xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 020c5260)
1 /*
2  * Core of Xen paravirt_ops implementation.
3  *
4  * This file contains the xen_paravirt_ops structure itself, and the
5  * implementations for:
6  * - privileged instructions
7  * - interrupt flags
8  * - segment operations
9  * - booting and setup
10  *
11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12  */
13 
14 #include <linux/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/export.h>
27 #include <linux/mm.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
36 
37 #include <xen/xen.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
47 #include <xen/page.h>
48 #include <xen/hvc-console.h>
49 #include <xen/acpi.h>
50 
51 #include <asm/paravirt.h>
52 #include <asm/apic.h>
53 #include <asm/page.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
64 #include <asm/desc.h>
65 #include <asm/pgalloc.h>
66 #include <asm/pgtable.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/stackprotector.h>
70 #include <asm/hypervisor.h>
71 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
74 #include <asm/cpu.h>
75 
76 #ifdef CONFIG_ACPI
77 #include <linux/acpi.h>
78 #include <asm/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
82 #endif
83 
84 #include "xen-ops.h"
85 #include "mmu.h"
86 #include "smp.h"
87 #include "multicalls.h"
88 #include "pmu.h"
89 
90 void *xen_initial_gdt;
91 
92 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
93 
94 static int xen_cpu_up_prepare_pv(unsigned int cpu);
95 static int xen_cpu_dead_pv(unsigned int cpu);
96 
97 struct tls_descs {
98 	struct desc_struct desc[3];
99 };
100 
101 /*
102  * Updating the 3 TLS descriptors in the GDT on every task switch is
103  * surprisingly expensive so we avoid updating them if they haven't
104  * changed.  Since Xen writes different descriptors than the one
105  * passed in the update_descriptor hypercall we keep shadow copies to
106  * compare against.
107  */
108 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
109 
110 /*
111  * On restore, set the vcpu placement up again.
112  * If it fails, then we're in a bad state, since
113  * we can't back out from using it...
114  */
115 void xen_vcpu_restore(void)
116 {
117 	int cpu;
118 
119 	for_each_possible_cpu(cpu) {
120 		bool other_cpu = (cpu != smp_processor_id());
121 		bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
122 						NULL);
123 
124 		if (other_cpu && is_up &&
125 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
126 			BUG();
127 
128 		xen_setup_runstate_info(cpu);
129 
130 		if (xen_have_vcpu_info_placement)
131 			xen_vcpu_setup(cpu);
132 
133 		if (other_cpu && is_up &&
134 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
135 			BUG();
136 	}
137 }
138 
139 static void __init xen_banner(void)
140 {
141 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
142 	struct xen_extraversion extra;
143 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
144 
145 	pr_info("Booting paravirtualized kernel %son %s\n",
146 		xen_feature(XENFEAT_auto_translated_physmap) ?
147 			"with PVH extensions " : "", pv_info.name);
148 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
149 	       version >> 16, version & 0xffff, extra.extraversion,
150 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
151 }
152 /* Check if running on Xen version (major, minor) or later */
153 bool
154 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
155 {
156 	unsigned int version;
157 
158 	if (!xen_domain())
159 		return false;
160 
161 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
162 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
163 		((version >> 16) > major))
164 		return true;
165 	return false;
166 }
167 
168 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
169 static __read_mostly unsigned int cpuid_leaf5_edx_val;
170 
171 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
172 		      unsigned int *cx, unsigned int *dx)
173 {
174 	unsigned maskebx = ~0;
175 
176 	/*
177 	 * Mask out inconvenient features, to try and disable as many
178 	 * unsupported kernel subsystems as possible.
179 	 */
180 	switch (*ax) {
181 	case CPUID_MWAIT_LEAF:
182 		/* Synthesize the values.. */
183 		*ax = 0;
184 		*bx = 0;
185 		*cx = cpuid_leaf5_ecx_val;
186 		*dx = cpuid_leaf5_edx_val;
187 		return;
188 
189 	case 0xb:
190 		/* Suppress extended topology stuff */
191 		maskebx = 0;
192 		break;
193 	}
194 
195 	asm(XEN_EMULATE_PREFIX "cpuid"
196 		: "=a" (*ax),
197 		  "=b" (*bx),
198 		  "=c" (*cx),
199 		  "=d" (*dx)
200 		: "0" (*ax), "2" (*cx));
201 
202 	*bx &= maskebx;
203 }
204 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
205 
206 static bool __init xen_check_mwait(void)
207 {
208 #ifdef CONFIG_ACPI
209 	struct xen_platform_op op = {
210 		.cmd			= XENPF_set_processor_pminfo,
211 		.u.set_pminfo.id	= -1,
212 		.u.set_pminfo.type	= XEN_PM_PDC,
213 	};
214 	uint32_t buf[3];
215 	unsigned int ax, bx, cx, dx;
216 	unsigned int mwait_mask;
217 
218 	/* We need to determine whether it is OK to expose the MWAIT
219 	 * capability to the kernel to harvest deeper than C3 states from ACPI
220 	 * _CST using the processor_harvest_xen.c module. For this to work, we
221 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
222 	 * checks against). The hypervisor won't expose the MWAIT flag because
223 	 * it would break backwards compatibility; so we will find out directly
224 	 * from the hardware and hypercall.
225 	 */
226 	if (!xen_initial_domain())
227 		return false;
228 
229 	/*
230 	 * When running under platform earlier than Xen4.2, do not expose
231 	 * mwait, to avoid the risk of loading native acpi pad driver
232 	 */
233 	if (!xen_running_on_version_or_later(4, 2))
234 		return false;
235 
236 	ax = 1;
237 	cx = 0;
238 
239 	native_cpuid(&ax, &bx, &cx, &dx);
240 
241 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
242 		     (1 << (X86_FEATURE_MWAIT % 32));
243 
244 	if ((cx & mwait_mask) != mwait_mask)
245 		return false;
246 
247 	/* We need to emulate the MWAIT_LEAF and for that we need both
248 	 * ecx and edx. The hypercall provides only partial information.
249 	 */
250 
251 	ax = CPUID_MWAIT_LEAF;
252 	bx = 0;
253 	cx = 0;
254 	dx = 0;
255 
256 	native_cpuid(&ax, &bx, &cx, &dx);
257 
258 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
259 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
260 	 */
261 	buf[0] = ACPI_PDC_REVISION_ID;
262 	buf[1] = 1;
263 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
264 
265 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
266 
267 	if ((HYPERVISOR_platform_op(&op) == 0) &&
268 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
269 		cpuid_leaf5_ecx_val = cx;
270 		cpuid_leaf5_edx_val = dx;
271 	}
272 	return true;
273 #else
274 	return false;
275 #endif
276 }
277 
278 static bool __init xen_check_xsave(void)
279 {
280 	unsigned int cx, xsave_mask;
281 
282 	cx = cpuid_ecx(1);
283 
284 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
285 		     (1 << (X86_FEATURE_OSXSAVE % 32));
286 
287 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
288 	return (cx & xsave_mask) == xsave_mask;
289 }
290 
291 static void __init xen_init_capabilities(void)
292 {
293 	setup_force_cpu_cap(X86_FEATURE_XENPV);
294 	setup_clear_cpu_cap(X86_FEATURE_DCA);
295 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
296 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
297 	setup_clear_cpu_cap(X86_FEATURE_ACC);
298 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
299 
300 	if (!xen_initial_domain())
301 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
302 
303 	if (xen_check_mwait())
304 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
305 	else
306 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
307 
308 	if (!xen_check_xsave()) {
309 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
310 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
311 	}
312 }
313 
314 static void xen_set_debugreg(int reg, unsigned long val)
315 {
316 	HYPERVISOR_set_debugreg(reg, val);
317 }
318 
319 static unsigned long xen_get_debugreg(int reg)
320 {
321 	return HYPERVISOR_get_debugreg(reg);
322 }
323 
324 static void xen_end_context_switch(struct task_struct *next)
325 {
326 	xen_mc_flush();
327 	paravirt_end_context_switch(next);
328 }
329 
330 static unsigned long xen_store_tr(void)
331 {
332 	return 0;
333 }
334 
335 /*
336  * Set the page permissions for a particular virtual address.  If the
337  * address is a vmalloc mapping (or other non-linear mapping), then
338  * find the linear mapping of the page and also set its protections to
339  * match.
340  */
341 static void set_aliased_prot(void *v, pgprot_t prot)
342 {
343 	int level;
344 	pte_t *ptep;
345 	pte_t pte;
346 	unsigned long pfn;
347 	struct page *page;
348 	unsigned char dummy;
349 
350 	ptep = lookup_address((unsigned long)v, &level);
351 	BUG_ON(ptep == NULL);
352 
353 	pfn = pte_pfn(*ptep);
354 	page = pfn_to_page(pfn);
355 
356 	pte = pfn_pte(pfn, prot);
357 
358 	/*
359 	 * Careful: update_va_mapping() will fail if the virtual address
360 	 * we're poking isn't populated in the page tables.  We don't
361 	 * need to worry about the direct map (that's always in the page
362 	 * tables), but we need to be careful about vmap space.  In
363 	 * particular, the top level page table can lazily propagate
364 	 * entries between processes, so if we've switched mms since we
365 	 * vmapped the target in the first place, we might not have the
366 	 * top-level page table entry populated.
367 	 *
368 	 * We disable preemption because we want the same mm active when
369 	 * we probe the target and when we issue the hypercall.  We'll
370 	 * have the same nominal mm, but if we're a kernel thread, lazy
371 	 * mm dropping could change our pgd.
372 	 *
373 	 * Out of an abundance of caution, this uses __get_user() to fault
374 	 * in the target address just in case there's some obscure case
375 	 * in which the target address isn't readable.
376 	 */
377 
378 	preempt_disable();
379 
380 	probe_kernel_read(&dummy, v, 1);
381 
382 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
383 		BUG();
384 
385 	if (!PageHighMem(page)) {
386 		void *av = __va(PFN_PHYS(pfn));
387 
388 		if (av != v)
389 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
390 				BUG();
391 	} else
392 		kmap_flush_unused();
393 
394 	preempt_enable();
395 }
396 
397 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
398 {
399 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
400 	int i;
401 
402 	/*
403 	 * We need to mark the all aliases of the LDT pages RO.  We
404 	 * don't need to call vm_flush_aliases(), though, since that's
405 	 * only responsible for flushing aliases out the TLBs, not the
406 	 * page tables, and Xen will flush the TLB for us if needed.
407 	 *
408 	 * To avoid confusing future readers: none of this is necessary
409 	 * to load the LDT.  The hypervisor only checks this when the
410 	 * LDT is faulted in due to subsequent descriptor access.
411 	 */
412 
413 	for (i = 0; i < entries; i += entries_per_page)
414 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
415 }
416 
417 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
418 {
419 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
420 	int i;
421 
422 	for (i = 0; i < entries; i += entries_per_page)
423 		set_aliased_prot(ldt + i, PAGE_KERNEL);
424 }
425 
426 static void xen_set_ldt(const void *addr, unsigned entries)
427 {
428 	struct mmuext_op *op;
429 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
430 
431 	trace_xen_cpu_set_ldt(addr, entries);
432 
433 	op = mcs.args;
434 	op->cmd = MMUEXT_SET_LDT;
435 	op->arg1.linear_addr = (unsigned long)addr;
436 	op->arg2.nr_ents = entries;
437 
438 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
439 
440 	xen_mc_issue(PARAVIRT_LAZY_CPU);
441 }
442 
443 static void xen_load_gdt(const struct desc_ptr *dtr)
444 {
445 	unsigned long va = dtr->address;
446 	unsigned int size = dtr->size + 1;
447 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
448 	unsigned long frames[pages];
449 	int f;
450 
451 	/*
452 	 * A GDT can be up to 64k in size, which corresponds to 8192
453 	 * 8-byte entries, or 16 4k pages..
454 	 */
455 
456 	BUG_ON(size > 65536);
457 	BUG_ON(va & ~PAGE_MASK);
458 
459 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
460 		int level;
461 		pte_t *ptep;
462 		unsigned long pfn, mfn;
463 		void *virt;
464 
465 		/*
466 		 * The GDT is per-cpu and is in the percpu data area.
467 		 * That can be virtually mapped, so we need to do a
468 		 * page-walk to get the underlying MFN for the
469 		 * hypercall.  The page can also be in the kernel's
470 		 * linear range, so we need to RO that mapping too.
471 		 */
472 		ptep = lookup_address(va, &level);
473 		BUG_ON(ptep == NULL);
474 
475 		pfn = pte_pfn(*ptep);
476 		mfn = pfn_to_mfn(pfn);
477 		virt = __va(PFN_PHYS(pfn));
478 
479 		frames[f] = mfn;
480 
481 		make_lowmem_page_readonly((void *)va);
482 		make_lowmem_page_readonly(virt);
483 	}
484 
485 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
486 		BUG();
487 }
488 
489 /*
490  * load_gdt for early boot, when the gdt is only mapped once
491  */
492 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
493 {
494 	unsigned long va = dtr->address;
495 	unsigned int size = dtr->size + 1;
496 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
497 	unsigned long frames[pages];
498 	int f;
499 
500 	/*
501 	 * A GDT can be up to 64k in size, which corresponds to 8192
502 	 * 8-byte entries, or 16 4k pages..
503 	 */
504 
505 	BUG_ON(size > 65536);
506 	BUG_ON(va & ~PAGE_MASK);
507 
508 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
509 		pte_t pte;
510 		unsigned long pfn, mfn;
511 
512 		pfn = virt_to_pfn(va);
513 		mfn = pfn_to_mfn(pfn);
514 
515 		pte = pfn_pte(pfn, PAGE_KERNEL_RO);
516 
517 		if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
518 			BUG();
519 
520 		frames[f] = mfn;
521 	}
522 
523 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
524 		BUG();
525 }
526 
527 static inline bool desc_equal(const struct desc_struct *d1,
528 			      const struct desc_struct *d2)
529 {
530 	return d1->a == d2->a && d1->b == d2->b;
531 }
532 
533 static void load_TLS_descriptor(struct thread_struct *t,
534 				unsigned int cpu, unsigned int i)
535 {
536 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
537 	struct desc_struct *gdt;
538 	xmaddr_t maddr;
539 	struct multicall_space mc;
540 
541 	if (desc_equal(shadow, &t->tls_array[i]))
542 		return;
543 
544 	*shadow = t->tls_array[i];
545 
546 	gdt = get_cpu_gdt_rw(cpu);
547 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
548 	mc = __xen_mc_entry(0);
549 
550 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
551 }
552 
553 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
554 {
555 	/*
556 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
557 	 * and lazy gs handling is enabled, it means we're in a
558 	 * context switch, and %gs has just been saved.  This means we
559 	 * can zero it out to prevent faults on exit from the
560 	 * hypervisor if the next process has no %gs.  Either way, it
561 	 * has been saved, and the new value will get loaded properly.
562 	 * This will go away as soon as Xen has been modified to not
563 	 * save/restore %gs for normal hypercalls.
564 	 *
565 	 * On x86_64, this hack is not used for %gs, because gs points
566 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
567 	 * must not zero %gs on x86_64
568 	 *
569 	 * For x86_64, we need to zero %fs, otherwise we may get an
570 	 * exception between the new %fs descriptor being loaded and
571 	 * %fs being effectively cleared at __switch_to().
572 	 */
573 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
574 #ifdef CONFIG_X86_32
575 		lazy_load_gs(0);
576 #else
577 		loadsegment(fs, 0);
578 #endif
579 	}
580 
581 	xen_mc_batch();
582 
583 	load_TLS_descriptor(t, cpu, 0);
584 	load_TLS_descriptor(t, cpu, 1);
585 	load_TLS_descriptor(t, cpu, 2);
586 
587 	xen_mc_issue(PARAVIRT_LAZY_CPU);
588 }
589 
590 #ifdef CONFIG_X86_64
591 static void xen_load_gs_index(unsigned int idx)
592 {
593 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
594 		BUG();
595 }
596 #endif
597 
598 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
599 				const void *ptr)
600 {
601 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
602 	u64 entry = *(u64 *)ptr;
603 
604 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
605 
606 	preempt_disable();
607 
608 	xen_mc_flush();
609 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
610 		BUG();
611 
612 	preempt_enable();
613 }
614 
615 static int cvt_gate_to_trap(int vector, const gate_desc *val,
616 			    struct trap_info *info)
617 {
618 	unsigned long addr;
619 
620 	if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
621 		return 0;
622 
623 	info->vector = vector;
624 
625 	addr = gate_offset(*val);
626 #ifdef CONFIG_X86_64
627 	/*
628 	 * Look for known traps using IST, and substitute them
629 	 * appropriately.  The debugger ones are the only ones we care
630 	 * about.  Xen will handle faults like double_fault,
631 	 * so we should never see them.  Warn if
632 	 * there's an unexpected IST-using fault handler.
633 	 */
634 	if (addr == (unsigned long)debug)
635 		addr = (unsigned long)xen_debug;
636 	else if (addr == (unsigned long)int3)
637 		addr = (unsigned long)xen_int3;
638 	else if (addr == (unsigned long)stack_segment)
639 		addr = (unsigned long)xen_stack_segment;
640 	else if (addr == (unsigned long)double_fault) {
641 		/* Don't need to handle these */
642 		return 0;
643 #ifdef CONFIG_X86_MCE
644 	} else if (addr == (unsigned long)machine_check) {
645 		/*
646 		 * when xen hypervisor inject vMCE to guest,
647 		 * use native mce handler to handle it
648 		 */
649 		;
650 #endif
651 	} else if (addr == (unsigned long)nmi)
652 		/*
653 		 * Use the native version as well.
654 		 */
655 		;
656 	else {
657 		/* Some other trap using IST? */
658 		if (WARN_ON(val->ist != 0))
659 			return 0;
660 	}
661 #endif	/* CONFIG_X86_64 */
662 	info->address = addr;
663 
664 	info->cs = gate_segment(*val);
665 	info->flags = val->dpl;
666 	/* interrupt gates clear IF */
667 	if (val->type == GATE_INTERRUPT)
668 		info->flags |= 1 << 2;
669 
670 	return 1;
671 }
672 
673 /* Locations of each CPU's IDT */
674 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
675 
676 /* Set an IDT entry.  If the entry is part of the current IDT, then
677    also update Xen. */
678 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
679 {
680 	unsigned long p = (unsigned long)&dt[entrynum];
681 	unsigned long start, end;
682 
683 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
684 
685 	preempt_disable();
686 
687 	start = __this_cpu_read(idt_desc.address);
688 	end = start + __this_cpu_read(idt_desc.size) + 1;
689 
690 	xen_mc_flush();
691 
692 	native_write_idt_entry(dt, entrynum, g);
693 
694 	if (p >= start && (p + 8) <= end) {
695 		struct trap_info info[2];
696 
697 		info[1].address = 0;
698 
699 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
700 			if (HYPERVISOR_set_trap_table(info))
701 				BUG();
702 	}
703 
704 	preempt_enable();
705 }
706 
707 static void xen_convert_trap_info(const struct desc_ptr *desc,
708 				  struct trap_info *traps)
709 {
710 	unsigned in, out, count;
711 
712 	count = (desc->size+1) / sizeof(gate_desc);
713 	BUG_ON(count > 256);
714 
715 	for (in = out = 0; in < count; in++) {
716 		gate_desc *entry = (gate_desc *)(desc->address) + in;
717 
718 		if (cvt_gate_to_trap(in, entry, &traps[out]))
719 			out++;
720 	}
721 	traps[out].address = 0;
722 }
723 
724 void xen_copy_trap_info(struct trap_info *traps)
725 {
726 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
727 
728 	xen_convert_trap_info(desc, traps);
729 }
730 
731 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
732    hold a spinlock to protect the static traps[] array (static because
733    it avoids allocation, and saves stack space). */
734 static void xen_load_idt(const struct desc_ptr *desc)
735 {
736 	static DEFINE_SPINLOCK(lock);
737 	static struct trap_info traps[257];
738 
739 	trace_xen_cpu_load_idt(desc);
740 
741 	spin_lock(&lock);
742 
743 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
744 
745 	xen_convert_trap_info(desc, traps);
746 
747 	xen_mc_flush();
748 	if (HYPERVISOR_set_trap_table(traps))
749 		BUG();
750 
751 	spin_unlock(&lock);
752 }
753 
754 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
755    they're handled differently. */
756 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
757 				const void *desc, int type)
758 {
759 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
760 
761 	preempt_disable();
762 
763 	switch (type) {
764 	case DESC_LDT:
765 	case DESC_TSS:
766 		/* ignore */
767 		break;
768 
769 	default: {
770 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
771 
772 		xen_mc_flush();
773 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
774 			BUG();
775 	}
776 
777 	}
778 
779 	preempt_enable();
780 }
781 
782 /*
783  * Version of write_gdt_entry for use at early boot-time needed to
784  * update an entry as simply as possible.
785  */
786 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
787 					    const void *desc, int type)
788 {
789 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
790 
791 	switch (type) {
792 	case DESC_LDT:
793 	case DESC_TSS:
794 		/* ignore */
795 		break;
796 
797 	default: {
798 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
799 
800 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
801 			dt[entry] = *(struct desc_struct *)desc;
802 	}
803 
804 	}
805 }
806 
807 static void xen_load_sp0(struct tss_struct *tss,
808 			 struct thread_struct *thread)
809 {
810 	struct multicall_space mcs;
811 
812 	mcs = xen_mc_entry(0);
813 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
814 	xen_mc_issue(PARAVIRT_LAZY_CPU);
815 	tss->x86_tss.sp0 = thread->sp0;
816 }
817 
818 void xen_set_iopl_mask(unsigned mask)
819 {
820 	struct physdev_set_iopl set_iopl;
821 
822 	/* Force the change at ring 0. */
823 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
824 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
825 }
826 
827 static void xen_io_delay(void)
828 {
829 }
830 
831 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
832 
833 static unsigned long xen_read_cr0(void)
834 {
835 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
836 
837 	if (unlikely(cr0 == 0)) {
838 		cr0 = native_read_cr0();
839 		this_cpu_write(xen_cr0_value, cr0);
840 	}
841 
842 	return cr0;
843 }
844 
845 static void xen_write_cr0(unsigned long cr0)
846 {
847 	struct multicall_space mcs;
848 
849 	this_cpu_write(xen_cr0_value, cr0);
850 
851 	/* Only pay attention to cr0.TS; everything else is
852 	   ignored. */
853 	mcs = xen_mc_entry(0);
854 
855 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
856 
857 	xen_mc_issue(PARAVIRT_LAZY_CPU);
858 }
859 
860 static void xen_write_cr4(unsigned long cr4)
861 {
862 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
863 
864 	native_write_cr4(cr4);
865 }
866 #ifdef CONFIG_X86_64
867 static inline unsigned long xen_read_cr8(void)
868 {
869 	return 0;
870 }
871 static inline void xen_write_cr8(unsigned long val)
872 {
873 	BUG_ON(val);
874 }
875 #endif
876 
877 static u64 xen_read_msr_safe(unsigned int msr, int *err)
878 {
879 	u64 val;
880 
881 	if (pmu_msr_read(msr, &val, err))
882 		return val;
883 
884 	val = native_read_msr_safe(msr, err);
885 	switch (msr) {
886 	case MSR_IA32_APICBASE:
887 #ifdef CONFIG_X86_X2APIC
888 		if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
889 #endif
890 			val &= ~X2APIC_ENABLE;
891 		break;
892 	}
893 	return val;
894 }
895 
896 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
897 {
898 	int ret;
899 
900 	ret = 0;
901 
902 	switch (msr) {
903 #ifdef CONFIG_X86_64
904 		unsigned which;
905 		u64 base;
906 
907 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
908 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
909 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
910 
911 	set:
912 		base = ((u64)high << 32) | low;
913 		if (HYPERVISOR_set_segment_base(which, base) != 0)
914 			ret = -EIO;
915 		break;
916 #endif
917 
918 	case MSR_STAR:
919 	case MSR_CSTAR:
920 	case MSR_LSTAR:
921 	case MSR_SYSCALL_MASK:
922 	case MSR_IA32_SYSENTER_CS:
923 	case MSR_IA32_SYSENTER_ESP:
924 	case MSR_IA32_SYSENTER_EIP:
925 		/* Fast syscall setup is all done in hypercalls, so
926 		   these are all ignored.  Stub them out here to stop
927 		   Xen console noise. */
928 		break;
929 
930 	default:
931 		if (!pmu_msr_write(msr, low, high, &ret))
932 			ret = native_write_msr_safe(msr, low, high);
933 	}
934 
935 	return ret;
936 }
937 
938 static u64 xen_read_msr(unsigned int msr)
939 {
940 	/*
941 	 * This will silently swallow a #GP from RDMSR.  It may be worth
942 	 * changing that.
943 	 */
944 	int err;
945 
946 	return xen_read_msr_safe(msr, &err);
947 }
948 
949 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
950 {
951 	/*
952 	 * This will silently swallow a #GP from WRMSR.  It may be worth
953 	 * changing that.
954 	 */
955 	xen_write_msr_safe(msr, low, high);
956 }
957 
958 void xen_setup_shared_info(void)
959 {
960 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
961 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
962 			   xen_start_info->shared_info);
963 
964 		HYPERVISOR_shared_info =
965 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
966 	} else
967 		HYPERVISOR_shared_info =
968 			(struct shared_info *)__va(xen_start_info->shared_info);
969 
970 #ifndef CONFIG_SMP
971 	/* In UP this is as good a place as any to set up shared info */
972 	xen_setup_vcpu_info_placement();
973 #endif
974 
975 	xen_setup_mfn_list_list();
976 
977 	/*
978 	 * Now that shared info is set up we can start using routines that
979 	 * point to pvclock area.
980 	 */
981 	if (system_state == SYSTEM_BOOTING)
982 		xen_init_time_ops();
983 }
984 
985 /* This is called once we have the cpu_possible_mask */
986 void xen_setup_vcpu_info_placement(void)
987 {
988 	int cpu;
989 
990 	for_each_possible_cpu(cpu) {
991 		/* Set up direct vCPU id mapping for PV guests. */
992 		per_cpu(xen_vcpu_id, cpu) = cpu;
993 		xen_vcpu_setup(cpu);
994 	}
995 
996 	/*
997 	 * xen_vcpu_setup managed to place the vcpu_info within the
998 	 * percpu area for all cpus, so make use of it.
999 	 */
1000 	if (xen_have_vcpu_info_placement) {
1001 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1002 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1003 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1004 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1005 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1006 	}
1007 }
1008 
1009 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1010 			  unsigned long addr, unsigned len)
1011 {
1012 	char *start, *end, *reloc;
1013 	unsigned ret;
1014 
1015 	start = end = reloc = NULL;
1016 
1017 #define SITE(op, x)							\
1018 	case PARAVIRT_PATCH(op.x):					\
1019 	if (xen_have_vcpu_info_placement) {				\
1020 		start = (char *)xen_##x##_direct;			\
1021 		end = xen_##x##_direct_end;				\
1022 		reloc = xen_##x##_direct_reloc;				\
1023 	}								\
1024 	goto patch_site
1025 
1026 	switch (type) {
1027 		SITE(pv_irq_ops, irq_enable);
1028 		SITE(pv_irq_ops, irq_disable);
1029 		SITE(pv_irq_ops, save_fl);
1030 		SITE(pv_irq_ops, restore_fl);
1031 #undef SITE
1032 
1033 	patch_site:
1034 		if (start == NULL || (end-start) > len)
1035 			goto default_patch;
1036 
1037 		ret = paravirt_patch_insns(insnbuf, len, start, end);
1038 
1039 		/* Note: because reloc is assigned from something that
1040 		   appears to be an array, gcc assumes it's non-null,
1041 		   but doesn't know its relationship with start and
1042 		   end. */
1043 		if (reloc > start && reloc < end) {
1044 			int reloc_off = reloc - start;
1045 			long *relocp = (long *)(insnbuf + reloc_off);
1046 			long delta = start - (char *)addr;
1047 
1048 			*relocp += delta;
1049 		}
1050 		break;
1051 
1052 	default_patch:
1053 	default:
1054 		ret = paravirt_patch_default(type, clobbers, insnbuf,
1055 					     addr, len);
1056 		break;
1057 	}
1058 
1059 	return ret;
1060 }
1061 
1062 static const struct pv_info xen_info __initconst = {
1063 	.shared_kernel_pmd = 0,
1064 
1065 #ifdef CONFIG_X86_64
1066 	.extra_user_64bit_cs = FLAT_USER_CS64,
1067 #endif
1068 	.name = "Xen",
1069 };
1070 
1071 static const struct pv_init_ops xen_init_ops __initconst = {
1072 	.patch = xen_patch,
1073 };
1074 
1075 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1076 	.cpuid = xen_cpuid,
1077 
1078 	.set_debugreg = xen_set_debugreg,
1079 	.get_debugreg = xen_get_debugreg,
1080 
1081 	.read_cr0 = xen_read_cr0,
1082 	.write_cr0 = xen_write_cr0,
1083 
1084 	.read_cr4 = native_read_cr4,
1085 	.write_cr4 = xen_write_cr4,
1086 
1087 #ifdef CONFIG_X86_64
1088 	.read_cr8 = xen_read_cr8,
1089 	.write_cr8 = xen_write_cr8,
1090 #endif
1091 
1092 	.wbinvd = native_wbinvd,
1093 
1094 	.read_msr = xen_read_msr,
1095 	.write_msr = xen_write_msr,
1096 
1097 	.read_msr_safe = xen_read_msr_safe,
1098 	.write_msr_safe = xen_write_msr_safe,
1099 
1100 	.read_pmc = xen_read_pmc,
1101 
1102 	.iret = xen_iret,
1103 #ifdef CONFIG_X86_64
1104 	.usergs_sysret64 = xen_sysret64,
1105 #endif
1106 
1107 	.load_tr_desc = paravirt_nop,
1108 	.set_ldt = xen_set_ldt,
1109 	.load_gdt = xen_load_gdt,
1110 	.load_idt = xen_load_idt,
1111 	.load_tls = xen_load_tls,
1112 #ifdef CONFIG_X86_64
1113 	.load_gs_index = xen_load_gs_index,
1114 #endif
1115 
1116 	.alloc_ldt = xen_alloc_ldt,
1117 	.free_ldt = xen_free_ldt,
1118 
1119 	.store_idt = native_store_idt,
1120 	.store_tr = xen_store_tr,
1121 
1122 	.write_ldt_entry = xen_write_ldt_entry,
1123 	.write_gdt_entry = xen_write_gdt_entry,
1124 	.write_idt_entry = xen_write_idt_entry,
1125 	.load_sp0 = xen_load_sp0,
1126 
1127 	.set_iopl_mask = xen_set_iopl_mask,
1128 	.io_delay = xen_io_delay,
1129 
1130 	/* Xen takes care of %gs when switching to usermode for us */
1131 	.swapgs = paravirt_nop,
1132 
1133 	.start_context_switch = paravirt_start_context_switch,
1134 	.end_context_switch = xen_end_context_switch,
1135 };
1136 
1137 static void xen_restart(char *msg)
1138 {
1139 	xen_reboot(SHUTDOWN_reboot);
1140 }
1141 
1142 static void xen_machine_halt(void)
1143 {
1144 	xen_reboot(SHUTDOWN_poweroff);
1145 }
1146 
1147 static void xen_machine_power_off(void)
1148 {
1149 	if (pm_power_off)
1150 		pm_power_off();
1151 	xen_reboot(SHUTDOWN_poweroff);
1152 }
1153 
1154 static void xen_crash_shutdown(struct pt_regs *regs)
1155 {
1156 	xen_reboot(SHUTDOWN_crash);
1157 }
1158 
1159 static const struct machine_ops xen_machine_ops __initconst = {
1160 	.restart = xen_restart,
1161 	.halt = xen_machine_halt,
1162 	.power_off = xen_machine_power_off,
1163 	.shutdown = xen_machine_halt,
1164 	.crash_shutdown = xen_crash_shutdown,
1165 	.emergency_restart = xen_emergency_restart,
1166 };
1167 
1168 static unsigned char xen_get_nmi_reason(void)
1169 {
1170 	unsigned char reason = 0;
1171 
1172 	/* Construct a value which looks like it came from port 0x61. */
1173 	if (test_bit(_XEN_NMIREASON_io_error,
1174 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1175 		reason |= NMI_REASON_IOCHK;
1176 	if (test_bit(_XEN_NMIREASON_pci_serr,
1177 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1178 		reason |= NMI_REASON_SERR;
1179 
1180 	return reason;
1181 }
1182 
1183 static void __init xen_boot_params_init_edd(void)
1184 {
1185 #if IS_ENABLED(CONFIG_EDD)
1186 	struct xen_platform_op op;
1187 	struct edd_info *edd_info;
1188 	u32 *mbr_signature;
1189 	unsigned nr;
1190 	int ret;
1191 
1192 	edd_info = boot_params.eddbuf;
1193 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1194 
1195 	op.cmd = XENPF_firmware_info;
1196 
1197 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1198 	for (nr = 0; nr < EDDMAXNR; nr++) {
1199 		struct edd_info *info = edd_info + nr;
1200 
1201 		op.u.firmware_info.index = nr;
1202 		info->params.length = sizeof(info->params);
1203 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1204 				     &info->params);
1205 		ret = HYPERVISOR_platform_op(&op);
1206 		if (ret)
1207 			break;
1208 
1209 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1210 		C(device);
1211 		C(version);
1212 		C(interface_support);
1213 		C(legacy_max_cylinder);
1214 		C(legacy_max_head);
1215 		C(legacy_sectors_per_track);
1216 #undef C
1217 	}
1218 	boot_params.eddbuf_entries = nr;
1219 
1220 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1221 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1222 		op.u.firmware_info.index = nr;
1223 		ret = HYPERVISOR_platform_op(&op);
1224 		if (ret)
1225 			break;
1226 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1227 	}
1228 	boot_params.edd_mbr_sig_buf_entries = nr;
1229 #endif
1230 }
1231 
1232 /*
1233  * Set up the GDT and segment registers for -fstack-protector.  Until
1234  * we do this, we have to be careful not to call any stack-protected
1235  * function, which is most of the kernel.
1236  */
1237 static void xen_setup_gdt(int cpu)
1238 {
1239 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1240 	pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1241 
1242 	setup_stack_canary_segment(0);
1243 	switch_to_new_gdt(0);
1244 
1245 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1246 	pv_cpu_ops.load_gdt = xen_load_gdt;
1247 }
1248 
1249 static void __init xen_dom0_set_legacy_features(void)
1250 {
1251 	x86_platform.legacy.rtc = 1;
1252 }
1253 
1254 /* First C function to be called on Xen boot */
1255 asmlinkage __visible void __init xen_start_kernel(void)
1256 {
1257 	struct physdev_set_iopl set_iopl;
1258 	unsigned long initrd_start = 0;
1259 	int rc;
1260 
1261 	if (!xen_start_info)
1262 		return;
1263 
1264 	xen_domain_type = XEN_PV_DOMAIN;
1265 
1266 	xen_setup_features();
1267 
1268 	xen_setup_machphys_mapping();
1269 
1270 	/* Install Xen paravirt ops */
1271 	pv_info = xen_info;
1272 	pv_init_ops = xen_init_ops;
1273 	pv_cpu_ops = xen_cpu_ops;
1274 
1275 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1276 
1277 	x86_init.resources.memory_setup = xen_memory_setup;
1278 	x86_init.oem.arch_setup = xen_arch_setup;
1279 	x86_init.oem.banner = xen_banner;
1280 
1281 	/*
1282 	 * Set up some pagetable state before starting to set any ptes.
1283 	 */
1284 
1285 	xen_init_mmu_ops();
1286 
1287 	/* Prevent unwanted bits from being set in PTEs. */
1288 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1289 
1290 	/*
1291 	 * Prevent page tables from being allocated in highmem, even
1292 	 * if CONFIG_HIGHPTE is enabled.
1293 	 */
1294 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1295 
1296 	/* Work out if we support NX */
1297 	x86_configure_nx();
1298 
1299 	/* Get mfn list */
1300 	xen_build_dynamic_phys_to_machine();
1301 
1302 	/*
1303 	 * Set up kernel GDT and segment registers, mainly so that
1304 	 * -fstack-protector code can be executed.
1305 	 */
1306 	xen_setup_gdt(0);
1307 
1308 	xen_init_irq_ops();
1309 	xen_init_capabilities();
1310 
1311 #ifdef CONFIG_X86_LOCAL_APIC
1312 	/*
1313 	 * set up the basic apic ops.
1314 	 */
1315 	xen_init_apic();
1316 #endif
1317 
1318 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1319 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1320 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1321 	}
1322 
1323 	machine_ops = xen_machine_ops;
1324 
1325 	/*
1326 	 * The only reliable way to retain the initial address of the
1327 	 * percpu gdt_page is to remember it here, so we can go and
1328 	 * mark it RW later, when the initial percpu area is freed.
1329 	 */
1330 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1331 
1332 	xen_smp_init();
1333 
1334 #ifdef CONFIG_ACPI_NUMA
1335 	/*
1336 	 * The pages we from Xen are not related to machine pages, so
1337 	 * any NUMA information the kernel tries to get from ACPI will
1338 	 * be meaningless.  Prevent it from trying.
1339 	 */
1340 	acpi_numa = -1;
1341 #endif
1342 	/* Don't do the full vcpu_info placement stuff until we have a
1343 	   possible map and a non-dummy shared_info. */
1344 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1345 
1346 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1347 
1348 	local_irq_disable();
1349 	early_boot_irqs_disabled = true;
1350 
1351 	xen_raw_console_write("mapping kernel into physical memory\n");
1352 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1353 				   xen_start_info->nr_pages);
1354 	xen_reserve_special_pages();
1355 
1356 	/* keep using Xen gdt for now; no urgent need to change it */
1357 
1358 #ifdef CONFIG_X86_32
1359 	pv_info.kernel_rpl = 1;
1360 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1361 		pv_info.kernel_rpl = 0;
1362 #else
1363 	pv_info.kernel_rpl = 0;
1364 #endif
1365 	/* set the limit of our address space */
1366 	xen_reserve_top();
1367 
1368 	/*
1369 	 * We used to do this in xen_arch_setup, but that is too late
1370 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1371 	 * early_amd_init which pokes 0xcf8 port.
1372 	 */
1373 	set_iopl.iopl = 1;
1374 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1375 	if (rc != 0)
1376 		xen_raw_printk("physdev_op failed %d\n", rc);
1377 
1378 #ifdef CONFIG_X86_32
1379 	/* set up basic CPUID stuff */
1380 	cpu_detect(&new_cpu_data);
1381 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1382 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1383 #endif
1384 
1385 	if (xen_start_info->mod_start) {
1386 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1387 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1388 	    else
1389 		initrd_start = __pa(xen_start_info->mod_start);
1390 	}
1391 
1392 	/* Poke various useful things into boot_params */
1393 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1394 	boot_params.hdr.ramdisk_image = initrd_start;
1395 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1396 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1397 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1398 
1399 	if (!xen_initial_domain()) {
1400 		add_preferred_console("xenboot", 0, NULL);
1401 		add_preferred_console("tty", 0, NULL);
1402 		add_preferred_console("hvc", 0, NULL);
1403 		if (pci_xen)
1404 			x86_init.pci.arch_init = pci_xen_init;
1405 	} else {
1406 		const struct dom0_vga_console_info *info =
1407 			(void *)((char *)xen_start_info +
1408 				 xen_start_info->console.dom0.info_off);
1409 		struct xen_platform_op op = {
1410 			.cmd = XENPF_firmware_info,
1411 			.interface_version = XENPF_INTERFACE_VERSION,
1412 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1413 		};
1414 
1415 		x86_platform.set_legacy_features =
1416 				xen_dom0_set_legacy_features;
1417 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1418 		xen_start_info->console.domU.mfn = 0;
1419 		xen_start_info->console.domU.evtchn = 0;
1420 
1421 		if (HYPERVISOR_platform_op(&op) == 0)
1422 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1423 
1424 		/* Make sure ACS will be enabled */
1425 		pci_request_acs();
1426 
1427 		xen_acpi_sleep_register();
1428 
1429 		/* Avoid searching for BIOS MP tables */
1430 		x86_init.mpparse.find_smp_config = x86_init_noop;
1431 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1432 
1433 		xen_boot_params_init_edd();
1434 	}
1435 #ifdef CONFIG_PCI
1436 	/* PCI BIOS service won't work from a PV guest. */
1437 	pci_probe &= ~PCI_PROBE_BIOS;
1438 #endif
1439 	xen_raw_console_write("about to get started...\n");
1440 
1441 	/* Let's presume PV guests always boot on vCPU with id 0. */
1442 	per_cpu(xen_vcpu_id, 0) = 0;
1443 
1444 	xen_setup_runstate_info(0);
1445 
1446 	xen_efi_init();
1447 
1448 	/* Start the world */
1449 #ifdef CONFIG_X86_32
1450 	i386_start_kernel();
1451 #else
1452 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1453 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1454 #endif
1455 }
1456 
1457 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1458 {
1459 	int rc;
1460 
1461 	xen_setup_timer(cpu);
1462 
1463 	rc = xen_smp_intr_init(cpu);
1464 	if (rc) {
1465 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1466 		     cpu, rc);
1467 		return rc;
1468 	}
1469 
1470 	rc = xen_smp_intr_init_pv(cpu);
1471 	if (rc) {
1472 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1473 		     cpu, rc);
1474 		return rc;
1475 	}
1476 
1477 	return 0;
1478 }
1479 
1480 static int xen_cpu_dead_pv(unsigned int cpu)
1481 {
1482 	xen_smp_intr_free(cpu);
1483 	xen_smp_intr_free_pv(cpu);
1484 
1485 	xen_teardown_timer(cpu);
1486 
1487 	return 0;
1488 }
1489 
1490 static uint32_t __init xen_platform_pv(void)
1491 {
1492 	if (xen_pv_domain())
1493 		return xen_cpuid_base();
1494 
1495 	return 0;
1496 }
1497 
1498 const struct hypervisor_x86 x86_hyper_xen_pv = {
1499 	.name                   = "Xen PV",
1500 	.detect                 = xen_platform_pv,
1501 	.pin_vcpu               = xen_pin_vcpu,
1502 };
1503 EXPORT_SYMBOL(x86_hyper_xen_pv);
1504