xref: /openbmc/linux/arch/x86/xen/enlighten_pv.c (revision 83b975b5)
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 __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
112 
113 static int __init parse_xen_msr_safe(char *str)
114 {
115 	if (str)
116 		return strtobool(str, &xen_msr_safe);
117 	return -EINVAL;
118 }
119 early_param("xen_msr_safe", parse_xen_msr_safe);
120 
121 static void __init xen_pv_init_platform(void)
122 {
123 	/* PV guests can't operate virtio devices without grants. */
124 	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
125 		virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
126 
127 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
128 
129 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
130 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
131 
132 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
133 	xen_vcpu_info_reset(0);
134 
135 	/* pvclock is in shared info area */
136 	xen_init_time_ops();
137 }
138 
139 static void __init xen_pv_guest_late_init(void)
140 {
141 #ifndef CONFIG_SMP
142 	/* Setup shared vcpu info for non-smp configurations */
143 	xen_setup_vcpu_info_placement();
144 #endif
145 }
146 
147 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
148 static __read_mostly unsigned int cpuid_leaf5_edx_val;
149 
150 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
151 		      unsigned int *cx, unsigned int *dx)
152 {
153 	unsigned maskebx = ~0;
154 
155 	/*
156 	 * Mask out inconvenient features, to try and disable as many
157 	 * unsupported kernel subsystems as possible.
158 	 */
159 	switch (*ax) {
160 	case CPUID_MWAIT_LEAF:
161 		/* Synthesize the values.. */
162 		*ax = 0;
163 		*bx = 0;
164 		*cx = cpuid_leaf5_ecx_val;
165 		*dx = cpuid_leaf5_edx_val;
166 		return;
167 
168 	case 0xb:
169 		/* Suppress extended topology stuff */
170 		maskebx = 0;
171 		break;
172 	}
173 
174 	asm(XEN_EMULATE_PREFIX "cpuid"
175 		: "=a" (*ax),
176 		  "=b" (*bx),
177 		  "=c" (*cx),
178 		  "=d" (*dx)
179 		: "0" (*ax), "2" (*cx));
180 
181 	*bx &= maskebx;
182 }
183 
184 static bool __init xen_check_mwait(void)
185 {
186 #ifdef CONFIG_ACPI
187 	struct xen_platform_op op = {
188 		.cmd			= XENPF_set_processor_pminfo,
189 		.u.set_pminfo.id	= -1,
190 		.u.set_pminfo.type	= XEN_PM_PDC,
191 	};
192 	uint32_t buf[3];
193 	unsigned int ax, bx, cx, dx;
194 	unsigned int mwait_mask;
195 
196 	/* We need to determine whether it is OK to expose the MWAIT
197 	 * capability to the kernel to harvest deeper than C3 states from ACPI
198 	 * _CST using the processor_harvest_xen.c module. For this to work, we
199 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
200 	 * checks against). The hypervisor won't expose the MWAIT flag because
201 	 * it would break backwards compatibility; so we will find out directly
202 	 * from the hardware and hypercall.
203 	 */
204 	if (!xen_initial_domain())
205 		return false;
206 
207 	/*
208 	 * When running under platform earlier than Xen4.2, do not expose
209 	 * mwait, to avoid the risk of loading native acpi pad driver
210 	 */
211 	if (!xen_running_on_version_or_later(4, 2))
212 		return false;
213 
214 	ax = 1;
215 	cx = 0;
216 
217 	native_cpuid(&ax, &bx, &cx, &dx);
218 
219 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
220 		     (1 << (X86_FEATURE_MWAIT % 32));
221 
222 	if ((cx & mwait_mask) != mwait_mask)
223 		return false;
224 
225 	/* We need to emulate the MWAIT_LEAF and for that we need both
226 	 * ecx and edx. The hypercall provides only partial information.
227 	 */
228 
229 	ax = CPUID_MWAIT_LEAF;
230 	bx = 0;
231 	cx = 0;
232 	dx = 0;
233 
234 	native_cpuid(&ax, &bx, &cx, &dx);
235 
236 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
237 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
238 	 */
239 	buf[0] = ACPI_PDC_REVISION_ID;
240 	buf[1] = 1;
241 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
242 
243 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
244 
245 	if ((HYPERVISOR_platform_op(&op) == 0) &&
246 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
247 		cpuid_leaf5_ecx_val = cx;
248 		cpuid_leaf5_edx_val = dx;
249 	}
250 	return true;
251 #else
252 	return false;
253 #endif
254 }
255 
256 static bool __init xen_check_xsave(void)
257 {
258 	unsigned int cx, xsave_mask;
259 
260 	cx = cpuid_ecx(1);
261 
262 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
263 		     (1 << (X86_FEATURE_OSXSAVE % 32));
264 
265 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
266 	return (cx & xsave_mask) == xsave_mask;
267 }
268 
269 static void __init xen_init_capabilities(void)
270 {
271 	setup_force_cpu_cap(X86_FEATURE_XENPV);
272 	setup_clear_cpu_cap(X86_FEATURE_DCA);
273 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
274 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
275 	setup_clear_cpu_cap(X86_FEATURE_ACC);
276 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
277 	setup_clear_cpu_cap(X86_FEATURE_SME);
278 
279 	/*
280 	 * Xen PV would need some work to support PCID: CR3 handling as well
281 	 * as xen_flush_tlb_others() would need updating.
282 	 */
283 	setup_clear_cpu_cap(X86_FEATURE_PCID);
284 
285 	if (!xen_initial_domain())
286 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
287 
288 	if (xen_check_mwait())
289 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
290 	else
291 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
292 
293 	if (!xen_check_xsave()) {
294 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
295 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
296 	}
297 }
298 
299 static noinstr void xen_set_debugreg(int reg, unsigned long val)
300 {
301 	HYPERVISOR_set_debugreg(reg, val);
302 }
303 
304 static noinstr unsigned long xen_get_debugreg(int reg)
305 {
306 	return HYPERVISOR_get_debugreg(reg);
307 }
308 
309 static void xen_end_context_switch(struct task_struct *next)
310 {
311 	xen_mc_flush();
312 	paravirt_end_context_switch(next);
313 }
314 
315 static unsigned long xen_store_tr(void)
316 {
317 	return 0;
318 }
319 
320 /*
321  * Set the page permissions for a particular virtual address.  If the
322  * address is a vmalloc mapping (or other non-linear mapping), then
323  * find the linear mapping of the page and also set its protections to
324  * match.
325  */
326 static void set_aliased_prot(void *v, pgprot_t prot)
327 {
328 	int level;
329 	pte_t *ptep;
330 	pte_t pte;
331 	unsigned long pfn;
332 	unsigned char dummy;
333 	void *va;
334 
335 	ptep = lookup_address((unsigned long)v, &level);
336 	BUG_ON(ptep == NULL);
337 
338 	pfn = pte_pfn(*ptep);
339 	pte = pfn_pte(pfn, prot);
340 
341 	/*
342 	 * Careful: update_va_mapping() will fail if the virtual address
343 	 * we're poking isn't populated in the page tables.  We don't
344 	 * need to worry about the direct map (that's always in the page
345 	 * tables), but we need to be careful about vmap space.  In
346 	 * particular, the top level page table can lazily propagate
347 	 * entries between processes, so if we've switched mms since we
348 	 * vmapped the target in the first place, we might not have the
349 	 * top-level page table entry populated.
350 	 *
351 	 * We disable preemption because we want the same mm active when
352 	 * we probe the target and when we issue the hypercall.  We'll
353 	 * have the same nominal mm, but if we're a kernel thread, lazy
354 	 * mm dropping could change our pgd.
355 	 *
356 	 * Out of an abundance of caution, this uses __get_user() to fault
357 	 * in the target address just in case there's some obscure case
358 	 * in which the target address isn't readable.
359 	 */
360 
361 	preempt_disable();
362 
363 	copy_from_kernel_nofault(&dummy, v, 1);
364 
365 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
366 		BUG();
367 
368 	va = __va(PFN_PHYS(pfn));
369 
370 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
371 		BUG();
372 
373 	preempt_enable();
374 }
375 
376 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
377 {
378 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
379 	int i;
380 
381 	/*
382 	 * We need to mark the all aliases of the LDT pages RO.  We
383 	 * don't need to call vm_flush_aliases(), though, since that's
384 	 * only responsible for flushing aliases out the TLBs, not the
385 	 * page tables, and Xen will flush the TLB for us if needed.
386 	 *
387 	 * To avoid confusing future readers: none of this is necessary
388 	 * to load the LDT.  The hypervisor only checks this when the
389 	 * LDT is faulted in due to subsequent descriptor access.
390 	 */
391 
392 	for (i = 0; i < entries; i += entries_per_page)
393 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
394 }
395 
396 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
397 {
398 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
399 	int i;
400 
401 	for (i = 0; i < entries; i += entries_per_page)
402 		set_aliased_prot(ldt + i, PAGE_KERNEL);
403 }
404 
405 static void xen_set_ldt(const void *addr, unsigned entries)
406 {
407 	struct mmuext_op *op;
408 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
409 
410 	trace_xen_cpu_set_ldt(addr, entries);
411 
412 	op = mcs.args;
413 	op->cmd = MMUEXT_SET_LDT;
414 	op->arg1.linear_addr = (unsigned long)addr;
415 	op->arg2.nr_ents = entries;
416 
417 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
418 
419 	xen_mc_issue(PARAVIRT_LAZY_CPU);
420 }
421 
422 static void xen_load_gdt(const struct desc_ptr *dtr)
423 {
424 	unsigned long va = dtr->address;
425 	unsigned int size = dtr->size + 1;
426 	unsigned long pfn, mfn;
427 	int level;
428 	pte_t *ptep;
429 	void *virt;
430 
431 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
432 	BUG_ON(size > PAGE_SIZE);
433 	BUG_ON(va & ~PAGE_MASK);
434 
435 	/*
436 	 * The GDT is per-cpu and is in the percpu data area.
437 	 * That can be virtually mapped, so we need to do a
438 	 * page-walk to get the underlying MFN for the
439 	 * hypercall.  The page can also be in the kernel's
440 	 * linear range, so we need to RO that mapping too.
441 	 */
442 	ptep = lookup_address(va, &level);
443 	BUG_ON(ptep == NULL);
444 
445 	pfn = pte_pfn(*ptep);
446 	mfn = pfn_to_mfn(pfn);
447 	virt = __va(PFN_PHYS(pfn));
448 
449 	make_lowmem_page_readonly((void *)va);
450 	make_lowmem_page_readonly(virt);
451 
452 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
453 		BUG();
454 }
455 
456 /*
457  * load_gdt for early boot, when the gdt is only mapped once
458  */
459 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
460 {
461 	unsigned long va = dtr->address;
462 	unsigned int size = dtr->size + 1;
463 	unsigned long pfn, mfn;
464 	pte_t pte;
465 
466 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
467 	BUG_ON(size > PAGE_SIZE);
468 	BUG_ON(va & ~PAGE_MASK);
469 
470 	pfn = virt_to_pfn(va);
471 	mfn = pfn_to_mfn(pfn);
472 
473 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
474 
475 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
476 		BUG();
477 
478 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
479 		BUG();
480 }
481 
482 static inline bool desc_equal(const struct desc_struct *d1,
483 			      const struct desc_struct *d2)
484 {
485 	return !memcmp(d1, d2, sizeof(*d1));
486 }
487 
488 static void load_TLS_descriptor(struct thread_struct *t,
489 				unsigned int cpu, unsigned int i)
490 {
491 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
492 	struct desc_struct *gdt;
493 	xmaddr_t maddr;
494 	struct multicall_space mc;
495 
496 	if (desc_equal(shadow, &t->tls_array[i]))
497 		return;
498 
499 	*shadow = t->tls_array[i];
500 
501 	gdt = get_cpu_gdt_rw(cpu);
502 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
503 	mc = __xen_mc_entry(0);
504 
505 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
506 }
507 
508 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
509 {
510 	/*
511 	 * In lazy mode we need to zero %fs, otherwise we may get an
512 	 * exception between the new %fs descriptor being loaded and
513 	 * %fs being effectively cleared at __switch_to().
514 	 */
515 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU)
516 		loadsegment(fs, 0);
517 
518 	xen_mc_batch();
519 
520 	load_TLS_descriptor(t, cpu, 0);
521 	load_TLS_descriptor(t, cpu, 1);
522 	load_TLS_descriptor(t, cpu, 2);
523 
524 	xen_mc_issue(PARAVIRT_LAZY_CPU);
525 }
526 
527 static void xen_load_gs_index(unsigned int idx)
528 {
529 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
530 		BUG();
531 }
532 
533 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
534 				const void *ptr)
535 {
536 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
537 	u64 entry = *(u64 *)ptr;
538 
539 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
540 
541 	preempt_disable();
542 
543 	xen_mc_flush();
544 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
545 		BUG();
546 
547 	preempt_enable();
548 }
549 
550 void noist_exc_debug(struct pt_regs *regs);
551 
552 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
553 {
554 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
555 	exc_nmi(regs);
556 }
557 
558 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
559 {
560 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
561 	exc_double_fault(regs, error_code);
562 }
563 
564 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
565 {
566 	/*
567 	 * There's no IST on Xen PV, but we still need to dispatch
568 	 * to the correct handler.
569 	 */
570 	if (user_mode(regs))
571 		noist_exc_debug(regs);
572 	else
573 		exc_debug(regs);
574 }
575 
576 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
577 {
578 	/* This should never happen and there is no way to handle it. */
579 	instrumentation_begin();
580 	pr_err("Unknown trap in Xen PV mode.");
581 	BUG();
582 	instrumentation_end();
583 }
584 
585 #ifdef CONFIG_X86_MCE
586 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
587 {
588 	/*
589 	 * There's no IST on Xen PV, but we still need to dispatch
590 	 * to the correct handler.
591 	 */
592 	if (user_mode(regs))
593 		noist_exc_machine_check(regs);
594 	else
595 		exc_machine_check(regs);
596 }
597 #endif
598 
599 struct trap_array_entry {
600 	void (*orig)(void);
601 	void (*xen)(void);
602 	bool ist_okay;
603 };
604 
605 #define TRAP_ENTRY(func, ist_ok) {			\
606 	.orig		= asm_##func,			\
607 	.xen		= xen_asm_##func,		\
608 	.ist_okay	= ist_ok }
609 
610 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
611 	.orig		= asm_##func,			\
612 	.xen		= xen_asm_xenpv_##func,		\
613 	.ist_okay	= ist_ok }
614 
615 static struct trap_array_entry trap_array[] = {
616 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
617 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
618 #ifdef CONFIG_X86_MCE
619 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
620 #endif
621 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
622 	TRAP_ENTRY(exc_int3,				false ),
623 	TRAP_ENTRY(exc_overflow,			false ),
624 #ifdef CONFIG_IA32_EMULATION
625 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
626 #endif
627 	TRAP_ENTRY(exc_page_fault,			false ),
628 	TRAP_ENTRY(exc_divide_error,			false ),
629 	TRAP_ENTRY(exc_bounds,				false ),
630 	TRAP_ENTRY(exc_invalid_op,			false ),
631 	TRAP_ENTRY(exc_device_not_available,		false ),
632 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
633 	TRAP_ENTRY(exc_invalid_tss,			false ),
634 	TRAP_ENTRY(exc_segment_not_present,		false ),
635 	TRAP_ENTRY(exc_stack_segment,			false ),
636 	TRAP_ENTRY(exc_general_protection,		false ),
637 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
638 	TRAP_ENTRY(exc_coprocessor_error,		false ),
639 	TRAP_ENTRY(exc_alignment_check,			false ),
640 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
641 #ifdef CONFIG_X86_KERNEL_IBT
642 	TRAP_ENTRY(exc_control_protection,		false ),
643 #endif
644 };
645 
646 static bool __ref get_trap_addr(void **addr, unsigned int ist)
647 {
648 	unsigned int nr;
649 	bool ist_okay = false;
650 	bool found = false;
651 
652 	/*
653 	 * Replace trap handler addresses by Xen specific ones.
654 	 * Check for known traps using IST and whitelist them.
655 	 * The debugger ones are the only ones we care about.
656 	 * Xen will handle faults like double_fault, so we should never see
657 	 * them.  Warn if there's an unexpected IST-using fault handler.
658 	 */
659 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
660 		struct trap_array_entry *entry = trap_array + nr;
661 
662 		if (*addr == entry->orig) {
663 			*addr = entry->xen;
664 			ist_okay = entry->ist_okay;
665 			found = true;
666 			break;
667 		}
668 	}
669 
670 	if (nr == ARRAY_SIZE(trap_array) &&
671 	    *addr >= (void *)early_idt_handler_array[0] &&
672 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
673 		nr = (*addr - (void *)early_idt_handler_array[0]) /
674 		     EARLY_IDT_HANDLER_SIZE;
675 		*addr = (void *)xen_early_idt_handler_array[nr];
676 		found = true;
677 	}
678 
679 	if (!found)
680 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
681 
682 	if (WARN_ON(found && ist != 0 && !ist_okay))
683 		return false;
684 
685 	return true;
686 }
687 
688 static int cvt_gate_to_trap(int vector, const gate_desc *val,
689 			    struct trap_info *info)
690 {
691 	unsigned long addr;
692 
693 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
694 		return 0;
695 
696 	info->vector = vector;
697 
698 	addr = gate_offset(val);
699 	if (!get_trap_addr((void **)&addr, val->bits.ist))
700 		return 0;
701 	info->address = addr;
702 
703 	info->cs = gate_segment(val);
704 	info->flags = val->bits.dpl;
705 	/* interrupt gates clear IF */
706 	if (val->bits.type == GATE_INTERRUPT)
707 		info->flags |= 1 << 2;
708 
709 	return 1;
710 }
711 
712 /* Locations of each CPU's IDT */
713 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
714 
715 /* Set an IDT entry.  If the entry is part of the current IDT, then
716    also update Xen. */
717 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
718 {
719 	unsigned long p = (unsigned long)&dt[entrynum];
720 	unsigned long start, end;
721 
722 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
723 
724 	preempt_disable();
725 
726 	start = __this_cpu_read(idt_desc.address);
727 	end = start + __this_cpu_read(idt_desc.size) + 1;
728 
729 	xen_mc_flush();
730 
731 	native_write_idt_entry(dt, entrynum, g);
732 
733 	if (p >= start && (p + 8) <= end) {
734 		struct trap_info info[2];
735 
736 		info[1].address = 0;
737 
738 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
739 			if (HYPERVISOR_set_trap_table(info))
740 				BUG();
741 	}
742 
743 	preempt_enable();
744 }
745 
746 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
747 				      struct trap_info *traps, bool full)
748 {
749 	unsigned in, out, count;
750 
751 	count = (desc->size+1) / sizeof(gate_desc);
752 	BUG_ON(count > 256);
753 
754 	for (in = out = 0; in < count; in++) {
755 		gate_desc *entry = (gate_desc *)(desc->address) + in;
756 
757 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
758 			out++;
759 	}
760 
761 	return out;
762 }
763 
764 void xen_copy_trap_info(struct trap_info *traps)
765 {
766 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
767 
768 	xen_convert_trap_info(desc, traps, true);
769 }
770 
771 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
772    hold a spinlock to protect the static traps[] array (static because
773    it avoids allocation, and saves stack space). */
774 static void xen_load_idt(const struct desc_ptr *desc)
775 {
776 	static DEFINE_SPINLOCK(lock);
777 	static struct trap_info traps[257];
778 	static const struct trap_info zero = { };
779 	unsigned out;
780 
781 	trace_xen_cpu_load_idt(desc);
782 
783 	spin_lock(&lock);
784 
785 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
786 
787 	out = xen_convert_trap_info(desc, traps, false);
788 	traps[out] = zero;
789 
790 	xen_mc_flush();
791 	if (HYPERVISOR_set_trap_table(traps))
792 		BUG();
793 
794 	spin_unlock(&lock);
795 }
796 
797 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
798    they're handled differently. */
799 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
800 				const void *desc, int type)
801 {
802 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
803 
804 	preempt_disable();
805 
806 	switch (type) {
807 	case DESC_LDT:
808 	case DESC_TSS:
809 		/* ignore */
810 		break;
811 
812 	default: {
813 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
814 
815 		xen_mc_flush();
816 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
817 			BUG();
818 	}
819 
820 	}
821 
822 	preempt_enable();
823 }
824 
825 /*
826  * Version of write_gdt_entry for use at early boot-time needed to
827  * update an entry as simply as possible.
828  */
829 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
830 					    const void *desc, int type)
831 {
832 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
833 
834 	switch (type) {
835 	case DESC_LDT:
836 	case DESC_TSS:
837 		/* ignore */
838 		break;
839 
840 	default: {
841 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
842 
843 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
844 			dt[entry] = *(struct desc_struct *)desc;
845 	}
846 
847 	}
848 }
849 
850 static void xen_load_sp0(unsigned long sp0)
851 {
852 	struct multicall_space mcs;
853 
854 	mcs = xen_mc_entry(0);
855 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
856 	xen_mc_issue(PARAVIRT_LAZY_CPU);
857 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
858 }
859 
860 #ifdef CONFIG_X86_IOPL_IOPERM
861 static void xen_invalidate_io_bitmap(void)
862 {
863 	struct physdev_set_iobitmap iobitmap = {
864 		.bitmap = NULL,
865 		.nr_ports = 0,
866 	};
867 
868 	native_tss_invalidate_io_bitmap();
869 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
870 }
871 
872 static void xen_update_io_bitmap(void)
873 {
874 	struct physdev_set_iobitmap iobitmap;
875 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
876 
877 	native_tss_update_io_bitmap();
878 
879 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
880 			  tss->x86_tss.io_bitmap_base;
881 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
882 		iobitmap.nr_ports = 0;
883 	else
884 		iobitmap.nr_ports = IO_BITMAP_BITS;
885 
886 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
887 }
888 #endif
889 
890 static void xen_io_delay(void)
891 {
892 }
893 
894 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
895 
896 static unsigned long xen_read_cr0(void)
897 {
898 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
899 
900 	if (unlikely(cr0 == 0)) {
901 		cr0 = native_read_cr0();
902 		this_cpu_write(xen_cr0_value, cr0);
903 	}
904 
905 	return cr0;
906 }
907 
908 static void xen_write_cr0(unsigned long cr0)
909 {
910 	struct multicall_space mcs;
911 
912 	this_cpu_write(xen_cr0_value, cr0);
913 
914 	/* Only pay attention to cr0.TS; everything else is
915 	   ignored. */
916 	mcs = xen_mc_entry(0);
917 
918 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
919 
920 	xen_mc_issue(PARAVIRT_LAZY_CPU);
921 }
922 
923 static void xen_write_cr4(unsigned long cr4)
924 {
925 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
926 
927 	native_write_cr4(cr4);
928 }
929 
930 static u64 xen_do_read_msr(unsigned int msr, int *err)
931 {
932 	u64 val = 0;	/* Avoid uninitialized value for safe variant. */
933 
934 	if (pmu_msr_read(msr, &val, err))
935 		return val;
936 
937 	if (err)
938 		val = native_read_msr_safe(msr, err);
939 	else
940 		val = native_read_msr(msr);
941 
942 	switch (msr) {
943 	case MSR_IA32_APICBASE:
944 		val &= ~X2APIC_ENABLE;
945 		break;
946 	}
947 	return val;
948 }
949 
950 static void set_seg(unsigned int which, unsigned int low, unsigned int high,
951 		    int *err)
952 {
953 	u64 base = ((u64)high << 32) | low;
954 
955 	if (HYPERVISOR_set_segment_base(which, base) == 0)
956 		return;
957 
958 	if (err)
959 		*err = -EIO;
960 	else
961 		WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
962 }
963 
964 /*
965  * Support write_msr_safe() and write_msr() semantics.
966  * With err == NULL write_msr() semantics are selected.
967  * Supplying an err pointer requires err to be pre-initialized with 0.
968  */
969 static void xen_do_write_msr(unsigned int msr, unsigned int low,
970 			     unsigned int high, int *err)
971 {
972 	switch (msr) {
973 	case MSR_FS_BASE:
974 		set_seg(SEGBASE_FS, low, high, err);
975 		break;
976 
977 	case MSR_KERNEL_GS_BASE:
978 		set_seg(SEGBASE_GS_USER, low, high, err);
979 		break;
980 
981 	case MSR_GS_BASE:
982 		set_seg(SEGBASE_GS_KERNEL, low, high, err);
983 		break;
984 
985 	case MSR_STAR:
986 	case MSR_CSTAR:
987 	case MSR_LSTAR:
988 	case MSR_SYSCALL_MASK:
989 	case MSR_IA32_SYSENTER_CS:
990 	case MSR_IA32_SYSENTER_ESP:
991 	case MSR_IA32_SYSENTER_EIP:
992 		/* Fast syscall setup is all done in hypercalls, so
993 		   these are all ignored.  Stub them out here to stop
994 		   Xen console noise. */
995 		break;
996 
997 	default:
998 		if (!pmu_msr_write(msr, low, high, err)) {
999 			if (err)
1000 				*err = native_write_msr_safe(msr, low, high);
1001 			else
1002 				native_write_msr(msr, low, high);
1003 		}
1004 	}
1005 }
1006 
1007 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1008 {
1009 	return xen_do_read_msr(msr, err);
1010 }
1011 
1012 static int xen_write_msr_safe(unsigned int msr, unsigned int low,
1013 			      unsigned int high)
1014 {
1015 	int err = 0;
1016 
1017 	xen_do_write_msr(msr, low, high, &err);
1018 
1019 	return err;
1020 }
1021 
1022 static u64 xen_read_msr(unsigned int msr)
1023 {
1024 	int err;
1025 
1026 	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1027 }
1028 
1029 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1030 {
1031 	int err;
1032 
1033 	xen_do_write_msr(msr, low, high, xen_msr_safe ? &err : NULL);
1034 }
1035 
1036 /* This is called once we have the cpu_possible_mask */
1037 void __init xen_setup_vcpu_info_placement(void)
1038 {
1039 	int cpu;
1040 
1041 	for_each_possible_cpu(cpu) {
1042 		/* Set up direct vCPU id mapping for PV guests. */
1043 		per_cpu(xen_vcpu_id, cpu) = cpu;
1044 		xen_vcpu_setup(cpu);
1045 	}
1046 
1047 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1048 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1049 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1050 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1051 }
1052 
1053 static const struct pv_info xen_info __initconst = {
1054 	.extra_user_64bit_cs = FLAT_USER_CS64,
1055 	.name = "Xen",
1056 };
1057 
1058 static const typeof(pv_ops) xen_cpu_ops __initconst = {
1059 	.cpu = {
1060 		.cpuid = xen_cpuid,
1061 
1062 		.set_debugreg = xen_set_debugreg,
1063 		.get_debugreg = xen_get_debugreg,
1064 
1065 		.read_cr0 = xen_read_cr0,
1066 		.write_cr0 = xen_write_cr0,
1067 
1068 		.write_cr4 = xen_write_cr4,
1069 
1070 		.wbinvd = native_wbinvd,
1071 
1072 		.read_msr = xen_read_msr,
1073 		.write_msr = xen_write_msr,
1074 
1075 		.read_msr_safe = xen_read_msr_safe,
1076 		.write_msr_safe = xen_write_msr_safe,
1077 
1078 		.read_pmc = xen_read_pmc,
1079 
1080 		.load_tr_desc = paravirt_nop,
1081 		.set_ldt = xen_set_ldt,
1082 		.load_gdt = xen_load_gdt,
1083 		.load_idt = xen_load_idt,
1084 		.load_tls = xen_load_tls,
1085 		.load_gs_index = xen_load_gs_index,
1086 
1087 		.alloc_ldt = xen_alloc_ldt,
1088 		.free_ldt = xen_free_ldt,
1089 
1090 		.store_tr = xen_store_tr,
1091 
1092 		.write_ldt_entry = xen_write_ldt_entry,
1093 		.write_gdt_entry = xen_write_gdt_entry,
1094 		.write_idt_entry = xen_write_idt_entry,
1095 		.load_sp0 = xen_load_sp0,
1096 
1097 #ifdef CONFIG_X86_IOPL_IOPERM
1098 		.invalidate_io_bitmap = xen_invalidate_io_bitmap,
1099 		.update_io_bitmap = xen_update_io_bitmap,
1100 #endif
1101 		.io_delay = xen_io_delay,
1102 
1103 		.start_context_switch = paravirt_start_context_switch,
1104 		.end_context_switch = xen_end_context_switch,
1105 	},
1106 };
1107 
1108 static void xen_restart(char *msg)
1109 {
1110 	xen_reboot(SHUTDOWN_reboot);
1111 }
1112 
1113 static void xen_machine_halt(void)
1114 {
1115 	xen_reboot(SHUTDOWN_poweroff);
1116 }
1117 
1118 static void xen_machine_power_off(void)
1119 {
1120 	do_kernel_power_off();
1121 	xen_reboot(SHUTDOWN_poweroff);
1122 }
1123 
1124 static void xen_crash_shutdown(struct pt_regs *regs)
1125 {
1126 	xen_reboot(SHUTDOWN_crash);
1127 }
1128 
1129 static const struct machine_ops xen_machine_ops __initconst = {
1130 	.restart = xen_restart,
1131 	.halt = xen_machine_halt,
1132 	.power_off = xen_machine_power_off,
1133 	.shutdown = xen_machine_halt,
1134 	.crash_shutdown = xen_crash_shutdown,
1135 	.emergency_restart = xen_emergency_restart,
1136 };
1137 
1138 static unsigned char xen_get_nmi_reason(void)
1139 {
1140 	unsigned char reason = 0;
1141 
1142 	/* Construct a value which looks like it came from port 0x61. */
1143 	if (test_bit(_XEN_NMIREASON_io_error,
1144 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1145 		reason |= NMI_REASON_IOCHK;
1146 	if (test_bit(_XEN_NMIREASON_pci_serr,
1147 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1148 		reason |= NMI_REASON_SERR;
1149 
1150 	return reason;
1151 }
1152 
1153 static void __init xen_boot_params_init_edd(void)
1154 {
1155 #if IS_ENABLED(CONFIG_EDD)
1156 	struct xen_platform_op op;
1157 	struct edd_info *edd_info;
1158 	u32 *mbr_signature;
1159 	unsigned nr;
1160 	int ret;
1161 
1162 	edd_info = boot_params.eddbuf;
1163 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1164 
1165 	op.cmd = XENPF_firmware_info;
1166 
1167 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1168 	for (nr = 0; nr < EDDMAXNR; nr++) {
1169 		struct edd_info *info = edd_info + nr;
1170 
1171 		op.u.firmware_info.index = nr;
1172 		info->params.length = sizeof(info->params);
1173 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1174 				     &info->params);
1175 		ret = HYPERVISOR_platform_op(&op);
1176 		if (ret)
1177 			break;
1178 
1179 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1180 		C(device);
1181 		C(version);
1182 		C(interface_support);
1183 		C(legacy_max_cylinder);
1184 		C(legacy_max_head);
1185 		C(legacy_sectors_per_track);
1186 #undef C
1187 	}
1188 	boot_params.eddbuf_entries = nr;
1189 
1190 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1191 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1192 		op.u.firmware_info.index = nr;
1193 		ret = HYPERVISOR_platform_op(&op);
1194 		if (ret)
1195 			break;
1196 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1197 	}
1198 	boot_params.edd_mbr_sig_buf_entries = nr;
1199 #endif
1200 }
1201 
1202 /*
1203  * Set up the GDT and segment registers for -fstack-protector.  Until
1204  * we do this, we have to be careful not to call any stack-protected
1205  * function, which is most of the kernel.
1206  */
1207 static void __init xen_setup_gdt(int cpu)
1208 {
1209 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1210 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1211 
1212 	switch_to_new_gdt(cpu);
1213 
1214 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1215 	pv_ops.cpu.load_gdt = xen_load_gdt;
1216 }
1217 
1218 static void __init xen_dom0_set_legacy_features(void)
1219 {
1220 	x86_platform.legacy.rtc = 1;
1221 }
1222 
1223 static void __init xen_domu_set_legacy_features(void)
1224 {
1225 	x86_platform.legacy.rtc = 0;
1226 }
1227 
1228 extern void early_xen_iret_patch(void);
1229 
1230 /* First C function to be called on Xen boot */
1231 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1232 {
1233 	struct physdev_set_iopl set_iopl;
1234 	unsigned long initrd_start = 0;
1235 	int rc;
1236 
1237 	if (!si)
1238 		return;
1239 
1240 	clear_bss();
1241 
1242 	xen_start_info = si;
1243 
1244 	__text_gen_insn(&early_xen_iret_patch,
1245 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1246 			JMP32_INSN_SIZE);
1247 
1248 	xen_domain_type = XEN_PV_DOMAIN;
1249 	xen_start_flags = xen_start_info->flags;
1250 
1251 	xen_setup_features();
1252 
1253 	/* Install Xen paravirt ops */
1254 	pv_info = xen_info;
1255 	pv_ops.cpu = xen_cpu_ops.cpu;
1256 	xen_init_irq_ops();
1257 
1258 	/*
1259 	 * Setup xen_vcpu early because it is needed for
1260 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1261 	 *
1262 	 * Don't do the full vcpu_info placement stuff until we have
1263 	 * the cpu_possible_mask and a non-dummy shared_info.
1264 	 */
1265 	xen_vcpu_info_reset(0);
1266 
1267 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1268 
1269 	x86_init.resources.memory_setup = xen_memory_setup;
1270 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1271 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1272 	x86_init.oem.arch_setup = xen_arch_setup;
1273 	x86_init.oem.banner = xen_banner;
1274 	x86_init.hyper.init_platform = xen_pv_init_platform;
1275 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1276 
1277 	/*
1278 	 * Set up some pagetable state before starting to set any ptes.
1279 	 */
1280 
1281 	xen_setup_machphys_mapping();
1282 	xen_init_mmu_ops();
1283 
1284 	/* Prevent unwanted bits from being set in PTEs. */
1285 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1286 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1287 
1288 	/* Get mfn list */
1289 	xen_build_dynamic_phys_to_machine();
1290 
1291 	/* Work out if we support NX */
1292 	get_cpu_cap(&boot_cpu_data);
1293 	x86_configure_nx();
1294 
1295 	/*
1296 	 * Set up kernel GDT and segment registers, mainly so that
1297 	 * -fstack-protector code can be executed.
1298 	 */
1299 	xen_setup_gdt(0);
1300 
1301 	/* Determine virtual and physical address sizes */
1302 	get_cpu_address_sizes(&boot_cpu_data);
1303 
1304 	/* Let's presume PV guests always boot on vCPU with id 0. */
1305 	per_cpu(xen_vcpu_id, 0) = 0;
1306 
1307 	idt_setup_early_handler();
1308 
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 	machine_ops = xen_machine_ops;
1319 
1320 	/*
1321 	 * The only reliable way to retain the initial address of the
1322 	 * percpu gdt_page is to remember it here, so we can go and
1323 	 * mark it RW later, when the initial percpu area is freed.
1324 	 */
1325 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1326 
1327 	xen_smp_init();
1328 
1329 #ifdef CONFIG_ACPI_NUMA
1330 	/*
1331 	 * The pages we from Xen are not related to machine pages, so
1332 	 * any NUMA information the kernel tries to get from ACPI will
1333 	 * be meaningless.  Prevent it from trying.
1334 	 */
1335 	disable_srat();
1336 #endif
1337 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1338 
1339 	local_irq_disable();
1340 	early_boot_irqs_disabled = true;
1341 
1342 	xen_raw_console_write("mapping kernel into physical memory\n");
1343 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1344 				   xen_start_info->nr_pages);
1345 	xen_reserve_special_pages();
1346 
1347 	/*
1348 	 * We used to do this in xen_arch_setup, but that is too late
1349 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1350 	 * early_amd_init which pokes 0xcf8 port.
1351 	 */
1352 	set_iopl.iopl = 1;
1353 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1354 	if (rc != 0)
1355 		xen_raw_printk("physdev_op failed %d\n", rc);
1356 
1357 
1358 	if (xen_start_info->mod_start) {
1359 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1360 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1361 	    else
1362 		initrd_start = __pa(xen_start_info->mod_start);
1363 	}
1364 
1365 	/* Poke various useful things into boot_params */
1366 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1367 	boot_params.hdr.ramdisk_image = initrd_start;
1368 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1369 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1370 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1371 
1372 	if (!xen_initial_domain()) {
1373 		if (pci_xen)
1374 			x86_init.pci.arch_init = pci_xen_init;
1375 		x86_platform.set_legacy_features =
1376 				xen_domu_set_legacy_features;
1377 	} else {
1378 		const struct dom0_vga_console_info *info =
1379 			(void *)((char *)xen_start_info +
1380 				 xen_start_info->console.dom0.info_off);
1381 		struct xen_platform_op op = {
1382 			.cmd = XENPF_firmware_info,
1383 			.interface_version = XENPF_INTERFACE_VERSION,
1384 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1385 		};
1386 
1387 		x86_platform.set_legacy_features =
1388 				xen_dom0_set_legacy_features;
1389 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1390 		xen_start_info->console.domU.mfn = 0;
1391 		xen_start_info->console.domU.evtchn = 0;
1392 
1393 		if (HYPERVISOR_platform_op(&op) == 0)
1394 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1395 
1396 		/* Make sure ACS will be enabled */
1397 		pci_request_acs();
1398 
1399 		xen_acpi_sleep_register();
1400 
1401 		xen_boot_params_init_edd();
1402 
1403 #ifdef CONFIG_ACPI
1404 		/*
1405 		 * Disable selecting "Firmware First mode" for correctable
1406 		 * memory errors, as this is the duty of the hypervisor to
1407 		 * decide.
1408 		 */
1409 		acpi_disable_cmcff = 1;
1410 #endif
1411 	}
1412 
1413 	xen_add_preferred_consoles();
1414 
1415 #ifdef CONFIG_PCI
1416 	/* PCI BIOS service won't work from a PV guest. */
1417 	pci_probe &= ~PCI_PROBE_BIOS;
1418 #endif
1419 	xen_raw_console_write("about to get started...\n");
1420 
1421 	/* We need this for printk timestamps */
1422 	xen_setup_runstate_info(0);
1423 
1424 	xen_efi_init(&boot_params);
1425 
1426 	/* Start the world */
1427 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1428 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1429 }
1430 
1431 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1432 {
1433 	int rc;
1434 
1435 	if (per_cpu(xen_vcpu, cpu) == NULL)
1436 		return -ENODEV;
1437 
1438 	xen_setup_timer(cpu);
1439 
1440 	rc = xen_smp_intr_init(cpu);
1441 	if (rc) {
1442 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1443 		     cpu, rc);
1444 		return rc;
1445 	}
1446 
1447 	rc = xen_smp_intr_init_pv(cpu);
1448 	if (rc) {
1449 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1450 		     cpu, rc);
1451 		return rc;
1452 	}
1453 
1454 	return 0;
1455 }
1456 
1457 static int xen_cpu_dead_pv(unsigned int cpu)
1458 {
1459 	xen_smp_intr_free(cpu);
1460 	xen_smp_intr_free_pv(cpu);
1461 
1462 	xen_teardown_timer(cpu);
1463 
1464 	return 0;
1465 }
1466 
1467 static uint32_t __init xen_platform_pv(void)
1468 {
1469 	if (xen_pv_domain())
1470 		return xen_cpuid_base();
1471 
1472 	return 0;
1473 }
1474 
1475 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1476 	.name                   = "Xen PV",
1477 	.detect                 = xen_platform_pv,
1478 	.type			= X86_HYPER_XEN_PV,
1479 	.runtime.pin_vcpu       = xen_pin_vcpu,
1480 	.ignore_nopv		= true,
1481 };
1482