• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/highmem.h>
31 #include <linux/console.h>
32 #include <linux/pci.h>
33 #include <linux/gfp.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
36 
37 #include <xen/xen.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
47 #include <xen/page.h>
48 #include <xen/hvc-console.h>
49 #include <xen/acpi.h>
50 
51 #include <asm/paravirt.h>
52 #include <asm/apic.h>
53 #include <asm/page.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
64 #include <asm/desc.h>
65 #include <asm/pgalloc.h>
66 #include <asm/pgtable.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/stackprotector.h>
70 #include <asm/hypervisor.h>
71 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
74 #include <asm/cpu.h>
75 
76 #ifdef CONFIG_ACPI
77 #include <linux/acpi.h>
78 #include <asm/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
82 #endif
83 
84 #include "xen-ops.h"
85 #include "mmu.h"
86 #include "smp.h"
87 #include "multicalls.h"
88 #include "pmu.h"
89 
90 #include "../kernel/cpu/cpu.h" /* get_cpu_cap() */
91 
92 void *xen_initial_gdt;
93 
94 static int xen_cpu_up_prepare_pv(unsigned int cpu);
95 static int xen_cpu_dead_pv(unsigned int cpu);
96 
97 struct tls_descs {
98 	struct desc_struct desc[3];
99 };
100 
101 /*
102  * Updating the 3 TLS descriptors in the GDT on every task switch is
103  * surprisingly expensive so we avoid updating them if they haven't
104  * changed.  Since Xen writes different descriptors than the one
105  * passed in the update_descriptor hypercall we keep shadow copies to
106  * compare against.
107  */
108 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
109 
xen_banner(void)110 static void __init xen_banner(void)
111 {
112 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
113 	struct xen_extraversion extra;
114 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
115 
116 	pr_info("Booting paravirtualized kernel on %s\n", pv_info.name);
117 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
118 	       version >> 16, version & 0xffff, extra.extraversion,
119 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
120 
121 #ifdef CONFIG_X86_32
122 	pr_warn("WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! WARNING!\n"
123 		"Support for running as 32-bit PV-guest under Xen will soon be removed\n"
124 		"from the Linux kernel!\n"
125 		"Please use either a 64-bit kernel or switch to HVM or PVH mode!\n"
126 		"WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! WARNING!\n");
127 #endif
128 }
129 
xen_pv_init_platform(void)130 static void __init xen_pv_init_platform(void)
131 {
132 	populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));
133 
134 	set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
135 	HYPERVISOR_shared_info = (void *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
136 
137 	/* xen clock uses per-cpu vcpu_info, need to init it for boot cpu */
138 	xen_vcpu_info_reset(0);
139 
140 	/* pvclock is in shared info area */
141 	xen_init_time_ops();
142 }
143 
xen_pv_guest_late_init(void)144 static void __init xen_pv_guest_late_init(void)
145 {
146 #ifndef CONFIG_SMP
147 	/* Setup shared vcpu info for non-smp configurations */
148 	xen_setup_vcpu_info_placement();
149 #endif
150 }
151 
152 /* Check if running on Xen version (major, minor) or later */
153 bool
xen_running_on_version_or_later(unsigned int major,unsigned int minor)154 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
155 {
156 	unsigned int version;
157 
158 	if (!xen_domain())
159 		return false;
160 
161 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
162 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
163 		((version >> 16) > major))
164 		return true;
165 	return false;
166 }
167 
168 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
169 static __read_mostly unsigned int cpuid_leaf5_edx_val;
170 
xen_cpuid(unsigned int * ax,unsigned int * bx,unsigned int * cx,unsigned int * dx)171 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
172 		      unsigned int *cx, unsigned int *dx)
173 {
174 	unsigned maskebx = ~0;
175 
176 	/*
177 	 * Mask out inconvenient features, to try and disable as many
178 	 * unsupported kernel subsystems as possible.
179 	 */
180 	switch (*ax) {
181 	case CPUID_MWAIT_LEAF:
182 		/* Synthesize the values.. */
183 		*ax = 0;
184 		*bx = 0;
185 		*cx = cpuid_leaf5_ecx_val;
186 		*dx = cpuid_leaf5_edx_val;
187 		return;
188 
189 	case 0xb:
190 		/* Suppress extended topology stuff */
191 		maskebx = 0;
192 		break;
193 	}
194 
195 	asm(XEN_EMULATE_PREFIX "cpuid"
196 		: "=a" (*ax),
197 		  "=b" (*bx),
198 		  "=c" (*cx),
199 		  "=d" (*dx)
200 		: "0" (*ax), "2" (*cx));
201 
202 	*bx &= maskebx;
203 }
204 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
205 
xen_check_mwait(void)206 static bool __init xen_check_mwait(void)
207 {
208 #ifdef CONFIG_ACPI
209 	struct xen_platform_op op = {
210 		.cmd			= XENPF_set_processor_pminfo,
211 		.u.set_pminfo.id	= -1,
212 		.u.set_pminfo.type	= XEN_PM_PDC,
213 	};
214 	uint32_t buf[3];
215 	unsigned int ax, bx, cx, dx;
216 	unsigned int mwait_mask;
217 
218 	/* We need to determine whether it is OK to expose the MWAIT
219 	 * capability to the kernel to harvest deeper than C3 states from ACPI
220 	 * _CST using the processor_harvest_xen.c module. For this to work, we
221 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
222 	 * checks against). The hypervisor won't expose the MWAIT flag because
223 	 * it would break backwards compatibility; so we will find out directly
224 	 * from the hardware and hypercall.
225 	 */
226 	if (!xen_initial_domain())
227 		return false;
228 
229 	/*
230 	 * When running under platform earlier than Xen4.2, do not expose
231 	 * mwait, to avoid the risk of loading native acpi pad driver
232 	 */
233 	if (!xen_running_on_version_or_later(4, 2))
234 		return false;
235 
236 	ax = 1;
237 	cx = 0;
238 
239 	native_cpuid(&ax, &bx, &cx, &dx);
240 
241 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
242 		     (1 << (X86_FEATURE_MWAIT % 32));
243 
244 	if ((cx & mwait_mask) != mwait_mask)
245 		return false;
246 
247 	/* We need to emulate the MWAIT_LEAF and for that we need both
248 	 * ecx and edx. The hypercall provides only partial information.
249 	 */
250 
251 	ax = CPUID_MWAIT_LEAF;
252 	bx = 0;
253 	cx = 0;
254 	dx = 0;
255 
256 	native_cpuid(&ax, &bx, &cx, &dx);
257 
258 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
259 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
260 	 */
261 	buf[0] = ACPI_PDC_REVISION_ID;
262 	buf[1] = 1;
263 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
264 
265 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
266 
267 	if ((HYPERVISOR_platform_op(&op) == 0) &&
268 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
269 		cpuid_leaf5_ecx_val = cx;
270 		cpuid_leaf5_edx_val = dx;
271 	}
272 	return true;
273 #else
274 	return false;
275 #endif
276 }
277 
xen_check_xsave(void)278 static bool __init xen_check_xsave(void)
279 {
280 	unsigned int cx, xsave_mask;
281 
282 	cx = cpuid_ecx(1);
283 
284 	xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) |
285 		     (1 << (X86_FEATURE_OSXSAVE % 32));
286 
287 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
288 	return (cx & xsave_mask) == xsave_mask;
289 }
290 
xen_init_capabilities(void)291 static void __init xen_init_capabilities(void)
292 {
293 	setup_force_cpu_cap(X86_FEATURE_XENPV);
294 	setup_clear_cpu_cap(X86_FEATURE_DCA);
295 	setup_clear_cpu_cap(X86_FEATURE_APERFMPERF);
296 	setup_clear_cpu_cap(X86_FEATURE_MTRR);
297 	setup_clear_cpu_cap(X86_FEATURE_ACC);
298 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
299 	setup_clear_cpu_cap(X86_FEATURE_SME);
300 
301 	/*
302 	 * Xen PV would need some work to support PCID: CR3 handling as well
303 	 * as xen_flush_tlb_others() would need updating.
304 	 */
305 	setup_clear_cpu_cap(X86_FEATURE_PCID);
306 
307 	if (!xen_initial_domain())
308 		setup_clear_cpu_cap(X86_FEATURE_ACPI);
309 
310 	if (xen_check_mwait())
311 		setup_force_cpu_cap(X86_FEATURE_MWAIT);
312 	else
313 		setup_clear_cpu_cap(X86_FEATURE_MWAIT);
314 
315 	if (!xen_check_xsave()) {
316 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
317 		setup_clear_cpu_cap(X86_FEATURE_OSXSAVE);
318 	}
319 }
320 
xen_set_debugreg(int reg,unsigned long val)321 static void xen_set_debugreg(int reg, unsigned long val)
322 {
323 	HYPERVISOR_set_debugreg(reg, val);
324 }
325 
xen_get_debugreg(int reg)326 static unsigned long xen_get_debugreg(int reg)
327 {
328 	return HYPERVISOR_get_debugreg(reg);
329 }
330 
xen_end_context_switch(struct task_struct * next)331 static void xen_end_context_switch(struct task_struct *next)
332 {
333 	xen_mc_flush();
334 	paravirt_end_context_switch(next);
335 }
336 
xen_store_tr(void)337 static unsigned long xen_store_tr(void)
338 {
339 	return 0;
340 }
341 
342 /*
343  * Set the page permissions for a particular virtual address.  If the
344  * address is a vmalloc mapping (or other non-linear mapping), then
345  * find the linear mapping of the page and also set its protections to
346  * match.
347  */
set_aliased_prot(void * v,pgprot_t prot)348 static void set_aliased_prot(void *v, pgprot_t prot)
349 {
350 	int level;
351 	pte_t *ptep;
352 	pte_t pte;
353 	unsigned long pfn;
354 	struct page *page;
355 	unsigned char dummy;
356 
357 	ptep = lookup_address((unsigned long)v, &level);
358 	BUG_ON(ptep == NULL);
359 
360 	pfn = pte_pfn(*ptep);
361 	page = pfn_to_page(pfn);
362 
363 	pte = pfn_pte(pfn, prot);
364 
365 	/*
366 	 * Careful: update_va_mapping() will fail if the virtual address
367 	 * we're poking isn't populated in the page tables.  We don't
368 	 * need to worry about the direct map (that's always in the page
369 	 * tables), but we need to be careful about vmap space.  In
370 	 * particular, the top level page table can lazily propagate
371 	 * entries between processes, so if we've switched mms since we
372 	 * vmapped the target in the first place, we might not have the
373 	 * top-level page table entry populated.
374 	 *
375 	 * We disable preemption because we want the same mm active when
376 	 * we probe the target and when we issue the hypercall.  We'll
377 	 * have the same nominal mm, but if we're a kernel thread, lazy
378 	 * mm dropping could change our pgd.
379 	 *
380 	 * Out of an abundance of caution, this uses __get_user() to fault
381 	 * in the target address just in case there's some obscure case
382 	 * in which the target address isn't readable.
383 	 */
384 
385 	preempt_disable();
386 
387 	probe_kernel_read(&dummy, v, 1);
388 
389 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
390 		BUG();
391 
392 	if (!PageHighMem(page)) {
393 		void *av = __va(PFN_PHYS(pfn));
394 
395 		if (av != v)
396 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
397 				BUG();
398 	} else
399 		kmap_flush_unused();
400 
401 	preempt_enable();
402 }
403 
xen_alloc_ldt(struct desc_struct * ldt,unsigned entries)404 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
405 {
406 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
407 	int i;
408 
409 	/*
410 	 * We need to mark the all aliases of the LDT pages RO.  We
411 	 * don't need to call vm_flush_aliases(), though, since that's
412 	 * only responsible for flushing aliases out the TLBs, not the
413 	 * page tables, and Xen will flush the TLB for us if needed.
414 	 *
415 	 * To avoid confusing future readers: none of this is necessary
416 	 * to load the LDT.  The hypervisor only checks this when the
417 	 * LDT is faulted in due to subsequent descriptor access.
418 	 */
419 
420 	for (i = 0; i < entries; i += entries_per_page)
421 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
422 }
423 
xen_free_ldt(struct desc_struct * ldt,unsigned entries)424 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
425 {
426 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
427 	int i;
428 
429 	for (i = 0; i < entries; i += entries_per_page)
430 		set_aliased_prot(ldt + i, PAGE_KERNEL);
431 }
432 
xen_set_ldt(const void * addr,unsigned entries)433 static void xen_set_ldt(const void *addr, unsigned entries)
434 {
435 	struct mmuext_op *op;
436 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
437 
438 	trace_xen_cpu_set_ldt(addr, entries);
439 
440 	op = mcs.args;
441 	op->cmd = MMUEXT_SET_LDT;
442 	op->arg1.linear_addr = (unsigned long)addr;
443 	op->arg2.nr_ents = entries;
444 
445 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
446 
447 	xen_mc_issue(PARAVIRT_LAZY_CPU);
448 }
449 
xen_load_gdt(const struct desc_ptr * dtr)450 static void xen_load_gdt(const struct desc_ptr *dtr)
451 {
452 	unsigned long va = dtr->address;
453 	unsigned int size = dtr->size + 1;
454 	unsigned long pfn, mfn;
455 	int level;
456 	pte_t *ptep;
457 	void *virt;
458 
459 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
460 	BUG_ON(size > PAGE_SIZE);
461 	BUG_ON(va & ~PAGE_MASK);
462 
463 	/*
464 	 * The GDT is per-cpu and is in the percpu data area.
465 	 * That can be virtually mapped, so we need to do a
466 	 * page-walk to get the underlying MFN for the
467 	 * hypercall.  The page can also be in the kernel's
468 	 * linear range, so we need to RO that mapping too.
469 	 */
470 	ptep = lookup_address(va, &level);
471 	BUG_ON(ptep == NULL);
472 
473 	pfn = pte_pfn(*ptep);
474 	mfn = pfn_to_mfn(pfn);
475 	virt = __va(PFN_PHYS(pfn));
476 
477 	make_lowmem_page_readonly((void *)va);
478 	make_lowmem_page_readonly(virt);
479 
480 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
481 		BUG();
482 }
483 
484 /*
485  * load_gdt for early boot, when the gdt is only mapped once
486  */
xen_load_gdt_boot(const struct desc_ptr * dtr)487 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
488 {
489 	unsigned long va = dtr->address;
490 	unsigned int size = dtr->size + 1;
491 	unsigned long pfn, mfn;
492 	pte_t pte;
493 
494 	/* @size should be at most GDT_SIZE which is smaller than PAGE_SIZE. */
495 	BUG_ON(size > PAGE_SIZE);
496 	BUG_ON(va & ~PAGE_MASK);
497 
498 	pfn = virt_to_pfn(va);
499 	mfn = pfn_to_mfn(pfn);
500 
501 	pte = pfn_pte(pfn, PAGE_KERNEL_RO);
502 
503 	if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
504 		BUG();
505 
506 	if (HYPERVISOR_set_gdt(&mfn, size / sizeof(struct desc_struct)))
507 		BUG();
508 }
509 
desc_equal(const struct desc_struct * d1,const struct desc_struct * d2)510 static inline bool desc_equal(const struct desc_struct *d1,
511 			      const struct desc_struct *d2)
512 {
513 	return !memcmp(d1, d2, sizeof(*d1));
514 }
515 
load_TLS_descriptor(struct thread_struct * t,unsigned int cpu,unsigned int i)516 static void load_TLS_descriptor(struct thread_struct *t,
517 				unsigned int cpu, unsigned int i)
518 {
519 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
520 	struct desc_struct *gdt;
521 	xmaddr_t maddr;
522 	struct multicall_space mc;
523 
524 	if (desc_equal(shadow, &t->tls_array[i]))
525 		return;
526 
527 	*shadow = t->tls_array[i];
528 
529 	gdt = get_cpu_gdt_rw(cpu);
530 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
531 	mc = __xen_mc_entry(0);
532 
533 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
534 }
535 
xen_load_tls(struct thread_struct * t,unsigned int cpu)536 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
537 {
538 	/*
539 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
540 	 * and lazy gs handling is enabled, it means we're in a
541 	 * context switch, and %gs has just been saved.  This means we
542 	 * can zero it out to prevent faults on exit from the
543 	 * hypervisor if the next process has no %gs.  Either way, it
544 	 * has been saved, and the new value will get loaded properly.
545 	 * This will go away as soon as Xen has been modified to not
546 	 * save/restore %gs for normal hypercalls.
547 	 *
548 	 * On x86_64, this hack is not used for %gs, because gs points
549 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
550 	 * must not zero %gs on x86_64
551 	 *
552 	 * For x86_64, we need to zero %fs, otherwise we may get an
553 	 * exception between the new %fs descriptor being loaded and
554 	 * %fs being effectively cleared at __switch_to().
555 	 */
556 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
557 #ifdef CONFIG_X86_32
558 		lazy_load_gs(0);
559 #else
560 		loadsegment(fs, 0);
561 #endif
562 	}
563 
564 	xen_mc_batch();
565 
566 	load_TLS_descriptor(t, cpu, 0);
567 	load_TLS_descriptor(t, cpu, 1);
568 	load_TLS_descriptor(t, cpu, 2);
569 
570 	xen_mc_issue(PARAVIRT_LAZY_CPU);
571 }
572 
573 #ifdef CONFIG_X86_64
xen_load_gs_index(unsigned int idx)574 static void xen_load_gs_index(unsigned int idx)
575 {
576 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
577 		BUG();
578 }
579 #endif
580 
xen_write_ldt_entry(struct desc_struct * dt,int entrynum,const void * ptr)581 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
582 				const void *ptr)
583 {
584 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
585 	u64 entry = *(u64 *)ptr;
586 
587 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
588 
589 	preempt_disable();
590 
591 	xen_mc_flush();
592 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
593 		BUG();
594 
595 	preempt_enable();
596 }
597 
598 #ifdef CONFIG_X86_64
599 struct trap_array_entry {
600 	void (*orig)(void);
601 	void (*xen)(void);
602 	bool ist_okay;
603 };
604 
605 static struct trap_array_entry trap_array[] = {
606 	{ debug,                       xen_xendebug,                    true },
607 	{ double_fault,                xen_double_fault,                true },
608 #ifdef CONFIG_X86_MCE
609 	{ machine_check,               xen_machine_check,               true },
610 #endif
611 	{ nmi,                         xen_xennmi,                      true },
612 	{ int3,                        xen_int3,                        false },
613 	{ overflow,                    xen_overflow,                    false },
614 #ifdef CONFIG_IA32_EMULATION
615 	{ entry_INT80_compat,          xen_entry_INT80_compat,          false },
616 #endif
617 	{ page_fault,                  xen_page_fault,                  false },
618 	{ divide_error,                xen_divide_error,                false },
619 	{ bounds,                      xen_bounds,                      false },
620 	{ invalid_op,                  xen_invalid_op,                  false },
621 	{ device_not_available,        xen_device_not_available,        false },
622 	{ coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false },
623 	{ invalid_TSS,                 xen_invalid_TSS,                 false },
624 	{ segment_not_present,         xen_segment_not_present,         false },
625 	{ stack_segment,               xen_stack_segment,               false },
626 	{ general_protection,          xen_general_protection,          false },
627 	{ spurious_interrupt_bug,      xen_spurious_interrupt_bug,      false },
628 	{ coprocessor_error,           xen_coprocessor_error,           false },
629 	{ alignment_check,             xen_alignment_check,             false },
630 	{ simd_coprocessor_error,      xen_simd_coprocessor_error,      false },
631 };
632 
get_trap_addr(void ** addr,unsigned int ist)633 static bool __ref get_trap_addr(void **addr, unsigned int ist)
634 {
635 	unsigned int nr;
636 	bool ist_okay = false;
637 
638 	/*
639 	 * Replace trap handler addresses by Xen specific ones.
640 	 * Check for known traps using IST and whitelist them.
641 	 * The debugger ones are the only ones we care about.
642 	 * Xen will handle faults like double_fault, * so we should never see
643 	 * them.  Warn if there's an unexpected IST-using fault handler.
644 	 */
645 	for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) {
646 		struct trap_array_entry *entry = trap_array + nr;
647 
648 		if (*addr == entry->orig) {
649 			*addr = entry->xen;
650 			ist_okay = entry->ist_okay;
651 			break;
652 		}
653 	}
654 
655 	if (nr == ARRAY_SIZE(trap_array) &&
656 	    *addr >= (void *)early_idt_handler_array[0] &&
657 	    *addr < (void *)early_idt_handler_array[NUM_EXCEPTION_VECTORS]) {
658 		nr = (*addr - (void *)early_idt_handler_array[0]) /
659 		     EARLY_IDT_HANDLER_SIZE;
660 		*addr = (void *)xen_early_idt_handler_array[nr];
661 	}
662 
663 	if (WARN_ON(ist != 0 && !ist_okay))
664 		return false;
665 
666 	return true;
667 }
668 #endif
669 
cvt_gate_to_trap(int vector,const gate_desc * val,struct trap_info * info)670 static int cvt_gate_to_trap(int vector, const gate_desc *val,
671 			    struct trap_info *info)
672 {
673 	unsigned long addr;
674 
675 	if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT)
676 		return 0;
677 
678 	info->vector = vector;
679 
680 	addr = gate_offset(val);
681 #ifdef CONFIG_X86_64
682 	if (!get_trap_addr((void **)&addr, val->bits.ist))
683 		return 0;
684 #endif	/* CONFIG_X86_64 */
685 	info->address = addr;
686 
687 	info->cs = gate_segment(val);
688 	info->flags = val->bits.dpl;
689 	/* interrupt gates clear IF */
690 	if (val->bits.type == GATE_INTERRUPT)
691 		info->flags |= 1 << 2;
692 
693 	return 1;
694 }
695 
696 /* Locations of each CPU's IDT */
697 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
698 
699 /* Set an IDT entry.  If the entry is part of the current IDT, then
700    also update Xen. */
xen_write_idt_entry(gate_desc * dt,int entrynum,const gate_desc * g)701 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
702 {
703 	unsigned long p = (unsigned long)&dt[entrynum];
704 	unsigned long start, end;
705 
706 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
707 
708 	preempt_disable();
709 
710 	start = __this_cpu_read(idt_desc.address);
711 	end = start + __this_cpu_read(idt_desc.size) + 1;
712 
713 	xen_mc_flush();
714 
715 	native_write_idt_entry(dt, entrynum, g);
716 
717 	if (p >= start && (p + 8) <= end) {
718 		struct trap_info info[2];
719 
720 		info[1].address = 0;
721 
722 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
723 			if (HYPERVISOR_set_trap_table(info))
724 				BUG();
725 	}
726 
727 	preempt_enable();
728 }
729 
xen_convert_trap_info(const struct desc_ptr * desc,struct trap_info * traps,bool full)730 static unsigned xen_convert_trap_info(const struct desc_ptr *desc,
731 				      struct trap_info *traps, bool full)
732 {
733 	unsigned in, out, count;
734 
735 	count = (desc->size+1) / sizeof(gate_desc);
736 	BUG_ON(count > 256);
737 
738 	for (in = out = 0; in < count; in++) {
739 		gate_desc *entry = (gate_desc *)(desc->address) + in;
740 
741 		if (cvt_gate_to_trap(in, entry, &traps[out]) || full)
742 			out++;
743 	}
744 
745 	return out;
746 }
747 
xen_copy_trap_info(struct trap_info * traps)748 void xen_copy_trap_info(struct trap_info *traps)
749 {
750 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
751 
752 	xen_convert_trap_info(desc, traps, true);
753 }
754 
755 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
756    hold a spinlock to protect the static traps[] array (static because
757    it avoids allocation, and saves stack space). */
xen_load_idt(const struct desc_ptr * desc)758 static void xen_load_idt(const struct desc_ptr *desc)
759 {
760 	static DEFINE_SPINLOCK(lock);
761 	static struct trap_info traps[257];
762 	unsigned out;
763 
764 	trace_xen_cpu_load_idt(desc);
765 
766 	spin_lock(&lock);
767 
768 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
769 
770 	out = xen_convert_trap_info(desc, traps, false);
771 	memset(&traps[out], 0, sizeof(traps[0]));
772 
773 	xen_mc_flush();
774 	if (HYPERVISOR_set_trap_table(traps))
775 		BUG();
776 
777 	spin_unlock(&lock);
778 }
779 
780 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
781    they're handled differently. */
xen_write_gdt_entry(struct desc_struct * dt,int entry,const void * desc,int type)782 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
783 				const void *desc, int type)
784 {
785 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
786 
787 	preempt_disable();
788 
789 	switch (type) {
790 	case DESC_LDT:
791 	case DESC_TSS:
792 		/* ignore */
793 		break;
794 
795 	default: {
796 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
797 
798 		xen_mc_flush();
799 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
800 			BUG();
801 	}
802 
803 	}
804 
805 	preempt_enable();
806 }
807 
808 /*
809  * Version of write_gdt_entry for use at early boot-time needed to
810  * update an entry as simply as possible.
811  */
xen_write_gdt_entry_boot(struct desc_struct * dt,int entry,const void * desc,int type)812 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
813 					    const void *desc, int type)
814 {
815 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
816 
817 	switch (type) {
818 	case DESC_LDT:
819 	case DESC_TSS:
820 		/* ignore */
821 		break;
822 
823 	default: {
824 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
825 
826 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
827 			dt[entry] = *(struct desc_struct *)desc;
828 	}
829 
830 	}
831 }
832 
xen_load_sp0(unsigned long sp0)833 static void xen_load_sp0(unsigned long sp0)
834 {
835 	struct multicall_space mcs;
836 
837 	mcs = xen_mc_entry(0);
838 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, sp0);
839 	xen_mc_issue(PARAVIRT_LAZY_CPU);
840 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
841 }
842 
xen_set_iopl_mask(unsigned mask)843 void xen_set_iopl_mask(unsigned mask)
844 {
845 	struct physdev_set_iopl set_iopl;
846 
847 	/* Force the change at ring 0. */
848 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
849 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
850 }
851 
xen_io_delay(void)852 static void xen_io_delay(void)
853 {
854 }
855 
856 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
857 
xen_read_cr0(void)858 static unsigned long xen_read_cr0(void)
859 {
860 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
861 
862 	if (unlikely(cr0 == 0)) {
863 		cr0 = native_read_cr0();
864 		this_cpu_write(xen_cr0_value, cr0);
865 	}
866 
867 	return cr0;
868 }
869 
xen_write_cr0(unsigned long cr0)870 static void xen_write_cr0(unsigned long cr0)
871 {
872 	struct multicall_space mcs;
873 
874 	this_cpu_write(xen_cr0_value, cr0);
875 
876 	/* Only pay attention to cr0.TS; everything else is
877 	   ignored. */
878 	mcs = xen_mc_entry(0);
879 
880 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
881 
882 	xen_mc_issue(PARAVIRT_LAZY_CPU);
883 }
884 
xen_write_cr4(unsigned long cr4)885 static void xen_write_cr4(unsigned long cr4)
886 {
887 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
888 
889 	native_write_cr4(cr4);
890 }
891 
xen_read_msr_safe(unsigned int msr,int * err)892 static u64 xen_read_msr_safe(unsigned int msr, int *err)
893 {
894 	u64 val;
895 
896 	if (pmu_msr_read(msr, &val, err))
897 		return val;
898 
899 	val = native_read_msr_safe(msr, err);
900 	switch (msr) {
901 	case MSR_IA32_APICBASE:
902 		val &= ~X2APIC_ENABLE;
903 		break;
904 	}
905 	return val;
906 }
907 
xen_write_msr_safe(unsigned int msr,unsigned low,unsigned high)908 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
909 {
910 	int ret;
911 #ifdef CONFIG_X86_64
912 	unsigned int which;
913 	u64 base;
914 #endif
915 
916 	ret = 0;
917 
918 	switch (msr) {
919 #ifdef CONFIG_X86_64
920 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
921 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
922 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
923 
924 	set:
925 		base = ((u64)high << 32) | low;
926 		if (HYPERVISOR_set_segment_base(which, base) != 0)
927 			ret = -EIO;
928 		break;
929 #endif
930 
931 	case MSR_STAR:
932 	case MSR_CSTAR:
933 	case MSR_LSTAR:
934 	case MSR_SYSCALL_MASK:
935 	case MSR_IA32_SYSENTER_CS:
936 	case MSR_IA32_SYSENTER_ESP:
937 	case MSR_IA32_SYSENTER_EIP:
938 		/* Fast syscall setup is all done in hypercalls, so
939 		   these are all ignored.  Stub them out here to stop
940 		   Xen console noise. */
941 		break;
942 
943 	default:
944 		if (!pmu_msr_write(msr, low, high, &ret))
945 			ret = native_write_msr_safe(msr, low, high);
946 	}
947 
948 	return ret;
949 }
950 
xen_read_msr(unsigned int msr)951 static u64 xen_read_msr(unsigned int msr)
952 {
953 	/*
954 	 * This will silently swallow a #GP from RDMSR.  It may be worth
955 	 * changing that.
956 	 */
957 	int err;
958 
959 	return xen_read_msr_safe(msr, &err);
960 }
961 
xen_write_msr(unsigned int msr,unsigned low,unsigned high)962 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
963 {
964 	/*
965 	 * This will silently swallow a #GP from WRMSR.  It may be worth
966 	 * changing that.
967 	 */
968 	xen_write_msr_safe(msr, low, high);
969 }
970 
971 /* This is called once we have the cpu_possible_mask */
xen_setup_vcpu_info_placement(void)972 void __init xen_setup_vcpu_info_placement(void)
973 {
974 	int cpu;
975 
976 	for_each_possible_cpu(cpu) {
977 		/* Set up direct vCPU id mapping for PV guests. */
978 		per_cpu(xen_vcpu_id, cpu) = cpu;
979 
980 		/*
981 		 * xen_vcpu_setup(cpu) can fail  -- in which case it
982 		 * falls back to the shared_info version for cpus
983 		 * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS.
984 		 *
985 		 * xen_cpu_up_prepare_pv() handles the rest by failing
986 		 * them in hotplug.
987 		 */
988 		(void) xen_vcpu_setup(cpu);
989 	}
990 
991 	/*
992 	 * xen_vcpu_setup managed to place the vcpu_info within the
993 	 * percpu area for all cpus, so make use of it.
994 	 */
995 	if (xen_have_vcpu_info_placement) {
996 		pv_ops.irq.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
997 		pv_ops.irq.restore_fl =
998 			__PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
999 		pv_ops.irq.irq_disable =
1000 			__PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1001 		pv_ops.irq.irq_enable =
1002 			__PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1003 		pv_ops.mmu.read_cr2 =
1004 			__PV_IS_CALLEE_SAVE(xen_read_cr2_direct);
1005 	}
1006 }
1007 
1008 static const struct pv_info xen_info __initconst = {
1009 	.shared_kernel_pmd = 0,
1010 
1011 #ifdef CONFIG_X86_64
1012 	.extra_user_64bit_cs = FLAT_USER_CS64,
1013 #endif
1014 	.name = "Xen",
1015 };
1016 
1017 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
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 	.iret = xen_iret,
1039 #ifdef CONFIG_X86_64
1040 	.usergs_sysret64 = xen_sysret64,
1041 #endif
1042 
1043 	.load_tr_desc = paravirt_nop,
1044 	.set_ldt = xen_set_ldt,
1045 	.load_gdt = xen_load_gdt,
1046 	.load_idt = xen_load_idt,
1047 	.load_tls = xen_load_tls,
1048 #ifdef CONFIG_X86_64
1049 	.load_gs_index = xen_load_gs_index,
1050 #endif
1051 
1052 	.alloc_ldt = xen_alloc_ldt,
1053 	.free_ldt = xen_free_ldt,
1054 
1055 	.store_tr = xen_store_tr,
1056 
1057 	.write_ldt_entry = xen_write_ldt_entry,
1058 	.write_gdt_entry = xen_write_gdt_entry,
1059 	.write_idt_entry = xen_write_idt_entry,
1060 	.load_sp0 = xen_load_sp0,
1061 
1062 	.set_iopl_mask = xen_set_iopl_mask,
1063 	.io_delay = xen_io_delay,
1064 
1065 	/* Xen takes care of %gs when switching to usermode for us */
1066 	.swapgs = paravirt_nop,
1067 
1068 	.start_context_switch = paravirt_start_context_switch,
1069 	.end_context_switch = xen_end_context_switch,
1070 };
1071 
xen_restart(char * msg)1072 static void xen_restart(char *msg)
1073 {
1074 	xen_reboot(SHUTDOWN_reboot);
1075 }
1076 
xen_machine_halt(void)1077 static void xen_machine_halt(void)
1078 {
1079 	xen_reboot(SHUTDOWN_poweroff);
1080 }
1081 
xen_machine_power_off(void)1082 static void xen_machine_power_off(void)
1083 {
1084 	if (pm_power_off)
1085 		pm_power_off();
1086 	xen_reboot(SHUTDOWN_poweroff);
1087 }
1088 
xen_crash_shutdown(struct pt_regs * regs)1089 static void xen_crash_shutdown(struct pt_regs *regs)
1090 {
1091 	xen_reboot(SHUTDOWN_crash);
1092 }
1093 
1094 static const struct machine_ops xen_machine_ops __initconst = {
1095 	.restart = xen_restart,
1096 	.halt = xen_machine_halt,
1097 	.power_off = xen_machine_power_off,
1098 	.shutdown = xen_machine_halt,
1099 	.crash_shutdown = xen_crash_shutdown,
1100 	.emergency_restart = xen_emergency_restart,
1101 };
1102 
xen_get_nmi_reason(void)1103 static unsigned char xen_get_nmi_reason(void)
1104 {
1105 	unsigned char reason = 0;
1106 
1107 	/* Construct a value which looks like it came from port 0x61. */
1108 	if (test_bit(_XEN_NMIREASON_io_error,
1109 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1110 		reason |= NMI_REASON_IOCHK;
1111 	if (test_bit(_XEN_NMIREASON_pci_serr,
1112 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1113 		reason |= NMI_REASON_SERR;
1114 
1115 	return reason;
1116 }
1117 
xen_boot_params_init_edd(void)1118 static void __init xen_boot_params_init_edd(void)
1119 {
1120 #if IS_ENABLED(CONFIG_EDD)
1121 	struct xen_platform_op op;
1122 	struct edd_info *edd_info;
1123 	u32 *mbr_signature;
1124 	unsigned nr;
1125 	int ret;
1126 
1127 	edd_info = boot_params.eddbuf;
1128 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1129 
1130 	op.cmd = XENPF_firmware_info;
1131 
1132 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1133 	for (nr = 0; nr < EDDMAXNR; nr++) {
1134 		struct edd_info *info = edd_info + nr;
1135 
1136 		op.u.firmware_info.index = nr;
1137 		info->params.length = sizeof(info->params);
1138 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1139 				     &info->params);
1140 		ret = HYPERVISOR_platform_op(&op);
1141 		if (ret)
1142 			break;
1143 
1144 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1145 		C(device);
1146 		C(version);
1147 		C(interface_support);
1148 		C(legacy_max_cylinder);
1149 		C(legacy_max_head);
1150 		C(legacy_sectors_per_track);
1151 #undef C
1152 	}
1153 	boot_params.eddbuf_entries = nr;
1154 
1155 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1156 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1157 		op.u.firmware_info.index = nr;
1158 		ret = HYPERVISOR_platform_op(&op);
1159 		if (ret)
1160 			break;
1161 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1162 	}
1163 	boot_params.edd_mbr_sig_buf_entries = nr;
1164 #endif
1165 }
1166 
1167 /*
1168  * Set up the GDT and segment registers for -fstack-protector.  Until
1169  * we do this, we have to be careful not to call any stack-protected
1170  * function, which is most of the kernel.
1171  */
xen_setup_gdt(int cpu)1172 static void __init xen_setup_gdt(int cpu)
1173 {
1174 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry_boot;
1175 	pv_ops.cpu.load_gdt = xen_load_gdt_boot;
1176 
1177 	setup_stack_canary_segment(cpu);
1178 	switch_to_new_gdt(cpu);
1179 
1180 	pv_ops.cpu.write_gdt_entry = xen_write_gdt_entry;
1181 	pv_ops.cpu.load_gdt = xen_load_gdt;
1182 }
1183 
xen_dom0_set_legacy_features(void)1184 static void __init xen_dom0_set_legacy_features(void)
1185 {
1186 	x86_platform.legacy.rtc = 1;
1187 }
1188 
xen_domu_set_legacy_features(void)1189 static void __init xen_domu_set_legacy_features(void)
1190 {
1191 	x86_platform.legacy.rtc = 0;
1192 }
1193 
1194 /* First C function to be called on Xen boot */
xen_start_kernel(void)1195 asmlinkage __visible void __init xen_start_kernel(void)
1196 {
1197 	struct physdev_set_iopl set_iopl;
1198 	unsigned long initrd_start = 0;
1199 	int rc;
1200 
1201 	if (!xen_start_info)
1202 		return;
1203 
1204 	xen_domain_type = XEN_PV_DOMAIN;
1205 	xen_start_flags = xen_start_info->flags;
1206 
1207 	xen_setup_features();
1208 
1209 	/* Install Xen paravirt ops */
1210 	pv_info = xen_info;
1211 	pv_ops.init.patch = paravirt_patch_default;
1212 	pv_ops.cpu = xen_cpu_ops;
1213 	xen_init_irq_ops();
1214 
1215 	/*
1216 	 * Setup xen_vcpu early because it is needed for
1217 	 * local_irq_disable(), irqs_disabled(), e.g. in printk().
1218 	 *
1219 	 * Don't do the full vcpu_info placement stuff until we have
1220 	 * the cpu_possible_mask and a non-dummy shared_info.
1221 	 */
1222 	xen_vcpu_info_reset(0);
1223 
1224 	x86_platform.get_nmi_reason = xen_get_nmi_reason;
1225 
1226 	x86_init.resources.memory_setup = xen_memory_setup;
1227 	x86_init.irqs.intr_mode_select	= x86_init_noop;
1228 	x86_init.irqs.intr_mode_init	= x86_init_noop;
1229 	x86_init.oem.arch_setup = xen_arch_setup;
1230 	x86_init.oem.banner = xen_banner;
1231 	x86_init.hyper.init_platform = xen_pv_init_platform;
1232 	x86_init.hyper.guest_late_init = xen_pv_guest_late_init;
1233 
1234 	/*
1235 	 * Set up some pagetable state before starting to set any ptes.
1236 	 */
1237 
1238 	xen_setup_machphys_mapping();
1239 	xen_init_mmu_ops();
1240 
1241 	/* Prevent unwanted bits from being set in PTEs. */
1242 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1243 	__default_kernel_pte_mask &= ~_PAGE_GLOBAL;
1244 
1245 	/*
1246 	 * Prevent page tables from being allocated in highmem, even
1247 	 * if CONFIG_HIGHPTE is enabled.
1248 	 */
1249 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1250 
1251 	/* Get mfn list */
1252 	xen_build_dynamic_phys_to_machine();
1253 
1254 	/*
1255 	 * Set up kernel GDT and segment registers, mainly so that
1256 	 * -fstack-protector code can be executed.
1257 	 */
1258 	xen_setup_gdt(0);
1259 
1260 	/* Work out if we support NX */
1261 	get_cpu_cap(&boot_cpu_data);
1262 	x86_configure_nx();
1263 
1264 	/* Determine virtual and physical address sizes */
1265 	get_cpu_address_sizes(&boot_cpu_data);
1266 
1267 	/* Let's presume PV guests always boot on vCPU with id 0. */
1268 	per_cpu(xen_vcpu_id, 0) = 0;
1269 
1270 	idt_setup_early_handler();
1271 
1272 	xen_init_capabilities();
1273 
1274 #ifdef CONFIG_X86_LOCAL_APIC
1275 	/*
1276 	 * set up the basic apic ops.
1277 	 */
1278 	xen_init_apic();
1279 #endif
1280 
1281 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1282 		pv_ops.mmu.ptep_modify_prot_start =
1283 			xen_ptep_modify_prot_start;
1284 		pv_ops.mmu.ptep_modify_prot_commit =
1285 			xen_ptep_modify_prot_commit;
1286 	}
1287 
1288 	machine_ops = xen_machine_ops;
1289 
1290 	/*
1291 	 * The only reliable way to retain the initial address of the
1292 	 * percpu gdt_page is to remember it here, so we can go and
1293 	 * mark it RW later, when the initial percpu area is freed.
1294 	 */
1295 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1296 
1297 	xen_smp_init();
1298 
1299 #ifdef CONFIG_ACPI_NUMA
1300 	/*
1301 	 * The pages we from Xen are not related to machine pages, so
1302 	 * any NUMA information the kernel tries to get from ACPI will
1303 	 * be meaningless.  Prevent it from trying.
1304 	 */
1305 	acpi_numa = -1;
1306 #endif
1307 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv));
1308 
1309 	local_irq_disable();
1310 	early_boot_irqs_disabled = true;
1311 
1312 	xen_raw_console_write("mapping kernel into physical memory\n");
1313 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1314 				   xen_start_info->nr_pages);
1315 	xen_reserve_special_pages();
1316 
1317 	/* keep using Xen gdt for now; no urgent need to change it */
1318 
1319 #ifdef CONFIG_X86_32
1320 	pv_info.kernel_rpl = 1;
1321 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1322 		pv_info.kernel_rpl = 0;
1323 #else
1324 	pv_info.kernel_rpl = 0;
1325 #endif
1326 	/* set the limit of our address space */
1327 	xen_reserve_top();
1328 
1329 	/*
1330 	 * We used to do this in xen_arch_setup, but that is too late
1331 	 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1332 	 * early_amd_init which pokes 0xcf8 port.
1333 	 */
1334 	set_iopl.iopl = 1;
1335 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1336 	if (rc != 0)
1337 		xen_raw_printk("physdev_op failed %d\n", rc);
1338 
1339 #ifdef CONFIG_X86_32
1340 	/* set up basic CPUID stuff */
1341 	cpu_detect(&new_cpu_data);
1342 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1343 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1344 #endif
1345 
1346 	if (xen_start_info->mod_start) {
1347 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1348 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1349 	    else
1350 		initrd_start = __pa(xen_start_info->mod_start);
1351 	}
1352 
1353 	/* Poke various useful things into boot_params */
1354 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1355 	boot_params.hdr.ramdisk_image = initrd_start;
1356 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1357 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1358 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1359 
1360 	if (!xen_initial_domain()) {
1361 		add_preferred_console("xenboot", 0, NULL);
1362 		if (pci_xen)
1363 			x86_init.pci.arch_init = pci_xen_init;
1364 		x86_platform.set_legacy_features =
1365 				xen_domu_set_legacy_features;
1366 	} else {
1367 		const struct dom0_vga_console_info *info =
1368 			(void *)((char *)xen_start_info +
1369 				 xen_start_info->console.dom0.info_off);
1370 		struct xen_platform_op op = {
1371 			.cmd = XENPF_firmware_info,
1372 			.interface_version = XENPF_INTERFACE_VERSION,
1373 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1374 		};
1375 
1376 		x86_platform.set_legacy_features =
1377 				xen_dom0_set_legacy_features;
1378 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1379 		xen_start_info->console.domU.mfn = 0;
1380 		xen_start_info->console.domU.evtchn = 0;
1381 
1382 		if (HYPERVISOR_platform_op(&op) == 0)
1383 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1384 
1385 		/* Make sure ACS will be enabled */
1386 		pci_request_acs();
1387 
1388 		xen_acpi_sleep_register();
1389 
1390 		/* Avoid searching for BIOS MP tables */
1391 		x86_init.mpparse.find_smp_config = x86_init_noop;
1392 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1393 
1394 		xen_boot_params_init_edd();
1395 
1396 #ifdef CONFIG_ACPI
1397 		/*
1398 		 * Disable selecting "Firmware First mode" for correctable
1399 		 * memory errors, as this is the duty of the hypervisor to
1400 		 * decide.
1401 		 */
1402 		acpi_disable_cmcff = 1;
1403 #endif
1404 	}
1405 
1406 	if (!boot_params.screen_info.orig_video_isVGA)
1407 		add_preferred_console("tty", 0, NULL);
1408 	add_preferred_console("hvc", 0, NULL);
1409 	if (boot_params.screen_info.orig_video_isVGA)
1410 		add_preferred_console("tty", 0, NULL);
1411 
1412 #ifdef CONFIG_PCI
1413 	/* PCI BIOS service won't work from a PV guest. */
1414 	pci_probe &= ~PCI_PROBE_BIOS;
1415 #endif
1416 	xen_raw_console_write("about to get started...\n");
1417 
1418 	/* We need this for printk timestamps */
1419 	xen_setup_runstate_info(0);
1420 
1421 	xen_efi_init(&boot_params);
1422 
1423 	/* Start the world */
1424 #ifdef CONFIG_X86_32
1425 	i386_start_kernel();
1426 #else
1427 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1428 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1429 #endif
1430 }
1431 
xen_cpu_up_prepare_pv(unsigned int cpu)1432 static int xen_cpu_up_prepare_pv(unsigned int cpu)
1433 {
1434 	int rc;
1435 
1436 	if (per_cpu(xen_vcpu, cpu) == NULL)
1437 		return -ENODEV;
1438 
1439 	xen_setup_timer(cpu);
1440 
1441 	rc = xen_smp_intr_init(cpu);
1442 	if (rc) {
1443 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1444 		     cpu, rc);
1445 		return rc;
1446 	}
1447 
1448 	rc = xen_smp_intr_init_pv(cpu);
1449 	if (rc) {
1450 		WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1451 		     cpu, rc);
1452 		return rc;
1453 	}
1454 
1455 	return 0;
1456 }
1457 
xen_cpu_dead_pv(unsigned int cpu)1458 static int xen_cpu_dead_pv(unsigned int cpu)
1459 {
1460 	xen_smp_intr_free(cpu);
1461 	xen_smp_intr_free_pv(cpu);
1462 
1463 	xen_teardown_timer(cpu);
1464 
1465 	return 0;
1466 }
1467 
xen_platform_pv(void)1468 static uint32_t __init xen_platform_pv(void)
1469 {
1470 	if (xen_pv_domain())
1471 		return xen_cpuid_base();
1472 
1473 	return 0;
1474 }
1475 
1476 const __initconst struct hypervisor_x86 x86_hyper_xen_pv = {
1477 	.name                   = "Xen PV",
1478 	.detect                 = xen_platform_pv,
1479 	.type			= X86_HYPER_XEN_PV,
1480 	.runtime.pin_vcpu       = xen_pin_vcpu,
1481 	.ignore_nopv		= true,
1482 };
1483