xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 249592bf)
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/console.h>
32 #include <linux/pci.h>
33 #include <linux/gfp.h>
34 #include <linux/edd.h>
35 #include <linux/objtool.h>
36 
37 #include <xen/xen.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
47 #include <xen/page.h>
48 #include <xen/hvc-console.h>
49 #include <xen/acpi.h>
50 
51 #include <asm/paravirt.h>
52 #include <asm/apic.h>
53 #include <asm/page.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
64 #include <asm/desc.h>
65 #include <asm/pgalloc.h>
66 #include <asm/tlbflush.h>
67 #include <asm/reboot.h>
68 #include <asm/stackprotector.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 void __init xen_banner(void)
113 {
114 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
115 	struct xen_extraversion extra;
116 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
117 
118 	pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
119 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
120 	       version >> 16, version & 0xffff, extra.extraversion,
121 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
122 }
123 
124 static void __init xen_pv_init_platform(void)
125 {
126 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
127 
128 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
129 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
130 
131 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
132 	xen_vcpu_info_reset(0);
133 
134 	/* pvclock is in shared info area */
135 	xen_init_time_ops();
136 }
137 
138 static void __init xen_pv_guest_late_init(void)
139 {
140 #ifndef CONFIG_SMP
141 	/* Setup shared vcpu info for non-smp configurations */
142 	xen_setup_vcpu_info_placement();
143 #endif
144 }
145 
146 /* Check if running on Xen version (major, minor) or later */
147 bool
148 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
149 {
150 	unsigned int version;
151 
152 	if (!xen_domain())
153 		return false;
154 
155 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
156 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
157 		((version >> 16) > major))
158 		return true;
159 	return false;
160 }
161 
162 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
163 static __read_mostly unsigned int cpuid_leaf5_edx_val;
164 
165 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
166 		      unsigned int *cx, unsigned int *dx)
167 {
168 	unsigned maskebx = ~0;
169 
170 	/*
171 	 * Mask out inconvenient features, to try and disable as many
172 	 * unsupported kernel subsystems as possible.
173 	 */
174 	switch (*ax) {
175 	case CPUID_MWAIT_LEAF:
176 		/* Synthesize the values.. */
177 		*ax = 0;
178 		*bx = 0;
179 		*cx = cpuid_leaf5_ecx_val;
180 		*dx = cpuid_leaf5_edx_val;
181 		return;
182 
183 	case 0xb:
184 		/* Suppress extended topology stuff */
185 		maskebx = 0;
186 		break;
187 	}
188 
189 	asm(XEN_EMULATE_PREFIX "cpuid"
190 		: "=a" (*ax),
191 		  "=b" (*bx),
192 		  "=c" (*cx),
193 		  "=d" (*dx)
194 		: "0" (*ax), "2" (*cx));
195 
196 	*bx &= maskebx;
197 }
198 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
199 
200 static bool __init xen_check_mwait(void)
201 {
202 #ifdef CONFIG_ACPI
203 	struct xen_platform_op op = {
204 		.cmd			= XENPF_set_processor_pminfo,
205 		.u.set_pminfo.id	= -1,
206 		.u.set_pminfo.type	= XEN_PM_PDC,
207 	};
208 	uint32_t buf[3];
209 	unsigned int ax, bx, cx, dx;
210 	unsigned int mwait_mask;
211 
212 	/* We need to determine whether it is OK to expose the MWAIT
213 	 * capability to the kernel to harvest deeper than C3 states from ACPI
214 	 * _CST using the processor_harvest_xen.c module. For this to work, we
215 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
216 	 * checks against). The hypervisor won't expose the MWAIT flag because
217 	 * it would break backwards compatibility; so we will find out directly
218 	 * from the hardware and hypercall.
219 	 */
220 	if (!xen_initial_domain())
221 		return false;
222 
223 	/*
224 	 * When running under platform earlier than Xen4.2, do not expose
225 	 * mwait, to avoid the risk of loading native acpi pad driver
226 	 */
227 	if (!xen_running_on_version_or_later(4, 2))
228 		return false;
229 
230 	ax = 1;
231 	cx = 0;
232 
233 	native_cpuid(&ax, &bx, &cx, &dx);
234 
235 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
236 		     (1 << (X86_FEATURE_MWAIT % 32));
237 
238 	if ((cx & mwait_mask) != mwait_mask)
239 		return false;
240 
241 	/* We need to emulate the MWAIT_LEAF and for that we need both
242 	 * ecx and edx. The hypercall provides only partial information.
243 	 */
244 
245 	ax = CPUID_MWAIT_LEAF;
246 	bx = 0;
247 	cx = 0;
248 	dx = 0;
249 
250 	native_cpuid(&ax, &bx, &cx, &dx);
251 
252 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
253 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
254 	 */
255 	buf[0] = ACPI_PDC_REVISION_ID;
256 	buf[1] = 1;
257 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
258 
259 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
260 
261 	if ((HYPERVISOR_platform_op(&op) == 0) &&
262 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
263 		cpuid_leaf5_ecx_val = cx;
264 		cpuid_leaf5_edx_val = dx;
265 	}
266 	return true;
267 #else
268 	return false;
269 #endif
270 }
271 
272 static bool __init xen_check_xsave(void)
273 {
274 	unsigned int cx, xsave_mask;
275 
276 	cx = cpuid_ecx(1);
277 
278 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
279 		     (1 << (X86_FEATURE_OSXSAVE % 32));
280 
281 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
282 	return (cx & xsave_mask) == xsave_mask;
283 }
284 
285 static void __init xen_init_capabilities(void)
286 {
287 	setup_force_cpu_cap(X86_FEATURE_XENPV);
288 	setup_clear_cpu_cap(X86_FEATURE_DCA);
289 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
290 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
291 	setup_clear_cpu_cap(X86_FEATURE_ACC);
292 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
293 	setup_clear_cpu_cap(X86_FEATURE_SME);
294 
295 	/*
296 	 * Xen PV would need some work to support PCID: CR3 handling as well
297 	 * as xen_flush_tlb_others() would need updating.
298 	 */
299 	setup_clear_cpu_cap(X86_FEATURE_PCID);
300 
301 	if (!xen_initial_domain())
302 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
303 
304 	if (xen_check_mwait())
305 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
306 	else
307 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
308 
309 	if (!xen_check_xsave()) {
310 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
311 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
312 	}
313 }
314 
315 static void xen_set_debugreg(int reg, unsigned long val)
316 {
317 	HYPERVISOR_set_debugreg(reg, val);
318 }
319 
320 static unsigned long xen_get_debugreg(int reg)
321 {
322 	return HYPERVISOR_get_debugreg(reg);
323 }
324 
325 static void xen_end_context_switch(struct task_struct *next)
326 {
327 	xen_mc_flush();
328 	paravirt_end_context_switch(next);
329 }
330 
331 static unsigned long xen_store_tr(void)
332 {
333 	return 0;
334 }
335 
336 /*
337  * Set the page permissions for a particular virtual address.  If the
338  * address is a vmalloc mapping (or other non-linear mapping), then
339  * find the linear mapping of the page and also set its protections to
340  * match.
341  */
342 static void set_aliased_prot(void *v, pgprot_t prot)
343 {
344 	int level;
345 	pte_t *ptep;
346 	pte_t pte;
347 	unsigned long pfn;
348 	unsigned char dummy;
349 	void *va;
350 
351 	ptep = lookup_address((unsigned long)v, &level);
352 	BUG_ON(ptep == NULL);
353 
354 	pfn = pte_pfn(*ptep);
355 	pte = pfn_pte(pfn, prot);
356 
357 	/*
358 	 * Careful: update_va_mapping() will fail if the virtual address
359 	 * we're poking isn't populated in the page tables.  We don't
360 	 * need to worry about the direct map (that's always in the page
361 	 * tables), but we need to be careful about vmap space.  In
362 	 * particular, the top level page table can lazily propagate
363 	 * entries between processes, so if we've switched mms since we
364 	 * vmapped the target in the first place, we might not have the
365 	 * top-level page table entry populated.
366 	 *
367 	 * We disable preemption because we want the same mm active when
368 	 * we probe the target and when we issue the hypercall.  We'll
369 	 * have the same nominal mm, but if we're a kernel thread, lazy
370 	 * mm dropping could change our pgd.
371 	 *
372 	 * Out of an abundance of caution, this uses __get_user() to fault
373 	 * in the target address just in case there's some obscure case
374 	 * in which the target address isn't readable.
375 	 */
376 
377 	preempt_disable();
378 
379 	copy_from_kernel_nofault(&dummy, v, 1);
380 
381 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
382 		BUG();
383 
384 	va = __va(PFN_PHYS(pfn));
385 
386 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
387 		BUG();
388 
389 	preempt_enable();
390 }
391 
392 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
393 {
394 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
395 	int i;
396 
397 	/*
398 	 * We need to mark the all aliases of the LDT pages RO.  We
399 	 * don't need to call vm_flush_aliases(), though, since that's
400 	 * only responsible for flushing aliases out the TLBs, not the
401 	 * page tables, and Xen will flush the TLB for us if needed.
402 	 *
403 	 * To avoid confusing future readers: none of this is necessary
404 	 * to load the LDT.  The hypervisor only checks this when the
405 	 * LDT is faulted in due to subsequent descriptor access.
406 	 */
407 
408 	for (i = 0; i < entries; i += entries_per_page)
409 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
410 }
411 
412 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
413 {
414 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
415 	int i;
416 
417 	for (i = 0; i < entries; i += entries_per_page)
418 		set_aliased_prot(ldt + i, PAGE_KERNEL);
419 }
420 
421 static void xen_set_ldt(const void *addr, unsigned entries)
422 {
423 	struct mmuext_op *op;
424 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
425 
426 	trace_xen_cpu_set_ldt(addr, entries);
427 
428 	op = mcs.args;
429 	op->cmd = MMUEXT_SET_LDT;
430 	op->arg1.linear_addr = (unsigned long)addr;
431 	op->arg2.nr_ents = entries;
432 
433 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
434 
435 	xen_mc_issue(PARAVIRT_LAZY_CPU);
436 }
437 
438 static void xen_load_gdt(const struct desc_ptr *dtr)
439 {
440 	unsigned long va = dtr->address;
441 	unsigned int size = dtr->size + 1;
442 	unsigned long pfn, mfn;
443 	int level;
444 	pte_t *ptep;
445 	void *virt;
446 
447 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
448 	BUG_ON(size > PAGE_SIZE);
449 	BUG_ON(va & ~PAGE_MASK);
450 
451 	/*
452 	 * The GDT is per-cpu and is in the percpu data area.
453 	 * That can be virtually mapped, so we need to do a
454 	 * page-walk to get the underlying MFN for the
455 	 * hypercall.  The page can also be in the kernel's
456 	 * linear range, so we need to RO that mapping too.
457 	 */
458 	ptep = lookup_address(va, &level);
459 	BUG_ON(ptep == NULL);
460 
461 	pfn = pte_pfn(*ptep);
462 	mfn = pfn_to_mfn(pfn);
463 	virt = __va(PFN_PHYS(pfn));
464 
465 	make_lowmem_page_readonly((void *)va);
466 	make_lowmem_page_readonly(virt);
467 
468 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
469 		BUG();
470 }
471 
472 /*
473  * load_gdt for early boot, when the gdt is only mapped once
474  */
475 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
476 {
477 	unsigned long va = dtr->address;
478 	unsigned int size = dtr->size + 1;
479 	unsigned long pfn, mfn;
480 	pte_t pte;
481 
482 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
483 	BUG_ON(size > PAGE_SIZE);
484 	BUG_ON(va & ~PAGE_MASK);
485 
486 	pfn = virt_to_pfn(va);
487 	mfn = pfn_to_mfn(pfn);
488 
489 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
490 
491 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
492 		BUG();
493 
494 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
495 		BUG();
496 }
497 
498 static inline bool desc_equal(const struct desc_struct *d1,
499 			      const struct desc_struct *d2)
500 {
501 	return !memcmp(d1, d2, sizeof(*d1));
502 }
503 
504 static void load_TLS_descriptor(struct thread_struct *t,
505 				unsigned int cpu, unsigned int i)
506 {
507 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
508 	struct desc_struct *gdt;
509 	xmaddr_t maddr;
510 	struct multicall_space mc;
511 
512 	if (desc_equal(shadow, &t->tls_array[i]))
513 		return;
514 
515 	*shadow = t->tls_array[i];
516 
517 	gdt = get_cpu_gdt_rw(cpu);
518 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
519 	mc = __xen_mc_entry(0);
520 
521 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
522 }
523 
524 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
525 {
526 	/*
527 	 * In lazy mode we need to zero %fs, otherwise we may get an
528 	 * exception between the new %fs descriptor being loaded and
529 	 * %fs being effectively cleared at __switch_to().
530 	 */
531 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
532 		loadsegment(fs, 0);
533 
534 	xen_mc_batch();
535 
536 	load_TLS_descriptor(t, cpu, 0);
537 	load_TLS_descriptor(t, cpu, 1);
538 	load_TLS_descriptor(t, cpu, 2);
539 
540 	xen_mc_issue(PARAVIRT_LAZY_CPU);
541 }
542 
543 static void xen_load_gs_index(unsigned int idx)
544 {
545 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
546 		BUG();
547 }
548 
549 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
550 				const void *ptr)
551 {
552 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
553 	u64 entry = *(u64 *)ptr;
554 
555 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
556 
557 	preempt_disable();
558 
559 	xen_mc_flush();
560 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
561 		BUG();
562 
563 	preempt_enable();
564 }
565 
566 void noist_exc_debug(struct pt_regs *regs);
567 
568 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
569 {
570 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
571 	exc_nmi(regs);
572 }
573 
574 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
575 {
576 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
577 	exc_double_fault(regs, error_code);
578 }
579 
580 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
581 {
582 	/*
583 	 * There's no IST on Xen PV, but we still need to dispatch
584 	 * to the correct handler.
585 	 */
586 	if (user_mode(regs))
587 		noist_exc_debug(regs);
588 	else
589 		exc_debug(regs);
590 }
591 
592 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
593 {
594 	/* This should never happen and there is no way to handle it. */
595 	pr_err("Unknown trap in Xen PV mode.");
596 	BUG();
597 }
598 
599 #ifdef CONFIG_X86_MCE
600 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
601 {
602 	/*
603 	 * There's no IST on Xen PV, but we still need to dispatch
604 	 * to the correct handler.
605 	 */
606 	if (user_mode(regs))
607 		noist_exc_machine_check(regs);
608 	else
609 		exc_machine_check(regs);
610 }
611 #endif
612 
613 struct trap_array_entry {
614 	void (*orig)(void);
615 	void (*xen)(void);
616 	bool ist_okay;
617 };
618 
619 #define TRAP_ENTRY(func, ist_ok) {			\
620 	.orig		= asm_##func,			\
621 	.xen		= xen_asm_##func,		\
622 	.ist_okay	= ist_ok }
623 
624 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
625 	.orig		= asm_##func,			\
626 	.xen		= xen_asm_xenpv_##func,		\
627 	.ist_okay	= ist_ok }
628 
629 static struct trap_array_entry trap_array[] = {
630 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
631 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
632 #ifdef CONFIG_X86_MCE
633 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
634 #endif
635 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
636 	TRAP_ENTRY(exc_int3,				false ),
637 	TRAP_ENTRY(exc_overflow,			false ),
638 #ifdef CONFIG_IA32_EMULATION
639 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
640 #endif
641 	TRAP_ENTRY(exc_page_fault,			false ),
642 	TRAP_ENTRY(exc_divide_error,			false ),
643 	TRAP_ENTRY(exc_bounds,				false ),
644 	TRAP_ENTRY(exc_invalid_op,			false ),
645 	TRAP_ENTRY(exc_device_not_available,		false ),
646 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
647 	TRAP_ENTRY(exc_invalid_tss,			false ),
648 	TRAP_ENTRY(exc_segment_not_present,		false ),
649 	TRAP_ENTRY(exc_stack_segment,			false ),
650 	TRAP_ENTRY(exc_general_protection,		false ),
651 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
652 	TRAP_ENTRY(exc_coprocessor_error,		false ),
653 	TRAP_ENTRY(exc_alignment_check,			false ),
654 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
655 };
656 
657 static bool __ref get_trap_addr(void **addr, unsigned int ist)
658 {
659 	unsigned int nr;
660 	bool ist_okay = false;
661 	bool found = false;
662 
663 	/*
664 	 * Replace trap handler addresses by Xen specific ones.
665 	 * Check for known traps using IST and whitelist them.
666 	 * The debugger ones are the only ones we care about.
667 	 * Xen will handle faults like double_fault, so we should never see
668 	 * them.  Warn if there's an unexpected IST-using fault handler.
669 	 */
670 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
671 		struct trap_array_entry *entry = trap_array + nr;
672 
673 		if (*addr == entry->orig) {
674 			*addr = entry->xen;
675 			ist_okay = entry->ist_okay;
676 			found = true;
677 			break;
678 		}
679 	}
680 
681 	if (nr == ARRAY_SIZE(trap_array) &&
682 	    *addr >= (void *)early_idt_handler_array[0] &&
683 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
684 		nr = (*addr - (void *)early_idt_handler_array[0]) /
685 		     EARLY_IDT_HANDLER_SIZE;
686 		*addr = (void *)xen_early_idt_handler_array[nr];
687 		found = true;
688 	}
689 
690 	if (!found)
691 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
692 
693 	if (WARN_ON(found && ist != 0 && !ist_okay))
694 		return false;
695 
696 	return true;
697 }
698 
699 static int cvt_gate_to_trap(int vector, const gate_desc *val,
700 			    struct trap_info *info)
701 {
702 	unsigned long addr;
703 
704 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
705 		return 0;
706 
707 	info->vector = vector;
708 
709 	addr = gate_offset(val);
710 	if (!get_trap_addr((void **)&addr, val->bits.ist))
711 		return 0;
712 	info->address = addr;
713 
714 	info->cs = gate_segment(val);
715 	info->flags = val->bits.dpl;
716 	/* interrupt gates clear IF */
717 	if (val->bits.type == GATE_INTERRUPT)
718 		info->flags |= 1 << 2;
719 
720 	return 1;
721 }
722 
723 /* Locations of each CPU's IDT */
724 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
725 
726 /* Set an IDT entry.  If the entry is part of the current IDT, then
727    also update Xen. */
728 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
729 {
730 	unsigned long p = (unsigned long)&dt[entrynum];
731 	unsigned long start, end;
732 
733 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
734 
735 	preempt_disable();
736 
737 	start = __this_cpu_read(idt_desc.address);
738 	end = start + __this_cpu_read(idt_desc.size) + 1;
739 
740 	xen_mc_flush();
741 
742 	native_write_idt_entry(dt, entrynum, g);
743 
744 	if (p >= start && (p + 8) <= end) {
745 		struct trap_info info[2];
746 
747 		info[1].address = 0;
748 
749 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
750 			if (HYPERVISOR_set_trap_table(info))
751 				BUG();
752 	}
753 
754 	preempt_enable();
755 }
756 
757 static void xen_convert_trap_info(const struct desc_ptr *desc,
758 				  struct trap_info *traps)
759 {
760 	unsigned in, out, count;
761 
762 	count = (desc->size+1) / sizeof(gate_desc);
763 	BUG_ON(count > 256);
764 
765 	for (in = out = 0; in < count; in++) {
766 		gate_desc *entry = (gate_desc *)(desc->address) + in;
767 
768 		if (cvt_gate_to_trap(in, entry, &traps[out]))
769 			out++;
770 	}
771 	traps[out].address = 0;
772 }
773 
774 void xen_copy_trap_info(struct trap_info *traps)
775 {
776 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
777 
778 	xen_convert_trap_info(desc, traps);
779 }
780 
781 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
782    hold a spinlock to protect the static traps[] array (static because
783    it avoids allocation, and saves stack space). */
784 static void xen_load_idt(const struct desc_ptr *desc)
785 {
786 	static DEFINE_SPINLOCK(lock);
787 	static struct trap_info traps[257];
788 
789 	trace_xen_cpu_load_idt(desc);
790 
791 	spin_lock(&lock);
792 
793 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
794 
795 	xen_convert_trap_info(desc, traps);
796 
797 	xen_mc_flush();
798 	if (HYPERVISOR_set_trap_table(traps))
799 		BUG();
800 
801 	spin_unlock(&lock);
802 }
803 
804 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
805    they're handled differently. */
806 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
807 				const void *desc, int type)
808 {
809 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
810 
811 	preempt_disable();
812 
813 	switch (type) {
814 	case DESC_LDT:
815 	case DESC_TSS:
816 		/* ignore */
817 		break;
818 
819 	default: {
820 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
821 
822 		xen_mc_flush();
823 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
824 			BUG();
825 	}
826 
827 	}
828 
829 	preempt_enable();
830 }
831 
832 /*
833  * Version of write_gdt_entry for use at early boot-time needed to
834  * update an entry as simply as possible.
835  */
836 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
837 					    const void *desc, int type)
838 {
839 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
840 
841 	switch (type) {
842 	case DESC_LDT:
843 	case DESC_TSS:
844 		/* ignore */
845 		break;
846 
847 	default: {
848 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
849 
850 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
851 			dt[entry] = *(struct desc_struct *)desc;
852 	}
853 
854 	}
855 }
856 
857 static void xen_load_sp0(unsigned long sp0)
858 {
859 	struct multicall_space mcs;
860 
861 	mcs = xen_mc_entry(0);
862 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
863 	xen_mc_issue(PARAVIRT_LAZY_CPU);
864 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
865 }
866 
867 #ifdef CONFIG_X86_IOPL_IOPERM
868 static void xen_invalidate_io_bitmap(void)
869 {
870 	struct physdev_set_iobitmap iobitmap = {
871 		.bitmap = NULL,
872 		.nr_ports = 0,
873 	};
874 
875 	native_tss_invalidate_io_bitmap();
876 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
877 }
878 
879 static void xen_update_io_bitmap(void)
880 {
881 	struct physdev_set_iobitmap iobitmap;
882 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
883 
884 	native_tss_update_io_bitmap();
885 
886 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
887 			  tss->x86_tss.io_bitmap_base;
888 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
889 		iobitmap.nr_ports = 0;
890 	else
891 		iobitmap.nr_ports = IO_BITMAP_BITS;
892 
893 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
894 }
895 #endif
896 
897 static void xen_io_delay(void)
898 {
899 }
900 
901 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
902 
903 static unsigned long xen_read_cr0(void)
904 {
905 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
906 
907 	if (unlikely(cr0 == 0)) {
908 		cr0 = native_read_cr0();
909 		this_cpu_write(xen_cr0_value, cr0);
910 	}
911 
912 	return cr0;
913 }
914 
915 static void xen_write_cr0(unsigned long cr0)
916 {
917 	struct multicall_space mcs;
918 
919 	this_cpu_write(xen_cr0_value, cr0);
920 
921 	/* Only pay attention to cr0.TS; everything else is
922 	   ignored. */
923 	mcs = xen_mc_entry(0);
924 
925 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
926 
927 	xen_mc_issue(PARAVIRT_LAZY_CPU);
928 }
929 
930 static void xen_write_cr4(unsigned long cr4)
931 {
932 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
933 
934 	native_write_cr4(cr4);
935 }
936 
937 static u64 xen_read_msr_safe(unsigned int msr, int *err)
938 {
939 	u64 val;
940 
941 	if (pmu_msr_read(msr, &val, err))
942 		return val;
943 
944 	val = native_read_msr_safe(msr, err);
945 	switch (msr) {
946 	case MSR_IA32_APICBASE:
947 		val &= ~X2APIC_ENABLE;
948 		break;
949 	}
950 	return val;
951 }
952 
953 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
954 {
955 	int ret;
956 	unsigned int which;
957 	u64 base;
958 
959 	ret = 0;
960 
961 	switch (msr) {
962 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
963 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
964 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
965 
966 	set:
967 		base = ((u64)high << 32) | low;
968 		if (HYPERVISOR_set_segment_base(which, base) != 0)
969 			ret = -EIO;
970 		break;
971 
972 	case MSR_STAR:
973 	case MSR_CSTAR:
974 	case MSR_LSTAR:
975 	case MSR_SYSCALL_MASK:
976 	case MSR_IA32_SYSENTER_CS:
977 	case MSR_IA32_SYSENTER_ESP:
978 	case MSR_IA32_SYSENTER_EIP:
979 		/* Fast syscall setup is all done in hypercalls, so
980 		   these are all ignored.  Stub them out here to stop
981 		   Xen console noise. */
982 		break;
983 
984 	default:
985 		if (!pmu_msr_write(msr, low, high, &ret))
986 			ret = native_write_msr_safe(msr, low, high);
987 	}
988 
989 	return ret;
990 }
991 
992 static u64 xen_read_msr(unsigned int msr)
993 {
994 	/*
995 	 * This will silently swallow a #GP from RDMSR.  It may be worth
996 	 * changing that.
997 	 */
998 	int err;
999 
1000 	return xen_read_msr_safe(msr, &err);
1001 }
1002 
1003 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1004 {
1005 	/*
1006 	 * This will silently swallow a #GP from WRMSR.  It may be worth
1007 	 * changing that.
1008 	 */
1009 	xen_write_msr_safe(msr, low, high);
1010 }
1011 
1012 /* This is called once we have the cpu_possible_mask */
1013 void __init xen_setup_vcpu_info_placement(void)
1014 {
1015 	int cpu;
1016 
1017 	for_each_possible_cpu(cpu) {
1018 		/* Set up direct vCPU id mapping for PV guests. */
1019 		per_cpu(xen_vcpu_id, cpu) = cpu;
1020 
1021 		/*
1022 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
1023 		 * falls back to the shared_info version for cpus
1024 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1025 		 *
1026 		 * xen_cpu_up_prepare_pv() handles the rest by failing
1027 		 * them in hotplug.
1028 		 */
1029 		(void) xen_vcpu_setup(cpu);
1030 	}
1031 
1032 	/*
1033 	 * xen_vcpu_setup managed to place the vcpu_info within the
1034 	 * percpu area for all cpus, so make use of it.
1035 	 */
1036 	if (xen_have_vcpu_info_placement) {
1037 		pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1038 		pv_ops.irq.irq_disable =
1039 			__PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1040 		pv_ops.irq.irq_enable =
1041 			__PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1042 		pv_ops.mmu.read_cr2 =
1043 			__PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1044 	}
1045 }
1046 
1047 static const struct pv_info xen_info __initconst = {
1048 	.extra_user_64bit_cs = FLAT_USER_CS64,
1049 	.name = "Xen",
1050 };
1051 
1052 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1053 	.cpuid = xen_cpuid,
1054 
1055 	.set_debugreg = xen_set_debugreg,
1056 	.get_debugreg = xen_get_debugreg,
1057 
1058 	.read_cr0 = xen_read_cr0,
1059 	.write_cr0 = xen_write_cr0,
1060 
1061 	.write_cr4 = xen_write_cr4,
1062 
1063 	.wbinvd = native_wbinvd,
1064 
1065 	.read_msr = xen_read_msr,
1066 	.write_msr = xen_write_msr,
1067 
1068 	.read_msr_safe = xen_read_msr_safe,
1069 	.write_msr_safe = xen_write_msr_safe,
1070 
1071 	.read_pmc = xen_read_pmc,
1072 
1073 	.load_tr_desc = paravirt_nop,
1074 	.set_ldt = xen_set_ldt,
1075 	.load_gdt = xen_load_gdt,
1076 	.load_idt = xen_load_idt,
1077 	.load_tls = xen_load_tls,
1078 	.load_gs_index = xen_load_gs_index,
1079 
1080 	.alloc_ldt = xen_alloc_ldt,
1081 	.free_ldt = xen_free_ldt,
1082 
1083 	.store_tr = xen_store_tr,
1084 
1085 	.write_ldt_entry = xen_write_ldt_entry,
1086 	.write_gdt_entry = xen_write_gdt_entry,
1087 	.write_idt_entry = xen_write_idt_entry,
1088 	.load_sp0 = xen_load_sp0,
1089 
1090 #ifdef CONFIG_X86_IOPL_IOPERM
1091 	.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1092 	.update_io_bitmap = xen_update_io_bitmap,
1093 #endif
1094 	.io_delay = xen_io_delay,
1095 
1096 	.start_context_switch = paravirt_start_context_switch,
1097 	.end_context_switch = xen_end_context_switch,
1098 };
1099 
1100 static void xen_restart(char *msg)
1101 {
1102 	xen_reboot(SHUTDOWN_reboot);
1103 }
1104 
1105 static void xen_machine_halt(void)
1106 {
1107 	xen_reboot(SHUTDOWN_poweroff);
1108 }
1109 
1110 static void xen_machine_power_off(void)
1111 {
1112 	if (pm_power_off)
1113 		pm_power_off();
1114 	xen_reboot(SHUTDOWN_poweroff);
1115 }
1116 
1117 static void xen_crash_shutdown(struct pt_regs *regs)
1118 {
1119 	xen_reboot(SHUTDOWN_crash);
1120 }
1121 
1122 static const struct machine_ops xen_machine_ops __initconst = {
1123 	.restart = xen_restart,
1124 	.halt = xen_machine_halt,
1125 	.power_off = xen_machine_power_off,
1126 	.shutdown = xen_machine_halt,
1127 	.crash_shutdown = xen_crash_shutdown,
1128 	.emergency_restart = xen_emergency_restart,
1129 };
1130 
1131 static unsigned char xen_get_nmi_reason(void)
1132 {
1133 	unsigned char reason = 0;
1134 
1135 	/* Construct a value which looks like it came from port 0x61. */
1136 	if (test_bit(_XEN_NMIREASON_io_error,
1137 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1138 		reason |= NMI_REASON_IOCHK;
1139 	if (test_bit(_XEN_NMIREASON_pci_serr,
1140 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1141 		reason |= NMI_REASON_SERR;
1142 
1143 	return reason;
1144 }
1145 
1146 static void __init xen_boot_params_init_edd(void)
1147 {
1148 #if IS_ENABLED(CONFIG_EDD)
1149 	struct xen_platform_op op;
1150 	struct edd_info *edd_info;
1151 	u32 *mbr_signature;
1152 	unsigned nr;
1153 	int ret;
1154 
1155 	edd_info = boot_params.eddbuf;
1156 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1157 
1158 	op.cmd = XENPF_firmware_info;
1159 
1160 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1161 	for (nr = 0; nr < EDDMAXNR; nr++) {
1162 		struct edd_info *info = edd_info + nr;
1163 
1164 		op.u.firmware_info.index = nr;
1165 		info->params.length = sizeof(info->params);
1166 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1167 				     &info->params);
1168 		ret = HYPERVISOR_platform_op(&op);
1169 		if (ret)
1170 			break;
1171 
1172 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1173 		C(device);
1174 		C(version);
1175 		C(interface_support);
1176 		C(legacy_max_cylinder);
1177 		C(legacy_max_head);
1178 		C(legacy_sectors_per_track);
1179 #undef C
1180 	}
1181 	boot_params.eddbuf_entries = nr;
1182 
1183 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1184 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1185 		op.u.firmware_info.index = nr;
1186 		ret = HYPERVISOR_platform_op(&op);
1187 		if (ret)
1188 			break;
1189 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1190 	}
1191 	boot_params.edd_mbr_sig_buf_entries = nr;
1192 #endif
1193 }
1194 
1195 /*
1196  * Set up the GDT and segment registers for -fstack-protector.  Until
1197  * we do this, we have to be careful not to call any stack-protected
1198  * function, which is most of the kernel.
1199  */
1200 static void __init xen_setup_gdt(int cpu)
1201 {
1202 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1203 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1204 
1205 	switch_to_new_gdt(cpu);
1206 
1207 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1208 	pv_ops.cpu.load_gdt = xen_load_gdt;
1209 }
1210 
1211 static void __init xen_dom0_set_legacy_features(void)
1212 {
1213 	x86_platform.legacy.rtc = 1;
1214 }
1215 
1216 /* First C function to be called on Xen boot */
1217 asmlinkage __visible void __init xen_start_kernel(void)
1218 {
1219 	struct physdev_set_iopl set_iopl;
1220 	unsigned long initrd_start = 0;
1221 	int rc;
1222 
1223 	if (!xen_start_info)
1224 		return;
1225 
1226 	xen_domain_type = XEN_PV_DOMAIN;
1227 	xen_start_flags = xen_start_info->flags;
1228 
1229 	xen_setup_features();
1230 
1231 	/* Install Xen paravirt ops */
1232 	pv_info = xen_info;
1233 	pv_ops.cpu = xen_cpu_ops;
1234 	paravirt_iret = xen_iret;
1235 	xen_init_irq_ops();
1236 
1237 	/*
1238 	 * Setup xen_vcpu early because it is needed for
1239 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1240 	 *
1241 	 * Don't do the full vcpu_info placement stuff until we have
1242 	 * the cpu_possible_mask and a non-dummy shared_info.
1243 	 */
1244 	xen_vcpu_info_reset(0);
1245 
1246 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1247 
1248 	x86_init.resources.memory_setup = xen_memory_setup;
1249 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1250 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1251 	x86_init.oem.arch_setup = xen_arch_setup;
1252 	x86_init.oem.banner = xen_banner;
1253 	x86_init.hyper.init_platform = xen_pv_init_platform;
1254 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1255 
1256 	/*
1257 	 * Set up some pagetable state before starting to set any ptes.
1258 	 */
1259 
1260 	xen_setup_machphys_mapping();
1261 	xen_init_mmu_ops();
1262 
1263 	/* Prevent unwanted bits from being set in PTEs. */
1264 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1265 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1266 
1267 	/*
1268 	 * Prevent page tables from being allocated in highmem, even
1269 	 * if CONFIG_HIGHPTE is enabled.
1270 	 */
1271 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1272 
1273 	/* Get mfn list */
1274 	xen_build_dynamic_phys_to_machine();
1275 
1276 	/*
1277 	 * Set up kernel GDT and segment registers, mainly so that
1278 	 * -fstack-protector code can be executed.
1279 	 */
1280 	xen_setup_gdt(0);
1281 
1282 	/* Work out if we support NX */
1283 	get_cpu_cap(&boot_cpu_data);
1284 	x86_configure_nx();
1285 
1286 	/* Determine virtual and physical address sizes */
1287 	get_cpu_address_sizes(&boot_cpu_data);
1288 
1289 	/* Let's presume PV guests always boot on vCPU with id 0. */
1290 	per_cpu(xen_vcpu_id, 0) = 0;
1291 
1292 	idt_setup_early_handler();
1293 
1294 	xen_init_capabilities();
1295 
1296 #ifdef CONFIG_X86_LOCAL_APIC
1297 	/*
1298 	 * set up the basic apic ops.
1299 	 */
1300 	xen_init_apic();
1301 #endif
1302 
1303 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1304 		pv_ops.mmu.ptep_modify_prot_start =
1305 			xen_ptep_modify_prot_start;
1306 		pv_ops.mmu.ptep_modify_prot_commit =
1307 			xen_ptep_modify_prot_commit;
1308 	}
1309 
1310 	machine_ops = xen_machine_ops;
1311 
1312 	/*
1313 	 * The only reliable way to retain the initial address of the
1314 	 * percpu gdt_page is to remember it here, so we can go and
1315 	 * mark it RW later, when the initial percpu area is freed.
1316 	 */
1317 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1318 
1319 	xen_smp_init();
1320 
1321 #ifdef CONFIG_ACPI_NUMA
1322 	/*
1323 	 * The pages we from Xen are not related to machine pages, so
1324 	 * any NUMA information the kernel tries to get from ACPI will
1325 	 * be meaningless.  Prevent it from trying.
1326 	 */
1327 	disable_srat();
1328 #endif
1329 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1330 
1331 	local_irq_disable();
1332 	early_boot_irqs_disabled = true;
1333 
1334 	xen_raw_console_write("mapping kernel into physical memory\n");
1335 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1336 				   xen_start_info->nr_pages);
1337 	xen_reserve_special_pages();
1338 
1339 	/*
1340 	 * We used to do this in xen_arch_setup, but that is too late
1341 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1342 	 * early_amd_init which pokes 0xcf8 port.
1343 	 */
1344 	set_iopl.iopl = 1;
1345 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1346 	if (rc != 0)
1347 		xen_raw_printk("physdev_op failed %d\n", rc);
1348 
1349 
1350 	if (xen_start_info->mod_start) {
1351 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1352 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1353 	    else
1354 		initrd_start = __pa(xen_start_info->mod_start);
1355 	}
1356 
1357 	/* Poke various useful things into boot_params */
1358 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1359 	boot_params.hdr.ramdisk_image = initrd_start;
1360 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1361 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1362 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1363 
1364 	if (!xen_initial_domain()) {
1365 		add_preferred_console("xenboot", 0, NULL);
1366 		if (pci_xen)
1367 			x86_init.pci.arch_init = pci_xen_init;
1368 	} else {
1369 		const struct dom0_vga_console_info *info =
1370 			(void *)((char *)xen_start_info +
1371 				 xen_start_info->console.dom0.info_off);
1372 		struct xen_platform_op op = {
1373 			.cmd = XENPF_firmware_info,
1374 			.interface_version = XENPF_INTERFACE_VERSION,
1375 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1376 		};
1377 
1378 		x86_platform.set_legacy_features =
1379 				xen_dom0_set_legacy_features;
1380 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1381 		xen_start_info->console.domU.mfn = 0;
1382 		xen_start_info->console.domU.evtchn = 0;
1383 
1384 		if (HYPERVISOR_platform_op(&op) == 0)
1385 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1386 
1387 		/* Make sure ACS will be enabled */
1388 		pci_request_acs();
1389 
1390 		xen_acpi_sleep_register();
1391 
1392 		/* Avoid searching for BIOS MP tables */
1393 		x86_init.mpparse.find_smp_config = x86_init_noop;
1394 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1395 
1396 		xen_boot_params_init_edd();
1397 
1398 #ifdef CONFIG_ACPI
1399 		/*
1400 		 * Disable selecting "Firmware First mode" for correctable
1401 		 * memory errors, as this is the duty of the hypervisor to
1402 		 * decide.
1403 		 */
1404 		acpi_disable_cmcff = 1;
1405 #endif
1406 	}
1407 
1408 	if (!boot_params.screen_info.orig_video_isVGA)
1409 		add_preferred_console("tty", 0, NULL);
1410 	add_preferred_console("hvc", 0, NULL);
1411 	if (boot_params.screen_info.orig_video_isVGA)
1412 		add_preferred_console("tty", 0, NULL);
1413 
1414 #ifdef CONFIG_PCI
1415 	/* PCI BIOS service won't work from a PV guest. */
1416 	pci_probe &= ~PCI_PROBE_BIOS;
1417 #endif
1418 	xen_raw_console_write("about to get started...\n");
1419 
1420 	/* We need this for printk timestamps */
1421 	xen_setup_runstate_info(0);
1422 
1423 	xen_efi_init(&boot_params);
1424 
1425 	/* Start the world */
1426 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1427 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1428 }
1429 
1430 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1431 {
1432 	int rc;
1433 
1434 	if (per_cpu(xen_vcpu, cpu) == NULL)
1435 		return -ENODEV;
1436 
1437 	xen_setup_timer(cpu);
1438 
1439 	rc = xen_smp_intr_init(cpu);
1440 	if (rc) {
1441 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1442 		     cpu, rc);
1443 		return rc;
1444 	}
1445 
1446 	rc = xen_smp_intr_init_pv(cpu);
1447 	if (rc) {
1448 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1449 		     cpu, rc);
1450 		return rc;
1451 	}
1452 
1453 	return 0;
1454 }
1455 
1456 static int xen_cpu_dead_pv(unsigned int cpu)
1457 {
1458 	xen_smp_intr_free(cpu);
1459 	xen_smp_intr_free_pv(cpu);
1460 
1461 	xen_teardown_timer(cpu);
1462 
1463 	return 0;
1464 }
1465 
1466 static uint32_t __init xen_platform_pv(void)
1467 {
1468 	if (xen_pv_domain())
1469 		return xen_cpuid_base();
1470 
1471 	return 0;
1472 }
1473 
1474 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1475 	.name                   = "Xen PV",
1476 	.detect                 = xen_platform_pv,
1477 	.type			= X86_HYPER_XEN_PV,
1478 	.runtime.pin_vcpu       = xen_pin_vcpu,
1479 	.ignore_nopv		= true,
1480 };
1481