xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 7cc39531)
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/memblock.h>
27 #include <linux/export.h>
28 #include <linux/mm.h>
29 #include <linux/page-flags.h>
30 #include <linux/pci.h>
31 #include <linux/gfp.h>
32 #include <linux/edd.h>
33 #include <linux/reboot.h>
34 #include <linux/virtio_anchor.h>
35 
36 #include <xen/xen.h>
37 #include <xen/events.h>
38 #include <xen/interface/xen.h>
39 #include <xen/interface/version.h>
40 #include <xen/interface/physdev.h>
41 #include <xen/interface/vcpu.h>
42 #include <xen/interface/memory.h>
43 #include <xen/interface/nmi.h>
44 #include <xen/interface/xen-mca.h>
45 #include <xen/features.h>
46 #include <xen/page.h>
47 #include <xen/hvc-console.h>
48 #include <xen/acpi.h>
49 
50 #include <asm/paravirt.h>
51 #include <asm/apic.h>
52 #include <asm/page.h>
53 #include <asm/xen/pci.h>
54 #include <asm/xen/hypercall.h>
55 #include <asm/xen/hypervisor.h>
56 #include <asm/xen/cpuid.h>
57 #include <asm/fixmap.h>
58 #include <asm/processor.h>
59 #include <asm/proto.h>
60 #include <asm/msr-index.h>
61 #include <asm/traps.h>
62 #include <asm/setup.h>
63 #include <asm/desc.h>
64 #include <asm/pgalloc.h>
65 #include <asm/tlbflush.h>
66 #include <asm/reboot.h>
67 #include <asm/stackprotector.h>
68 #include <asm/hypervisor.h>
69 #include <asm/mach_traps.h>
70 #include <asm/mwait.h>
71 #include <asm/pci_x86.h>
72 #include <asm/cpu.h>
73 #ifdef CONFIG_X86_IOPL_IOPERM
74 #include <asm/io_bitmap.h>
75 #endif
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_pv_init_platform(void)
112 {
113 	/* PV guests can't operate virtio devices without grants. */
114 	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
115 		virtio_set_mem_acc_cb(virtio_require_restricted_mem_acc);
116 
117 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
118 
119 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
120 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
121 
122 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
123 	xen_vcpu_info_reset(0);
124 
125 	/* pvclock is in shared info area */
126 	xen_init_time_ops();
127 }
128 
129 static void __init xen_pv_guest_late_init(void)
130 {
131 #ifndef CONFIG_SMP
132 	/* Setup shared vcpu info for non-smp configurations */
133 	xen_setup_vcpu_info_placement();
134 #endif
135 }
136 
137 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
138 static __read_mostly unsigned int cpuid_leaf5_edx_val;
139 
140 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
141 		      unsigned int *cx, unsigned int *dx)
142 {
143 	unsigned maskebx = ~0;
144 
145 	/*
146 	 * Mask out inconvenient features, to try and disable as many
147 	 * unsupported kernel subsystems as possible.
148 	 */
149 	switch (*ax) {
150 	case CPUID_MWAIT_LEAF:
151 		/* Synthesize the values.. */
152 		*ax = 0;
153 		*bx = 0;
154 		*cx = cpuid_leaf5_ecx_val;
155 		*dx = cpuid_leaf5_edx_val;
156 		return;
157 
158 	case 0xb:
159 		/* Suppress extended topology stuff */
160 		maskebx = 0;
161 		break;
162 	}
163 
164 	asm(XEN_EMULATE_PREFIX "cpuid"
165 		: "=a" (*ax),
166 		  "=b" (*bx),
167 		  "=c" (*cx),
168 		  "=d" (*dx)
169 		: "0" (*ax), "2" (*cx));
170 
171 	*bx &= maskebx;
172 }
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 noinstr void xen_set_debugreg(int reg, unsigned long val)
290 {
291 	HYPERVISOR_set_debugreg(reg, val);
292 }
293 
294 static noinstr 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 	unsigned char dummy;
323 	void *va;
324 
325 	ptep = lookup_address((unsigned long)v, &level);
326 	BUG_ON(ptep == NULL);
327 
328 	pfn = pte_pfn(*ptep);
329 	pte = pfn_pte(pfn, prot);
330 
331 	/*
332 	 * Careful: update_va_mapping() will fail if the virtual address
333 	 * we're poking isn't populated in the page tables.  We don't
334 	 * need to worry about the direct map (that's always in the page
335 	 * tables), but we need to be careful about vmap space.  In
336 	 * particular, the top level page table can lazily propagate
337 	 * entries between processes, so if we've switched mms since we
338 	 * vmapped the target in the first place, we might not have the
339 	 * top-level page table entry populated.
340 	 *
341 	 * We disable preemption because we want the same mm active when
342 	 * we probe the target and when we issue the hypercall.  We'll
343 	 * have the same nominal mm, but if we're a kernel thread, lazy
344 	 * mm dropping could change our pgd.
345 	 *
346 	 * Out of an abundance of caution, this uses __get_user() to fault
347 	 * in the target address just in case there's some obscure case
348 	 * in which the target address isn't readable.
349 	 */
350 
351 	preempt_disable();
352 
353 	copy_from_kernel_nofault(&dummy, v, 1);
354 
355 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
356 		BUG();
357 
358 	va = __va(PFN_PHYS(pfn));
359 
360 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
361 		BUG();
362 
363 	preempt_enable();
364 }
365 
366 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
367 {
368 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
369 	int i;
370 
371 	/*
372 	 * We need to mark the all aliases of the LDT pages RO.  We
373 	 * don't need to call vm_flush_aliases(), though, since that's
374 	 * only responsible for flushing aliases out the TLBs, not the
375 	 * page tables, and Xen will flush the TLB for us if needed.
376 	 *
377 	 * To avoid confusing future readers: none of this is necessary
378 	 * to load the LDT.  The hypervisor only checks this when the
379 	 * LDT is faulted in due to subsequent descriptor access.
380 	 */
381 
382 	for (i = 0; i < entries; i += entries_per_page)
383 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
384 }
385 
386 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
387 {
388 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
389 	int i;
390 
391 	for (i = 0; i < entries; i += entries_per_page)
392 		set_aliased_prot(ldt + i, PAGE_KERNEL);
393 }
394 
395 static void xen_set_ldt(const void *addr, unsigned entries)
396 {
397 	struct mmuext_op *op;
398 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
399 
400 	trace_xen_cpu_set_ldt(addr, entries);
401 
402 	op = mcs.args;
403 	op->cmd = MMUEXT_SET_LDT;
404 	op->arg1.linear_addr = (unsigned long)addr;
405 	op->arg2.nr_ents = entries;
406 
407 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
408 
409 	xen_mc_issue(PARAVIRT_LAZY_CPU);
410 }
411 
412 static void xen_load_gdt(const struct desc_ptr *dtr)
413 {
414 	unsigned long va = dtr->address;
415 	unsigned int size = dtr->size + 1;
416 	unsigned long pfn, mfn;
417 	int level;
418 	pte_t *ptep;
419 	void *virt;
420 
421 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
422 	BUG_ON(size > PAGE_SIZE);
423 	BUG_ON(va & ~PAGE_MASK);
424 
425 	/*
426 	 * The GDT is per-cpu and is in the percpu data area.
427 	 * That can be virtually mapped, so we need to do a
428 	 * page-walk to get the underlying MFN for the
429 	 * hypercall.  The page can also be in the kernel's
430 	 * linear range, so we need to RO that mapping too.
431 	 */
432 	ptep = lookup_address(va, &level);
433 	BUG_ON(ptep == NULL);
434 
435 	pfn = pte_pfn(*ptep);
436 	mfn = pfn_to_mfn(pfn);
437 	virt = __va(PFN_PHYS(pfn));
438 
439 	make_lowmem_page_readonly((void *)va);
440 	make_lowmem_page_readonly(virt);
441 
442 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
443 		BUG();
444 }
445 
446 /*
447  * load_gdt for early boot, when the gdt is only mapped once
448  */
449 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
450 {
451 	unsigned long va = dtr->address;
452 	unsigned int size = dtr->size + 1;
453 	unsigned long pfn, mfn;
454 	pte_t pte;
455 
456 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
457 	BUG_ON(size > PAGE_SIZE);
458 	BUG_ON(va & ~PAGE_MASK);
459 
460 	pfn = virt_to_pfn(va);
461 	mfn = pfn_to_mfn(pfn);
462 
463 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
464 
465 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
466 		BUG();
467 
468 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
469 		BUG();
470 }
471 
472 static inline bool desc_equal(const struct desc_struct *d1,
473 			      const struct desc_struct *d2)
474 {
475 	return !memcmp(d1, d2, sizeof(*d1));
476 }
477 
478 static void load_TLS_descriptor(struct thread_struct *t,
479 				unsigned int cpu, unsigned int i)
480 {
481 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
482 	struct desc_struct *gdt;
483 	xmaddr_t maddr;
484 	struct multicall_space mc;
485 
486 	if (desc_equal(shadow, &t->tls_array[i]))
487 		return;
488 
489 	*shadow = t->tls_array[i];
490 
491 	gdt = get_cpu_gdt_rw(cpu);
492 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
493 	mc = __xen_mc_entry(0);
494 
495 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
496 }
497 
498 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
499 {
500 	/*
501 	 * In lazy mode we need to zero %fs, otherwise we may get an
502 	 * exception between the new %fs descriptor being loaded and
503 	 * %fs being effectively cleared at __switch_to().
504 	 */
505 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
506 		loadsegment(fs, 0);
507 
508 	xen_mc_batch();
509 
510 	load_TLS_descriptor(t, cpu, 0);
511 	load_TLS_descriptor(t, cpu, 1);
512 	load_TLS_descriptor(t, cpu, 2);
513 
514 	xen_mc_issue(PARAVIRT_LAZY_CPU);
515 }
516 
517 static void xen_load_gs_index(unsigned int idx)
518 {
519 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
520 		BUG();
521 }
522 
523 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
524 				const void *ptr)
525 {
526 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
527 	u64 entry = *(u64 *)ptr;
528 
529 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
530 
531 	preempt_disable();
532 
533 	xen_mc_flush();
534 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
535 		BUG();
536 
537 	preempt_enable();
538 }
539 
540 void noist_exc_debug(struct pt_regs *regs);
541 
542 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
543 {
544 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
545 	exc_nmi(regs);
546 }
547 
548 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
549 {
550 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
551 	exc_double_fault(regs, error_code);
552 }
553 
554 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
555 {
556 	/*
557 	 * There's no IST on Xen PV, but we still need to dispatch
558 	 * to the correct handler.
559 	 */
560 	if (user_mode(regs))
561 		noist_exc_debug(regs);
562 	else
563 		exc_debug(regs);
564 }
565 
566 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
567 {
568 	/* This should never happen and there is no way to handle it. */
569 	instrumentation_begin();
570 	pr_err("Unknown trap in Xen PV mode.");
571 	BUG();
572 	instrumentation_end();
573 }
574 
575 #ifdef CONFIG_X86_MCE
576 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
577 {
578 	/*
579 	 * There's no IST on Xen PV, but we still need to dispatch
580 	 * to the correct handler.
581 	 */
582 	if (user_mode(regs))
583 		noist_exc_machine_check(regs);
584 	else
585 		exc_machine_check(regs);
586 }
587 #endif
588 
589 struct trap_array_entry {
590 	void (*orig)(void);
591 	void (*xen)(void);
592 	bool ist_okay;
593 };
594 
595 #define TRAP_ENTRY(func, ist_ok) {			\
596 	.orig		= asm_##func,			\
597 	.xen		= xen_asm_##func,		\
598 	.ist_okay	= ist_ok }
599 
600 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
601 	.orig		= asm_##func,			\
602 	.xen		= xen_asm_xenpv_##func,		\
603 	.ist_okay	= ist_ok }
604 
605 static struct trap_array_entry trap_array[] = {
606 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
607 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
608 #ifdef CONFIG_X86_MCE
609 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
610 #endif
611 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
612 	TRAP_ENTRY(exc_int3,				false ),
613 	TRAP_ENTRY(exc_overflow,			false ),
614 #ifdef CONFIG_IA32_EMULATION
615 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
616 #endif
617 	TRAP_ENTRY(exc_page_fault,			false ),
618 	TRAP_ENTRY(exc_divide_error,			false ),
619 	TRAP_ENTRY(exc_bounds,				false ),
620 	TRAP_ENTRY(exc_invalid_op,			false ),
621 	TRAP_ENTRY(exc_device_not_available,		false ),
622 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
623 	TRAP_ENTRY(exc_invalid_tss,			false ),
624 	TRAP_ENTRY(exc_segment_not_present,		false ),
625 	TRAP_ENTRY(exc_stack_segment,			false ),
626 	TRAP_ENTRY(exc_general_protection,		false ),
627 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
628 	TRAP_ENTRY(exc_coprocessor_error,		false ),
629 	TRAP_ENTRY(exc_alignment_check,			false ),
630 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
631 #ifdef CONFIG_X86_KERNEL_IBT
632 	TRAP_ENTRY(exc_control_protection,		false ),
633 #endif
634 };
635 
636 static bool __ref get_trap_addr(void **addr, unsigned int ist)
637 {
638 	unsigned int nr;
639 	bool ist_okay = false;
640 	bool found = false;
641 
642 	/*
643 	 * Replace trap handler addresses by Xen specific ones.
644 	 * Check for known traps using IST and whitelist them.
645 	 * The debugger ones are the only ones we care about.
646 	 * Xen will handle faults like double_fault, so we should never see
647 	 * them.  Warn if there's an unexpected IST-using fault handler.
648 	 */
649 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
650 		struct trap_array_entry *entry = trap_array + nr;
651 
652 		if (*addr == entry->orig) {
653 			*addr = entry->xen;
654 			ist_okay = entry->ist_okay;
655 			found = true;
656 			break;
657 		}
658 	}
659 
660 	if (nr == ARRAY_SIZE(trap_array) &&
661 	    *addr >= (void *)early_idt_handler_array[0] &&
662 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
663 		nr = (*addr - (void *)early_idt_handler_array[0]) /
664 		     EARLY_IDT_HANDLER_SIZE;
665 		*addr = (void *)xen_early_idt_handler_array[nr];
666 		found = true;
667 	}
668 
669 	if (!found)
670 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
671 
672 	if (WARN_ON(found && ist != 0 && !ist_okay))
673 		return false;
674 
675 	return true;
676 }
677 
678 static int cvt_gate_to_trap(int vector, const gate_desc *val,
679 			    struct trap_info *info)
680 {
681 	unsigned long addr;
682 
683 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
684 		return 0;
685 
686 	info->vector = vector;
687 
688 	addr = gate_offset(val);
689 	if (!get_trap_addr((void **)&addr, val->bits.ist))
690 		return 0;
691 	info->address = addr;
692 
693 	info->cs = gate_segment(val);
694 	info->flags = val->bits.dpl;
695 	/* interrupt gates clear IF */
696 	if (val->bits.type == GATE_INTERRUPT)
697 		info->flags |= 1 << 2;
698 
699 	return 1;
700 }
701 
702 /* Locations of each CPU's IDT */
703 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
704 
705 /* Set an IDT entry.  If the entry is part of the current IDT, then
706    also update Xen. */
707 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
708 {
709 	unsigned long p = (unsigned long)&dt[entrynum];
710 	unsigned long start, end;
711 
712 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
713 
714 	preempt_disable();
715 
716 	start = __this_cpu_read(idt_desc.address);
717 	end = start + __this_cpu_read(idt_desc.size) + 1;
718 
719 	xen_mc_flush();
720 
721 	native_write_idt_entry(dt, entrynum, g);
722 
723 	if (p >= start && (p + 8) <= end) {
724 		struct trap_info info[2];
725 
726 		info[1].address = 0;
727 
728 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
729 			if (HYPERVISOR_set_trap_table(info))
730 				BUG();
731 	}
732 
733 	preempt_enable();
734 }
735 
736 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
737 				      struct trap_info *traps, bool full)
738 {
739 	unsigned in, out, count;
740 
741 	count = (desc->size+1) / sizeof(gate_desc);
742 	BUG_ON(count > 256);
743 
744 	for (in = out = 0; in < count; in++) {
745 		gate_desc *entry = (gate_desc *)(desc->address) + in;
746 
747 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
748 			out++;
749 	}
750 
751 	return out;
752 }
753 
754 void xen_copy_trap_info(struct trap_info *traps)
755 {
756 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
757 
758 	xen_convert_trap_info(desc, traps, true);
759 }
760 
761 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
762    hold a spinlock to protect the static traps[] array (static because
763    it avoids allocation, and saves stack space). */
764 static void xen_load_idt(const struct desc_ptr *desc)
765 {
766 	static DEFINE_SPINLOCK(lock);
767 	static struct trap_info traps[257];
768 	unsigned out;
769 
770 	trace_xen_cpu_load_idt(desc);
771 
772 	spin_lock(&lock);
773 
774 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
775 
776 	out = xen_convert_trap_info(desc, traps, false);
777 	memset(&traps[out], 0, sizeof(traps[0]));
778 
779 	xen_mc_flush();
780 	if (HYPERVISOR_set_trap_table(traps))
781 		BUG();
782 
783 	spin_unlock(&lock);
784 }
785 
786 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
787    they're handled differently. */
788 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
789 				const void *desc, int type)
790 {
791 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
792 
793 	preempt_disable();
794 
795 	switch (type) {
796 	case DESC_LDT:
797 	case DESC_TSS:
798 		/* ignore */
799 		break;
800 
801 	default: {
802 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
803 
804 		xen_mc_flush();
805 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
806 			BUG();
807 	}
808 
809 	}
810 
811 	preempt_enable();
812 }
813 
814 /*
815  * Version of write_gdt_entry for use at early boot-time needed to
816  * update an entry as simply as possible.
817  */
818 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
819 					    const void *desc, int type)
820 {
821 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
822 
823 	switch (type) {
824 	case DESC_LDT:
825 	case DESC_TSS:
826 		/* ignore */
827 		break;
828 
829 	default: {
830 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
831 
832 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
833 			dt[entry] = *(struct desc_struct *)desc;
834 	}
835 
836 	}
837 }
838 
839 static void xen_load_sp0(unsigned long sp0)
840 {
841 	struct multicall_space mcs;
842 
843 	mcs = xen_mc_entry(0);
844 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
845 	xen_mc_issue(PARAVIRT_LAZY_CPU);
846 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
847 }
848 
849 #ifdef CONFIG_X86_IOPL_IOPERM
850 static void xen_invalidate_io_bitmap(void)
851 {
852 	struct physdev_set_iobitmap iobitmap = {
853 		.bitmap = NULL,
854 		.nr_ports = 0,
855 	};
856 
857 	native_tss_invalidate_io_bitmap();
858 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
859 }
860 
861 static void xen_update_io_bitmap(void)
862 {
863 	struct physdev_set_iobitmap iobitmap;
864 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
865 
866 	native_tss_update_io_bitmap();
867 
868 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
869 			  tss->x86_tss.io_bitmap_base;
870 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
871 		iobitmap.nr_ports = 0;
872 	else
873 		iobitmap.nr_ports = IO_BITMAP_BITS;
874 
875 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
876 }
877 #endif
878 
879 static void xen_io_delay(void)
880 {
881 }
882 
883 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
884 
885 static unsigned long xen_read_cr0(void)
886 {
887 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
888 
889 	if (unlikely(cr0 == 0)) {
890 		cr0 = native_read_cr0();
891 		this_cpu_write(xen_cr0_value, cr0);
892 	}
893 
894 	return cr0;
895 }
896 
897 static void xen_write_cr0(unsigned long cr0)
898 {
899 	struct multicall_space mcs;
900 
901 	this_cpu_write(xen_cr0_value, cr0);
902 
903 	/* Only pay attention to cr0.TS; everything else is
904 	   ignored. */
905 	mcs = xen_mc_entry(0);
906 
907 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
908 
909 	xen_mc_issue(PARAVIRT_LAZY_CPU);
910 }
911 
912 static void xen_write_cr4(unsigned long cr4)
913 {
914 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
915 
916 	native_write_cr4(cr4);
917 }
918 
919 static u64 xen_read_msr_safe(unsigned int msr, int *err)
920 {
921 	u64 val;
922 
923 	if (pmu_msr_read(msr, &val, err))
924 		return val;
925 
926 	val = native_read_msr_safe(msr, err);
927 	switch (msr) {
928 	case MSR_IA32_APICBASE:
929 		val &= ~X2APIC_ENABLE;
930 		break;
931 	}
932 	return val;
933 }
934 
935 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
936 {
937 	int ret;
938 	unsigned int which;
939 	u64 base;
940 
941 	ret = 0;
942 
943 	switch (msr) {
944 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
945 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
946 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
947 
948 	set:
949 		base = ((u64)high << 32) | low;
950 		if (HYPERVISOR_set_segment_base(which, base) != 0)
951 			ret = -EIO;
952 		break;
953 
954 	case MSR_STAR:
955 	case MSR_CSTAR:
956 	case MSR_LSTAR:
957 	case MSR_SYSCALL_MASK:
958 	case MSR_IA32_SYSENTER_CS:
959 	case MSR_IA32_SYSENTER_ESP:
960 	case MSR_IA32_SYSENTER_EIP:
961 		/* Fast syscall setup is all done in hypercalls, so
962 		   these are all ignored.  Stub them out here to stop
963 		   Xen console noise. */
964 		break;
965 
966 	default:
967 		if (!pmu_msr_write(msr, low, high, &ret))
968 			ret = native_write_msr_safe(msr, low, high);
969 	}
970 
971 	return ret;
972 }
973 
974 static u64 xen_read_msr(unsigned int msr)
975 {
976 	/*
977 	 * This will silently swallow a #GP from RDMSR.  It may be worth
978 	 * changing that.
979 	 */
980 	int err;
981 
982 	return xen_read_msr_safe(msr, &err);
983 }
984 
985 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
986 {
987 	/*
988 	 * This will silently swallow a #GP from WRMSR.  It may be worth
989 	 * changing that.
990 	 */
991 	xen_write_msr_safe(msr, low, high);
992 }
993 
994 /* This is called once we have the cpu_possible_mask */
995 void __init xen_setup_vcpu_info_placement(void)
996 {
997 	int cpu;
998 
999 	for_each_possible_cpu(cpu) {
1000 		/* Set up direct vCPU id mapping for PV guests. */
1001 		per_cpu(xen_vcpu_id, cpu) = cpu;
1002 		xen_vcpu_setup(cpu);
1003 	}
1004 
1005 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1006 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1007 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1008 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1009 }
1010 
1011 static const struct pv_info xen_info __initconst = {
1012 	.extra_user_64bit_cs = FLAT_USER_CS64,
1013 	.name = "Xen",
1014 };
1015 
1016 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1017 	.cpu = {
1018 		.cpuid = xen_cpuid,
1019 
1020 		.set_debugreg = xen_set_debugreg,
1021 		.get_debugreg = xen_get_debugreg,
1022 
1023 		.read_cr0 = xen_read_cr0,
1024 		.write_cr0 = xen_write_cr0,
1025 
1026 		.write_cr4 = xen_write_cr4,
1027 
1028 		.wbinvd = native_wbinvd,
1029 
1030 		.read_msr = xen_read_msr,
1031 		.write_msr = xen_write_msr,
1032 
1033 		.read_msr_safe = xen_read_msr_safe,
1034 		.write_msr_safe = xen_write_msr_safe,
1035 
1036 		.read_pmc = xen_read_pmc,
1037 
1038 		.load_tr_desc = paravirt_nop,
1039 		.set_ldt = xen_set_ldt,
1040 		.load_gdt = xen_load_gdt,
1041 		.load_idt = xen_load_idt,
1042 		.load_tls = xen_load_tls,
1043 		.load_gs_index = xen_load_gs_index,
1044 
1045 		.alloc_ldt = xen_alloc_ldt,
1046 		.free_ldt = xen_free_ldt,
1047 
1048 		.store_tr = xen_store_tr,
1049 
1050 		.write_ldt_entry = xen_write_ldt_entry,
1051 		.write_gdt_entry = xen_write_gdt_entry,
1052 		.write_idt_entry = xen_write_idt_entry,
1053 		.load_sp0 = xen_load_sp0,
1054 
1055 #ifdef CONFIG_X86_IOPL_IOPERM
1056 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1057 		.update_io_bitmap = xen_update_io_bitmap,
1058 #endif
1059 		.io_delay = xen_io_delay,
1060 
1061 		.start_context_switch = paravirt_start_context_switch,
1062 		.end_context_switch = xen_end_context_switch,
1063 	},
1064 };
1065 
1066 static void xen_restart(char *msg)
1067 {
1068 	xen_reboot(SHUTDOWN_reboot);
1069 }
1070 
1071 static void xen_machine_halt(void)
1072 {
1073 	xen_reboot(SHUTDOWN_poweroff);
1074 }
1075 
1076 static void xen_machine_power_off(void)
1077 {
1078 	do_kernel_power_off();
1079 	xen_reboot(SHUTDOWN_poweroff);
1080 }
1081 
1082 static void xen_crash_shutdown(struct pt_regs *regs)
1083 {
1084 	xen_reboot(SHUTDOWN_crash);
1085 }
1086 
1087 static const struct machine_ops xen_machine_ops __initconst = {
1088 	.restart = xen_restart,
1089 	.halt = xen_machine_halt,
1090 	.power_off = xen_machine_power_off,
1091 	.shutdown = xen_machine_halt,
1092 	.crash_shutdown = xen_crash_shutdown,
1093 	.emergency_restart = xen_emergency_restart,
1094 };
1095 
1096 static unsigned char xen_get_nmi_reason(void)
1097 {
1098 	unsigned char reason = 0;
1099 
1100 	/* Construct a value which looks like it came from port 0x61. */
1101 	if (test_bit(_XEN_NMIREASON_io_error,
1102 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1103 		reason |= NMI_REASON_IOCHK;
1104 	if (test_bit(_XEN_NMIREASON_pci_serr,
1105 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1106 		reason |= NMI_REASON_SERR;
1107 
1108 	return reason;
1109 }
1110 
1111 static void __init xen_boot_params_init_edd(void)
1112 {
1113 #if IS_ENABLED(CONFIG_EDD)
1114 	struct xen_platform_op op;
1115 	struct edd_info *edd_info;
1116 	u32 *mbr_signature;
1117 	unsigned nr;
1118 	int ret;
1119 
1120 	edd_info = boot_params.eddbuf;
1121 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1122 
1123 	op.cmd = XENPF_firmware_info;
1124 
1125 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1126 	for (nr = 0; nr < EDDMAXNR; nr++) {
1127 		struct edd_info *info = edd_info + nr;
1128 
1129 		op.u.firmware_info.index = nr;
1130 		info->params.length = sizeof(info->params);
1131 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1132 				     &info->params);
1133 		ret = HYPERVISOR_platform_op(&op);
1134 		if (ret)
1135 			break;
1136 
1137 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1138 		C(device);
1139 		C(version);
1140 		C(interface_support);
1141 		C(legacy_max_cylinder);
1142 		C(legacy_max_head);
1143 		C(legacy_sectors_per_track);
1144 #undef C
1145 	}
1146 	boot_params.eddbuf_entries = nr;
1147 
1148 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1149 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1150 		op.u.firmware_info.index = nr;
1151 		ret = HYPERVISOR_platform_op(&op);
1152 		if (ret)
1153 			break;
1154 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1155 	}
1156 	boot_params.edd_mbr_sig_buf_entries = nr;
1157 #endif
1158 }
1159 
1160 /*
1161  * Set up the GDT and segment registers for -fstack-protector.  Until
1162  * we do this, we have to be careful not to call any stack-protected
1163  * function, which is most of the kernel.
1164  */
1165 static void __init xen_setup_gdt(int cpu)
1166 {
1167 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1168 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1169 
1170 	switch_to_new_gdt(cpu);
1171 
1172 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1173 	pv_ops.cpu.load_gdt = xen_load_gdt;
1174 }
1175 
1176 static void __init xen_dom0_set_legacy_features(void)
1177 {
1178 	x86_platform.legacy.rtc = 1;
1179 }
1180 
1181 static void __init xen_domu_set_legacy_features(void)
1182 {
1183 	x86_platform.legacy.rtc = 0;
1184 }
1185 
1186 extern void early_xen_iret_patch(void);
1187 
1188 /* First C function to be called on Xen boot */
1189 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1190 {
1191 	struct physdev_set_iopl set_iopl;
1192 	unsigned long initrd_start = 0;
1193 	int rc;
1194 
1195 	if (!si)
1196 		return;
1197 
1198 	clear_bss();
1199 
1200 	xen_start_info = si;
1201 
1202 	__text_gen_insn(&early_xen_iret_patch,
1203 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1204 			JMP32_INSN_SIZE);
1205 
1206 	xen_domain_type = XEN_PV_DOMAIN;
1207 	xen_start_flags = xen_start_info->flags;
1208 
1209 	xen_setup_features();
1210 
1211 	/* Install Xen paravirt ops */
1212 	pv_info = xen_info;
1213 	pv_ops.cpu = xen_cpu_ops.cpu;
1214 	xen_init_irq_ops();
1215 
1216 	/*
1217 	 * Setup xen_vcpu early because it is needed for
1218 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1219 	 *
1220 	 * Don't do the full vcpu_info placement stuff until we have
1221 	 * the cpu_possible_mask and a non-dummy shared_info.
1222 	 */
1223 	xen_vcpu_info_reset(0);
1224 
1225 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1226 
1227 	x86_init.resources.memory_setup = xen_memory_setup;
1228 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1229 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1230 	x86_init.oem.arch_setup = xen_arch_setup;
1231 	x86_init.oem.banner = xen_banner;
1232 	x86_init.hyper.init_platform = xen_pv_init_platform;
1233 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1234 
1235 	/*
1236 	 * Set up some pagetable state before starting to set any ptes.
1237 	 */
1238 
1239 	xen_setup_machphys_mapping();
1240 	xen_init_mmu_ops();
1241 
1242 	/* Prevent unwanted bits from being set in PTEs. */
1243 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1244 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1245 
1246 	/* Get mfn list */
1247 	xen_build_dynamic_phys_to_machine();
1248 
1249 	/* Work out if we support NX */
1250 	get_cpu_cap(&boot_cpu_data);
1251 	x86_configure_nx();
1252 
1253 	/*
1254 	 * Set up kernel GDT and segment registers, mainly so that
1255 	 * -fstack-protector code can be executed.
1256 	 */
1257 	xen_setup_gdt(0);
1258 
1259 	/* Determine virtual and physical address sizes */
1260 	get_cpu_address_sizes(&boot_cpu_data);
1261 
1262 	/* Let's presume PV guests always boot on vCPU with id 0. */
1263 	per_cpu(xen_vcpu_id, 0) = 0;
1264 
1265 	idt_setup_early_handler();
1266 
1267 	xen_init_capabilities();
1268 
1269 #ifdef CONFIG_X86_LOCAL_APIC
1270 	/*
1271 	 * set up the basic apic ops.
1272 	 */
1273 	xen_init_apic();
1274 #endif
1275 
1276 	machine_ops = xen_machine_ops;
1277 
1278 	/*
1279 	 * The only reliable way to retain the initial address of the
1280 	 * percpu gdt_page is to remember it here, so we can go and
1281 	 * mark it RW later, when the initial percpu area is freed.
1282 	 */
1283 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1284 
1285 	xen_smp_init();
1286 
1287 #ifdef CONFIG_ACPI_NUMA
1288 	/*
1289 	 * The pages we from Xen are not related to machine pages, so
1290 	 * any NUMA information the kernel tries to get from ACPI will
1291 	 * be meaningless.  Prevent it from trying.
1292 	 */
1293 	disable_srat();
1294 #endif
1295 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1296 
1297 	local_irq_disable();
1298 	early_boot_irqs_disabled = true;
1299 
1300 	xen_raw_console_write("mapping kernel into physical memory\n");
1301 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1302 				   xen_start_info->nr_pages);
1303 	xen_reserve_special_pages();
1304 
1305 	/*
1306 	 * We used to do this in xen_arch_setup, but that is too late
1307 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1308 	 * early_amd_init which pokes 0xcf8 port.
1309 	 */
1310 	set_iopl.iopl = 1;
1311 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1312 	if (rc != 0)
1313 		xen_raw_printk("physdev_op failed %d\n", rc);
1314 
1315 
1316 	if (xen_start_info->mod_start) {
1317 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1318 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1319 	    else
1320 		initrd_start = __pa(xen_start_info->mod_start);
1321 	}
1322 
1323 	/* Poke various useful things into boot_params */
1324 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1325 	boot_params.hdr.ramdisk_image = initrd_start;
1326 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1327 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1328 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1329 
1330 	if (!xen_initial_domain()) {
1331 		if (pci_xen)
1332 			x86_init.pci.arch_init = pci_xen_init;
1333 		x86_platform.set_legacy_features =
1334 				xen_domu_set_legacy_features;
1335 	} else {
1336 		const struct dom0_vga_console_info *info =
1337 			(void *)((char *)xen_start_info +
1338 				 xen_start_info->console.dom0.info_off);
1339 		struct xen_platform_op op = {
1340 			.cmd = XENPF_firmware_info,
1341 			.interface_version = XENPF_INTERFACE_VERSION,
1342 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1343 		};
1344 
1345 		x86_platform.set_legacy_features =
1346 				xen_dom0_set_legacy_features;
1347 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1348 		xen_start_info->console.domU.mfn = 0;
1349 		xen_start_info->console.domU.evtchn = 0;
1350 
1351 		if (HYPERVISOR_platform_op(&op) == 0)
1352 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1353 
1354 		/* Make sure ACS will be enabled */
1355 		pci_request_acs();
1356 
1357 		xen_acpi_sleep_register();
1358 
1359 		xen_boot_params_init_edd();
1360 
1361 #ifdef CONFIG_ACPI
1362 		/*
1363 		 * Disable selecting "Firmware First mode" for correctable
1364 		 * memory errors, as this is the duty of the hypervisor to
1365 		 * decide.
1366 		 */
1367 		acpi_disable_cmcff = 1;
1368 #endif
1369 	}
1370 
1371 	xen_add_preferred_consoles();
1372 
1373 #ifdef CONFIG_PCI
1374 	/* PCI BIOS service won't work from a PV guest. */
1375 	pci_probe &= ~PCI_PROBE_BIOS;
1376 #endif
1377 	xen_raw_console_write("about to get started...\n");
1378 
1379 	/* We need this for printk timestamps */
1380 	xen_setup_runstate_info(0);
1381 
1382 	xen_efi_init(&boot_params);
1383 
1384 	/* Start the world */
1385 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1386 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1387 }
1388 
1389 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1390 {
1391 	int rc;
1392 
1393 	if (per_cpu(xen_vcpu, cpu) == NULL)
1394 		return -ENODEV;
1395 
1396 	xen_setup_timer(cpu);
1397 
1398 	rc = xen_smp_intr_init(cpu);
1399 	if (rc) {
1400 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1401 		     cpu, rc);
1402 		return rc;
1403 	}
1404 
1405 	rc = xen_smp_intr_init_pv(cpu);
1406 	if (rc) {
1407 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1408 		     cpu, rc);
1409 		return rc;
1410 	}
1411 
1412 	return 0;
1413 }
1414 
1415 static int xen_cpu_dead_pv(unsigned int cpu)
1416 {
1417 	xen_smp_intr_free(cpu);
1418 	xen_smp_intr_free_pv(cpu);
1419 
1420 	xen_teardown_timer(cpu);
1421 
1422 	return 0;
1423 }
1424 
1425 static uint32_t __init xen_platform_pv(void)
1426 {
1427 	if (xen_pv_domain())
1428 		return xen_cpuid_base();
1429 
1430 	return 0;
1431 }
1432 
1433 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1434 	.name                   = "Xen PV",
1435 	.detect                 = xen_platform_pv,
1436 	.type			= X86_HYPER_XEN_PV,
1437 	.runtime.pin_vcpu       = xen_pin_vcpu,
1438 	.ignore_nopv		= true,
1439 };
1440