xref: /openbmc/linux/arch/x86/kvm/x86.c (revision b04b4f78)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *   Amit Shah    <amit.shah@qumranet.com>
14  *   Ben-Ami Yassour <benami@il.ibm.com>
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.  See
17  * the COPYING file in the top-level directory.
18  *
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "irq.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "x86.h"
28 
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 
41 #include <asm/uaccess.h>
42 #include <asm/msr.h>
43 #include <asm/desc.h>
44 #include <asm/mtrr.h>
45 
46 #define MAX_IO_MSRS 256
47 #define CR0_RESERVED_BITS						\
48 	(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
49 			  | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
50 			  | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
51 #define CR4_RESERVED_BITS						\
52 	(~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
53 			  | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE	\
54 			  | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR	\
55 			  | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
56 
57 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
58 /* EFER defaults:
59  * - enable syscall per default because its emulated by KVM
60  * - enable LME and LMA per default on 64 bit KVM
61  */
62 #ifdef CONFIG_X86_64
63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
64 #else
65 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
66 #endif
67 
68 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
69 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
70 
71 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
72 				    struct kvm_cpuid_entry2 __user *entries);
73 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
74 					      u32 function, u32 index);
75 
76 struct kvm_x86_ops *kvm_x86_ops;
77 EXPORT_SYMBOL_GPL(kvm_x86_ops);
78 
79 struct kvm_stats_debugfs_item debugfs_entries[] = {
80 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
81 	{ "pf_guest", VCPU_STAT(pf_guest) },
82 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
83 	{ "invlpg", VCPU_STAT(invlpg) },
84 	{ "exits", VCPU_STAT(exits) },
85 	{ "io_exits", VCPU_STAT(io_exits) },
86 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
87 	{ "signal_exits", VCPU_STAT(signal_exits) },
88 	{ "irq_window", VCPU_STAT(irq_window_exits) },
89 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
90 	{ "halt_exits", VCPU_STAT(halt_exits) },
91 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
92 	{ "hypercalls", VCPU_STAT(hypercalls) },
93 	{ "request_irq", VCPU_STAT(request_irq_exits) },
94 	{ "request_nmi", VCPU_STAT(request_nmi_exits) },
95 	{ "irq_exits", VCPU_STAT(irq_exits) },
96 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
97 	{ "efer_reload", VCPU_STAT(efer_reload) },
98 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
99 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
100 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
101 	{ "irq_injections", VCPU_STAT(irq_injections) },
102 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
103 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
104 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
105 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
106 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
107 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
108 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
109 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
110 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
111 	{ "mmu_unsync_global", VM_STAT(mmu_unsync_global) },
112 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
113 	{ "largepages", VM_STAT(lpages) },
114 	{ NULL }
115 };
116 
117 unsigned long segment_base(u16 selector)
118 {
119 	struct descriptor_table gdt;
120 	struct desc_struct *d;
121 	unsigned long table_base;
122 	unsigned long v;
123 
124 	if (selector == 0)
125 		return 0;
126 
127 	asm("sgdt %0" : "=m"(gdt));
128 	table_base = gdt.base;
129 
130 	if (selector & 4) {           /* from ldt */
131 		u16 ldt_selector;
132 
133 		asm("sldt %0" : "=g"(ldt_selector));
134 		table_base = segment_base(ldt_selector);
135 	}
136 	d = (struct desc_struct *)(table_base + (selector & ~7));
137 	v = d->base0 | ((unsigned long)d->base1 << 16) |
138 		((unsigned long)d->base2 << 24);
139 #ifdef CONFIG_X86_64
140 	if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
141 		v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
142 #endif
143 	return v;
144 }
145 EXPORT_SYMBOL_GPL(segment_base);
146 
147 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
148 {
149 	if (irqchip_in_kernel(vcpu->kvm))
150 		return vcpu->arch.apic_base;
151 	else
152 		return vcpu->arch.apic_base;
153 }
154 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
155 
156 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
157 {
158 	/* TODO: reserve bits check */
159 	if (irqchip_in_kernel(vcpu->kvm))
160 		kvm_lapic_set_base(vcpu, data);
161 	else
162 		vcpu->arch.apic_base = data;
163 }
164 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
165 
166 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
167 {
168 	WARN_ON(vcpu->arch.exception.pending);
169 	vcpu->arch.exception.pending = true;
170 	vcpu->arch.exception.has_error_code = false;
171 	vcpu->arch.exception.nr = nr;
172 }
173 EXPORT_SYMBOL_GPL(kvm_queue_exception);
174 
175 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
176 			   u32 error_code)
177 {
178 	++vcpu->stat.pf_guest;
179 
180 	if (vcpu->arch.exception.pending) {
181 		if (vcpu->arch.exception.nr == PF_VECTOR) {
182 			printk(KERN_DEBUG "kvm: inject_page_fault:"
183 					" double fault 0x%lx\n", addr);
184 			vcpu->arch.exception.nr = DF_VECTOR;
185 			vcpu->arch.exception.error_code = 0;
186 		} else if (vcpu->arch.exception.nr == DF_VECTOR) {
187 			/* triple fault -> shutdown */
188 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
189 		}
190 		return;
191 	}
192 	vcpu->arch.cr2 = addr;
193 	kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
194 }
195 
196 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
197 {
198 	vcpu->arch.nmi_pending = 1;
199 }
200 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
201 
202 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
203 {
204 	WARN_ON(vcpu->arch.exception.pending);
205 	vcpu->arch.exception.pending = true;
206 	vcpu->arch.exception.has_error_code = true;
207 	vcpu->arch.exception.nr = nr;
208 	vcpu->arch.exception.error_code = error_code;
209 }
210 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
211 
212 static void __queue_exception(struct kvm_vcpu *vcpu)
213 {
214 	kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
215 				     vcpu->arch.exception.has_error_code,
216 				     vcpu->arch.exception.error_code);
217 }
218 
219 /*
220  * Load the pae pdptrs.  Return true is they are all valid.
221  */
222 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
223 {
224 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
225 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
226 	int i;
227 	int ret;
228 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
229 
230 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
231 				  offset * sizeof(u64), sizeof(pdpte));
232 	if (ret < 0) {
233 		ret = 0;
234 		goto out;
235 	}
236 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
237 		if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
238 			ret = 0;
239 			goto out;
240 		}
241 	}
242 	ret = 1;
243 
244 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
245 out:
246 
247 	return ret;
248 }
249 EXPORT_SYMBOL_GPL(load_pdptrs);
250 
251 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
252 {
253 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
254 	bool changed = true;
255 	int r;
256 
257 	if (is_long_mode(vcpu) || !is_pae(vcpu))
258 		return false;
259 
260 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
261 	if (r < 0)
262 		goto out;
263 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
264 out:
265 
266 	return changed;
267 }
268 
269 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
270 {
271 	if (cr0 & CR0_RESERVED_BITS) {
272 		printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
273 		       cr0, vcpu->arch.cr0);
274 		kvm_inject_gp(vcpu, 0);
275 		return;
276 	}
277 
278 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
279 		printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
280 		kvm_inject_gp(vcpu, 0);
281 		return;
282 	}
283 
284 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
285 		printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
286 		       "and a clear PE flag\n");
287 		kvm_inject_gp(vcpu, 0);
288 		return;
289 	}
290 
291 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
292 #ifdef CONFIG_X86_64
293 		if ((vcpu->arch.shadow_efer & EFER_LME)) {
294 			int cs_db, cs_l;
295 
296 			if (!is_pae(vcpu)) {
297 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
298 				       "in long mode while PAE is disabled\n");
299 				kvm_inject_gp(vcpu, 0);
300 				return;
301 			}
302 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
303 			if (cs_l) {
304 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
305 				       "in long mode while CS.L == 1\n");
306 				kvm_inject_gp(vcpu, 0);
307 				return;
308 
309 			}
310 		} else
311 #endif
312 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
313 			printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
314 			       "reserved bits\n");
315 			kvm_inject_gp(vcpu, 0);
316 			return;
317 		}
318 
319 	}
320 
321 	kvm_x86_ops->set_cr0(vcpu, cr0);
322 	vcpu->arch.cr0 = cr0;
323 
324 	kvm_mmu_sync_global(vcpu);
325 	kvm_mmu_reset_context(vcpu);
326 	return;
327 }
328 EXPORT_SYMBOL_GPL(kvm_set_cr0);
329 
330 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
331 {
332 	kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
333 	KVMTRACE_1D(LMSW, vcpu,
334 		    (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
335 		    handler);
336 }
337 EXPORT_SYMBOL_GPL(kvm_lmsw);
338 
339 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
340 {
341 	if (cr4 & CR4_RESERVED_BITS) {
342 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
343 		kvm_inject_gp(vcpu, 0);
344 		return;
345 	}
346 
347 	if (is_long_mode(vcpu)) {
348 		if (!(cr4 & X86_CR4_PAE)) {
349 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
350 			       "in long mode\n");
351 			kvm_inject_gp(vcpu, 0);
352 			return;
353 		}
354 	} else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
355 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
356 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
357 		kvm_inject_gp(vcpu, 0);
358 		return;
359 	}
360 
361 	if (cr4 & X86_CR4_VMXE) {
362 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
363 		kvm_inject_gp(vcpu, 0);
364 		return;
365 	}
366 	kvm_x86_ops->set_cr4(vcpu, cr4);
367 	vcpu->arch.cr4 = cr4;
368 	vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
369 	kvm_mmu_sync_global(vcpu);
370 	kvm_mmu_reset_context(vcpu);
371 }
372 EXPORT_SYMBOL_GPL(kvm_set_cr4);
373 
374 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
375 {
376 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
377 		kvm_mmu_sync_roots(vcpu);
378 		kvm_mmu_flush_tlb(vcpu);
379 		return;
380 	}
381 
382 	if (is_long_mode(vcpu)) {
383 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
384 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
385 			kvm_inject_gp(vcpu, 0);
386 			return;
387 		}
388 	} else {
389 		if (is_pae(vcpu)) {
390 			if (cr3 & CR3_PAE_RESERVED_BITS) {
391 				printk(KERN_DEBUG
392 				       "set_cr3: #GP, reserved bits\n");
393 				kvm_inject_gp(vcpu, 0);
394 				return;
395 			}
396 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
397 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
398 				       "reserved bits\n");
399 				kvm_inject_gp(vcpu, 0);
400 				return;
401 			}
402 		}
403 		/*
404 		 * We don't check reserved bits in nonpae mode, because
405 		 * this isn't enforced, and VMware depends on this.
406 		 */
407 	}
408 
409 	/*
410 	 * Does the new cr3 value map to physical memory? (Note, we
411 	 * catch an invalid cr3 even in real-mode, because it would
412 	 * cause trouble later on when we turn on paging anyway.)
413 	 *
414 	 * A real CPU would silently accept an invalid cr3 and would
415 	 * attempt to use it - with largely undefined (and often hard
416 	 * to debug) behavior on the guest side.
417 	 */
418 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
419 		kvm_inject_gp(vcpu, 0);
420 	else {
421 		vcpu->arch.cr3 = cr3;
422 		vcpu->arch.mmu.new_cr3(vcpu);
423 	}
424 }
425 EXPORT_SYMBOL_GPL(kvm_set_cr3);
426 
427 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
428 {
429 	if (cr8 & CR8_RESERVED_BITS) {
430 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
431 		kvm_inject_gp(vcpu, 0);
432 		return;
433 	}
434 	if (irqchip_in_kernel(vcpu->kvm))
435 		kvm_lapic_set_tpr(vcpu, cr8);
436 	else
437 		vcpu->arch.cr8 = cr8;
438 }
439 EXPORT_SYMBOL_GPL(kvm_set_cr8);
440 
441 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
442 {
443 	if (irqchip_in_kernel(vcpu->kvm))
444 		return kvm_lapic_get_cr8(vcpu);
445 	else
446 		return vcpu->arch.cr8;
447 }
448 EXPORT_SYMBOL_GPL(kvm_get_cr8);
449 
450 static inline u32 bit(int bitno)
451 {
452 	return 1 << (bitno & 31);
453 }
454 
455 /*
456  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
457  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
458  *
459  * This list is modified at module load time to reflect the
460  * capabilities of the host cpu.
461  */
462 static u32 msrs_to_save[] = {
463 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
464 	MSR_K6_STAR,
465 #ifdef CONFIG_X86_64
466 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
467 #endif
468 	MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
469 	MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
470 };
471 
472 static unsigned num_msrs_to_save;
473 
474 static u32 emulated_msrs[] = {
475 	MSR_IA32_MISC_ENABLE,
476 };
477 
478 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
479 {
480 	if (efer & efer_reserved_bits) {
481 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
482 		       efer);
483 		kvm_inject_gp(vcpu, 0);
484 		return;
485 	}
486 
487 	if (is_paging(vcpu)
488 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
489 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
490 		kvm_inject_gp(vcpu, 0);
491 		return;
492 	}
493 
494 	if (efer & EFER_FFXSR) {
495 		struct kvm_cpuid_entry2 *feat;
496 
497 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
498 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
499 			printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
500 			kvm_inject_gp(vcpu, 0);
501 			return;
502 		}
503 	}
504 
505 	if (efer & EFER_SVME) {
506 		struct kvm_cpuid_entry2 *feat;
507 
508 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
509 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
510 			printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
511 			kvm_inject_gp(vcpu, 0);
512 			return;
513 		}
514 	}
515 
516 	kvm_x86_ops->set_efer(vcpu, efer);
517 
518 	efer &= ~EFER_LMA;
519 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
520 
521 	vcpu->arch.shadow_efer = efer;
522 }
523 
524 void kvm_enable_efer_bits(u64 mask)
525 {
526        efer_reserved_bits &= ~mask;
527 }
528 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
529 
530 
531 /*
532  * Writes msr value into into the appropriate "register".
533  * Returns 0 on success, non-0 otherwise.
534  * Assumes vcpu_load() was already called.
535  */
536 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
537 {
538 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
539 }
540 
541 /*
542  * Adapt set_msr() to msr_io()'s calling convention
543  */
544 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
545 {
546 	return kvm_set_msr(vcpu, index, *data);
547 }
548 
549 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
550 {
551 	static int version;
552 	struct pvclock_wall_clock wc;
553 	struct timespec now, sys, boot;
554 
555 	if (!wall_clock)
556 		return;
557 
558 	version++;
559 
560 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
561 
562 	/*
563 	 * The guest calculates current wall clock time by adding
564 	 * system time (updated by kvm_write_guest_time below) to the
565 	 * wall clock specified here.  guest system time equals host
566 	 * system time for us, thus we must fill in host boot time here.
567 	 */
568 	now = current_kernel_time();
569 	ktime_get_ts(&sys);
570 	boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
571 
572 	wc.sec = boot.tv_sec;
573 	wc.nsec = boot.tv_nsec;
574 	wc.version = version;
575 
576 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
577 
578 	version++;
579 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
580 }
581 
582 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
583 {
584 	uint32_t quotient, remainder;
585 
586 	/* Don't try to replace with do_div(), this one calculates
587 	 * "(dividend << 32) / divisor" */
588 	__asm__ ( "divl %4"
589 		  : "=a" (quotient), "=d" (remainder)
590 		  : "0" (0), "1" (dividend), "r" (divisor) );
591 	return quotient;
592 }
593 
594 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
595 {
596 	uint64_t nsecs = 1000000000LL;
597 	int32_t  shift = 0;
598 	uint64_t tps64;
599 	uint32_t tps32;
600 
601 	tps64 = tsc_khz * 1000LL;
602 	while (tps64 > nsecs*2) {
603 		tps64 >>= 1;
604 		shift--;
605 	}
606 
607 	tps32 = (uint32_t)tps64;
608 	while (tps32 <= (uint32_t)nsecs) {
609 		tps32 <<= 1;
610 		shift++;
611 	}
612 
613 	hv_clock->tsc_shift = shift;
614 	hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
615 
616 	pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
617 		 __func__, tsc_khz, hv_clock->tsc_shift,
618 		 hv_clock->tsc_to_system_mul);
619 }
620 
621 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
622 
623 static void kvm_write_guest_time(struct kvm_vcpu *v)
624 {
625 	struct timespec ts;
626 	unsigned long flags;
627 	struct kvm_vcpu_arch *vcpu = &v->arch;
628 	void *shared_kaddr;
629 
630 	if ((!vcpu->time_page))
631 		return;
632 
633 	if (unlikely(vcpu->hv_clock_tsc_khz != __get_cpu_var(cpu_tsc_khz))) {
634 		kvm_set_time_scale(__get_cpu_var(cpu_tsc_khz), &vcpu->hv_clock);
635 		vcpu->hv_clock_tsc_khz = __get_cpu_var(cpu_tsc_khz);
636 	}
637 
638 	/* Keep irq disabled to prevent changes to the clock */
639 	local_irq_save(flags);
640 	kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
641 			  &vcpu->hv_clock.tsc_timestamp);
642 	ktime_get_ts(&ts);
643 	local_irq_restore(flags);
644 
645 	/* With all the info we got, fill in the values */
646 
647 	vcpu->hv_clock.system_time = ts.tv_nsec +
648 				     (NSEC_PER_SEC * (u64)ts.tv_sec);
649 	/*
650 	 * The interface expects us to write an even number signaling that the
651 	 * update is finished. Since the guest won't see the intermediate
652 	 * state, we just increase by 2 at the end.
653 	 */
654 	vcpu->hv_clock.version += 2;
655 
656 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
657 
658 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
659 	       sizeof(vcpu->hv_clock));
660 
661 	kunmap_atomic(shared_kaddr, KM_USER0);
662 
663 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
664 }
665 
666 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
667 {
668 	struct kvm_vcpu_arch *vcpu = &v->arch;
669 
670 	if (!vcpu->time_page)
671 		return 0;
672 	set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
673 	return 1;
674 }
675 
676 static bool msr_mtrr_valid(unsigned msr)
677 {
678 	switch (msr) {
679 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
680 	case MSR_MTRRfix64K_00000:
681 	case MSR_MTRRfix16K_80000:
682 	case MSR_MTRRfix16K_A0000:
683 	case MSR_MTRRfix4K_C0000:
684 	case MSR_MTRRfix4K_C8000:
685 	case MSR_MTRRfix4K_D0000:
686 	case MSR_MTRRfix4K_D8000:
687 	case MSR_MTRRfix4K_E0000:
688 	case MSR_MTRRfix4K_E8000:
689 	case MSR_MTRRfix4K_F0000:
690 	case MSR_MTRRfix4K_F8000:
691 	case MSR_MTRRdefType:
692 	case MSR_IA32_CR_PAT:
693 		return true;
694 	case 0x2f8:
695 		return true;
696 	}
697 	return false;
698 }
699 
700 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
701 {
702 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
703 
704 	if (!msr_mtrr_valid(msr))
705 		return 1;
706 
707 	if (msr == MSR_MTRRdefType) {
708 		vcpu->arch.mtrr_state.def_type = data;
709 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
710 	} else if (msr == MSR_MTRRfix64K_00000)
711 		p[0] = data;
712 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
713 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
714 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
715 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
716 	else if (msr == MSR_IA32_CR_PAT)
717 		vcpu->arch.pat = data;
718 	else {	/* Variable MTRRs */
719 		int idx, is_mtrr_mask;
720 		u64 *pt;
721 
722 		idx = (msr - 0x200) / 2;
723 		is_mtrr_mask = msr - 0x200 - 2 * idx;
724 		if (!is_mtrr_mask)
725 			pt =
726 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
727 		else
728 			pt =
729 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
730 		*pt = data;
731 	}
732 
733 	kvm_mmu_reset_context(vcpu);
734 	return 0;
735 }
736 
737 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
738 {
739 	switch (msr) {
740 	case MSR_EFER:
741 		set_efer(vcpu, data);
742 		break;
743 	case MSR_IA32_MC0_STATUS:
744 		pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
745 		       __func__, data);
746 		break;
747 	case MSR_IA32_MCG_STATUS:
748 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
749 			__func__, data);
750 		break;
751 	case MSR_IA32_MCG_CTL:
752 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
753 			__func__, data);
754 		break;
755 	case MSR_IA32_DEBUGCTLMSR:
756 		if (!data) {
757 			/* We support the non-activated case already */
758 			break;
759 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
760 			/* Values other than LBR and BTF are vendor-specific,
761 			   thus reserved and should throw a #GP */
762 			return 1;
763 		}
764 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
765 			__func__, data);
766 		break;
767 	case MSR_IA32_UCODE_REV:
768 	case MSR_IA32_UCODE_WRITE:
769 	case MSR_VM_HSAVE_PA:
770 		break;
771 	case 0x200 ... 0x2ff:
772 		return set_msr_mtrr(vcpu, msr, data);
773 	case MSR_IA32_APICBASE:
774 		kvm_set_apic_base(vcpu, data);
775 		break;
776 	case MSR_IA32_MISC_ENABLE:
777 		vcpu->arch.ia32_misc_enable_msr = data;
778 		break;
779 	case MSR_KVM_WALL_CLOCK:
780 		vcpu->kvm->arch.wall_clock = data;
781 		kvm_write_wall_clock(vcpu->kvm, data);
782 		break;
783 	case MSR_KVM_SYSTEM_TIME: {
784 		if (vcpu->arch.time_page) {
785 			kvm_release_page_dirty(vcpu->arch.time_page);
786 			vcpu->arch.time_page = NULL;
787 		}
788 
789 		vcpu->arch.time = data;
790 
791 		/* we verify if the enable bit is set... */
792 		if (!(data & 1))
793 			break;
794 
795 		/* ...but clean it before doing the actual write */
796 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
797 
798 		vcpu->arch.time_page =
799 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
800 
801 		if (is_error_page(vcpu->arch.time_page)) {
802 			kvm_release_page_clean(vcpu->arch.time_page);
803 			vcpu->arch.time_page = NULL;
804 		}
805 
806 		kvm_request_guest_time_update(vcpu);
807 		break;
808 	}
809 	default:
810 		pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
811 		return 1;
812 	}
813 	return 0;
814 }
815 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
816 
817 
818 /*
819  * Reads an msr value (of 'msr_index') into 'pdata'.
820  * Returns 0 on success, non-0 otherwise.
821  * Assumes vcpu_load() was already called.
822  */
823 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
824 {
825 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
826 }
827 
828 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
829 {
830 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
831 
832 	if (!msr_mtrr_valid(msr))
833 		return 1;
834 
835 	if (msr == MSR_MTRRdefType)
836 		*pdata = vcpu->arch.mtrr_state.def_type +
837 			 (vcpu->arch.mtrr_state.enabled << 10);
838 	else if (msr == MSR_MTRRfix64K_00000)
839 		*pdata = p[0];
840 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
841 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
842 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
843 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
844 	else if (msr == MSR_IA32_CR_PAT)
845 		*pdata = vcpu->arch.pat;
846 	else {	/* Variable MTRRs */
847 		int idx, is_mtrr_mask;
848 		u64 *pt;
849 
850 		idx = (msr - 0x200) / 2;
851 		is_mtrr_mask = msr - 0x200 - 2 * idx;
852 		if (!is_mtrr_mask)
853 			pt =
854 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
855 		else
856 			pt =
857 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
858 		*pdata = *pt;
859 	}
860 
861 	return 0;
862 }
863 
864 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
865 {
866 	u64 data;
867 
868 	switch (msr) {
869 	case 0xc0010010: /* SYSCFG */
870 	case 0xc0010015: /* HWCR */
871 	case MSR_IA32_PLATFORM_ID:
872 	case MSR_IA32_P5_MC_ADDR:
873 	case MSR_IA32_P5_MC_TYPE:
874 	case MSR_IA32_MC0_CTL:
875 	case MSR_IA32_MCG_STATUS:
876 	case MSR_IA32_MCG_CAP:
877 	case MSR_IA32_MCG_CTL:
878 	case MSR_IA32_MC0_MISC:
879 	case MSR_IA32_MC0_MISC+4:
880 	case MSR_IA32_MC0_MISC+8:
881 	case MSR_IA32_MC0_MISC+12:
882 	case MSR_IA32_MC0_MISC+16:
883 	case MSR_IA32_MC0_MISC+20:
884 	case MSR_IA32_UCODE_REV:
885 	case MSR_IA32_EBL_CR_POWERON:
886 	case MSR_IA32_DEBUGCTLMSR:
887 	case MSR_IA32_LASTBRANCHFROMIP:
888 	case MSR_IA32_LASTBRANCHTOIP:
889 	case MSR_IA32_LASTINTFROMIP:
890 	case MSR_IA32_LASTINTTOIP:
891 	case MSR_VM_HSAVE_PA:
892 		data = 0;
893 		break;
894 	case MSR_MTRRcap:
895 		data = 0x500 | KVM_NR_VAR_MTRR;
896 		break;
897 	case 0x200 ... 0x2ff:
898 		return get_msr_mtrr(vcpu, msr, pdata);
899 	case 0xcd: /* fsb frequency */
900 		data = 3;
901 		break;
902 	case MSR_IA32_APICBASE:
903 		data = kvm_get_apic_base(vcpu);
904 		break;
905 	case MSR_IA32_MISC_ENABLE:
906 		data = vcpu->arch.ia32_misc_enable_msr;
907 		break;
908 	case MSR_IA32_PERF_STATUS:
909 		/* TSC increment by tick */
910 		data = 1000ULL;
911 		/* CPU multiplier */
912 		data |= (((uint64_t)4ULL) << 40);
913 		break;
914 	case MSR_EFER:
915 		data = vcpu->arch.shadow_efer;
916 		break;
917 	case MSR_KVM_WALL_CLOCK:
918 		data = vcpu->kvm->arch.wall_clock;
919 		break;
920 	case MSR_KVM_SYSTEM_TIME:
921 		data = vcpu->arch.time;
922 		break;
923 	default:
924 		pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
925 		return 1;
926 	}
927 	*pdata = data;
928 	return 0;
929 }
930 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
931 
932 /*
933  * Read or write a bunch of msrs. All parameters are kernel addresses.
934  *
935  * @return number of msrs set successfully.
936  */
937 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
938 		    struct kvm_msr_entry *entries,
939 		    int (*do_msr)(struct kvm_vcpu *vcpu,
940 				  unsigned index, u64 *data))
941 {
942 	int i;
943 
944 	vcpu_load(vcpu);
945 
946 	down_read(&vcpu->kvm->slots_lock);
947 	for (i = 0; i < msrs->nmsrs; ++i)
948 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
949 			break;
950 	up_read(&vcpu->kvm->slots_lock);
951 
952 	vcpu_put(vcpu);
953 
954 	return i;
955 }
956 
957 /*
958  * Read or write a bunch of msrs. Parameters are user addresses.
959  *
960  * @return number of msrs set successfully.
961  */
962 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
963 		  int (*do_msr)(struct kvm_vcpu *vcpu,
964 				unsigned index, u64 *data),
965 		  int writeback)
966 {
967 	struct kvm_msrs msrs;
968 	struct kvm_msr_entry *entries;
969 	int r, n;
970 	unsigned size;
971 
972 	r = -EFAULT;
973 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
974 		goto out;
975 
976 	r = -E2BIG;
977 	if (msrs.nmsrs >= MAX_IO_MSRS)
978 		goto out;
979 
980 	r = -ENOMEM;
981 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
982 	entries = vmalloc(size);
983 	if (!entries)
984 		goto out;
985 
986 	r = -EFAULT;
987 	if (copy_from_user(entries, user_msrs->entries, size))
988 		goto out_free;
989 
990 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
991 	if (r < 0)
992 		goto out_free;
993 
994 	r = -EFAULT;
995 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
996 		goto out_free;
997 
998 	r = n;
999 
1000 out_free:
1001 	vfree(entries);
1002 out:
1003 	return r;
1004 }
1005 
1006 int kvm_dev_ioctl_check_extension(long ext)
1007 {
1008 	int r;
1009 
1010 	switch (ext) {
1011 	case KVM_CAP_IRQCHIP:
1012 	case KVM_CAP_HLT:
1013 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1014 	case KVM_CAP_SET_TSS_ADDR:
1015 	case KVM_CAP_EXT_CPUID:
1016 	case KVM_CAP_CLOCKSOURCE:
1017 	case KVM_CAP_PIT:
1018 	case KVM_CAP_NOP_IO_DELAY:
1019 	case KVM_CAP_MP_STATE:
1020 	case KVM_CAP_SYNC_MMU:
1021 	case KVM_CAP_REINJECT_CONTROL:
1022 	case KVM_CAP_IRQ_INJECT_STATUS:
1023 		r = 1;
1024 		break;
1025 	case KVM_CAP_COALESCED_MMIO:
1026 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1027 		break;
1028 	case KVM_CAP_VAPIC:
1029 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1030 		break;
1031 	case KVM_CAP_NR_VCPUS:
1032 		r = KVM_MAX_VCPUS;
1033 		break;
1034 	case KVM_CAP_NR_MEMSLOTS:
1035 		r = KVM_MEMORY_SLOTS;
1036 		break;
1037 	case KVM_CAP_PV_MMU:
1038 		r = !tdp_enabled;
1039 		break;
1040 	case KVM_CAP_IOMMU:
1041 		r = iommu_found();
1042 		break;
1043 	default:
1044 		r = 0;
1045 		break;
1046 	}
1047 	return r;
1048 
1049 }
1050 
1051 long kvm_arch_dev_ioctl(struct file *filp,
1052 			unsigned int ioctl, unsigned long arg)
1053 {
1054 	void __user *argp = (void __user *)arg;
1055 	long r;
1056 
1057 	switch (ioctl) {
1058 	case KVM_GET_MSR_INDEX_LIST: {
1059 		struct kvm_msr_list __user *user_msr_list = argp;
1060 		struct kvm_msr_list msr_list;
1061 		unsigned n;
1062 
1063 		r = -EFAULT;
1064 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1065 			goto out;
1066 		n = msr_list.nmsrs;
1067 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1068 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1069 			goto out;
1070 		r = -E2BIG;
1071 		if (n < num_msrs_to_save)
1072 			goto out;
1073 		r = -EFAULT;
1074 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1075 				 num_msrs_to_save * sizeof(u32)))
1076 			goto out;
1077 		if (copy_to_user(user_msr_list->indices
1078 				 + num_msrs_to_save * sizeof(u32),
1079 				 &emulated_msrs,
1080 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1081 			goto out;
1082 		r = 0;
1083 		break;
1084 	}
1085 	case KVM_GET_SUPPORTED_CPUID: {
1086 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1087 		struct kvm_cpuid2 cpuid;
1088 
1089 		r = -EFAULT;
1090 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1091 			goto out;
1092 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1093 						      cpuid_arg->entries);
1094 		if (r)
1095 			goto out;
1096 
1097 		r = -EFAULT;
1098 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1099 			goto out;
1100 		r = 0;
1101 		break;
1102 	}
1103 	default:
1104 		r = -EINVAL;
1105 	}
1106 out:
1107 	return r;
1108 }
1109 
1110 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1111 {
1112 	kvm_x86_ops->vcpu_load(vcpu, cpu);
1113 	kvm_request_guest_time_update(vcpu);
1114 }
1115 
1116 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1117 {
1118 	kvm_x86_ops->vcpu_put(vcpu);
1119 	kvm_put_guest_fpu(vcpu);
1120 }
1121 
1122 static int is_efer_nx(void)
1123 {
1124 	u64 efer;
1125 
1126 	rdmsrl(MSR_EFER, efer);
1127 	return efer & EFER_NX;
1128 }
1129 
1130 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1131 {
1132 	int i;
1133 	struct kvm_cpuid_entry2 *e, *entry;
1134 
1135 	entry = NULL;
1136 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1137 		e = &vcpu->arch.cpuid_entries[i];
1138 		if (e->function == 0x80000001) {
1139 			entry = e;
1140 			break;
1141 		}
1142 	}
1143 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1144 		entry->edx &= ~(1 << 20);
1145 		printk(KERN_INFO "kvm: guest NX capability removed\n");
1146 	}
1147 }
1148 
1149 /* when an old userspace process fills a new kernel module */
1150 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1151 				    struct kvm_cpuid *cpuid,
1152 				    struct kvm_cpuid_entry __user *entries)
1153 {
1154 	int r, i;
1155 	struct kvm_cpuid_entry *cpuid_entries;
1156 
1157 	r = -E2BIG;
1158 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1159 		goto out;
1160 	r = -ENOMEM;
1161 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1162 	if (!cpuid_entries)
1163 		goto out;
1164 	r = -EFAULT;
1165 	if (copy_from_user(cpuid_entries, entries,
1166 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1167 		goto out_free;
1168 	for (i = 0; i < cpuid->nent; i++) {
1169 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1170 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1171 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1172 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1173 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1174 		vcpu->arch.cpuid_entries[i].index = 0;
1175 		vcpu->arch.cpuid_entries[i].flags = 0;
1176 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
1177 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
1178 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
1179 	}
1180 	vcpu->arch.cpuid_nent = cpuid->nent;
1181 	cpuid_fix_nx_cap(vcpu);
1182 	r = 0;
1183 
1184 out_free:
1185 	vfree(cpuid_entries);
1186 out:
1187 	return r;
1188 }
1189 
1190 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1191 				     struct kvm_cpuid2 *cpuid,
1192 				     struct kvm_cpuid_entry2 __user *entries)
1193 {
1194 	int r;
1195 
1196 	r = -E2BIG;
1197 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1198 		goto out;
1199 	r = -EFAULT;
1200 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1201 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1202 		goto out;
1203 	vcpu->arch.cpuid_nent = cpuid->nent;
1204 	return 0;
1205 
1206 out:
1207 	return r;
1208 }
1209 
1210 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1211 				     struct kvm_cpuid2 *cpuid,
1212 				     struct kvm_cpuid_entry2 __user *entries)
1213 {
1214 	int r;
1215 
1216 	r = -E2BIG;
1217 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1218 		goto out;
1219 	r = -EFAULT;
1220 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1221 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1222 		goto out;
1223 	return 0;
1224 
1225 out:
1226 	cpuid->nent = vcpu->arch.cpuid_nent;
1227 	return r;
1228 }
1229 
1230 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1231 			   u32 index)
1232 {
1233 	entry->function = function;
1234 	entry->index = index;
1235 	cpuid_count(entry->function, entry->index,
1236 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1237 	entry->flags = 0;
1238 }
1239 
1240 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1241 			 u32 index, int *nent, int maxnent)
1242 {
1243 	const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1244 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1245 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1246 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1247 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1248 		bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1249 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1250 		bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1251 		bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1252 		bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1253 	const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1254 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1255 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1256 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1257 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1258 		bit(X86_FEATURE_PGE) |
1259 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1260 		bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1261 		bit(X86_FEATURE_SYSCALL) |
1262 		(bit(X86_FEATURE_NX) && is_efer_nx()) |
1263 #ifdef CONFIG_X86_64
1264 		bit(X86_FEATURE_LM) |
1265 #endif
1266 		bit(X86_FEATURE_FXSR_OPT) |
1267 		bit(X86_FEATURE_MMXEXT) |
1268 		bit(X86_FEATURE_3DNOWEXT) |
1269 		bit(X86_FEATURE_3DNOW);
1270 	const u32 kvm_supported_word3_x86_features =
1271 		bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1272 	const u32 kvm_supported_word6_x86_features =
1273 		bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1274 		bit(X86_FEATURE_SVM);
1275 
1276 	/* all calls to cpuid_count() should be made on the same cpu */
1277 	get_cpu();
1278 	do_cpuid_1_ent(entry, function, index);
1279 	++*nent;
1280 
1281 	switch (function) {
1282 	case 0:
1283 		entry->eax = min(entry->eax, (u32)0xb);
1284 		break;
1285 	case 1:
1286 		entry->edx &= kvm_supported_word0_x86_features;
1287 		entry->ecx &= kvm_supported_word3_x86_features;
1288 		break;
1289 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1290 	 * may return different values. This forces us to get_cpu() before
1291 	 * issuing the first command, and also to emulate this annoying behavior
1292 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1293 	case 2: {
1294 		int t, times = entry->eax & 0xff;
1295 
1296 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1297 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1298 		for (t = 1; t < times && *nent < maxnent; ++t) {
1299 			do_cpuid_1_ent(&entry[t], function, 0);
1300 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1301 			++*nent;
1302 		}
1303 		break;
1304 	}
1305 	/* function 4 and 0xb have additional index. */
1306 	case 4: {
1307 		int i, cache_type;
1308 
1309 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1310 		/* read more entries until cache_type is zero */
1311 		for (i = 1; *nent < maxnent; ++i) {
1312 			cache_type = entry[i - 1].eax & 0x1f;
1313 			if (!cache_type)
1314 				break;
1315 			do_cpuid_1_ent(&entry[i], function, i);
1316 			entry[i].flags |=
1317 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1318 			++*nent;
1319 		}
1320 		break;
1321 	}
1322 	case 0xb: {
1323 		int i, level_type;
1324 
1325 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1326 		/* read more entries until level_type is zero */
1327 		for (i = 1; *nent < maxnent; ++i) {
1328 			level_type = entry[i - 1].ecx & 0xff00;
1329 			if (!level_type)
1330 				break;
1331 			do_cpuid_1_ent(&entry[i], function, i);
1332 			entry[i].flags |=
1333 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1334 			++*nent;
1335 		}
1336 		break;
1337 	}
1338 	case 0x80000000:
1339 		entry->eax = min(entry->eax, 0x8000001a);
1340 		break;
1341 	case 0x80000001:
1342 		entry->edx &= kvm_supported_word1_x86_features;
1343 		entry->ecx &= kvm_supported_word6_x86_features;
1344 		break;
1345 	}
1346 	put_cpu();
1347 }
1348 
1349 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1350 				     struct kvm_cpuid_entry2 __user *entries)
1351 {
1352 	struct kvm_cpuid_entry2 *cpuid_entries;
1353 	int limit, nent = 0, r = -E2BIG;
1354 	u32 func;
1355 
1356 	if (cpuid->nent < 1)
1357 		goto out;
1358 	r = -ENOMEM;
1359 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1360 	if (!cpuid_entries)
1361 		goto out;
1362 
1363 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1364 	limit = cpuid_entries[0].eax;
1365 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1366 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1367 			     &nent, cpuid->nent);
1368 	r = -E2BIG;
1369 	if (nent >= cpuid->nent)
1370 		goto out_free;
1371 
1372 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1373 	limit = cpuid_entries[nent - 1].eax;
1374 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1375 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1376 			     &nent, cpuid->nent);
1377 	r = -EFAULT;
1378 	if (copy_to_user(entries, cpuid_entries,
1379 			 nent * sizeof(struct kvm_cpuid_entry2)))
1380 		goto out_free;
1381 	cpuid->nent = nent;
1382 	r = 0;
1383 
1384 out_free:
1385 	vfree(cpuid_entries);
1386 out:
1387 	return r;
1388 }
1389 
1390 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1391 				    struct kvm_lapic_state *s)
1392 {
1393 	vcpu_load(vcpu);
1394 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1395 	vcpu_put(vcpu);
1396 
1397 	return 0;
1398 }
1399 
1400 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1401 				    struct kvm_lapic_state *s)
1402 {
1403 	vcpu_load(vcpu);
1404 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1405 	kvm_apic_post_state_restore(vcpu);
1406 	vcpu_put(vcpu);
1407 
1408 	return 0;
1409 }
1410 
1411 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1412 				    struct kvm_interrupt *irq)
1413 {
1414 	if (irq->irq < 0 || irq->irq >= 256)
1415 		return -EINVAL;
1416 	if (irqchip_in_kernel(vcpu->kvm))
1417 		return -ENXIO;
1418 	vcpu_load(vcpu);
1419 
1420 	set_bit(irq->irq, vcpu->arch.irq_pending);
1421 	set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1422 
1423 	vcpu_put(vcpu);
1424 
1425 	return 0;
1426 }
1427 
1428 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1429 {
1430 	vcpu_load(vcpu);
1431 	kvm_inject_nmi(vcpu);
1432 	vcpu_put(vcpu);
1433 
1434 	return 0;
1435 }
1436 
1437 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1438 					   struct kvm_tpr_access_ctl *tac)
1439 {
1440 	if (tac->flags)
1441 		return -EINVAL;
1442 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1443 	return 0;
1444 }
1445 
1446 long kvm_arch_vcpu_ioctl(struct file *filp,
1447 			 unsigned int ioctl, unsigned long arg)
1448 {
1449 	struct kvm_vcpu *vcpu = filp->private_data;
1450 	void __user *argp = (void __user *)arg;
1451 	int r;
1452 	struct kvm_lapic_state *lapic = NULL;
1453 
1454 	switch (ioctl) {
1455 	case KVM_GET_LAPIC: {
1456 		lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1457 
1458 		r = -ENOMEM;
1459 		if (!lapic)
1460 			goto out;
1461 		r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1462 		if (r)
1463 			goto out;
1464 		r = -EFAULT;
1465 		if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1466 			goto out;
1467 		r = 0;
1468 		break;
1469 	}
1470 	case KVM_SET_LAPIC: {
1471 		lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1472 		r = -ENOMEM;
1473 		if (!lapic)
1474 			goto out;
1475 		r = -EFAULT;
1476 		if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1477 			goto out;
1478 		r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1479 		if (r)
1480 			goto out;
1481 		r = 0;
1482 		break;
1483 	}
1484 	case KVM_INTERRUPT: {
1485 		struct kvm_interrupt irq;
1486 
1487 		r = -EFAULT;
1488 		if (copy_from_user(&irq, argp, sizeof irq))
1489 			goto out;
1490 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1491 		if (r)
1492 			goto out;
1493 		r = 0;
1494 		break;
1495 	}
1496 	case KVM_NMI: {
1497 		r = kvm_vcpu_ioctl_nmi(vcpu);
1498 		if (r)
1499 			goto out;
1500 		r = 0;
1501 		break;
1502 	}
1503 	case KVM_SET_CPUID: {
1504 		struct kvm_cpuid __user *cpuid_arg = argp;
1505 		struct kvm_cpuid cpuid;
1506 
1507 		r = -EFAULT;
1508 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1509 			goto out;
1510 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1511 		if (r)
1512 			goto out;
1513 		break;
1514 	}
1515 	case KVM_SET_CPUID2: {
1516 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1517 		struct kvm_cpuid2 cpuid;
1518 
1519 		r = -EFAULT;
1520 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1521 			goto out;
1522 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1523 					      cpuid_arg->entries);
1524 		if (r)
1525 			goto out;
1526 		break;
1527 	}
1528 	case KVM_GET_CPUID2: {
1529 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1530 		struct kvm_cpuid2 cpuid;
1531 
1532 		r = -EFAULT;
1533 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1534 			goto out;
1535 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1536 					      cpuid_arg->entries);
1537 		if (r)
1538 			goto out;
1539 		r = -EFAULT;
1540 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1541 			goto out;
1542 		r = 0;
1543 		break;
1544 	}
1545 	case KVM_GET_MSRS:
1546 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1547 		break;
1548 	case KVM_SET_MSRS:
1549 		r = msr_io(vcpu, argp, do_set_msr, 0);
1550 		break;
1551 	case KVM_TPR_ACCESS_REPORTING: {
1552 		struct kvm_tpr_access_ctl tac;
1553 
1554 		r = -EFAULT;
1555 		if (copy_from_user(&tac, argp, sizeof tac))
1556 			goto out;
1557 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1558 		if (r)
1559 			goto out;
1560 		r = -EFAULT;
1561 		if (copy_to_user(argp, &tac, sizeof tac))
1562 			goto out;
1563 		r = 0;
1564 		break;
1565 	};
1566 	case KVM_SET_VAPIC_ADDR: {
1567 		struct kvm_vapic_addr va;
1568 
1569 		r = -EINVAL;
1570 		if (!irqchip_in_kernel(vcpu->kvm))
1571 			goto out;
1572 		r = -EFAULT;
1573 		if (copy_from_user(&va, argp, sizeof va))
1574 			goto out;
1575 		r = 0;
1576 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1577 		break;
1578 	}
1579 	default:
1580 		r = -EINVAL;
1581 	}
1582 out:
1583 	if (lapic)
1584 		kfree(lapic);
1585 	return r;
1586 }
1587 
1588 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1589 {
1590 	int ret;
1591 
1592 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1593 		return -1;
1594 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1595 	return ret;
1596 }
1597 
1598 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1599 					  u32 kvm_nr_mmu_pages)
1600 {
1601 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1602 		return -EINVAL;
1603 
1604 	down_write(&kvm->slots_lock);
1605 
1606 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1607 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1608 
1609 	up_write(&kvm->slots_lock);
1610 	return 0;
1611 }
1612 
1613 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1614 {
1615 	return kvm->arch.n_alloc_mmu_pages;
1616 }
1617 
1618 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1619 {
1620 	int i;
1621 	struct kvm_mem_alias *alias;
1622 
1623 	for (i = 0; i < kvm->arch.naliases; ++i) {
1624 		alias = &kvm->arch.aliases[i];
1625 		if (gfn >= alias->base_gfn
1626 		    && gfn < alias->base_gfn + alias->npages)
1627 			return alias->target_gfn + gfn - alias->base_gfn;
1628 	}
1629 	return gfn;
1630 }
1631 
1632 /*
1633  * Set a new alias region.  Aliases map a portion of physical memory into
1634  * another portion.  This is useful for memory windows, for example the PC
1635  * VGA region.
1636  */
1637 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1638 					 struct kvm_memory_alias *alias)
1639 {
1640 	int r, n;
1641 	struct kvm_mem_alias *p;
1642 
1643 	r = -EINVAL;
1644 	/* General sanity checks */
1645 	if (alias->memory_size & (PAGE_SIZE - 1))
1646 		goto out;
1647 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1648 		goto out;
1649 	if (alias->slot >= KVM_ALIAS_SLOTS)
1650 		goto out;
1651 	if (alias->guest_phys_addr + alias->memory_size
1652 	    < alias->guest_phys_addr)
1653 		goto out;
1654 	if (alias->target_phys_addr + alias->memory_size
1655 	    < alias->target_phys_addr)
1656 		goto out;
1657 
1658 	down_write(&kvm->slots_lock);
1659 	spin_lock(&kvm->mmu_lock);
1660 
1661 	p = &kvm->arch.aliases[alias->slot];
1662 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1663 	p->npages = alias->memory_size >> PAGE_SHIFT;
1664 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1665 
1666 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1667 		if (kvm->arch.aliases[n - 1].npages)
1668 			break;
1669 	kvm->arch.naliases = n;
1670 
1671 	spin_unlock(&kvm->mmu_lock);
1672 	kvm_mmu_zap_all(kvm);
1673 
1674 	up_write(&kvm->slots_lock);
1675 
1676 	return 0;
1677 
1678 out:
1679 	return r;
1680 }
1681 
1682 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1683 {
1684 	int r;
1685 
1686 	r = 0;
1687 	switch (chip->chip_id) {
1688 	case KVM_IRQCHIP_PIC_MASTER:
1689 		memcpy(&chip->chip.pic,
1690 			&pic_irqchip(kvm)->pics[0],
1691 			sizeof(struct kvm_pic_state));
1692 		break;
1693 	case KVM_IRQCHIP_PIC_SLAVE:
1694 		memcpy(&chip->chip.pic,
1695 			&pic_irqchip(kvm)->pics[1],
1696 			sizeof(struct kvm_pic_state));
1697 		break;
1698 	case KVM_IRQCHIP_IOAPIC:
1699 		memcpy(&chip->chip.ioapic,
1700 			ioapic_irqchip(kvm),
1701 			sizeof(struct kvm_ioapic_state));
1702 		break;
1703 	default:
1704 		r = -EINVAL;
1705 		break;
1706 	}
1707 	return r;
1708 }
1709 
1710 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1711 {
1712 	int r;
1713 
1714 	r = 0;
1715 	switch (chip->chip_id) {
1716 	case KVM_IRQCHIP_PIC_MASTER:
1717 		memcpy(&pic_irqchip(kvm)->pics[0],
1718 			&chip->chip.pic,
1719 			sizeof(struct kvm_pic_state));
1720 		break;
1721 	case KVM_IRQCHIP_PIC_SLAVE:
1722 		memcpy(&pic_irqchip(kvm)->pics[1],
1723 			&chip->chip.pic,
1724 			sizeof(struct kvm_pic_state));
1725 		break;
1726 	case KVM_IRQCHIP_IOAPIC:
1727 		memcpy(ioapic_irqchip(kvm),
1728 			&chip->chip.ioapic,
1729 			sizeof(struct kvm_ioapic_state));
1730 		break;
1731 	default:
1732 		r = -EINVAL;
1733 		break;
1734 	}
1735 	kvm_pic_update_irq(pic_irqchip(kvm));
1736 	return r;
1737 }
1738 
1739 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1740 {
1741 	int r = 0;
1742 
1743 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1744 	return r;
1745 }
1746 
1747 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1748 {
1749 	int r = 0;
1750 
1751 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1752 	kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1753 	return r;
1754 }
1755 
1756 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1757 				 struct kvm_reinject_control *control)
1758 {
1759 	if (!kvm->arch.vpit)
1760 		return -ENXIO;
1761 	kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1762 	return 0;
1763 }
1764 
1765 /*
1766  * Get (and clear) the dirty memory log for a memory slot.
1767  */
1768 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1769 				      struct kvm_dirty_log *log)
1770 {
1771 	int r;
1772 	int n;
1773 	struct kvm_memory_slot *memslot;
1774 	int is_dirty = 0;
1775 
1776 	down_write(&kvm->slots_lock);
1777 
1778 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1779 	if (r)
1780 		goto out;
1781 
1782 	/* If nothing is dirty, don't bother messing with page tables. */
1783 	if (is_dirty) {
1784 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
1785 		kvm_flush_remote_tlbs(kvm);
1786 		memslot = &kvm->memslots[log->slot];
1787 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1788 		memset(memslot->dirty_bitmap, 0, n);
1789 	}
1790 	r = 0;
1791 out:
1792 	up_write(&kvm->slots_lock);
1793 	return r;
1794 }
1795 
1796 long kvm_arch_vm_ioctl(struct file *filp,
1797 		       unsigned int ioctl, unsigned long arg)
1798 {
1799 	struct kvm *kvm = filp->private_data;
1800 	void __user *argp = (void __user *)arg;
1801 	int r = -EINVAL;
1802 	/*
1803 	 * This union makes it completely explicit to gcc-3.x
1804 	 * that these two variables' stack usage should be
1805 	 * combined, not added together.
1806 	 */
1807 	union {
1808 		struct kvm_pit_state ps;
1809 		struct kvm_memory_alias alias;
1810 	} u;
1811 
1812 	switch (ioctl) {
1813 	case KVM_SET_TSS_ADDR:
1814 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1815 		if (r < 0)
1816 			goto out;
1817 		break;
1818 	case KVM_SET_MEMORY_REGION: {
1819 		struct kvm_memory_region kvm_mem;
1820 		struct kvm_userspace_memory_region kvm_userspace_mem;
1821 
1822 		r = -EFAULT;
1823 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1824 			goto out;
1825 		kvm_userspace_mem.slot = kvm_mem.slot;
1826 		kvm_userspace_mem.flags = kvm_mem.flags;
1827 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1828 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1829 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1830 		if (r)
1831 			goto out;
1832 		break;
1833 	}
1834 	case KVM_SET_NR_MMU_PAGES:
1835 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1836 		if (r)
1837 			goto out;
1838 		break;
1839 	case KVM_GET_NR_MMU_PAGES:
1840 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1841 		break;
1842 	case KVM_SET_MEMORY_ALIAS:
1843 		r = -EFAULT;
1844 		if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1845 			goto out;
1846 		r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1847 		if (r)
1848 			goto out;
1849 		break;
1850 	case KVM_CREATE_IRQCHIP:
1851 		r = -ENOMEM;
1852 		kvm->arch.vpic = kvm_create_pic(kvm);
1853 		if (kvm->arch.vpic) {
1854 			r = kvm_ioapic_init(kvm);
1855 			if (r) {
1856 				kfree(kvm->arch.vpic);
1857 				kvm->arch.vpic = NULL;
1858 				goto out;
1859 			}
1860 		} else
1861 			goto out;
1862 		r = kvm_setup_default_irq_routing(kvm);
1863 		if (r) {
1864 			kfree(kvm->arch.vpic);
1865 			kfree(kvm->arch.vioapic);
1866 			goto out;
1867 		}
1868 		break;
1869 	case KVM_CREATE_PIT:
1870 		mutex_lock(&kvm->lock);
1871 		r = -EEXIST;
1872 		if (kvm->arch.vpit)
1873 			goto create_pit_unlock;
1874 		r = -ENOMEM;
1875 		kvm->arch.vpit = kvm_create_pit(kvm);
1876 		if (kvm->arch.vpit)
1877 			r = 0;
1878 	create_pit_unlock:
1879 		mutex_unlock(&kvm->lock);
1880 		break;
1881 	case KVM_IRQ_LINE_STATUS:
1882 	case KVM_IRQ_LINE: {
1883 		struct kvm_irq_level irq_event;
1884 
1885 		r = -EFAULT;
1886 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
1887 			goto out;
1888 		if (irqchip_in_kernel(kvm)) {
1889 			__s32 status;
1890 			mutex_lock(&kvm->lock);
1891 			status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1892 					irq_event.irq, irq_event.level);
1893 			mutex_unlock(&kvm->lock);
1894 			if (ioctl == KVM_IRQ_LINE_STATUS) {
1895 				irq_event.status = status;
1896 				if (copy_to_user(argp, &irq_event,
1897 							sizeof irq_event))
1898 					goto out;
1899 			}
1900 			r = 0;
1901 		}
1902 		break;
1903 	}
1904 	case KVM_GET_IRQCHIP: {
1905 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1906 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1907 
1908 		r = -ENOMEM;
1909 		if (!chip)
1910 			goto out;
1911 		r = -EFAULT;
1912 		if (copy_from_user(chip, argp, sizeof *chip))
1913 			goto get_irqchip_out;
1914 		r = -ENXIO;
1915 		if (!irqchip_in_kernel(kvm))
1916 			goto get_irqchip_out;
1917 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1918 		if (r)
1919 			goto get_irqchip_out;
1920 		r = -EFAULT;
1921 		if (copy_to_user(argp, chip, sizeof *chip))
1922 			goto get_irqchip_out;
1923 		r = 0;
1924 	get_irqchip_out:
1925 		kfree(chip);
1926 		if (r)
1927 			goto out;
1928 		break;
1929 	}
1930 	case KVM_SET_IRQCHIP: {
1931 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1932 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1933 
1934 		r = -ENOMEM;
1935 		if (!chip)
1936 			goto out;
1937 		r = -EFAULT;
1938 		if (copy_from_user(chip, argp, sizeof *chip))
1939 			goto set_irqchip_out;
1940 		r = -ENXIO;
1941 		if (!irqchip_in_kernel(kvm))
1942 			goto set_irqchip_out;
1943 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1944 		if (r)
1945 			goto set_irqchip_out;
1946 		r = 0;
1947 	set_irqchip_out:
1948 		kfree(chip);
1949 		if (r)
1950 			goto out;
1951 		break;
1952 	}
1953 	case KVM_GET_PIT: {
1954 		r = -EFAULT;
1955 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1956 			goto out;
1957 		r = -ENXIO;
1958 		if (!kvm->arch.vpit)
1959 			goto out;
1960 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1961 		if (r)
1962 			goto out;
1963 		r = -EFAULT;
1964 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1965 			goto out;
1966 		r = 0;
1967 		break;
1968 	}
1969 	case KVM_SET_PIT: {
1970 		r = -EFAULT;
1971 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
1972 			goto out;
1973 		r = -ENXIO;
1974 		if (!kvm->arch.vpit)
1975 			goto out;
1976 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1977 		if (r)
1978 			goto out;
1979 		r = 0;
1980 		break;
1981 	}
1982 	case KVM_REINJECT_CONTROL: {
1983 		struct kvm_reinject_control control;
1984 		r =  -EFAULT;
1985 		if (copy_from_user(&control, argp, sizeof(control)))
1986 			goto out;
1987 		r = kvm_vm_ioctl_reinject(kvm, &control);
1988 		if (r)
1989 			goto out;
1990 		r = 0;
1991 		break;
1992 	}
1993 	default:
1994 		;
1995 	}
1996 out:
1997 	return r;
1998 }
1999 
2000 static void kvm_init_msr_list(void)
2001 {
2002 	u32 dummy[2];
2003 	unsigned i, j;
2004 
2005 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2006 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2007 			continue;
2008 		if (j < i)
2009 			msrs_to_save[j] = msrs_to_save[i];
2010 		j++;
2011 	}
2012 	num_msrs_to_save = j;
2013 }
2014 
2015 /*
2016  * Only apic need an MMIO device hook, so shortcut now..
2017  */
2018 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2019 						gpa_t addr, int len,
2020 						int is_write)
2021 {
2022 	struct kvm_io_device *dev;
2023 
2024 	if (vcpu->arch.apic) {
2025 		dev = &vcpu->arch.apic->dev;
2026 		if (dev->in_range(dev, addr, len, is_write))
2027 			return dev;
2028 	}
2029 	return NULL;
2030 }
2031 
2032 
2033 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2034 						gpa_t addr, int len,
2035 						int is_write)
2036 {
2037 	struct kvm_io_device *dev;
2038 
2039 	dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2040 	if (dev == NULL)
2041 		dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2042 					  is_write);
2043 	return dev;
2044 }
2045 
2046 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2047 			       struct kvm_vcpu *vcpu)
2048 {
2049 	void *data = val;
2050 	int r = X86EMUL_CONTINUE;
2051 
2052 	while (bytes) {
2053 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2054 		unsigned offset = addr & (PAGE_SIZE-1);
2055 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2056 		int ret;
2057 
2058 		if (gpa == UNMAPPED_GVA) {
2059 			r = X86EMUL_PROPAGATE_FAULT;
2060 			goto out;
2061 		}
2062 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2063 		if (ret < 0) {
2064 			r = X86EMUL_UNHANDLEABLE;
2065 			goto out;
2066 		}
2067 
2068 		bytes -= toread;
2069 		data += toread;
2070 		addr += toread;
2071 	}
2072 out:
2073 	return r;
2074 }
2075 
2076 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2077 				struct kvm_vcpu *vcpu)
2078 {
2079 	void *data = val;
2080 	int r = X86EMUL_CONTINUE;
2081 
2082 	while (bytes) {
2083 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2084 		unsigned offset = addr & (PAGE_SIZE-1);
2085 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2086 		int ret;
2087 
2088 		if (gpa == UNMAPPED_GVA) {
2089 			r = X86EMUL_PROPAGATE_FAULT;
2090 			goto out;
2091 		}
2092 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2093 		if (ret < 0) {
2094 			r = X86EMUL_UNHANDLEABLE;
2095 			goto out;
2096 		}
2097 
2098 		bytes -= towrite;
2099 		data += towrite;
2100 		addr += towrite;
2101 	}
2102 out:
2103 	return r;
2104 }
2105 
2106 
2107 static int emulator_read_emulated(unsigned long addr,
2108 				  void *val,
2109 				  unsigned int bytes,
2110 				  struct kvm_vcpu *vcpu)
2111 {
2112 	struct kvm_io_device *mmio_dev;
2113 	gpa_t                 gpa;
2114 
2115 	if (vcpu->mmio_read_completed) {
2116 		memcpy(val, vcpu->mmio_data, bytes);
2117 		vcpu->mmio_read_completed = 0;
2118 		return X86EMUL_CONTINUE;
2119 	}
2120 
2121 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2122 
2123 	/* For APIC access vmexit */
2124 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2125 		goto mmio;
2126 
2127 	if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2128 				== X86EMUL_CONTINUE)
2129 		return X86EMUL_CONTINUE;
2130 	if (gpa == UNMAPPED_GVA)
2131 		return X86EMUL_PROPAGATE_FAULT;
2132 
2133 mmio:
2134 	/*
2135 	 * Is this MMIO handled locally?
2136 	 */
2137 	mutex_lock(&vcpu->kvm->lock);
2138 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2139 	if (mmio_dev) {
2140 		kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2141 		mutex_unlock(&vcpu->kvm->lock);
2142 		return X86EMUL_CONTINUE;
2143 	}
2144 	mutex_unlock(&vcpu->kvm->lock);
2145 
2146 	vcpu->mmio_needed = 1;
2147 	vcpu->mmio_phys_addr = gpa;
2148 	vcpu->mmio_size = bytes;
2149 	vcpu->mmio_is_write = 0;
2150 
2151 	return X86EMUL_UNHANDLEABLE;
2152 }
2153 
2154 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2155 			  const void *val, int bytes)
2156 {
2157 	int ret;
2158 
2159 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2160 	if (ret < 0)
2161 		return 0;
2162 	kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2163 	return 1;
2164 }
2165 
2166 static int emulator_write_emulated_onepage(unsigned long addr,
2167 					   const void *val,
2168 					   unsigned int bytes,
2169 					   struct kvm_vcpu *vcpu)
2170 {
2171 	struct kvm_io_device *mmio_dev;
2172 	gpa_t                 gpa;
2173 
2174 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2175 
2176 	if (gpa == UNMAPPED_GVA) {
2177 		kvm_inject_page_fault(vcpu, addr, 2);
2178 		return X86EMUL_PROPAGATE_FAULT;
2179 	}
2180 
2181 	/* For APIC access vmexit */
2182 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2183 		goto mmio;
2184 
2185 	if (emulator_write_phys(vcpu, gpa, val, bytes))
2186 		return X86EMUL_CONTINUE;
2187 
2188 mmio:
2189 	/*
2190 	 * Is this MMIO handled locally?
2191 	 */
2192 	mutex_lock(&vcpu->kvm->lock);
2193 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2194 	if (mmio_dev) {
2195 		kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2196 		mutex_unlock(&vcpu->kvm->lock);
2197 		return X86EMUL_CONTINUE;
2198 	}
2199 	mutex_unlock(&vcpu->kvm->lock);
2200 
2201 	vcpu->mmio_needed = 1;
2202 	vcpu->mmio_phys_addr = gpa;
2203 	vcpu->mmio_size = bytes;
2204 	vcpu->mmio_is_write = 1;
2205 	memcpy(vcpu->mmio_data, val, bytes);
2206 
2207 	return X86EMUL_CONTINUE;
2208 }
2209 
2210 int emulator_write_emulated(unsigned long addr,
2211 				   const void *val,
2212 				   unsigned int bytes,
2213 				   struct kvm_vcpu *vcpu)
2214 {
2215 	/* Crossing a page boundary? */
2216 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2217 		int rc, now;
2218 
2219 		now = -addr & ~PAGE_MASK;
2220 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2221 		if (rc != X86EMUL_CONTINUE)
2222 			return rc;
2223 		addr += now;
2224 		val += now;
2225 		bytes -= now;
2226 	}
2227 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2228 }
2229 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2230 
2231 static int emulator_cmpxchg_emulated(unsigned long addr,
2232 				     const void *old,
2233 				     const void *new,
2234 				     unsigned int bytes,
2235 				     struct kvm_vcpu *vcpu)
2236 {
2237 	static int reported;
2238 
2239 	if (!reported) {
2240 		reported = 1;
2241 		printk(KERN_WARNING "kvm: emulating exchange as write\n");
2242 	}
2243 #ifndef CONFIG_X86_64
2244 	/* guests cmpxchg8b have to be emulated atomically */
2245 	if (bytes == 8) {
2246 		gpa_t gpa;
2247 		struct page *page;
2248 		char *kaddr;
2249 		u64 val;
2250 
2251 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2252 
2253 		if (gpa == UNMAPPED_GVA ||
2254 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2255 			goto emul_write;
2256 
2257 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2258 			goto emul_write;
2259 
2260 		val = *(u64 *)new;
2261 
2262 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2263 
2264 		kaddr = kmap_atomic(page, KM_USER0);
2265 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2266 		kunmap_atomic(kaddr, KM_USER0);
2267 		kvm_release_page_dirty(page);
2268 	}
2269 emul_write:
2270 #endif
2271 
2272 	return emulator_write_emulated(addr, new, bytes, vcpu);
2273 }
2274 
2275 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2276 {
2277 	return kvm_x86_ops->get_segment_base(vcpu, seg);
2278 }
2279 
2280 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2281 {
2282 	kvm_mmu_invlpg(vcpu, address);
2283 	return X86EMUL_CONTINUE;
2284 }
2285 
2286 int emulate_clts(struct kvm_vcpu *vcpu)
2287 {
2288 	KVMTRACE_0D(CLTS, vcpu, handler);
2289 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2290 	return X86EMUL_CONTINUE;
2291 }
2292 
2293 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2294 {
2295 	struct kvm_vcpu *vcpu = ctxt->vcpu;
2296 
2297 	switch (dr) {
2298 	case 0 ... 3:
2299 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
2300 		return X86EMUL_CONTINUE;
2301 	default:
2302 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2303 		return X86EMUL_UNHANDLEABLE;
2304 	}
2305 }
2306 
2307 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2308 {
2309 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2310 	int exception;
2311 
2312 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2313 	if (exception) {
2314 		/* FIXME: better handling */
2315 		return X86EMUL_UNHANDLEABLE;
2316 	}
2317 	return X86EMUL_CONTINUE;
2318 }
2319 
2320 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2321 {
2322 	u8 opcodes[4];
2323 	unsigned long rip = kvm_rip_read(vcpu);
2324 	unsigned long rip_linear;
2325 
2326 	if (!printk_ratelimit())
2327 		return;
2328 
2329 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2330 
2331 	kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2332 
2333 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2334 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2335 }
2336 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2337 
2338 static struct x86_emulate_ops emulate_ops = {
2339 	.read_std            = kvm_read_guest_virt,
2340 	.read_emulated       = emulator_read_emulated,
2341 	.write_emulated      = emulator_write_emulated,
2342 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2343 };
2344 
2345 static void cache_all_regs(struct kvm_vcpu *vcpu)
2346 {
2347 	kvm_register_read(vcpu, VCPU_REGS_RAX);
2348 	kvm_register_read(vcpu, VCPU_REGS_RSP);
2349 	kvm_register_read(vcpu, VCPU_REGS_RIP);
2350 	vcpu->arch.regs_dirty = ~0;
2351 }
2352 
2353 int emulate_instruction(struct kvm_vcpu *vcpu,
2354 			struct kvm_run *run,
2355 			unsigned long cr2,
2356 			u16 error_code,
2357 			int emulation_type)
2358 {
2359 	int r;
2360 	struct decode_cache *c;
2361 
2362 	kvm_clear_exception_queue(vcpu);
2363 	vcpu->arch.mmio_fault_cr2 = cr2;
2364 	/*
2365 	 * TODO: fix x86_emulate.c to use guest_read/write_register
2366 	 * instead of direct ->regs accesses, can save hundred cycles
2367 	 * on Intel for instructions that don't read/change RSP, for
2368 	 * for example.
2369 	 */
2370 	cache_all_regs(vcpu);
2371 
2372 	vcpu->mmio_is_write = 0;
2373 	vcpu->arch.pio.string = 0;
2374 
2375 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2376 		int cs_db, cs_l;
2377 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2378 
2379 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2380 		vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2381 		vcpu->arch.emulate_ctxt.mode =
2382 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2383 			? X86EMUL_MODE_REAL : cs_l
2384 			? X86EMUL_MODE_PROT64 :	cs_db
2385 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2386 
2387 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2388 
2389 		/* Reject the instructions other than VMCALL/VMMCALL when
2390 		 * try to emulate invalid opcode */
2391 		c = &vcpu->arch.emulate_ctxt.decode;
2392 		if ((emulation_type & EMULTYPE_TRAP_UD) &&
2393 		    (!(c->twobyte && c->b == 0x01 &&
2394 		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2395 		       c->modrm_mod == 3 && c->modrm_rm == 1)))
2396 			return EMULATE_FAIL;
2397 
2398 		++vcpu->stat.insn_emulation;
2399 		if (r)  {
2400 			++vcpu->stat.insn_emulation_fail;
2401 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2402 				return EMULATE_DONE;
2403 			return EMULATE_FAIL;
2404 		}
2405 	}
2406 
2407 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2408 
2409 	if (vcpu->arch.pio.string)
2410 		return EMULATE_DO_MMIO;
2411 
2412 	if ((r || vcpu->mmio_is_write) && run) {
2413 		run->exit_reason = KVM_EXIT_MMIO;
2414 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2415 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2416 		run->mmio.len = vcpu->mmio_size;
2417 		run->mmio.is_write = vcpu->mmio_is_write;
2418 	}
2419 
2420 	if (r) {
2421 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2422 			return EMULATE_DONE;
2423 		if (!vcpu->mmio_needed) {
2424 			kvm_report_emulation_failure(vcpu, "mmio");
2425 			return EMULATE_FAIL;
2426 		}
2427 		return EMULATE_DO_MMIO;
2428 	}
2429 
2430 	kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2431 
2432 	if (vcpu->mmio_is_write) {
2433 		vcpu->mmio_needed = 0;
2434 		return EMULATE_DO_MMIO;
2435 	}
2436 
2437 	return EMULATE_DONE;
2438 }
2439 EXPORT_SYMBOL_GPL(emulate_instruction);
2440 
2441 static int pio_copy_data(struct kvm_vcpu *vcpu)
2442 {
2443 	void *p = vcpu->arch.pio_data;
2444 	gva_t q = vcpu->arch.pio.guest_gva;
2445 	unsigned bytes;
2446 	int ret;
2447 
2448 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2449 	if (vcpu->arch.pio.in)
2450 		ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2451 	else
2452 		ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2453 	return ret;
2454 }
2455 
2456 int complete_pio(struct kvm_vcpu *vcpu)
2457 {
2458 	struct kvm_pio_request *io = &vcpu->arch.pio;
2459 	long delta;
2460 	int r;
2461 	unsigned long val;
2462 
2463 	if (!io->string) {
2464 		if (io->in) {
2465 			val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2466 			memcpy(&val, vcpu->arch.pio_data, io->size);
2467 			kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2468 		}
2469 	} else {
2470 		if (io->in) {
2471 			r = pio_copy_data(vcpu);
2472 			if (r)
2473 				return r;
2474 		}
2475 
2476 		delta = 1;
2477 		if (io->rep) {
2478 			delta *= io->cur_count;
2479 			/*
2480 			 * The size of the register should really depend on
2481 			 * current address size.
2482 			 */
2483 			val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2484 			val -= delta;
2485 			kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2486 		}
2487 		if (io->down)
2488 			delta = -delta;
2489 		delta *= io->size;
2490 		if (io->in) {
2491 			val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2492 			val += delta;
2493 			kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2494 		} else {
2495 			val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2496 			val += delta;
2497 			kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2498 		}
2499 	}
2500 
2501 	io->count -= io->cur_count;
2502 	io->cur_count = 0;
2503 
2504 	return 0;
2505 }
2506 
2507 static void kernel_pio(struct kvm_io_device *pio_dev,
2508 		       struct kvm_vcpu *vcpu,
2509 		       void *pd)
2510 {
2511 	/* TODO: String I/O for in kernel device */
2512 
2513 	mutex_lock(&vcpu->kvm->lock);
2514 	if (vcpu->arch.pio.in)
2515 		kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2516 				  vcpu->arch.pio.size,
2517 				  pd);
2518 	else
2519 		kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2520 				   vcpu->arch.pio.size,
2521 				   pd);
2522 	mutex_unlock(&vcpu->kvm->lock);
2523 }
2524 
2525 static void pio_string_write(struct kvm_io_device *pio_dev,
2526 			     struct kvm_vcpu *vcpu)
2527 {
2528 	struct kvm_pio_request *io = &vcpu->arch.pio;
2529 	void *pd = vcpu->arch.pio_data;
2530 	int i;
2531 
2532 	mutex_lock(&vcpu->kvm->lock);
2533 	for (i = 0; i < io->cur_count; i++) {
2534 		kvm_iodevice_write(pio_dev, io->port,
2535 				   io->size,
2536 				   pd);
2537 		pd += io->size;
2538 	}
2539 	mutex_unlock(&vcpu->kvm->lock);
2540 }
2541 
2542 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2543 					       gpa_t addr, int len,
2544 					       int is_write)
2545 {
2546 	return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2547 }
2548 
2549 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2550 		  int size, unsigned port)
2551 {
2552 	struct kvm_io_device *pio_dev;
2553 	unsigned long val;
2554 
2555 	vcpu->run->exit_reason = KVM_EXIT_IO;
2556 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2557 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2558 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2559 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2560 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2561 	vcpu->arch.pio.in = in;
2562 	vcpu->arch.pio.string = 0;
2563 	vcpu->arch.pio.down = 0;
2564 	vcpu->arch.pio.rep = 0;
2565 
2566 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2567 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2568 			    handler);
2569 	else
2570 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2571 			    handler);
2572 
2573 	val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2574 	memcpy(vcpu->arch.pio_data, &val, 4);
2575 
2576 	pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2577 	if (pio_dev) {
2578 		kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2579 		complete_pio(vcpu);
2580 		return 1;
2581 	}
2582 	return 0;
2583 }
2584 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2585 
2586 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2587 		  int size, unsigned long count, int down,
2588 		  gva_t address, int rep, unsigned port)
2589 {
2590 	unsigned now, in_page;
2591 	int ret = 0;
2592 	struct kvm_io_device *pio_dev;
2593 
2594 	vcpu->run->exit_reason = KVM_EXIT_IO;
2595 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2596 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2597 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2598 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2599 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2600 	vcpu->arch.pio.in = in;
2601 	vcpu->arch.pio.string = 1;
2602 	vcpu->arch.pio.down = down;
2603 	vcpu->arch.pio.rep = rep;
2604 
2605 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2606 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2607 			    handler);
2608 	else
2609 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2610 			    handler);
2611 
2612 	if (!count) {
2613 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2614 		return 1;
2615 	}
2616 
2617 	if (!down)
2618 		in_page = PAGE_SIZE - offset_in_page(address);
2619 	else
2620 		in_page = offset_in_page(address) + size;
2621 	now = min(count, (unsigned long)in_page / size);
2622 	if (!now)
2623 		now = 1;
2624 	if (down) {
2625 		/*
2626 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2627 		 */
2628 		pr_unimpl(vcpu, "guest string pio down\n");
2629 		kvm_inject_gp(vcpu, 0);
2630 		return 1;
2631 	}
2632 	vcpu->run->io.count = now;
2633 	vcpu->arch.pio.cur_count = now;
2634 
2635 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2636 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2637 
2638 	vcpu->arch.pio.guest_gva = address;
2639 
2640 	pio_dev = vcpu_find_pio_dev(vcpu, port,
2641 				    vcpu->arch.pio.cur_count,
2642 				    !vcpu->arch.pio.in);
2643 	if (!vcpu->arch.pio.in) {
2644 		/* string PIO write */
2645 		ret = pio_copy_data(vcpu);
2646 		if (ret == X86EMUL_PROPAGATE_FAULT) {
2647 			kvm_inject_gp(vcpu, 0);
2648 			return 1;
2649 		}
2650 		if (ret == 0 && pio_dev) {
2651 			pio_string_write(pio_dev, vcpu);
2652 			complete_pio(vcpu);
2653 			if (vcpu->arch.pio.count == 0)
2654 				ret = 1;
2655 		}
2656 	} else if (pio_dev)
2657 		pr_unimpl(vcpu, "no string pio read support yet, "
2658 		       "port %x size %d count %ld\n",
2659 			port, size, count);
2660 
2661 	return ret;
2662 }
2663 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2664 
2665 static void bounce_off(void *info)
2666 {
2667 	/* nothing */
2668 }
2669 
2670 static unsigned int  ref_freq;
2671 static unsigned long tsc_khz_ref;
2672 
2673 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2674 				     void *data)
2675 {
2676 	struct cpufreq_freqs *freq = data;
2677 	struct kvm *kvm;
2678 	struct kvm_vcpu *vcpu;
2679 	int i, send_ipi = 0;
2680 
2681 	if (!ref_freq)
2682 		ref_freq = freq->old;
2683 
2684 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2685 		return 0;
2686 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2687 		return 0;
2688 	per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2689 
2690 	spin_lock(&kvm_lock);
2691 	list_for_each_entry(kvm, &vm_list, vm_list) {
2692 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2693 			vcpu = kvm->vcpus[i];
2694 			if (!vcpu)
2695 				continue;
2696 			if (vcpu->cpu != freq->cpu)
2697 				continue;
2698 			if (!kvm_request_guest_time_update(vcpu))
2699 				continue;
2700 			if (vcpu->cpu != smp_processor_id())
2701 				send_ipi++;
2702 		}
2703 	}
2704 	spin_unlock(&kvm_lock);
2705 
2706 	if (freq->old < freq->new && send_ipi) {
2707 		/*
2708 		 * We upscale the frequency.  Must make the guest
2709 		 * doesn't see old kvmclock values while running with
2710 		 * the new frequency, otherwise we risk the guest sees
2711 		 * time go backwards.
2712 		 *
2713 		 * In case we update the frequency for another cpu
2714 		 * (which might be in guest context) send an interrupt
2715 		 * to kick the cpu out of guest context.  Next time
2716 		 * guest context is entered kvmclock will be updated,
2717 		 * so the guest will not see stale values.
2718 		 */
2719 		smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2720 	}
2721 	return 0;
2722 }
2723 
2724 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2725         .notifier_call  = kvmclock_cpufreq_notifier
2726 };
2727 
2728 int kvm_arch_init(void *opaque)
2729 {
2730 	int r, cpu;
2731 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2732 
2733 	if (kvm_x86_ops) {
2734 		printk(KERN_ERR "kvm: already loaded the other module\n");
2735 		r = -EEXIST;
2736 		goto out;
2737 	}
2738 
2739 	if (!ops->cpu_has_kvm_support()) {
2740 		printk(KERN_ERR "kvm: no hardware support\n");
2741 		r = -EOPNOTSUPP;
2742 		goto out;
2743 	}
2744 	if (ops->disabled_by_bios()) {
2745 		printk(KERN_ERR "kvm: disabled by bios\n");
2746 		r = -EOPNOTSUPP;
2747 		goto out;
2748 	}
2749 
2750 	r = kvm_mmu_module_init();
2751 	if (r)
2752 		goto out;
2753 
2754 	kvm_init_msr_list();
2755 
2756 	kvm_x86_ops = ops;
2757 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2758 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2759 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2760 			PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2761 
2762 	for_each_possible_cpu(cpu)
2763 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2764 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2765 		tsc_khz_ref = tsc_khz;
2766 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2767 					  CPUFREQ_TRANSITION_NOTIFIER);
2768 	}
2769 
2770 	return 0;
2771 
2772 out:
2773 	return r;
2774 }
2775 
2776 void kvm_arch_exit(void)
2777 {
2778 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2779 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2780 					    CPUFREQ_TRANSITION_NOTIFIER);
2781 	kvm_x86_ops = NULL;
2782 	kvm_mmu_module_exit();
2783 }
2784 
2785 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2786 {
2787 	++vcpu->stat.halt_exits;
2788 	KVMTRACE_0D(HLT, vcpu, handler);
2789 	if (irqchip_in_kernel(vcpu->kvm)) {
2790 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2791 		return 1;
2792 	} else {
2793 		vcpu->run->exit_reason = KVM_EXIT_HLT;
2794 		return 0;
2795 	}
2796 }
2797 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2798 
2799 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2800 			   unsigned long a1)
2801 {
2802 	if (is_long_mode(vcpu))
2803 		return a0;
2804 	else
2805 		return a0 | ((gpa_t)a1 << 32);
2806 }
2807 
2808 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2809 {
2810 	unsigned long nr, a0, a1, a2, a3, ret;
2811 	int r = 1;
2812 
2813 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2814 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2815 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2816 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2817 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2818 
2819 	KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2820 
2821 	if (!is_long_mode(vcpu)) {
2822 		nr &= 0xFFFFFFFF;
2823 		a0 &= 0xFFFFFFFF;
2824 		a1 &= 0xFFFFFFFF;
2825 		a2 &= 0xFFFFFFFF;
2826 		a3 &= 0xFFFFFFFF;
2827 	}
2828 
2829 	switch (nr) {
2830 	case KVM_HC_VAPIC_POLL_IRQ:
2831 		ret = 0;
2832 		break;
2833 	case KVM_HC_MMU_OP:
2834 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2835 		break;
2836 	default:
2837 		ret = -KVM_ENOSYS;
2838 		break;
2839 	}
2840 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2841 	++vcpu->stat.hypercalls;
2842 	return r;
2843 }
2844 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2845 
2846 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2847 {
2848 	char instruction[3];
2849 	int ret = 0;
2850 	unsigned long rip = kvm_rip_read(vcpu);
2851 
2852 
2853 	/*
2854 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
2855 	 * to ensure that the updated hypercall appears atomically across all
2856 	 * VCPUs.
2857 	 */
2858 	kvm_mmu_zap_all(vcpu->kvm);
2859 
2860 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
2861 	if (emulator_write_emulated(rip, instruction, 3, vcpu)
2862 	    != X86EMUL_CONTINUE)
2863 		ret = -EFAULT;
2864 
2865 	return ret;
2866 }
2867 
2868 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2869 {
2870 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2871 }
2872 
2873 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2874 {
2875 	struct descriptor_table dt = { limit, base };
2876 
2877 	kvm_x86_ops->set_gdt(vcpu, &dt);
2878 }
2879 
2880 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2881 {
2882 	struct descriptor_table dt = { limit, base };
2883 
2884 	kvm_x86_ops->set_idt(vcpu, &dt);
2885 }
2886 
2887 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2888 		   unsigned long *rflags)
2889 {
2890 	kvm_lmsw(vcpu, msw);
2891 	*rflags = kvm_x86_ops->get_rflags(vcpu);
2892 }
2893 
2894 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2895 {
2896 	unsigned long value;
2897 
2898 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2899 	switch (cr) {
2900 	case 0:
2901 		value = vcpu->arch.cr0;
2902 		break;
2903 	case 2:
2904 		value = vcpu->arch.cr2;
2905 		break;
2906 	case 3:
2907 		value = vcpu->arch.cr3;
2908 		break;
2909 	case 4:
2910 		value = vcpu->arch.cr4;
2911 		break;
2912 	case 8:
2913 		value = kvm_get_cr8(vcpu);
2914 		break;
2915 	default:
2916 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2917 		return 0;
2918 	}
2919 	KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2920 		    (u32)((u64)value >> 32), handler);
2921 
2922 	return value;
2923 }
2924 
2925 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2926 		     unsigned long *rflags)
2927 {
2928 	KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2929 		    (u32)((u64)val >> 32), handler);
2930 
2931 	switch (cr) {
2932 	case 0:
2933 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2934 		*rflags = kvm_x86_ops->get_rflags(vcpu);
2935 		break;
2936 	case 2:
2937 		vcpu->arch.cr2 = val;
2938 		break;
2939 	case 3:
2940 		kvm_set_cr3(vcpu, val);
2941 		break;
2942 	case 4:
2943 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2944 		break;
2945 	case 8:
2946 		kvm_set_cr8(vcpu, val & 0xfUL);
2947 		break;
2948 	default:
2949 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2950 	}
2951 }
2952 
2953 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2954 {
2955 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2956 	int j, nent = vcpu->arch.cpuid_nent;
2957 
2958 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2959 	/* when no next entry is found, the current entry[i] is reselected */
2960 	for (j = i + 1; ; j = (j + 1) % nent) {
2961 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2962 		if (ej->function == e->function) {
2963 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2964 			return j;
2965 		}
2966 	}
2967 	return 0; /* silence gcc, even though control never reaches here */
2968 }
2969 
2970 /* find an entry with matching function, matching index (if needed), and that
2971  * should be read next (if it's stateful) */
2972 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2973 	u32 function, u32 index)
2974 {
2975 	if (e->function != function)
2976 		return 0;
2977 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2978 		return 0;
2979 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2980 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2981 		return 0;
2982 	return 1;
2983 }
2984 
2985 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2986 					      u32 function, u32 index)
2987 {
2988 	int i;
2989 	struct kvm_cpuid_entry2 *best = NULL;
2990 
2991 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2992 		struct kvm_cpuid_entry2 *e;
2993 
2994 		e = &vcpu->arch.cpuid_entries[i];
2995 		if (is_matching_cpuid_entry(e, function, index)) {
2996 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2997 				move_to_next_stateful_cpuid_entry(vcpu, i);
2998 			best = e;
2999 			break;
3000 		}
3001 		/*
3002 		 * Both basic or both extended?
3003 		 */
3004 		if (((e->function ^ function) & 0x80000000) == 0)
3005 			if (!best || e->function > best->function)
3006 				best = e;
3007 	}
3008 	return best;
3009 }
3010 
3011 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3012 {
3013 	u32 function, index;
3014 	struct kvm_cpuid_entry2 *best;
3015 
3016 	function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3017 	index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3018 	kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3019 	kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3020 	kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3021 	kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3022 	best = kvm_find_cpuid_entry(vcpu, function, index);
3023 	if (best) {
3024 		kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3025 		kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3026 		kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3027 		kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3028 	}
3029 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3030 	KVMTRACE_5D(CPUID, vcpu, function,
3031 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3032 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3033 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3034 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3035 }
3036 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3037 
3038 /*
3039  * Check if userspace requested an interrupt window, and that the
3040  * interrupt window is open.
3041  *
3042  * No need to exit to userspace if we already have an interrupt queued.
3043  */
3044 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3045 					  struct kvm_run *kvm_run)
3046 {
3047 	return (!vcpu->arch.irq_summary &&
3048 		kvm_run->request_interrupt_window &&
3049 		vcpu->arch.interrupt_window_open &&
3050 		(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3051 }
3052 
3053 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3054 			      struct kvm_run *kvm_run)
3055 {
3056 	kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3057 	kvm_run->cr8 = kvm_get_cr8(vcpu);
3058 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
3059 	if (irqchip_in_kernel(vcpu->kvm))
3060 		kvm_run->ready_for_interrupt_injection = 1;
3061 	else
3062 		kvm_run->ready_for_interrupt_injection =
3063 					(vcpu->arch.interrupt_window_open &&
3064 					 vcpu->arch.irq_summary == 0);
3065 }
3066 
3067 static void vapic_enter(struct kvm_vcpu *vcpu)
3068 {
3069 	struct kvm_lapic *apic = vcpu->arch.apic;
3070 	struct page *page;
3071 
3072 	if (!apic || !apic->vapic_addr)
3073 		return;
3074 
3075 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3076 
3077 	vcpu->arch.apic->vapic_page = page;
3078 }
3079 
3080 static void vapic_exit(struct kvm_vcpu *vcpu)
3081 {
3082 	struct kvm_lapic *apic = vcpu->arch.apic;
3083 
3084 	if (!apic || !apic->vapic_addr)
3085 		return;
3086 
3087 	down_read(&vcpu->kvm->slots_lock);
3088 	kvm_release_page_dirty(apic->vapic_page);
3089 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3090 	up_read(&vcpu->kvm->slots_lock);
3091 }
3092 
3093 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3094 {
3095 	int r;
3096 
3097 	if (vcpu->requests)
3098 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3099 			kvm_mmu_unload(vcpu);
3100 
3101 	r = kvm_mmu_reload(vcpu);
3102 	if (unlikely(r))
3103 		goto out;
3104 
3105 	if (vcpu->requests) {
3106 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3107 			__kvm_migrate_timers(vcpu);
3108 		if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3109 			kvm_write_guest_time(vcpu);
3110 		if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3111 			kvm_mmu_sync_roots(vcpu);
3112 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3113 			kvm_x86_ops->tlb_flush(vcpu);
3114 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3115 				       &vcpu->requests)) {
3116 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3117 			r = 0;
3118 			goto out;
3119 		}
3120 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3121 			kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3122 			r = 0;
3123 			goto out;
3124 		}
3125 	}
3126 
3127 	clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3128 	kvm_inject_pending_timer_irqs(vcpu);
3129 
3130 	preempt_disable();
3131 
3132 	kvm_x86_ops->prepare_guest_switch(vcpu);
3133 	kvm_load_guest_fpu(vcpu);
3134 
3135 	local_irq_disable();
3136 
3137 	if (vcpu->requests || need_resched() || signal_pending(current)) {
3138 		local_irq_enable();
3139 		preempt_enable();
3140 		r = 1;
3141 		goto out;
3142 	}
3143 
3144 	vcpu->guest_mode = 1;
3145 	/*
3146 	 * Make sure that guest_mode assignment won't happen after
3147 	 * testing the pending IRQ vector bitmap.
3148 	 */
3149 	smp_wmb();
3150 
3151 	if (vcpu->arch.exception.pending)
3152 		__queue_exception(vcpu);
3153 	else if (irqchip_in_kernel(vcpu->kvm))
3154 		kvm_x86_ops->inject_pending_irq(vcpu);
3155 	else
3156 		kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3157 
3158 	kvm_lapic_sync_to_vapic(vcpu);
3159 
3160 	up_read(&vcpu->kvm->slots_lock);
3161 
3162 	kvm_guest_enter();
3163 
3164 	get_debugreg(vcpu->arch.host_dr6, 6);
3165 	get_debugreg(vcpu->arch.host_dr7, 7);
3166 	if (unlikely(vcpu->arch.switch_db_regs)) {
3167 		get_debugreg(vcpu->arch.host_db[0], 0);
3168 		get_debugreg(vcpu->arch.host_db[1], 1);
3169 		get_debugreg(vcpu->arch.host_db[2], 2);
3170 		get_debugreg(vcpu->arch.host_db[3], 3);
3171 
3172 		set_debugreg(0, 7);
3173 		set_debugreg(vcpu->arch.eff_db[0], 0);
3174 		set_debugreg(vcpu->arch.eff_db[1], 1);
3175 		set_debugreg(vcpu->arch.eff_db[2], 2);
3176 		set_debugreg(vcpu->arch.eff_db[3], 3);
3177 	}
3178 
3179 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3180 	kvm_x86_ops->run(vcpu, kvm_run);
3181 
3182 	if (unlikely(vcpu->arch.switch_db_regs)) {
3183 		set_debugreg(0, 7);
3184 		set_debugreg(vcpu->arch.host_db[0], 0);
3185 		set_debugreg(vcpu->arch.host_db[1], 1);
3186 		set_debugreg(vcpu->arch.host_db[2], 2);
3187 		set_debugreg(vcpu->arch.host_db[3], 3);
3188 	}
3189 	set_debugreg(vcpu->arch.host_dr6, 6);
3190 	set_debugreg(vcpu->arch.host_dr7, 7);
3191 
3192 	vcpu->guest_mode = 0;
3193 	local_irq_enable();
3194 
3195 	++vcpu->stat.exits;
3196 
3197 	/*
3198 	 * We must have an instruction between local_irq_enable() and
3199 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3200 	 * the interrupt shadow.  The stat.exits increment will do nicely.
3201 	 * But we need to prevent reordering, hence this barrier():
3202 	 */
3203 	barrier();
3204 
3205 	kvm_guest_exit();
3206 
3207 	preempt_enable();
3208 
3209 	down_read(&vcpu->kvm->slots_lock);
3210 
3211 	/*
3212 	 * Profile KVM exit RIPs:
3213 	 */
3214 	if (unlikely(prof_on == KVM_PROFILING)) {
3215 		unsigned long rip = kvm_rip_read(vcpu);
3216 		profile_hit(KVM_PROFILING, (void *)rip);
3217 	}
3218 
3219 	if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3220 		vcpu->arch.exception.pending = false;
3221 
3222 	kvm_lapic_sync_from_vapic(vcpu);
3223 
3224 	r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3225 out:
3226 	return r;
3227 }
3228 
3229 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3230 {
3231 	int r;
3232 
3233 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3234 		pr_debug("vcpu %d received sipi with vector # %x\n",
3235 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3236 		kvm_lapic_reset(vcpu);
3237 		r = kvm_arch_vcpu_reset(vcpu);
3238 		if (r)
3239 			return r;
3240 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3241 	}
3242 
3243 	down_read(&vcpu->kvm->slots_lock);
3244 	vapic_enter(vcpu);
3245 
3246 	r = 1;
3247 	while (r > 0) {
3248 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3249 			r = vcpu_enter_guest(vcpu, kvm_run);
3250 		else {
3251 			up_read(&vcpu->kvm->slots_lock);
3252 			kvm_vcpu_block(vcpu);
3253 			down_read(&vcpu->kvm->slots_lock);
3254 			if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3255 				if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3256 					vcpu->arch.mp_state =
3257 							KVM_MP_STATE_RUNNABLE;
3258 			if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3259 				r = -EINTR;
3260 		}
3261 
3262 		if (r > 0) {
3263 			if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3264 				r = -EINTR;
3265 				kvm_run->exit_reason = KVM_EXIT_INTR;
3266 				++vcpu->stat.request_irq_exits;
3267 			}
3268 			if (signal_pending(current)) {
3269 				r = -EINTR;
3270 				kvm_run->exit_reason = KVM_EXIT_INTR;
3271 				++vcpu->stat.signal_exits;
3272 			}
3273 			if (need_resched()) {
3274 				up_read(&vcpu->kvm->slots_lock);
3275 				kvm_resched(vcpu);
3276 				down_read(&vcpu->kvm->slots_lock);
3277 			}
3278 		}
3279 	}
3280 
3281 	up_read(&vcpu->kvm->slots_lock);
3282 	post_kvm_run_save(vcpu, kvm_run);
3283 
3284 	vapic_exit(vcpu);
3285 
3286 	return r;
3287 }
3288 
3289 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3290 {
3291 	int r;
3292 	sigset_t sigsaved;
3293 
3294 	vcpu_load(vcpu);
3295 
3296 	if (vcpu->sigset_active)
3297 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3298 
3299 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3300 		kvm_vcpu_block(vcpu);
3301 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3302 		r = -EAGAIN;
3303 		goto out;
3304 	}
3305 
3306 	/* re-sync apic's tpr */
3307 	if (!irqchip_in_kernel(vcpu->kvm))
3308 		kvm_set_cr8(vcpu, kvm_run->cr8);
3309 
3310 	if (vcpu->arch.pio.cur_count) {
3311 		r = complete_pio(vcpu);
3312 		if (r)
3313 			goto out;
3314 	}
3315 #if CONFIG_HAS_IOMEM
3316 	if (vcpu->mmio_needed) {
3317 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3318 		vcpu->mmio_read_completed = 1;
3319 		vcpu->mmio_needed = 0;
3320 
3321 		down_read(&vcpu->kvm->slots_lock);
3322 		r = emulate_instruction(vcpu, kvm_run,
3323 					vcpu->arch.mmio_fault_cr2, 0,
3324 					EMULTYPE_NO_DECODE);
3325 		up_read(&vcpu->kvm->slots_lock);
3326 		if (r == EMULATE_DO_MMIO) {
3327 			/*
3328 			 * Read-modify-write.  Back to userspace.
3329 			 */
3330 			r = 0;
3331 			goto out;
3332 		}
3333 	}
3334 #endif
3335 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3336 		kvm_register_write(vcpu, VCPU_REGS_RAX,
3337 				     kvm_run->hypercall.ret);
3338 
3339 	r = __vcpu_run(vcpu, kvm_run);
3340 
3341 out:
3342 	if (vcpu->sigset_active)
3343 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3344 
3345 	vcpu_put(vcpu);
3346 	return r;
3347 }
3348 
3349 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3350 {
3351 	vcpu_load(vcpu);
3352 
3353 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3354 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3355 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3356 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3357 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3358 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3359 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3360 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3361 #ifdef CONFIG_X86_64
3362 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3363 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3364 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3365 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3366 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3367 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3368 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3369 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3370 #endif
3371 
3372 	regs->rip = kvm_rip_read(vcpu);
3373 	regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3374 
3375 	/*
3376 	 * Don't leak debug flags in case they were set for guest debugging
3377 	 */
3378 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3379 		regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3380 
3381 	vcpu_put(vcpu);
3382 
3383 	return 0;
3384 }
3385 
3386 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3387 {
3388 	vcpu_load(vcpu);
3389 
3390 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3391 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3392 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3393 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3394 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3395 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3396 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3397 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3398 #ifdef CONFIG_X86_64
3399 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3400 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3401 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3402 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3403 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3404 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3405 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3406 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3407 
3408 #endif
3409 
3410 	kvm_rip_write(vcpu, regs->rip);
3411 	kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3412 
3413 
3414 	vcpu->arch.exception.pending = false;
3415 
3416 	vcpu_put(vcpu);
3417 
3418 	return 0;
3419 }
3420 
3421 void kvm_get_segment(struct kvm_vcpu *vcpu,
3422 		     struct kvm_segment *var, int seg)
3423 {
3424 	kvm_x86_ops->get_segment(vcpu, var, seg);
3425 }
3426 
3427 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3428 {
3429 	struct kvm_segment cs;
3430 
3431 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3432 	*db = cs.db;
3433 	*l = cs.l;
3434 }
3435 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3436 
3437 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3438 				  struct kvm_sregs *sregs)
3439 {
3440 	struct descriptor_table dt;
3441 	int pending_vec;
3442 
3443 	vcpu_load(vcpu);
3444 
3445 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3446 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3447 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3448 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3449 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3450 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3451 
3452 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3453 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3454 
3455 	kvm_x86_ops->get_idt(vcpu, &dt);
3456 	sregs->idt.limit = dt.limit;
3457 	sregs->idt.base = dt.base;
3458 	kvm_x86_ops->get_gdt(vcpu, &dt);
3459 	sregs->gdt.limit = dt.limit;
3460 	sregs->gdt.base = dt.base;
3461 
3462 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3463 	sregs->cr0 = vcpu->arch.cr0;
3464 	sregs->cr2 = vcpu->arch.cr2;
3465 	sregs->cr3 = vcpu->arch.cr3;
3466 	sregs->cr4 = vcpu->arch.cr4;
3467 	sregs->cr8 = kvm_get_cr8(vcpu);
3468 	sregs->efer = vcpu->arch.shadow_efer;
3469 	sregs->apic_base = kvm_get_apic_base(vcpu);
3470 
3471 	if (irqchip_in_kernel(vcpu->kvm)) {
3472 		memset(sregs->interrupt_bitmap, 0,
3473 		       sizeof sregs->interrupt_bitmap);
3474 		pending_vec = kvm_x86_ops->get_irq(vcpu);
3475 		if (pending_vec >= 0)
3476 			set_bit(pending_vec,
3477 				(unsigned long *)sregs->interrupt_bitmap);
3478 	} else
3479 		memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3480 		       sizeof sregs->interrupt_bitmap);
3481 
3482 	vcpu_put(vcpu);
3483 
3484 	return 0;
3485 }
3486 
3487 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3488 				    struct kvm_mp_state *mp_state)
3489 {
3490 	vcpu_load(vcpu);
3491 	mp_state->mp_state = vcpu->arch.mp_state;
3492 	vcpu_put(vcpu);
3493 	return 0;
3494 }
3495 
3496 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3497 				    struct kvm_mp_state *mp_state)
3498 {
3499 	vcpu_load(vcpu);
3500 	vcpu->arch.mp_state = mp_state->mp_state;
3501 	vcpu_put(vcpu);
3502 	return 0;
3503 }
3504 
3505 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3506 			struct kvm_segment *var, int seg)
3507 {
3508 	kvm_x86_ops->set_segment(vcpu, var, seg);
3509 }
3510 
3511 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3512 				   struct kvm_segment *kvm_desct)
3513 {
3514 	kvm_desct->base = seg_desc->base0;
3515 	kvm_desct->base |= seg_desc->base1 << 16;
3516 	kvm_desct->base |= seg_desc->base2 << 24;
3517 	kvm_desct->limit = seg_desc->limit0;
3518 	kvm_desct->limit |= seg_desc->limit << 16;
3519 	if (seg_desc->g) {
3520 		kvm_desct->limit <<= 12;
3521 		kvm_desct->limit |= 0xfff;
3522 	}
3523 	kvm_desct->selector = selector;
3524 	kvm_desct->type = seg_desc->type;
3525 	kvm_desct->present = seg_desc->p;
3526 	kvm_desct->dpl = seg_desc->dpl;
3527 	kvm_desct->db = seg_desc->d;
3528 	kvm_desct->s = seg_desc->s;
3529 	kvm_desct->l = seg_desc->l;
3530 	kvm_desct->g = seg_desc->g;
3531 	kvm_desct->avl = seg_desc->avl;
3532 	if (!selector)
3533 		kvm_desct->unusable = 1;
3534 	else
3535 		kvm_desct->unusable = 0;
3536 	kvm_desct->padding = 0;
3537 }
3538 
3539 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3540 					  u16 selector,
3541 					  struct descriptor_table *dtable)
3542 {
3543 	if (selector & 1 << 2) {
3544 		struct kvm_segment kvm_seg;
3545 
3546 		kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3547 
3548 		if (kvm_seg.unusable)
3549 			dtable->limit = 0;
3550 		else
3551 			dtable->limit = kvm_seg.limit;
3552 		dtable->base = kvm_seg.base;
3553 	}
3554 	else
3555 		kvm_x86_ops->get_gdt(vcpu, dtable);
3556 }
3557 
3558 /* allowed just for 8 bytes segments */
3559 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3560 					 struct desc_struct *seg_desc)
3561 {
3562 	gpa_t gpa;
3563 	struct descriptor_table dtable;
3564 	u16 index = selector >> 3;
3565 
3566 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
3567 
3568 	if (dtable.limit < index * 8 + 7) {
3569 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3570 		return 1;
3571 	}
3572 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3573 	gpa += index * 8;
3574 	return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3575 }
3576 
3577 /* allowed just for 8 bytes segments */
3578 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3579 					 struct desc_struct *seg_desc)
3580 {
3581 	gpa_t gpa;
3582 	struct descriptor_table dtable;
3583 	u16 index = selector >> 3;
3584 
3585 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
3586 
3587 	if (dtable.limit < index * 8 + 7)
3588 		return 1;
3589 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3590 	gpa += index * 8;
3591 	return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3592 }
3593 
3594 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3595 			     struct desc_struct *seg_desc)
3596 {
3597 	u32 base_addr;
3598 
3599 	base_addr = seg_desc->base0;
3600 	base_addr |= (seg_desc->base1 << 16);
3601 	base_addr |= (seg_desc->base2 << 24);
3602 
3603 	return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3604 }
3605 
3606 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3607 {
3608 	struct kvm_segment kvm_seg;
3609 
3610 	kvm_get_segment(vcpu, &kvm_seg, seg);
3611 	return kvm_seg.selector;
3612 }
3613 
3614 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3615 						u16 selector,
3616 						struct kvm_segment *kvm_seg)
3617 {
3618 	struct desc_struct seg_desc;
3619 
3620 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3621 		return 1;
3622 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3623 	return 0;
3624 }
3625 
3626 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3627 {
3628 	struct kvm_segment segvar = {
3629 		.base = selector << 4,
3630 		.limit = 0xffff,
3631 		.selector = selector,
3632 		.type = 3,
3633 		.present = 1,
3634 		.dpl = 3,
3635 		.db = 0,
3636 		.s = 1,
3637 		.l = 0,
3638 		.g = 0,
3639 		.avl = 0,
3640 		.unusable = 0,
3641 	};
3642 	kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3643 	return 0;
3644 }
3645 
3646 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3647 				int type_bits, int seg)
3648 {
3649 	struct kvm_segment kvm_seg;
3650 
3651 	if (!(vcpu->arch.cr0 & X86_CR0_PE))
3652 		return kvm_load_realmode_segment(vcpu, selector, seg);
3653 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3654 		return 1;
3655 	kvm_seg.type |= type_bits;
3656 
3657 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3658 	    seg != VCPU_SREG_LDTR)
3659 		if (!kvm_seg.s)
3660 			kvm_seg.unusable = 1;
3661 
3662 	kvm_set_segment(vcpu, &kvm_seg, seg);
3663 	return 0;
3664 }
3665 
3666 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3667 				struct tss_segment_32 *tss)
3668 {
3669 	tss->cr3 = vcpu->arch.cr3;
3670 	tss->eip = kvm_rip_read(vcpu);
3671 	tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3672 	tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3673 	tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3674 	tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3675 	tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3676 	tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3677 	tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3678 	tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3679 	tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3680 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3681 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3682 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3683 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3684 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3685 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3686 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3687 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3688 }
3689 
3690 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3691 				  struct tss_segment_32 *tss)
3692 {
3693 	kvm_set_cr3(vcpu, tss->cr3);
3694 
3695 	kvm_rip_write(vcpu, tss->eip);
3696 	kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3697 
3698 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3699 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3700 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3701 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3702 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3703 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3704 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3705 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3706 
3707 	if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3708 		return 1;
3709 
3710 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3711 		return 1;
3712 
3713 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3714 		return 1;
3715 
3716 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3717 		return 1;
3718 
3719 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3720 		return 1;
3721 
3722 	if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3723 		return 1;
3724 
3725 	if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3726 		return 1;
3727 	return 0;
3728 }
3729 
3730 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3731 				struct tss_segment_16 *tss)
3732 {
3733 	tss->ip = kvm_rip_read(vcpu);
3734 	tss->flag = kvm_x86_ops->get_rflags(vcpu);
3735 	tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3736 	tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3737 	tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3738 	tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3739 	tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3740 	tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3741 	tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3742 	tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3743 
3744 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3745 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3746 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3747 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3748 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3749 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3750 }
3751 
3752 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3753 				 struct tss_segment_16 *tss)
3754 {
3755 	kvm_rip_write(vcpu, tss->ip);
3756 	kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3757 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3758 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3759 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3760 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3761 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3762 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3763 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3764 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3765 
3766 	if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3767 		return 1;
3768 
3769 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3770 		return 1;
3771 
3772 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3773 		return 1;
3774 
3775 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3776 		return 1;
3777 
3778 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3779 		return 1;
3780 	return 0;
3781 }
3782 
3783 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3784 		       u32 old_tss_base,
3785 		       struct desc_struct *nseg_desc)
3786 {
3787 	struct tss_segment_16 tss_segment_16;
3788 	int ret = 0;
3789 
3790 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3791 			   sizeof tss_segment_16))
3792 		goto out;
3793 
3794 	save_state_to_tss16(vcpu, &tss_segment_16);
3795 
3796 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3797 			    sizeof tss_segment_16))
3798 		goto out;
3799 
3800 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3801 			   &tss_segment_16, sizeof tss_segment_16))
3802 		goto out;
3803 
3804 	if (load_state_from_tss16(vcpu, &tss_segment_16))
3805 		goto out;
3806 
3807 	ret = 1;
3808 out:
3809 	return ret;
3810 }
3811 
3812 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3813 		       u32 old_tss_base,
3814 		       struct desc_struct *nseg_desc)
3815 {
3816 	struct tss_segment_32 tss_segment_32;
3817 	int ret = 0;
3818 
3819 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3820 			   sizeof tss_segment_32))
3821 		goto out;
3822 
3823 	save_state_to_tss32(vcpu, &tss_segment_32);
3824 
3825 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3826 			    sizeof tss_segment_32))
3827 		goto out;
3828 
3829 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3830 			   &tss_segment_32, sizeof tss_segment_32))
3831 		goto out;
3832 
3833 	if (load_state_from_tss32(vcpu, &tss_segment_32))
3834 		goto out;
3835 
3836 	ret = 1;
3837 out:
3838 	return ret;
3839 }
3840 
3841 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3842 {
3843 	struct kvm_segment tr_seg;
3844 	struct desc_struct cseg_desc;
3845 	struct desc_struct nseg_desc;
3846 	int ret = 0;
3847 	u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3848 	u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3849 
3850 	old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3851 
3852 	/* FIXME: Handle errors. Failure to read either TSS or their
3853 	 * descriptors should generate a pagefault.
3854 	 */
3855 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3856 		goto out;
3857 
3858 	if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3859 		goto out;
3860 
3861 	if (reason != TASK_SWITCH_IRET) {
3862 		int cpl;
3863 
3864 		cpl = kvm_x86_ops->get_cpl(vcpu);
3865 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3866 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3867 			return 1;
3868 		}
3869 	}
3870 
3871 	if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3872 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3873 		return 1;
3874 	}
3875 
3876 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3877 		cseg_desc.type &= ~(1 << 1); //clear the B flag
3878 		save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3879 	}
3880 
3881 	if (reason == TASK_SWITCH_IRET) {
3882 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3883 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3884 	}
3885 
3886 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3887 
3888 	if (nseg_desc.type & 8)
3889 		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3890 					 &nseg_desc);
3891 	else
3892 		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3893 					 &nseg_desc);
3894 
3895 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3896 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3897 		kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3898 	}
3899 
3900 	if (reason != TASK_SWITCH_IRET) {
3901 		nseg_desc.type |= (1 << 1);
3902 		save_guest_segment_descriptor(vcpu, tss_selector,
3903 					      &nseg_desc);
3904 	}
3905 
3906 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3907 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3908 	tr_seg.type = 11;
3909 	kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3910 out:
3911 	return ret;
3912 }
3913 EXPORT_SYMBOL_GPL(kvm_task_switch);
3914 
3915 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3916 				  struct kvm_sregs *sregs)
3917 {
3918 	int mmu_reset_needed = 0;
3919 	int i, pending_vec, max_bits;
3920 	struct descriptor_table dt;
3921 
3922 	vcpu_load(vcpu);
3923 
3924 	dt.limit = sregs->idt.limit;
3925 	dt.base = sregs->idt.base;
3926 	kvm_x86_ops->set_idt(vcpu, &dt);
3927 	dt.limit = sregs->gdt.limit;
3928 	dt.base = sregs->gdt.base;
3929 	kvm_x86_ops->set_gdt(vcpu, &dt);
3930 
3931 	vcpu->arch.cr2 = sregs->cr2;
3932 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3933 	vcpu->arch.cr3 = sregs->cr3;
3934 
3935 	kvm_set_cr8(vcpu, sregs->cr8);
3936 
3937 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3938 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
3939 	kvm_set_apic_base(vcpu, sregs->apic_base);
3940 
3941 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3942 
3943 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3944 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3945 	vcpu->arch.cr0 = sregs->cr0;
3946 
3947 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3948 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3949 	if (!is_long_mode(vcpu) && is_pae(vcpu))
3950 		load_pdptrs(vcpu, vcpu->arch.cr3);
3951 
3952 	if (mmu_reset_needed)
3953 		kvm_mmu_reset_context(vcpu);
3954 
3955 	if (!irqchip_in_kernel(vcpu->kvm)) {
3956 		memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3957 		       sizeof vcpu->arch.irq_pending);
3958 		vcpu->arch.irq_summary = 0;
3959 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3960 			if (vcpu->arch.irq_pending[i])
3961 				__set_bit(i, &vcpu->arch.irq_summary);
3962 	} else {
3963 		max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3964 		pending_vec = find_first_bit(
3965 			(const unsigned long *)sregs->interrupt_bitmap,
3966 			max_bits);
3967 		/* Only pending external irq is handled here */
3968 		if (pending_vec < max_bits) {
3969 			kvm_x86_ops->set_irq(vcpu, pending_vec);
3970 			pr_debug("Set back pending irq %d\n",
3971 				 pending_vec);
3972 		}
3973 		kvm_pic_clear_isr_ack(vcpu->kvm);
3974 	}
3975 
3976 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3977 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3978 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3979 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3980 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3981 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3982 
3983 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3984 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3985 
3986 	/* Older userspace won't unhalt the vcpu on reset. */
3987 	if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3988 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3989 	    !(vcpu->arch.cr0 & X86_CR0_PE))
3990 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3991 
3992 	vcpu_put(vcpu);
3993 
3994 	return 0;
3995 }
3996 
3997 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
3998 					struct kvm_guest_debug *dbg)
3999 {
4000 	int i, r;
4001 
4002 	vcpu_load(vcpu);
4003 
4004 	if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4005 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4006 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
4007 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4008 		vcpu->arch.switch_db_regs =
4009 			(dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4010 	} else {
4011 		for (i = 0; i < KVM_NR_DB_REGS; i++)
4012 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4013 		vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4014 	}
4015 
4016 	r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4017 
4018 	if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4019 		kvm_queue_exception(vcpu, DB_VECTOR);
4020 	else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4021 		kvm_queue_exception(vcpu, BP_VECTOR);
4022 
4023 	vcpu_put(vcpu);
4024 
4025 	return r;
4026 }
4027 
4028 /*
4029  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4030  * we have asm/x86/processor.h
4031  */
4032 struct fxsave {
4033 	u16	cwd;
4034 	u16	swd;
4035 	u16	twd;
4036 	u16	fop;
4037 	u64	rip;
4038 	u64	rdp;
4039 	u32	mxcsr;
4040 	u32	mxcsr_mask;
4041 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
4042 #ifdef CONFIG_X86_64
4043 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
4044 #else
4045 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
4046 #endif
4047 };
4048 
4049 /*
4050  * Translate a guest virtual address to a guest physical address.
4051  */
4052 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4053 				    struct kvm_translation *tr)
4054 {
4055 	unsigned long vaddr = tr->linear_address;
4056 	gpa_t gpa;
4057 
4058 	vcpu_load(vcpu);
4059 	down_read(&vcpu->kvm->slots_lock);
4060 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4061 	up_read(&vcpu->kvm->slots_lock);
4062 	tr->physical_address = gpa;
4063 	tr->valid = gpa != UNMAPPED_GVA;
4064 	tr->writeable = 1;
4065 	tr->usermode = 0;
4066 	vcpu_put(vcpu);
4067 
4068 	return 0;
4069 }
4070 
4071 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4072 {
4073 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4074 
4075 	vcpu_load(vcpu);
4076 
4077 	memcpy(fpu->fpr, fxsave->st_space, 128);
4078 	fpu->fcw = fxsave->cwd;
4079 	fpu->fsw = fxsave->swd;
4080 	fpu->ftwx = fxsave->twd;
4081 	fpu->last_opcode = fxsave->fop;
4082 	fpu->last_ip = fxsave->rip;
4083 	fpu->last_dp = fxsave->rdp;
4084 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4085 
4086 	vcpu_put(vcpu);
4087 
4088 	return 0;
4089 }
4090 
4091 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4092 {
4093 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4094 
4095 	vcpu_load(vcpu);
4096 
4097 	memcpy(fxsave->st_space, fpu->fpr, 128);
4098 	fxsave->cwd = fpu->fcw;
4099 	fxsave->swd = fpu->fsw;
4100 	fxsave->twd = fpu->ftwx;
4101 	fxsave->fop = fpu->last_opcode;
4102 	fxsave->rip = fpu->last_ip;
4103 	fxsave->rdp = fpu->last_dp;
4104 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4105 
4106 	vcpu_put(vcpu);
4107 
4108 	return 0;
4109 }
4110 
4111 void fx_init(struct kvm_vcpu *vcpu)
4112 {
4113 	unsigned after_mxcsr_mask;
4114 
4115 	/*
4116 	 * Touch the fpu the first time in non atomic context as if
4117 	 * this is the first fpu instruction the exception handler
4118 	 * will fire before the instruction returns and it'll have to
4119 	 * allocate ram with GFP_KERNEL.
4120 	 */
4121 	if (!used_math())
4122 		kvm_fx_save(&vcpu->arch.host_fx_image);
4123 
4124 	/* Initialize guest FPU by resetting ours and saving into guest's */
4125 	preempt_disable();
4126 	kvm_fx_save(&vcpu->arch.host_fx_image);
4127 	kvm_fx_finit();
4128 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4129 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4130 	preempt_enable();
4131 
4132 	vcpu->arch.cr0 |= X86_CR0_ET;
4133 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4134 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4135 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4136 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4137 }
4138 EXPORT_SYMBOL_GPL(fx_init);
4139 
4140 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4141 {
4142 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4143 		return;
4144 
4145 	vcpu->guest_fpu_loaded = 1;
4146 	kvm_fx_save(&vcpu->arch.host_fx_image);
4147 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
4148 }
4149 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4150 
4151 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4152 {
4153 	if (!vcpu->guest_fpu_loaded)
4154 		return;
4155 
4156 	vcpu->guest_fpu_loaded = 0;
4157 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4158 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4159 	++vcpu->stat.fpu_reload;
4160 }
4161 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4162 
4163 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4164 {
4165 	if (vcpu->arch.time_page) {
4166 		kvm_release_page_dirty(vcpu->arch.time_page);
4167 		vcpu->arch.time_page = NULL;
4168 	}
4169 
4170 	kvm_x86_ops->vcpu_free(vcpu);
4171 }
4172 
4173 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4174 						unsigned int id)
4175 {
4176 	return kvm_x86_ops->vcpu_create(kvm, id);
4177 }
4178 
4179 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4180 {
4181 	int r;
4182 
4183 	/* We do fxsave: this must be aligned. */
4184 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4185 
4186 	vcpu->arch.mtrr_state.have_fixed = 1;
4187 	vcpu_load(vcpu);
4188 	r = kvm_arch_vcpu_reset(vcpu);
4189 	if (r == 0)
4190 		r = kvm_mmu_setup(vcpu);
4191 	vcpu_put(vcpu);
4192 	if (r < 0)
4193 		goto free_vcpu;
4194 
4195 	return 0;
4196 free_vcpu:
4197 	kvm_x86_ops->vcpu_free(vcpu);
4198 	return r;
4199 }
4200 
4201 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4202 {
4203 	vcpu_load(vcpu);
4204 	kvm_mmu_unload(vcpu);
4205 	vcpu_put(vcpu);
4206 
4207 	kvm_x86_ops->vcpu_free(vcpu);
4208 }
4209 
4210 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4211 {
4212 	vcpu->arch.nmi_pending = false;
4213 	vcpu->arch.nmi_injected = false;
4214 
4215 	vcpu->arch.switch_db_regs = 0;
4216 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4217 	vcpu->arch.dr6 = DR6_FIXED_1;
4218 	vcpu->arch.dr7 = DR7_FIXED_1;
4219 
4220 	return kvm_x86_ops->vcpu_reset(vcpu);
4221 }
4222 
4223 void kvm_arch_hardware_enable(void *garbage)
4224 {
4225 	kvm_x86_ops->hardware_enable(garbage);
4226 }
4227 
4228 void kvm_arch_hardware_disable(void *garbage)
4229 {
4230 	kvm_x86_ops->hardware_disable(garbage);
4231 }
4232 
4233 int kvm_arch_hardware_setup(void)
4234 {
4235 	return kvm_x86_ops->hardware_setup();
4236 }
4237 
4238 void kvm_arch_hardware_unsetup(void)
4239 {
4240 	kvm_x86_ops->hardware_unsetup();
4241 }
4242 
4243 void kvm_arch_check_processor_compat(void *rtn)
4244 {
4245 	kvm_x86_ops->check_processor_compatibility(rtn);
4246 }
4247 
4248 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4249 {
4250 	struct page *page;
4251 	struct kvm *kvm;
4252 	int r;
4253 
4254 	BUG_ON(vcpu->kvm == NULL);
4255 	kvm = vcpu->kvm;
4256 
4257 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4258 	if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4259 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4260 	else
4261 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4262 
4263 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4264 	if (!page) {
4265 		r = -ENOMEM;
4266 		goto fail;
4267 	}
4268 	vcpu->arch.pio_data = page_address(page);
4269 
4270 	r = kvm_mmu_create(vcpu);
4271 	if (r < 0)
4272 		goto fail_free_pio_data;
4273 
4274 	if (irqchip_in_kernel(kvm)) {
4275 		r = kvm_create_lapic(vcpu);
4276 		if (r < 0)
4277 			goto fail_mmu_destroy;
4278 	}
4279 
4280 	return 0;
4281 
4282 fail_mmu_destroy:
4283 	kvm_mmu_destroy(vcpu);
4284 fail_free_pio_data:
4285 	free_page((unsigned long)vcpu->arch.pio_data);
4286 fail:
4287 	return r;
4288 }
4289 
4290 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4291 {
4292 	kvm_free_lapic(vcpu);
4293 	down_read(&vcpu->kvm->slots_lock);
4294 	kvm_mmu_destroy(vcpu);
4295 	up_read(&vcpu->kvm->slots_lock);
4296 	free_page((unsigned long)vcpu->arch.pio_data);
4297 }
4298 
4299 struct  kvm *kvm_arch_create_vm(void)
4300 {
4301 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4302 
4303 	if (!kvm)
4304 		return ERR_PTR(-ENOMEM);
4305 
4306 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4307 	INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4308 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4309 
4310 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4311 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4312 
4313 	rdtscll(kvm->arch.vm_init_tsc);
4314 
4315 	return kvm;
4316 }
4317 
4318 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4319 {
4320 	vcpu_load(vcpu);
4321 	kvm_mmu_unload(vcpu);
4322 	vcpu_put(vcpu);
4323 }
4324 
4325 static void kvm_free_vcpus(struct kvm *kvm)
4326 {
4327 	unsigned int i;
4328 
4329 	/*
4330 	 * Unpin any mmu pages first.
4331 	 */
4332 	for (i = 0; i < KVM_MAX_VCPUS; ++i)
4333 		if (kvm->vcpus[i])
4334 			kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4335 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4336 		if (kvm->vcpus[i]) {
4337 			kvm_arch_vcpu_free(kvm->vcpus[i]);
4338 			kvm->vcpus[i] = NULL;
4339 		}
4340 	}
4341 
4342 }
4343 
4344 void kvm_arch_sync_events(struct kvm *kvm)
4345 {
4346 	kvm_free_all_assigned_devices(kvm);
4347 }
4348 
4349 void kvm_arch_destroy_vm(struct kvm *kvm)
4350 {
4351 	kvm_iommu_unmap_guest(kvm);
4352 	kvm_free_pit(kvm);
4353 	kfree(kvm->arch.vpic);
4354 	kfree(kvm->arch.vioapic);
4355 	kvm_free_vcpus(kvm);
4356 	kvm_free_physmem(kvm);
4357 	if (kvm->arch.apic_access_page)
4358 		put_page(kvm->arch.apic_access_page);
4359 	if (kvm->arch.ept_identity_pagetable)
4360 		put_page(kvm->arch.ept_identity_pagetable);
4361 	kfree(kvm);
4362 }
4363 
4364 int kvm_arch_set_memory_region(struct kvm *kvm,
4365 				struct kvm_userspace_memory_region *mem,
4366 				struct kvm_memory_slot old,
4367 				int user_alloc)
4368 {
4369 	int npages = mem->memory_size >> PAGE_SHIFT;
4370 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4371 
4372 	/*To keep backward compatibility with older userspace,
4373 	 *x86 needs to hanlde !user_alloc case.
4374 	 */
4375 	if (!user_alloc) {
4376 		if (npages && !old.rmap) {
4377 			unsigned long userspace_addr;
4378 
4379 			down_write(&current->mm->mmap_sem);
4380 			userspace_addr = do_mmap(NULL, 0,
4381 						 npages * PAGE_SIZE,
4382 						 PROT_READ | PROT_WRITE,
4383 						 MAP_PRIVATE | MAP_ANONYMOUS,
4384 						 0);
4385 			up_write(&current->mm->mmap_sem);
4386 
4387 			if (IS_ERR((void *)userspace_addr))
4388 				return PTR_ERR((void *)userspace_addr);
4389 
4390 			/* set userspace_addr atomically for kvm_hva_to_rmapp */
4391 			spin_lock(&kvm->mmu_lock);
4392 			memslot->userspace_addr = userspace_addr;
4393 			spin_unlock(&kvm->mmu_lock);
4394 		} else {
4395 			if (!old.user_alloc && old.rmap) {
4396 				int ret;
4397 
4398 				down_write(&current->mm->mmap_sem);
4399 				ret = do_munmap(current->mm, old.userspace_addr,
4400 						old.npages * PAGE_SIZE);
4401 				up_write(&current->mm->mmap_sem);
4402 				if (ret < 0)
4403 					printk(KERN_WARNING
4404 				       "kvm_vm_ioctl_set_memory_region: "
4405 				       "failed to munmap memory\n");
4406 			}
4407 		}
4408 	}
4409 
4410 	if (!kvm->arch.n_requested_mmu_pages) {
4411 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4412 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4413 	}
4414 
4415 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4416 	kvm_flush_remote_tlbs(kvm);
4417 
4418 	return 0;
4419 }
4420 
4421 void kvm_arch_flush_shadow(struct kvm *kvm)
4422 {
4423 	kvm_mmu_zap_all(kvm);
4424 }
4425 
4426 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4427 {
4428 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4429 	       || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4430 	       || vcpu->arch.nmi_pending;
4431 }
4432 
4433 static void vcpu_kick_intr(void *info)
4434 {
4435 #ifdef DEBUG
4436 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4437 	printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4438 #endif
4439 }
4440 
4441 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4442 {
4443 	int ipi_pcpu = vcpu->cpu;
4444 	int cpu = get_cpu();
4445 
4446 	if (waitqueue_active(&vcpu->wq)) {
4447 		wake_up_interruptible(&vcpu->wq);
4448 		++vcpu->stat.halt_wakeup;
4449 	}
4450 	/*
4451 	 * We may be called synchronously with irqs disabled in guest mode,
4452 	 * So need not to call smp_call_function_single() in that case.
4453 	 */
4454 	if (vcpu->guest_mode && vcpu->cpu != cpu)
4455 		smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4456 	put_cpu();
4457 }
4458