xref: /linux/arch/x86/xen/enlighten_pv.c (revision 320fefa9e2edc67011e235ea1d50f0d00ddfe004)
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 
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 __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 
280 	/*
281 	 * Xen PV would need some work to support PCID: CR3 handling as well
282 	 * as xen_flush_tlb_others() would need updating.
283 	 */
284 	setup_clear_cpu_cap(X86_FEATURE_PCID);
285 
286 	if (!xen_initial_domain())
287 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
288 
289 	if (xen_check_mwait())
290 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
291 	else
292 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
293 
294 	if (!xen_check_xsave()) {
295 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
296 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
297 	}
298 }
299 
300 static noinstr void xen_set_debugreg(int reg, unsigned long val)
301 {
302 	HYPERVISOR_set_debugreg(reg, val);
303 }
304 
305 static noinstr unsigned long xen_get_debugreg(int reg)
306 {
307 	return HYPERVISOR_get_debugreg(reg);
308 }
309 
310 static void xen_end_context_switch(struct task_struct *next)
311 {
312 	xen_mc_flush();
313 	paravirt_end_context_switch(next);
314 }
315 
316 static unsigned long xen_store_tr(void)
317 {
318 	return 0;
319 }
320 
321 /*
322  * Set the page permissions for a particular virtual address.  If the
323  * address is a vmalloc mapping (or other non-linear mapping), then
324  * find the linear mapping of the page and also set its protections to
325  * match.
326  */
327 static void set_aliased_prot(void *v, pgprot_t prot)
328 {
329 	int level;
330 	pte_t *ptep;
331 	pte_t pte;
332 	unsigned long pfn;
333 	unsigned char dummy;
334 	void *va;
335 
336 	ptep = lookup_address((unsigned long)v, &level);
337 	BUG_ON(ptep == NULL);
338 
339 	pfn = pte_pfn(*ptep);
340 	pte = pfn_pte(pfn, prot);
341 
342 	/*
343 	 * Careful: update_va_mapping() will fail if the virtual address
344 	 * we're poking isn't populated in the page tables.  We don't
345 	 * need to worry about the direct map (that's always in the page
346 	 * tables), but we need to be careful about vmap space.  In
347 	 * particular, the top level page table can lazily propagate
348 	 * entries between processes, so if we've switched mms since we
349 	 * vmapped the target in the first place, we might not have the
350 	 * top-level page table entry populated.
351 	 *
352 	 * We disable preemption because we want the same mm active when
353 	 * we probe the target and when we issue the hypercall.  We'll
354 	 * have the same nominal mm, but if we're a kernel thread, lazy
355 	 * mm dropping could change our pgd.
356 	 *
357 	 * Out of an abundance of caution, this uses __get_user() to fault
358 	 * in the target address just in case there's some obscure case
359 	 * in which the target address isn't readable.
360 	 */
361 
362 	preempt_disable();
363 
364 	copy_from_kernel_nofault(&dummy, v, 1);
365 
366 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
367 		BUG();
368 
369 	va = __va(PFN_PHYS(pfn));
370 
371 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
372 		BUG();
373 
374 	preempt_enable();
375 }
376 
377 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
378 {
379 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
380 	int i;
381 
382 	/*
383 	 * We need to mark the all aliases of the LDT pages RO.  We
384 	 * don't need to call vm_flush_aliases(), though, since that's
385 	 * only responsible for flushing aliases out the TLBs, not the
386 	 * page tables, and Xen will flush the TLB for us if needed.
387 	 *
388 	 * To avoid confusing future readers: none of this is necessary
389 	 * to load the LDT.  The hypervisor only checks this when the
390 	 * LDT is faulted in due to subsequent descriptor access.
391 	 */
392 
393 	for (i = 0; i < entries; i += entries_per_page)
394 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
395 }
396 
397 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
398 {
399 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
400 	int i;
401 
402 	for (i = 0; i < entries; i += entries_per_page)
403 		set_aliased_prot(ldt + i, PAGE_KERNEL);
404 }
405 
406 static void xen_set_ldt(const void *addr, unsigned entries)
407 {
408 	struct mmuext_op *op;
409 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
410 
411 	trace_xen_cpu_set_ldt(addr, entries);
412 
413 	op = mcs.args;
414 	op->cmd = MMUEXT_SET_LDT;
415 	op->arg1.linear_addr = (unsigned long)addr;
416 	op->arg2.nr_ents = entries;
417 
418 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
419 
420 	xen_mc_issue(PARAVIRT_LAZY_CPU);
421 }
422 
423 static void xen_load_gdt(const struct desc_ptr *dtr)
424 {
425 	unsigned long va = dtr->address;
426 	unsigned int size = dtr->size + 1;
427 	unsigned long pfn, mfn;
428 	int level;
429 	pte_t *ptep;
430 	void *virt;
431 
432 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
433 	BUG_ON(size > PAGE_SIZE);
434 	BUG_ON(va & ~PAGE_MASK);
435 
436 	/*
437 	 * The GDT is per-cpu and is in the percpu data area.
438 	 * That can be virtually mapped, so we need to do a
439 	 * page-walk to get the underlying MFN for the
440 	 * hypercall.  The page can also be in the kernel's
441 	 * linear range, so we need to RO that mapping too.
442 	 */
443 	ptep = lookup_address(va, &level);
444 	BUG_ON(ptep == NULL);
445 
446 	pfn = pte_pfn(*ptep);
447 	mfn = pfn_to_mfn(pfn);
448 	virt = __va(PFN_PHYS(pfn));
449 
450 	make_lowmem_page_readonly((void *)va);
451 	make_lowmem_page_readonly(virt);
452 
453 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
454 		BUG();
455 }
456 
457 /*
458  * load_gdt for early boot, when the gdt is only mapped once
459  */
460 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
461 {
462 	unsigned long va = dtr->address;
463 	unsigned int size = dtr->size + 1;
464 	unsigned long pfn, mfn;
465 	pte_t pte;
466 
467 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
468 	BUG_ON(size > PAGE_SIZE);
469 	BUG_ON(va & ~PAGE_MASK);
470 
471 	pfn = virt_to_pfn(va);
472 	mfn = pfn_to_mfn(pfn);
473 
474 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
475 
476 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
477 		BUG();
478 
479 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
480 		BUG();
481 }
482 
483 static inline bool desc_equal(const struct desc_struct *d1,
484 			      const struct desc_struct *d2)
485 {
486 	return !memcmp(d1, d2, sizeof(*d1));
487 }
488 
489 static void load_TLS_descriptor(struct thread_struct *t,
490 				unsigned int cpu, unsigned int i)
491 {
492 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
493 	struct desc_struct *gdt;
494 	xmaddr_t maddr;
495 	struct multicall_space mc;
496 
497 	if (desc_equal(shadow, &t->tls_array[i]))
498 		return;
499 
500 	*shadow = t->tls_array[i];
501 
502 	gdt = get_cpu_gdt_rw(cpu);
503 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
504 	mc = __xen_mc_entry(0);
505 
506 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
507 }
508 
509 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
510 {
511 	/*
512 	 * In lazy mode we need to zero %fs, otherwise we may get an
513 	 * exception between the new %fs descriptor being loaded and
514 	 * %fs being effectively cleared at __switch_to().
515 	 */
516 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
517 		loadsegment(fs, 0);
518 
519 	xen_mc_batch();
520 
521 	load_TLS_descriptor(t, cpu, 0);
522 	load_TLS_descriptor(t, cpu, 1);
523 	load_TLS_descriptor(t, cpu, 2);
524 
525 	xen_mc_issue(PARAVIRT_LAZY_CPU);
526 }
527 
528 static void xen_load_gs_index(unsigned int idx)
529 {
530 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
531 		BUG();
532 }
533 
534 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
535 				const void *ptr)
536 {
537 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
538 	u64 entry = *(u64 *)ptr;
539 
540 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
541 
542 	preempt_disable();
543 
544 	xen_mc_flush();
545 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
546 		BUG();
547 
548 	preempt_enable();
549 }
550 
551 void noist_exc_debug(struct pt_regs *regs);
552 
553 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
554 {
555 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
556 	exc_nmi(regs);
557 }
558 
559 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
560 {
561 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
562 	exc_double_fault(regs, error_code);
563 }
564 
565 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
566 {
567 	/*
568 	 * There's no IST on Xen PV, but we still need to dispatch
569 	 * to the correct handler.
570 	 */
571 	if (user_mode(regs))
572 		noist_exc_debug(regs);
573 	else
574 		exc_debug(regs);
575 }
576 
577 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
578 {
579 	/* This should never happen and there is no way to handle it. */
580 	instrumentation_begin();
581 	pr_err("Unknown trap in Xen PV mode.");
582 	BUG();
583 	instrumentation_end();
584 }
585 
586 #ifdef CONFIG_X86_MCE
587 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
588 {
589 	/*
590 	 * There's no IST on Xen PV, but we still need to dispatch
591 	 * to the correct handler.
592 	 */
593 	if (user_mode(regs))
594 		noist_exc_machine_check(regs);
595 	else
596 		exc_machine_check(regs);
597 }
598 #endif
599 
600 struct trap_array_entry {
601 	void (*orig)(void);
602 	void (*xen)(void);
603 	bool ist_okay;
604 };
605 
606 #define TRAP_ENTRY(func, ist_ok) {			\
607 	.orig		= asm_##func,			\
608 	.xen		= xen_asm_##func,		\
609 	.ist_okay	= ist_ok }
610 
611 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
612 	.orig		= asm_##func,			\
613 	.xen		= xen_asm_xenpv_##func,		\
614 	.ist_okay	= ist_ok }
615 
616 static struct trap_array_entry trap_array[] = {
617 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
618 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
619 #ifdef CONFIG_X86_MCE
620 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
621 #endif
622 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
623 	TRAP_ENTRY(exc_int3,				false ),
624 	TRAP_ENTRY(exc_overflow,			false ),
625 #ifdef CONFIG_IA32_EMULATION
626 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
627 #endif
628 	TRAP_ENTRY(exc_page_fault,			false ),
629 	TRAP_ENTRY(exc_divide_error,			false ),
630 	TRAP_ENTRY(exc_bounds,				false ),
631 	TRAP_ENTRY(exc_invalid_op,			false ),
632 	TRAP_ENTRY(exc_device_not_available,		false ),
633 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
634 	TRAP_ENTRY(exc_invalid_tss,			false ),
635 	TRAP_ENTRY(exc_segment_not_present,		false ),
636 	TRAP_ENTRY(exc_stack_segment,			false ),
637 	TRAP_ENTRY(exc_general_protection,		false ),
638 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
639 	TRAP_ENTRY(exc_coprocessor_error,		false ),
640 	TRAP_ENTRY(exc_alignment_check,			false ),
641 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
642 #ifdef CONFIG_X86_KERNEL_IBT
643 	TRAP_ENTRY(exc_control_protection,		false ),
644 #endif
645 };
646 
647 static bool __ref get_trap_addr(void **addr, unsigned int ist)
648 {
649 	unsigned int nr;
650 	bool ist_okay = false;
651 	bool found = false;
652 
653 	/*
654 	 * Replace trap handler addresses by Xen specific ones.
655 	 * Check for known traps using IST and whitelist them.
656 	 * The debugger ones are the only ones we care about.
657 	 * Xen will handle faults like double_fault, so we should never see
658 	 * them.  Warn if there's an unexpected IST-using fault handler.
659 	 */
660 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
661 		struct trap_array_entry *entry = trap_array + nr;
662 
663 		if (*addr == entry->orig) {
664 			*addr = entry->xen;
665 			ist_okay = entry->ist_okay;
666 			found = true;
667 			break;
668 		}
669 	}
670 
671 	if (nr == ARRAY_SIZE(trap_array) &&
672 	    *addr >= (void *)early_idt_handler_array[0] &&
673 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
674 		nr = (*addr - (void *)early_idt_handler_array[0]) /
675 		     EARLY_IDT_HANDLER_SIZE;
676 		*addr = (void *)xen_early_idt_handler_array[nr];
677 		found = true;
678 	}
679 
680 	if (!found)
681 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
682 
683 	if (WARN_ON(found && ist != 0 && !ist_okay))
684 		return false;
685 
686 	return true;
687 }
688 
689 static int cvt_gate_to_trap(int vector, const gate_desc *val,
690 			    struct trap_info *info)
691 {
692 	unsigned long addr;
693 
694 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
695 		return 0;
696 
697 	info->vector = vector;
698 
699 	addr = gate_offset(val);
700 	if (!get_trap_addr((void **)&addr, val->bits.ist))
701 		return 0;
702 	info->address = addr;
703 
704 	info->cs = gate_segment(val);
705 	info->flags = val->bits.dpl;
706 	/* interrupt gates clear IF */
707 	if (val->bits.type == GATE_INTERRUPT)
708 		info->flags |= 1 << 2;
709 
710 	return 1;
711 }
712 
713 /* Locations of each CPU's IDT */
714 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
715 
716 /* Set an IDT entry.  If the entry is part of the current IDT, then
717    also update Xen. */
718 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
719 {
720 	unsigned long p = (unsigned long)&dt[entrynum];
721 	unsigned long start, end;
722 
723 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
724 
725 	preempt_disable();
726 
727 	start = __this_cpu_read(idt_desc.address);
728 	end = start + __this_cpu_read(idt_desc.size) + 1;
729 
730 	xen_mc_flush();
731 
732 	native_write_idt_entry(dt, entrynum, g);
733 
734 	if (p >= start && (p + 8) <= end) {
735 		struct trap_info info[2];
736 
737 		info[1].address = 0;
738 
739 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
740 			if (HYPERVISOR_set_trap_table(info))
741 				BUG();
742 	}
743 
744 	preempt_enable();
745 }
746 
747 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
748 				      struct trap_info *traps, bool full)
749 {
750 	unsigned in, out, count;
751 
752 	count = (desc->size+1) / sizeof(gate_desc);
753 	BUG_ON(count > 256);
754 
755 	for (in = out = 0; in < count; in++) {
756 		gate_desc *entry = (gate_desc *)(desc->address) + in;
757 
758 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
759 			out++;
760 	}
761 
762 	return out;
763 }
764 
765 void xen_copy_trap_info(struct trap_info *traps)
766 {
767 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
768 
769 	xen_convert_trap_info(desc, traps, true);
770 }
771 
772 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
773    hold a spinlock to protect the static traps[] array (static because
774    it avoids allocation, and saves stack space). */
775 static void xen_load_idt(const struct desc_ptr *desc)
776 {
777 	static DEFINE_SPINLOCK(lock);
778 	static struct trap_info traps[257];
779 	static const struct trap_info zero = { };
780 	unsigned out;
781 
782 	trace_xen_cpu_load_idt(desc);
783 
784 	spin_lock(&lock);
785 
786 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
787 
788 	out = xen_convert_trap_info(desc, traps, false);
789 	traps[out] = zero;
790 
791 	xen_mc_flush();
792 	if (HYPERVISOR_set_trap_table(traps))
793 		BUG();
794 
795 	spin_unlock(&lock);
796 }
797 
798 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
799    they're handled differently. */
800 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
801 				const void *desc, int type)
802 {
803 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
804 
805 	preempt_disable();
806 
807 	switch (type) {
808 	case DESC_LDT:
809 	case DESC_TSS:
810 		/* ignore */
811 		break;
812 
813 	default: {
814 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
815 
816 		xen_mc_flush();
817 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
818 			BUG();
819 	}
820 
821 	}
822 
823 	preempt_enable();
824 }
825 
826 /*
827  * Version of write_gdt_entry for use at early boot-time needed to
828  * update an entry as simply as possible.
829  */
830 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
831 					    const void *desc, int type)
832 {
833 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
834 
835 	switch (type) {
836 	case DESC_LDT:
837 	case DESC_TSS:
838 		/* ignore */
839 		break;
840 
841 	default: {
842 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
843 
844 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
845 			dt[entry] = *(struct desc_struct *)desc;
846 	}
847 
848 	}
849 }
850 
851 static void xen_load_sp0(unsigned long sp0)
852 {
853 	struct multicall_space mcs;
854 
855 	mcs = xen_mc_entry(0);
856 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
857 	xen_mc_issue(PARAVIRT_LAZY_CPU);
858 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
859 }
860 
861 #ifdef CONFIG_X86_IOPL_IOPERM
862 static void xen_invalidate_io_bitmap(void)
863 {
864 	struct physdev_set_iobitmap iobitmap = {
865 		.bitmap = NULL,
866 		.nr_ports = 0,
867 	};
868 
869 	native_tss_invalidate_io_bitmap();
870 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
871 }
872 
873 static void xen_update_io_bitmap(void)
874 {
875 	struct physdev_set_iobitmap iobitmap;
876 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
877 
878 	native_tss_update_io_bitmap();
879 
880 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
881 			  tss->x86_tss.io_bitmap_base;
882 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
883 		iobitmap.nr_ports = 0;
884 	else
885 		iobitmap.nr_ports = IO_BITMAP_BITS;
886 
887 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
888 }
889 #endif
890 
891 static void xen_io_delay(void)
892 {
893 }
894 
895 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
896 
897 static unsigned long xen_read_cr0(void)
898 {
899 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
900 
901 	if (unlikely(cr0 == 0)) {
902 		cr0 = native_read_cr0();
903 		this_cpu_write(xen_cr0_value, cr0);
904 	}
905 
906 	return cr0;
907 }
908 
909 static void xen_write_cr0(unsigned long cr0)
910 {
911 	struct multicall_space mcs;
912 
913 	this_cpu_write(xen_cr0_value, cr0);
914 
915 	/* Only pay attention to cr0.TS; everything else is
916 	   ignored. */
917 	mcs = xen_mc_entry(0);
918 
919 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
920 
921 	xen_mc_issue(PARAVIRT_LAZY_CPU);
922 }
923 
924 static void xen_write_cr4(unsigned long cr4)
925 {
926 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
927 
928 	native_write_cr4(cr4);
929 }
930 
931 static u64 xen_do_read_msr(unsigned int msr, int *err)
932 {
933 	u64 val = 0;	/* Avoid uninitialized value for safe variant. */
934 
935 	if (pmu_msr_read(msr, &val, err))
936 		return val;
937 
938 	if (err)
939 		val = native_read_msr_safe(msr, err);
940 	else
941 		val = native_read_msr(msr);
942 
943 	switch (msr) {
944 	case MSR_IA32_APICBASE:
945 		val &= ~X2APIC_ENABLE;
946 		break;
947 	}
948 	return val;
949 }
950 
951 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
952 		    int *err)
953 {
954 	u64 base = ((u64)high << 32) | low;
955 
956 	if (HYPERVISOR_set_segment_base(which, base) == 0)
957 		return;
958 
959 	if (err)
960 		*err = -EIO;
961 	else
962 		WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
963 }
964 
965 /*
966  * Support write_msr_safe() and write_msr() semantics.
967  * With err == NULL write_msr() semantics are selected.
968  * Supplying an err pointer requires err to be pre-initialized with 0.
969  */
970 static void xen_do_write_msr(unsigned int msr, unsigned int low,
971 			     unsigned int high, int *err)
972 {
973 	switch (msr) {
974 	case MSR_FS_BASE:
975 		set_seg(SEGBASE_FS, low, high, err);
976 		break;
977 
978 	case MSR_KERNEL_GS_BASE:
979 		set_seg(SEGBASE_GS_USER, low, high, err);
980 		break;
981 
982 	case MSR_GS_BASE:
983 		set_seg(SEGBASE_GS_KERNEL, low, high, err);
984 		break;
985 
986 	case MSR_STAR:
987 	case MSR_CSTAR:
988 	case MSR_LSTAR:
989 	case MSR_SYSCALL_MASK:
990 	case MSR_IA32_SYSENTER_CS:
991 	case MSR_IA32_SYSENTER_ESP:
992 	case MSR_IA32_SYSENTER_EIP:
993 		/* Fast syscall setup is all done in hypercalls, so
994 		   these are all ignored.  Stub them out here to stop
995 		   Xen console noise. */
996 		break;
997 
998 	default:
999 		if (!pmu_msr_write(msr, low, high, err)) {
1000 			if (err)
1001 				*err = native_write_msr_safe(msr, low, high);
1002 			else
1003 				native_write_msr(msr, low, high);
1004 		}
1005 	}
1006 }
1007 
1008 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1009 {
1010 	return xen_do_read_msr(msr, err);
1011 }
1012 
1013 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1014 			      unsigned int high)
1015 {
1016 	int err = 0;
1017 
1018 	xen_do_write_msr(msr, low, high, &err);
1019 
1020 	return err;
1021 }
1022 
1023 static u64 xen_read_msr(unsigned int msr)
1024 {
1025 	int err;
1026 
1027 	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1028 }
1029 
1030 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1031 {
1032 	int err;
1033 
1034 	xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1035 }
1036 
1037 /* This is called once we have the cpu_possible_mask */
1038 void __init xen_setup_vcpu_info_placement(void)
1039 {
1040 	int cpu;
1041 
1042 	for_each_possible_cpu(cpu) {
1043 		/* Set up direct vCPU id mapping for PV guests. */
1044 		per_cpu(xen_vcpu_id, cpu) = cpu;
1045 		xen_vcpu_setup(cpu);
1046 	}
1047 
1048 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1049 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1050 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1051 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1052 }
1053 
1054 static const struct pv_info xen_info __initconst = {
1055 	.extra_user_64bit_cs = FLAT_USER_CS64,
1056 	.name = "Xen",
1057 };
1058 
1059 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1060 	.cpu = {
1061 		.cpuid = xen_cpuid,
1062 
1063 		.set_debugreg = xen_set_debugreg,
1064 		.get_debugreg = xen_get_debugreg,
1065 
1066 		.read_cr0 = xen_read_cr0,
1067 		.write_cr0 = xen_write_cr0,
1068 
1069 		.write_cr4 = xen_write_cr4,
1070 
1071 		.wbinvd = native_wbinvd,
1072 
1073 		.read_msr = xen_read_msr,
1074 		.write_msr = xen_write_msr,
1075 
1076 		.read_msr_safe = xen_read_msr_safe,
1077 		.write_msr_safe = xen_write_msr_safe,
1078 
1079 		.read_pmc = xen_read_pmc,
1080 
1081 		.load_tr_desc = paravirt_nop,
1082 		.set_ldt = xen_set_ldt,
1083 		.load_gdt = xen_load_gdt,
1084 		.load_idt = xen_load_idt,
1085 		.load_tls = xen_load_tls,
1086 		.load_gs_index = xen_load_gs_index,
1087 
1088 		.alloc_ldt = xen_alloc_ldt,
1089 		.free_ldt = xen_free_ldt,
1090 
1091 		.store_tr = xen_store_tr,
1092 
1093 		.write_ldt_entry = xen_write_ldt_entry,
1094 		.write_gdt_entry = xen_write_gdt_entry,
1095 		.write_idt_entry = xen_write_idt_entry,
1096 		.load_sp0 = xen_load_sp0,
1097 
1098 #ifdef CONFIG_X86_IOPL_IOPERM
1099 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1100 		.update_io_bitmap = xen_update_io_bitmap,
1101 #endif
1102 		.io_delay = xen_io_delay,
1103 
1104 		.start_context_switch = paravirt_start_context_switch,
1105 		.end_context_switch = xen_end_context_switch,
1106 	},
1107 };
1108 
1109 static void xen_restart(char *msg)
1110 {
1111 	xen_reboot(SHUTDOWN_reboot);
1112 }
1113 
1114 static void xen_machine_halt(void)
1115 {
1116 	xen_reboot(SHUTDOWN_poweroff);
1117 }
1118 
1119 static void xen_machine_power_off(void)
1120 {
1121 	do_kernel_power_off();
1122 	xen_reboot(SHUTDOWN_poweroff);
1123 }
1124 
1125 static void xen_crash_shutdown(struct pt_regs *regs)
1126 {
1127 	xen_reboot(SHUTDOWN_crash);
1128 }
1129 
1130 static const struct machine_ops xen_machine_ops __initconst = {
1131 	.restart = xen_restart,
1132 	.halt = xen_machine_halt,
1133 	.power_off = xen_machine_power_off,
1134 	.shutdown = xen_machine_halt,
1135 	.crash_shutdown = xen_crash_shutdown,
1136 	.emergency_restart = xen_emergency_restart,
1137 };
1138 
1139 static unsigned char xen_get_nmi_reason(void)
1140 {
1141 	unsigned char reason = 0;
1142 
1143 	/* Construct a value which looks like it came from port 0x61. */
1144 	if (test_bit(_XEN_NMIREASON_io_error,
1145 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1146 		reason |= NMI_REASON_IOCHK;
1147 	if (test_bit(_XEN_NMIREASON_pci_serr,
1148 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1149 		reason |= NMI_REASON_SERR;
1150 
1151 	return reason;
1152 }
1153 
1154 static void __init xen_boot_params_init_edd(void)
1155 {
1156 #if IS_ENABLED(CONFIG_EDD)
1157 	struct xen_platform_op op;
1158 	struct edd_info *edd_info;
1159 	u32 *mbr_signature;
1160 	unsigned nr;
1161 	int ret;
1162 
1163 	edd_info = boot_params.eddbuf;
1164 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1165 
1166 	op.cmd = XENPF_firmware_info;
1167 
1168 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1169 	for (nr = 0; nr < EDDMAXNR; nr++) {
1170 		struct edd_info *info = edd_info + nr;
1171 
1172 		op.u.firmware_info.index = nr;
1173 		info->params.length = sizeof(info->params);
1174 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1175 				     &info->params);
1176 		ret = HYPERVISOR_platform_op(&op);
1177 		if (ret)
1178 			break;
1179 
1180 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1181 		C(device);
1182 		C(version);
1183 		C(interface_support);
1184 		C(legacy_max_cylinder);
1185 		C(legacy_max_head);
1186 		C(legacy_sectors_per_track);
1187 #undef C
1188 	}
1189 	boot_params.eddbuf_entries = nr;
1190 
1191 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1192 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1193 		op.u.firmware_info.index = nr;
1194 		ret = HYPERVISOR_platform_op(&op);
1195 		if (ret)
1196 			break;
1197 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1198 	}
1199 	boot_params.edd_mbr_sig_buf_entries = nr;
1200 #endif
1201 }
1202 
1203 /*
1204  * Set up the GDT and segment registers for -fstack-protector.  Until
1205  * we do this, we have to be careful not to call any stack-protected
1206  * function, which is most of the kernel.
1207  */
1208 static void __init xen_setup_gdt(int cpu)
1209 {
1210 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1211 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1212 
1213 	switch_to_new_gdt(cpu);
1214 
1215 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1216 	pv_ops.cpu.load_gdt = xen_load_gdt;
1217 }
1218 
1219 static void __init xen_dom0_set_legacy_features(void)
1220 {
1221 	x86_platform.legacy.rtc = 1;
1222 }
1223 
1224 static void __init xen_domu_set_legacy_features(void)
1225 {
1226 	x86_platform.legacy.rtc = 0;
1227 }
1228 
1229 extern void early_xen_iret_patch(void);
1230 
1231 /* First C function to be called on Xen boot */
1232 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1233 {
1234 	struct physdev_set_iopl set_iopl;
1235 	unsigned long initrd_start = 0;
1236 	int rc;
1237 
1238 	if (!si)
1239 		return;
1240 
1241 	clear_bss();
1242 
1243 	xen_start_info = si;
1244 
1245 	__text_gen_insn(&early_xen_iret_patch,
1246 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1247 			JMP32_INSN_SIZE);
1248 
1249 	xen_domain_type = XEN_PV_DOMAIN;
1250 	xen_start_flags = xen_start_info->flags;
1251 
1252 	xen_setup_features();
1253 
1254 	/* Install Xen paravirt ops */
1255 	pv_info = xen_info;
1256 	pv_ops.cpu = xen_cpu_ops.cpu;
1257 	xen_init_irq_ops();
1258 
1259 	/*
1260 	 * Setup xen_vcpu early because it is needed for
1261 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1262 	 *
1263 	 * Don't do the full vcpu_info placement stuff until we have
1264 	 * the cpu_possible_mask and a non-dummy shared_info.
1265 	 */
1266 	xen_vcpu_info_reset(0);
1267 
1268 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1269 
1270 	x86_init.resources.memory_setup = xen_memory_setup;
1271 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1272 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1273 	x86_init.oem.arch_setup = xen_arch_setup;
1274 	x86_init.oem.banner = xen_banner;
1275 	x86_init.hyper.init_platform = xen_pv_init_platform;
1276 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1277 
1278 	/*
1279 	 * Set up some pagetable state before starting to set any ptes.
1280 	 */
1281 
1282 	xen_setup_machphys_mapping();
1283 	xen_init_mmu_ops();
1284 
1285 	/* Prevent unwanted bits from being set in PTEs. */
1286 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1287 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1288 
1289 	/* Get mfn list */
1290 	xen_build_dynamic_phys_to_machine();
1291 
1292 	/* Work out if we support NX */
1293 	get_cpu_cap(&boot_cpu_data);
1294 	x86_configure_nx();
1295 
1296 	/*
1297 	 * Set up kernel GDT and segment registers, mainly so that
1298 	 * -fstack-protector code can be executed.
1299 	 */
1300 	xen_setup_gdt(0);
1301 
1302 	/* Determine virtual and physical address sizes */
1303 	get_cpu_address_sizes(&boot_cpu_data);
1304 
1305 	/* Let's presume PV guests always boot on vCPU with id 0. */
1306 	per_cpu(xen_vcpu_id, 0) = 0;
1307 
1308 	idt_setup_early_handler();
1309 
1310 	xen_init_capabilities();
1311 
1312 #ifdef CONFIG_X86_LOCAL_APIC
1313 	/*
1314 	 * set up the basic apic ops.
1315 	 */
1316 	xen_init_apic();
1317 #endif
1318 
1319 	machine_ops = xen_machine_ops;
1320 
1321 	/*
1322 	 * The only reliable way to retain the initial address of the
1323 	 * percpu gdt_page is to remember it here, so we can go and
1324 	 * mark it RW later, when the initial percpu area is freed.
1325 	 */
1326 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1327 
1328 	xen_smp_init();
1329 
1330 #ifdef CONFIG_ACPI_NUMA
1331 	/*
1332 	 * The pages we from Xen are not related to machine pages, so
1333 	 * any NUMA information the kernel tries to get from ACPI will
1334 	 * be meaningless.  Prevent it from trying.
1335 	 */
1336 	disable_srat();
1337 #endif
1338 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1339 
1340 	local_irq_disable();
1341 	early_boot_irqs_disabled = true;
1342 
1343 	xen_raw_console_write("mapping kernel into physical memory\n");
1344 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1345 				   xen_start_info->nr_pages);
1346 	xen_reserve_special_pages();
1347 
1348 	/*
1349 	 * We used to do this in xen_arch_setup, but that is too late
1350 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1351 	 * early_amd_init which pokes 0xcf8 port.
1352 	 */
1353 	set_iopl.iopl = 1;
1354 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1355 	if (rc != 0)
1356 		xen_raw_printk("physdev_op failed %d\n", rc);
1357 
1358 
1359 	if (xen_start_info->mod_start) {
1360 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1361 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1362 	    else
1363 		initrd_start = __pa(xen_start_info->mod_start);
1364 	}
1365 
1366 	/* Poke various useful things into boot_params */
1367 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1368 	boot_params.hdr.ramdisk_image = initrd_start;
1369 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1370 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1371 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1372 
1373 	if (!xen_initial_domain()) {
1374 		if (pci_xen)
1375 			x86_init.pci.arch_init = pci_xen_init;
1376 		x86_platform.set_legacy_features =
1377 				xen_domu_set_legacy_features;
1378 	} else {
1379 		const struct dom0_vga_console_info *info =
1380 			(void *)((char *)xen_start_info +
1381 				 xen_start_info->console.dom0.info_off);
1382 		struct xen_platform_op op = {
1383 			.cmd = XENPF_firmware_info,
1384 			.interface_version = XENPF_INTERFACE_VERSION,
1385 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1386 		};
1387 
1388 		x86_platform.set_legacy_features =
1389 				xen_dom0_set_legacy_features;
1390 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1391 		xen_start_info->console.domU.mfn = 0;
1392 		xen_start_info->console.domU.evtchn = 0;
1393 
1394 		if (HYPERVISOR_platform_op(&op) == 0)
1395 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1396 
1397 		/* Make sure ACS will be enabled */
1398 		pci_request_acs();
1399 
1400 		xen_acpi_sleep_register();
1401 
1402 		xen_boot_params_init_edd();
1403 
1404 #ifdef CONFIG_ACPI
1405 		/*
1406 		 * Disable selecting "Firmware First mode" for correctable
1407 		 * memory errors, as this is the duty of the hypervisor to
1408 		 * decide.
1409 		 */
1410 		acpi_disable_cmcff = 1;
1411 #endif
1412 	}
1413 
1414 	xen_add_preferred_consoles();
1415 
1416 #ifdef CONFIG_PCI
1417 	/* PCI BIOS service won't work from a PV guest. */
1418 	pci_probe &= ~PCI_PROBE_BIOS;
1419 #endif
1420 	xen_raw_console_write("about to get started...\n");
1421 
1422 	/* We need this for printk timestamps */
1423 	xen_setup_runstate_info(0);
1424 
1425 	xen_efi_init(&boot_params);
1426 
1427 	/* Start the world */
1428 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1429 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1430 }
1431 
1432 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1433 {
1434 	int rc;
1435 
1436 	if (per_cpu(xen_vcpu, cpu) == NULL)
1437 		return -ENODEV;
1438 
1439 	xen_setup_timer(cpu);
1440 
1441 	rc = xen_smp_intr_init(cpu);
1442 	if (rc) {
1443 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1444 		     cpu, rc);
1445 		return rc;
1446 	}
1447 
1448 	rc = xen_smp_intr_init_pv(cpu);
1449 	if (rc) {
1450 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1451 		     cpu, rc);
1452 		return rc;
1453 	}
1454 
1455 	return 0;
1456 }
1457 
1458 static int xen_cpu_dead_pv(unsigned int cpu)
1459 {
1460 	xen_smp_intr_free(cpu);
1461 	xen_smp_intr_free_pv(cpu);
1462 
1463 	xen_teardown_timer(cpu);
1464 
1465 	return 0;
1466 }
1467 
1468 static uint32_t __init xen_platform_pv(void)
1469 {
1470 	if (xen_pv_domain())
1471 		return xen_cpuid_base();
1472 
1473 	return 0;
1474 }
1475 
1476 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1477 	.name                   = "Xen PV",
1478 	.detect                 = xen_platform_pv,
1479 	.type			= X86_HYPER_XEN_PV,
1480 	.runtime.pin_vcpu       = xen_pin_vcpu,
1481 	.ignore_nopv		= true,
1482 };
1483