• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Core of Xen paravirt_ops implementation.
3  *
4  * This file contains the xen_paravirt_ops structure itself, and the
5  * implementations for:
6  * - privileged instructions
7  * - interrupt flags
8  * - segment operations
9  * - booting and setup
10  *
11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12  */
13 
14 #include <linux/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/export.h>
27 #include <linux/mm.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
36 
37 #include <linux/kexec.h>
38 
39 #include <xen/xen.h>
40 #include <xen/events.h>
41 #include <xen/interface/xen.h>
42 #include <xen/interface/version.h>
43 #include <xen/interface/physdev.h>
44 #include <xen/interface/vcpu.h>
45 #include <xen/interface/memory.h>
46 #include <xen/interface/nmi.h>
47 #include <xen/interface/xen-mca.h>
48 #include <xen/features.h>
49 #include <xen/page.h>
50 #include <xen/hvm.h>
51 #include <xen/hvc-console.h>
52 #include <xen/acpi.h>
53 
54 #include <asm/paravirt.h>
55 #include <asm/apic.h>
56 #include <asm/page.h>
57 #include <asm/xen/pci.h>
58 #include <asm/xen/hypercall.h>
59 #include <asm/xen/hypervisor.h>
60 #include <asm/xen/cpuid.h>
61 #include <asm/fixmap.h>
62 #include <asm/processor.h>
63 #include <asm/proto.h>
64 #include <asm/msr-index.h>
65 #include <asm/traps.h>
66 #include <asm/setup.h>
67 #include <asm/desc.h>
68 #include <asm/pgalloc.h>
69 #include <asm/pgtable.h>
70 #include <asm/tlbflush.h>
71 #include <asm/reboot.h>
72 #include <asm/stackprotector.h>
73 #include <asm/hypervisor.h>
74 #include <asm/mach_traps.h>
75 #include <asm/mwait.h>
76 #include <asm/pci_x86.h>
77 #include <asm/cpu.h>
78 
79 #ifdef CONFIG_ACPI
80 #include <linux/acpi.h>
81 #include <asm/acpi.h>
82 #include <acpi/pdc_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
85 #endif
86 
87 #include "xen-ops.h"
88 #include "mmu.h"
89 #include "smp.h"
90 #include "multicalls.h"
91 #include "pmu.h"
92 
93 EXPORT_SYMBOL_GPL(hypercall_page);
94 
95 /*
96  * Pointer to the xen_vcpu_info structure or
97  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
98  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
99  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
100  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
101  * acknowledge pending events.
102  * Also more subtly it is used by the patched version of irq enable/disable
103  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
104  *
105  * The desire to be able to do those mask/unmask operations as a single
106  * instruction by using the per-cpu offset held in %gs is the real reason
107  * vcpu info is in a per-cpu pointer and the original reason for this
108  * hypercall.
109  *
110  */
111 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
112 
113 /*
114  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
115  * hypercall. This can be used both in PV and PVHVM mode. The structure
116  * overrides the default per_cpu(xen_vcpu, cpu) value.
117  */
118 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
119 
120 /* Linux <-> Xen vCPU id mapping */
121 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
122 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
123 
124 enum xen_domain_type xen_domain_type = XEN_NATIVE;
125 EXPORT_SYMBOL_GPL(xen_domain_type);
126 
127 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
128 EXPORT_SYMBOL(machine_to_phys_mapping);
129 unsigned long  machine_to_phys_nr;
130 EXPORT_SYMBOL(machine_to_phys_nr);
131 
132 struct start_info *xen_start_info;
133 EXPORT_SYMBOL_GPL(xen_start_info);
134 
135 struct shared_info xen_dummy_shared_info;
136 
137 void *xen_initial_gdt;
138 
139 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
140 __read_mostly int xen_have_vector_callback;
141 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
142 
143 static int xen_cpu_up_prepare(unsigned int cpu);
144 static int xen_cpu_up_online(unsigned int cpu);
145 static int xen_cpu_dead(unsigned int cpu);
146 
147 /*
148  * Point at some empty memory to start with. We map the real shared_info
149  * page as soon as fixmap is up and running.
150  */
151 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
152 
153 /*
154  * Flag to determine whether vcpu info placement is available on all
155  * VCPUs.  We assume it is to start with, and then set it to zero on
156  * the first failure.  This is because it can succeed on some VCPUs
157  * and not others, since it can involve hypervisor memory allocation,
158  * or because the guest failed to guarantee all the appropriate
159  * constraints on all VCPUs (ie buffer can't cross a page boundary).
160  *
161  * Note that any particular CPU may be using a placed vcpu structure,
162  * but we can only optimise if the all are.
163  *
164  * 0: not available, 1: available
165  */
166 static int have_vcpu_info_placement = 1;
167 
168 struct tls_descs {
169 	struct desc_struct desc[3];
170 };
171 
172 /*
173  * Updating the 3 TLS descriptors in the GDT on every task switch is
174  * surprisingly expensive so we avoid updating them if they haven't
175  * changed.  Since Xen writes different descriptors than the one
176  * passed in the update_descriptor hypercall we keep shadow copies to
177  * compare against.
178  */
179 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
180 
clamp_max_cpus(void)181 static void clamp_max_cpus(void)
182 {
183 #ifdef CONFIG_SMP
184 	if (setup_max_cpus > MAX_VIRT_CPUS)
185 		setup_max_cpus = MAX_VIRT_CPUS;
186 #endif
187 }
188 
xen_vcpu_setup(int cpu)189 void xen_vcpu_setup(int cpu)
190 {
191 	struct vcpu_register_vcpu_info info;
192 	int err;
193 	struct vcpu_info *vcpup;
194 
195 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
196 
197 	/*
198 	 * This path is called twice on PVHVM - first during bootup via
199 	 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
200 	 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
201 	 * As we can only do the VCPUOP_register_vcpu_info once lets
202 	 * not over-write its result.
203 	 *
204 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
205 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
206 	 * use this function.
207 	 */
208 	if (xen_hvm_domain()) {
209 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
210 			return;
211 	}
212 	if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
213 		per_cpu(xen_vcpu, cpu) =
214 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
215 
216 	if (!have_vcpu_info_placement) {
217 		if (cpu >= MAX_VIRT_CPUS)
218 			clamp_max_cpus();
219 		return;
220 	}
221 
222 	vcpup = &per_cpu(xen_vcpu_info, cpu);
223 	info.mfn = arbitrary_virt_to_mfn(vcpup);
224 	info.offset = offset_in_page(vcpup);
225 
226 	/* Check to see if the hypervisor will put the vcpu_info
227 	   structure where we want it, which allows direct access via
228 	   a percpu-variable.
229 	   N.B. This hypercall can _only_ be called once per CPU. Subsequent
230 	   calls will error out with -EINVAL. This is due to the fact that
231 	   hypervisor has no unregister variant and this hypercall does not
232 	   allow to over-write info.mfn and info.offset.
233 	 */
234 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
235 				 &info);
236 
237 	if (err) {
238 		printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
239 		have_vcpu_info_placement = 0;
240 		clamp_max_cpus();
241 	} else {
242 		/* This cpu is using the registered vcpu info, even if
243 		   later ones fail to. */
244 		per_cpu(xen_vcpu, cpu) = vcpup;
245 	}
246 }
247 
248 /*
249  * On restore, set the vcpu placement up again.
250  * If it fails, then we're in a bad state, since
251  * we can't back out from using it...
252  */
xen_vcpu_restore(void)253 void xen_vcpu_restore(void)
254 {
255 	int cpu;
256 
257 	for_each_possible_cpu(cpu) {
258 		bool other_cpu = (cpu != smp_processor_id());
259 		bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
260 						NULL);
261 
262 		if (other_cpu && is_up &&
263 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
264 			BUG();
265 
266 		xen_setup_runstate_info(cpu);
267 
268 		if (have_vcpu_info_placement)
269 			xen_vcpu_setup(cpu);
270 
271 		if (other_cpu && is_up &&
272 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
273 			BUG();
274 	}
275 }
276 
xen_banner(void)277 static void __init xen_banner(void)
278 {
279 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
280 	struct xen_extraversion extra;
281 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
282 
283 	pr_info("Booting paravirtualized kernel %son %s\n",
284 		xen_feature(XENFEAT_auto_translated_physmap) ?
285 			"with PVH extensions " : "", pv_info.name);
286 	printk(KERN_INFO "Xen version: %d.%d%s%s\n",
287 	       version >> 16, version & 0xffff, extra.extraversion,
288 	       xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
289 }
290 /* Check if running on Xen version (major, minor) or later */
291 bool
xen_running_on_version_or_later(unsigned int major,unsigned int minor)292 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
293 {
294 	unsigned int version;
295 
296 	if (!xen_domain())
297 		return false;
298 
299 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
300 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
301 		((version >> 16) > major))
302 		return true;
303 	return false;
304 }
305 
306 #define CPUID_THERM_POWER_LEAF 6
307 #define APERFMPERF_PRESENT 0
308 
309 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
310 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
311 
312 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
313 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
314 static __read_mostly unsigned int cpuid_leaf5_edx_val;
315 
xen_cpuid(unsigned int * ax,unsigned int * bx,unsigned int * cx,unsigned int * dx)316 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
317 		      unsigned int *cx, unsigned int *dx)
318 {
319 	unsigned maskebx = ~0;
320 	unsigned maskecx = ~0;
321 	unsigned maskedx = ~0;
322 	unsigned setecx = 0;
323 	/*
324 	 * Mask out inconvenient features, to try and disable as many
325 	 * unsupported kernel subsystems as possible.
326 	 */
327 	switch (*ax) {
328 	case 1:
329 		maskecx = cpuid_leaf1_ecx_mask;
330 		setecx = cpuid_leaf1_ecx_set_mask;
331 		maskedx = cpuid_leaf1_edx_mask;
332 		break;
333 
334 	case CPUID_MWAIT_LEAF:
335 		/* Synthesize the values.. */
336 		*ax = 0;
337 		*bx = 0;
338 		*cx = cpuid_leaf5_ecx_val;
339 		*dx = cpuid_leaf5_edx_val;
340 		return;
341 
342 	case CPUID_THERM_POWER_LEAF:
343 		/* Disabling APERFMPERF for kernel usage */
344 		maskecx = ~(1 << APERFMPERF_PRESENT);
345 		break;
346 
347 	case 0xb:
348 		/* Suppress extended topology stuff */
349 		maskebx = 0;
350 		break;
351 	}
352 
353 	asm(XEN_EMULATE_PREFIX "cpuid"
354 		: "=a" (*ax),
355 		  "=b" (*bx),
356 		  "=c" (*cx),
357 		  "=d" (*dx)
358 		: "0" (*ax), "2" (*cx));
359 
360 	*bx &= maskebx;
361 	*cx &= maskecx;
362 	*cx |= setecx;
363 	*dx &= maskedx;
364 }
365 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
366 
xen_check_mwait(void)367 static bool __init xen_check_mwait(void)
368 {
369 #ifdef CONFIG_ACPI
370 	struct xen_platform_op op = {
371 		.cmd			= XENPF_set_processor_pminfo,
372 		.u.set_pminfo.id	= -1,
373 		.u.set_pminfo.type	= XEN_PM_PDC,
374 	};
375 	uint32_t buf[3];
376 	unsigned int ax, bx, cx, dx;
377 	unsigned int mwait_mask;
378 
379 	/* We need to determine whether it is OK to expose the MWAIT
380 	 * capability to the kernel to harvest deeper than C3 states from ACPI
381 	 * _CST using the processor_harvest_xen.c module. For this to work, we
382 	 * need to gather the MWAIT_LEAF values (which the cstate.c code
383 	 * checks against). The hypervisor won't expose the MWAIT flag because
384 	 * it would break backwards compatibility; so we will find out directly
385 	 * from the hardware and hypercall.
386 	 */
387 	if (!xen_initial_domain())
388 		return false;
389 
390 	/*
391 	 * When running under platform earlier than Xen4.2, do not expose
392 	 * mwait, to avoid the risk of loading native acpi pad driver
393 	 */
394 	if (!xen_running_on_version_or_later(4, 2))
395 		return false;
396 
397 	ax = 1;
398 	cx = 0;
399 
400 	native_cpuid(&ax, &bx, &cx, &dx);
401 
402 	mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
403 		     (1 << (X86_FEATURE_MWAIT % 32));
404 
405 	if ((cx & mwait_mask) != mwait_mask)
406 		return false;
407 
408 	/* We need to emulate the MWAIT_LEAF and for that we need both
409 	 * ecx and edx. The hypercall provides only partial information.
410 	 */
411 
412 	ax = CPUID_MWAIT_LEAF;
413 	bx = 0;
414 	cx = 0;
415 	dx = 0;
416 
417 	native_cpuid(&ax, &bx, &cx, &dx);
418 
419 	/* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
420 	 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
421 	 */
422 	buf[0] = ACPI_PDC_REVISION_ID;
423 	buf[1] = 1;
424 	buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
425 
426 	set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
427 
428 	if ((HYPERVISOR_platform_op(&op) == 0) &&
429 	    (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
430 		cpuid_leaf5_ecx_val = cx;
431 		cpuid_leaf5_edx_val = dx;
432 	}
433 	return true;
434 #else
435 	return false;
436 #endif
437 }
xen_init_cpuid_mask(void)438 static void __init xen_init_cpuid_mask(void)
439 {
440 	unsigned int ax, bx, cx, dx;
441 	unsigned int xsave_mask;
442 
443 	cpuid_leaf1_edx_mask =
444 		~((1 << X86_FEATURE_MTRR) |  /* disable MTRR */
445 		  (1 << X86_FEATURE_ACC));   /* thermal monitoring */
446 
447 	/*
448 	 * Xen PV would need some work to support PCID: CR3 handling as well
449 	 * as xen_flush_tlb_others() would need updating.
450 	 */
451 	cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_PCID % 32));  /* disable PCID */
452 
453 	if (!xen_initial_domain())
454 		cpuid_leaf1_edx_mask &=
455 			~((1 << X86_FEATURE_ACPI));  /* disable ACPI */
456 
457 	cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
458 
459 	ax = 1;
460 	cx = 0;
461 	cpuid(1, &ax, &bx, &cx, &dx);
462 
463 	xsave_mask =
464 		(1 << (X86_FEATURE_XSAVE % 32)) |
465 		(1 << (X86_FEATURE_OSXSAVE % 32));
466 
467 	/* Xen will set CR4.OSXSAVE if supported and not disabled by force */
468 	if ((cx & xsave_mask) != xsave_mask)
469 		cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
470 	if (xen_check_mwait())
471 		cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
472 }
473 
xen_set_debugreg(int reg,unsigned long val)474 static void xen_set_debugreg(int reg, unsigned long val)
475 {
476 	HYPERVISOR_set_debugreg(reg, val);
477 }
478 
xen_get_debugreg(int reg)479 static unsigned long xen_get_debugreg(int reg)
480 {
481 	return HYPERVISOR_get_debugreg(reg);
482 }
483 
xen_end_context_switch(struct task_struct * next)484 static void xen_end_context_switch(struct task_struct *next)
485 {
486 	xen_mc_flush();
487 	paravirt_end_context_switch(next);
488 }
489 
xen_store_tr(void)490 static unsigned long xen_store_tr(void)
491 {
492 	return 0;
493 }
494 
495 /*
496  * Set the page permissions for a particular virtual address.  If the
497  * address is a vmalloc mapping (or other non-linear mapping), then
498  * find the linear mapping of the page and also set its protections to
499  * match.
500  */
set_aliased_prot(void * v,pgprot_t prot)501 static void set_aliased_prot(void *v, pgprot_t prot)
502 {
503 	int level;
504 	pte_t *ptep;
505 	pte_t pte;
506 	unsigned long pfn;
507 	struct page *page;
508 	unsigned char dummy;
509 
510 	ptep = lookup_address((unsigned long)v, &level);
511 	BUG_ON(ptep == NULL);
512 
513 	pfn = pte_pfn(*ptep);
514 	page = pfn_to_page(pfn);
515 
516 	pte = pfn_pte(pfn, prot);
517 
518 	/*
519 	 * Careful: update_va_mapping() will fail if the virtual address
520 	 * we're poking isn't populated in the page tables.  We don't
521 	 * need to worry about the direct map (that's always in the page
522 	 * tables), but we need to be careful about vmap space.  In
523 	 * particular, the top level page table can lazily propagate
524 	 * entries between processes, so if we've switched mms since we
525 	 * vmapped the target in the first place, we might not have the
526 	 * top-level page table entry populated.
527 	 *
528 	 * We disable preemption because we want the same mm active when
529 	 * we probe the target and when we issue the hypercall.  We'll
530 	 * have the same nominal mm, but if we're a kernel thread, lazy
531 	 * mm dropping could change our pgd.
532 	 *
533 	 * Out of an abundance of caution, this uses __get_user() to fault
534 	 * in the target address just in case there's some obscure case
535 	 * in which the target address isn't readable.
536 	 */
537 
538 	preempt_disable();
539 
540 	probe_kernel_read(&dummy, v, 1);
541 
542 	if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
543 		BUG();
544 
545 	if (!PageHighMem(page)) {
546 		void *av = __va(PFN_PHYS(pfn));
547 
548 		if (av != v)
549 			if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
550 				BUG();
551 	} else
552 		kmap_flush_unused();
553 
554 	preempt_enable();
555 }
556 
xen_alloc_ldt(struct desc_struct * ldt,unsigned entries)557 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
558 {
559 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
560 	int i;
561 
562 	/*
563 	 * We need to mark the all aliases of the LDT pages RO.  We
564 	 * don't need to call vm_flush_aliases(), though, since that's
565 	 * only responsible for flushing aliases out the TLBs, not the
566 	 * page tables, and Xen will flush the TLB for us if needed.
567 	 *
568 	 * To avoid confusing future readers: none of this is necessary
569 	 * to load the LDT.  The hypervisor only checks this when the
570 	 * LDT is faulted in due to subsequent descriptor access.
571 	 */
572 
573 	for(i = 0; i < entries; i += entries_per_page)
574 		set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
575 }
576 
xen_free_ldt(struct desc_struct * ldt,unsigned entries)577 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
578 {
579 	const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
580 	int i;
581 
582 	for(i = 0; i < entries; i += entries_per_page)
583 		set_aliased_prot(ldt + i, PAGE_KERNEL);
584 }
585 
xen_set_ldt(const void * addr,unsigned entries)586 static void xen_set_ldt(const void *addr, unsigned entries)
587 {
588 	struct mmuext_op *op;
589 	struct multicall_space mcs = xen_mc_entry(sizeof(*op));
590 
591 	trace_xen_cpu_set_ldt(addr, entries);
592 
593 	op = mcs.args;
594 	op->cmd = MMUEXT_SET_LDT;
595 	op->arg1.linear_addr = (unsigned long)addr;
596 	op->arg2.nr_ents = entries;
597 
598 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
599 
600 	xen_mc_issue(PARAVIRT_LAZY_CPU);
601 }
602 
xen_load_gdt(const struct desc_ptr * dtr)603 static void xen_load_gdt(const struct desc_ptr *dtr)
604 {
605 	unsigned long va = dtr->address;
606 	unsigned int size = dtr->size + 1;
607 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
608 	unsigned long frames[pages];
609 	int f;
610 
611 	/*
612 	 * A GDT can be up to 64k in size, which corresponds to 8192
613 	 * 8-byte entries, or 16 4k pages..
614 	 */
615 
616 	BUG_ON(size > 65536);
617 	BUG_ON(va & ~PAGE_MASK);
618 
619 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
620 		int level;
621 		pte_t *ptep;
622 		unsigned long pfn, mfn;
623 		void *virt;
624 
625 		/*
626 		 * The GDT is per-cpu and is in the percpu data area.
627 		 * That can be virtually mapped, so we need to do a
628 		 * page-walk to get the underlying MFN for the
629 		 * hypercall.  The page can also be in the kernel's
630 		 * linear range, so we need to RO that mapping too.
631 		 */
632 		ptep = lookup_address(va, &level);
633 		BUG_ON(ptep == NULL);
634 
635 		pfn = pte_pfn(*ptep);
636 		mfn = pfn_to_mfn(pfn);
637 		virt = __va(PFN_PHYS(pfn));
638 
639 		frames[f] = mfn;
640 
641 		make_lowmem_page_readonly((void *)va);
642 		make_lowmem_page_readonly(virt);
643 	}
644 
645 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
646 		BUG();
647 }
648 
649 /*
650  * load_gdt for early boot, when the gdt is only mapped once
651  */
xen_load_gdt_boot(const struct desc_ptr * dtr)652 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
653 {
654 	unsigned long va = dtr->address;
655 	unsigned int size = dtr->size + 1;
656 	unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
657 	unsigned long frames[pages];
658 	int f;
659 
660 	/*
661 	 * A GDT can be up to 64k in size, which corresponds to 8192
662 	 * 8-byte entries, or 16 4k pages..
663 	 */
664 
665 	BUG_ON(size > 65536);
666 	BUG_ON(va & ~PAGE_MASK);
667 
668 	for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
669 		pte_t pte;
670 		unsigned long pfn, mfn;
671 
672 		pfn = virt_to_pfn(va);
673 		mfn = pfn_to_mfn(pfn);
674 
675 		pte = pfn_pte(pfn, PAGE_KERNEL_RO);
676 
677 		if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
678 			BUG();
679 
680 		frames[f] = mfn;
681 	}
682 
683 	if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
684 		BUG();
685 }
686 
desc_equal(const struct desc_struct * d1,const struct desc_struct * d2)687 static inline bool desc_equal(const struct desc_struct *d1,
688 			      const struct desc_struct *d2)
689 {
690 	return d1->a == d2->a && d1->b == d2->b;
691 }
692 
load_TLS_descriptor(struct thread_struct * t,unsigned int cpu,unsigned int i)693 static void load_TLS_descriptor(struct thread_struct *t,
694 				unsigned int cpu, unsigned int i)
695 {
696 	struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
697 	struct desc_struct *gdt;
698 	xmaddr_t maddr;
699 	struct multicall_space mc;
700 
701 	if (desc_equal(shadow, &t->tls_array[i]))
702 		return;
703 
704 	*shadow = t->tls_array[i];
705 
706 	gdt = get_cpu_gdt_table(cpu);
707 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
708 	mc = __xen_mc_entry(0);
709 
710 	MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
711 }
712 
xen_load_tls(struct thread_struct * t,unsigned int cpu)713 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
714 {
715 	/*
716 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone
717 	 * and lazy gs handling is enabled, it means we're in a
718 	 * context switch, and %gs has just been saved.  This means we
719 	 * can zero it out to prevent faults on exit from the
720 	 * hypervisor if the next process has no %gs.  Either way, it
721 	 * has been saved, and the new value will get loaded properly.
722 	 * This will go away as soon as Xen has been modified to not
723 	 * save/restore %gs for normal hypercalls.
724 	 *
725 	 * On x86_64, this hack is not used for %gs, because gs points
726 	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
727 	 * must not zero %gs on x86_64
728 	 *
729 	 * For x86_64, we need to zero %fs, otherwise we may get an
730 	 * exception between the new %fs descriptor being loaded and
731 	 * %fs being effectively cleared at __switch_to().
732 	 */
733 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
734 #ifdef CONFIG_X86_32
735 		lazy_load_gs(0);
736 #else
737 		loadsegment(fs, 0);
738 #endif
739 	}
740 
741 	xen_mc_batch();
742 
743 	load_TLS_descriptor(t, cpu, 0);
744 	load_TLS_descriptor(t, cpu, 1);
745 	load_TLS_descriptor(t, cpu, 2);
746 
747 	xen_mc_issue(PARAVIRT_LAZY_CPU);
748 }
749 
750 #ifdef CONFIG_X86_64
xen_load_gs_index(unsigned int idx)751 static void xen_load_gs_index(unsigned int idx)
752 {
753 	if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
754 		BUG();
755 }
756 #endif
757 
xen_write_ldt_entry(struct desc_struct * dt,int entrynum,const void * ptr)758 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
759 				const void *ptr)
760 {
761 	xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
762 	u64 entry = *(u64 *)ptr;
763 
764 	trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
765 
766 	preempt_disable();
767 
768 	xen_mc_flush();
769 	if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
770 		BUG();
771 
772 	preempt_enable();
773 }
774 
cvt_gate_to_trap(int vector,const gate_desc * val,struct trap_info * info)775 static int cvt_gate_to_trap(int vector, const gate_desc *val,
776 			    struct trap_info *info)
777 {
778 	unsigned long addr;
779 
780 	if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
781 		return 0;
782 
783 	info->vector = vector;
784 
785 	addr = gate_offset(*val);
786 #ifdef CONFIG_X86_64
787 	/*
788 	 * Look for known traps using IST, and substitute them
789 	 * appropriately.  The debugger ones are the only ones we care
790 	 * about.  Xen will handle faults like double_fault,
791 	 * so we should never see them.  Warn if
792 	 * there's an unexpected IST-using fault handler.
793 	 */
794 	if (addr == (unsigned long)debug)
795 		addr = (unsigned long)xen_debug;
796 	else if (addr == (unsigned long)int3)
797 		addr = (unsigned long)xen_int3;
798 	else if (addr == (unsigned long)stack_segment)
799 		addr = (unsigned long)xen_stack_segment;
800 	else if (addr == (unsigned long)double_fault) {
801 		/* Don't need to handle these */
802 		return 0;
803 #ifdef CONFIG_X86_MCE
804 	} else if (addr == (unsigned long)machine_check) {
805 		/*
806 		 * when xen hypervisor inject vMCE to guest,
807 		 * use native mce handler to handle it
808 		 */
809 		;
810 #endif
811 	} else if (addr == (unsigned long)nmi)
812 		/*
813 		 * Use the native version as well.
814 		 */
815 		;
816 	else {
817 		/* Some other trap using IST? */
818 		if (WARN_ON(val->ist != 0))
819 			return 0;
820 	}
821 #endif	/* CONFIG_X86_64 */
822 	info->address = addr;
823 
824 	info->cs = gate_segment(*val);
825 	info->flags = val->dpl;
826 	/* interrupt gates clear IF */
827 	if (val->type == GATE_INTERRUPT)
828 		info->flags |= 1 << 2;
829 
830 	return 1;
831 }
832 
833 /* Locations of each CPU's IDT */
834 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
835 
836 /* Set an IDT entry.  If the entry is part of the current IDT, then
837    also update Xen. */
xen_write_idt_entry(gate_desc * dt,int entrynum,const gate_desc * g)838 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
839 {
840 	unsigned long p = (unsigned long)&dt[entrynum];
841 	unsigned long start, end;
842 
843 	trace_xen_cpu_write_idt_entry(dt, entrynum, g);
844 
845 	preempt_disable();
846 
847 	start = __this_cpu_read(idt_desc.address);
848 	end = start + __this_cpu_read(idt_desc.size) + 1;
849 
850 	xen_mc_flush();
851 
852 	native_write_idt_entry(dt, entrynum, g);
853 
854 	if (p >= start && (p + 8) <= end) {
855 		struct trap_info info[2];
856 
857 		info[1].address = 0;
858 
859 		if (cvt_gate_to_trap(entrynum, g, &info[0]))
860 			if (HYPERVISOR_set_trap_table(info))
861 				BUG();
862 	}
863 
864 	preempt_enable();
865 }
866 
xen_convert_trap_info(const struct desc_ptr * desc,struct trap_info * traps)867 static void xen_convert_trap_info(const struct desc_ptr *desc,
868 				  struct trap_info *traps)
869 {
870 	unsigned in, out, count;
871 
872 	count = (desc->size+1) / sizeof(gate_desc);
873 	BUG_ON(count > 256);
874 
875 	for (in = out = 0; in < count; in++) {
876 		gate_desc *entry = (gate_desc*)(desc->address) + in;
877 
878 		if (cvt_gate_to_trap(in, entry, &traps[out]))
879 			out++;
880 	}
881 	traps[out].address = 0;
882 }
883 
xen_copy_trap_info(struct trap_info * traps)884 void xen_copy_trap_info(struct trap_info *traps)
885 {
886 	const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
887 
888 	xen_convert_trap_info(desc, traps);
889 }
890 
891 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
892    hold a spinlock to protect the static traps[] array (static because
893    it avoids allocation, and saves stack space). */
xen_load_idt(const struct desc_ptr * desc)894 static void xen_load_idt(const struct desc_ptr *desc)
895 {
896 	static DEFINE_SPINLOCK(lock);
897 	static struct trap_info traps[257];
898 
899 	trace_xen_cpu_load_idt(desc);
900 
901 	spin_lock(&lock);
902 
903 	memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
904 
905 	xen_convert_trap_info(desc, traps);
906 
907 	xen_mc_flush();
908 	if (HYPERVISOR_set_trap_table(traps))
909 		BUG();
910 
911 	spin_unlock(&lock);
912 }
913 
914 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
915    they're handled differently. */
xen_write_gdt_entry(struct desc_struct * dt,int entry,const void * desc,int type)916 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
917 				const void *desc, int type)
918 {
919 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
920 
921 	preempt_disable();
922 
923 	switch (type) {
924 	case DESC_LDT:
925 	case DESC_TSS:
926 		/* ignore */
927 		break;
928 
929 	default: {
930 		xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
931 
932 		xen_mc_flush();
933 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
934 			BUG();
935 	}
936 
937 	}
938 
939 	preempt_enable();
940 }
941 
942 /*
943  * Version of write_gdt_entry for use at early boot-time needed to
944  * update an entry as simply as possible.
945  */
xen_write_gdt_entry_boot(struct desc_struct * dt,int entry,const void * desc,int type)946 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
947 					    const void *desc, int type)
948 {
949 	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
950 
951 	switch (type) {
952 	case DESC_LDT:
953 	case DESC_TSS:
954 		/* ignore */
955 		break;
956 
957 	default: {
958 		xmaddr_t maddr = virt_to_machine(&dt[entry]);
959 
960 		if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
961 			dt[entry] = *(struct desc_struct *)desc;
962 	}
963 
964 	}
965 }
966 
xen_load_sp0(struct tss_struct * tss,struct thread_struct * thread)967 static void xen_load_sp0(struct tss_struct *tss,
968 			 struct thread_struct *thread)
969 {
970 	struct multicall_space mcs;
971 
972 	mcs = xen_mc_entry(0);
973 	MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
974 	xen_mc_issue(PARAVIRT_LAZY_CPU);
975 	tss->x86_tss.sp0 = thread->sp0;
976 }
977 
xen_set_iopl_mask(unsigned mask)978 void xen_set_iopl_mask(unsigned mask)
979 {
980 	struct physdev_set_iopl set_iopl;
981 
982 	/* Force the change at ring 0. */
983 	set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
984 	HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
985 }
986 
xen_io_delay(void)987 static void xen_io_delay(void)
988 {
989 }
990 
xen_clts(void)991 static void xen_clts(void)
992 {
993 	struct multicall_space mcs;
994 
995 	mcs = xen_mc_entry(0);
996 
997 	MULTI_fpu_taskswitch(mcs.mc, 0);
998 
999 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1000 }
1001 
1002 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
1003 
xen_read_cr0(void)1004 static unsigned long xen_read_cr0(void)
1005 {
1006 	unsigned long cr0 = this_cpu_read(xen_cr0_value);
1007 
1008 	if (unlikely(cr0 == 0)) {
1009 		cr0 = native_read_cr0();
1010 		this_cpu_write(xen_cr0_value, cr0);
1011 	}
1012 
1013 	return cr0;
1014 }
1015 
xen_write_cr0(unsigned long cr0)1016 static void xen_write_cr0(unsigned long cr0)
1017 {
1018 	struct multicall_space mcs;
1019 
1020 	this_cpu_write(xen_cr0_value, cr0);
1021 
1022 	/* Only pay attention to cr0.TS; everything else is
1023 	   ignored. */
1024 	mcs = xen_mc_entry(0);
1025 
1026 	MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1027 
1028 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1029 }
1030 
xen_write_cr4(unsigned long cr4)1031 static void xen_write_cr4(unsigned long cr4)
1032 {
1033 	cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1034 
1035 	native_write_cr4(cr4);
1036 }
1037 #ifdef CONFIG_X86_64
xen_read_cr8(void)1038 static inline unsigned long xen_read_cr8(void)
1039 {
1040 	return 0;
1041 }
xen_write_cr8(unsigned long val)1042 static inline void xen_write_cr8(unsigned long val)
1043 {
1044 	BUG_ON(val);
1045 }
1046 #endif
1047 
xen_read_msr_safe(unsigned int msr,int * err)1048 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1049 {
1050 	u64 val;
1051 
1052 	if (pmu_msr_read(msr, &val, err))
1053 		return val;
1054 
1055 	val = native_read_msr_safe(msr, err);
1056 	switch (msr) {
1057 	case MSR_IA32_APICBASE:
1058 #ifdef CONFIG_X86_X2APIC
1059 		if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1060 #endif
1061 			val &= ~X2APIC_ENABLE;
1062 		break;
1063 	}
1064 	return val;
1065 }
1066 
xen_write_msr_safe(unsigned int msr,unsigned low,unsigned high)1067 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1068 {
1069 	int ret;
1070 
1071 	ret = 0;
1072 
1073 	switch (msr) {
1074 #ifdef CONFIG_X86_64
1075 		unsigned which;
1076 		u64 base;
1077 
1078 	case MSR_FS_BASE:		which = SEGBASE_FS; goto set;
1079 	case MSR_KERNEL_GS_BASE:	which = SEGBASE_GS_USER; goto set;
1080 	case MSR_GS_BASE:		which = SEGBASE_GS_KERNEL; goto set;
1081 
1082 	set:
1083 		base = ((u64)high << 32) | low;
1084 		if (HYPERVISOR_set_segment_base(which, base) != 0)
1085 			ret = -EIO;
1086 		break;
1087 #endif
1088 
1089 	case MSR_STAR:
1090 	case MSR_CSTAR:
1091 	case MSR_LSTAR:
1092 	case MSR_SYSCALL_MASK:
1093 	case MSR_IA32_SYSENTER_CS:
1094 	case MSR_IA32_SYSENTER_ESP:
1095 	case MSR_IA32_SYSENTER_EIP:
1096 		/* Fast syscall setup is all done in hypercalls, so
1097 		   these are all ignored.  Stub them out here to stop
1098 		   Xen console noise. */
1099 		break;
1100 
1101 	default:
1102 		if (!pmu_msr_write(msr, low, high, &ret))
1103 			ret = native_write_msr_safe(msr, low, high);
1104 	}
1105 
1106 	return ret;
1107 }
1108 
xen_read_msr(unsigned int msr)1109 static u64 xen_read_msr(unsigned int msr)
1110 {
1111 	/*
1112 	 * This will silently swallow a #GP from RDMSR.  It may be worth
1113 	 * changing that.
1114 	 */
1115 	int err;
1116 
1117 	return xen_read_msr_safe(msr, &err);
1118 }
1119 
xen_write_msr(unsigned int msr,unsigned low,unsigned high)1120 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1121 {
1122 	/*
1123 	 * This will silently swallow a #GP from WRMSR.  It may be worth
1124 	 * changing that.
1125 	 */
1126 	xen_write_msr_safe(msr, low, high);
1127 }
1128 
xen_setup_shared_info(void)1129 void xen_setup_shared_info(void)
1130 {
1131 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1132 		set_fixmap(FIX_PARAVIRT_BOOTMAP,
1133 			   xen_start_info->shared_info);
1134 
1135 		HYPERVISOR_shared_info =
1136 			(struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1137 	} else
1138 		HYPERVISOR_shared_info =
1139 			(struct shared_info *)__va(xen_start_info->shared_info);
1140 
1141 #ifndef CONFIG_SMP
1142 	/* In UP this is as good a place as any to set up shared info */
1143 	xen_setup_vcpu_info_placement();
1144 #endif
1145 
1146 	xen_setup_mfn_list_list();
1147 }
1148 
1149 /* This is called once we have the cpu_possible_mask */
xen_setup_vcpu_info_placement(void)1150 void xen_setup_vcpu_info_placement(void)
1151 {
1152 	int cpu;
1153 
1154 	for_each_possible_cpu(cpu) {
1155 		/* Set up direct vCPU id mapping for PV guests. */
1156 		per_cpu(xen_vcpu_id, cpu) = cpu;
1157 		xen_vcpu_setup(cpu);
1158 	}
1159 
1160 	/* xen_vcpu_setup managed to place the vcpu_info within the
1161 	 * percpu area for all cpus, so make use of it. Note that for
1162 	 * PVH we want to use native IRQ mechanism. */
1163 	if (have_vcpu_info_placement && !xen_pvh_domain()) {
1164 		pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1165 		pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1166 		pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1167 		pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1168 		pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1169 	}
1170 }
1171 
xen_patch(u8 type,u16 clobbers,void * insnbuf,unsigned long addr,unsigned len)1172 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1173 			  unsigned long addr, unsigned len)
1174 {
1175 	char *start, *end, *reloc;
1176 	unsigned ret;
1177 
1178 	start = end = reloc = NULL;
1179 
1180 #define SITE(op, x)							\
1181 	case PARAVIRT_PATCH(op.x):					\
1182 	if (have_vcpu_info_placement) {					\
1183 		start = (char *)xen_##x##_direct;			\
1184 		end = xen_##x##_direct_end;				\
1185 		reloc = xen_##x##_direct_reloc;				\
1186 	}								\
1187 	goto patch_site
1188 
1189 	switch (type) {
1190 		SITE(pv_irq_ops, irq_enable);
1191 		SITE(pv_irq_ops, irq_disable);
1192 		SITE(pv_irq_ops, save_fl);
1193 		SITE(pv_irq_ops, restore_fl);
1194 #undef SITE
1195 
1196 	patch_site:
1197 		if (start == NULL || (end-start) > len)
1198 			goto default_patch;
1199 
1200 		ret = paravirt_patch_insns(insnbuf, len, start, end);
1201 
1202 		/* Note: because reloc is assigned from something that
1203 		   appears to be an array, gcc assumes it's non-null,
1204 		   but doesn't know its relationship with start and
1205 		   end. */
1206 		if (reloc > start && reloc < end) {
1207 			int reloc_off = reloc - start;
1208 			long *relocp = (long *)(insnbuf + reloc_off);
1209 			long delta = start - (char *)addr;
1210 
1211 			*relocp += delta;
1212 		}
1213 		break;
1214 
1215 	default_patch:
1216 	default:
1217 		ret = paravirt_patch_default(type, clobbers, insnbuf,
1218 					     addr, len);
1219 		break;
1220 	}
1221 
1222 	return ret;
1223 }
1224 
1225 static const struct pv_info xen_info __initconst = {
1226 	.shared_kernel_pmd = 0,
1227 
1228 #ifdef CONFIG_X86_64
1229 	.extra_user_64bit_cs = FLAT_USER_CS64,
1230 #endif
1231 	.name = "Xen",
1232 };
1233 
1234 static const struct pv_init_ops xen_init_ops __initconst = {
1235 	.patch = xen_patch,
1236 };
1237 
1238 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1239 	.cpuid = xen_cpuid,
1240 
1241 	.set_debugreg = xen_set_debugreg,
1242 	.get_debugreg = xen_get_debugreg,
1243 
1244 	.clts = xen_clts,
1245 
1246 	.read_cr0 = xen_read_cr0,
1247 	.write_cr0 = xen_write_cr0,
1248 
1249 	.read_cr4 = native_read_cr4,
1250 	.write_cr4 = xen_write_cr4,
1251 
1252 #ifdef CONFIG_X86_64
1253 	.read_cr8 = xen_read_cr8,
1254 	.write_cr8 = xen_write_cr8,
1255 #endif
1256 
1257 	.wbinvd = native_wbinvd,
1258 
1259 	.read_msr = xen_read_msr,
1260 	.write_msr = xen_write_msr,
1261 
1262 	.read_msr_safe = xen_read_msr_safe,
1263 	.write_msr_safe = xen_write_msr_safe,
1264 
1265 	.read_pmc = xen_read_pmc,
1266 
1267 	.iret = xen_iret,
1268 #ifdef CONFIG_X86_64
1269 	.usergs_sysret64 = xen_sysret64,
1270 #endif
1271 
1272 	.load_tr_desc = paravirt_nop,
1273 	.set_ldt = xen_set_ldt,
1274 	.load_gdt = xen_load_gdt,
1275 	.load_idt = xen_load_idt,
1276 	.load_tls = xen_load_tls,
1277 #ifdef CONFIG_X86_64
1278 	.load_gs_index = xen_load_gs_index,
1279 #endif
1280 
1281 	.alloc_ldt = xen_alloc_ldt,
1282 	.free_ldt = xen_free_ldt,
1283 
1284 	.store_idt = native_store_idt,
1285 	.store_tr = xen_store_tr,
1286 
1287 	.write_ldt_entry = xen_write_ldt_entry,
1288 	.write_gdt_entry = xen_write_gdt_entry,
1289 	.write_idt_entry = xen_write_idt_entry,
1290 	.load_sp0 = xen_load_sp0,
1291 
1292 	.set_iopl_mask = xen_set_iopl_mask,
1293 	.io_delay = xen_io_delay,
1294 
1295 	/* Xen takes care of %gs when switching to usermode for us */
1296 	.swapgs = paravirt_nop,
1297 
1298 	.start_context_switch = paravirt_start_context_switch,
1299 	.end_context_switch = xen_end_context_switch,
1300 };
1301 
xen_reboot(int reason)1302 static void xen_reboot(int reason)
1303 {
1304 	struct sched_shutdown r = { .reason = reason };
1305 	int cpu;
1306 
1307 	for_each_online_cpu(cpu)
1308 		xen_pmu_finish(cpu);
1309 
1310 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1311 		BUG();
1312 }
1313 
xen_restart(char * msg)1314 static void xen_restart(char *msg)
1315 {
1316 	xen_reboot(SHUTDOWN_reboot);
1317 }
1318 
xen_emergency_restart(void)1319 static void xen_emergency_restart(void)
1320 {
1321 	xen_reboot(SHUTDOWN_reboot);
1322 }
1323 
xen_machine_halt(void)1324 static void xen_machine_halt(void)
1325 {
1326 	xen_reboot(SHUTDOWN_poweroff);
1327 }
1328 
xen_machine_power_off(void)1329 static void xen_machine_power_off(void)
1330 {
1331 	if (pm_power_off)
1332 		pm_power_off();
1333 	xen_reboot(SHUTDOWN_poweroff);
1334 }
1335 
xen_crash_shutdown(struct pt_regs * regs)1336 static void xen_crash_shutdown(struct pt_regs *regs)
1337 {
1338 	xen_reboot(SHUTDOWN_crash);
1339 }
1340 
1341 static int
xen_panic_event(struct notifier_block * this,unsigned long event,void * ptr)1342 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1343 {
1344 	if (!kexec_crash_loaded())
1345 		xen_reboot(SHUTDOWN_crash);
1346 	return NOTIFY_DONE;
1347 }
1348 
1349 static struct notifier_block xen_panic_block = {
1350 	.notifier_call= xen_panic_event,
1351 	.priority = INT_MIN
1352 };
1353 
xen_panic_handler_init(void)1354 int xen_panic_handler_init(void)
1355 {
1356 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1357 	return 0;
1358 }
1359 
1360 static const struct machine_ops xen_machine_ops __initconst = {
1361 	.restart = xen_restart,
1362 	.halt = xen_machine_halt,
1363 	.power_off = xen_machine_power_off,
1364 	.shutdown = xen_machine_halt,
1365 	.crash_shutdown = xen_crash_shutdown,
1366 	.emergency_restart = xen_emergency_restart,
1367 };
1368 
xen_get_nmi_reason(void)1369 static unsigned char xen_get_nmi_reason(void)
1370 {
1371 	unsigned char reason = 0;
1372 
1373 	/* Construct a value which looks like it came from port 0x61. */
1374 	if (test_bit(_XEN_NMIREASON_io_error,
1375 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1376 		reason |= NMI_REASON_IOCHK;
1377 	if (test_bit(_XEN_NMIREASON_pci_serr,
1378 		     &HYPERVISOR_shared_info->arch.nmi_reason))
1379 		reason |= NMI_REASON_SERR;
1380 
1381 	return reason;
1382 }
1383 
xen_boot_params_init_edd(void)1384 static void __init xen_boot_params_init_edd(void)
1385 {
1386 #if IS_ENABLED(CONFIG_EDD)
1387 	struct xen_platform_op op;
1388 	struct edd_info *edd_info;
1389 	u32 *mbr_signature;
1390 	unsigned nr;
1391 	int ret;
1392 
1393 	edd_info = boot_params.eddbuf;
1394 	mbr_signature = boot_params.edd_mbr_sig_buffer;
1395 
1396 	op.cmd = XENPF_firmware_info;
1397 
1398 	op.u.firmware_info.type = XEN_FW_DISK_INFO;
1399 	for (nr = 0; nr < EDDMAXNR; nr++) {
1400 		struct edd_info *info = edd_info + nr;
1401 
1402 		op.u.firmware_info.index = nr;
1403 		info->params.length = sizeof(info->params);
1404 		set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1405 				     &info->params);
1406 		ret = HYPERVISOR_platform_op(&op);
1407 		if (ret)
1408 			break;
1409 
1410 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1411 		C(device);
1412 		C(version);
1413 		C(interface_support);
1414 		C(legacy_max_cylinder);
1415 		C(legacy_max_head);
1416 		C(legacy_sectors_per_track);
1417 #undef C
1418 	}
1419 	boot_params.eddbuf_entries = nr;
1420 
1421 	op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1422 	for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1423 		op.u.firmware_info.index = nr;
1424 		ret = HYPERVISOR_platform_op(&op);
1425 		if (ret)
1426 			break;
1427 		mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1428 	}
1429 	boot_params.edd_mbr_sig_buf_entries = nr;
1430 #endif
1431 }
1432 
1433 /*
1434  * Set up the GDT and segment registers for -fstack-protector.  Until
1435  * we do this, we have to be careful not to call any stack-protected
1436  * function, which is most of the kernel.
1437  *
1438  * Note, that it is __ref because the only caller of this after init
1439  * is PVH which is not going to use xen_load_gdt_boot or other
1440  * __init functions.
1441  */
xen_setup_gdt(int cpu)1442 static void __ref xen_setup_gdt(int cpu)
1443 {
1444 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
1445 #ifdef CONFIG_X86_64
1446 		unsigned long dummy;
1447 
1448 		load_percpu_segment(cpu); /* We need to access per-cpu area */
1449 		switch_to_new_gdt(cpu); /* GDT and GS set */
1450 
1451 		/* We are switching of the Xen provided GDT to our HVM mode
1452 		 * GDT. The new GDT has  __KERNEL_CS with CS.L = 1
1453 		 * and we are jumping to reload it.
1454 		 */
1455 		asm volatile ("pushq %0\n"
1456 			      "leaq 1f(%%rip),%0\n"
1457 			      "pushq %0\n"
1458 			      "lretq\n"
1459 			      "1:\n"
1460 			      : "=&r" (dummy) : "0" (__KERNEL_CS));
1461 
1462 		/*
1463 		 * While not needed, we also set the %es, %ds, and %fs
1464 		 * to zero. We don't care about %ss as it is NULL.
1465 		 * Strictly speaking this is not needed as Xen zeros those
1466 		 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1467 		 *
1468 		 * Linux zeros them in cpu_init() and in secondary_startup_64
1469 		 * (for BSP).
1470 		 */
1471 		loadsegment(es, 0);
1472 		loadsegment(ds, 0);
1473 		loadsegment(fs, 0);
1474 #else
1475 		/* PVH: TODO Implement. */
1476 		BUG();
1477 #endif
1478 		return; /* PVH does not need any PV GDT ops. */
1479 	}
1480 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1481 	pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1482 
1483 	setup_stack_canary_segment(0);
1484 	switch_to_new_gdt(0);
1485 
1486 	pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1487 	pv_cpu_ops.load_gdt = xen_load_gdt;
1488 }
1489 
1490 #ifdef CONFIG_XEN_PVH
1491 /*
1492  * A PV guest starts with default flags that are not set for PVH, set them
1493  * here asap.
1494  */
xen_pvh_set_cr_flags(int cpu)1495 static void xen_pvh_set_cr_flags(int cpu)
1496 {
1497 
1498 	/* Some of these are setup in 'secondary_startup_64'. The others:
1499 	 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1500 	 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1501 	write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
1502 
1503 	if (!cpu)
1504 		return;
1505 	/*
1506 	 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1507 	 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
1508 	*/
1509 	if (boot_cpu_has(X86_FEATURE_PSE))
1510 		cr4_set_bits_and_update_boot(X86_CR4_PSE);
1511 
1512 	if (boot_cpu_has(X86_FEATURE_PGE))
1513 		cr4_set_bits_and_update_boot(X86_CR4_PGE);
1514 }
1515 
1516 /*
1517  * Note, that it is ref - because the only caller of this after init
1518  * is PVH which is not going to use xen_load_gdt_boot or other
1519  * __init functions.
1520  */
xen_pvh_secondary_vcpu_init(int cpu)1521 void __ref xen_pvh_secondary_vcpu_init(int cpu)
1522 {
1523 	xen_setup_gdt(cpu);
1524 	xen_pvh_set_cr_flags(cpu);
1525 }
1526 
xen_pvh_early_guest_init(void)1527 static void __init xen_pvh_early_guest_init(void)
1528 {
1529 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1530 		return;
1531 
1532 	if (!xen_feature(XENFEAT_hvm_callback_vector))
1533 		return;
1534 
1535 	xen_have_vector_callback = 1;
1536 
1537 	xen_pvh_early_cpu_init(0, false);
1538 	xen_pvh_set_cr_flags(0);
1539 
1540 #ifdef CONFIG_X86_32
1541 	BUG(); /* PVH: Implement proper support. */
1542 #endif
1543 }
1544 #endif    /* CONFIG_XEN_PVH */
1545 
xen_dom0_set_legacy_features(void)1546 static void __init xen_dom0_set_legacy_features(void)
1547 {
1548 	x86_platform.legacy.rtc = 1;
1549 }
1550 
xen_cpuhp_setup(void)1551 static int xen_cpuhp_setup(void)
1552 {
1553 	int rc;
1554 
1555 	rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
1556 				       "XEN_HVM_GUEST_PREPARE",
1557 				       xen_cpu_up_prepare, xen_cpu_dead);
1558 	if (rc >= 0) {
1559 		rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
1560 					       "XEN_HVM_GUEST_ONLINE",
1561 					       xen_cpu_up_online, NULL);
1562 		if (rc < 0)
1563 			cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
1564 	}
1565 
1566 	return rc >= 0 ? 0 : rc;
1567 }
1568 
1569 /* First C function to be called on Xen boot */
xen_start_kernel(void)1570 asmlinkage __visible void __init xen_start_kernel(void)
1571 {
1572 	struct physdev_set_iopl set_iopl;
1573 	unsigned long initrd_start = 0;
1574 	int rc;
1575 
1576 	if (!xen_start_info)
1577 		return;
1578 
1579 	xen_domain_type = XEN_PV_DOMAIN;
1580 
1581 	xen_setup_features();
1582 #ifdef CONFIG_XEN_PVH
1583 	xen_pvh_early_guest_init();
1584 #endif
1585 	xen_setup_machphys_mapping();
1586 
1587 	/* Install Xen paravirt ops */
1588 	pv_info = xen_info;
1589 	pv_init_ops = xen_init_ops;
1590 	if (!xen_pvh_domain()) {
1591 		pv_cpu_ops = xen_cpu_ops;
1592 
1593 		x86_platform.get_nmi_reason = xen_get_nmi_reason;
1594 	}
1595 
1596 	if (xen_feature(XENFEAT_auto_translated_physmap))
1597 		x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1598 	else
1599 		x86_init.resources.memory_setup = xen_memory_setup;
1600 	x86_init.oem.arch_setup = xen_arch_setup;
1601 	x86_init.oem.banner = xen_banner;
1602 
1603 	xen_init_time_ops();
1604 
1605 	/*
1606 	 * Set up some pagetable state before starting to set any ptes.
1607 	 */
1608 
1609 	xen_init_mmu_ops();
1610 
1611 	/* Prevent unwanted bits from being set in PTEs. */
1612 	__supported_pte_mask &= ~_PAGE_GLOBAL;
1613 
1614 	/*
1615 	 * Prevent page tables from being allocated in highmem, even
1616 	 * if CONFIG_HIGHPTE is enabled.
1617 	 */
1618 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1619 
1620 	/* Work out if we support NX */
1621 	x86_configure_nx();
1622 
1623 	/* Get mfn list */
1624 	xen_build_dynamic_phys_to_machine();
1625 
1626 	/*
1627 	 * Set up kernel GDT and segment registers, mainly so that
1628 	 * -fstack-protector code can be executed.
1629 	 */
1630 	xen_setup_gdt(0);
1631 
1632 	xen_init_irq_ops();
1633 	xen_init_cpuid_mask();
1634 
1635 #ifdef CONFIG_X86_LOCAL_APIC
1636 	/*
1637 	 * set up the basic apic ops.
1638 	 */
1639 	xen_init_apic();
1640 #endif
1641 
1642 	if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1643 		pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1644 		pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1645 	}
1646 
1647 	machine_ops = xen_machine_ops;
1648 
1649 	/*
1650 	 * The only reliable way to retain the initial address of the
1651 	 * percpu gdt_page is to remember it here, so we can go and
1652 	 * mark it RW later, when the initial percpu area is freed.
1653 	 */
1654 	xen_initial_gdt = &per_cpu(gdt_page, 0);
1655 
1656 	xen_smp_init();
1657 
1658 #ifdef CONFIG_ACPI_NUMA
1659 	/*
1660 	 * The pages we from Xen are not related to machine pages, so
1661 	 * any NUMA information the kernel tries to get from ACPI will
1662 	 * be meaningless.  Prevent it from trying.
1663 	 */
1664 	acpi_numa = -1;
1665 #endif
1666 	/* Don't do the full vcpu_info placement stuff until we have a
1667 	   possible map and a non-dummy shared_info. */
1668 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1669 
1670 	WARN_ON(xen_cpuhp_setup());
1671 
1672 	local_irq_disable();
1673 	early_boot_irqs_disabled = true;
1674 
1675 	xen_raw_console_write("mapping kernel into physical memory\n");
1676 	xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1677 				   xen_start_info->nr_pages);
1678 	xen_reserve_special_pages();
1679 
1680 	/* keep using Xen gdt for now; no urgent need to change it */
1681 
1682 #ifdef CONFIG_X86_32
1683 	pv_info.kernel_rpl = 1;
1684 	if (xen_feature(XENFEAT_supervisor_mode_kernel))
1685 		pv_info.kernel_rpl = 0;
1686 #else
1687 	pv_info.kernel_rpl = 0;
1688 #endif
1689 	/* set the limit of our address space */
1690 	xen_reserve_top();
1691 
1692 	/* PVH: runs at default kernel iopl of 0 */
1693 	if (!xen_pvh_domain()) {
1694 		/*
1695 		 * We used to do this in xen_arch_setup, but that is too late
1696 		 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1697 		 * early_amd_init which pokes 0xcf8 port.
1698 		 */
1699 		set_iopl.iopl = 1;
1700 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1701 		if (rc != 0)
1702 			xen_raw_printk("physdev_op failed %d\n", rc);
1703 	}
1704 
1705 #ifdef CONFIG_X86_32
1706 	/* set up basic CPUID stuff */
1707 	cpu_detect(&new_cpu_data);
1708 	set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1709 	new_cpu_data.wp_works_ok = 1;
1710 	new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1711 #endif
1712 
1713 	if (xen_start_info->mod_start) {
1714 	    if (xen_start_info->flags & SIF_MOD_START_PFN)
1715 		initrd_start = PFN_PHYS(xen_start_info->mod_start);
1716 	    else
1717 		initrd_start = __pa(xen_start_info->mod_start);
1718 	}
1719 
1720 	/* Poke various useful things into boot_params */
1721 	boot_params.hdr.type_of_loader = (9 << 4) | 0;
1722 	boot_params.hdr.ramdisk_image = initrd_start;
1723 	boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1724 	boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1725 	boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1726 
1727 	if (!xen_initial_domain()) {
1728 		add_preferred_console("xenboot", 0, NULL);
1729 		add_preferred_console("tty", 0, NULL);
1730 		add_preferred_console("hvc", 0, NULL);
1731 		if (pci_xen)
1732 			x86_init.pci.arch_init = pci_xen_init;
1733 	} else {
1734 		const struct dom0_vga_console_info *info =
1735 			(void *)((char *)xen_start_info +
1736 				 xen_start_info->console.dom0.info_off);
1737 		struct xen_platform_op op = {
1738 			.cmd = XENPF_firmware_info,
1739 			.interface_version = XENPF_INTERFACE_VERSION,
1740 			.u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1741 		};
1742 
1743 		x86_platform.set_legacy_features =
1744 				xen_dom0_set_legacy_features;
1745 		xen_init_vga(info, xen_start_info->console.dom0.info_size);
1746 		xen_start_info->console.domU.mfn = 0;
1747 		xen_start_info->console.domU.evtchn = 0;
1748 
1749 		if (HYPERVISOR_platform_op(&op) == 0)
1750 			boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1751 
1752 		/* Make sure ACS will be enabled */
1753 		pci_request_acs();
1754 
1755 		xen_acpi_sleep_register();
1756 
1757 		/* Avoid searching for BIOS MP tables */
1758 		x86_init.mpparse.find_smp_config = x86_init_noop;
1759 		x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1760 
1761 		xen_boot_params_init_edd();
1762 	}
1763 #ifdef CONFIG_PCI
1764 	/* PCI BIOS service won't work from a PV guest. */
1765 	pci_probe &= ~PCI_PROBE_BIOS;
1766 #endif
1767 	xen_raw_console_write("about to get started...\n");
1768 
1769 	/* Let's presume PV guests always boot on vCPU with id 0. */
1770 	per_cpu(xen_vcpu_id, 0) = 0;
1771 
1772 	xen_setup_runstate_info(0);
1773 
1774 	xen_efi_init();
1775 
1776 	/* Start the world */
1777 #ifdef CONFIG_X86_32
1778 	i386_start_kernel();
1779 #else
1780 	cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1781 	x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1782 #endif
1783 }
1784 
xen_hvm_init_shared_info(void)1785 void __ref xen_hvm_init_shared_info(void)
1786 {
1787 	int cpu;
1788 	struct xen_add_to_physmap xatp;
1789 	static struct shared_info *shared_info_page = 0;
1790 
1791 	if (!shared_info_page)
1792 		shared_info_page = (struct shared_info *)
1793 			extend_brk(PAGE_SIZE, PAGE_SIZE);
1794 	xatp.domid = DOMID_SELF;
1795 	xatp.idx = 0;
1796 	xatp.space = XENMAPSPACE_shared_info;
1797 	xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1798 	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1799 		BUG();
1800 
1801 	HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1802 
1803 	/* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1804 	 * page, we use it in the event channel upcall and in some pvclock
1805 	 * related functions. We don't need the vcpu_info placement
1806 	 * optimizations because we don't use any pv_mmu or pv_irq op on
1807 	 * HVM.
1808 	 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1809 	 * online but xen_hvm_init_shared_info is run at resume time too and
1810 	 * in that case multiple vcpus might be online. */
1811 	for_each_online_cpu(cpu) {
1812 		/* Leave it to be NULL. */
1813 		if (xen_vcpu_nr(cpu) >= MAX_VIRT_CPUS)
1814 			continue;
1815 		per_cpu(xen_vcpu, cpu) =
1816 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
1817 	}
1818 }
1819 
1820 #ifdef CONFIG_XEN_PVHVM
init_hvm_pv_info(void)1821 static void __init init_hvm_pv_info(void)
1822 {
1823 	int major, minor;
1824 	uint32_t eax, ebx, ecx, edx, pages, msr, base;
1825 	u64 pfn;
1826 
1827 	base = xen_cpuid_base();
1828 	cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1829 
1830 	major = eax >> 16;
1831 	minor = eax & 0xffff;
1832 	printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1833 
1834 	cpuid(base + 2, &pages, &msr, &ecx, &edx);
1835 
1836 	pfn = __pa(hypercall_page);
1837 	wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1838 
1839 	xen_setup_features();
1840 
1841 	cpuid(base + 4, &eax, &ebx, &ecx, &edx);
1842 	if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
1843 		this_cpu_write(xen_vcpu_id, ebx);
1844 	else
1845 		this_cpu_write(xen_vcpu_id, smp_processor_id());
1846 
1847 	pv_info.name = "Xen HVM";
1848 
1849 	xen_domain_type = XEN_HVM_DOMAIN;
1850 }
1851 #endif
1852 
xen_cpu_up_prepare(unsigned int cpu)1853 static int xen_cpu_up_prepare(unsigned int cpu)
1854 {
1855 	int rc;
1856 
1857 	if (xen_hvm_domain()) {
1858 		/*
1859 		 * This can happen if CPU was offlined earlier and
1860 		 * offlining timed out in common_cpu_die().
1861 		 */
1862 		if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
1863 			xen_smp_intr_free(cpu);
1864 			xen_uninit_lock_cpu(cpu);
1865 		}
1866 
1867 		if (cpu_acpi_id(cpu) != U32_MAX)
1868 			per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
1869 		else
1870 			per_cpu(xen_vcpu_id, cpu) = cpu;
1871 		xen_vcpu_setup(cpu);
1872 	}
1873 
1874 	if (xen_pv_domain() ||
1875 	    (xen_have_vector_callback &&
1876 	     xen_feature(XENFEAT_hvm_safe_pvclock)))
1877 		xen_setup_timer(cpu);
1878 
1879 	rc = xen_smp_intr_init(cpu);
1880 	if (rc) {
1881 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1882 		     cpu, rc);
1883 		return rc;
1884 	}
1885 	return 0;
1886 }
1887 
xen_cpu_dead(unsigned int cpu)1888 static int xen_cpu_dead(unsigned int cpu)
1889 {
1890 	xen_smp_intr_free(cpu);
1891 
1892 	if (xen_pv_domain() ||
1893 	    (xen_have_vector_callback &&
1894 	     xen_feature(XENFEAT_hvm_safe_pvclock)))
1895 		xen_teardown_timer(cpu);
1896 
1897 	return 0;
1898 }
1899 
xen_cpu_up_online(unsigned int cpu)1900 static int xen_cpu_up_online(unsigned int cpu)
1901 {
1902 	xen_init_lock_cpu(cpu);
1903 	return 0;
1904 }
1905 
1906 #ifdef CONFIG_XEN_PVHVM
1907 #ifdef CONFIG_KEXEC_CORE
xen_hvm_shutdown(void)1908 static void xen_hvm_shutdown(void)
1909 {
1910 	native_machine_shutdown();
1911 	if (kexec_in_progress)
1912 		xen_reboot(SHUTDOWN_soft_reset);
1913 }
1914 
xen_hvm_crash_shutdown(struct pt_regs * regs)1915 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1916 {
1917 	native_machine_crash_shutdown(regs);
1918 	xen_reboot(SHUTDOWN_soft_reset);
1919 }
1920 #endif
1921 
xen_hvm_guest_init(void)1922 static void __init xen_hvm_guest_init(void)
1923 {
1924 	if (xen_pv_domain())
1925 		return;
1926 
1927 	init_hvm_pv_info();
1928 
1929 	xen_hvm_init_shared_info();
1930 
1931 	xen_panic_handler_init();
1932 
1933 	if (xen_feature(XENFEAT_hvm_callback_vector))
1934 		xen_have_vector_callback = 1;
1935 	xen_hvm_smp_init();
1936 	WARN_ON(xen_cpuhp_setup());
1937 	xen_unplug_emulated_devices();
1938 	x86_init.irqs.intr_init = xen_init_IRQ;
1939 	xen_hvm_init_time_ops();
1940 	xen_hvm_init_mmu_ops();
1941 #ifdef CONFIG_KEXEC_CORE
1942 	machine_ops.shutdown = xen_hvm_shutdown;
1943 	machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1944 #endif
1945 }
1946 #endif
1947 
1948 static bool xen_nopv = false;
xen_parse_nopv(char * arg)1949 static __init int xen_parse_nopv(char *arg)
1950 {
1951        xen_nopv = true;
1952        return 0;
1953 }
1954 early_param("xen_nopv", xen_parse_nopv);
1955 
xen_platform(void)1956 static uint32_t __init xen_platform(void)
1957 {
1958 	if (xen_nopv)
1959 		return 0;
1960 
1961 	return xen_cpuid_base();
1962 }
1963 
xen_hvm_need_lapic(void)1964 bool xen_hvm_need_lapic(void)
1965 {
1966 	if (xen_nopv)
1967 		return false;
1968 	if (xen_pv_domain())
1969 		return false;
1970 	if (!xen_hvm_domain())
1971 		return false;
1972 	if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1973 		return false;
1974 	return true;
1975 }
1976 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1977 
xen_set_cpu_features(struct cpuinfo_x86 * c)1978 static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1979 {
1980 	if (xen_pv_domain()) {
1981 		clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1982 		set_cpu_cap(c, X86_FEATURE_XENPV);
1983 	}
1984 }
1985 
xen_pin_vcpu(int cpu)1986 static void xen_pin_vcpu(int cpu)
1987 {
1988 	static bool disable_pinning;
1989 	struct sched_pin_override pin_override;
1990 	int ret;
1991 
1992 	if (disable_pinning)
1993 		return;
1994 
1995 	pin_override.pcpu = cpu;
1996 	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
1997 
1998 	/* Ignore errors when removing override. */
1999 	if (cpu < 0)
2000 		return;
2001 
2002 	switch (ret) {
2003 	case -ENOSYS:
2004 		pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
2005 			cpu);
2006 		disable_pinning = true;
2007 		break;
2008 	case -EPERM:
2009 		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
2010 		disable_pinning = true;
2011 		break;
2012 	case -EINVAL:
2013 	case -EBUSY:
2014 		pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
2015 			cpu);
2016 		break;
2017 	case 0:
2018 		break;
2019 	default:
2020 		WARN(1, "rc %d while trying to pin vcpu\n", ret);
2021 		disable_pinning = true;
2022 	}
2023 }
2024 
2025 const struct hypervisor_x86 x86_hyper_xen = {
2026 	.name			= "Xen",
2027 	.detect			= xen_platform,
2028 #ifdef CONFIG_XEN_PVHVM
2029 	.init_platform		= xen_hvm_guest_init,
2030 #endif
2031 	.x2apic_available	= xen_x2apic_para_available,
2032 	.set_cpu_features       = xen_set_cpu_features,
2033 	.pin_vcpu               = xen_pin_vcpu,
2034 };
2035 EXPORT_SYMBOL(x86_hyper_xen);
2036 
2037 #ifdef CONFIG_HOTPLUG_CPU
xen_arch_register_cpu(int num)2038 void xen_arch_register_cpu(int num)
2039 {
2040 	arch_register_cpu(num);
2041 }
2042 EXPORT_SYMBOL(xen_arch_register_cpu);
2043 
xen_arch_unregister_cpu(int num)2044 void xen_arch_unregister_cpu(int num)
2045 {
2046 	arch_unregister_cpu(num);
2047 }
2048 EXPORT_SYMBOL(xen_arch_unregister_cpu);
2049 #endif
2050