xref: /linux/arch/x86/xen/enlighten_pv.c (revision fada1935590f66dc6784981e0d557ca09013c847)
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/bootmem.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/memblock.h>
35 #include <linux/edd.h>
36 #include <linux/frame.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/pgtable.h>
68 #include <asm/tlbflush.h>
69 #include <asm/reboot.h>
70 #include <asm/stackprotector.h>
71 #include <asm/hypervisor.h>
72 #include <asm/mach_traps.h>
73 #include <asm/mwait.h>
74 #include <asm/pci_x86.h>
75 #include <asm/cpu.h>
76 
77 #ifdef CONFIG_ACPI
78 #include <linux/acpi.h>
79 #include <asm/acpi.h>
80 #include <acpi/pdc_intel.h>
81 #include <acpi/processor.h>
82 #include <xen/interface/platform.h>
83 #endif
84 
85 #include "xen-ops.h"
86 #include "mmu.h"
87 #include "smp.h"
88 #include "multicalls.h"
89 #include "pmu.h"
90 
91 void *xen_initial_gdt;
92 
93 static int xen_cpu_up_prepare_pv(unsigned int cpu);
94 static int xen_cpu_dead_pv(unsigned int cpu);
95 
96 struct tls_descs {
97 	struct desc_struct desc[3];
98 };
99 
100 /*
101  * Updating the 3 TLS descriptors in the GDT on every task switch is
102  * surprisingly expensive so we avoid updating them if they haven't
103  * changed.  Since Xen writes different descriptors than the one
104  * passed in the update_descriptor hypercall we keep shadow copies to
105  * compare against.
106  */
107 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
108 
109 static void __init xen_banner(void)
110 {
111 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
112 	struct xen_extraversion extra;
113 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
114 
115 	pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
116 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
117 	       version >> 16, version & 0xffff, extra.extraversion,
118 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
119 }
120 /* Check if running on Xen version (major, minor) or later */
121 bool
122 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
123 {
124 	unsigned int version;
125 
126 	if (!xen_domain())
127 		return false;
128 
129 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
130 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
131 		((version >> 16) > major))
132 		return true;
133 	return false;
134 }
135 
136 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
137 static __read_mostly unsigned int cpuid_leaf5_edx_val;
138 
139 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
140 		      unsigned int *cx, unsigned int *dx)
141 {
142 	unsigned maskebx = ~0;
143 
144 	/*
145 	 * Mask out inconvenient features, to try and disable as many
146 	 * unsupported kernel subsystems as possible.
147 	 */
148 	switch (*ax) {
149 	case CPUID_MWAIT_LEAF:
150 		/* Synthesize the values.. */
151 		*ax = 0;
152 		*bx = 0;
153 		*cx = cpuid_leaf5_ecx_val;
154 		*dx = cpuid_leaf5_edx_val;
155 		return;
156 
157 	case 0xb:
158 		/* Suppress extended topology stuff */
159 		maskebx = 0;
160 		break;
161 	}
162 
163 	asm(XEN_EMULATE_PREFIX "cpuid"
164 		: "=a" (*ax),
165 		  "=b" (*bx),
166 		  "=c" (*cx),
167 		  "=d" (*dx)
168 		: "0" (*ax), "2" (*cx));
169 
170 	*bx &= maskebx;
171 }
172 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
173 
174 static bool __init xen_check_mwait(void)
175 {
176 #ifdef CONFIG_ACPI
177 	struct xen_platform_op op = {
178 		.cmd			= XENPF_set_processor_pminfo,
179 		.u.set_pminfo.id	= -1,
180 		.u.set_pminfo.type	= XEN_PM_PDC,
181 	};
182 	uint32_t buf[3];
183 	unsigned int ax, bx, cx, dx;
184 	unsigned int mwait_mask;
185 
186 	/* We need to determine whether it is OK to expose the MWAIT
187 	 * capability to the kernel to harvest deeper than C3 states from ACPI
188 	 * _CST using the processor_harvest_xen.c module. For this to work, we
189 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
190 	 * checks against). The hypervisor won't expose the MWAIT flag because
191 	 * it would break backwards compatibility; so we will find out directly
192 	 * from the hardware and hypercall.
193 	 */
194 	if (!xen_initial_domain())
195 		return false;
196 
197 	/*
198 	 * When running under platform earlier than Xen4.2, do not expose
199 	 * mwait, to avoid the risk of loading native acpi pad driver
200 	 */
201 	if (!xen_running_on_version_or_later(4, 2))
202 		return false;
203 
204 	ax = 1;
205 	cx = 0;
206 
207 	native_cpuid(&ax, &bx, &cx, &dx);
208 
209 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
210 		     (1 << (X86_FEATURE_MWAIT % 32));
211 
212 	if ((cx & mwait_mask) != mwait_mask)
213 		return false;
214 
215 	/* We need to emulate the MWAIT_LEAF and for that we need both
216 	 * ecx and edx. The hypercall provides only partial information.
217 	 */
218 
219 	ax = CPUID_MWAIT_LEAF;
220 	bx = 0;
221 	cx = 0;
222 	dx = 0;
223 
224 	native_cpuid(&ax, &bx, &cx, &dx);
225 
226 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
227 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
228 	 */
229 	buf[0] = ACPI_PDC_REVISION_ID;
230 	buf[1] = 1;
231 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
232 
233 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
234 
235 	if ((HYPERVISOR_platform_op(&op) == 0) &&
236 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
237 		cpuid_leaf5_ecx_val = cx;
238 		cpuid_leaf5_edx_val = dx;
239 	}
240 	return true;
241 #else
242 	return false;
243 #endif
244 }
245 
246 static bool __init xen_check_xsave(void)
247 {
248 	unsigned int cx, xsave_mask;
249 
250 	cx = cpuid_ecx(1);
251 
252 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
253 		     (1 << (X86_FEATURE_OSXSAVE % 32));
254 
255 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
256 	return (cx & xsave_mask) == xsave_mask;
257 }
258 
259 static void __init xen_init_capabilities(void)
260 {
261 	setup_force_cpu_cap(X86_FEATURE_XENPV);
262 	setup_clear_cpu_cap(X86_FEATURE_DCA);
263 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
264 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
265 	setup_clear_cpu_cap(X86_FEATURE_ACC);
266 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
267 	setup_clear_cpu_cap(X86_FEATURE_SME);
268 
269 	/*
270 	 * Xen PV would need some work to support PCID: CR3 handling as well
271 	 * as xen_flush_tlb_others() would need updating.
272 	 */
273 	setup_clear_cpu_cap(X86_FEATURE_PCID);
274 
275 	if (!xen_initial_domain())
276 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
277 
278 	if (xen_check_mwait())
279 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
280 	else
281 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
282 
283 	if (!xen_check_xsave()) {
284 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
285 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
286 	}
287 }
288 
289 static void xen_set_debugreg(int reg, unsigned long val)
290 {
291 	HYPERVISOR_set_debugreg(reg, val);
292 }
293 
294 static unsigned long xen_get_debugreg(int reg)
295 {
296 	return HYPERVISOR_get_debugreg(reg);
297 }
298 
299 static void xen_end_context_switch(struct task_struct *next)
300 {
301 	xen_mc_flush();
302 	paravirt_end_context_switch(next);
303 }
304 
305 static unsigned long xen_store_tr(void)
306 {
307 	return 0;
308 }
309 
310 /*
311  * Set the page permissions for a particular virtual address.  If the
312  * address is a vmalloc mapping (or other non-linear mapping), then
313  * find the linear mapping of the page and also set its protections to
314  * match.
315  */
316 static void set_aliased_prot(void *v, pgprot_t prot)
317 {
318 	int level;
319 	pte_t *ptep;
320 	pte_t pte;
321 	unsigned long pfn;
322 	struct page *page;
323 	unsigned char dummy;
324 
325 	ptep = lookup_address((unsigned long)v, &level);
326 	BUG_ON(ptep == NULL);
327 
328 	pfn = pte_pfn(*ptep);
329 	page = pfn_to_page(pfn);
330 
331 	pte = pfn_pte(pfn, prot);
332 
333 	/*
334 	 * Careful: update_va_mapping() will fail if the virtual address
335 	 * we're poking isn't populated in the page tables.  We don't
336 	 * need to worry about the direct map (that's always in the page
337 	 * tables), but we need to be careful about vmap space.  In
338 	 * particular, the top level page table can lazily propagate
339 	 * entries between processes, so if we've switched mms since we
340 	 * vmapped the target in the first place, we might not have the
341 	 * top-level page table entry populated.
342 	 *
343 	 * We disable preemption because we want the same mm active when
344 	 * we probe the target and when we issue the hypercall.  We'll
345 	 * have the same nominal mm, but if we're a kernel thread, lazy
346 	 * mm dropping could change our pgd.
347 	 *
348 	 * Out of an abundance of caution, this uses __get_user() to fault
349 	 * in the target address just in case there's some obscure case
350 	 * in which the target address isn't readable.
351 	 */
352 
353 	preempt_disable();
354 
355 	probe_kernel_read(&dummy, v, 1);
356 
357 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
358 		BUG();
359 
360 	if (!PageHighMem(page)) {
361 		void *av = __va(PFN_PHYS(pfn));
362 
363 		if (av != v)
364 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
365 				BUG();
366 	} else
367 		kmap_flush_unused();
368 
369 	preempt_enable();
370 }
371 
372 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
373 {
374 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
375 	int i;
376 
377 	/*
378 	 * We need to mark the all aliases of the LDT pages RO.  We
379 	 * don't need to call vm_flush_aliases(), though, since that's
380 	 * only responsible for flushing aliases out the TLBs, not the
381 	 * page tables, and Xen will flush the TLB for us if needed.
382 	 *
383 	 * To avoid confusing future readers: none of this is necessary
384 	 * to load the LDT.  The hypervisor only checks this when the
385 	 * LDT is faulted in due to subsequent descriptor access.
386 	 */
387 
388 	for (i = 0; i < entries; i += entries_per_page)
389 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
390 }
391 
392 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
393 {
394 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
395 	int i;
396 
397 	for (i = 0; i < entries; i += entries_per_page)
398 		set_aliased_prot(ldt + i, PAGE_KERNEL);
399 }
400 
401 static void xen_set_ldt(const void *addr, unsigned entries)
402 {
403 	struct mmuext_op *op;
404 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
405 
406 	trace_xen_cpu_set_ldt(addr, entries);
407 
408 	op = mcs.args;
409 	op->cmd = MMUEXT_SET_LDT;
410 	op->arg1.linear_addr = (unsigned long)addr;
411 	op->arg2.nr_ents = entries;
412 
413 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
414 
415 	xen_mc_issue(PARAVIRT_LAZY_CPU);
416 }
417 
418 static void xen_load_gdt(const struct desc_ptr *dtr)
419 {
420 	unsigned long va = dtr->address;
421 	unsigned int size = dtr->size + 1;
422 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
423 	unsigned long frames[pages];
424 	int f;
425 
426 	/*
427 	 * A GDT can be up to 64k in size, which corresponds to 8192
428 	 * 8-byte entries, or 16 4k pages..
429 	 */
430 
431 	BUG_ON(size > 65536);
432 	BUG_ON(va & ~PAGE_MASK);
433 
434 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
435 		int level;
436 		pte_t *ptep;
437 		unsigned long pfn, mfn;
438 		void *virt;
439 
440 		/*
441 		 * The GDT is per-cpu and is in the percpu data area.
442 		 * That can be virtually mapped, so we need to do a
443 		 * page-walk to get the underlying MFN for the
444 		 * hypercall.  The page can also be in the kernel's
445 		 * linear range, so we need to RO that mapping too.
446 		 */
447 		ptep = lookup_address(va, &level);
448 		BUG_ON(ptep == NULL);
449 
450 		pfn = pte_pfn(*ptep);
451 		mfn = pfn_to_mfn(pfn);
452 		virt = __va(PFN_PHYS(pfn));
453 
454 		frames[f] = mfn;
455 
456 		make_lowmem_page_readonly((void *)va);
457 		make_lowmem_page_readonly(virt);
458 	}
459 
460 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
461 		BUG();
462 }
463 
464 /*
465  * load_gdt for early boot, when the gdt is only mapped once
466  */
467 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
468 {
469 	unsigned long va = dtr->address;
470 	unsigned int size = dtr->size + 1;
471 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
472 	unsigned long frames[pages];
473 	int f;
474 
475 	/*
476 	 * A GDT can be up to 64k in size, which corresponds to 8192
477 	 * 8-byte entries, or 16 4k pages..
478 	 */
479 
480 	BUG_ON(size > 65536);
481 	BUG_ON(va & ~PAGE_MASK);
482 
483 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
484 		pte_t pte;
485 		unsigned long pfn, mfn;
486 
487 		pfn = virt_to_pfn(va);
488 		mfn = pfn_to_mfn(pfn);
489 
490 		pte = pfn_pte(pfn, PAGE_KERNEL_RO);
491 
492 		if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
493 			BUG();
494 
495 		frames[f] = mfn;
496 	}
497 
498 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
499 		BUG();
500 }
501 
502 static inline bool desc_equal(const struct desc_struct *d1,
503 			      const struct desc_struct *d2)
504 {
505 	return !memcmp(d1, d2, sizeof(*d1));
506 }
507 
508 static void load_TLS_descriptor(struct thread_struct *t,
509 				unsigned int cpu, unsigned int i)
510 {
511 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
512 	struct desc_struct *gdt;
513 	xmaddr_t maddr;
514 	struct multicall_space mc;
515 
516 	if (desc_equal(shadow, &t->tls_array[i]))
517 		return;
518 
519 	*shadow = t->tls_array[i];
520 
521 	gdt = get_cpu_gdt_rw(cpu);
522 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
523 	mc = __xen_mc_entry(0);
524 
525 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
526 }
527 
528 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
529 {
530 	/*
531 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
532 	 * and lazy gs handling is enabled, it means we're in a
533 	 * context switch, and %gs has just been saved.  This means we
534 	 * can zero it out to prevent faults on exit from the
535 	 * hypervisor if the next process has no %gs.  Either way, it
536 	 * has been saved, and the new value will get loaded properly.
537 	 * This will go away as soon as Xen has been modified to not
538 	 * save/restore %gs for normal hypercalls.
539 	 *
540 	 * On x86_64, this hack is not used for %gs, because gs points
541 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
542 	 * must not zero %gs on x86_64
543 	 *
544 	 * For x86_64, we need to zero %fs, otherwise we may get an
545 	 * exception between the new %fs descriptor being loaded and
546 	 * %fs being effectively cleared at __switch_to().
547 	 */
548 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
549 #ifdef CONFIG_X86_32
550 		lazy_load_gs(0);
551 #else
552 		loadsegment(fs, 0);
553 #endif
554 	}
555 
556 	xen_mc_batch();
557 
558 	load_TLS_descriptor(t, cpu, 0);
559 	load_TLS_descriptor(t, cpu, 1);
560 	load_TLS_descriptor(t, cpu, 2);
561 
562 	xen_mc_issue(PARAVIRT_LAZY_CPU);
563 }
564 
565 #ifdef CONFIG_X86_64
566 static void xen_load_gs_index(unsigned int idx)
567 {
568 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
569 		BUG();
570 }
571 #endif
572 
573 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
574 				const void *ptr)
575 {
576 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
577 	u64 entry = *(u64 *)ptr;
578 
579 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
580 
581 	preempt_disable();
582 
583 	xen_mc_flush();
584 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
585 		BUG();
586 
587 	preempt_enable();
588 }
589 
590 #ifdef CONFIG_X86_64
591 struct trap_array_entry {
592 	void (*orig)(void);
593 	void (*xen)(void);
594 	bool ist_okay;
595 };
596 
597 static struct trap_array_entry trap_array[] = {
598 	{ debug,                       xen_xendebug,                    true },
599 	{ int3,                        xen_xenint3,                     true },
600 	{ double_fault,                xen_double_fault,                true },
601 #ifdef CONFIG_X86_MCE
602 	{ machine_check,               xen_machine_check,               true },
603 #endif
604 	{ nmi,                         xen_xennmi,                      true },
605 	{ overflow,                    xen_overflow,                    false },
606 #ifdef CONFIG_IA32_EMULATION
607 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
608 #endif
609 	{ page_fault,                  xen_page_fault,                  false },
610 	{ divide_error,                xen_divide_error,                false },
611 	{ bounds,                      xen_bounds,                      false },
612 	{ invalid_op,                  xen_invalid_op,                  false },
613 	{ device_not_available,        xen_device_not_available,        false },
614 	{ coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
615 	{ invalid_TSS,                 xen_invalid_TSS,                 false },
616 	{ segment_not_present,         xen_segment_not_present,         false },
617 	{ stack_segment,               xen_stack_segment,               false },
618 	{ general_protection,          xen_general_protection,          false },
619 	{ spurious_interrupt_bug,      xen_spurious_interrupt_bug,      false },
620 	{ coprocessor_error,           xen_coprocessor_error,           false },
621 	{ alignment_check,             xen_alignment_check,             false },
622 	{ simd_coprocessor_error,      xen_simd_coprocessor_error,      false },
623 };
624 
625 static bool __ref get_trap_addr(void **addr, unsigned int ist)
626 {
627 	unsigned int nr;
628 	bool ist_okay = false;
629 
630 	/*
631 	 * Replace trap handler addresses by Xen specific ones.
632 	 * Check for known traps using IST and whitelist them.
633 	 * The debugger ones are the only ones we care about.
634 	 * Xen will handle faults like double_fault, * so we should never see
635 	 * them.  Warn if there's an unexpected IST-using fault handler.
636 	 */
637 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
638 		struct trap_array_entry *entry = trap_array + nr;
639 
640 		if (*addr == entry->orig) {
641 			*addr = entry->xen;
642 			ist_okay = entry->ist_okay;
643 			break;
644 		}
645 	}
646 
647 	if (nr == ARRAY_SIZE(trap_array) &&
648 	    *addr >= (void *)early_idt_handler_array[0] &&
649 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
650 		nr = (*addr - (void *)early_idt_handler_array[0]) /
651 		     EARLY_IDT_HANDLER_SIZE;
652 		*addr = (void *)xen_early_idt_handler_array[nr];
653 	}
654 
655 	if (WARN_ON(ist != 0 && !ist_okay))
656 		return false;
657 
658 	return true;
659 }
660 #endif
661 
662 static int cvt_gate_to_trap(int vector, const gate_desc *val,
663 			    struct trap_info *info)
664 {
665 	unsigned long addr;
666 
667 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
668 		return 0;
669 
670 	info->vector = vector;
671 
672 	addr = gate_offset(val);
673 #ifdef CONFIG_X86_64
674 	if (!get_trap_addr((void **)&addr, val->bits.ist))
675 		return 0;
676 #endif	/* CONFIG_X86_64 */
677 	info->address = addr;
678 
679 	info->cs = gate_segment(val);
680 	info->flags = val->bits.dpl;
681 	/* interrupt gates clear IF */
682 	if (val->bits.type == GATE_INTERRUPT)
683 		info->flags |= 1 << 2;
684 
685 	return 1;
686 }
687 
688 /* Locations of each CPU's IDT */
689 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
690 
691 /* Set an IDT entry.  If the entry is part of the current IDT, then
692    also update Xen. */
693 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
694 {
695 	unsigned long p = (unsigned long)&dt[entrynum];
696 	unsigned long start, end;
697 
698 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
699 
700 	preempt_disable();
701 
702 	start = __this_cpu_read(idt_desc.address);
703 	end = start + __this_cpu_read(idt_desc.size) + 1;
704 
705 	xen_mc_flush();
706 
707 	native_write_idt_entry(dt, entrynum, g);
708 
709 	if (p >= start && (p + 8) <= end) {
710 		struct trap_info info[2];
711 
712 		info[1].address = 0;
713 
714 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
715 			if (HYPERVISOR_set_trap_table(info))
716 				BUG();
717 	}
718 
719 	preempt_enable();
720 }
721 
722 static void xen_convert_trap_info(const struct desc_ptr *desc,
723 				  struct trap_info *traps)
724 {
725 	unsigned in, out, count;
726 
727 	count = (desc->size+1) / sizeof(gate_desc);
728 	BUG_ON(count > 256);
729 
730 	for (in = out = 0; in < count; in++) {
731 		gate_desc *entry = (gate_desc *)(desc->address) + in;
732 
733 		if (cvt_gate_to_trap(in, entry, &traps[out]))
734 			out++;
735 	}
736 	traps[out].address = 0;
737 }
738 
739 void xen_copy_trap_info(struct trap_info *traps)
740 {
741 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
742 
743 	xen_convert_trap_info(desc, traps);
744 }
745 
746 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
747    hold a spinlock to protect the static traps[] array (static because
748    it avoids allocation, and saves stack space). */
749 static void xen_load_idt(const struct desc_ptr *desc)
750 {
751 	static DEFINE_SPINLOCK(lock);
752 	static struct trap_info traps[257];
753 
754 	trace_xen_cpu_load_idt(desc);
755 
756 	spin_lock(&lock);
757 
758 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
759 
760 	xen_convert_trap_info(desc, traps);
761 
762 	xen_mc_flush();
763 	if (HYPERVISOR_set_trap_table(traps))
764 		BUG();
765 
766 	spin_unlock(&lock);
767 }
768 
769 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
770    they're handled differently. */
771 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
772 				const void *desc, int type)
773 {
774 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
775 
776 	preempt_disable();
777 
778 	switch (type) {
779 	case DESC_LDT:
780 	case DESC_TSS:
781 		/* ignore */
782 		break;
783 
784 	default: {
785 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
786 
787 		xen_mc_flush();
788 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
789 			BUG();
790 	}
791 
792 	}
793 
794 	preempt_enable();
795 }
796 
797 /*
798  * Version of write_gdt_entry for use at early boot-time needed to
799  * update an entry as simply as possible.
800  */
801 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
802 					    const void *desc, int type)
803 {
804 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
805 
806 	switch (type) {
807 	case DESC_LDT:
808 	case DESC_TSS:
809 		/* ignore */
810 		break;
811 
812 	default: {
813 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
814 
815 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
816 			dt[entry] = *(struct desc_struct *)desc;
817 	}
818 
819 	}
820 }
821 
822 static void xen_load_sp0(unsigned long sp0)
823 {
824 	struct multicall_space mcs;
825 
826 	mcs = xen_mc_entry(0);
827 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
828 	xen_mc_issue(PARAVIRT_LAZY_CPU);
829 	this_cpu_write(cpu_tss.x86_tss.sp0, sp0);
830 }
831 
832 void xen_set_iopl_mask(unsigned mask)
833 {
834 	struct physdev_set_iopl set_iopl;
835 
836 	/* Force the change at ring 0. */
837 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
838 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
839 }
840 
841 static void xen_io_delay(void)
842 {
843 }
844 
845 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
846 
847 static unsigned long xen_read_cr0(void)
848 {
849 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
850 
851 	if (unlikely(cr0 == 0)) {
852 		cr0 = native_read_cr0();
853 		this_cpu_write(xen_cr0_value, cr0);
854 	}
855 
856 	return cr0;
857 }
858 
859 static void xen_write_cr0(unsigned long cr0)
860 {
861 	struct multicall_space mcs;
862 
863 	this_cpu_write(xen_cr0_value, cr0);
864 
865 	/* Only pay attention to cr0.TS; everything else is
866 	   ignored. */
867 	mcs = xen_mc_entry(0);
868 
869 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
870 
871 	xen_mc_issue(PARAVIRT_LAZY_CPU);
872 }
873 
874 static void xen_write_cr4(unsigned long cr4)
875 {
876 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
877 
878 	native_write_cr4(cr4);
879 }
880 #ifdef CONFIG_X86_64
881 static inline unsigned long xen_read_cr8(void)
882 {
883 	return 0;
884 }
885 static inline void xen_write_cr8(unsigned long val)
886 {
887 	BUG_ON(val);
888 }
889 #endif
890 
891 static u64 xen_read_msr_safe(unsigned int msr, int *err)
892 {
893 	u64 val;
894 
895 	if (pmu_msr_read(msr, &val, err))
896 		return val;
897 
898 	val = native_read_msr_safe(msr, err);
899 	switch (msr) {
900 	case MSR_IA32_APICBASE:
901 #ifdef CONFIG_X86_X2APIC
902 		if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
903 #endif
904 			val &= ~X2APIC_ENABLE;
905 		break;
906 	}
907 	return val;
908 }
909 
910 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
911 {
912 	int ret;
913 
914 	ret = 0;
915 
916 	switch (msr) {
917 #ifdef CONFIG_X86_64
918 		unsigned which;
919 		u64 base;
920 
921 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
922 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
923 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
924 
925 	set:
926 		base = ((u64)high << 32) | low;
927 		if (HYPERVISOR_set_segment_base(which, base) != 0)
928 			ret = -EIO;
929 		break;
930 #endif
931 
932 	case MSR_STAR:
933 	case MSR_CSTAR:
934 	case MSR_LSTAR:
935 	case MSR_SYSCALL_MASK:
936 	case MSR_IA32_SYSENTER_CS:
937 	case MSR_IA32_SYSENTER_ESP:
938 	case MSR_IA32_SYSENTER_EIP:
939 		/* Fast syscall setup is all done in hypercalls, so
940 		   these are all ignored.  Stub them out here to stop
941 		   Xen console noise. */
942 		break;
943 
944 	default:
945 		if (!pmu_msr_write(msr, low, high, &ret))
946 			ret = native_write_msr_safe(msr, low, high);
947 	}
948 
949 	return ret;
950 }
951 
952 static u64 xen_read_msr(unsigned int msr)
953 {
954 	/*
955 	 * This will silently swallow a #GP from RDMSR.  It may be worth
956 	 * changing that.
957 	 */
958 	int err;
959 
960 	return xen_read_msr_safe(msr, &err);
961 }
962 
963 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
964 {
965 	/*
966 	 * This will silently swallow a #GP from WRMSR.  It may be worth
967 	 * changing that.
968 	 */
969 	xen_write_msr_safe(msr, low, high);
970 }
971 
972 void xen_setup_shared_info(void)
973 {
974 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
975 
976 	HYPERVISOR_shared_info =
977 		(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
978 
979 	xen_setup_mfn_list_list();
980 
981 	if (system_state == SYSTEM_BOOTING) {
982 #ifndef CONFIG_SMP
983 		/*
984 		 * In UP this is as good a place as any to set up shared info.
985 		 * Limit this to boot only, at restore vcpu setup is done via
986 		 * xen_vcpu_restore().
987 		 */
988 		xen_setup_vcpu_info_placement();
989 #endif
990 		/*
991 		 * Now that shared info is set up we can start using routines
992 		 * that point to pvclock area.
993 		 */
994 		xen_init_time_ops();
995 	}
996 }
997 
998 /* This is called once we have the cpu_possible_mask */
999 void __ref xen_setup_vcpu_info_placement(void)
1000 {
1001 	int cpu;
1002 
1003 	for_each_possible_cpu(cpu) {
1004 		/* Set up direct vCPU id mapping for PV guests. */
1005 		per_cpu(xen_vcpu_id, cpu) = cpu;
1006 
1007 		/*
1008 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
1009 		 * falls back to the shared_info version for cpus
1010 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1011 		 *
1012 		 * xen_cpu_up_prepare_pv() handles the rest by failing
1013 		 * them in hotplug.
1014 		 */
1015 		(void) xen_vcpu_setup(cpu);
1016 	}
1017 
1018 	/*
1019 	 * xen_vcpu_setup managed to place the vcpu_info within the
1020 	 * percpu area for all cpus, so make use of it.
1021 	 */
1022 	if (xen_have_vcpu_info_placement) {
1023 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1024 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1025 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1026 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1027 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1028 	}
1029 }
1030 
1031 static const struct pv_info xen_info __initconst = {
1032 	.shared_kernel_pmd = 0,
1033 
1034 #ifdef CONFIG_X86_64
1035 	.extra_user_64bit_cs = FLAT_USER_CS64,
1036 #endif
1037 	.name = "Xen",
1038 };
1039 
1040 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1041 	.cpuid = xen_cpuid,
1042 
1043 	.set_debugreg = xen_set_debugreg,
1044 	.get_debugreg = xen_get_debugreg,
1045 
1046 	.read_cr0 = xen_read_cr0,
1047 	.write_cr0 = xen_write_cr0,
1048 
1049 	.write_cr4 = xen_write_cr4,
1050 
1051 #ifdef CONFIG_X86_64
1052 	.read_cr8 = xen_read_cr8,
1053 	.write_cr8 = xen_write_cr8,
1054 #endif
1055 
1056 	.wbinvd = native_wbinvd,
1057 
1058 	.read_msr = xen_read_msr,
1059 	.write_msr = xen_write_msr,
1060 
1061 	.read_msr_safe = xen_read_msr_safe,
1062 	.write_msr_safe = xen_write_msr_safe,
1063 
1064 	.read_pmc = xen_read_pmc,
1065 
1066 	.iret = xen_iret,
1067 #ifdef CONFIG_X86_64
1068 	.usergs_sysret64 = xen_sysret64,
1069 #endif
1070 
1071 	.load_tr_desc = paravirt_nop,
1072 	.set_ldt = xen_set_ldt,
1073 	.load_gdt = xen_load_gdt,
1074 	.load_idt = xen_load_idt,
1075 	.load_tls = xen_load_tls,
1076 #ifdef CONFIG_X86_64
1077 	.load_gs_index = xen_load_gs_index,
1078 #endif
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 	.set_iopl_mask = xen_set_iopl_mask,
1091 	.io_delay = xen_io_delay,
1092 
1093 	/* Xen takes care of %gs when switching to usermode for us */
1094 	.swapgs = paravirt_nop,
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 xen_setup_gdt(int cpu)
1201 {
1202 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1203 	pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1204 
1205 	setup_stack_canary_segment(0);
1206 	switch_to_new_gdt(0);
1207 
1208 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1209 	pv_cpu_ops.load_gdt = xen_load_gdt;
1210 }
1211 
1212 static void __init xen_dom0_set_legacy_features(void)
1213 {
1214 	x86_platform.legacy.rtc = 1;
1215 }
1216 
1217 /* First C function to be called on Xen boot */
1218 asmlinkage __visible void __init xen_start_kernel(void)
1219 {
1220 	struct physdev_set_iopl set_iopl;
1221 	unsigned long initrd_start = 0;
1222 	int rc;
1223 
1224 	if (!xen_start_info)
1225 		return;
1226 
1227 	xen_domain_type = XEN_PV_DOMAIN;
1228 
1229 	xen_setup_features();
1230 
1231 	xen_setup_machphys_mapping();
1232 
1233 	/* Install Xen paravirt ops */
1234 	pv_info = xen_info;
1235 	pv_init_ops.patch = paravirt_patch_default;
1236 	pv_cpu_ops = xen_cpu_ops;
1237 
1238 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1239 
1240 	x86_init.resources.memory_setup = xen_memory_setup;
1241 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1242 	x86_init.oem.arch_setup = xen_arch_setup;
1243 	x86_init.oem.banner = xen_banner;
1244 
1245 	/*
1246 	 * Set up some pagetable state before starting to set any ptes.
1247 	 */
1248 
1249 	xen_init_mmu_ops();
1250 
1251 	/* Prevent unwanted bits from being set in PTEs. */
1252 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1253 
1254 	/*
1255 	 * Prevent page tables from being allocated in highmem, even
1256 	 * if CONFIG_HIGHPTE is enabled.
1257 	 */
1258 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1259 
1260 	/* Work out if we support NX */
1261 	x86_configure_nx();
1262 
1263 	/* Get mfn list */
1264 	xen_build_dynamic_phys_to_machine();
1265 
1266 	/*
1267 	 * Set up kernel GDT and segment registers, mainly so that
1268 	 * -fstack-protector code can be executed.
1269 	 */
1270 	xen_setup_gdt(0);
1271 
1272 	xen_init_irq_ops();
1273 
1274 	/* Let's presume PV guests always boot on vCPU with id 0. */
1275 	per_cpu(xen_vcpu_id, 0) = 0;
1276 
1277 	/*
1278 	 * Setup xen_vcpu early because idt_setup_early_handler needs it for
1279 	 * local_irq_disable(), irqs_disabled().
1280 	 *
1281 	 * Don't do the full vcpu_info placement stuff until we have
1282 	 * the cpu_possible_mask and a non-dummy shared_info.
1283 	 */
1284 	xen_vcpu_info_reset(0);
1285 
1286 	idt_setup_early_handler();
1287 
1288 	xen_init_capabilities();
1289 
1290 #ifdef CONFIG_X86_LOCAL_APIC
1291 	/*
1292 	 * set up the basic apic ops.
1293 	 */
1294 	xen_init_apic();
1295 #endif
1296 
1297 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1298 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1299 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1300 	}
1301 
1302 	machine_ops = xen_machine_ops;
1303 
1304 	/*
1305 	 * The only reliable way to retain the initial address of the
1306 	 * percpu gdt_page is to remember it here, so we can go and
1307 	 * mark it RW later, when the initial percpu area is freed.
1308 	 */
1309 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1310 
1311 	xen_smp_init();
1312 
1313 #ifdef CONFIG_ACPI_NUMA
1314 	/*
1315 	 * The pages we from Xen are not related to machine pages, so
1316 	 * any NUMA information the kernel tries to get from ACPI will
1317 	 * be meaningless.  Prevent it from trying.
1318 	 */
1319 	acpi_numa = -1;
1320 #endif
1321 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1322 
1323 	local_irq_disable();
1324 	early_boot_irqs_disabled = true;
1325 
1326 	xen_raw_console_write("mapping kernel into physical memory\n");
1327 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1328 				   xen_start_info->nr_pages);
1329 	xen_reserve_special_pages();
1330 
1331 	/* keep using Xen gdt for now; no urgent need to change it */
1332 
1333 #ifdef CONFIG_X86_32
1334 	pv_info.kernel_rpl = 1;
1335 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1336 		pv_info.kernel_rpl = 0;
1337 #else
1338 	pv_info.kernel_rpl = 0;
1339 #endif
1340 	/* set the limit of our address space */
1341 	xen_reserve_top();
1342 
1343 	/*
1344 	 * We used to do this in xen_arch_setup, but that is too late
1345 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1346 	 * early_amd_init which pokes 0xcf8 port.
1347 	 */
1348 	set_iopl.iopl = 1;
1349 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1350 	if (rc != 0)
1351 		xen_raw_printk("physdev_op failed %d\n", rc);
1352 
1353 #ifdef CONFIG_X86_32
1354 	/* set up basic CPUID stuff */
1355 	cpu_detect(&new_cpu_data);
1356 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1357 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1358 #endif
1359 
1360 	if (xen_start_info->mod_start) {
1361 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1362 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1363 	    else
1364 		initrd_start = __pa(xen_start_info->mod_start);
1365 	}
1366 
1367 	/* Poke various useful things into boot_params */
1368 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1369 	boot_params.hdr.ramdisk_image = initrd_start;
1370 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1371 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1372 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1373 
1374 	if (!xen_initial_domain()) {
1375 		add_preferred_console("xenboot", 0, NULL);
1376 		add_preferred_console("tty", 0, NULL);
1377 		add_preferred_console("hvc", 0, NULL);
1378 		if (pci_xen)
1379 			x86_init.pci.arch_init = pci_xen_init;
1380 	} else {
1381 		const struct dom0_vga_console_info *info =
1382 			(void *)((char *)xen_start_info +
1383 				 xen_start_info->console.dom0.info_off);
1384 		struct xen_platform_op op = {
1385 			.cmd = XENPF_firmware_info,
1386 			.interface_version = XENPF_INTERFACE_VERSION,
1387 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1388 		};
1389 
1390 		x86_platform.set_legacy_features =
1391 				xen_dom0_set_legacy_features;
1392 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1393 		xen_start_info->console.domU.mfn = 0;
1394 		xen_start_info->console.domU.evtchn = 0;
1395 
1396 		if (HYPERVISOR_platform_op(&op) == 0)
1397 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1398 
1399 		/* Make sure ACS will be enabled */
1400 		pci_request_acs();
1401 
1402 		xen_acpi_sleep_register();
1403 
1404 		/* Avoid searching for BIOS MP tables */
1405 		x86_init.mpparse.find_smp_config = x86_init_noop;
1406 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1407 
1408 		xen_boot_params_init_edd();
1409 	}
1410 #ifdef CONFIG_PCI
1411 	/* PCI BIOS service won't work from a PV guest. */
1412 	pci_probe &= ~PCI_PROBE_BIOS;
1413 #endif
1414 	xen_raw_console_write("about to get started...\n");
1415 
1416 	/* We need this for printk timestamps */
1417 	xen_setup_runstate_info(0);
1418 
1419 	xen_efi_init();
1420 
1421 	/* Start the world */
1422 #ifdef CONFIG_X86_32
1423 	i386_start_kernel();
1424 #else
1425 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1426 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1427 #endif
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 };
1480