_sizes(&boot_cpu_data);
1307 
1308 	/* Let's presume PV guests always boot on vCPU with id 0. */
1309 	per_cpu(xen_vcpu_id, 0) = 0;
1310 
1311 	idt_setup_early_handler();
1312 
1313 	xen_init_capabilities();
1314 
1315 #ifdef CONFIG_X86_LOCAL_APIC
1316 	/*
1317 	 * set up the basic apic ops.
1318 	 */
1319 	xen_init_apic();
1320 #endif
1321 
1322 	machine_ops = xen_machine_ops;
1323 
1324 	/*
1325 	 * The only reliable way to retain the initial address of the
1326 	 * percpu gdt_page is to remember it here, so we can go and
1327 	 * mark it RW later, when the initial percpu area is freed.
1328 	 */
1329 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1330 
1331 	xen_smp_init();
1332 
1333 #ifdef CONFIG_ACPI_NUMA
1334 	/*
1335 	 * The pages we from Xen are not related to machine pages, so
1336 	 * any NUMA information the kernel tries to get from ACPI will
1337 	 * be meaningless.  Prevent it from trying.
1338 	 */
1339 	disable_srat();
1340 #endif
1341 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1342 
1343 	local_irq_disable();
1344 	early_boot_irqs_disabled = true;
1345 
1346 	xen_raw_console_write("mapping kernel into physical memory\n");
1347 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1348 				   xen_start_info->nr_pages);
1349 	xen_reserve_special_pages();
1350 
1351 	/*
1352 	 * We used to do this in xen_arch_setup, but that is too late
1353 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1354 	 * early_amd_init which pokes 0xcf8 port.
1355 	 */
1356 	set_iopl.iopl = 1;
1357 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1358 	if (rc != 0)
1359 		xen_raw_printk("physdev_op failed %d\n", rc);
1360 
1361 
1362 	if (xen_start_info->mod_start) {
1363 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1364 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1365 	    else
1366 		initrd_start = __pa(xen_start_info->mod_start);
1367 	}
1368 
1369 	/* Poke various useful things into boot_params */
1370 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1371 	boot_params.hdr.ramdisk_image = initrd_start;
1372 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1373 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1374 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1375 
1376 	if (!xen_initial_domain()) {
1377 		if (pci_xen)
1378 			x86_init.pci.arch_init = pci_xen_init;
1379 		x86_platform.set_legacy_features =
1380 				xen_domu_set_legacy_features;
1381 	} else {
1382 		const struct dom0_vga_console_info *info =
1383 			(void *)((char *)xen_start_info +
1384 				 xen_start_info->console.dom0.info_off);
1385 		struct xen_platform_op op = {
1386 			.cmd = XENPF_firmware_info,
1387 			.interface_version = XENPF_INTERFACE_VERSION,
1388 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1389 		};
1390 
1391 		x86_platform.set_legacy_features =
1392 				xen_dom0_set_legacy_features;
1393 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1394 		xen_start_info->console.domU.mfn = 0;
1395 		xen_start_info->console.domU.evtchn = 0;
1396 
1397 		if (HYPERVISOR_platform_op(&op) == 0)
1398 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1399 
1400 		/* Make sure ACS will be enabled */
1401 		pci_request_acs();
1402 
1403 		xen_acpi_sleep_register();
1404 
1405 		xen_boot_params_init_edd();
1406 
1407 #ifdef CONFIG_ACPI
1408 		/*
1409 		 * Disable selecting "Firmware First mode" for correctable
1410 		 * memory errors, as this is the duty of the hypervisor to
1411 		 * decide.
1412 		 */
1413 		acpi_disable_cmcff = 1;
1414 #endif
1415 	}
1416 
1417 	xen_add_preferred_consoles();
1418 
1419 #ifdef CONFIG_PCI
1420 	/* PCI BIOS service won't work from a PV guest. */
1421 	pci_probe &= ~PCI_PROBE_BIOS;
1422 #endif
1423 	xen_raw_console_write("about to get started...\n");
1424 
1425 	/* We need this for printk timestamps */
1426 	xen_setup_runstate_info(0);
1427 
1428 	xen_efi_init(&boot_params);
1429 
1430 	/* Start the world */
1431 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1432 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1433 }
1434 
1435 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1436 {
1437 	int rc;
1438 
1439 	if (per_cpu(xen_vcpu, cpu) == NULL)
1440 		return -ENODEV;
1441 
1442 	xen_setup_timer(cpu);
1443 
1444 	rc = xen_smp_intr_init(cpu);
1445 	if (rc) {
1446 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1447 		     cpu, rc);
1448 		return rc;
1449 	}
1450 
1451 	rc = xen_smp_intr_init_pv(cpu);
1452 	if (rc) {
1453 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1454 		     cpu, rc);
1455 		return rc;
1456 	}
1457 
1458 	return 0;
1459 }
1460 
1461 static int xen_cpu_dead_pv(unsigned int cpu)
1462 {
1463 	xen_smp_intr_free(cpu);
1464 	xen_smp_intr_free_pv(cpu);
1465 
1466 	xen_teardown_timer(cpu);
1467 
1468 	return 0;
1469 }
1470 
1471 static uint32_t __init xen_platform_pv(void)
1472 {
1473 	if (xen_pv_domain())
1474 		return xen_cpuid_base();
1475 
1476 	return 0;
1477 }
1478 
1479 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1480 	.name                   = "Xen PV",
1481 	.detect                 = xen_platform_pv,
1482 	.type			= X86_HYPER_XEN_PV,
1483 	.runtime.pin_vcpu       = xen_pin_vcpu,
1484 	.ignore_nopv		= true,
1485 };
1486