xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 6724ed7f)
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 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
92 
93 void *xen_initial_gdt;
94 
95 static int xen_cpu_up_prepare_pv(unsigned int cpu);
96 static int xen_cpu_dead_pv(unsigned int cpu);
97 
98 struct tls_descs {
99 	struct desc_struct desc[3];
100 };
101 
102 /*
103  * Updating the 3 TLS descriptors in the GDT on every task switch is
104  * surprisingly expensive so we avoid updating them if they haven't
105  * changed.  Since Xen writes different descriptors than the one
106  * passed in the update_descriptor hypercall we keep shadow copies to
107  * compare against.
108  */
109 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
110 
111 static void __init xen_banner(void)
112 {
113 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
114 	struct xen_extraversion extra;
115 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
116 
117 	pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
118 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
119 	       version >> 16, version & 0xffff, extra.extraversion,
120 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
121 }
122 /* Check if running on Xen version (major, minor) or later */
123 bool
124 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
125 {
126 	unsigned int version;
127 
128 	if (!xen_domain())
129 		return false;
130 
131 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
132 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
133 		((version >> 16) > major))
134 		return true;
135 	return false;
136 }
137 
138 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
139 static __read_mostly unsigned int cpuid_leaf5_edx_val;
140 
141 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
142 		      unsigned int *cx, unsigned int *dx)
143 {
144 	unsigned maskebx = ~0;
145 
146 	/*
147 	 * Mask out inconvenient features, to try and disable as many
148 	 * unsupported kernel subsystems as possible.
149 	 */
150 	switch (*ax) {
151 	case CPUID_MWAIT_LEAF:
152 		/* Synthesize the values.. */
153 		*ax = 0;
154 		*bx = 0;
155 		*cx = cpuid_leaf5_ecx_val;
156 		*dx = cpuid_leaf5_edx_val;
157 		return;
158 
159 	case 0xb:
160 		/* Suppress extended topology stuff */
161 		maskebx = 0;
162 		break;
163 	}
164 
165 	asm(XEN_EMULATE_PREFIX "cpuid"
166 		: "=a" (*ax),
167 		  "=b" (*bx),
168 		  "=c" (*cx),
169 		  "=d" (*dx)
170 		: "0" (*ax), "2" (*cx));
171 
172 	*bx &= maskebx;
173 }
174 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
175 
176 static bool __init xen_check_mwait(void)
177 {
178 #ifdef CONFIG_ACPI
179 	struct xen_platform_op op = {
180 		.cmd			= XENPF_set_processor_pminfo,
181 		.u.set_pminfo.id	= -1,
182 		.u.set_pminfo.type	= XEN_PM_PDC,
183 	};
184 	uint32_t buf[3];
185 	unsigned int ax, bx, cx, dx;
186 	unsigned int mwait_mask;
187 
188 	/* We need to determine whether it is OK to expose the MWAIT
189 	 * capability to the kernel to harvest deeper than C3 states from ACPI
190 	 * _CST using the processor_harvest_xen.c module. For this to work, we
191 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
192 	 * checks against). The hypervisor won't expose the MWAIT flag because
193 	 * it would break backwards compatibility; so we will find out directly
194 	 * from the hardware and hypercall.
195 	 */
196 	if (!xen_initial_domain())
197 		return false;
198 
199 	/*
200 	 * When running under platform earlier than Xen4.2, do not expose
201 	 * mwait, to avoid the risk of loading native acpi pad driver
202 	 */
203 	if (!xen_running_on_version_or_later(4, 2))
204 		return false;
205 
206 	ax = 1;
207 	cx = 0;
208 
209 	native_cpuid(&ax, &bx, &cx, &dx);
210 
211 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
212 		     (1 << (X86_FEATURE_MWAIT % 32));
213 
214 	if ((cx & mwait_mask) != mwait_mask)
215 		return false;
216 
217 	/* We need to emulate the MWAIT_LEAF and for that we need both
218 	 * ecx and edx. The hypercall provides only partial information.
219 	 */
220 
221 	ax = CPUID_MWAIT_LEAF;
222 	bx = 0;
223 	cx = 0;
224 	dx = 0;
225 
226 	native_cpuid(&ax, &bx, &cx, &dx);
227 
228 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
229 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
230 	 */
231 	buf[0] = ACPI_PDC_REVISION_ID;
232 	buf[1] = 1;
233 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
234 
235 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
236 
237 	if ((HYPERVISOR_platform_op(&op) == 0) &&
238 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
239 		cpuid_leaf5_ecx_val = cx;
240 		cpuid_leaf5_edx_val = dx;
241 	}
242 	return true;
243 #else
244 	return false;
245 #endif
246 }
247 
248 static bool __init xen_check_xsave(void)
249 {
250 	unsigned int cx, xsave_mask;
251 
252 	cx = cpuid_ecx(1);
253 
254 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
255 		     (1 << (X86_FEATURE_OSXSAVE % 32));
256 
257 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
258 	return (cx & xsave_mask) == xsave_mask;
259 }
260 
261 static void __init xen_init_capabilities(void)
262 {
263 	setup_force_cpu_cap(X86_FEATURE_XENPV);
264 	setup_clear_cpu_cap(X86_FEATURE_DCA);
265 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
266 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
267 	setup_clear_cpu_cap(X86_FEATURE_ACC);
268 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
269 	setup_clear_cpu_cap(X86_FEATURE_SME);
270 
271 	/*
272 	 * Xen PV would need some work to support PCID: CR3 handling as well
273 	 * as xen_flush_tlb_others() would need updating.
274 	 */
275 	setup_clear_cpu_cap(X86_FEATURE_PCID);
276 
277 	if (!xen_initial_domain())
278 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
279 
280 	if (xen_check_mwait())
281 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
282 	else
283 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
284 
285 	if (!xen_check_xsave()) {
286 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
287 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
288 	}
289 }
290 
291 static void xen_set_debugreg(int reg, unsigned long val)
292 {
293 	HYPERVISOR_set_debugreg(reg, val);
294 }
295 
296 static unsigned long xen_get_debugreg(int reg)
297 {
298 	return HYPERVISOR_get_debugreg(reg);
299 }
300 
301 static void xen_end_context_switch(struct task_struct *next)
302 {
303 	xen_mc_flush();
304 	paravirt_end_context_switch(next);
305 }
306 
307 static unsigned long xen_store_tr(void)
308 {
309 	return 0;
310 }
311 
312 /*
313  * Set the page permissions for a particular virtual address.  If the
314  * address is a vmalloc mapping (or other non-linear mapping), then
315  * find the linear mapping of the page and also set its protections to
316  * match.
317  */
318 static void set_aliased_prot(void *v, pgprot_t prot)
319 {
320 	int level;
321 	pte_t *ptep;
322 	pte_t pte;
323 	unsigned long pfn;
324 	struct page *page;
325 	unsigned char dummy;
326 
327 	ptep = lookup_address((unsigned long)v, &level);
328 	BUG_ON(ptep == NULL);
329 
330 	pfn = pte_pfn(*ptep);
331 	page = pfn_to_page(pfn);
332 
333 	pte = pfn_pte(pfn, prot);
334 
335 	/*
336 	 * Careful: update_va_mapping() will fail if the virtual address
337 	 * we're poking isn't populated in the page tables.  We don't
338 	 * need to worry about the direct map (that's always in the page
339 	 * tables), but we need to be careful about vmap space.  In
340 	 * particular, the top level page table can lazily propagate
341 	 * entries between processes, so if we've switched mms since we
342 	 * vmapped the target in the first place, we might not have the
343 	 * top-level page table entry populated.
344 	 *
345 	 * We disable preemption because we want the same mm active when
346 	 * we probe the target and when we issue the hypercall.  We'll
347 	 * have the same nominal mm, but if we're a kernel thread, lazy
348 	 * mm dropping could change our pgd.
349 	 *
350 	 * Out of an abundance of caution, this uses __get_user() to fault
351 	 * in the target address just in case there's some obscure case
352 	 * in which the target address isn't readable.
353 	 */
354 
355 	preempt_disable();
356 
357 	probe_kernel_read(&dummy, v, 1);
358 
359 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
360 		BUG();
361 
362 	if (!PageHighMem(page)) {
363 		void *av = __va(PFN_PHYS(pfn));
364 
365 		if (av != v)
366 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
367 				BUG();
368 	} else
369 		kmap_flush_unused();
370 
371 	preempt_enable();
372 }
373 
374 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
375 {
376 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
377 	int i;
378 
379 	/*
380 	 * We need to mark the all aliases of the LDT pages RO.  We
381 	 * don't need to call vm_flush_aliases(), though, since that's
382 	 * only responsible for flushing aliases out the TLBs, not the
383 	 * page tables, and Xen will flush the TLB for us if needed.
384 	 *
385 	 * To avoid confusing future readers: none of this is necessary
386 	 * to load the LDT.  The hypervisor only checks this when the
387 	 * LDT is faulted in due to subsequent descriptor access.
388 	 */
389 
390 	for (i = 0; i < entries; i += entries_per_page)
391 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
392 }
393 
394 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
395 {
396 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
397 	int i;
398 
399 	for (i = 0; i < entries; i += entries_per_page)
400 		set_aliased_prot(ldt + i, PAGE_KERNEL);
401 }
402 
403 static void xen_set_ldt(const void *addr, unsigned entries)
404 {
405 	struct mmuext_op *op;
406 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
407 
408 	trace_xen_cpu_set_ldt(addr, entries);
409 
410 	op = mcs.args;
411 	op->cmd = MMUEXT_SET_LDT;
412 	op->arg1.linear_addr = (unsigned long)addr;
413 	op->arg2.nr_ents = entries;
414 
415 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
416 
417 	xen_mc_issue(PARAVIRT_LAZY_CPU);
418 }
419 
420 static void xen_load_gdt(const struct desc_ptr *dtr)
421 {
422 	unsigned long va = dtr->address;
423 	unsigned int size = dtr->size + 1;
424 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
425 	unsigned long frames[pages];
426 	int f;
427 
428 	/*
429 	 * A GDT can be up to 64k in size, which corresponds to 8192
430 	 * 8-byte entries, or 16 4k pages..
431 	 */
432 
433 	BUG_ON(size > 65536);
434 	BUG_ON(va & ~PAGE_MASK);
435 
436 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
437 		int level;
438 		pte_t *ptep;
439 		unsigned long pfn, mfn;
440 		void *virt;
441 
442 		/*
443 		 * The GDT is per-cpu and is in the percpu data area.
444 		 * That can be virtually mapped, so we need to do a
445 		 * page-walk to get the underlying MFN for the
446 		 * hypercall.  The page can also be in the kernel's
447 		 * linear range, so we need to RO that mapping too.
448 		 */
449 		ptep = lookup_address(va, &level);
450 		BUG_ON(ptep == NULL);
451 
452 		pfn = pte_pfn(*ptep);
453 		mfn = pfn_to_mfn(pfn);
454 		virt = __va(PFN_PHYS(pfn));
455 
456 		frames[f] = mfn;
457 
458 		make_lowmem_page_readonly((void *)va);
459 		make_lowmem_page_readonly(virt);
460 	}
461 
462 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
463 		BUG();
464 }
465 
466 /*
467  * load_gdt for early boot, when the gdt is only mapped once
468  */
469 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
470 {
471 	unsigned long va = dtr->address;
472 	unsigned int size = dtr->size + 1;
473 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
474 	unsigned long frames[pages];
475 	int f;
476 
477 	/*
478 	 * A GDT can be up to 64k in size, which corresponds to 8192
479 	 * 8-byte entries, or 16 4k pages..
480 	 */
481 
482 	BUG_ON(size > 65536);
483 	BUG_ON(va & ~PAGE_MASK);
484 
485 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
486 		pte_t pte;
487 		unsigned long pfn, mfn;
488 
489 		pfn = virt_to_pfn(va);
490 		mfn = pfn_to_mfn(pfn);
491 
492 		pte = pfn_pte(pfn, PAGE_KERNEL_RO);
493 
494 		if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
495 			BUG();
496 
497 		frames[f] = mfn;
498 	}
499 
500 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
501 		BUG();
502 }
503 
504 static inline bool desc_equal(const struct desc_struct *d1,
505 			      const struct desc_struct *d2)
506 {
507 	return !memcmp(d1, d2, sizeof(*d1));
508 }
509 
510 static void load_TLS_descriptor(struct thread_struct *t,
511 				unsigned int cpu, unsigned int i)
512 {
513 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
514 	struct desc_struct *gdt;
515 	xmaddr_t maddr;
516 	struct multicall_space mc;
517 
518 	if (desc_equal(shadow, &t->tls_array[i]))
519 		return;
520 
521 	*shadow = t->tls_array[i];
522 
523 	gdt = get_cpu_gdt_rw(cpu);
524 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
525 	mc = __xen_mc_entry(0);
526 
527 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
528 }
529 
530 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
531 {
532 	/*
533 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
534 	 * and lazy gs handling is enabled, it means we're in a
535 	 * context switch, and %gs has just been saved.  This means we
536 	 * can zero it out to prevent faults on exit from the
537 	 * hypervisor if the next process has no %gs.  Either way, it
538 	 * has been saved, and the new value will get loaded properly.
539 	 * This will go away as soon as Xen has been modified to not
540 	 * save/restore %gs for normal hypercalls.
541 	 *
542 	 * On x86_64, this hack is not used for %gs, because gs points
543 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
544 	 * must not zero %gs on x86_64
545 	 *
546 	 * For x86_64, we need to zero %fs, otherwise we may get an
547 	 * exception between the new %fs descriptor being loaded and
548 	 * %fs being effectively cleared at __switch_to().
549 	 */
550 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
551 #ifdef CONFIG_X86_32
552 		lazy_load_gs(0);
553 #else
554 		loadsegment(fs, 0);
555 #endif
556 	}
557 
558 	xen_mc_batch();
559 
560 	load_TLS_descriptor(t, cpu, 0);
561 	load_TLS_descriptor(t, cpu, 1);
562 	load_TLS_descriptor(t, cpu, 2);
563 
564 	xen_mc_issue(PARAVIRT_LAZY_CPU);
565 }
566 
567 #ifdef CONFIG_X86_64
568 static void xen_load_gs_index(unsigned int idx)
569 {
570 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
571 		BUG();
572 }
573 #endif
574 
575 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
576 				const void *ptr)
577 {
578 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
579 	u64 entry = *(u64 *)ptr;
580 
581 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
582 
583 	preempt_disable();
584 
585 	xen_mc_flush();
586 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
587 		BUG();
588 
589 	preempt_enable();
590 }
591 
592 #ifdef CONFIG_X86_64
593 struct trap_array_entry {
594 	void (*orig)(void);
595 	void (*xen)(void);
596 	bool ist_okay;
597 };
598 
599 static struct trap_array_entry trap_array[] = {
600 	{ debug,                       xen_xendebug,                    true },
601 	{ int3,                        xen_xenint3,                     true },
602 	{ double_fault,                xen_double_fault,                true },
603 #ifdef CONFIG_X86_MCE
604 	{ machine_check,               xen_machine_check,               true },
605 #endif
606 	{ nmi,                         xen_xennmi,                      true },
607 	{ overflow,                    xen_overflow,                    false },
608 #ifdef CONFIG_IA32_EMULATION
609 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
610 #endif
611 	{ page_fault,                  xen_page_fault,                  false },
612 	{ divide_error,                xen_divide_error,                false },
613 	{ bounds,                      xen_bounds,                      false },
614 	{ invalid_op,                  xen_invalid_op,                  false },
615 	{ device_not_available,        xen_device_not_available,        false },
616 	{ coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
617 	{ invalid_TSS,                 xen_invalid_TSS,                 false },
618 	{ segment_not_present,         xen_segment_not_present,         false },
619 	{ stack_segment,               xen_stack_segment,               false },
620 	{ general_protection,          xen_general_protection,          false },
621 	{ spurious_interrupt_bug,      xen_spurious_interrupt_bug,      false },
622 	{ coprocessor_error,           xen_coprocessor_error,           false },
623 	{ alignment_check,             xen_alignment_check,             false },
624 	{ simd_coprocessor_error,      xen_simd_coprocessor_error,      false },
625 };
626 
627 static bool __ref get_trap_addr(void **addr, unsigned int ist)
628 {
629 	unsigned int nr;
630 	bool ist_okay = false;
631 
632 	/*
633 	 * Replace trap handler addresses by Xen specific ones.
634 	 * Check for known traps using IST and whitelist them.
635 	 * The debugger ones are the only ones we care about.
636 	 * Xen will handle faults like double_fault, * so we should never see
637 	 * them.  Warn if there's an unexpected IST-using fault handler.
638 	 */
639 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
640 		struct trap_array_entry *entry = trap_array + nr;
641 
642 		if (*addr == entry->orig) {
643 			*addr = entry->xen;
644 			ist_okay = entry->ist_okay;
645 			break;
646 		}
647 	}
648 
649 	if (nr == ARRAY_SIZE(trap_array) &&
650 	    *addr >= (void *)early_idt_handler_array[0] &&
651 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
652 		nr = (*addr - (void *)early_idt_handler_array[0]) /
653 		     EARLY_IDT_HANDLER_SIZE;
654 		*addr = (void *)xen_early_idt_handler_array[nr];
655 	}
656 
657 	if (WARN_ON(ist != 0 && !ist_okay))
658 		return false;
659 
660 	return true;
661 }
662 #endif
663 
664 static int cvt_gate_to_trap(int vector, const gate_desc *val,
665 			    struct trap_info *info)
666 {
667 	unsigned long addr;
668 
669 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
670 		return 0;
671 
672 	info->vector = vector;
673 
674 	addr = gate_offset(val);
675 #ifdef CONFIG_X86_64
676 	if (!get_trap_addr((void **)&addr, val->bits.ist))
677 		return 0;
678 #endif	/* CONFIG_X86_64 */
679 	info->address = addr;
680 
681 	info->cs = gate_segment(val);
682 	info->flags = val->bits.dpl;
683 	/* interrupt gates clear IF */
684 	if (val->bits.type == GATE_INTERRUPT)
685 		info->flags |= 1 << 2;
686 
687 	return 1;
688 }
689 
690 /* Locations of each CPU's IDT */
691 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
692 
693 /* Set an IDT entry.  If the entry is part of the current IDT, then
694    also update Xen. */
695 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
696 {
697 	unsigned long p = (unsigned long)&dt[entrynum];
698 	unsigned long start, end;
699 
700 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
701 
702 	preempt_disable();
703 
704 	start = __this_cpu_read(idt_desc.address);
705 	end = start + __this_cpu_read(idt_desc.size) + 1;
706 
707 	xen_mc_flush();
708 
709 	native_write_idt_entry(dt, entrynum, g);
710 
711 	if (p >= start && (p + 8) <= end) {
712 		struct trap_info info[2];
713 
714 		info[1].address = 0;
715 
716 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
717 			if (HYPERVISOR_set_trap_table(info))
718 				BUG();
719 	}
720 
721 	preempt_enable();
722 }
723 
724 static void xen_convert_trap_info(const struct desc_ptr *desc,
725 				  struct trap_info *traps)
726 {
727 	unsigned in, out, count;
728 
729 	count = (desc->size+1) / sizeof(gate_desc);
730 	BUG_ON(count > 256);
731 
732 	for (in = out = 0; in < count; in++) {
733 		gate_desc *entry = (gate_desc *)(desc->address) + in;
734 
735 		if (cvt_gate_to_trap(in, entry, &traps[out]))
736 			out++;
737 	}
738 	traps[out].address = 0;
739 }
740 
741 void xen_copy_trap_info(struct trap_info *traps)
742 {
743 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
744 
745 	xen_convert_trap_info(desc, traps);
746 }
747 
748 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
749    hold a spinlock to protect the static traps[] array (static because
750    it avoids allocation, and saves stack space). */
751 static void xen_load_idt(const struct desc_ptr *desc)
752 {
753 	static DEFINE_SPINLOCK(lock);
754 	static struct trap_info traps[257];
755 
756 	trace_xen_cpu_load_idt(desc);
757 
758 	spin_lock(&lock);
759 
760 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
761 
762 	xen_convert_trap_info(desc, traps);
763 
764 	xen_mc_flush();
765 	if (HYPERVISOR_set_trap_table(traps))
766 		BUG();
767 
768 	spin_unlock(&lock);
769 }
770 
771 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
772    they're handled differently. */
773 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
774 				const void *desc, int type)
775 {
776 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
777 
778 	preempt_disable();
779 
780 	switch (type) {
781 	case DESC_LDT:
782 	case DESC_TSS:
783 		/* ignore */
784 		break;
785 
786 	default: {
787 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
788 
789 		xen_mc_flush();
790 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
791 			BUG();
792 	}
793 
794 	}
795 
796 	preempt_enable();
797 }
798 
799 /*
800  * Version of write_gdt_entry for use at early boot-time needed to
801  * update an entry as simply as possible.
802  */
803 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
804 					    const void *desc, int type)
805 {
806 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
807 
808 	switch (type) {
809 	case DESC_LDT:
810 	case DESC_TSS:
811 		/* ignore */
812 		break;
813 
814 	default: {
815 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
816 
817 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
818 			dt[entry] = *(struct desc_struct *)desc;
819 	}
820 
821 	}
822 }
823 
824 static void xen_load_sp0(unsigned long sp0)
825 {
826 	struct multicall_space mcs;
827 
828 	mcs = xen_mc_entry(0);
829 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
830 	xen_mc_issue(PARAVIRT_LAZY_CPU);
831 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
832 }
833 
834 void xen_set_iopl_mask(unsigned mask)
835 {
836 	struct physdev_set_iopl set_iopl;
837 
838 	/* Force the change at ring 0. */
839 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
840 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
841 }
842 
843 static void xen_io_delay(void)
844 {
845 }
846 
847 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
848 
849 static unsigned long xen_read_cr0(void)
850 {
851 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
852 
853 	if (unlikely(cr0 == 0)) {
854 		cr0 = native_read_cr0();
855 		this_cpu_write(xen_cr0_value, cr0);
856 	}
857 
858 	return cr0;
859 }
860 
861 static void xen_write_cr0(unsigned long cr0)
862 {
863 	struct multicall_space mcs;
864 
865 	this_cpu_write(xen_cr0_value, cr0);
866 
867 	/* Only pay attention to cr0.TS; everything else is
868 	   ignored. */
869 	mcs = xen_mc_entry(0);
870 
871 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
872 
873 	xen_mc_issue(PARAVIRT_LAZY_CPU);
874 }
875 
876 static void xen_write_cr4(unsigned long cr4)
877 {
878 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
879 
880 	native_write_cr4(cr4);
881 }
882 #ifdef CONFIG_X86_64
883 static inline unsigned long xen_read_cr8(void)
884 {
885 	return 0;
886 }
887 static inline void xen_write_cr8(unsigned long val)
888 {
889 	BUG_ON(val);
890 }
891 #endif
892 
893 static u64 xen_read_msr_safe(unsigned int msr, int *err)
894 {
895 	u64 val;
896 
897 	if (pmu_msr_read(msr, &val, err))
898 		return val;
899 
900 	val = native_read_msr_safe(msr, err);
901 	switch (msr) {
902 	case MSR_IA32_APICBASE:
903 #ifdef CONFIG_X86_X2APIC
904 		if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
905 #endif
906 			val &= ~X2APIC_ENABLE;
907 		break;
908 	}
909 	return val;
910 }
911 
912 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
913 {
914 	int ret;
915 
916 	ret = 0;
917 
918 	switch (msr) {
919 #ifdef CONFIG_X86_64
920 		unsigned which;
921 		u64 base;
922 
923 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
924 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
925 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
926 
927 	set:
928 		base = ((u64)high << 32) | low;
929 		if (HYPERVISOR_set_segment_base(which, base) != 0)
930 			ret = -EIO;
931 		break;
932 #endif
933 
934 	case MSR_STAR:
935 	case MSR_CSTAR:
936 	case MSR_LSTAR:
937 	case MSR_SYSCALL_MASK:
938 	case MSR_IA32_SYSENTER_CS:
939 	case MSR_IA32_SYSENTER_ESP:
940 	case MSR_IA32_SYSENTER_EIP:
941 		/* Fast syscall setup is all done in hypercalls, so
942 		   these are all ignored.  Stub them out here to stop
943 		   Xen console noise. */
944 		break;
945 
946 	default:
947 		if (!pmu_msr_write(msr, low, high, &ret))
948 			ret = native_write_msr_safe(msr, low, high);
949 	}
950 
951 	return ret;
952 }
953 
954 static u64 xen_read_msr(unsigned int msr)
955 {
956 	/*
957 	 * This will silently swallow a #GP from RDMSR.  It may be worth
958 	 * changing that.
959 	 */
960 	int err;
961 
962 	return xen_read_msr_safe(msr, &err);
963 }
964 
965 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
966 {
967 	/*
968 	 * This will silently swallow a #GP from WRMSR.  It may be worth
969 	 * changing that.
970 	 */
971 	xen_write_msr_safe(msr, low, high);
972 }
973 
974 void xen_setup_shared_info(void)
975 {
976 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
977 
978 	HYPERVISOR_shared_info =
979 		(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
980 
981 	xen_setup_mfn_list_list();
982 
983 	if (system_state == SYSTEM_BOOTING) {
984 #ifndef CONFIG_SMP
985 		/*
986 		 * In UP this is as good a place as any to set up shared info.
987 		 * Limit this to boot only, at restore vcpu setup is done via
988 		 * xen_vcpu_restore().
989 		 */
990 		xen_setup_vcpu_info_placement();
991 #endif
992 		/*
993 		 * Now that shared info is set up we can start using routines
994 		 * that point to pvclock area.
995 		 */
996 		xen_init_time_ops();
997 	}
998 }
999 
1000 /* This is called once we have the cpu_possible_mask */
1001 void __ref xen_setup_vcpu_info_placement(void)
1002 {
1003 	int cpu;
1004 
1005 	for_each_possible_cpu(cpu) {
1006 		/* Set up direct vCPU id mapping for PV guests. */
1007 		per_cpu(xen_vcpu_id, cpu) = cpu;
1008 
1009 		/*
1010 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
1011 		 * falls back to the shared_info version for cpus
1012 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
1013 		 *
1014 		 * xen_cpu_up_prepare_pv() handles the rest by failing
1015 		 * them in hotplug.
1016 		 */
1017 		(void) xen_vcpu_setup(cpu);
1018 	}
1019 
1020 	/*
1021 	 * xen_vcpu_setup managed to place the vcpu_info within the
1022 	 * percpu area for all cpus, so make use of it.
1023 	 */
1024 	if (xen_have_vcpu_info_placement) {
1025 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1026 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1027 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1028 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1029 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1030 	}
1031 }
1032 
1033 static const struct pv_info xen_info __initconst = {
1034 	.shared_kernel_pmd = 0,
1035 
1036 #ifdef CONFIG_X86_64
1037 	.extra_user_64bit_cs = FLAT_USER_CS64,
1038 #endif
1039 	.name = "Xen",
1040 };
1041 
1042 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1043 	.cpuid = xen_cpuid,
1044 
1045 	.set_debugreg = xen_set_debugreg,
1046 	.get_debugreg = xen_get_debugreg,
1047 
1048 	.read_cr0 = xen_read_cr0,
1049 	.write_cr0 = xen_write_cr0,
1050 
1051 	.write_cr4 = xen_write_cr4,
1052 
1053 #ifdef CONFIG_X86_64
1054 	.read_cr8 = xen_read_cr8,
1055 	.write_cr8 = xen_write_cr8,
1056 #endif
1057 
1058 	.wbinvd = native_wbinvd,
1059 
1060 	.read_msr = xen_read_msr,
1061 	.write_msr = xen_write_msr,
1062 
1063 	.read_msr_safe = xen_read_msr_safe,
1064 	.write_msr_safe = xen_write_msr_safe,
1065 
1066 	.read_pmc = xen_read_pmc,
1067 
1068 	.iret = xen_iret,
1069 #ifdef CONFIG_X86_64
1070 	.usergs_sysret64 = xen_sysret64,
1071 #endif
1072 
1073 	.load_tr_desc = paravirt_nop,
1074 	.set_ldt = xen_set_ldt,
1075 	.load_gdt = xen_load_gdt,
1076 	.load_idt = xen_load_idt,
1077 	.load_tls = xen_load_tls,
1078 #ifdef CONFIG_X86_64
1079 	.load_gs_index = xen_load_gs_index,
1080 #endif
1081 
1082 	.alloc_ldt = xen_alloc_ldt,
1083 	.free_ldt = xen_free_ldt,
1084 
1085 	.store_tr = xen_store_tr,
1086 
1087 	.write_ldt_entry = xen_write_ldt_entry,
1088 	.write_gdt_entry = xen_write_gdt_entry,
1089 	.write_idt_entry = xen_write_idt_entry,
1090 	.load_sp0 = xen_load_sp0,
1091 
1092 	.set_iopl_mask = xen_set_iopl_mask,
1093 	.io_delay = xen_io_delay,
1094 
1095 	/* Xen takes care of %gs when switching to usermode for us */
1096 	.swapgs = paravirt_nop,
1097 
1098 	.start_context_switch = paravirt_start_context_switch,
1099 	.end_context_switch = xen_end_context_switch,
1100 };
1101 
1102 static void xen_restart(char *msg)
1103 {
1104 	xen_reboot(SHUTDOWN_reboot);
1105 }
1106 
1107 static void xen_machine_halt(void)
1108 {
1109 	xen_reboot(SHUTDOWN_poweroff);
1110 }
1111 
1112 static void xen_machine_power_off(void)
1113 {
1114 	if (pm_power_off)
1115 		pm_power_off();
1116 	xen_reboot(SHUTDOWN_poweroff);
1117 }
1118 
1119 static void xen_crash_shutdown(struct pt_regs *regs)
1120 {
1121 	xen_reboot(SHUTDOWN_crash);
1122 }
1123 
1124 static const struct machine_ops xen_machine_ops __initconst = {
1125 	.restart = xen_restart,
1126 	.halt = xen_machine_halt,
1127 	.power_off = xen_machine_power_off,
1128 	.shutdown = xen_machine_halt,
1129 	.crash_shutdown = xen_crash_shutdown,
1130 	.emergency_restart = xen_emergency_restart,
1131 };
1132 
1133 static unsigned char xen_get_nmi_reason(void)
1134 {
1135 	unsigned char reason = 0;
1136 
1137 	/* Construct a value which looks like it came from port 0x61. */
1138 	if (test_bit(_XEN_NMIREASON_io_error,
1139 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1140 		reason |= NMI_REASON_IOCHK;
1141 	if (test_bit(_XEN_NMIREASON_pci_serr,
1142 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1143 		reason |= NMI_REASON_SERR;
1144 
1145 	return reason;
1146 }
1147 
1148 static void __init xen_boot_params_init_edd(void)
1149 {
1150 #if IS_ENABLED(CONFIG_EDD)
1151 	struct xen_platform_op op;
1152 	struct edd_info *edd_info;
1153 	u32 *mbr_signature;
1154 	unsigned nr;
1155 	int ret;
1156 
1157 	edd_info = boot_params.eddbuf;
1158 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1159 
1160 	op.cmd = XENPF_firmware_info;
1161 
1162 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1163 	for (nr = 0; nr < EDDMAXNR; nr++) {
1164 		struct edd_info *info = edd_info + nr;
1165 
1166 		op.u.firmware_info.index = nr;
1167 		info->params.length = sizeof(info->params);
1168 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1169 				     &info->params);
1170 		ret = HYPERVISOR_platform_op(&op);
1171 		if (ret)
1172 			break;
1173 
1174 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1175 		C(device);
1176 		C(version);
1177 		C(interface_support);
1178 		C(legacy_max_cylinder);
1179 		C(legacy_max_head);
1180 		C(legacy_sectors_per_track);
1181 #undef C
1182 	}
1183 	boot_params.eddbuf_entries = nr;
1184 
1185 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1186 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1187 		op.u.firmware_info.index = nr;
1188 		ret = HYPERVISOR_platform_op(&op);
1189 		if (ret)
1190 			break;
1191 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1192 	}
1193 	boot_params.edd_mbr_sig_buf_entries = nr;
1194 #endif
1195 }
1196 
1197 /*
1198  * Set up the GDT and segment registers for -fstack-protector.  Until
1199  * we do this, we have to be careful not to call any stack-protected
1200  * function, which is most of the kernel.
1201  */
1202 static void xen_setup_gdt(int cpu)
1203 {
1204 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1205 	pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1206 
1207 	setup_stack_canary_segment(0);
1208 	switch_to_new_gdt(0);
1209 
1210 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1211 	pv_cpu_ops.load_gdt = xen_load_gdt;
1212 }
1213 
1214 static void __init xen_dom0_set_legacy_features(void)
1215 {
1216 	x86_platform.legacy.rtc = 1;
1217 }
1218 
1219 /* First C function to be called on Xen boot */
1220 asmlinkage __visible void __init xen_start_kernel(void)
1221 {
1222 	struct physdev_set_iopl set_iopl;
1223 	unsigned long initrd_start = 0;
1224 	int rc;
1225 
1226 	if (!xen_start_info)
1227 		return;
1228 
1229 	xen_domain_type = XEN_PV_DOMAIN;
1230 
1231 	xen_setup_features();
1232 
1233 	xen_setup_machphys_mapping();
1234 
1235 	/* Install Xen paravirt ops */
1236 	pv_info = xen_info;
1237 	pv_init_ops.patch = paravirt_patch_default;
1238 	pv_cpu_ops = xen_cpu_ops;
1239 
1240 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1241 
1242 	x86_init.resources.memory_setup = xen_memory_setup;
1243 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1244 	x86_init.oem.arch_setup = xen_arch_setup;
1245 	x86_init.oem.banner = xen_banner;
1246 
1247 	/*
1248 	 * Set up some pagetable state before starting to set any ptes.
1249 	 */
1250 
1251 	xen_init_mmu_ops();
1252 
1253 	/* Prevent unwanted bits from being set in PTEs. */
1254 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1255 
1256 	/*
1257 	 * Prevent page tables from being allocated in highmem, even
1258 	 * if CONFIG_HIGHPTE is enabled.
1259 	 */
1260 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1261 
1262 	/* Work out if we support NX */
1263 	get_cpu_cap(&boot_cpu_data);
1264 	x86_configure_nx();
1265 
1266 	/* Get mfn list */
1267 	xen_build_dynamic_phys_to_machine();
1268 
1269 	/*
1270 	 * Set up kernel GDT and segment registers, mainly so that
1271 	 * -fstack-protector code can be executed.
1272 	 */
1273 	xen_setup_gdt(0);
1274 
1275 	xen_init_irq_ops();
1276 
1277 	/* Let's presume PV guests always boot on vCPU with id 0. */
1278 	per_cpu(xen_vcpu_id, 0) = 0;
1279 
1280 	/*
1281 	 * Setup xen_vcpu early because idt_setup_early_handler needs it for
1282 	 * local_irq_disable(), irqs_disabled().
1283 	 *
1284 	 * Don't do the full vcpu_info placement stuff until we have
1285 	 * the cpu_possible_mask and a non-dummy shared_info.
1286 	 */
1287 	xen_vcpu_info_reset(0);
1288 
1289 	idt_setup_early_handler();
1290 
1291 	xen_init_capabilities();
1292 
1293 #ifdef CONFIG_X86_LOCAL_APIC
1294 	/*
1295 	 * set up the basic apic ops.
1296 	 */
1297 	xen_init_apic();
1298 #endif
1299 
1300 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1301 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1302 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1303 	}
1304 
1305 	machine_ops = xen_machine_ops;
1306 
1307 	/*
1308 	 * The only reliable way to retain the initial address of the
1309 	 * percpu gdt_page is to remember it here, so we can go and
1310 	 * mark it RW later, when the initial percpu area is freed.
1311 	 */
1312 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1313 
1314 	xen_smp_init();
1315 
1316 #ifdef CONFIG_ACPI_NUMA
1317 	/*
1318 	 * The pages we from Xen are not related to machine pages, so
1319 	 * any NUMA information the kernel tries to get from ACPI will
1320 	 * be meaningless.  Prevent it from trying.
1321 	 */
1322 	acpi_numa = -1;
1323 #endif
1324 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1325 
1326 	local_irq_disable();
1327 	early_boot_irqs_disabled = true;
1328 
1329 	xen_raw_console_write("mapping kernel into physical memory\n");
1330 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1331 				   xen_start_info->nr_pages);
1332 	xen_reserve_special_pages();
1333 
1334 	/* keep using Xen gdt for now; no urgent need to change it */
1335 
1336 #ifdef CONFIG_X86_32
1337 	pv_info.kernel_rpl = 1;
1338 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1339 		pv_info.kernel_rpl = 0;
1340 #else
1341 	pv_info.kernel_rpl = 0;
1342 #endif
1343 	/* set the limit of our address space */
1344 	xen_reserve_top();
1345 
1346 	/*
1347 	 * We used to do this in xen_arch_setup, but that is too late
1348 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1349 	 * early_amd_init which pokes 0xcf8 port.
1350 	 */
1351 	set_iopl.iopl = 1;
1352 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1353 	if (rc != 0)
1354 		xen_raw_printk("physdev_op failed %d\n", rc);
1355 
1356 #ifdef CONFIG_X86_32
1357 	/* set up basic CPUID stuff */
1358 	cpu_detect(&new_cpu_data);
1359 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1360 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1361 #endif
1362 
1363 	if (xen_start_info->mod_start) {
1364 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1365 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1366 	    else
1367 		initrd_start = __pa(xen_start_info->mod_start);
1368 	}
1369 
1370 	/* Poke various useful things into boot_params */
1371 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1372 	boot_params.hdr.ramdisk_image = initrd_start;
1373 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1374 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1375 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1376 
1377 	if (!xen_initial_domain()) {
1378 		add_preferred_console("xenboot", 0, NULL);
1379 		add_preferred_console("tty", 0, NULL);
1380 		add_preferred_console("hvc", 0, NULL);
1381 		if (pci_xen)
1382 			x86_init.pci.arch_init = pci_xen_init;
1383 	} else {
1384 		const struct dom0_vga_console_info *info =
1385 			(void *)((char *)xen_start_info +
1386 				 xen_start_info->console.dom0.info_off);
1387 		struct xen_platform_op op = {
1388 			.cmd = XENPF_firmware_info,
1389 			.interface_version = XENPF_INTERFACE_VERSION,
1390 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1391 		};
1392 
1393 		x86_platform.set_legacy_features =
1394 				xen_dom0_set_legacy_features;
1395 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1396 		xen_start_info->console.domU.mfn = 0;
1397 		xen_start_info->console.domU.evtchn = 0;
1398 
1399 		if (HYPERVISOR_platform_op(&op) == 0)
1400 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1401 
1402 		/* Make sure ACS will be enabled */
1403 		pci_request_acs();
1404 
1405 		xen_acpi_sleep_register();
1406 
1407 		/* Avoid searching for BIOS MP tables */
1408 		x86_init.mpparse.find_smp_config = x86_init_noop;
1409 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1410 
1411 		xen_boot_params_init_edd();
1412 	}
1413 #ifdef CONFIG_PCI
1414 	/* PCI BIOS service won't work from a PV guest. */
1415 	pci_probe &= ~PCI_PROBE_BIOS;
1416 #endif
1417 	xen_raw_console_write("about to get started...\n");
1418 
1419 	/* We need this for printk timestamps */
1420 	xen_setup_runstate_info(0);
1421 
1422 	xen_efi_init();
1423 
1424 	/* Start the world */
1425 #ifdef CONFIG_X86_32
1426 	i386_start_kernel();
1427 #else
1428 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1429 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1430 #endif
1431 }
1432 
1433 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1434 {
1435 	int rc;
1436 
1437 	if (per_cpu(xen_vcpu, cpu) == NULL)
1438 		return -ENODEV;
1439 
1440 	xen_setup_timer(cpu);
1441 
1442 	rc = xen_smp_intr_init(cpu);
1443 	if (rc) {
1444 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1445 		     cpu, rc);
1446 		return rc;
1447 	}
1448 
1449 	rc = xen_smp_intr_init_pv(cpu);
1450 	if (rc) {
1451 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1452 		     cpu, rc);
1453 		return rc;
1454 	}
1455 
1456 	return 0;
1457 }
1458 
1459 static int xen_cpu_dead_pv(unsigned int cpu)
1460 {
1461 	xen_smp_intr_free(cpu);
1462 	xen_smp_intr_free_pv(cpu);
1463 
1464 	xen_teardown_timer(cpu);
1465 
1466 	return 0;
1467 }
1468 
1469 static uint32_t __init xen_platform_pv(void)
1470 {
1471 	if (xen_pv_domain())
1472 		return xen_cpuid_base();
1473 
1474 	return 0;
1475 }
1476 
1477 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1478 	.name                   = "Xen PV",
1479 	.detect                 = xen_platform_pv,
1480 	.type			= X86_HYPER_XEN_PV,
1481 	.runtime.pin_vcpu       = xen_pin_vcpu,
1482 };
1483