xref: /linux/arch/x86/xen/enlighten_pv.c (revision 6a4aee277740d04ac0fd54cfa17cc28261932ddc)
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/mtrr.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
74 #include <asm/cpu.h>
75 #ifdef CONFIG_X86_IOPL_IOPERM
76 #include <asm/io_bitmap.h>
77 #endif
78 
79 #ifdef CONFIG_ACPI
80 #include <linux/acpi.h>
81 #include <asm/acpi.h>
82 #include <acpi/proc_cap_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
85 #endif
86 
87 #include "xen-ops.h"
88 #include "mmu.h"
89 #include "smp.h"
90 #include "multicalls.h"
91 #include "pmu.h"
92 
93 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
94 
95 void *xen_initial_gdt;
96 
97 static int xen_cpu_up_prepare_pv(unsigned int cpu);
98 static int xen_cpu_dead_pv(unsigned int cpu);
99 
100 struct tls_descs {
101 	struct desc_struct desc[3];
102 };
103 
104 DEFINE_PER_CPU(enum xen_lazy_mode, xen_lazy_mode) = XEN_LAZY_NONE;
105 DEFINE_PER_CPU(unsigned int, xen_lazy_nesting);
106 
107 enum xen_lazy_mode xen_get_lazy_mode(void)
108 {
109 	if (in_interrupt())
110 		return XEN_LAZY_NONE;
111 
112 	return this_cpu_read(xen_lazy_mode);
113 }
114 
115 /*
116  * Updating the 3 TLS descriptors in the GDT on every task switch is
117  * surprisingly expensive so we avoid updating them if they haven't
118  * changed.  Since Xen writes different descriptors than the one
119  * passed in the update_descriptor hypercall we keep shadow copies to
120  * compare against.
121  */
122 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
123 
124 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
125 
126 static int __init parse_xen_msr_safe(char *str)
127 {
128 	if (str)
129 		return kstrtobool(str, &xen_msr_safe);
130 	return -EINVAL;
131 }
132 early_param("xen_msr_safe", parse_xen_msr_safe);
133 
134 /* Get MTRR settings from Xen and put them into mtrr_state. */
135 static void __init xen_set_mtrr_data(void)
136 {
137 #ifdef CONFIG_MTRR
138 	struct xen_platform_op op = {
139 		.cmd = XENPF_read_memtype,
140 		.interface_version = XENPF_INTERFACE_VERSION,
141 	};
142 	unsigned int reg;
143 	unsigned long mask;
144 	uint32_t eax, width;
145 	static struct mtrr_var_range var[MTRR_MAX_VAR_RANGES] __initdata;
146 
147 	/* Get physical address width (only 64-bit cpus supported). */
148 	width = 36;
149 	eax = cpuid_eax(0x80000000);
150 	if ((eax >> 16) == 0x8000 && eax >= 0x80000008) {
151 		eax = cpuid_eax(0x80000008);
152 		width = eax & 0xff;
153 	}
154 
155 	for (reg = 0; reg < MTRR_MAX_VAR_RANGES; reg++) {
156 		op.u.read_memtype.reg = reg;
157 		if (HYPERVISOR_platform_op(&op))
158 			break;
159 
160 		/*
161 		 * Only called in dom0, which has all RAM PFNs mapped at
162 		 * RAM MFNs, and all PCI space etc. is identity mapped.
163 		 * This means we can treat MFN == PFN regarding MTRR settings.
164 		 */
165 		var[reg].base_lo = op.u.read_memtype.type;
166 		var[reg].base_lo |= op.u.read_memtype.mfn << PAGE_SHIFT;
167 		var[reg].base_hi = op.u.read_memtype.mfn >> (32 - PAGE_SHIFT);
168 		mask = ~((op.u.read_memtype.nr_mfns << PAGE_SHIFT) - 1);
169 		mask &= (1UL << width) - 1;
170 		if (mask)
171 			mask |= MTRR_PHYSMASK_V;
172 		var[reg].mask_lo = mask;
173 		var[reg].mask_hi = mask >> 32;
174 	}
175 
176 	/* Only overwrite MTRR state if any MTRR could be got from Xen. */
177 	if (reg)
178 		mtrr_overwrite_state(var, reg, MTRR_TYPE_UNCACHABLE);
179 #endif
180 }
181 
182 static void __init xen_pv_init_platform(void)
183 {
184 	/* PV guests can't operate virtio devices without grants. */
185 	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
186 		virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
187 
188 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
189 
190 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
191 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
192 
193 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
194 	xen_vcpu_info_reset(0);
195 
196 	/* pvclock is in shared info area */
197 	xen_init_time_ops();
198 
199 	if (xen_initial_domain())
200 		xen_set_mtrr_data();
201 	else
202 		mtrr_overwrite_state(NULL, 0, MTRR_TYPE_WRBACK);
203 
204 	/* Adjust nr_cpu_ids before "enumeration" happens */
205 	xen_smp_count_cpus();
206 }
207 
208 static void __init xen_pv_guest_late_init(void)
209 {
210 #ifndef CONFIG_SMP
211 	/* Setup shared vcpu info for non-smp configurations */
212 	xen_setup_vcpu_info_placement();
213 #endif
214 }
215 
216 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
217 static __read_mostly unsigned int cpuid_leaf5_edx_val;
218 
219 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
220 		      unsigned int *cx, unsigned int *dx)
221 {
222 	unsigned maskebx = ~0;
223 
224 	/*
225 	 * Mask out inconvenient features, to try and disable as many
226 	 * unsupported kernel subsystems as possible.
227 	 */
228 	switch (*ax) {
229 	case CPUID_MWAIT_LEAF:
230 		/* Synthesize the values.. */
231 		*ax = 0;
232 		*bx = 0;
233 		*cx = cpuid_leaf5_ecx_val;
234 		*dx = cpuid_leaf5_edx_val;
235 		return;
236 
237 	case 0xb:
238 		/* Suppress extended topology stuff */
239 		maskebx = 0;
240 		break;
241 	}
242 
243 	asm(XEN_EMULATE_PREFIX "cpuid"
244 		: "=a" (*ax),
245 		  "=b" (*bx),
246 		  "=c" (*cx),
247 		  "=d" (*dx)
248 		: "0" (*ax), "2" (*cx));
249 
250 	*bx &= maskebx;
251 }
252 
253 static bool __init xen_check_mwait(void)
254 {
255 #ifdef CONFIG_ACPI
256 	struct xen_platform_op op = {
257 		.cmd			= XENPF_set_processor_pminfo,
258 		.u.set_pminfo.id	= -1,
259 		.u.set_pminfo.type	= XEN_PM_PDC,
260 	};
261 	uint32_t buf[3];
262 	unsigned int ax, bx, cx, dx;
263 	unsigned int mwait_mask;
264 
265 	/* We need to determine whether it is OK to expose the MWAIT
266 	 * capability to the kernel to harvest deeper than C3 states from ACPI
267 	 * _CST using the processor_harvest_xen.c module. For this to work, we
268 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
269 	 * checks against). The hypervisor won't expose the MWAIT flag because
270 	 * it would break backwards compatibility; so we will find out directly
271 	 * from the hardware and hypercall.
272 	 */
273 	if (!xen_initial_domain())
274 		return false;
275 
276 	/*
277 	 * When running under platform earlier than Xen4.2, do not expose
278 	 * mwait, to avoid the risk of loading native acpi pad driver
279 	 */
280 	if (!xen_running_on_version_or_later(4, 2))
281 		return false;
282 
283 	ax = 1;
284 	cx = 0;
285 
286 	native_cpuid(&ax, &bx, &cx, &dx);
287 
288 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
289 		     (1 << (X86_FEATURE_MWAIT % 32));
290 
291 	if ((cx & mwait_mask) != mwait_mask)
292 		return false;
293 
294 	/* We need to emulate the MWAIT_LEAF and for that we need both
295 	 * ecx and edx. The hypercall provides only partial information.
296 	 */
297 
298 	ax = CPUID_MWAIT_LEAF;
299 	bx = 0;
300 	cx = 0;
301 	dx = 0;
302 
303 	native_cpuid(&ax, &bx, &cx, &dx);
304 
305 	/* Ask the Hypervisor whether to clear ACPI_PROC_CAP_C_C2C3_FFH. If so,
306 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
307 	 */
308 	buf[0] = ACPI_PDC_REVISION_ID;
309 	buf[1] = 1;
310 	buf[2] = (ACPI_PROC_CAP_C_CAPABILITY_SMP | ACPI_PROC_CAP_EST_CAPABILITY_SWSMP);
311 
312 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
313 
314 	if ((HYPERVISOR_platform_op(&op) == 0) &&
315 	    (buf[2] & (ACPI_PROC_CAP_C_C1_FFH | ACPI_PROC_CAP_C_C2C3_FFH))) {
316 		cpuid_leaf5_ecx_val = cx;
317 		cpuid_leaf5_edx_val = dx;
318 	}
319 	return true;
320 #else
321 	return false;
322 #endif
323 }
324 
325 static bool __init xen_check_xsave(void)
326 {
327 	unsigned int cx, xsave_mask;
328 
329 	cx = cpuid_ecx(1);
330 
331 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
332 		     (1 << (X86_FEATURE_OSXSAVE % 32));
333 
334 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
335 	return (cx & xsave_mask) == xsave_mask;
336 }
337 
338 static void __init xen_init_capabilities(void)
339 {
340 	setup_force_cpu_cap(X86_FEATURE_XENPV);
341 	setup_clear_cpu_cap(X86_FEATURE_DCA);
342 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
343 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
344 	setup_clear_cpu_cap(X86_FEATURE_ACC);
345 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
346 	setup_clear_cpu_cap(X86_FEATURE_SME);
347 	setup_clear_cpu_cap(X86_FEATURE_LKGS);
348 
349 	/*
350 	 * Xen PV would need some work to support PCID: CR3 handling as well
351 	 * as xen_flush_tlb_others() would need updating.
352 	 */
353 	setup_clear_cpu_cap(X86_FEATURE_PCID);
354 
355 	if (!xen_initial_domain())
356 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
357 
358 	if (xen_check_mwait())
359 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
360 	else
361 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
362 
363 	if (!xen_check_xsave()) {
364 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
365 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
366 	}
367 }
368 
369 static noinstr void xen_set_debugreg(int reg, unsigned long val)
370 {
371 	HYPERVISOR_set_debugreg(reg, val);
372 }
373 
374 static noinstr unsigned long xen_get_debugreg(int reg)
375 {
376 	return HYPERVISOR_get_debugreg(reg);
377 }
378 
379 static void xen_start_context_switch(struct task_struct *prev)
380 {
381 	BUG_ON(preemptible());
382 
383 	if (this_cpu_read(xen_lazy_mode) == XEN_LAZY_MMU) {
384 		arch_leave_lazy_mmu_mode();
385 		set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
386 	}
387 	enter_lazy(XEN_LAZY_CPU);
388 }
389 
390 static void xen_end_context_switch(struct task_struct *next)
391 {
392 	BUG_ON(preemptible());
393 
394 	xen_mc_flush();
395 	leave_lazy(XEN_LAZY_CPU);
396 	if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
397 		arch_enter_lazy_mmu_mode();
398 }
399 
400 static unsigned long xen_store_tr(void)
401 {
402 	return 0;
403 }
404 
405 /*
406  * Set the page permissions for a particular virtual address.  If the
407  * address is a vmalloc mapping (or other non-linear mapping), then
408  * find the linear mapping of the page and also set its protections to
409  * match.
410  */
411 static void set_aliased_prot(void *v, pgprot_t prot)
412 {
413 	int level;
414 	pte_t *ptep;
415 	pte_t pte;
416 	unsigned long pfn;
417 	unsigned char dummy;
418 	void *va;
419 
420 	ptep = lookup_address((unsigned long)v, &level);
421 	BUG_ON(ptep == NULL);
422 
423 	pfn = pte_pfn(*ptep);
424 	pte = pfn_pte(pfn, prot);
425 
426 	/*
427 	 * Careful: update_va_mapping() will fail if the virtual address
428 	 * we're poking isn't populated in the page tables.  We don't
429 	 * need to worry about the direct map (that's always in the page
430 	 * tables), but we need to be careful about vmap space.  In
431 	 * particular, the top level page table can lazily propagate
432 	 * entries between processes, so if we've switched mms since we
433 	 * vmapped the target in the first place, we might not have the
434 	 * top-level page table entry populated.
435 	 *
436 	 * We disable preemption because we want the same mm active when
437 	 * we probe the target and when we issue the hypercall.  We'll
438 	 * have the same nominal mm, but if we're a kernel thread, lazy
439 	 * mm dropping could change our pgd.
440 	 *
441 	 * Out of an abundance of caution, this uses __get_user() to fault
442 	 * in the target address just in case there's some obscure case
443 	 * in which the target address isn't readable.
444 	 */
445 
446 	preempt_disable();
447 
448 	copy_from_kernel_nofault(&dummy, v, 1);
449 
450 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
451 		BUG();
452 
453 	va = __va(PFN_PHYS(pfn));
454 
455 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
456 		BUG();
457 
458 	preempt_enable();
459 }
460 
461 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
462 {
463 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
464 	int i;
465 
466 	/*
467 	 * We need to mark the all aliases of the LDT pages RO.  We
468 	 * don't need to call vm_flush_aliases(), though, since that's
469 	 * only responsible for flushing aliases out the TLBs, not the
470 	 * page tables, and Xen will flush the TLB for us if needed.
471 	 *
472 	 * To avoid confusing future readers: none of this is necessary
473 	 * to load the LDT.  The hypervisor only checks this when the
474 	 * LDT is faulted in due to subsequent descriptor access.
475 	 */
476 
477 	for (i = 0; i < entries; i += entries_per_page)
478 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
479 }
480 
481 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
482 {
483 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
484 	int i;
485 
486 	for (i = 0; i < entries; i += entries_per_page)
487 		set_aliased_prot(ldt + i, PAGE_KERNEL);
488 }
489 
490 static void xen_set_ldt(const void *addr, unsigned entries)
491 {
492 	struct mmuext_op *op;
493 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
494 
495 	trace_xen_cpu_set_ldt(addr, entries);
496 
497 	op = mcs.args;
498 	op->cmd = MMUEXT_SET_LDT;
499 	op->arg1.linear_addr = (unsigned long)addr;
500 	op->arg2.nr_ents = entries;
501 
502 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
503 
504 	xen_mc_issue(XEN_LAZY_CPU);
505 }
506 
507 static void xen_load_gdt(const struct desc_ptr *dtr)
508 {
509 	unsigned long va = dtr->address;
510 	unsigned int size = dtr->size + 1;
511 	unsigned long pfn, mfn;
512 	int level;
513 	pte_t *ptep;
514 	void *virt;
515 
516 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
517 	BUG_ON(size > PAGE_SIZE);
518 	BUG_ON(va & ~PAGE_MASK);
519 
520 	/*
521 	 * The GDT is per-cpu and is in the percpu data area.
522 	 * That can be virtually mapped, so we need to do a
523 	 * page-walk to get the underlying MFN for the
524 	 * hypercall.  The page can also be in the kernel's
525 	 * linear range, so we need to RO that mapping too.
526 	 */
527 	ptep = lookup_address(va, &level);
528 	BUG_ON(ptep == NULL);
529 
530 	pfn = pte_pfn(*ptep);
531 	mfn = pfn_to_mfn(pfn);
532 	virt = __va(PFN_PHYS(pfn));
533 
534 	make_lowmem_page_readonly((void *)va);
535 	make_lowmem_page_readonly(virt);
536 
537 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
538 		BUG();
539 }
540 
541 /*
542  * load_gdt for early boot, when the gdt is only mapped once
543  */
544 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
545 {
546 	unsigned long va = dtr->address;
547 	unsigned int size = dtr->size + 1;
548 	unsigned long pfn, mfn;
549 	pte_t pte;
550 
551 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
552 	BUG_ON(size > PAGE_SIZE);
553 	BUG_ON(va & ~PAGE_MASK);
554 
555 	pfn = virt_to_pfn((void *)va);
556 	mfn = pfn_to_mfn(pfn);
557 
558 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
559 
560 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
561 		BUG();
562 
563 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
564 		BUG();
565 }
566 
567 static inline bool desc_equal(const struct desc_struct *d1,
568 			      const struct desc_struct *d2)
569 {
570 	return !memcmp(d1, d2, sizeof(*d1));
571 }
572 
573 static void load_TLS_descriptor(struct thread_struct *t,
574 				unsigned int cpu, unsigned int i)
575 {
576 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
577 	struct desc_struct *gdt;
578 	xmaddr_t maddr;
579 	struct multicall_space mc;
580 
581 	if (desc_equal(shadow, &t->tls_array[i]))
582 		return;
583 
584 	*shadow = t->tls_array[i];
585 
586 	gdt = get_cpu_gdt_rw(cpu);
587 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
588 	mc = __xen_mc_entry(0);
589 
590 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
591 }
592 
593 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
594 {
595 	/*
596 	 * In lazy mode we need to zero %fs, otherwise we may get an
597 	 * exception between the new %fs descriptor being loaded and
598 	 * %fs being effectively cleared at __switch_to().
599 	 */
600 	if (xen_get_lazy_mode() == XEN_LAZY_CPU)
601 		loadsegment(fs, 0);
602 
603 	xen_mc_batch();
604 
605 	load_TLS_descriptor(t, cpu, 0);
606 	load_TLS_descriptor(t, cpu, 1);
607 	load_TLS_descriptor(t, cpu, 2);
608 
609 	xen_mc_issue(XEN_LAZY_CPU);
610 }
611 
612 static void xen_load_gs_index(unsigned int idx)
613 {
614 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
615 		BUG();
616 }
617 
618 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
619 				const void *ptr)
620 {
621 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
622 	u64 entry = *(u64 *)ptr;
623 
624 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
625 
626 	preempt_disable();
627 
628 	xen_mc_flush();
629 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
630 		BUG();
631 
632 	preempt_enable();
633 }
634 
635 void noist_exc_debug(struct pt_regs *regs);
636 
637 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
638 {
639 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
640 	exc_nmi(regs);
641 }
642 
643 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
644 {
645 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
646 	exc_double_fault(regs, error_code);
647 }
648 
649 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
650 {
651 	/*
652 	 * There's no IST on Xen PV, but we still need to dispatch
653 	 * to the correct handler.
654 	 */
655 	if (user_mode(regs))
656 		noist_exc_debug(regs);
657 	else
658 		exc_debug(regs);
659 }
660 
661 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
662 {
663 	/* This should never happen and there is no way to handle it. */
664 	instrumentation_begin();
665 	pr_err("Unknown trap in Xen PV mode.");
666 	BUG();
667 	instrumentation_end();
668 }
669 
670 #ifdef CONFIG_X86_MCE
671 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
672 {
673 	/*
674 	 * There's no IST on Xen PV, but we still need to dispatch
675 	 * to the correct handler.
676 	 */
677 	if (user_mode(regs))
678 		noist_exc_machine_check(regs);
679 	else
680 		exc_machine_check(regs);
681 }
682 #endif
683 
684 struct trap_array_entry {
685 	void (*orig)(void);
686 	void (*xen)(void);
687 	bool ist_okay;
688 };
689 
690 #define TRAP_ENTRY(func, ist_ok) {			\
691 	.orig		= asm_##func,			\
692 	.xen		= xen_asm_##func,		\
693 	.ist_okay	= ist_ok }
694 
695 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
696 	.orig		= asm_##func,			\
697 	.xen		= xen_asm_xenpv_##func,		\
698 	.ist_okay	= ist_ok }
699 
700 static struct trap_array_entry trap_array[] = {
701 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
702 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
703 #ifdef CONFIG_X86_MCE
704 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
705 #endif
706 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
707 	TRAP_ENTRY(exc_int3,				false ),
708 	TRAP_ENTRY(exc_overflow,			false ),
709 #ifdef CONFIG_IA32_EMULATION
710 	TRAP_ENTRY(int80_emulation,			false ),
711 #endif
712 	TRAP_ENTRY(exc_page_fault,			false ),
713 	TRAP_ENTRY(exc_divide_error,			false ),
714 	TRAP_ENTRY(exc_bounds,				false ),
715 	TRAP_ENTRY(exc_invalid_op,			false ),
716 	TRAP_ENTRY(exc_device_not_available,		false ),
717 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
718 	TRAP_ENTRY(exc_invalid_tss,			false ),
719 	TRAP_ENTRY(exc_segment_not_present,		false ),
720 	TRAP_ENTRY(exc_stack_segment,			false ),
721 	TRAP_ENTRY(exc_general_protection,		false ),
722 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
723 	TRAP_ENTRY(exc_coprocessor_error,		false ),
724 	TRAP_ENTRY(exc_alignment_check,			false ),
725 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
726 #ifdef CONFIG_X86_CET
727 	TRAP_ENTRY(exc_control_protection,		false ),
728 #endif
729 };
730 
731 static bool __ref get_trap_addr(void **addr, unsigned int ist)
732 {
733 	unsigned int nr;
734 	bool ist_okay = false;
735 	bool found = false;
736 
737 	/*
738 	 * Replace trap handler addresses by Xen specific ones.
739 	 * Check for known traps using IST and whitelist them.
740 	 * The debugger ones are the only ones we care about.
741 	 * Xen will handle faults like double_fault, so we should never see
742 	 * them.  Warn if there's an unexpected IST-using fault handler.
743 	 */
744 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
745 		struct trap_array_entry *entry = trap_array + nr;
746 
747 		if (*addr == entry->orig) {
748 			*addr = entry->xen;
749 			ist_okay = entry->ist_okay;
750 			found = true;
751 			break;
752 		}
753 	}
754 
755 	if (nr == ARRAY_SIZE(trap_array) &&
756 	    *addr >= (void *)early_idt_handler_array[0] &&
757 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
758 		nr = (*addr - (void *)early_idt_handler_array[0]) /
759 		     EARLY_IDT_HANDLER_SIZE;
760 		*addr = (void *)xen_early_idt_handler_array[nr];
761 		found = true;
762 	}
763 
764 	if (!found)
765 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
766 
767 	if (WARN_ON(found && ist != 0 && !ist_okay))
768 		return false;
769 
770 	return true;
771 }
772 
773 static int cvt_gate_to_trap(int vector, const gate_desc *val,
774 			    struct trap_info *info)
775 {
776 	unsigned long addr;
777 
778 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
779 		return 0;
780 
781 	info->vector = vector;
782 
783 	addr = gate_offset(val);
784 	if (!get_trap_addr((void **)&addr, val->bits.ist))
785 		return 0;
786 	info->address = addr;
787 
788 	info->cs = gate_segment(val);
789 	info->flags = val->bits.dpl;
790 	/* interrupt gates clear IF */
791 	if (val->bits.type == GATE_INTERRUPT)
792 		info->flags |= 1 << 2;
793 
794 	return 1;
795 }
796 
797 /* Locations of each CPU's IDT */
798 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
799 
800 /* Set an IDT entry.  If the entry is part of the current IDT, then
801    also update Xen. */
802 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
803 {
804 	unsigned long p = (unsigned long)&dt[entrynum];
805 	unsigned long start, end;
806 
807 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
808 
809 	preempt_disable();
810 
811 	start = __this_cpu_read(idt_desc.address);
812 	end = start + __this_cpu_read(idt_desc.size) + 1;
813 
814 	xen_mc_flush();
815 
816 	native_write_idt_entry(dt, entrynum, g);
817 
818 	if (p >= start && (p + 8) <= end) {
819 		struct trap_info info[2];
820 
821 		info[1].address = 0;
822 
823 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
824 			if (HYPERVISOR_set_trap_table(info))
825 				BUG();
826 	}
827 
828 	preempt_enable();
829 }
830 
831 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
832 				      struct trap_info *traps, bool full)
833 {
834 	unsigned in, out, count;
835 
836 	count = (desc->size+1) / sizeof(gate_desc);
837 	BUG_ON(count > 256);
838 
839 	for (in = out = 0; in < count; in++) {
840 		gate_desc *entry = (gate_desc *)(desc->address) + in;
841 
842 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
843 			out++;
844 	}
845 
846 	return out;
847 }
848 
849 void xen_copy_trap_info(struct trap_info *traps)
850 {
851 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
852 
853 	xen_convert_trap_info(desc, traps, true);
854 }
855 
856 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
857    hold a spinlock to protect the static traps[] array (static because
858    it avoids allocation, and saves stack space). */
859 static void xen_load_idt(const struct desc_ptr *desc)
860 {
861 	static DEFINE_SPINLOCK(lock);
862 	static struct trap_info traps[257];
863 	static const struct trap_info zero = { };
864 	unsigned out;
865 
866 	trace_xen_cpu_load_idt(desc);
867 
868 	spin_lock(&lock);
869 
870 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
871 
872 	out = xen_convert_trap_info(desc, traps, false);
873 	traps[out] = zero;
874 
875 	xen_mc_flush();
876 	if (HYPERVISOR_set_trap_table(traps))
877 		BUG();
878 
879 	spin_unlock(&lock);
880 }
881 
882 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
883    they're handled differently. */
884 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
885 				const void *desc, int type)
886 {
887 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
888 
889 	preempt_disable();
890 
891 	switch (type) {
892 	case DESC_LDT:
893 	case DESC_TSS:
894 		/* ignore */
895 		break;
896 
897 	default: {
898 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
899 
900 		xen_mc_flush();
901 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
902 			BUG();
903 	}
904 
905 	}
906 
907 	preempt_enable();
908 }
909 
910 /*
911  * Version of write_gdt_entry for use at early boot-time needed to
912  * update an entry as simply as possible.
913  */
914 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
915 					    const void *desc, int type)
916 {
917 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
918 
919 	switch (type) {
920 	case DESC_LDT:
921 	case DESC_TSS:
922 		/* ignore */
923 		break;
924 
925 	default: {
926 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
927 
928 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
929 			dt[entry] = *(struct desc_struct *)desc;
930 	}
931 
932 	}
933 }
934 
935 static void xen_load_sp0(unsigned long sp0)
936 {
937 	struct multicall_space mcs;
938 
939 	mcs = xen_mc_entry(0);
940 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
941 	xen_mc_issue(XEN_LAZY_CPU);
942 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
943 }
944 
945 #ifdef CONFIG_X86_IOPL_IOPERM
946 static void xen_invalidate_io_bitmap(void)
947 {
948 	struct physdev_set_iobitmap iobitmap = {
949 		.bitmap = NULL,
950 		.nr_ports = 0,
951 	};
952 
953 	native_tss_invalidate_io_bitmap();
954 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
955 }
956 
957 static void xen_update_io_bitmap(void)
958 {
959 	struct physdev_set_iobitmap iobitmap;
960 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
961 
962 	native_tss_update_io_bitmap();
963 
964 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
965 			  tss->x86_tss.io_bitmap_base;
966 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
967 		iobitmap.nr_ports = 0;
968 	else
969 		iobitmap.nr_ports = IO_BITMAP_BITS;
970 
971 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
972 }
973 #endif
974 
975 static void xen_io_delay(void)
976 {
977 }
978 
979 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
980 
981 static unsigned long xen_read_cr0(void)
982 {
983 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
984 
985 	if (unlikely(cr0 == 0)) {
986 		cr0 = native_read_cr0();
987 		this_cpu_write(xen_cr0_value, cr0);
988 	}
989 
990 	return cr0;
991 }
992 
993 static void xen_write_cr0(unsigned long cr0)
994 {
995 	struct multicall_space mcs;
996 
997 	this_cpu_write(xen_cr0_value, cr0);
998 
999 	/* Only pay attention to cr0.TS; everything else is
1000 	   ignored. */
1001 	mcs = xen_mc_entry(0);
1002 
1003 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1004 
1005 	xen_mc_issue(XEN_LAZY_CPU);
1006 }
1007 
1008 static void xen_write_cr4(unsigned long cr4)
1009 {
1010 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1011 
1012 	native_write_cr4(cr4);
1013 }
1014 
1015 static u64 xen_do_read_msr(unsigned int msr, int *err)
1016 {
1017 	u64 val = 0;	/* Avoid uninitialized value for safe variant. */
1018 
1019 	if (pmu_msr_read(msr, &val, err))
1020 		return val;
1021 
1022 	if (err)
1023 		val = native_read_msr_safe(msr, err);
1024 	else
1025 		val = native_read_msr(msr);
1026 
1027 	switch (msr) {
1028 	case MSR_IA32_APICBASE:
1029 		val &= ~X2APIC_ENABLE;
1030 		break;
1031 	}
1032 	return val;
1033 }
1034 
1035 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
1036 		    int *err)
1037 {
1038 	u64 base = ((u64)high << 32) | low;
1039 
1040 	if (HYPERVISOR_set_segment_base(which, base) == 0)
1041 		return;
1042 
1043 	if (err)
1044 		*err = -EIO;
1045 	else
1046 		WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
1047 }
1048 
1049 /*
1050  * Support write_msr_safe() and write_msr() semantics.
1051  * With err == NULL write_msr() semantics are selected.
1052  * Supplying an err pointer requires err to be pre-initialized with 0.
1053  */
1054 static void xen_do_write_msr(unsigned int msr, unsigned int low,
1055 			     unsigned int high, int *err)
1056 {
1057 	switch (msr) {
1058 	case MSR_FS_BASE:
1059 		set_seg(SEGBASE_FS, low, high, err);
1060 		break;
1061 
1062 	case MSR_KERNEL_GS_BASE:
1063 		set_seg(SEGBASE_GS_USER, low, high, err);
1064 		break;
1065 
1066 	case MSR_GS_BASE:
1067 		set_seg(SEGBASE_GS_KERNEL, low, high, err);
1068 		break;
1069 
1070 	case MSR_STAR:
1071 	case MSR_CSTAR:
1072 	case MSR_LSTAR:
1073 	case MSR_SYSCALL_MASK:
1074 	case MSR_IA32_SYSENTER_CS:
1075 	case MSR_IA32_SYSENTER_ESP:
1076 	case MSR_IA32_SYSENTER_EIP:
1077 		/* Fast syscall setup is all done in hypercalls, so
1078 		   these are all ignored.  Stub them out here to stop
1079 		   Xen console noise. */
1080 		break;
1081 
1082 	default:
1083 		if (!pmu_msr_write(msr, low, high, err)) {
1084 			if (err)
1085 				*err = native_write_msr_safe(msr, low, high);
1086 			else
1087 				native_write_msr(msr, low, high);
1088 		}
1089 	}
1090 }
1091 
1092 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1093 {
1094 	return xen_do_read_msr(msr, err);
1095 }
1096 
1097 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1098 			      unsigned int high)
1099 {
1100 	int err = 0;
1101 
1102 	xen_do_write_msr(msr, low, high, &err);
1103 
1104 	return err;
1105 }
1106 
1107 static u64 xen_read_msr(unsigned int msr)
1108 {
1109 	int err;
1110 
1111 	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1112 }
1113 
1114 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1115 {
1116 	int err;
1117 
1118 	xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1119 }
1120 
1121 /* This is called once we have the cpu_possible_mask */
1122 void __init xen_setup_vcpu_info_placement(void)
1123 {
1124 	int cpu;
1125 
1126 	for_each_possible_cpu(cpu) {
1127 		/* Set up direct vCPU id mapping for PV guests. */
1128 		per_cpu(xen_vcpu_id, cpu) = cpu;
1129 		xen_vcpu_setup(cpu);
1130 	}
1131 
1132 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1133 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1134 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1135 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1136 }
1137 
1138 static const struct pv_info xen_info __initconst = {
1139 	.extra_user_64bit_cs = FLAT_USER_CS64,
1140 	.name = "Xen",
1141 };
1142 
1143 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1144 	.cpu = {
1145 		.cpuid = xen_cpuid,
1146 
1147 		.set_debugreg = xen_set_debugreg,
1148 		.get_debugreg = xen_get_debugreg,
1149 
1150 		.read_cr0 = xen_read_cr0,
1151 		.write_cr0 = xen_write_cr0,
1152 
1153 		.write_cr4 = xen_write_cr4,
1154 
1155 		.wbinvd = pv_native_wbinvd,
1156 
1157 		.read_msr = xen_read_msr,
1158 		.write_msr = xen_write_msr,
1159 
1160 		.read_msr_safe = xen_read_msr_safe,
1161 		.write_msr_safe = xen_write_msr_safe,
1162 
1163 		.read_pmc = xen_read_pmc,
1164 
1165 		.load_tr_desc = paravirt_nop,
1166 		.set_ldt = xen_set_ldt,
1167 		.load_gdt = xen_load_gdt,
1168 		.load_idt = xen_load_idt,
1169 		.load_tls = xen_load_tls,
1170 		.load_gs_index = xen_load_gs_index,
1171 
1172 		.alloc_ldt = xen_alloc_ldt,
1173 		.free_ldt = xen_free_ldt,
1174 
1175 		.store_tr = xen_store_tr,
1176 
1177 		.write_ldt_entry = xen_write_ldt_entry,
1178 		.write_gdt_entry = xen_write_gdt_entry,
1179 		.write_idt_entry = xen_write_idt_entry,
1180 		.load_sp0 = xen_load_sp0,
1181 
1182 #ifdef CONFIG_X86_IOPL_IOPERM
1183 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1184 		.update_io_bitmap = xen_update_io_bitmap,
1185 #endif
1186 		.io_delay = xen_io_delay,
1187 
1188 		.start_context_switch = xen_start_context_switch,
1189 		.end_context_switch = xen_end_context_switch,
1190 	},
1191 };
1192 
1193 static void xen_restart(char *msg)
1194 {
1195 	xen_reboot(SHUTDOWN_reboot);
1196 }
1197 
1198 static void xen_machine_halt(void)
1199 {
1200 	xen_reboot(SHUTDOWN_poweroff);
1201 }
1202 
1203 static void xen_machine_power_off(void)
1204 {
1205 	do_kernel_power_off();
1206 	xen_reboot(SHUTDOWN_poweroff);
1207 }
1208 
1209 static void xen_crash_shutdown(struct pt_regs *regs)
1210 {
1211 	xen_reboot(SHUTDOWN_crash);
1212 }
1213 
1214 static const struct machine_ops xen_machine_ops __initconst = {
1215 	.restart = xen_restart,
1216 	.halt = xen_machine_halt,
1217 	.power_off = xen_machine_power_off,
1218 	.shutdown = xen_machine_halt,
1219 	.crash_shutdown = xen_crash_shutdown,
1220 	.emergency_restart = xen_emergency_restart,
1221 };
1222 
1223 static unsigned char xen_get_nmi_reason(void)
1224 {
1225 	unsigned char reason = 0;
1226 
1227 	/* Construct a value which looks like it came from port 0x61. */
1228 	if (test_bit(_XEN_NMIREASON_io_error,
1229 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1230 		reason |= NMI_REASON_IOCHK;
1231 	if (test_bit(_XEN_NMIREASON_pci_serr,
1232 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1233 		reason |= NMI_REASON_SERR;
1234 
1235 	return reason;
1236 }
1237 
1238 static void __init xen_boot_params_init_edd(void)
1239 {
1240 #if IS_ENABLED(CONFIG_EDD)
1241 	struct xen_platform_op op;
1242 	struct edd_info *edd_info;
1243 	u32 *mbr_signature;
1244 	unsigned nr;
1245 	int ret;
1246 
1247 	edd_info = boot_params.eddbuf;
1248 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1249 
1250 	op.cmd = XENPF_firmware_info;
1251 
1252 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1253 	for (nr = 0; nr < EDDMAXNR; nr++) {
1254 		struct edd_info *info = edd_info + nr;
1255 
1256 		op.u.firmware_info.index = nr;
1257 		info->params.length = sizeof(info->params);
1258 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1259 				     &info->params);
1260 		ret = HYPERVISOR_platform_op(&op);
1261 		if (ret)
1262 			break;
1263 
1264 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1265 		C(device);
1266 		C(version);
1267 		C(interface_support);
1268 		C(legacy_max_cylinder);
1269 		C(legacy_max_head);
1270 		C(legacy_sectors_per_track);
1271 #undef C
1272 	}
1273 	boot_params.eddbuf_entries = nr;
1274 
1275 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1276 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1277 		op.u.firmware_info.index = nr;
1278 		ret = HYPERVISOR_platform_op(&op);
1279 		if (ret)
1280 			break;
1281 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1282 	}
1283 	boot_params.edd_mbr_sig_buf_entries = nr;
1284 #endif
1285 }
1286 
1287 /*
1288  * Set up the GDT and segment registers for -fstack-protector.  Until
1289  * we do this, we have to be careful not to call any stack-protected
1290  * function, which is most of the kernel.
1291  */
1292 static void __init xen_setup_gdt(int cpu)
1293 {
1294 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1295 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1296 
1297 	switch_gdt_and_percpu_base(cpu);
1298 
1299 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1300 	pv_ops.cpu.load_gdt = xen_load_gdt;
1301 }
1302 
1303 static void __init xen_dom0_set_legacy_features(void)
1304 {
1305 	x86_platform.legacy.rtc = 1;
1306 }
1307 
1308 static void __init xen_domu_set_legacy_features(void)
1309 {
1310 	x86_platform.legacy.rtc = 0;
1311 }
1312 
1313 extern void early_xen_iret_patch(void);
1314 
1315 /* First C function to be called on Xen boot */
1316 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1317 {
1318 	struct physdev_set_iopl set_iopl;
1319 	unsigned long initrd_start = 0;
1320 	int rc;
1321 
1322 	if (!si)
1323 		return;
1324 
1325 	clear_bss();
1326 
1327 	xen_start_info = si;
1328 
1329 	__text_gen_insn(&early_xen_iret_patch,
1330 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1331 			JMP32_INSN_SIZE);
1332 
1333 	xen_domain_type = XEN_PV_DOMAIN;
1334 	xen_start_flags = xen_start_info->flags;
1335 
1336 	xen_setup_features();
1337 
1338 	/* Install Xen paravirt ops */
1339 	pv_info = xen_info;
1340 	pv_ops.cpu = xen_cpu_ops.cpu;
1341 	xen_init_irq_ops();
1342 
1343 	/*
1344 	 * Setup xen_vcpu early because it is needed for
1345 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1346 	 *
1347 	 * Don't do the full vcpu_info placement stuff until we have
1348 	 * the cpu_possible_mask and a non-dummy shared_info.
1349 	 */
1350 	xen_vcpu_info_reset(0);
1351 
1352 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1353 	x86_platform.realmode_reserve = x86_init_noop;
1354 	x86_platform.realmode_init = x86_init_noop;
1355 
1356 	x86_init.resources.memory_setup = xen_memory_setup;
1357 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1358 	x86_init.irqs.intr_mode_init	= x86_64_probe_apic;
1359 	x86_init.oem.arch_setup = xen_arch_setup;
1360 	x86_init.oem.banner = xen_banner;
1361 	x86_init.hyper.init_platform = xen_pv_init_platform;
1362 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1363 
1364 	/*
1365 	 * Set up some pagetable state before starting to set any ptes.
1366 	 */
1367 
1368 	xen_setup_machphys_mapping();
1369 	xen_init_mmu_ops();
1370 
1371 	/* Prevent unwanted bits from being set in PTEs. */
1372 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1373 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1374 
1375 	/* Get mfn list */
1376 	xen_build_dynamic_phys_to_machine();
1377 
1378 	/* Work out if we support NX */
1379 	get_cpu_cap(&boot_cpu_data);
1380 	x86_configure_nx();
1381 
1382 	/*
1383 	 * Set up kernel GDT and segment registers, mainly so that
1384 	 * -fstack-protector code can be executed.
1385 	 */
1386 	xen_setup_gdt(0);
1387 
1388 	/* Determine virtual and physical address sizes */
1389 	get_cpu_address_sizes(&boot_cpu_data);
1390 
1391 	/* Let's presume PV guests always boot on vCPU with id 0. */
1392 	per_cpu(xen_vcpu_id, 0) = 0;
1393 
1394 	idt_setup_early_handler();
1395 
1396 	xen_init_capabilities();
1397 
1398 	/*
1399 	 * set up the basic apic ops.
1400 	 */
1401 	xen_init_apic();
1402 
1403 	machine_ops = xen_machine_ops;
1404 
1405 	/*
1406 	 * The only reliable way to retain the initial address of the
1407 	 * percpu gdt_page is to remember it here, so we can go and
1408 	 * mark it RW later, when the initial percpu area is freed.
1409 	 */
1410 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1411 
1412 	xen_smp_init();
1413 
1414 #ifdef CONFIG_ACPI_NUMA
1415 	/*
1416 	 * The pages we from Xen are not related to machine pages, so
1417 	 * any NUMA information the kernel tries to get from ACPI will
1418 	 * be meaningless.  Prevent it from trying.
1419 	 */
1420 	disable_srat();
1421 #endif
1422 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1423 
1424 	local_irq_disable();
1425 	early_boot_irqs_disabled = true;
1426 
1427 	xen_raw_console_write("mapping kernel into physical memory\n");
1428 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1429 				   xen_start_info->nr_pages);
1430 	xen_reserve_special_pages();
1431 
1432 	/*
1433 	 * We used to do this in xen_arch_setup, but that is too late
1434 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1435 	 * early_amd_init which pokes 0xcf8 port.
1436 	 */
1437 	set_iopl.iopl = 1;
1438 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1439 	if (rc != 0)
1440 		xen_raw_printk("physdev_op failed %d\n", rc);
1441 
1442 
1443 	if (xen_start_info->mod_start) {
1444 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1445 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1446 	    else
1447 		initrd_start = __pa(xen_start_info->mod_start);
1448 	}
1449 
1450 	/* Poke various useful things into boot_params */
1451 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1452 	boot_params.hdr.ramdisk_image = initrd_start;
1453 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1454 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1455 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1456 
1457 	if (!xen_initial_domain()) {
1458 		if (pci_xen)
1459 			x86_init.pci.arch_init = pci_xen_init;
1460 		x86_platform.set_legacy_features =
1461 				xen_domu_set_legacy_features;
1462 	} else {
1463 		const struct dom0_vga_console_info *info =
1464 			(void *)((char *)xen_start_info +
1465 				 xen_start_info->console.dom0.info_off);
1466 		struct xen_platform_op op = {
1467 			.cmd = XENPF_firmware_info,
1468 			.interface_version = XENPF_INTERFACE_VERSION,
1469 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1470 		};
1471 
1472 		x86_platform.set_legacy_features =
1473 				xen_dom0_set_legacy_features;
1474 		xen_init_vga(info, xen_start_info->console.dom0.info_size,
1475 			     &boot_params.screen_info);
1476 		xen_start_info->console.domU.mfn = 0;
1477 		xen_start_info->console.domU.evtchn = 0;
1478 
1479 		if (HYPERVISOR_platform_op(&op) == 0)
1480 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1481 
1482 		/* Make sure ACS will be enabled */
1483 		pci_request_acs();
1484 
1485 		xen_acpi_sleep_register();
1486 
1487 		xen_boot_params_init_edd();
1488 
1489 #ifdef CONFIG_ACPI
1490 		/*
1491 		 * Disable selecting "Firmware First mode" for correctable
1492 		 * memory errors, as this is the duty of the hypervisor to
1493 		 * decide.
1494 		 */
1495 		acpi_disable_cmcff = 1;
1496 #endif
1497 	}
1498 
1499 	xen_add_preferred_consoles();
1500 
1501 #ifdef CONFIG_PCI
1502 	/* PCI BIOS service won't work from a PV guest. */
1503 	pci_probe &= ~PCI_PROBE_BIOS;
1504 #endif
1505 	xen_raw_console_write("about to get started...\n");
1506 
1507 	/* We need this for printk timestamps */
1508 	xen_setup_runstate_info(0);
1509 
1510 	xen_efi_init(&boot_params);
1511 
1512 	/* Start the world */
1513 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1514 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1515 }
1516 
1517 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1518 {
1519 	int rc;
1520 
1521 	if (per_cpu(xen_vcpu, cpu) == NULL)
1522 		return -ENODEV;
1523 
1524 	xen_setup_timer(cpu);
1525 
1526 	rc = xen_smp_intr_init(cpu);
1527 	if (rc) {
1528 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1529 		     cpu, rc);
1530 		return rc;
1531 	}
1532 
1533 	rc = xen_smp_intr_init_pv(cpu);
1534 	if (rc) {
1535 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1536 		     cpu, rc);
1537 		return rc;
1538 	}
1539 
1540 	return 0;
1541 }
1542 
1543 static int xen_cpu_dead_pv(unsigned int cpu)
1544 {
1545 	xen_smp_intr_free(cpu);
1546 	xen_smp_intr_free_pv(cpu);
1547 
1548 	xen_teardown_timer(cpu);
1549 
1550 	return 0;
1551 }
1552 
1553 static uint32_t __init xen_platform_pv(void)
1554 {
1555 	if (xen_pv_domain())
1556 		return xen_cpuid_base();
1557 
1558 	return 0;
1559 }
1560 
1561 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1562 	.name                   = "Xen PV",
1563 	.detect                 = xen_platform_pv,
1564 	.type			= X86_HYPER_XEN_PV,
1565 	.runtime.pin_vcpu       = xen_pin_vcpu,
1566 	.ignore_nopv		= true,
1567 };
1568