xref: /linux/arch/x86/xen/enlighten_pv.c (revision ff5ccdb8d5bd242f1064c6f7996603e47e28d095)
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/kstrtox.h>
27 #include <linux/memblock.h>
28 #include <linux/export.h>
29 #include <linux/mm.h>
30 #include <linux/page-flags.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/edd.h>
34 #include <linux/reboot.h>
35 #include <linux/virtio_anchor.h>
36 #include <linux/stackprotector.h>
37 
38 #include <xen/xen.h>
39 #include <xen/events.h>
40 #include <xen/interface/xen.h>
41 #include <xen/interface/version.h>
42 #include <xen/interface/physdev.h>
43 #include <xen/interface/vcpu.h>
44 #include <xen/interface/memory.h>
45 #include <xen/interface/nmi.h>
46 #include <xen/interface/xen-mca.h>
47 #include <xen/features.h>
48 #include <xen/page.h>
49 #include <xen/hvc-console.h>
50 #include <xen/acpi.h>
51 
52 #include <asm/cpuid/api.h>
53 #include <asm/paravirt.h>
54 #include <asm/apic.h>
55 #include <asm/page.h>
56 #include <asm/xen/pci.h>
57 #include <asm/xen/hypercall.h>
58 #include <asm/xen/hypervisor.h>
59 #include <asm/xen/cpuid.h>
60 #include <asm/fixmap.h>
61 #include <asm/processor.h>
62 #include <asm/proto.h>
63 #include <asm/msr-index.h>
64 #include <asm/msr.h>
65 #include <asm/traps.h>
66 #include <asm/setup.h>
67 #include <asm/desc.h>
68 #include <asm/pgalloc.h>
69 #include <asm/tlbflush.h>
70 #include <asm/reboot.h>
71 #include <asm/hypervisor.h>
72 #include <asm/mach_traps.h>
73 #include <asm/mtrr.h>
74 #include <asm/mwait.h>
75 #include <asm/pci_x86.h>
76 #include <asm/cpu.h>
77 #include <asm/irq_stack.h>
78 #ifdef CONFIG_X86_IOPL_IOPERM
79 #include <asm/io_bitmap.h>
80 #endif
81 
82 #ifdef CONFIG_ACPI
83 #include <linux/acpi.h>
84 #include <asm/acpi.h>
85 #include <acpi/proc_cap_intel.h>
86 #include <acpi/processor.h>
87 #include <xen/interface/platform.h>
88 #endif
89 
90 #include "xen-ops.h"
91 
92 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
93 
94 void *xen_initial_gdt;
95 
96 static int xen_cpu_up_prepare_pv(unsigned int cpu);
97 static int xen_cpu_dead_pv(unsigned int cpu);
98 
99 #ifndef CONFIG_PREEMPTION
100 /*
101  * Some hypercalls issued by the toolstack can take many 10s of
102  * seconds. Allow tasks running hypercalls via the privcmd driver to
103  * be voluntarily preempted even if full kernel preemption is
104  * disabled.
105  *
106  * Such preemptible hypercalls are bracketed by
107  * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
108  * calls.
109  */
110 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
111 EXPORT_PER_CPU_SYMBOL_GPL(xen_in_preemptible_hcall);
112 
113 /*
114  * In case of scheduling the flag must be cleared and restored after
115  * returning from schedule as the task might move to a different CPU.
116  */
117 static __always_inline bool get_and_clear_inhcall(void)
118 {
119 	bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
120 
121 	__this_cpu_write(xen_in_preemptible_hcall, false);
122 	return inhcall;
123 }
124 
125 static __always_inline void restore_inhcall(bool inhcall)
126 {
127 	__this_cpu_write(xen_in_preemptible_hcall, inhcall);
128 }
129 
130 #else
131 
132 static __always_inline bool get_and_clear_inhcall(void) { return false; }
133 static __always_inline void restore_inhcall(bool inhcall) { }
134 
135 #endif
136 
137 struct tls_descs {
138 	struct desc_struct desc[3];
139 };
140 
141 static DEFINE_PER_CPU(bool, xen_cpu_lazy_mode);
142 
143 bool xen_is_cpu_lazy_mode(void)
144 {
145 	return !in_interrupt() && this_cpu_read(xen_cpu_lazy_mode);
146 }
147 
148 /*
149  * Updating the 3 TLS descriptors in the GDT on every task switch is
150  * surprisingly expensive so we avoid updating them if they haven't
151  * changed.  Since Xen writes different descriptors than the one
152  * passed in the update_descriptor hypercall we keep shadow copies to
153  * compare against.
154  */
155 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
156 
157 static __read_mostly bool xen_msr_safe = IS_ENABLED(CONFIG_XEN_PV_MSR_SAFE);
158 
159 static int __init parse_xen_msr_safe(char *str)
160 {
161 	if (str)
162 		return kstrtobool(str, &xen_msr_safe);
163 	return -EINVAL;
164 }
165 early_param("xen_msr_safe", parse_xen_msr_safe);
166 
167 /* Get MTRR settings from Xen and put them into mtrr_state. */
168 static void __init xen_set_mtrr_data(void)
169 {
170 #ifdef CONFIG_MTRR
171 	struct xen_platform_op op = {
172 		.cmd = XENPF_read_memtype,
173 		.interface_version = XENPF_INTERFACE_VERSION,
174 	};
175 	unsigned int reg;
176 	unsigned long mask;
177 	uint32_t eax, width;
178 	static struct mtrr_var_range var[MTRR_MAX_VAR_RANGES] __initdata;
179 
180 	/* Get physical address width (only 64-bit cpus supported). */
181 	width = 36;
182 	eax = cpuid_eax(0x80000000);
183 	if ((eax >> 16) == 0x8000 && eax >= 0x80000008) {
184 		eax = cpuid_eax(0x80000008);
185 		width = eax & 0xff;
186 	}
187 
188 	for (reg = 0; reg < MTRR_MAX_VAR_RANGES; reg++) {
189 		op.u.read_memtype.reg = reg;
190 		if (HYPERVISOR_platform_op(&op))
191 			break;
192 
193 		/*
194 		 * Only called in dom0, which has all RAM PFNs mapped at
195 		 * RAM MFNs, and all PCI space etc. is identity mapped.
196 		 * This means we can treat MFN == PFN regarding MTRR settings.
197 		 */
198 		var[reg].base_lo = op.u.read_memtype.type;
199 		var[reg].base_lo |= op.u.read_memtype.mfn << PAGE_SHIFT;
200 		var[reg].base_hi = op.u.read_memtype.mfn >> (32 - PAGE_SHIFT);
201 		mask = ~((op.u.read_memtype.nr_mfns << PAGE_SHIFT) - 1);
202 		mask &= (1UL << width) - 1;
203 		if (mask)
204 			mask |= MTRR_PHYSMASK_V;
205 		var[reg].mask_lo = mask;
206 		var[reg].mask_hi = mask >> 32;
207 	}
208 
209 	/* Only overwrite MTRR state if any MTRR could be got from Xen. */
210 	if (reg)
211 		guest_force_mtrr_state(var, reg, MTRR_TYPE_UNCACHABLE);
212 #endif
213 }
214 
215 static void __init xen_pv_init_platform(void)
216 {
217 	/* PV guests can't operate virtio devices without grants. */
218 	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
219 		virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
220 
221 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
222 
223 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
224 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
225 
226 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
227 	xen_vcpu_info_reset(0);
228 
229 	/* pvclock is in shared info area */
230 	xen_init_time_ops();
231 
232 	if (xen_initial_domain())
233 		xen_set_mtrr_data();
234 	else
235 		guest_force_mtrr_state(NULL, 0, MTRR_TYPE_WRBACK);
236 
237 	/* Adjust nr_cpu_ids before "enumeration" happens */
238 	xen_smp_count_cpus();
239 }
240 
241 static void __init xen_pv_guest_late_init(void)
242 {
243 #ifndef CONFIG_SMP
244 	/* Setup shared vcpu info for non-smp configurations */
245 	xen_setup_vcpu_info_placement();
246 #endif
247 }
248 
249 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
250 static __read_mostly unsigned int cpuid_leaf5_edx_val;
251 
252 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
253 		      unsigned int *cx, unsigned int *dx)
254 {
255 	unsigned int maskebx = ~0;
256 	unsigned int or_ebx = 0;
257 
258 	/*
259 	 * Mask out inconvenient features, to try and disable as many
260 	 * unsupported kernel subsystems as possible.
261 	 */
262 	switch (*ax) {
263 	case 0x1:
264 		/* Replace initial APIC ID in bits 24-31 of EBX. */
265 		/* See xen_pv_smp_config() for related topology preparations. */
266 		maskebx = 0x00ffffff;
267 		or_ebx = smp_processor_id() << 24;
268 		break;
269 
270 	case CPUID_LEAF_MWAIT:
271 		/* Synthesize the values.. */
272 		*ax = 0;
273 		*bx = 0;
274 		*cx = cpuid_leaf5_ecx_val;
275 		*dx = cpuid_leaf5_edx_val;
276 		return;
277 
278 	case 0xb:
279 		/* Suppress extended topology stuff */
280 		maskebx = 0;
281 		break;
282 	}
283 
284 	asm(XEN_EMULATE_PREFIX "cpuid"
285 		: "=a" (*ax),
286 		  "=b" (*bx),
287 		  "=c" (*cx),
288 		  "=d" (*dx)
289 		: "0" (*ax), "2" (*cx));
290 
291 	*bx &= maskebx;
292 	*bx |= or_ebx;
293 }
294 
295 static bool __init xen_check_mwait(void)
296 {
297 #ifdef CONFIG_ACPI
298 	struct xen_platform_op op = {
299 		.cmd			= XENPF_set_processor_pminfo,
300 		.u.set_pminfo.id	= -1,
301 		.u.set_pminfo.type	= XEN_PM_PDC,
302 	};
303 	uint32_t buf[3];
304 	unsigned int ax, bx, cx, dx;
305 	unsigned int mwait_mask;
306 
307 	/* We need to determine whether it is OK to expose the MWAIT
308 	 * capability to the kernel to harvest deeper than C3 states from ACPI
309 	 * _CST using the processor_harvest_xen.c module. For this to work, we
310 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
311 	 * checks against). The hypervisor won't expose the MWAIT flag because
312 	 * it would break backwards compatibility; so we will find out directly
313 	 * from the hardware and hypercall.
314 	 */
315 	if (!xen_initial_domain())
316 		return false;
317 
318 	/*
319 	 * When running under platform earlier than Xen4.2, do not expose
320 	 * mwait, to avoid the risk of loading native acpi pad driver
321 	 */
322 	if (!xen_running_on_version_or_later(4, 2))
323 		return false;
324 
325 	ax = 1;
326 	cx = 0;
327 
328 	native_cpuid(&ax, &bx, &cx, &dx);
329 
330 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
331 		     (1 << (X86_FEATURE_MWAIT % 32));
332 
333 	if ((cx & mwait_mask) != mwait_mask)
334 		return false;
335 
336 	/* We need to emulate the MWAIT_LEAF and for that we need both
337 	 * ecx and edx. The hypercall provides only partial information.
338 	 */
339 
340 	ax = CPUID_LEAF_MWAIT;
341 	bx = 0;
342 	cx = 0;
343 	dx = 0;
344 
345 	native_cpuid(&ax, &bx, &cx, &dx);
346 
347 	/* Ask the Hypervisor whether to clear ACPI_PROC_CAP_C_C2C3_FFH. If so,
348 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
349 	 */
350 	buf[0] = ACPI_PDC_REVISION_ID;
351 	buf[1] = 1;
352 	buf[2] = (ACPI_PROC_CAP_C_CAPABILITY_SMP | ACPI_PROC_CAP_EST_CAPABILITY_SWSMP);
353 
354 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
355 
356 	if ((HYPERVISOR_platform_op(&op) == 0) &&
357 	    (buf[2] & (ACPI_PROC_CAP_C_C1_FFH | ACPI_PROC_CAP_C_C2C3_FFH))) {
358 		cpuid_leaf5_ecx_val = cx;
359 		cpuid_leaf5_edx_val = dx;
360 	}
361 	return true;
362 #else
363 	return false;
364 #endif
365 }
366 
367 static bool __init xen_check_xsave(void)
368 {
369 	unsigned int cx, xsave_mask;
370 
371 	cx = cpuid_ecx(1);
372 
373 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
374 		     (1 << (X86_FEATURE_OSXSAVE % 32));
375 
376 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
377 	return (cx & xsave_mask) == xsave_mask;
378 }
379 
380 static void __init xen_init_capabilities(void)
381 {
382 	setup_clear_cpu_cap(X86_FEATURE_DCA);
383 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
384 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
385 	setup_clear_cpu_cap(X86_FEATURE_ACC);
386 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
387 	setup_clear_cpu_cap(X86_FEATURE_SME);
388 	setup_clear_cpu_cap(X86_FEATURE_LKGS);
389 
390 	/*
391 	 * Xen PV would need some work to support PCID: CR3 handling as well
392 	 * as xen_flush_tlb_multi() would need updating.
393 	 */
394 	setup_clear_cpu_cap(X86_FEATURE_PCID);
395 
396 	if (!xen_initial_domain())
397 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
398 
399 	if (xen_check_mwait())
400 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
401 	else
402 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
403 
404 	if (!xen_check_xsave()) {
405 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
406 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
407 	}
408 }
409 
410 static noinstr void xen_set_debugreg(int reg, unsigned long val)
411 {
412 	HYPERVISOR_set_debugreg(reg, val);
413 }
414 
415 static noinstr unsigned long xen_get_debugreg(int reg)
416 {
417 	return HYPERVISOR_get_debugreg(reg);
418 }
419 
420 static void xen_start_context_switch(struct task_struct *prev)
421 {
422 	BUG_ON(preemptible());
423 
424 	__task_lazy_mmu_mode_pause(prev);
425 	this_cpu_write(xen_cpu_lazy_mode, true);
426 }
427 
428 static void xen_end_context_switch(struct task_struct *next)
429 {
430 	BUG_ON(preemptible());
431 
432 	xen_mc_flush();
433 	this_cpu_write(xen_cpu_lazy_mode, false);
434 	__task_lazy_mmu_mode_resume(next);
435 }
436 
437 static unsigned long xen_store_tr(void)
438 {
439 	return 0;
440 }
441 
442 /*
443  * Set the page permissions for a particular virtual address.  If the
444  * address is a vmalloc mapping (or other non-linear mapping), then
445  * find the linear mapping of the page and also set its protections to
446  * match.
447  */
448 static void set_aliased_prot(void *v, pgprot_t prot)
449 {
450 	int level;
451 	pte_t *ptep;
452 	pte_t pte;
453 	unsigned long pfn;
454 	unsigned char dummy;
455 	void *va;
456 
457 	ptep = lookup_address((unsigned long)v, &level);
458 	BUG_ON(ptep == NULL);
459 
460 	pfn = pte_pfn(*ptep);
461 	pte = pfn_pte(pfn, prot);
462 
463 	/*
464 	 * Careful: update_va_mapping() will fail if the virtual address
465 	 * we're poking isn't populated in the page tables.  We don't
466 	 * need to worry about the direct map (that's always in the page
467 	 * tables), but we need to be careful about vmap space.  In
468 	 * particular, the top level page table can lazily propagate
469 	 * entries between processes, so if we've switched mms since we
470 	 * vmapped the target in the first place, we might not have the
471 	 * top-level page table entry populated.
472 	 *
473 	 * We disable preemption because we want the same mm active when
474 	 * we probe the target and when we issue the hypercall.  We'll
475 	 * have the same nominal mm, but if we're a kernel thread, lazy
476 	 * mm dropping could change our pgd.
477 	 *
478 	 * Out of an abundance of caution, this uses __get_user() to fault
479 	 * in the target address just in case there's some obscure case
480 	 * in which the target address isn't readable.
481 	 */
482 
483 	preempt_disable();
484 
485 	copy_from_kernel_nofault(&dummy, v, 1);
486 
487 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
488 		BUG();
489 
490 	va = __va(PFN_PHYS(pfn));
491 
492 	if (va != v && HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
493 		BUG();
494 
495 	preempt_enable();
496 }
497 
498 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
499 {
500 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
501 	int i;
502 
503 	/*
504 	 * We need to mark the all aliases of the LDT pages RO.  We
505 	 * don't need to call vm_flush_aliases(), though, since that's
506 	 * only responsible for flushing aliases out the TLBs, not the
507 	 * page tables, and Xen will flush the TLB for us if needed.
508 	 *
509 	 * To avoid confusing future readers: none of this is necessary
510 	 * to load the LDT.  The hypervisor only checks this when the
511 	 * LDT is faulted in due to subsequent descriptor access.
512 	 */
513 
514 	for (i = 0; i < entries; i += entries_per_page)
515 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
516 }
517 
518 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
519 {
520 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
521 	int i;
522 
523 	for (i = 0; i < entries; i += entries_per_page)
524 		set_aliased_prot(ldt + i, PAGE_KERNEL);
525 }
526 
527 static void xen_set_ldt(const void *addr, unsigned entries)
528 {
529 	struct mmuext_op *op;
530 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
531 
532 	trace_xen_cpu_set_ldt(addr, entries);
533 
534 	op = mcs.args;
535 	op->cmd = MMUEXT_SET_LDT;
536 	op->arg1.linear_addr = (unsigned long)addr;
537 	op->arg2.nr_ents = entries;
538 
539 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
540 
541 	xen_mc_issue(!xen_is_cpu_lazy_mode());
542 }
543 
544 static void xen_load_gdt(const struct desc_ptr *dtr)
545 {
546 	unsigned long va = dtr->address;
547 	unsigned int size = dtr->size + 1;
548 	unsigned long pfn, mfn;
549 	int level;
550 	pte_t *ptep;
551 	void *virt;
552 
553 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
554 	BUG_ON(size > PAGE_SIZE);
555 	BUG_ON(va & ~PAGE_MASK);
556 
557 	/*
558 	 * The GDT is per-cpu and is in the percpu data area.
559 	 * That can be virtually mapped, so we need to do a
560 	 * page-walk to get the underlying MFN for the
561 	 * hypercall.  The page can also be in the kernel's
562 	 * linear range, so we need to RO that mapping too.
563 	 */
564 	ptep = lookup_address(va, &level);
565 	BUG_ON(ptep == NULL);
566 
567 	pfn = pte_pfn(*ptep);
568 	mfn = pfn_to_mfn(pfn);
569 	virt = __va(PFN_PHYS(pfn));
570 
571 	make_lowmem_page_readonly((void *)va);
572 	make_lowmem_page_readonly(virt);
573 
574 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
575 		BUG();
576 }
577 
578 /*
579  * load_gdt for early boot, when the gdt is only mapped once
580  */
581 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
582 {
583 	unsigned long va = dtr->address;
584 	unsigned int size = dtr->size + 1;
585 	unsigned long pfn, mfn;
586 	pte_t pte;
587 
588 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
589 	BUG_ON(size > PAGE_SIZE);
590 	BUG_ON(va & ~PAGE_MASK);
591 
592 	pfn = virt_to_pfn((void *)va);
593 	mfn = pfn_to_mfn(pfn);
594 
595 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
596 
597 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
598 		BUG();
599 
600 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
601 		BUG();
602 }
603 
604 static inline bool desc_equal(const struct desc_struct *d1,
605 			      const struct desc_struct *d2)
606 {
607 	return !memcmp(d1, d2, sizeof(*d1));
608 }
609 
610 static void load_TLS_descriptor(struct thread_struct *t,
611 				unsigned int cpu, unsigned int i)
612 {
613 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
614 	struct desc_struct *gdt;
615 	xmaddr_t maddr;
616 	struct multicall_space mc;
617 
618 	if (desc_equal(shadow, &t->tls_array[i]))
619 		return;
620 
621 	*shadow = t->tls_array[i];
622 
623 	gdt = get_cpu_gdt_rw(cpu);
624 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
625 	mc = __xen_mc_entry(0);
626 
627 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
628 }
629 
630 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
631 {
632 	/*
633 	 * In lazy mode we need to zero %fs, otherwise we may get an
634 	 * exception between the new %fs descriptor being loaded and
635 	 * %fs being effectively cleared at __switch_to().
636 	 */
637 	if (xen_is_cpu_lazy_mode())
638 		loadsegment(fs, 0);
639 
640 	xen_mc_batch();
641 
642 	load_TLS_descriptor(t, cpu, 0);
643 	load_TLS_descriptor(t, cpu, 1);
644 	load_TLS_descriptor(t, cpu, 2);
645 
646 	xen_mc_issue(!xen_is_cpu_lazy_mode());
647 }
648 
649 static void xen_load_gs_index(unsigned int idx)
650 {
651 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
652 		BUG();
653 }
654 
655 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
656 				const void *ptr)
657 {
658 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
659 	u64 entry = *(u64 *)ptr;
660 
661 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
662 
663 	preempt_disable();
664 
665 	xen_mc_flush();
666 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
667 		BUG();
668 
669 	preempt_enable();
670 }
671 
672 void noist_exc_debug(struct pt_regs *regs);
673 
674 DEFINE_IDTENTRY_RAW(xenpv_exc_nmi)
675 {
676 	/* On Xen PV, NMI doesn't use IST.  The C part is the same as native. */
677 	exc_nmi(regs);
678 }
679 
680 DEFINE_IDTENTRY_RAW_ERRORCODE(xenpv_exc_double_fault)
681 {
682 	/* On Xen PV, DF doesn't use IST.  The C part is the same as native. */
683 	exc_double_fault(regs, error_code);
684 }
685 
686 DEFINE_IDTENTRY_RAW(xenpv_exc_debug)
687 {
688 	/*
689 	 * There's no IST on Xen PV, but we still need to dispatch
690 	 * to the correct handler.
691 	 */
692 	if (user_mode(regs))
693 		noist_exc_debug(regs);
694 	else
695 		exc_debug(regs);
696 }
697 
698 DEFINE_IDTENTRY_RAW(exc_xen_unknown_trap)
699 {
700 	/* This should never happen and there is no way to handle it. */
701 	instrumentation_begin();
702 	pr_err("Unknown trap in Xen PV mode.");
703 	BUG();
704 	instrumentation_end();
705 }
706 
707 #ifdef CONFIG_X86_MCE
708 DEFINE_IDTENTRY_RAW(xenpv_exc_machine_check)
709 {
710 	/*
711 	 * There's no IST on Xen PV, but we still need to dispatch
712 	 * to the correct handler.
713 	 */
714 	if (user_mode(regs))
715 		noist_exc_machine_check(regs);
716 	else
717 		exc_machine_check(regs);
718 }
719 #endif
720 
721 static void __xen_pv_evtchn_do_upcall(struct pt_regs *regs)
722 {
723 	struct pt_regs *old_regs = set_irq_regs(regs);
724 
725 	inc_irq_stat(HYPERVISOR_CALLBACK);
726 
727 	xen_evtchn_do_upcall();
728 
729 	set_irq_regs(old_regs);
730 }
731 
732 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
733 {
734 	irqentry_state_t state = irqentry_enter(regs);
735 	bool inhcall;
736 
737 	instrumentation_begin();
738 	run_sysvec_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
739 
740 	inhcall = get_and_clear_inhcall();
741 	if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
742 		irqentry_exit_cond_resched();
743 		instrumentation_end();
744 		restore_inhcall(inhcall);
745 	} else {
746 		instrumentation_end();
747 		irqentry_exit(regs, state);
748 	}
749 }
750 
751 struct trap_array_entry {
752 	void (*orig)(void);
753 	void (*xen)(void);
754 	bool ist_okay;
755 };
756 
757 #define TRAP_ENTRY(func, ist_ok) {			\
758 	.orig		= asm_##func,			\
759 	.xen		= xen_asm_##func,		\
760 	.ist_okay	= ist_ok }
761 
762 #define TRAP_ENTRY_REDIR(func, ist_ok) {		\
763 	.orig		= asm_##func,			\
764 	.xen		= xen_asm_xenpv_##func,		\
765 	.ist_okay	= ist_ok }
766 
767 static struct trap_array_entry trap_array[] = {
768 	TRAP_ENTRY_REDIR(exc_debug,			true  ),
769 	TRAP_ENTRY_REDIR(exc_double_fault,		true  ),
770 #ifdef CONFIG_X86_MCE
771 	TRAP_ENTRY_REDIR(exc_machine_check,		true  ),
772 #endif
773 	TRAP_ENTRY_REDIR(exc_nmi,			true  ),
774 	TRAP_ENTRY(exc_int3,				false ),
775 	TRAP_ENTRY(exc_overflow,			false ),
776 #ifdef CONFIG_IA32_EMULATION
777 	TRAP_ENTRY(int80_emulation,			false ),
778 #endif
779 	TRAP_ENTRY(exc_page_fault,			false ),
780 	TRAP_ENTRY(exc_divide_error,			false ),
781 	TRAP_ENTRY(exc_bounds,				false ),
782 	TRAP_ENTRY(exc_invalid_op,			false ),
783 	TRAP_ENTRY(exc_device_not_available,		false ),
784 	TRAP_ENTRY(exc_coproc_segment_overrun,		false ),
785 	TRAP_ENTRY(exc_invalid_tss,			false ),
786 	TRAP_ENTRY(exc_segment_not_present,		false ),
787 	TRAP_ENTRY(exc_stack_segment,			false ),
788 	TRAP_ENTRY(exc_general_protection,		false ),
789 	TRAP_ENTRY(exc_spurious_interrupt_bug,		false ),
790 	TRAP_ENTRY(exc_coprocessor_error,		false ),
791 	TRAP_ENTRY(exc_alignment_check,			false ),
792 	TRAP_ENTRY(exc_simd_coprocessor_error,		false ),
793 #ifdef CONFIG_X86_CET
794 	TRAP_ENTRY(exc_control_protection,		false ),
795 #endif
796 };
797 
798 static bool __ref get_trap_addr(void **addr, unsigned int ist)
799 {
800 	unsigned int nr;
801 	bool ist_okay = false;
802 	bool found = false;
803 
804 	/*
805 	 * Replace trap handler addresses by Xen specific ones.
806 	 * Check for known traps using IST and whitelist them.
807 	 * The debugger ones are the only ones we care about.
808 	 * Xen will handle faults like double_fault, so we should never see
809 	 * them.  Warn if there's an unexpected IST-using fault handler.
810 	 */
811 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
812 		struct trap_array_entry *entry = trap_array + nr;
813 
814 		if (*addr == entry->orig) {
815 			*addr = entry->xen;
816 			ist_okay = entry->ist_okay;
817 			found = true;
818 			break;
819 		}
820 	}
821 
822 	if (nr == ARRAY_SIZE(trap_array) &&
823 	    *addr >= (void *)early_idt_handler_array[0] &&
824 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
825 		nr = (*addr - (void *)early_idt_handler_array[0]) /
826 		     EARLY_IDT_HANDLER_SIZE;
827 		*addr = (void *)xen_early_idt_handler_array[nr];
828 		found = true;
829 	}
830 
831 	if (!found)
832 		*addr = (void *)xen_asm_exc_xen_unknown_trap;
833 
834 	if (WARN_ON(found && ist != 0 && !ist_okay))
835 		return false;
836 
837 	return true;
838 }
839 
840 static int cvt_gate_to_trap(int vector, const gate_desc *val,
841 			    struct trap_info *info)
842 {
843 	unsigned long addr;
844 
845 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
846 		return 0;
847 
848 	info->vector = vector;
849 
850 	addr = gate_offset(val);
851 	if (!get_trap_addr((void **)&addr, val->bits.ist))
852 		return 0;
853 	info->address = addr;
854 
855 	info->cs = gate_segment(val);
856 	info->flags = val->bits.dpl;
857 	/* interrupt gates clear IF */
858 	if (val->bits.type == GATE_INTERRUPT)
859 		info->flags |= 1 << 2;
860 
861 	return 1;
862 }
863 
864 /* Locations of each CPU's IDT */
865 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
866 
867 /* Set an IDT entry.  If the entry is part of the current IDT, then
868    also update Xen. */
869 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
870 {
871 	unsigned long p = (unsigned long)&dt[entrynum];
872 	unsigned long start, end;
873 
874 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
875 
876 	preempt_disable();
877 
878 	start = __this_cpu_read(idt_desc.address);
879 	end = start + __this_cpu_read(idt_desc.size) + 1;
880 
881 	xen_mc_flush();
882 
883 	native_write_idt_entry(dt, entrynum, g);
884 
885 	if (p >= start && (p + 8) <= end) {
886 		struct trap_info info[2];
887 
888 		info[1].address = 0;
889 
890 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
891 			if (HYPERVISOR_set_trap_table(info))
892 				BUG();
893 	}
894 
895 	preempt_enable();
896 }
897 
898 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
899 				      struct trap_info *traps, bool full)
900 {
901 	unsigned in, out, count;
902 
903 	count = (desc->size+1) / sizeof(gate_desc);
904 	BUG_ON(count > 256);
905 
906 	for (in = out = 0; in < count; in++) {
907 		gate_desc *entry = (gate_desc *)(desc->address) + in;
908 
909 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
910 			out++;
911 	}
912 
913 	return out;
914 }
915 
916 void xen_copy_trap_info(struct trap_info *traps)
917 {
918 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
919 
920 	xen_convert_trap_info(desc, traps, true);
921 }
922 
923 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
924    hold a spinlock to protect the static traps[] array (static because
925    it avoids allocation, and saves stack space). */
926 static void xen_load_idt(const struct desc_ptr *desc)
927 {
928 	static DEFINE_SPINLOCK(lock);
929 	static struct trap_info traps[257];
930 	static const struct trap_info zero = { };
931 	unsigned out;
932 
933 	trace_xen_cpu_load_idt(desc);
934 
935 	spin_lock(&lock);
936 
937 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
938 
939 	out = xen_convert_trap_info(desc, traps, false);
940 	traps[out] = zero;
941 
942 	xen_mc_flush();
943 	if (HYPERVISOR_set_trap_table(traps))
944 		BUG();
945 
946 	spin_unlock(&lock);
947 }
948 
949 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
950    they're handled differently. */
951 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
952 				const void *desc, int type)
953 {
954 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
955 
956 	preempt_disable();
957 
958 	switch (type) {
959 	case DESC_LDT:
960 	case DESC_TSS:
961 		/* ignore */
962 		break;
963 
964 	default: {
965 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
966 
967 		xen_mc_flush();
968 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
969 			BUG();
970 	}
971 
972 	}
973 
974 	preempt_enable();
975 }
976 
977 /*
978  * Version of write_gdt_entry for use at early boot-time needed to
979  * update an entry as simply as possible.
980  */
981 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
982 					    const void *desc, int type)
983 {
984 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
985 
986 	switch (type) {
987 	case DESC_LDT:
988 	case DESC_TSS:
989 		/* ignore */
990 		break;
991 
992 	default: {
993 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
994 
995 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
996 			dt[entry] = *(struct desc_struct *)desc;
997 	}
998 
999 	}
1000 }
1001 
1002 static void xen_load_sp0(unsigned long sp0)
1003 {
1004 	struct multicall_space mcs;
1005 
1006 	mcs = xen_mc_entry(0);
1007 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
1008 	xen_mc_issue(!xen_is_cpu_lazy_mode());
1009 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
1010 }
1011 
1012 #ifdef CONFIG_X86_IOPL_IOPERM
1013 static void xen_invalidate_io_bitmap(void)
1014 {
1015 	struct physdev_set_iobitmap iobitmap = {
1016 		.bitmap = NULL,
1017 		.nr_ports = 0,
1018 	};
1019 
1020 	native_tss_invalidate_io_bitmap();
1021 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
1022 }
1023 
1024 static void xen_update_io_bitmap(void)
1025 {
1026 	struct physdev_set_iobitmap iobitmap;
1027 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1028 
1029 	native_tss_update_io_bitmap();
1030 
1031 	iobitmap.bitmap = (uint8_t *)(&tss->x86_tss) +
1032 			  tss->x86_tss.io_bitmap_base;
1033 	if (tss->x86_tss.io_bitmap_base == IO_BITMAP_OFFSET_INVALID)
1034 		iobitmap.nr_ports = 0;
1035 	else
1036 		iobitmap.nr_ports = IO_BITMAP_BITS;
1037 
1038 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobitmap);
1039 }
1040 #endif
1041 
1042 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
1043 
1044 static unsigned long xen_read_cr0(void)
1045 {
1046 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
1047 
1048 	if (unlikely(cr0 == 0)) {
1049 		cr0 = native_read_cr0();
1050 		this_cpu_write(xen_cr0_value, cr0);
1051 	}
1052 
1053 	return cr0;
1054 }
1055 
1056 static void xen_write_cr0(unsigned long cr0)
1057 {
1058 	struct multicall_space mcs;
1059 
1060 	this_cpu_write(xen_cr0_value, cr0);
1061 
1062 	/* Only pay attention to cr0.TS; everything else is
1063 	   ignored. */
1064 	mcs = xen_mc_entry(0);
1065 
1066 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1067 
1068 	xen_mc_issue(!xen_is_cpu_lazy_mode());
1069 }
1070 
1071 static void xen_write_cr4(unsigned long cr4)
1072 {
1073 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1074 
1075 	native_write_cr4(cr4);
1076 }
1077 
1078 static u64 xen_do_read_msr(u32 msr, int *err)
1079 {
1080 	u64 val = 0;	/* Avoid uninitialized value for safe variant. */
1081 
1082 	if (pmu_msr_chk_emulated(msr, &val, true))
1083 		return val;
1084 
1085 	if (err)
1086 		*err = native_read_msr_safe(msr, &val);
1087 	else
1088 		val = native_read_msr(msr);
1089 
1090 	switch (msr) {
1091 	case MSR_IA32_APICBASE:
1092 		val &= ~X2APIC_ENABLE;
1093 		if (smp_processor_id() == 0)
1094 			val |= MSR_IA32_APICBASE_BSP;
1095 		else
1096 			val &= ~MSR_IA32_APICBASE_BSP;
1097 		break;
1098 	}
1099 	return val;
1100 }
1101 
1102 static void set_seg(u32 which, u64 base)
1103 {
1104 	if (HYPERVISOR_set_segment_base(which, base))
1105 		WARN(1, "Xen set_segment_base(%u, %llx) failed\n", which, base);
1106 }
1107 
1108 /*
1109  * Support write_msr_safe() and write_msr() semantics.
1110  * With err == NULL write_msr() semantics are selected.
1111  * Supplying an err pointer requires err to be pre-initialized with 0.
1112  */
1113 static void xen_do_write_msr(u32 msr, u64 val, int *err)
1114 {
1115 	switch (msr) {
1116 	case MSR_FS_BASE:
1117 		set_seg(SEGBASE_FS, val);
1118 		break;
1119 
1120 	case MSR_KERNEL_GS_BASE:
1121 		set_seg(SEGBASE_GS_USER, val);
1122 		break;
1123 
1124 	case MSR_GS_BASE:
1125 		set_seg(SEGBASE_GS_KERNEL, val);
1126 		break;
1127 
1128 	case MSR_STAR:
1129 	case MSR_CSTAR:
1130 	case MSR_LSTAR:
1131 	case MSR_SYSCALL_MASK:
1132 	case MSR_IA32_SYSENTER_CS:
1133 	case MSR_IA32_SYSENTER_ESP:
1134 	case MSR_IA32_SYSENTER_EIP:
1135 		/* Fast syscall setup is all done in hypercalls, so
1136 		   these are all ignored.  Stub them out here to stop
1137 		   Xen console noise. */
1138 		break;
1139 
1140 	default:
1141 		if (pmu_msr_chk_emulated(msr, &val, false))
1142 			return;
1143 
1144 		if (err)
1145 			*err = native_write_msr_safe(msr, val);
1146 		else
1147 			native_write_msr(msr, val);
1148 	}
1149 }
1150 
1151 static int xen_read_msr_safe(u32 msr, u64 *val)
1152 {
1153 	int err = 0;
1154 
1155 	*val = xen_do_read_msr(msr, &err);
1156 	return err;
1157 }
1158 
1159 static int xen_write_msr_safe(u32 msr, u64 val)
1160 {
1161 	int err = 0;
1162 
1163 	xen_do_write_msr(msr, val, &err);
1164 
1165 	return err;
1166 }
1167 
1168 static u64 xen_read_msr(u32 msr)
1169 {
1170 	int err = 0;
1171 
1172 	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
1173 }
1174 
1175 static void xen_write_msr(u32 msr, u64 val)
1176 {
1177 	int err;
1178 
1179 	xen_do_write_msr(msr, val, xen_msr_safe ? &err : NULL);
1180 }
1181 
1182 /* This is called once we have the cpu_possible_mask */
1183 void __init xen_setup_vcpu_info_placement(void)
1184 {
1185 	int cpu;
1186 
1187 	for_each_possible_cpu(cpu) {
1188 		/* Set up direct vCPU id mapping for PV guests. */
1189 		per_cpu(xen_vcpu_id, cpu) = cpu;
1190 		xen_vcpu_setup(cpu);
1191 	}
1192 
1193 	pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1194 	pv_ops.irq.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1195 	pv_ops.irq.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1196 	pv_ops.mmu.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1197 }
1198 
1199 static const struct pv_info xen_info __initconst = {
1200 	.extra_user_64bit_cs = FLAT_USER_CS64,
1201 	.io_delay = false,
1202 	.name = "Xen",
1203 };
1204 
1205 static void xen_restart(char *msg)
1206 {
1207 	xen_reboot(SHUTDOWN_reboot);
1208 }
1209 
1210 static void xen_machine_halt(void)
1211 {
1212 	xen_reboot(SHUTDOWN_poweroff);
1213 }
1214 
1215 static void xen_machine_power_off(void)
1216 {
1217 	do_kernel_power_off();
1218 	xen_reboot(SHUTDOWN_poweroff);
1219 }
1220 
1221 static void xen_crash_shutdown(struct pt_regs *regs)
1222 {
1223 	xen_reboot(SHUTDOWN_crash);
1224 }
1225 
1226 static const struct machine_ops xen_machine_ops __initconst = {
1227 	.restart = xen_restart,
1228 	.halt = xen_machine_halt,
1229 	.power_off = xen_machine_power_off,
1230 	.shutdown = xen_machine_halt,
1231 	.crash_shutdown = xen_crash_shutdown,
1232 	.emergency_restart = xen_emergency_restart,
1233 };
1234 
1235 static unsigned char xen_get_nmi_reason(void)
1236 {
1237 	unsigned char reason = 0;
1238 
1239 	/* Construct a value which looks like it came from port 0x61. */
1240 	if (test_bit(_XEN_NMIREASON_io_error,
1241 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1242 		reason |= NMI_REASON_IOCHK;
1243 	if (test_bit(_XEN_NMIREASON_pci_serr,
1244 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1245 		reason |= NMI_REASON_SERR;
1246 
1247 	return reason;
1248 }
1249 
1250 static void __init xen_boot_params_init_edd(void)
1251 {
1252 #if IS_ENABLED(CONFIG_EDD)
1253 	struct xen_platform_op op;
1254 	struct edd_info *edd_info;
1255 	u32 *mbr_signature;
1256 	unsigned nr;
1257 	int ret;
1258 
1259 	edd_info = boot_params.eddbuf;
1260 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1261 
1262 	op.cmd = XENPF_firmware_info;
1263 
1264 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1265 	for (nr = 0; nr < EDDMAXNR; nr++) {
1266 		struct edd_info *info = edd_info + nr;
1267 
1268 		op.u.firmware_info.index = nr;
1269 		info->params.length = sizeof(info->params);
1270 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1271 				     &info->params);
1272 		ret = HYPERVISOR_platform_op(&op);
1273 		if (ret)
1274 			break;
1275 
1276 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1277 		C(device);
1278 		C(version);
1279 		C(interface_support);
1280 		C(legacy_max_cylinder);
1281 		C(legacy_max_head);
1282 		C(legacy_sectors_per_track);
1283 #undef C
1284 	}
1285 	boot_params.eddbuf_entries = nr;
1286 
1287 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1288 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1289 		op.u.firmware_info.index = nr;
1290 		ret = HYPERVISOR_platform_op(&op);
1291 		if (ret)
1292 			break;
1293 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1294 	}
1295 	boot_params.edd_mbr_sig_buf_entries = nr;
1296 #endif
1297 }
1298 
1299 /*
1300  * Set up the GDT and segment registers for -fstack-protector.  Until
1301  * we do this, we have to be careful not to call any stack-protected
1302  * function, which is most of the kernel.
1303  */
1304 static void __init xen_setup_gdt(int cpu)
1305 {
1306 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1307 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1308 
1309 	switch_gdt_and_percpu_base(cpu);
1310 
1311 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1312 	pv_ops.cpu.load_gdt = xen_load_gdt;
1313 }
1314 
1315 static void __init xen_dom0_set_legacy_features(void)
1316 {
1317 	x86_platform.legacy.rtc = 1;
1318 }
1319 
1320 static void __init xen_domu_set_legacy_features(void)
1321 {
1322 	x86_platform.legacy.rtc = 0;
1323 }
1324 
1325 extern void early_xen_iret_patch(void);
1326 
1327 /* First C function to be called on Xen boot */
1328 asmlinkage __visible void __init xen_start_kernel(struct start_info *si)
1329 {
1330 	struct physdev_set_iopl set_iopl;
1331 	unsigned long initrd_start = 0;
1332 	int rc;
1333 
1334 	if (!si)
1335 		return;
1336 
1337 	clear_bss();
1338 
1339 	xen_start_info = si;
1340 
1341 	__text_gen_insn(&early_xen_iret_patch,
1342 			JMP32_INSN_OPCODE, &early_xen_iret_patch, &xen_iret,
1343 			JMP32_INSN_SIZE);
1344 
1345 	xen_domain_type = XEN_PV_DOMAIN;
1346 	setup_force_cpu_cap(X86_FEATURE_XENPV);
1347 	xen_start_flags = xen_start_info->flags;
1348 	/* Interrupts are guaranteed to be off initially. */
1349 	early_boot_irqs_disabled = true;
1350 	static_call_update_early(xen_hypercall, xen_hypercall_pv);
1351 
1352 	xen_setup_features();
1353 
1354 	/* Install Xen paravirt ops */
1355 	pv_info = xen_info;
1356 
1357 	pv_ops.cpu.cpuid = xen_cpuid;
1358 	pv_ops.cpu.set_debugreg = xen_set_debugreg;
1359 	pv_ops.cpu.get_debugreg = xen_get_debugreg;
1360 	pv_ops.cpu.read_cr0 = xen_read_cr0;
1361 	pv_ops.cpu.write_cr0 = xen_write_cr0;
1362 	pv_ops.cpu.write_cr4 = xen_write_cr4;
1363 	pv_ops.cpu.read_msr = xen_read_msr;
1364 	pv_ops.cpu.write_msr = xen_write_msr;
1365 	pv_ops.cpu.read_msr_safe = xen_read_msr_safe;
1366 	pv_ops.cpu.write_msr_safe = xen_write_msr_safe;
1367 	pv_ops.cpu.read_pmc = xen_read_pmc;
1368 	pv_ops.cpu.load_tr_desc = paravirt_nop;
1369 	pv_ops.cpu.set_ldt = xen_set_ldt;
1370 	pv_ops.cpu.load_gdt = xen_load_gdt;
1371 	pv_ops.cpu.load_idt = xen_load_idt;
1372 	pv_ops.cpu.load_tls = xen_load_tls;
1373 	pv_ops.cpu.load_gs_index = xen_load_gs_index;
1374 	pv_ops.cpu.alloc_ldt = xen_alloc_ldt;
1375 	pv_ops.cpu.free_ldt = xen_free_ldt;
1376 	pv_ops.cpu.store_tr = xen_store_tr;
1377 	pv_ops.cpu.write_ldt_entry = xen_write_ldt_entry;
1378 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1379 	pv_ops.cpu.write_idt_entry = xen_write_idt_entry;
1380 	pv_ops.cpu.load_sp0 = xen_load_sp0;
1381 #ifdef CONFIG_X86_IOPL_IOPERM
1382 	pv_ops.cpu.invalidate_io_bitmap = xen_invalidate_io_bitmap;
1383 	pv_ops.cpu.update_io_bitmap = xen_update_io_bitmap;
1384 #endif
1385 	pv_ops.cpu.start_context_switch = xen_start_context_switch;
1386 	pv_ops.cpu.end_context_switch = xen_end_context_switch;
1387 
1388 	xen_init_irq_ops();
1389 
1390 	/*
1391 	 * Setup xen_vcpu early because it is needed for
1392 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1393 	 *
1394 	 * Don't do the full vcpu_info placement stuff until we have
1395 	 * the cpu_possible_mask and a non-dummy shared_info.
1396 	 */
1397 	xen_vcpu_info_reset(0);
1398 
1399 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1400 	x86_platform.realmode_reserve = x86_init_noop;
1401 	x86_platform.realmode_init = x86_init_noop;
1402 
1403 	x86_init.resources.memory_setup = xen_memory_setup;
1404 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1405 	x86_init.irqs.intr_mode_init	= x86_64_probe_apic;
1406 	x86_init.oem.arch_setup = xen_arch_setup;
1407 	x86_init.oem.banner = xen_banner;
1408 	x86_init.hyper.init_platform = xen_pv_init_platform;
1409 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1410 
1411 	/*
1412 	 * Set up some pagetable state before starting to set any ptes.
1413 	 */
1414 
1415 	xen_setup_machphys_mapping();
1416 	xen_init_mmu_ops();
1417 
1418 	/* Prevent unwanted bits from being set in PTEs. */
1419 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1420 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1421 
1422 	/* Get mfn list */
1423 	xen_build_dynamic_phys_to_machine();
1424 
1425 	/* Work out if we support NX */
1426 	cpuid_scan_cpu(&boot_cpu_data);
1427 	get_cpu_cap(&boot_cpu_data);
1428 	x86_configure_nx();
1429 
1430 	/*
1431 	 * Set up kernel GDT and segment registers, mainly so that
1432 	 * -fstack-protector code can be executed.
1433 	 */
1434 	xen_setup_gdt(0);
1435 
1436 	/* Determine virtual and physical address sizes */
1437 	get_cpu_address_sizes(&boot_cpu_data);
1438 
1439 	/* Let's presume PV guests always boot on vCPU with id 0. */
1440 	per_cpu(xen_vcpu_id, 0) = 0;
1441 
1442 	idt_setup_early_handler();
1443 
1444 	xen_init_capabilities();
1445 
1446 	/*
1447 	 * set up the basic apic ops.
1448 	 */
1449 	xen_init_apic();
1450 
1451 	machine_ops = xen_machine_ops;
1452 
1453 	/*
1454 	 * The only reliable way to retain the initial address of the
1455 	 * percpu gdt_page is to remember it here, so we can go and
1456 	 * mark it RW later, when the initial percpu area is freed.
1457 	 */
1458 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1459 
1460 	xen_smp_init();
1461 
1462 #ifdef CONFIG_ACPI_NUMA
1463 	/*
1464 	 * The pages we from Xen are not related to machine pages, so
1465 	 * any NUMA information the kernel tries to get from ACPI will
1466 	 * be meaningless.  Prevent it from trying.
1467 	 */
1468 	disable_srat();
1469 #endif
1470 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1471 
1472 	local_irq_disable();
1473 
1474 	xen_raw_console_write("mapping kernel into physical memory\n");
1475 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1476 				   xen_start_info->nr_pages);
1477 	xen_reserve_special_pages();
1478 
1479 	/*
1480 	 * We used to do this in xen_arch_setup, but that is too late
1481 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1482 	 * early_amd_init which pokes 0xcf8 port.
1483 	 */
1484 	set_iopl.iopl = 1;
1485 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1486 	if (rc != 0)
1487 		xen_raw_printk("physdev_op failed %d\n", rc);
1488 
1489 
1490 	if (xen_start_info->mod_start) {
1491 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1492 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1493 	    else
1494 		initrd_start = __pa(xen_start_info->mod_start);
1495 	}
1496 
1497 	/* Poke various useful things into boot_params */
1498 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1499 	boot_params.hdr.ramdisk_image = initrd_start;
1500 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1501 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1502 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1503 
1504 	if (!xen_initial_domain()) {
1505 		if (pci_xen)
1506 			x86_init.pci.arch_init = pci_xen_init;
1507 		x86_platform.set_legacy_features =
1508 				xen_domu_set_legacy_features;
1509 	} else {
1510 		const struct dom0_vga_console_info *info =
1511 			(void *)((char *)xen_start_info +
1512 				 xen_start_info->console.dom0.info_off);
1513 		struct xen_platform_op op = {
1514 			.cmd = XENPF_firmware_info,
1515 			.interface_version = XENPF_INTERFACE_VERSION,
1516 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1517 		};
1518 
1519 		x86_platform.set_legacy_features =
1520 				xen_dom0_set_legacy_features;
1521 		xen_init_vga(info, xen_start_info->console.dom0.info_size,
1522 			     &boot_params.screen_info);
1523 		xen_start_info->console.domU.mfn = 0;
1524 		xen_start_info->console.domU.evtchn = 0;
1525 
1526 		if (HYPERVISOR_platform_op(&op) == 0)
1527 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1528 
1529 		/* Make sure ACS will be enabled */
1530 		pci_request_acs();
1531 
1532 		xen_acpi_sleep_register();
1533 
1534 		xen_boot_params_init_edd();
1535 
1536 #ifdef CONFIG_ACPI
1537 		/*
1538 		 * Disable selecting "Firmware First mode" for correctable
1539 		 * memory errors, as this is the duty of the hypervisor to
1540 		 * decide.
1541 		 */
1542 		acpi_disable_cmcff = 1;
1543 #endif
1544 	}
1545 
1546 	xen_add_preferred_consoles();
1547 
1548 #ifdef CONFIG_PCI
1549 	/* PCI BIOS service won't work from a PV guest. */
1550 	pci_probe &= ~PCI_PROBE_BIOS;
1551 #endif
1552 	xen_raw_console_write("about to get started...\n");
1553 
1554 	/* We need this for printk timestamps */
1555 	xen_setup_runstate_info(0);
1556 
1557 	xen_efi_init(&boot_params);
1558 
1559 	/* Start the world */
1560 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1561 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1562 }
1563 
1564 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1565 {
1566 	int rc;
1567 
1568 	if (per_cpu(xen_vcpu, cpu) == NULL)
1569 		return -ENODEV;
1570 
1571 	xen_setup_timer(cpu);
1572 
1573 	rc = xen_smp_intr_init(cpu);
1574 	if (rc) {
1575 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1576 		     cpu, rc);
1577 		return rc;
1578 	}
1579 
1580 	rc = xen_smp_intr_init_pv(cpu);
1581 	if (rc) {
1582 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1583 		     cpu, rc);
1584 		return rc;
1585 	}
1586 
1587 	return 0;
1588 }
1589 
1590 static int xen_cpu_dead_pv(unsigned int cpu)
1591 {
1592 	xen_smp_intr_free(cpu);
1593 	xen_smp_intr_free_pv(cpu);
1594 
1595 	xen_teardown_timer(cpu);
1596 
1597 	return 0;
1598 }
1599 
1600 static uint32_t __init xen_platform_pv(void)
1601 {
1602 	if (xen_pv_domain())
1603 		return xen_cpuid_base();
1604 
1605 	return 0;
1606 }
1607 
1608 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1609 	.name                   = "Xen PV",
1610 	.detect                 = xen_platform_pv,
1611 	.type			= X86_HYPER_XEN_PV,
1612 	.runtime.pin_vcpu       = xen_pin_vcpu,
1613 	.ignore_nopv		= true,
1614 };
1615