xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 82ced6fd)
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 	unsigned long old_cr4 = vcpu->arch.cr4;
342 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
343 
344 	if (cr4 & CR4_RESERVED_BITS) {
345 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
346 		kvm_inject_gp(vcpu, 0);
347 		return;
348 	}
349 
350 	if (is_long_mode(vcpu)) {
351 		if (!(cr4 & X86_CR4_PAE)) {
352 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
353 			       "in long mode\n");
354 			kvm_inject_gp(vcpu, 0);
355 			return;
356 		}
357 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
358 		   && ((cr4 ^ old_cr4) & pdptr_bits)
359 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
360 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
361 		kvm_inject_gp(vcpu, 0);
362 		return;
363 	}
364 
365 	if (cr4 & X86_CR4_VMXE) {
366 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
367 		kvm_inject_gp(vcpu, 0);
368 		return;
369 	}
370 	kvm_x86_ops->set_cr4(vcpu, cr4);
371 	vcpu->arch.cr4 = cr4;
372 	vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
373 	kvm_mmu_sync_global(vcpu);
374 	kvm_mmu_reset_context(vcpu);
375 }
376 EXPORT_SYMBOL_GPL(kvm_set_cr4);
377 
378 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
379 {
380 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
381 		kvm_mmu_sync_roots(vcpu);
382 		kvm_mmu_flush_tlb(vcpu);
383 		return;
384 	}
385 
386 	if (is_long_mode(vcpu)) {
387 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
388 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
389 			kvm_inject_gp(vcpu, 0);
390 			return;
391 		}
392 	} else {
393 		if (is_pae(vcpu)) {
394 			if (cr3 & CR3_PAE_RESERVED_BITS) {
395 				printk(KERN_DEBUG
396 				       "set_cr3: #GP, reserved bits\n");
397 				kvm_inject_gp(vcpu, 0);
398 				return;
399 			}
400 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
401 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
402 				       "reserved bits\n");
403 				kvm_inject_gp(vcpu, 0);
404 				return;
405 			}
406 		}
407 		/*
408 		 * We don't check reserved bits in nonpae mode, because
409 		 * this isn't enforced, and VMware depends on this.
410 		 */
411 	}
412 
413 	/*
414 	 * Does the new cr3 value map to physical memory? (Note, we
415 	 * catch an invalid cr3 even in real-mode, because it would
416 	 * cause trouble later on when we turn on paging anyway.)
417 	 *
418 	 * A real CPU would silently accept an invalid cr3 and would
419 	 * attempt to use it - with largely undefined (and often hard
420 	 * to debug) behavior on the guest side.
421 	 */
422 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
423 		kvm_inject_gp(vcpu, 0);
424 	else {
425 		vcpu->arch.cr3 = cr3;
426 		vcpu->arch.mmu.new_cr3(vcpu);
427 	}
428 }
429 EXPORT_SYMBOL_GPL(kvm_set_cr3);
430 
431 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
432 {
433 	if (cr8 & CR8_RESERVED_BITS) {
434 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
435 		kvm_inject_gp(vcpu, 0);
436 		return;
437 	}
438 	if (irqchip_in_kernel(vcpu->kvm))
439 		kvm_lapic_set_tpr(vcpu, cr8);
440 	else
441 		vcpu->arch.cr8 = cr8;
442 }
443 EXPORT_SYMBOL_GPL(kvm_set_cr8);
444 
445 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
446 {
447 	if (irqchip_in_kernel(vcpu->kvm))
448 		return kvm_lapic_get_cr8(vcpu);
449 	else
450 		return vcpu->arch.cr8;
451 }
452 EXPORT_SYMBOL_GPL(kvm_get_cr8);
453 
454 static inline u32 bit(int bitno)
455 {
456 	return 1 << (bitno & 31);
457 }
458 
459 /*
460  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
461  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
462  *
463  * This list is modified at module load time to reflect the
464  * capabilities of the host cpu.
465  */
466 static u32 msrs_to_save[] = {
467 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
468 	MSR_K6_STAR,
469 #ifdef CONFIG_X86_64
470 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
471 #endif
472 	MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
473 	MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
474 };
475 
476 static unsigned num_msrs_to_save;
477 
478 static u32 emulated_msrs[] = {
479 	MSR_IA32_MISC_ENABLE,
480 };
481 
482 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
483 {
484 	if (efer & efer_reserved_bits) {
485 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
486 		       efer);
487 		kvm_inject_gp(vcpu, 0);
488 		return;
489 	}
490 
491 	if (is_paging(vcpu)
492 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
493 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
494 		kvm_inject_gp(vcpu, 0);
495 		return;
496 	}
497 
498 	if (efer & EFER_FFXSR) {
499 		struct kvm_cpuid_entry2 *feat;
500 
501 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
502 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
503 			printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
504 			kvm_inject_gp(vcpu, 0);
505 			return;
506 		}
507 	}
508 
509 	if (efer & EFER_SVME) {
510 		struct kvm_cpuid_entry2 *feat;
511 
512 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
513 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
514 			printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
515 			kvm_inject_gp(vcpu, 0);
516 			return;
517 		}
518 	}
519 
520 	kvm_x86_ops->set_efer(vcpu, efer);
521 
522 	efer &= ~EFER_LMA;
523 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
524 
525 	vcpu->arch.shadow_efer = efer;
526 }
527 
528 void kvm_enable_efer_bits(u64 mask)
529 {
530        efer_reserved_bits &= ~mask;
531 }
532 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
533 
534 
535 /*
536  * Writes msr value into into the appropriate "register".
537  * Returns 0 on success, non-0 otherwise.
538  * Assumes vcpu_load() was already called.
539  */
540 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
541 {
542 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
543 }
544 
545 /*
546  * Adapt set_msr() to msr_io()'s calling convention
547  */
548 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
549 {
550 	return kvm_set_msr(vcpu, index, *data);
551 }
552 
553 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
554 {
555 	static int version;
556 	struct pvclock_wall_clock wc;
557 	struct timespec now, sys, boot;
558 
559 	if (!wall_clock)
560 		return;
561 
562 	version++;
563 
564 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
565 
566 	/*
567 	 * The guest calculates current wall clock time by adding
568 	 * system time (updated by kvm_write_guest_time below) to the
569 	 * wall clock specified here.  guest system time equals host
570 	 * system time for us, thus we must fill in host boot time here.
571 	 */
572 	now = current_kernel_time();
573 	ktime_get_ts(&sys);
574 	boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
575 
576 	wc.sec = boot.tv_sec;
577 	wc.nsec = boot.tv_nsec;
578 	wc.version = version;
579 
580 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
581 
582 	version++;
583 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
584 }
585 
586 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
587 {
588 	uint32_t quotient, remainder;
589 
590 	/* Don't try to replace with do_div(), this one calculates
591 	 * "(dividend << 32) / divisor" */
592 	__asm__ ( "divl %4"
593 		  : "=a" (quotient), "=d" (remainder)
594 		  : "0" (0), "1" (dividend), "r" (divisor) );
595 	return quotient;
596 }
597 
598 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
599 {
600 	uint64_t nsecs = 1000000000LL;
601 	int32_t  shift = 0;
602 	uint64_t tps64;
603 	uint32_t tps32;
604 
605 	tps64 = tsc_khz * 1000LL;
606 	while (tps64 > nsecs*2) {
607 		tps64 >>= 1;
608 		shift--;
609 	}
610 
611 	tps32 = (uint32_t)tps64;
612 	while (tps32 <= (uint32_t)nsecs) {
613 		tps32 <<= 1;
614 		shift++;
615 	}
616 
617 	hv_clock->tsc_shift = shift;
618 	hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
619 
620 	pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
621 		 __func__, tsc_khz, hv_clock->tsc_shift,
622 		 hv_clock->tsc_to_system_mul);
623 }
624 
625 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
626 
627 static void kvm_write_guest_time(struct kvm_vcpu *v)
628 {
629 	struct timespec ts;
630 	unsigned long flags;
631 	struct kvm_vcpu_arch *vcpu = &v->arch;
632 	void *shared_kaddr;
633 
634 	if ((!vcpu->time_page))
635 		return;
636 
637 	if (unlikely(vcpu->hv_clock_tsc_khz != __get_cpu_var(cpu_tsc_khz))) {
638 		kvm_set_time_scale(__get_cpu_var(cpu_tsc_khz), &vcpu->hv_clock);
639 		vcpu->hv_clock_tsc_khz = __get_cpu_var(cpu_tsc_khz);
640 	}
641 
642 	/* Keep irq disabled to prevent changes to the clock */
643 	local_irq_save(flags);
644 	kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
645 			  &vcpu->hv_clock.tsc_timestamp);
646 	ktime_get_ts(&ts);
647 	local_irq_restore(flags);
648 
649 	/* With all the info we got, fill in the values */
650 
651 	vcpu->hv_clock.system_time = ts.tv_nsec +
652 				     (NSEC_PER_SEC * (u64)ts.tv_sec);
653 	/*
654 	 * The interface expects us to write an even number signaling that the
655 	 * update is finished. Since the guest won't see the intermediate
656 	 * state, we just increase by 2 at the end.
657 	 */
658 	vcpu->hv_clock.version += 2;
659 
660 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
661 
662 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
663 	       sizeof(vcpu->hv_clock));
664 
665 	kunmap_atomic(shared_kaddr, KM_USER0);
666 
667 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
668 }
669 
670 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
671 {
672 	struct kvm_vcpu_arch *vcpu = &v->arch;
673 
674 	if (!vcpu->time_page)
675 		return 0;
676 	set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
677 	return 1;
678 }
679 
680 static bool msr_mtrr_valid(unsigned msr)
681 {
682 	switch (msr) {
683 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
684 	case MSR_MTRRfix64K_00000:
685 	case MSR_MTRRfix16K_80000:
686 	case MSR_MTRRfix16K_A0000:
687 	case MSR_MTRRfix4K_C0000:
688 	case MSR_MTRRfix4K_C8000:
689 	case MSR_MTRRfix4K_D0000:
690 	case MSR_MTRRfix4K_D8000:
691 	case MSR_MTRRfix4K_E0000:
692 	case MSR_MTRRfix4K_E8000:
693 	case MSR_MTRRfix4K_F0000:
694 	case MSR_MTRRfix4K_F8000:
695 	case MSR_MTRRdefType:
696 	case MSR_IA32_CR_PAT:
697 		return true;
698 	case 0x2f8:
699 		return true;
700 	}
701 	return false;
702 }
703 
704 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
705 {
706 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
707 
708 	if (!msr_mtrr_valid(msr))
709 		return 1;
710 
711 	if (msr == MSR_MTRRdefType) {
712 		vcpu->arch.mtrr_state.def_type = data;
713 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
714 	} else if (msr == MSR_MTRRfix64K_00000)
715 		p[0] = data;
716 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
717 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
718 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
719 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
720 	else if (msr == MSR_IA32_CR_PAT)
721 		vcpu->arch.pat = data;
722 	else {	/* Variable MTRRs */
723 		int idx, is_mtrr_mask;
724 		u64 *pt;
725 
726 		idx = (msr - 0x200) / 2;
727 		is_mtrr_mask = msr - 0x200 - 2 * idx;
728 		if (!is_mtrr_mask)
729 			pt =
730 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
731 		else
732 			pt =
733 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
734 		*pt = data;
735 	}
736 
737 	kvm_mmu_reset_context(vcpu);
738 	return 0;
739 }
740 
741 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
742 {
743 	switch (msr) {
744 	case MSR_EFER:
745 		set_efer(vcpu, data);
746 		break;
747 	case MSR_IA32_MC0_STATUS:
748 		pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
749 		       __func__, data);
750 		break;
751 	case MSR_IA32_MCG_STATUS:
752 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
753 			__func__, data);
754 		break;
755 	case MSR_IA32_MCG_CTL:
756 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
757 			__func__, data);
758 		break;
759 	case MSR_IA32_DEBUGCTLMSR:
760 		if (!data) {
761 			/* We support the non-activated case already */
762 			break;
763 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
764 			/* Values other than LBR and BTF are vendor-specific,
765 			   thus reserved and should throw a #GP */
766 			return 1;
767 		}
768 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
769 			__func__, data);
770 		break;
771 	case MSR_IA32_UCODE_REV:
772 	case MSR_IA32_UCODE_WRITE:
773 	case MSR_VM_HSAVE_PA:
774 		break;
775 	case 0x200 ... 0x2ff:
776 		return set_msr_mtrr(vcpu, msr, data);
777 	case MSR_IA32_APICBASE:
778 		kvm_set_apic_base(vcpu, data);
779 		break;
780 	case MSR_IA32_MISC_ENABLE:
781 		vcpu->arch.ia32_misc_enable_msr = data;
782 		break;
783 	case MSR_KVM_WALL_CLOCK:
784 		vcpu->kvm->arch.wall_clock = data;
785 		kvm_write_wall_clock(vcpu->kvm, data);
786 		break;
787 	case MSR_KVM_SYSTEM_TIME: {
788 		if (vcpu->arch.time_page) {
789 			kvm_release_page_dirty(vcpu->arch.time_page);
790 			vcpu->arch.time_page = NULL;
791 		}
792 
793 		vcpu->arch.time = data;
794 
795 		/* we verify if the enable bit is set... */
796 		if (!(data & 1))
797 			break;
798 
799 		/* ...but clean it before doing the actual write */
800 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
801 
802 		vcpu->arch.time_page =
803 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
804 
805 		if (is_error_page(vcpu->arch.time_page)) {
806 			kvm_release_page_clean(vcpu->arch.time_page);
807 			vcpu->arch.time_page = NULL;
808 		}
809 
810 		kvm_request_guest_time_update(vcpu);
811 		break;
812 	}
813 	default:
814 		pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
815 		return 1;
816 	}
817 	return 0;
818 }
819 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
820 
821 
822 /*
823  * Reads an msr value (of 'msr_index') into 'pdata'.
824  * Returns 0 on success, non-0 otherwise.
825  * Assumes vcpu_load() was already called.
826  */
827 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
828 {
829 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
830 }
831 
832 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
833 {
834 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
835 
836 	if (!msr_mtrr_valid(msr))
837 		return 1;
838 
839 	if (msr == MSR_MTRRdefType)
840 		*pdata = vcpu->arch.mtrr_state.def_type +
841 			 (vcpu->arch.mtrr_state.enabled << 10);
842 	else if (msr == MSR_MTRRfix64K_00000)
843 		*pdata = p[0];
844 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
845 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
846 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
847 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
848 	else if (msr == MSR_IA32_CR_PAT)
849 		*pdata = vcpu->arch.pat;
850 	else {	/* Variable MTRRs */
851 		int idx, is_mtrr_mask;
852 		u64 *pt;
853 
854 		idx = (msr - 0x200) / 2;
855 		is_mtrr_mask = msr - 0x200 - 2 * idx;
856 		if (!is_mtrr_mask)
857 			pt =
858 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
859 		else
860 			pt =
861 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
862 		*pdata = *pt;
863 	}
864 
865 	return 0;
866 }
867 
868 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
869 {
870 	u64 data;
871 
872 	switch (msr) {
873 	case 0xc0010010: /* SYSCFG */
874 	case 0xc0010015: /* HWCR */
875 	case MSR_IA32_PLATFORM_ID:
876 	case MSR_IA32_P5_MC_ADDR:
877 	case MSR_IA32_P5_MC_TYPE:
878 	case MSR_IA32_MC0_CTL:
879 	case MSR_IA32_MCG_STATUS:
880 	case MSR_IA32_MCG_CAP:
881 	case MSR_IA32_MCG_CTL:
882 	case MSR_IA32_MC0_MISC:
883 	case MSR_IA32_MC0_MISC+4:
884 	case MSR_IA32_MC0_MISC+8:
885 	case MSR_IA32_MC0_MISC+12:
886 	case MSR_IA32_MC0_MISC+16:
887 	case MSR_IA32_MC0_MISC+20:
888 	case MSR_IA32_UCODE_REV:
889 	case MSR_IA32_EBL_CR_POWERON:
890 	case MSR_IA32_DEBUGCTLMSR:
891 	case MSR_IA32_LASTBRANCHFROMIP:
892 	case MSR_IA32_LASTBRANCHTOIP:
893 	case MSR_IA32_LASTINTFROMIP:
894 	case MSR_IA32_LASTINTTOIP:
895 	case MSR_VM_HSAVE_PA:
896 		data = 0;
897 		break;
898 	case MSR_MTRRcap:
899 		data = 0x500 | KVM_NR_VAR_MTRR;
900 		break;
901 	case 0x200 ... 0x2ff:
902 		return get_msr_mtrr(vcpu, msr, pdata);
903 	case 0xcd: /* fsb frequency */
904 		data = 3;
905 		break;
906 	case MSR_IA32_APICBASE:
907 		data = kvm_get_apic_base(vcpu);
908 		break;
909 	case MSR_IA32_MISC_ENABLE:
910 		data = vcpu->arch.ia32_misc_enable_msr;
911 		break;
912 	case MSR_IA32_PERF_STATUS:
913 		/* TSC increment by tick */
914 		data = 1000ULL;
915 		/* CPU multiplier */
916 		data |= (((uint64_t)4ULL) << 40);
917 		break;
918 	case MSR_EFER:
919 		data = vcpu->arch.shadow_efer;
920 		break;
921 	case MSR_KVM_WALL_CLOCK:
922 		data = vcpu->kvm->arch.wall_clock;
923 		break;
924 	case MSR_KVM_SYSTEM_TIME:
925 		data = vcpu->arch.time;
926 		break;
927 	default:
928 		pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
929 		return 1;
930 	}
931 	*pdata = data;
932 	return 0;
933 }
934 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
935 
936 /*
937  * Read or write a bunch of msrs. All parameters are kernel addresses.
938  *
939  * @return number of msrs set successfully.
940  */
941 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
942 		    struct kvm_msr_entry *entries,
943 		    int (*do_msr)(struct kvm_vcpu *vcpu,
944 				  unsigned index, u64 *data))
945 {
946 	int i;
947 
948 	vcpu_load(vcpu);
949 
950 	down_read(&vcpu->kvm->slots_lock);
951 	for (i = 0; i < msrs->nmsrs; ++i)
952 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
953 			break;
954 	up_read(&vcpu->kvm->slots_lock);
955 
956 	vcpu_put(vcpu);
957 
958 	return i;
959 }
960 
961 /*
962  * Read or write a bunch of msrs. Parameters are user addresses.
963  *
964  * @return number of msrs set successfully.
965  */
966 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
967 		  int (*do_msr)(struct kvm_vcpu *vcpu,
968 				unsigned index, u64 *data),
969 		  int writeback)
970 {
971 	struct kvm_msrs msrs;
972 	struct kvm_msr_entry *entries;
973 	int r, n;
974 	unsigned size;
975 
976 	r = -EFAULT;
977 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
978 		goto out;
979 
980 	r = -E2BIG;
981 	if (msrs.nmsrs >= MAX_IO_MSRS)
982 		goto out;
983 
984 	r = -ENOMEM;
985 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
986 	entries = vmalloc(size);
987 	if (!entries)
988 		goto out;
989 
990 	r = -EFAULT;
991 	if (copy_from_user(entries, user_msrs->entries, size))
992 		goto out_free;
993 
994 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
995 	if (r < 0)
996 		goto out_free;
997 
998 	r = -EFAULT;
999 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
1000 		goto out_free;
1001 
1002 	r = n;
1003 
1004 out_free:
1005 	vfree(entries);
1006 out:
1007 	return r;
1008 }
1009 
1010 int kvm_dev_ioctl_check_extension(long ext)
1011 {
1012 	int r;
1013 
1014 	switch (ext) {
1015 	case KVM_CAP_IRQCHIP:
1016 	case KVM_CAP_HLT:
1017 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1018 	case KVM_CAP_SET_TSS_ADDR:
1019 	case KVM_CAP_EXT_CPUID:
1020 	case KVM_CAP_CLOCKSOURCE:
1021 	case KVM_CAP_PIT:
1022 	case KVM_CAP_NOP_IO_DELAY:
1023 	case KVM_CAP_MP_STATE:
1024 	case KVM_CAP_SYNC_MMU:
1025 	case KVM_CAP_REINJECT_CONTROL:
1026 	case KVM_CAP_IRQ_INJECT_STATUS:
1027 		r = 1;
1028 		break;
1029 	case KVM_CAP_COALESCED_MMIO:
1030 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1031 		break;
1032 	case KVM_CAP_VAPIC:
1033 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1034 		break;
1035 	case KVM_CAP_NR_VCPUS:
1036 		r = KVM_MAX_VCPUS;
1037 		break;
1038 	case KVM_CAP_NR_MEMSLOTS:
1039 		r = KVM_MEMORY_SLOTS;
1040 		break;
1041 	case KVM_CAP_PV_MMU:
1042 		r = !tdp_enabled;
1043 		break;
1044 	case KVM_CAP_IOMMU:
1045 		r = iommu_found();
1046 		break;
1047 	default:
1048 		r = 0;
1049 		break;
1050 	}
1051 	return r;
1052 
1053 }
1054 
1055 long kvm_arch_dev_ioctl(struct file *filp,
1056 			unsigned int ioctl, unsigned long arg)
1057 {
1058 	void __user *argp = (void __user *)arg;
1059 	long r;
1060 
1061 	switch (ioctl) {
1062 	case KVM_GET_MSR_INDEX_LIST: {
1063 		struct kvm_msr_list __user *user_msr_list = argp;
1064 		struct kvm_msr_list msr_list;
1065 		unsigned n;
1066 
1067 		r = -EFAULT;
1068 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1069 			goto out;
1070 		n = msr_list.nmsrs;
1071 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1072 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1073 			goto out;
1074 		r = -E2BIG;
1075 		if (n < num_msrs_to_save)
1076 			goto out;
1077 		r = -EFAULT;
1078 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1079 				 num_msrs_to_save * sizeof(u32)))
1080 			goto out;
1081 		if (copy_to_user(user_msr_list->indices
1082 				 + num_msrs_to_save * sizeof(u32),
1083 				 &emulated_msrs,
1084 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1085 			goto out;
1086 		r = 0;
1087 		break;
1088 	}
1089 	case KVM_GET_SUPPORTED_CPUID: {
1090 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1091 		struct kvm_cpuid2 cpuid;
1092 
1093 		r = -EFAULT;
1094 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1095 			goto out;
1096 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1097 						      cpuid_arg->entries);
1098 		if (r)
1099 			goto out;
1100 
1101 		r = -EFAULT;
1102 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1103 			goto out;
1104 		r = 0;
1105 		break;
1106 	}
1107 	default:
1108 		r = -EINVAL;
1109 	}
1110 out:
1111 	return r;
1112 }
1113 
1114 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1115 {
1116 	kvm_x86_ops->vcpu_load(vcpu, cpu);
1117 	kvm_request_guest_time_update(vcpu);
1118 }
1119 
1120 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1121 {
1122 	kvm_x86_ops->vcpu_put(vcpu);
1123 	kvm_put_guest_fpu(vcpu);
1124 }
1125 
1126 static int is_efer_nx(void)
1127 {
1128 	unsigned long long efer = 0;
1129 
1130 	rdmsrl_safe(MSR_EFER, &efer);
1131 	return efer & EFER_NX;
1132 }
1133 
1134 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1135 {
1136 	int i;
1137 	struct kvm_cpuid_entry2 *e, *entry;
1138 
1139 	entry = NULL;
1140 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1141 		e = &vcpu->arch.cpuid_entries[i];
1142 		if (e->function == 0x80000001) {
1143 			entry = e;
1144 			break;
1145 		}
1146 	}
1147 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1148 		entry->edx &= ~(1 << 20);
1149 		printk(KERN_INFO "kvm: guest NX capability removed\n");
1150 	}
1151 }
1152 
1153 /* when an old userspace process fills a new kernel module */
1154 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1155 				    struct kvm_cpuid *cpuid,
1156 				    struct kvm_cpuid_entry __user *entries)
1157 {
1158 	int r, i;
1159 	struct kvm_cpuid_entry *cpuid_entries;
1160 
1161 	r = -E2BIG;
1162 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1163 		goto out;
1164 	r = -ENOMEM;
1165 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1166 	if (!cpuid_entries)
1167 		goto out;
1168 	r = -EFAULT;
1169 	if (copy_from_user(cpuid_entries, entries,
1170 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1171 		goto out_free;
1172 	for (i = 0; i < cpuid->nent; i++) {
1173 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1174 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1175 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1176 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1177 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1178 		vcpu->arch.cpuid_entries[i].index = 0;
1179 		vcpu->arch.cpuid_entries[i].flags = 0;
1180 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
1181 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
1182 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
1183 	}
1184 	vcpu->arch.cpuid_nent = cpuid->nent;
1185 	cpuid_fix_nx_cap(vcpu);
1186 	r = 0;
1187 
1188 out_free:
1189 	vfree(cpuid_entries);
1190 out:
1191 	return r;
1192 }
1193 
1194 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1195 				     struct kvm_cpuid2 *cpuid,
1196 				     struct kvm_cpuid_entry2 __user *entries)
1197 {
1198 	int r;
1199 
1200 	r = -E2BIG;
1201 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1202 		goto out;
1203 	r = -EFAULT;
1204 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1205 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1206 		goto out;
1207 	vcpu->arch.cpuid_nent = cpuid->nent;
1208 	return 0;
1209 
1210 out:
1211 	return r;
1212 }
1213 
1214 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1215 				     struct kvm_cpuid2 *cpuid,
1216 				     struct kvm_cpuid_entry2 __user *entries)
1217 {
1218 	int r;
1219 
1220 	r = -E2BIG;
1221 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1222 		goto out;
1223 	r = -EFAULT;
1224 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1225 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1226 		goto out;
1227 	return 0;
1228 
1229 out:
1230 	cpuid->nent = vcpu->arch.cpuid_nent;
1231 	return r;
1232 }
1233 
1234 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1235 			   u32 index)
1236 {
1237 	entry->function = function;
1238 	entry->index = index;
1239 	cpuid_count(entry->function, entry->index,
1240 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1241 	entry->flags = 0;
1242 }
1243 
1244 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1245 			 u32 index, int *nent, int maxnent)
1246 {
1247 	const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1248 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1249 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1250 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1251 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1252 		bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1253 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1254 		bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1255 		bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1256 		bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1257 	const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1258 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1259 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1260 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1261 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1262 		bit(X86_FEATURE_PGE) |
1263 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1264 		bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1265 		bit(X86_FEATURE_SYSCALL) |
1266 		(is_efer_nx() ? bit(X86_FEATURE_NX) : 0) |
1267 #ifdef CONFIG_X86_64
1268 		bit(X86_FEATURE_LM) |
1269 #endif
1270 		bit(X86_FEATURE_FXSR_OPT) |
1271 		bit(X86_FEATURE_MMXEXT) |
1272 		bit(X86_FEATURE_3DNOWEXT) |
1273 		bit(X86_FEATURE_3DNOW);
1274 	const u32 kvm_supported_word3_x86_features =
1275 		bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1276 	const u32 kvm_supported_word6_x86_features =
1277 		bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1278 		bit(X86_FEATURE_SVM);
1279 
1280 	/* all calls to cpuid_count() should be made on the same cpu */
1281 	get_cpu();
1282 	do_cpuid_1_ent(entry, function, index);
1283 	++*nent;
1284 
1285 	switch (function) {
1286 	case 0:
1287 		entry->eax = min(entry->eax, (u32)0xb);
1288 		break;
1289 	case 1:
1290 		entry->edx &= kvm_supported_word0_x86_features;
1291 		entry->ecx &= kvm_supported_word3_x86_features;
1292 		break;
1293 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1294 	 * may return different values. This forces us to get_cpu() before
1295 	 * issuing the first command, and also to emulate this annoying behavior
1296 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1297 	case 2: {
1298 		int t, times = entry->eax & 0xff;
1299 
1300 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1301 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1302 		for (t = 1; t < times && *nent < maxnent; ++t) {
1303 			do_cpuid_1_ent(&entry[t], function, 0);
1304 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1305 			++*nent;
1306 		}
1307 		break;
1308 	}
1309 	/* function 4 and 0xb have additional index. */
1310 	case 4: {
1311 		int i, cache_type;
1312 
1313 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1314 		/* read more entries until cache_type is zero */
1315 		for (i = 1; *nent < maxnent; ++i) {
1316 			cache_type = entry[i - 1].eax & 0x1f;
1317 			if (!cache_type)
1318 				break;
1319 			do_cpuid_1_ent(&entry[i], function, i);
1320 			entry[i].flags |=
1321 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1322 			++*nent;
1323 		}
1324 		break;
1325 	}
1326 	case 0xb: {
1327 		int i, level_type;
1328 
1329 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1330 		/* read more entries until level_type is zero */
1331 		for (i = 1; *nent < maxnent; ++i) {
1332 			level_type = entry[i - 1].ecx & 0xff00;
1333 			if (!level_type)
1334 				break;
1335 			do_cpuid_1_ent(&entry[i], function, i);
1336 			entry[i].flags |=
1337 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1338 			++*nent;
1339 		}
1340 		break;
1341 	}
1342 	case 0x80000000:
1343 		entry->eax = min(entry->eax, 0x8000001a);
1344 		break;
1345 	case 0x80000001:
1346 		entry->edx &= kvm_supported_word1_x86_features;
1347 		entry->ecx &= kvm_supported_word6_x86_features;
1348 		break;
1349 	}
1350 	put_cpu();
1351 }
1352 
1353 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1354 				     struct kvm_cpuid_entry2 __user *entries)
1355 {
1356 	struct kvm_cpuid_entry2 *cpuid_entries;
1357 	int limit, nent = 0, r = -E2BIG;
1358 	u32 func;
1359 
1360 	if (cpuid->nent < 1)
1361 		goto out;
1362 	r = -ENOMEM;
1363 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1364 	if (!cpuid_entries)
1365 		goto out;
1366 
1367 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1368 	limit = cpuid_entries[0].eax;
1369 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1370 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1371 			     &nent, cpuid->nent);
1372 	r = -E2BIG;
1373 	if (nent >= cpuid->nent)
1374 		goto out_free;
1375 
1376 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1377 	limit = cpuid_entries[nent - 1].eax;
1378 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1379 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1380 			     &nent, cpuid->nent);
1381 	r = -EFAULT;
1382 	if (copy_to_user(entries, cpuid_entries,
1383 			 nent * sizeof(struct kvm_cpuid_entry2)))
1384 		goto out_free;
1385 	cpuid->nent = nent;
1386 	r = 0;
1387 
1388 out_free:
1389 	vfree(cpuid_entries);
1390 out:
1391 	return r;
1392 }
1393 
1394 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1395 				    struct kvm_lapic_state *s)
1396 {
1397 	vcpu_load(vcpu);
1398 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1399 	vcpu_put(vcpu);
1400 
1401 	return 0;
1402 }
1403 
1404 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1405 				    struct kvm_lapic_state *s)
1406 {
1407 	vcpu_load(vcpu);
1408 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1409 	kvm_apic_post_state_restore(vcpu);
1410 	vcpu_put(vcpu);
1411 
1412 	return 0;
1413 }
1414 
1415 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1416 				    struct kvm_interrupt *irq)
1417 {
1418 	if (irq->irq < 0 || irq->irq >= 256)
1419 		return -EINVAL;
1420 	if (irqchip_in_kernel(vcpu->kvm))
1421 		return -ENXIO;
1422 	vcpu_load(vcpu);
1423 
1424 	set_bit(irq->irq, vcpu->arch.irq_pending);
1425 	set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1426 
1427 	vcpu_put(vcpu);
1428 
1429 	return 0;
1430 }
1431 
1432 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1433 {
1434 	vcpu_load(vcpu);
1435 	kvm_inject_nmi(vcpu);
1436 	vcpu_put(vcpu);
1437 
1438 	return 0;
1439 }
1440 
1441 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1442 					   struct kvm_tpr_access_ctl *tac)
1443 {
1444 	if (tac->flags)
1445 		return -EINVAL;
1446 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1447 	return 0;
1448 }
1449 
1450 long kvm_arch_vcpu_ioctl(struct file *filp,
1451 			 unsigned int ioctl, unsigned long arg)
1452 {
1453 	struct kvm_vcpu *vcpu = filp->private_data;
1454 	void __user *argp = (void __user *)arg;
1455 	int r;
1456 	struct kvm_lapic_state *lapic = NULL;
1457 
1458 	switch (ioctl) {
1459 	case KVM_GET_LAPIC: {
1460 		lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1461 
1462 		r = -ENOMEM;
1463 		if (!lapic)
1464 			goto out;
1465 		r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1466 		if (r)
1467 			goto out;
1468 		r = -EFAULT;
1469 		if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1470 			goto out;
1471 		r = 0;
1472 		break;
1473 	}
1474 	case KVM_SET_LAPIC: {
1475 		lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1476 		r = -ENOMEM;
1477 		if (!lapic)
1478 			goto out;
1479 		r = -EFAULT;
1480 		if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1481 			goto out;
1482 		r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1483 		if (r)
1484 			goto out;
1485 		r = 0;
1486 		break;
1487 	}
1488 	case KVM_INTERRUPT: {
1489 		struct kvm_interrupt irq;
1490 
1491 		r = -EFAULT;
1492 		if (copy_from_user(&irq, argp, sizeof irq))
1493 			goto out;
1494 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1495 		if (r)
1496 			goto out;
1497 		r = 0;
1498 		break;
1499 	}
1500 	case KVM_NMI: {
1501 		r = kvm_vcpu_ioctl_nmi(vcpu);
1502 		if (r)
1503 			goto out;
1504 		r = 0;
1505 		break;
1506 	}
1507 	case KVM_SET_CPUID: {
1508 		struct kvm_cpuid __user *cpuid_arg = argp;
1509 		struct kvm_cpuid cpuid;
1510 
1511 		r = -EFAULT;
1512 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1513 			goto out;
1514 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1515 		if (r)
1516 			goto out;
1517 		break;
1518 	}
1519 	case KVM_SET_CPUID2: {
1520 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1521 		struct kvm_cpuid2 cpuid;
1522 
1523 		r = -EFAULT;
1524 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1525 			goto out;
1526 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1527 					      cpuid_arg->entries);
1528 		if (r)
1529 			goto out;
1530 		break;
1531 	}
1532 	case KVM_GET_CPUID2: {
1533 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1534 		struct kvm_cpuid2 cpuid;
1535 
1536 		r = -EFAULT;
1537 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1538 			goto out;
1539 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1540 					      cpuid_arg->entries);
1541 		if (r)
1542 			goto out;
1543 		r = -EFAULT;
1544 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1545 			goto out;
1546 		r = 0;
1547 		break;
1548 	}
1549 	case KVM_GET_MSRS:
1550 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1551 		break;
1552 	case KVM_SET_MSRS:
1553 		r = msr_io(vcpu, argp, do_set_msr, 0);
1554 		break;
1555 	case KVM_TPR_ACCESS_REPORTING: {
1556 		struct kvm_tpr_access_ctl tac;
1557 
1558 		r = -EFAULT;
1559 		if (copy_from_user(&tac, argp, sizeof tac))
1560 			goto out;
1561 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1562 		if (r)
1563 			goto out;
1564 		r = -EFAULT;
1565 		if (copy_to_user(argp, &tac, sizeof tac))
1566 			goto out;
1567 		r = 0;
1568 		break;
1569 	};
1570 	case KVM_SET_VAPIC_ADDR: {
1571 		struct kvm_vapic_addr va;
1572 
1573 		r = -EINVAL;
1574 		if (!irqchip_in_kernel(vcpu->kvm))
1575 			goto out;
1576 		r = -EFAULT;
1577 		if (copy_from_user(&va, argp, sizeof va))
1578 			goto out;
1579 		r = 0;
1580 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1581 		break;
1582 	}
1583 	default:
1584 		r = -EINVAL;
1585 	}
1586 out:
1587 	if (lapic)
1588 		kfree(lapic);
1589 	return r;
1590 }
1591 
1592 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1593 {
1594 	int ret;
1595 
1596 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1597 		return -1;
1598 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1599 	return ret;
1600 }
1601 
1602 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1603 					  u32 kvm_nr_mmu_pages)
1604 {
1605 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1606 		return -EINVAL;
1607 
1608 	down_write(&kvm->slots_lock);
1609 
1610 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1611 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1612 
1613 	up_write(&kvm->slots_lock);
1614 	return 0;
1615 }
1616 
1617 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1618 {
1619 	return kvm->arch.n_alloc_mmu_pages;
1620 }
1621 
1622 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1623 {
1624 	int i;
1625 	struct kvm_mem_alias *alias;
1626 
1627 	for (i = 0; i < kvm->arch.naliases; ++i) {
1628 		alias = &kvm->arch.aliases[i];
1629 		if (gfn >= alias->base_gfn
1630 		    && gfn < alias->base_gfn + alias->npages)
1631 			return alias->target_gfn + gfn - alias->base_gfn;
1632 	}
1633 	return gfn;
1634 }
1635 
1636 /*
1637  * Set a new alias region.  Aliases map a portion of physical memory into
1638  * another portion.  This is useful for memory windows, for example the PC
1639  * VGA region.
1640  */
1641 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1642 					 struct kvm_memory_alias *alias)
1643 {
1644 	int r, n;
1645 	struct kvm_mem_alias *p;
1646 
1647 	r = -EINVAL;
1648 	/* General sanity checks */
1649 	if (alias->memory_size & (PAGE_SIZE - 1))
1650 		goto out;
1651 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1652 		goto out;
1653 	if (alias->slot >= KVM_ALIAS_SLOTS)
1654 		goto out;
1655 	if (alias->guest_phys_addr + alias->memory_size
1656 	    < alias->guest_phys_addr)
1657 		goto out;
1658 	if (alias->target_phys_addr + alias->memory_size
1659 	    < alias->target_phys_addr)
1660 		goto out;
1661 
1662 	down_write(&kvm->slots_lock);
1663 	spin_lock(&kvm->mmu_lock);
1664 
1665 	p = &kvm->arch.aliases[alias->slot];
1666 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1667 	p->npages = alias->memory_size >> PAGE_SHIFT;
1668 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1669 
1670 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1671 		if (kvm->arch.aliases[n - 1].npages)
1672 			break;
1673 	kvm->arch.naliases = n;
1674 
1675 	spin_unlock(&kvm->mmu_lock);
1676 	kvm_mmu_zap_all(kvm);
1677 
1678 	up_write(&kvm->slots_lock);
1679 
1680 	return 0;
1681 
1682 out:
1683 	return r;
1684 }
1685 
1686 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1687 {
1688 	int r;
1689 
1690 	r = 0;
1691 	switch (chip->chip_id) {
1692 	case KVM_IRQCHIP_PIC_MASTER:
1693 		memcpy(&chip->chip.pic,
1694 			&pic_irqchip(kvm)->pics[0],
1695 			sizeof(struct kvm_pic_state));
1696 		break;
1697 	case KVM_IRQCHIP_PIC_SLAVE:
1698 		memcpy(&chip->chip.pic,
1699 			&pic_irqchip(kvm)->pics[1],
1700 			sizeof(struct kvm_pic_state));
1701 		break;
1702 	case KVM_IRQCHIP_IOAPIC:
1703 		memcpy(&chip->chip.ioapic,
1704 			ioapic_irqchip(kvm),
1705 			sizeof(struct kvm_ioapic_state));
1706 		break;
1707 	default:
1708 		r = -EINVAL;
1709 		break;
1710 	}
1711 	return r;
1712 }
1713 
1714 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1715 {
1716 	int r;
1717 
1718 	r = 0;
1719 	switch (chip->chip_id) {
1720 	case KVM_IRQCHIP_PIC_MASTER:
1721 		memcpy(&pic_irqchip(kvm)->pics[0],
1722 			&chip->chip.pic,
1723 			sizeof(struct kvm_pic_state));
1724 		break;
1725 	case KVM_IRQCHIP_PIC_SLAVE:
1726 		memcpy(&pic_irqchip(kvm)->pics[1],
1727 			&chip->chip.pic,
1728 			sizeof(struct kvm_pic_state));
1729 		break;
1730 	case KVM_IRQCHIP_IOAPIC:
1731 		memcpy(ioapic_irqchip(kvm),
1732 			&chip->chip.ioapic,
1733 			sizeof(struct kvm_ioapic_state));
1734 		break;
1735 	default:
1736 		r = -EINVAL;
1737 		break;
1738 	}
1739 	kvm_pic_update_irq(pic_irqchip(kvm));
1740 	return r;
1741 }
1742 
1743 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1744 {
1745 	int r = 0;
1746 
1747 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1748 	return r;
1749 }
1750 
1751 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1752 {
1753 	int r = 0;
1754 
1755 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1756 	kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1757 	return r;
1758 }
1759 
1760 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1761 				 struct kvm_reinject_control *control)
1762 {
1763 	if (!kvm->arch.vpit)
1764 		return -ENXIO;
1765 	kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1766 	return 0;
1767 }
1768 
1769 /*
1770  * Get (and clear) the dirty memory log for a memory slot.
1771  */
1772 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1773 				      struct kvm_dirty_log *log)
1774 {
1775 	int r;
1776 	int n;
1777 	struct kvm_memory_slot *memslot;
1778 	int is_dirty = 0;
1779 
1780 	down_write(&kvm->slots_lock);
1781 
1782 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1783 	if (r)
1784 		goto out;
1785 
1786 	/* If nothing is dirty, don't bother messing with page tables. */
1787 	if (is_dirty) {
1788 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
1789 		kvm_flush_remote_tlbs(kvm);
1790 		memslot = &kvm->memslots[log->slot];
1791 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1792 		memset(memslot->dirty_bitmap, 0, n);
1793 	}
1794 	r = 0;
1795 out:
1796 	up_write(&kvm->slots_lock);
1797 	return r;
1798 }
1799 
1800 long kvm_arch_vm_ioctl(struct file *filp,
1801 		       unsigned int ioctl, unsigned long arg)
1802 {
1803 	struct kvm *kvm = filp->private_data;
1804 	void __user *argp = (void __user *)arg;
1805 	int r = -EINVAL;
1806 	/*
1807 	 * This union makes it completely explicit to gcc-3.x
1808 	 * that these two variables' stack usage should be
1809 	 * combined, not added together.
1810 	 */
1811 	union {
1812 		struct kvm_pit_state ps;
1813 		struct kvm_memory_alias alias;
1814 	} u;
1815 
1816 	switch (ioctl) {
1817 	case KVM_SET_TSS_ADDR:
1818 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1819 		if (r < 0)
1820 			goto out;
1821 		break;
1822 	case KVM_SET_MEMORY_REGION: {
1823 		struct kvm_memory_region kvm_mem;
1824 		struct kvm_userspace_memory_region kvm_userspace_mem;
1825 
1826 		r = -EFAULT;
1827 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1828 			goto out;
1829 		kvm_userspace_mem.slot = kvm_mem.slot;
1830 		kvm_userspace_mem.flags = kvm_mem.flags;
1831 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1832 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1833 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1834 		if (r)
1835 			goto out;
1836 		break;
1837 	}
1838 	case KVM_SET_NR_MMU_PAGES:
1839 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1840 		if (r)
1841 			goto out;
1842 		break;
1843 	case KVM_GET_NR_MMU_PAGES:
1844 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1845 		break;
1846 	case KVM_SET_MEMORY_ALIAS:
1847 		r = -EFAULT;
1848 		if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1849 			goto out;
1850 		r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1851 		if (r)
1852 			goto out;
1853 		break;
1854 	case KVM_CREATE_IRQCHIP:
1855 		r = -ENOMEM;
1856 		kvm->arch.vpic = kvm_create_pic(kvm);
1857 		if (kvm->arch.vpic) {
1858 			r = kvm_ioapic_init(kvm);
1859 			if (r) {
1860 				kfree(kvm->arch.vpic);
1861 				kvm->arch.vpic = NULL;
1862 				goto out;
1863 			}
1864 		} else
1865 			goto out;
1866 		r = kvm_setup_default_irq_routing(kvm);
1867 		if (r) {
1868 			kfree(kvm->arch.vpic);
1869 			kfree(kvm->arch.vioapic);
1870 			goto out;
1871 		}
1872 		break;
1873 	case KVM_CREATE_PIT:
1874 		mutex_lock(&kvm->lock);
1875 		r = -EEXIST;
1876 		if (kvm->arch.vpit)
1877 			goto create_pit_unlock;
1878 		r = -ENOMEM;
1879 		kvm->arch.vpit = kvm_create_pit(kvm);
1880 		if (kvm->arch.vpit)
1881 			r = 0;
1882 	create_pit_unlock:
1883 		mutex_unlock(&kvm->lock);
1884 		break;
1885 	case KVM_IRQ_LINE_STATUS:
1886 	case KVM_IRQ_LINE: {
1887 		struct kvm_irq_level irq_event;
1888 
1889 		r = -EFAULT;
1890 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
1891 			goto out;
1892 		if (irqchip_in_kernel(kvm)) {
1893 			__s32 status;
1894 			mutex_lock(&kvm->lock);
1895 			status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1896 					irq_event.irq, irq_event.level);
1897 			mutex_unlock(&kvm->lock);
1898 			if (ioctl == KVM_IRQ_LINE_STATUS) {
1899 				irq_event.status = status;
1900 				if (copy_to_user(argp, &irq_event,
1901 							sizeof irq_event))
1902 					goto out;
1903 			}
1904 			r = 0;
1905 		}
1906 		break;
1907 	}
1908 	case KVM_GET_IRQCHIP: {
1909 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1910 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1911 
1912 		r = -ENOMEM;
1913 		if (!chip)
1914 			goto out;
1915 		r = -EFAULT;
1916 		if (copy_from_user(chip, argp, sizeof *chip))
1917 			goto get_irqchip_out;
1918 		r = -ENXIO;
1919 		if (!irqchip_in_kernel(kvm))
1920 			goto get_irqchip_out;
1921 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1922 		if (r)
1923 			goto get_irqchip_out;
1924 		r = -EFAULT;
1925 		if (copy_to_user(argp, chip, sizeof *chip))
1926 			goto get_irqchip_out;
1927 		r = 0;
1928 	get_irqchip_out:
1929 		kfree(chip);
1930 		if (r)
1931 			goto out;
1932 		break;
1933 	}
1934 	case KVM_SET_IRQCHIP: {
1935 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1936 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1937 
1938 		r = -ENOMEM;
1939 		if (!chip)
1940 			goto out;
1941 		r = -EFAULT;
1942 		if (copy_from_user(chip, argp, sizeof *chip))
1943 			goto set_irqchip_out;
1944 		r = -ENXIO;
1945 		if (!irqchip_in_kernel(kvm))
1946 			goto set_irqchip_out;
1947 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1948 		if (r)
1949 			goto set_irqchip_out;
1950 		r = 0;
1951 	set_irqchip_out:
1952 		kfree(chip);
1953 		if (r)
1954 			goto out;
1955 		break;
1956 	}
1957 	case KVM_GET_PIT: {
1958 		r = -EFAULT;
1959 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1960 			goto out;
1961 		r = -ENXIO;
1962 		if (!kvm->arch.vpit)
1963 			goto out;
1964 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1965 		if (r)
1966 			goto out;
1967 		r = -EFAULT;
1968 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1969 			goto out;
1970 		r = 0;
1971 		break;
1972 	}
1973 	case KVM_SET_PIT: {
1974 		r = -EFAULT;
1975 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
1976 			goto out;
1977 		r = -ENXIO;
1978 		if (!kvm->arch.vpit)
1979 			goto out;
1980 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1981 		if (r)
1982 			goto out;
1983 		r = 0;
1984 		break;
1985 	}
1986 	case KVM_REINJECT_CONTROL: {
1987 		struct kvm_reinject_control control;
1988 		r =  -EFAULT;
1989 		if (copy_from_user(&control, argp, sizeof(control)))
1990 			goto out;
1991 		r = kvm_vm_ioctl_reinject(kvm, &control);
1992 		if (r)
1993 			goto out;
1994 		r = 0;
1995 		break;
1996 	}
1997 	default:
1998 		;
1999 	}
2000 out:
2001 	return r;
2002 }
2003 
2004 static void kvm_init_msr_list(void)
2005 {
2006 	u32 dummy[2];
2007 	unsigned i, j;
2008 
2009 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2010 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2011 			continue;
2012 		if (j < i)
2013 			msrs_to_save[j] = msrs_to_save[i];
2014 		j++;
2015 	}
2016 	num_msrs_to_save = j;
2017 }
2018 
2019 /*
2020  * Only apic need an MMIO device hook, so shortcut now..
2021  */
2022 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2023 						gpa_t addr, int len,
2024 						int is_write)
2025 {
2026 	struct kvm_io_device *dev;
2027 
2028 	if (vcpu->arch.apic) {
2029 		dev = &vcpu->arch.apic->dev;
2030 		if (dev->in_range(dev, addr, len, is_write))
2031 			return dev;
2032 	}
2033 	return NULL;
2034 }
2035 
2036 
2037 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2038 						gpa_t addr, int len,
2039 						int is_write)
2040 {
2041 	struct kvm_io_device *dev;
2042 
2043 	dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2044 	if (dev == NULL)
2045 		dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2046 					  is_write);
2047 	return dev;
2048 }
2049 
2050 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2051 			       struct kvm_vcpu *vcpu)
2052 {
2053 	void *data = val;
2054 	int r = X86EMUL_CONTINUE;
2055 
2056 	while (bytes) {
2057 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2058 		unsigned offset = addr & (PAGE_SIZE-1);
2059 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2060 		int ret;
2061 
2062 		if (gpa == UNMAPPED_GVA) {
2063 			r = X86EMUL_PROPAGATE_FAULT;
2064 			goto out;
2065 		}
2066 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2067 		if (ret < 0) {
2068 			r = X86EMUL_UNHANDLEABLE;
2069 			goto out;
2070 		}
2071 
2072 		bytes -= toread;
2073 		data += toread;
2074 		addr += toread;
2075 	}
2076 out:
2077 	return r;
2078 }
2079 
2080 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2081 				struct kvm_vcpu *vcpu)
2082 {
2083 	void *data = val;
2084 	int r = X86EMUL_CONTINUE;
2085 
2086 	while (bytes) {
2087 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2088 		unsigned offset = addr & (PAGE_SIZE-1);
2089 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2090 		int ret;
2091 
2092 		if (gpa == UNMAPPED_GVA) {
2093 			r = X86EMUL_PROPAGATE_FAULT;
2094 			goto out;
2095 		}
2096 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2097 		if (ret < 0) {
2098 			r = X86EMUL_UNHANDLEABLE;
2099 			goto out;
2100 		}
2101 
2102 		bytes -= towrite;
2103 		data += towrite;
2104 		addr += towrite;
2105 	}
2106 out:
2107 	return r;
2108 }
2109 
2110 
2111 static int emulator_read_emulated(unsigned long addr,
2112 				  void *val,
2113 				  unsigned int bytes,
2114 				  struct kvm_vcpu *vcpu)
2115 {
2116 	struct kvm_io_device *mmio_dev;
2117 	gpa_t                 gpa;
2118 
2119 	if (vcpu->mmio_read_completed) {
2120 		memcpy(val, vcpu->mmio_data, bytes);
2121 		vcpu->mmio_read_completed = 0;
2122 		return X86EMUL_CONTINUE;
2123 	}
2124 
2125 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2126 
2127 	/* For APIC access vmexit */
2128 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2129 		goto mmio;
2130 
2131 	if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2132 				== X86EMUL_CONTINUE)
2133 		return X86EMUL_CONTINUE;
2134 	if (gpa == UNMAPPED_GVA)
2135 		return X86EMUL_PROPAGATE_FAULT;
2136 
2137 mmio:
2138 	/*
2139 	 * Is this MMIO handled locally?
2140 	 */
2141 	mutex_lock(&vcpu->kvm->lock);
2142 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2143 	if (mmio_dev) {
2144 		kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2145 		mutex_unlock(&vcpu->kvm->lock);
2146 		return X86EMUL_CONTINUE;
2147 	}
2148 	mutex_unlock(&vcpu->kvm->lock);
2149 
2150 	vcpu->mmio_needed = 1;
2151 	vcpu->mmio_phys_addr = gpa;
2152 	vcpu->mmio_size = bytes;
2153 	vcpu->mmio_is_write = 0;
2154 
2155 	return X86EMUL_UNHANDLEABLE;
2156 }
2157 
2158 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2159 			  const void *val, int bytes)
2160 {
2161 	int ret;
2162 
2163 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2164 	if (ret < 0)
2165 		return 0;
2166 	kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2167 	return 1;
2168 }
2169 
2170 static int emulator_write_emulated_onepage(unsigned long addr,
2171 					   const void *val,
2172 					   unsigned int bytes,
2173 					   struct kvm_vcpu *vcpu)
2174 {
2175 	struct kvm_io_device *mmio_dev;
2176 	gpa_t                 gpa;
2177 
2178 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2179 
2180 	if (gpa == UNMAPPED_GVA) {
2181 		kvm_inject_page_fault(vcpu, addr, 2);
2182 		return X86EMUL_PROPAGATE_FAULT;
2183 	}
2184 
2185 	/* For APIC access vmexit */
2186 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2187 		goto mmio;
2188 
2189 	if (emulator_write_phys(vcpu, gpa, val, bytes))
2190 		return X86EMUL_CONTINUE;
2191 
2192 mmio:
2193 	/*
2194 	 * Is this MMIO handled locally?
2195 	 */
2196 	mutex_lock(&vcpu->kvm->lock);
2197 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2198 	if (mmio_dev) {
2199 		kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2200 		mutex_unlock(&vcpu->kvm->lock);
2201 		return X86EMUL_CONTINUE;
2202 	}
2203 	mutex_unlock(&vcpu->kvm->lock);
2204 
2205 	vcpu->mmio_needed = 1;
2206 	vcpu->mmio_phys_addr = gpa;
2207 	vcpu->mmio_size = bytes;
2208 	vcpu->mmio_is_write = 1;
2209 	memcpy(vcpu->mmio_data, val, bytes);
2210 
2211 	return X86EMUL_CONTINUE;
2212 }
2213 
2214 int emulator_write_emulated(unsigned long addr,
2215 				   const void *val,
2216 				   unsigned int bytes,
2217 				   struct kvm_vcpu *vcpu)
2218 {
2219 	/* Crossing a page boundary? */
2220 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2221 		int rc, now;
2222 
2223 		now = -addr & ~PAGE_MASK;
2224 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2225 		if (rc != X86EMUL_CONTINUE)
2226 			return rc;
2227 		addr += now;
2228 		val += now;
2229 		bytes -= now;
2230 	}
2231 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2232 }
2233 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2234 
2235 static int emulator_cmpxchg_emulated(unsigned long addr,
2236 				     const void *old,
2237 				     const void *new,
2238 				     unsigned int bytes,
2239 				     struct kvm_vcpu *vcpu)
2240 {
2241 	static int reported;
2242 
2243 	if (!reported) {
2244 		reported = 1;
2245 		printk(KERN_WARNING "kvm: emulating exchange as write\n");
2246 	}
2247 #ifndef CONFIG_X86_64
2248 	/* guests cmpxchg8b have to be emulated atomically */
2249 	if (bytes == 8) {
2250 		gpa_t gpa;
2251 		struct page *page;
2252 		char *kaddr;
2253 		u64 val;
2254 
2255 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2256 
2257 		if (gpa == UNMAPPED_GVA ||
2258 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2259 			goto emul_write;
2260 
2261 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2262 			goto emul_write;
2263 
2264 		val = *(u64 *)new;
2265 
2266 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2267 
2268 		kaddr = kmap_atomic(page, KM_USER0);
2269 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2270 		kunmap_atomic(kaddr, KM_USER0);
2271 		kvm_release_page_dirty(page);
2272 	}
2273 emul_write:
2274 #endif
2275 
2276 	return emulator_write_emulated(addr, new, bytes, vcpu);
2277 }
2278 
2279 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2280 {
2281 	return kvm_x86_ops->get_segment_base(vcpu, seg);
2282 }
2283 
2284 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2285 {
2286 	kvm_mmu_invlpg(vcpu, address);
2287 	return X86EMUL_CONTINUE;
2288 }
2289 
2290 int emulate_clts(struct kvm_vcpu *vcpu)
2291 {
2292 	KVMTRACE_0D(CLTS, vcpu, handler);
2293 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2294 	return X86EMUL_CONTINUE;
2295 }
2296 
2297 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2298 {
2299 	struct kvm_vcpu *vcpu = ctxt->vcpu;
2300 
2301 	switch (dr) {
2302 	case 0 ... 3:
2303 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
2304 		return X86EMUL_CONTINUE;
2305 	default:
2306 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2307 		return X86EMUL_UNHANDLEABLE;
2308 	}
2309 }
2310 
2311 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2312 {
2313 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2314 	int exception;
2315 
2316 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2317 	if (exception) {
2318 		/* FIXME: better handling */
2319 		return X86EMUL_UNHANDLEABLE;
2320 	}
2321 	return X86EMUL_CONTINUE;
2322 }
2323 
2324 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2325 {
2326 	u8 opcodes[4];
2327 	unsigned long rip = kvm_rip_read(vcpu);
2328 	unsigned long rip_linear;
2329 
2330 	if (!printk_ratelimit())
2331 		return;
2332 
2333 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2334 
2335 	kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2336 
2337 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2338 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2339 }
2340 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2341 
2342 static struct x86_emulate_ops emulate_ops = {
2343 	.read_std            = kvm_read_guest_virt,
2344 	.read_emulated       = emulator_read_emulated,
2345 	.write_emulated      = emulator_write_emulated,
2346 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2347 };
2348 
2349 static void cache_all_regs(struct kvm_vcpu *vcpu)
2350 {
2351 	kvm_register_read(vcpu, VCPU_REGS_RAX);
2352 	kvm_register_read(vcpu, VCPU_REGS_RSP);
2353 	kvm_register_read(vcpu, VCPU_REGS_RIP);
2354 	vcpu->arch.regs_dirty = ~0;
2355 }
2356 
2357 int emulate_instruction(struct kvm_vcpu *vcpu,
2358 			struct kvm_run *run,
2359 			unsigned long cr2,
2360 			u16 error_code,
2361 			int emulation_type)
2362 {
2363 	int r;
2364 	struct decode_cache *c;
2365 
2366 	kvm_clear_exception_queue(vcpu);
2367 	vcpu->arch.mmio_fault_cr2 = cr2;
2368 	/*
2369 	 * TODO: fix x86_emulate.c to use guest_read/write_register
2370 	 * instead of direct ->regs accesses, can save hundred cycles
2371 	 * on Intel for instructions that don't read/change RSP, for
2372 	 * for example.
2373 	 */
2374 	cache_all_regs(vcpu);
2375 
2376 	vcpu->mmio_is_write = 0;
2377 	vcpu->arch.pio.string = 0;
2378 
2379 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2380 		int cs_db, cs_l;
2381 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2382 
2383 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2384 		vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2385 		vcpu->arch.emulate_ctxt.mode =
2386 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2387 			? X86EMUL_MODE_REAL : cs_l
2388 			? X86EMUL_MODE_PROT64 :	cs_db
2389 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2390 
2391 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2392 
2393 		/* Reject the instructions other than VMCALL/VMMCALL when
2394 		 * try to emulate invalid opcode */
2395 		c = &vcpu->arch.emulate_ctxt.decode;
2396 		if ((emulation_type & EMULTYPE_TRAP_UD) &&
2397 		    (!(c->twobyte && c->b == 0x01 &&
2398 		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2399 		       c->modrm_mod == 3 && c->modrm_rm == 1)))
2400 			return EMULATE_FAIL;
2401 
2402 		++vcpu->stat.insn_emulation;
2403 		if (r)  {
2404 			++vcpu->stat.insn_emulation_fail;
2405 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2406 				return EMULATE_DONE;
2407 			return EMULATE_FAIL;
2408 		}
2409 	}
2410 
2411 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2412 
2413 	if (vcpu->arch.pio.string)
2414 		return EMULATE_DO_MMIO;
2415 
2416 	if ((r || vcpu->mmio_is_write) && run) {
2417 		run->exit_reason = KVM_EXIT_MMIO;
2418 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2419 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2420 		run->mmio.len = vcpu->mmio_size;
2421 		run->mmio.is_write = vcpu->mmio_is_write;
2422 	}
2423 
2424 	if (r) {
2425 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2426 			return EMULATE_DONE;
2427 		if (!vcpu->mmio_needed) {
2428 			kvm_report_emulation_failure(vcpu, "mmio");
2429 			return EMULATE_FAIL;
2430 		}
2431 		return EMULATE_DO_MMIO;
2432 	}
2433 
2434 	kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2435 
2436 	if (vcpu->mmio_is_write) {
2437 		vcpu->mmio_needed = 0;
2438 		return EMULATE_DO_MMIO;
2439 	}
2440 
2441 	return EMULATE_DONE;
2442 }
2443 EXPORT_SYMBOL_GPL(emulate_instruction);
2444 
2445 static int pio_copy_data(struct kvm_vcpu *vcpu)
2446 {
2447 	void *p = vcpu->arch.pio_data;
2448 	gva_t q = vcpu->arch.pio.guest_gva;
2449 	unsigned bytes;
2450 	int ret;
2451 
2452 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2453 	if (vcpu->arch.pio.in)
2454 		ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2455 	else
2456 		ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2457 	return ret;
2458 }
2459 
2460 int complete_pio(struct kvm_vcpu *vcpu)
2461 {
2462 	struct kvm_pio_request *io = &vcpu->arch.pio;
2463 	long delta;
2464 	int r;
2465 	unsigned long val;
2466 
2467 	if (!io->string) {
2468 		if (io->in) {
2469 			val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2470 			memcpy(&val, vcpu->arch.pio_data, io->size);
2471 			kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2472 		}
2473 	} else {
2474 		if (io->in) {
2475 			r = pio_copy_data(vcpu);
2476 			if (r)
2477 				return r;
2478 		}
2479 
2480 		delta = 1;
2481 		if (io->rep) {
2482 			delta *= io->cur_count;
2483 			/*
2484 			 * The size of the register should really depend on
2485 			 * current address size.
2486 			 */
2487 			val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2488 			val -= delta;
2489 			kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2490 		}
2491 		if (io->down)
2492 			delta = -delta;
2493 		delta *= io->size;
2494 		if (io->in) {
2495 			val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2496 			val += delta;
2497 			kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2498 		} else {
2499 			val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2500 			val += delta;
2501 			kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2502 		}
2503 	}
2504 
2505 	io->count -= io->cur_count;
2506 	io->cur_count = 0;
2507 
2508 	return 0;
2509 }
2510 
2511 static void kernel_pio(struct kvm_io_device *pio_dev,
2512 		       struct kvm_vcpu *vcpu,
2513 		       void *pd)
2514 {
2515 	/* TODO: String I/O for in kernel device */
2516 
2517 	mutex_lock(&vcpu->kvm->lock);
2518 	if (vcpu->arch.pio.in)
2519 		kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2520 				  vcpu->arch.pio.size,
2521 				  pd);
2522 	else
2523 		kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2524 				   vcpu->arch.pio.size,
2525 				   pd);
2526 	mutex_unlock(&vcpu->kvm->lock);
2527 }
2528 
2529 static void pio_string_write(struct kvm_io_device *pio_dev,
2530 			     struct kvm_vcpu *vcpu)
2531 {
2532 	struct kvm_pio_request *io = &vcpu->arch.pio;
2533 	void *pd = vcpu->arch.pio_data;
2534 	int i;
2535 
2536 	mutex_lock(&vcpu->kvm->lock);
2537 	for (i = 0; i < io->cur_count; i++) {
2538 		kvm_iodevice_write(pio_dev, io->port,
2539 				   io->size,
2540 				   pd);
2541 		pd += io->size;
2542 	}
2543 	mutex_unlock(&vcpu->kvm->lock);
2544 }
2545 
2546 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2547 					       gpa_t addr, int len,
2548 					       int is_write)
2549 {
2550 	return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2551 }
2552 
2553 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2554 		  int size, unsigned port)
2555 {
2556 	struct kvm_io_device *pio_dev;
2557 	unsigned long val;
2558 
2559 	vcpu->run->exit_reason = KVM_EXIT_IO;
2560 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2561 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2562 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2563 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2564 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2565 	vcpu->arch.pio.in = in;
2566 	vcpu->arch.pio.string = 0;
2567 	vcpu->arch.pio.down = 0;
2568 	vcpu->arch.pio.rep = 0;
2569 
2570 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2571 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2572 			    handler);
2573 	else
2574 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2575 			    handler);
2576 
2577 	val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2578 	memcpy(vcpu->arch.pio_data, &val, 4);
2579 
2580 	pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2581 	if (pio_dev) {
2582 		kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2583 		complete_pio(vcpu);
2584 		return 1;
2585 	}
2586 	return 0;
2587 }
2588 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2589 
2590 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2591 		  int size, unsigned long count, int down,
2592 		  gva_t address, int rep, unsigned port)
2593 {
2594 	unsigned now, in_page;
2595 	int ret = 0;
2596 	struct kvm_io_device *pio_dev;
2597 
2598 	vcpu->run->exit_reason = KVM_EXIT_IO;
2599 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2600 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2601 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2602 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2603 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2604 	vcpu->arch.pio.in = in;
2605 	vcpu->arch.pio.string = 1;
2606 	vcpu->arch.pio.down = down;
2607 	vcpu->arch.pio.rep = rep;
2608 
2609 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2610 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2611 			    handler);
2612 	else
2613 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2614 			    handler);
2615 
2616 	if (!count) {
2617 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2618 		return 1;
2619 	}
2620 
2621 	if (!down)
2622 		in_page = PAGE_SIZE - offset_in_page(address);
2623 	else
2624 		in_page = offset_in_page(address) + size;
2625 	now = min(count, (unsigned long)in_page / size);
2626 	if (!now)
2627 		now = 1;
2628 	if (down) {
2629 		/*
2630 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2631 		 */
2632 		pr_unimpl(vcpu, "guest string pio down\n");
2633 		kvm_inject_gp(vcpu, 0);
2634 		return 1;
2635 	}
2636 	vcpu->run->io.count = now;
2637 	vcpu->arch.pio.cur_count = now;
2638 
2639 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2640 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2641 
2642 	vcpu->arch.pio.guest_gva = address;
2643 
2644 	pio_dev = vcpu_find_pio_dev(vcpu, port,
2645 				    vcpu->arch.pio.cur_count,
2646 				    !vcpu->arch.pio.in);
2647 	if (!vcpu->arch.pio.in) {
2648 		/* string PIO write */
2649 		ret = pio_copy_data(vcpu);
2650 		if (ret == X86EMUL_PROPAGATE_FAULT) {
2651 			kvm_inject_gp(vcpu, 0);
2652 			return 1;
2653 		}
2654 		if (ret == 0 && pio_dev) {
2655 			pio_string_write(pio_dev, vcpu);
2656 			complete_pio(vcpu);
2657 			if (vcpu->arch.pio.count == 0)
2658 				ret = 1;
2659 		}
2660 	} else if (pio_dev)
2661 		pr_unimpl(vcpu, "no string pio read support yet, "
2662 		       "port %x size %d count %ld\n",
2663 			port, size, count);
2664 
2665 	return ret;
2666 }
2667 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2668 
2669 static void bounce_off(void *info)
2670 {
2671 	/* nothing */
2672 }
2673 
2674 static unsigned int  ref_freq;
2675 static unsigned long tsc_khz_ref;
2676 
2677 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2678 				     void *data)
2679 {
2680 	struct cpufreq_freqs *freq = data;
2681 	struct kvm *kvm;
2682 	struct kvm_vcpu *vcpu;
2683 	int i, send_ipi = 0;
2684 
2685 	if (!ref_freq)
2686 		ref_freq = freq->old;
2687 
2688 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2689 		return 0;
2690 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2691 		return 0;
2692 	per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2693 
2694 	spin_lock(&kvm_lock);
2695 	list_for_each_entry(kvm, &vm_list, vm_list) {
2696 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2697 			vcpu = kvm->vcpus[i];
2698 			if (!vcpu)
2699 				continue;
2700 			if (vcpu->cpu != freq->cpu)
2701 				continue;
2702 			if (!kvm_request_guest_time_update(vcpu))
2703 				continue;
2704 			if (vcpu->cpu != smp_processor_id())
2705 				send_ipi++;
2706 		}
2707 	}
2708 	spin_unlock(&kvm_lock);
2709 
2710 	if (freq->old < freq->new && send_ipi) {
2711 		/*
2712 		 * We upscale the frequency.  Must make the guest
2713 		 * doesn't see old kvmclock values while running with
2714 		 * the new frequency, otherwise we risk the guest sees
2715 		 * time go backwards.
2716 		 *
2717 		 * In case we update the frequency for another cpu
2718 		 * (which might be in guest context) send an interrupt
2719 		 * to kick the cpu out of guest context.  Next time
2720 		 * guest context is entered kvmclock will be updated,
2721 		 * so the guest will not see stale values.
2722 		 */
2723 		smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2724 	}
2725 	return 0;
2726 }
2727 
2728 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2729         .notifier_call  = kvmclock_cpufreq_notifier
2730 };
2731 
2732 int kvm_arch_init(void *opaque)
2733 {
2734 	int r, cpu;
2735 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2736 
2737 	if (kvm_x86_ops) {
2738 		printk(KERN_ERR "kvm: already loaded the other module\n");
2739 		r = -EEXIST;
2740 		goto out;
2741 	}
2742 
2743 	if (!ops->cpu_has_kvm_support()) {
2744 		printk(KERN_ERR "kvm: no hardware support\n");
2745 		r = -EOPNOTSUPP;
2746 		goto out;
2747 	}
2748 	if (ops->disabled_by_bios()) {
2749 		printk(KERN_ERR "kvm: disabled by bios\n");
2750 		r = -EOPNOTSUPP;
2751 		goto out;
2752 	}
2753 
2754 	r = kvm_mmu_module_init();
2755 	if (r)
2756 		goto out;
2757 
2758 	kvm_init_msr_list();
2759 
2760 	kvm_x86_ops = ops;
2761 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2762 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2763 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2764 			PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2765 
2766 	for_each_possible_cpu(cpu)
2767 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2768 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2769 		tsc_khz_ref = tsc_khz;
2770 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2771 					  CPUFREQ_TRANSITION_NOTIFIER);
2772 	}
2773 
2774 	return 0;
2775 
2776 out:
2777 	return r;
2778 }
2779 
2780 void kvm_arch_exit(void)
2781 {
2782 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2783 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2784 					    CPUFREQ_TRANSITION_NOTIFIER);
2785 	kvm_x86_ops = NULL;
2786 	kvm_mmu_module_exit();
2787 }
2788 
2789 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2790 {
2791 	++vcpu->stat.halt_exits;
2792 	KVMTRACE_0D(HLT, vcpu, handler);
2793 	if (irqchip_in_kernel(vcpu->kvm)) {
2794 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2795 		return 1;
2796 	} else {
2797 		vcpu->run->exit_reason = KVM_EXIT_HLT;
2798 		return 0;
2799 	}
2800 }
2801 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2802 
2803 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2804 			   unsigned long a1)
2805 {
2806 	if (is_long_mode(vcpu))
2807 		return a0;
2808 	else
2809 		return a0 | ((gpa_t)a1 << 32);
2810 }
2811 
2812 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2813 {
2814 	unsigned long nr, a0, a1, a2, a3, ret;
2815 	int r = 1;
2816 
2817 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2818 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2819 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2820 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2821 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2822 
2823 	KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2824 
2825 	if (!is_long_mode(vcpu)) {
2826 		nr &= 0xFFFFFFFF;
2827 		a0 &= 0xFFFFFFFF;
2828 		a1 &= 0xFFFFFFFF;
2829 		a2 &= 0xFFFFFFFF;
2830 		a3 &= 0xFFFFFFFF;
2831 	}
2832 
2833 	switch (nr) {
2834 	case KVM_HC_VAPIC_POLL_IRQ:
2835 		ret = 0;
2836 		break;
2837 	case KVM_HC_MMU_OP:
2838 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2839 		break;
2840 	default:
2841 		ret = -KVM_ENOSYS;
2842 		break;
2843 	}
2844 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2845 	++vcpu->stat.hypercalls;
2846 	return r;
2847 }
2848 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2849 
2850 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2851 {
2852 	char instruction[3];
2853 	int ret = 0;
2854 	unsigned long rip = kvm_rip_read(vcpu);
2855 
2856 
2857 	/*
2858 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
2859 	 * to ensure that the updated hypercall appears atomically across all
2860 	 * VCPUs.
2861 	 */
2862 	kvm_mmu_zap_all(vcpu->kvm);
2863 
2864 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
2865 	if (emulator_write_emulated(rip, instruction, 3, vcpu)
2866 	    != X86EMUL_CONTINUE)
2867 		ret = -EFAULT;
2868 
2869 	return ret;
2870 }
2871 
2872 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2873 {
2874 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2875 }
2876 
2877 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2878 {
2879 	struct descriptor_table dt = { limit, base };
2880 
2881 	kvm_x86_ops->set_gdt(vcpu, &dt);
2882 }
2883 
2884 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2885 {
2886 	struct descriptor_table dt = { limit, base };
2887 
2888 	kvm_x86_ops->set_idt(vcpu, &dt);
2889 }
2890 
2891 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2892 		   unsigned long *rflags)
2893 {
2894 	kvm_lmsw(vcpu, msw);
2895 	*rflags = kvm_x86_ops->get_rflags(vcpu);
2896 }
2897 
2898 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2899 {
2900 	unsigned long value;
2901 
2902 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2903 	switch (cr) {
2904 	case 0:
2905 		value = vcpu->arch.cr0;
2906 		break;
2907 	case 2:
2908 		value = vcpu->arch.cr2;
2909 		break;
2910 	case 3:
2911 		value = vcpu->arch.cr3;
2912 		break;
2913 	case 4:
2914 		value = vcpu->arch.cr4;
2915 		break;
2916 	case 8:
2917 		value = kvm_get_cr8(vcpu);
2918 		break;
2919 	default:
2920 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2921 		return 0;
2922 	}
2923 	KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2924 		    (u32)((u64)value >> 32), handler);
2925 
2926 	return value;
2927 }
2928 
2929 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2930 		     unsigned long *rflags)
2931 {
2932 	KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2933 		    (u32)((u64)val >> 32), handler);
2934 
2935 	switch (cr) {
2936 	case 0:
2937 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2938 		*rflags = kvm_x86_ops->get_rflags(vcpu);
2939 		break;
2940 	case 2:
2941 		vcpu->arch.cr2 = val;
2942 		break;
2943 	case 3:
2944 		kvm_set_cr3(vcpu, val);
2945 		break;
2946 	case 4:
2947 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2948 		break;
2949 	case 8:
2950 		kvm_set_cr8(vcpu, val & 0xfUL);
2951 		break;
2952 	default:
2953 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2954 	}
2955 }
2956 
2957 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2958 {
2959 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2960 	int j, nent = vcpu->arch.cpuid_nent;
2961 
2962 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2963 	/* when no next entry is found, the current entry[i] is reselected */
2964 	for (j = i + 1; ; j = (j + 1) % nent) {
2965 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2966 		if (ej->function == e->function) {
2967 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2968 			return j;
2969 		}
2970 	}
2971 	return 0; /* silence gcc, even though control never reaches here */
2972 }
2973 
2974 /* find an entry with matching function, matching index (if needed), and that
2975  * should be read next (if it's stateful) */
2976 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2977 	u32 function, u32 index)
2978 {
2979 	if (e->function != function)
2980 		return 0;
2981 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2982 		return 0;
2983 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2984 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2985 		return 0;
2986 	return 1;
2987 }
2988 
2989 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2990 					      u32 function, u32 index)
2991 {
2992 	int i;
2993 	struct kvm_cpuid_entry2 *best = NULL;
2994 
2995 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2996 		struct kvm_cpuid_entry2 *e;
2997 
2998 		e = &vcpu->arch.cpuid_entries[i];
2999 		if (is_matching_cpuid_entry(e, function, index)) {
3000 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3001 				move_to_next_stateful_cpuid_entry(vcpu, i);
3002 			best = e;
3003 			break;
3004 		}
3005 		/*
3006 		 * Both basic or both extended?
3007 		 */
3008 		if (((e->function ^ function) & 0x80000000) == 0)
3009 			if (!best || e->function > best->function)
3010 				best = e;
3011 	}
3012 	return best;
3013 }
3014 
3015 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3016 {
3017 	u32 function, index;
3018 	struct kvm_cpuid_entry2 *best;
3019 
3020 	function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3021 	index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3022 	kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3023 	kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3024 	kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3025 	kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3026 	best = kvm_find_cpuid_entry(vcpu, function, index);
3027 	if (best) {
3028 		kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3029 		kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3030 		kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3031 		kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3032 	}
3033 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3034 	KVMTRACE_5D(CPUID, vcpu, function,
3035 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3036 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3037 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3038 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3039 }
3040 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3041 
3042 /*
3043  * Check if userspace requested an interrupt window, and that the
3044  * interrupt window is open.
3045  *
3046  * No need to exit to userspace if we already have an interrupt queued.
3047  */
3048 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3049 					  struct kvm_run *kvm_run)
3050 {
3051 	return (!vcpu->arch.irq_summary &&
3052 		kvm_run->request_interrupt_window &&
3053 		vcpu->arch.interrupt_window_open &&
3054 		(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3055 }
3056 
3057 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3058 			      struct kvm_run *kvm_run)
3059 {
3060 	kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3061 	kvm_run->cr8 = kvm_get_cr8(vcpu);
3062 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
3063 	if (irqchip_in_kernel(vcpu->kvm))
3064 		kvm_run->ready_for_interrupt_injection = 1;
3065 	else
3066 		kvm_run->ready_for_interrupt_injection =
3067 					(vcpu->arch.interrupt_window_open &&
3068 					 vcpu->arch.irq_summary == 0);
3069 }
3070 
3071 static void vapic_enter(struct kvm_vcpu *vcpu)
3072 {
3073 	struct kvm_lapic *apic = vcpu->arch.apic;
3074 	struct page *page;
3075 
3076 	if (!apic || !apic->vapic_addr)
3077 		return;
3078 
3079 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3080 
3081 	vcpu->arch.apic->vapic_page = page;
3082 }
3083 
3084 static void vapic_exit(struct kvm_vcpu *vcpu)
3085 {
3086 	struct kvm_lapic *apic = vcpu->arch.apic;
3087 
3088 	if (!apic || !apic->vapic_addr)
3089 		return;
3090 
3091 	down_read(&vcpu->kvm->slots_lock);
3092 	kvm_release_page_dirty(apic->vapic_page);
3093 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3094 	up_read(&vcpu->kvm->slots_lock);
3095 }
3096 
3097 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3098 {
3099 	int r;
3100 
3101 	if (vcpu->requests)
3102 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3103 			kvm_mmu_unload(vcpu);
3104 
3105 	r = kvm_mmu_reload(vcpu);
3106 	if (unlikely(r))
3107 		goto out;
3108 
3109 	if (vcpu->requests) {
3110 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3111 			__kvm_migrate_timers(vcpu);
3112 		if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3113 			kvm_write_guest_time(vcpu);
3114 		if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3115 			kvm_mmu_sync_roots(vcpu);
3116 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3117 			kvm_x86_ops->tlb_flush(vcpu);
3118 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3119 				       &vcpu->requests)) {
3120 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3121 			r = 0;
3122 			goto out;
3123 		}
3124 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3125 			kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3126 			r = 0;
3127 			goto out;
3128 		}
3129 	}
3130 
3131 	clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3132 	kvm_inject_pending_timer_irqs(vcpu);
3133 
3134 	preempt_disable();
3135 
3136 	kvm_x86_ops->prepare_guest_switch(vcpu);
3137 	kvm_load_guest_fpu(vcpu);
3138 
3139 	local_irq_disable();
3140 
3141 	if (vcpu->requests || need_resched() || signal_pending(current)) {
3142 		local_irq_enable();
3143 		preempt_enable();
3144 		r = 1;
3145 		goto out;
3146 	}
3147 
3148 	vcpu->guest_mode = 1;
3149 	/*
3150 	 * Make sure that guest_mode assignment won't happen after
3151 	 * testing the pending IRQ vector bitmap.
3152 	 */
3153 	smp_wmb();
3154 
3155 	if (vcpu->arch.exception.pending)
3156 		__queue_exception(vcpu);
3157 	else if (irqchip_in_kernel(vcpu->kvm))
3158 		kvm_x86_ops->inject_pending_irq(vcpu);
3159 	else
3160 		kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3161 
3162 	kvm_lapic_sync_to_vapic(vcpu);
3163 
3164 	up_read(&vcpu->kvm->slots_lock);
3165 
3166 	kvm_guest_enter();
3167 
3168 	get_debugreg(vcpu->arch.host_dr6, 6);
3169 	get_debugreg(vcpu->arch.host_dr7, 7);
3170 	if (unlikely(vcpu->arch.switch_db_regs)) {
3171 		get_debugreg(vcpu->arch.host_db[0], 0);
3172 		get_debugreg(vcpu->arch.host_db[1], 1);
3173 		get_debugreg(vcpu->arch.host_db[2], 2);
3174 		get_debugreg(vcpu->arch.host_db[3], 3);
3175 
3176 		set_debugreg(0, 7);
3177 		set_debugreg(vcpu->arch.eff_db[0], 0);
3178 		set_debugreg(vcpu->arch.eff_db[1], 1);
3179 		set_debugreg(vcpu->arch.eff_db[2], 2);
3180 		set_debugreg(vcpu->arch.eff_db[3], 3);
3181 	}
3182 
3183 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3184 	kvm_x86_ops->run(vcpu, kvm_run);
3185 
3186 	if (unlikely(vcpu->arch.switch_db_regs)) {
3187 		set_debugreg(0, 7);
3188 		set_debugreg(vcpu->arch.host_db[0], 0);
3189 		set_debugreg(vcpu->arch.host_db[1], 1);
3190 		set_debugreg(vcpu->arch.host_db[2], 2);
3191 		set_debugreg(vcpu->arch.host_db[3], 3);
3192 	}
3193 	set_debugreg(vcpu->arch.host_dr6, 6);
3194 	set_debugreg(vcpu->arch.host_dr7, 7);
3195 
3196 	vcpu->guest_mode = 0;
3197 	local_irq_enable();
3198 
3199 	++vcpu->stat.exits;
3200 
3201 	/*
3202 	 * We must have an instruction between local_irq_enable() and
3203 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3204 	 * the interrupt shadow.  The stat.exits increment will do nicely.
3205 	 * But we need to prevent reordering, hence this barrier():
3206 	 */
3207 	barrier();
3208 
3209 	kvm_guest_exit();
3210 
3211 	preempt_enable();
3212 
3213 	down_read(&vcpu->kvm->slots_lock);
3214 
3215 	/*
3216 	 * Profile KVM exit RIPs:
3217 	 */
3218 	if (unlikely(prof_on == KVM_PROFILING)) {
3219 		unsigned long rip = kvm_rip_read(vcpu);
3220 		profile_hit(KVM_PROFILING, (void *)rip);
3221 	}
3222 
3223 	if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3224 		vcpu->arch.exception.pending = false;
3225 
3226 	kvm_lapic_sync_from_vapic(vcpu);
3227 
3228 	r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3229 out:
3230 	return r;
3231 }
3232 
3233 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3234 {
3235 	int r;
3236 
3237 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3238 		pr_debug("vcpu %d received sipi with vector # %x\n",
3239 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3240 		kvm_lapic_reset(vcpu);
3241 		r = kvm_arch_vcpu_reset(vcpu);
3242 		if (r)
3243 			return r;
3244 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3245 	}
3246 
3247 	down_read(&vcpu->kvm->slots_lock);
3248 	vapic_enter(vcpu);
3249 
3250 	r = 1;
3251 	while (r > 0) {
3252 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3253 			r = vcpu_enter_guest(vcpu, kvm_run);
3254 		else {
3255 			up_read(&vcpu->kvm->slots_lock);
3256 			kvm_vcpu_block(vcpu);
3257 			down_read(&vcpu->kvm->slots_lock);
3258 			if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3259 				if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3260 					vcpu->arch.mp_state =
3261 							KVM_MP_STATE_RUNNABLE;
3262 			if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3263 				r = -EINTR;
3264 		}
3265 
3266 		if (r > 0) {
3267 			if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3268 				r = -EINTR;
3269 				kvm_run->exit_reason = KVM_EXIT_INTR;
3270 				++vcpu->stat.request_irq_exits;
3271 			}
3272 			if (signal_pending(current)) {
3273 				r = -EINTR;
3274 				kvm_run->exit_reason = KVM_EXIT_INTR;
3275 				++vcpu->stat.signal_exits;
3276 			}
3277 			if (need_resched()) {
3278 				up_read(&vcpu->kvm->slots_lock);
3279 				kvm_resched(vcpu);
3280 				down_read(&vcpu->kvm->slots_lock);
3281 			}
3282 		}
3283 	}
3284 
3285 	up_read(&vcpu->kvm->slots_lock);
3286 	post_kvm_run_save(vcpu, kvm_run);
3287 
3288 	vapic_exit(vcpu);
3289 
3290 	return r;
3291 }
3292 
3293 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3294 {
3295 	int r;
3296 	sigset_t sigsaved;
3297 
3298 	vcpu_load(vcpu);
3299 
3300 	if (vcpu->sigset_active)
3301 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3302 
3303 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3304 		kvm_vcpu_block(vcpu);
3305 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3306 		r = -EAGAIN;
3307 		goto out;
3308 	}
3309 
3310 	/* re-sync apic's tpr */
3311 	if (!irqchip_in_kernel(vcpu->kvm))
3312 		kvm_set_cr8(vcpu, kvm_run->cr8);
3313 
3314 	if (vcpu->arch.pio.cur_count) {
3315 		r = complete_pio(vcpu);
3316 		if (r)
3317 			goto out;
3318 	}
3319 #if CONFIG_HAS_IOMEM
3320 	if (vcpu->mmio_needed) {
3321 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3322 		vcpu->mmio_read_completed = 1;
3323 		vcpu->mmio_needed = 0;
3324 
3325 		down_read(&vcpu->kvm->slots_lock);
3326 		r = emulate_instruction(vcpu, kvm_run,
3327 					vcpu->arch.mmio_fault_cr2, 0,
3328 					EMULTYPE_NO_DECODE);
3329 		up_read(&vcpu->kvm->slots_lock);
3330 		if (r == EMULATE_DO_MMIO) {
3331 			/*
3332 			 * Read-modify-write.  Back to userspace.
3333 			 */
3334 			r = 0;
3335 			goto out;
3336 		}
3337 	}
3338 #endif
3339 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3340 		kvm_register_write(vcpu, VCPU_REGS_RAX,
3341 				     kvm_run->hypercall.ret);
3342 
3343 	r = __vcpu_run(vcpu, kvm_run);
3344 
3345 out:
3346 	if (vcpu->sigset_active)
3347 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3348 
3349 	vcpu_put(vcpu);
3350 	return r;
3351 }
3352 
3353 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3354 {
3355 	vcpu_load(vcpu);
3356 
3357 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3358 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3359 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3360 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3361 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3362 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3363 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3364 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3365 #ifdef CONFIG_X86_64
3366 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3367 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3368 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3369 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3370 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3371 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3372 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3373 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3374 #endif
3375 
3376 	regs->rip = kvm_rip_read(vcpu);
3377 	regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3378 
3379 	/*
3380 	 * Don't leak debug flags in case they were set for guest debugging
3381 	 */
3382 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3383 		regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3384 
3385 	vcpu_put(vcpu);
3386 
3387 	return 0;
3388 }
3389 
3390 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3391 {
3392 	vcpu_load(vcpu);
3393 
3394 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3395 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3396 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3397 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3398 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3399 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3400 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3401 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3402 #ifdef CONFIG_X86_64
3403 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3404 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3405 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3406 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3407 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3408 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3409 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3410 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3411 
3412 #endif
3413 
3414 	kvm_rip_write(vcpu, regs->rip);
3415 	kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3416 
3417 
3418 	vcpu->arch.exception.pending = false;
3419 
3420 	vcpu_put(vcpu);
3421 
3422 	return 0;
3423 }
3424 
3425 void kvm_get_segment(struct kvm_vcpu *vcpu,
3426 		     struct kvm_segment *var, int seg)
3427 {
3428 	kvm_x86_ops->get_segment(vcpu, var, seg);
3429 }
3430 
3431 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3432 {
3433 	struct kvm_segment cs;
3434 
3435 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3436 	*db = cs.db;
3437 	*l = cs.l;
3438 }
3439 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3440 
3441 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3442 				  struct kvm_sregs *sregs)
3443 {
3444 	struct descriptor_table dt;
3445 	int pending_vec;
3446 
3447 	vcpu_load(vcpu);
3448 
3449 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3450 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3451 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3452 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3453 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3454 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3455 
3456 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3457 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3458 
3459 	kvm_x86_ops->get_idt(vcpu, &dt);
3460 	sregs->idt.limit = dt.limit;
3461 	sregs->idt.base = dt.base;
3462 	kvm_x86_ops->get_gdt(vcpu, &dt);
3463 	sregs->gdt.limit = dt.limit;
3464 	sregs->gdt.base = dt.base;
3465 
3466 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3467 	sregs->cr0 = vcpu->arch.cr0;
3468 	sregs->cr2 = vcpu->arch.cr2;
3469 	sregs->cr3 = vcpu->arch.cr3;
3470 	sregs->cr4 = vcpu->arch.cr4;
3471 	sregs->cr8 = kvm_get_cr8(vcpu);
3472 	sregs->efer = vcpu->arch.shadow_efer;
3473 	sregs->apic_base = kvm_get_apic_base(vcpu);
3474 
3475 	if (irqchip_in_kernel(vcpu->kvm)) {
3476 		memset(sregs->interrupt_bitmap, 0,
3477 		       sizeof sregs->interrupt_bitmap);
3478 		pending_vec = kvm_x86_ops->get_irq(vcpu);
3479 		if (pending_vec >= 0)
3480 			set_bit(pending_vec,
3481 				(unsigned long *)sregs->interrupt_bitmap);
3482 	} else
3483 		memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3484 		       sizeof sregs->interrupt_bitmap);
3485 
3486 	vcpu_put(vcpu);
3487 
3488 	return 0;
3489 }
3490 
3491 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3492 				    struct kvm_mp_state *mp_state)
3493 {
3494 	vcpu_load(vcpu);
3495 	mp_state->mp_state = vcpu->arch.mp_state;
3496 	vcpu_put(vcpu);
3497 	return 0;
3498 }
3499 
3500 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3501 				    struct kvm_mp_state *mp_state)
3502 {
3503 	vcpu_load(vcpu);
3504 	vcpu->arch.mp_state = mp_state->mp_state;
3505 	vcpu_put(vcpu);
3506 	return 0;
3507 }
3508 
3509 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3510 			struct kvm_segment *var, int seg)
3511 {
3512 	kvm_x86_ops->set_segment(vcpu, var, seg);
3513 }
3514 
3515 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3516 				   struct kvm_segment *kvm_desct)
3517 {
3518 	kvm_desct->base = seg_desc->base0;
3519 	kvm_desct->base |= seg_desc->base1 << 16;
3520 	kvm_desct->base |= seg_desc->base2 << 24;
3521 	kvm_desct->limit = seg_desc->limit0;
3522 	kvm_desct->limit |= seg_desc->limit << 16;
3523 	if (seg_desc->g) {
3524 		kvm_desct->limit <<= 12;
3525 		kvm_desct->limit |= 0xfff;
3526 	}
3527 	kvm_desct->selector = selector;
3528 	kvm_desct->type = seg_desc->type;
3529 	kvm_desct->present = seg_desc->p;
3530 	kvm_desct->dpl = seg_desc->dpl;
3531 	kvm_desct->db = seg_desc->d;
3532 	kvm_desct->s = seg_desc->s;
3533 	kvm_desct->l = seg_desc->l;
3534 	kvm_desct->g = seg_desc->g;
3535 	kvm_desct->avl = seg_desc->avl;
3536 	if (!selector)
3537 		kvm_desct->unusable = 1;
3538 	else
3539 		kvm_desct->unusable = 0;
3540 	kvm_desct->padding = 0;
3541 }
3542 
3543 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3544 					  u16 selector,
3545 					  struct descriptor_table *dtable)
3546 {
3547 	if (selector & 1 << 2) {
3548 		struct kvm_segment kvm_seg;
3549 
3550 		kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3551 
3552 		if (kvm_seg.unusable)
3553 			dtable->limit = 0;
3554 		else
3555 			dtable->limit = kvm_seg.limit;
3556 		dtable->base = kvm_seg.base;
3557 	}
3558 	else
3559 		kvm_x86_ops->get_gdt(vcpu, dtable);
3560 }
3561 
3562 /* allowed just for 8 bytes segments */
3563 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3564 					 struct desc_struct *seg_desc)
3565 {
3566 	gpa_t gpa;
3567 	struct descriptor_table dtable;
3568 	u16 index = selector >> 3;
3569 
3570 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
3571 
3572 	if (dtable.limit < index * 8 + 7) {
3573 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3574 		return 1;
3575 	}
3576 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3577 	gpa += index * 8;
3578 	return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3579 }
3580 
3581 /* allowed just for 8 bytes segments */
3582 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3583 					 struct desc_struct *seg_desc)
3584 {
3585 	gpa_t gpa;
3586 	struct descriptor_table dtable;
3587 	u16 index = selector >> 3;
3588 
3589 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
3590 
3591 	if (dtable.limit < index * 8 + 7)
3592 		return 1;
3593 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3594 	gpa += index * 8;
3595 	return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3596 }
3597 
3598 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3599 			     struct desc_struct *seg_desc)
3600 {
3601 	u32 base_addr;
3602 
3603 	base_addr = seg_desc->base0;
3604 	base_addr |= (seg_desc->base1 << 16);
3605 	base_addr |= (seg_desc->base2 << 24);
3606 
3607 	return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3608 }
3609 
3610 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3611 {
3612 	struct kvm_segment kvm_seg;
3613 
3614 	kvm_get_segment(vcpu, &kvm_seg, seg);
3615 	return kvm_seg.selector;
3616 }
3617 
3618 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3619 						u16 selector,
3620 						struct kvm_segment *kvm_seg)
3621 {
3622 	struct desc_struct seg_desc;
3623 
3624 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3625 		return 1;
3626 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3627 	return 0;
3628 }
3629 
3630 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3631 {
3632 	struct kvm_segment segvar = {
3633 		.base = selector << 4,
3634 		.limit = 0xffff,
3635 		.selector = selector,
3636 		.type = 3,
3637 		.present = 1,
3638 		.dpl = 3,
3639 		.db = 0,
3640 		.s = 1,
3641 		.l = 0,
3642 		.g = 0,
3643 		.avl = 0,
3644 		.unusable = 0,
3645 	};
3646 	kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3647 	return 0;
3648 }
3649 
3650 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3651 				int type_bits, int seg)
3652 {
3653 	struct kvm_segment kvm_seg;
3654 
3655 	if (!(vcpu->arch.cr0 & X86_CR0_PE))
3656 		return kvm_load_realmode_segment(vcpu, selector, seg);
3657 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3658 		return 1;
3659 	kvm_seg.type |= type_bits;
3660 
3661 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3662 	    seg != VCPU_SREG_LDTR)
3663 		if (!kvm_seg.s)
3664 			kvm_seg.unusable = 1;
3665 
3666 	kvm_set_segment(vcpu, &kvm_seg, seg);
3667 	return 0;
3668 }
3669 
3670 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3671 				struct tss_segment_32 *tss)
3672 {
3673 	tss->cr3 = vcpu->arch.cr3;
3674 	tss->eip = kvm_rip_read(vcpu);
3675 	tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3676 	tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3677 	tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3678 	tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3679 	tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3680 	tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3681 	tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3682 	tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3683 	tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3684 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3685 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3686 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3687 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3688 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3689 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3690 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3691 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3692 }
3693 
3694 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3695 				  struct tss_segment_32 *tss)
3696 {
3697 	kvm_set_cr3(vcpu, tss->cr3);
3698 
3699 	kvm_rip_write(vcpu, tss->eip);
3700 	kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3701 
3702 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3703 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3704 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3705 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3706 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3707 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3708 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3709 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3710 
3711 	if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3712 		return 1;
3713 
3714 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3715 		return 1;
3716 
3717 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3718 		return 1;
3719 
3720 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3721 		return 1;
3722 
3723 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3724 		return 1;
3725 
3726 	if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3727 		return 1;
3728 
3729 	if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3730 		return 1;
3731 	return 0;
3732 }
3733 
3734 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3735 				struct tss_segment_16 *tss)
3736 {
3737 	tss->ip = kvm_rip_read(vcpu);
3738 	tss->flag = kvm_x86_ops->get_rflags(vcpu);
3739 	tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3740 	tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3741 	tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3742 	tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3743 	tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3744 	tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3745 	tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3746 	tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3747 
3748 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3749 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3750 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3751 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3752 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3753 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3754 }
3755 
3756 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3757 				 struct tss_segment_16 *tss)
3758 {
3759 	kvm_rip_write(vcpu, tss->ip);
3760 	kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3761 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3762 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3763 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3764 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3765 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3766 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3767 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3768 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3769 
3770 	if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3771 		return 1;
3772 
3773 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3774 		return 1;
3775 
3776 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3777 		return 1;
3778 
3779 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3780 		return 1;
3781 
3782 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3783 		return 1;
3784 	return 0;
3785 }
3786 
3787 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3788 		       u32 old_tss_base,
3789 		       struct desc_struct *nseg_desc)
3790 {
3791 	struct tss_segment_16 tss_segment_16;
3792 	int ret = 0;
3793 
3794 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3795 			   sizeof tss_segment_16))
3796 		goto out;
3797 
3798 	save_state_to_tss16(vcpu, &tss_segment_16);
3799 
3800 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3801 			    sizeof tss_segment_16))
3802 		goto out;
3803 
3804 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3805 			   &tss_segment_16, sizeof tss_segment_16))
3806 		goto out;
3807 
3808 	if (load_state_from_tss16(vcpu, &tss_segment_16))
3809 		goto out;
3810 
3811 	ret = 1;
3812 out:
3813 	return ret;
3814 }
3815 
3816 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3817 		       u32 old_tss_base,
3818 		       struct desc_struct *nseg_desc)
3819 {
3820 	struct tss_segment_32 tss_segment_32;
3821 	int ret = 0;
3822 
3823 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3824 			   sizeof tss_segment_32))
3825 		goto out;
3826 
3827 	save_state_to_tss32(vcpu, &tss_segment_32);
3828 
3829 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3830 			    sizeof tss_segment_32))
3831 		goto out;
3832 
3833 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3834 			   &tss_segment_32, sizeof tss_segment_32))
3835 		goto out;
3836 
3837 	if (load_state_from_tss32(vcpu, &tss_segment_32))
3838 		goto out;
3839 
3840 	ret = 1;
3841 out:
3842 	return ret;
3843 }
3844 
3845 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3846 {
3847 	struct kvm_segment tr_seg;
3848 	struct desc_struct cseg_desc;
3849 	struct desc_struct nseg_desc;
3850 	int ret = 0;
3851 	u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3852 	u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3853 
3854 	old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3855 
3856 	/* FIXME: Handle errors. Failure to read either TSS or their
3857 	 * descriptors should generate a pagefault.
3858 	 */
3859 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3860 		goto out;
3861 
3862 	if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3863 		goto out;
3864 
3865 	if (reason != TASK_SWITCH_IRET) {
3866 		int cpl;
3867 
3868 		cpl = kvm_x86_ops->get_cpl(vcpu);
3869 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3870 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3871 			return 1;
3872 		}
3873 	}
3874 
3875 	if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3876 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3877 		return 1;
3878 	}
3879 
3880 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3881 		cseg_desc.type &= ~(1 << 1); //clear the B flag
3882 		save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3883 	}
3884 
3885 	if (reason == TASK_SWITCH_IRET) {
3886 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3887 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3888 	}
3889 
3890 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3891 
3892 	if (nseg_desc.type & 8)
3893 		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3894 					 &nseg_desc);
3895 	else
3896 		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3897 					 &nseg_desc);
3898 
3899 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3900 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3901 		kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3902 	}
3903 
3904 	if (reason != TASK_SWITCH_IRET) {
3905 		nseg_desc.type |= (1 << 1);
3906 		save_guest_segment_descriptor(vcpu, tss_selector,
3907 					      &nseg_desc);
3908 	}
3909 
3910 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3911 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3912 	tr_seg.type = 11;
3913 	kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3914 out:
3915 	return ret;
3916 }
3917 EXPORT_SYMBOL_GPL(kvm_task_switch);
3918 
3919 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3920 				  struct kvm_sregs *sregs)
3921 {
3922 	int mmu_reset_needed = 0;
3923 	int i, pending_vec, max_bits;
3924 	struct descriptor_table dt;
3925 
3926 	vcpu_load(vcpu);
3927 
3928 	dt.limit = sregs->idt.limit;
3929 	dt.base = sregs->idt.base;
3930 	kvm_x86_ops->set_idt(vcpu, &dt);
3931 	dt.limit = sregs->gdt.limit;
3932 	dt.base = sregs->gdt.base;
3933 	kvm_x86_ops->set_gdt(vcpu, &dt);
3934 
3935 	vcpu->arch.cr2 = sregs->cr2;
3936 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3937 	vcpu->arch.cr3 = sregs->cr3;
3938 
3939 	kvm_set_cr8(vcpu, sregs->cr8);
3940 
3941 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3942 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
3943 	kvm_set_apic_base(vcpu, sregs->apic_base);
3944 
3945 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3946 
3947 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3948 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3949 	vcpu->arch.cr0 = sregs->cr0;
3950 
3951 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3952 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3953 	if (!is_long_mode(vcpu) && is_pae(vcpu))
3954 		load_pdptrs(vcpu, vcpu->arch.cr3);
3955 
3956 	if (mmu_reset_needed)
3957 		kvm_mmu_reset_context(vcpu);
3958 
3959 	if (!irqchip_in_kernel(vcpu->kvm)) {
3960 		memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3961 		       sizeof vcpu->arch.irq_pending);
3962 		vcpu->arch.irq_summary = 0;
3963 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3964 			if (vcpu->arch.irq_pending[i])
3965 				__set_bit(i, &vcpu->arch.irq_summary);
3966 	} else {
3967 		max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3968 		pending_vec = find_first_bit(
3969 			(const unsigned long *)sregs->interrupt_bitmap,
3970 			max_bits);
3971 		/* Only pending external irq is handled here */
3972 		if (pending_vec < max_bits) {
3973 			kvm_x86_ops->set_irq(vcpu, pending_vec);
3974 			pr_debug("Set back pending irq %d\n",
3975 				 pending_vec);
3976 		}
3977 		kvm_pic_clear_isr_ack(vcpu->kvm);
3978 	}
3979 
3980 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3981 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3982 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3983 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3984 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3985 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3986 
3987 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3988 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3989 
3990 	/* Older userspace won't unhalt the vcpu on reset. */
3991 	if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3992 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3993 	    !(vcpu->arch.cr0 & X86_CR0_PE))
3994 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3995 
3996 	vcpu_put(vcpu);
3997 
3998 	return 0;
3999 }
4000 
4001 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4002 					struct kvm_guest_debug *dbg)
4003 {
4004 	int i, r;
4005 
4006 	vcpu_load(vcpu);
4007 
4008 	if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4009 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4010 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
4011 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4012 		vcpu->arch.switch_db_regs =
4013 			(dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4014 	} else {
4015 		for (i = 0; i < KVM_NR_DB_REGS; i++)
4016 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4017 		vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4018 	}
4019 
4020 	r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4021 
4022 	if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4023 		kvm_queue_exception(vcpu, DB_VECTOR);
4024 	else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4025 		kvm_queue_exception(vcpu, BP_VECTOR);
4026 
4027 	vcpu_put(vcpu);
4028 
4029 	return r;
4030 }
4031 
4032 /*
4033  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4034  * we have asm/x86/processor.h
4035  */
4036 struct fxsave {
4037 	u16	cwd;
4038 	u16	swd;
4039 	u16	twd;
4040 	u16	fop;
4041 	u64	rip;
4042 	u64	rdp;
4043 	u32	mxcsr;
4044 	u32	mxcsr_mask;
4045 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
4046 #ifdef CONFIG_X86_64
4047 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
4048 #else
4049 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
4050 #endif
4051 };
4052 
4053 /*
4054  * Translate a guest virtual address to a guest physical address.
4055  */
4056 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4057 				    struct kvm_translation *tr)
4058 {
4059 	unsigned long vaddr = tr->linear_address;
4060 	gpa_t gpa;
4061 
4062 	vcpu_load(vcpu);
4063 	down_read(&vcpu->kvm->slots_lock);
4064 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4065 	up_read(&vcpu->kvm->slots_lock);
4066 	tr->physical_address = gpa;
4067 	tr->valid = gpa != UNMAPPED_GVA;
4068 	tr->writeable = 1;
4069 	tr->usermode = 0;
4070 	vcpu_put(vcpu);
4071 
4072 	return 0;
4073 }
4074 
4075 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4076 {
4077 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4078 
4079 	vcpu_load(vcpu);
4080 
4081 	memcpy(fpu->fpr, fxsave->st_space, 128);
4082 	fpu->fcw = fxsave->cwd;
4083 	fpu->fsw = fxsave->swd;
4084 	fpu->ftwx = fxsave->twd;
4085 	fpu->last_opcode = fxsave->fop;
4086 	fpu->last_ip = fxsave->rip;
4087 	fpu->last_dp = fxsave->rdp;
4088 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4089 
4090 	vcpu_put(vcpu);
4091 
4092 	return 0;
4093 }
4094 
4095 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4096 {
4097 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4098 
4099 	vcpu_load(vcpu);
4100 
4101 	memcpy(fxsave->st_space, fpu->fpr, 128);
4102 	fxsave->cwd = fpu->fcw;
4103 	fxsave->swd = fpu->fsw;
4104 	fxsave->twd = fpu->ftwx;
4105 	fxsave->fop = fpu->last_opcode;
4106 	fxsave->rip = fpu->last_ip;
4107 	fxsave->rdp = fpu->last_dp;
4108 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4109 
4110 	vcpu_put(vcpu);
4111 
4112 	return 0;
4113 }
4114 
4115 void fx_init(struct kvm_vcpu *vcpu)
4116 {
4117 	unsigned after_mxcsr_mask;
4118 
4119 	/*
4120 	 * Touch the fpu the first time in non atomic context as if
4121 	 * this is the first fpu instruction the exception handler
4122 	 * will fire before the instruction returns and it'll have to
4123 	 * allocate ram with GFP_KERNEL.
4124 	 */
4125 	if (!used_math())
4126 		kvm_fx_save(&vcpu->arch.host_fx_image);
4127 
4128 	/* Initialize guest FPU by resetting ours and saving into guest's */
4129 	preempt_disable();
4130 	kvm_fx_save(&vcpu->arch.host_fx_image);
4131 	kvm_fx_finit();
4132 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4133 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4134 	preempt_enable();
4135 
4136 	vcpu->arch.cr0 |= X86_CR0_ET;
4137 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4138 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4139 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4140 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4141 }
4142 EXPORT_SYMBOL_GPL(fx_init);
4143 
4144 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4145 {
4146 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4147 		return;
4148 
4149 	vcpu->guest_fpu_loaded = 1;
4150 	kvm_fx_save(&vcpu->arch.host_fx_image);
4151 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
4152 }
4153 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4154 
4155 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4156 {
4157 	if (!vcpu->guest_fpu_loaded)
4158 		return;
4159 
4160 	vcpu->guest_fpu_loaded = 0;
4161 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4162 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4163 	++vcpu->stat.fpu_reload;
4164 }
4165 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4166 
4167 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4168 {
4169 	if (vcpu->arch.time_page) {
4170 		kvm_release_page_dirty(vcpu->arch.time_page);
4171 		vcpu->arch.time_page = NULL;
4172 	}
4173 
4174 	kvm_x86_ops->vcpu_free(vcpu);
4175 }
4176 
4177 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4178 						unsigned int id)
4179 {
4180 	return kvm_x86_ops->vcpu_create(kvm, id);
4181 }
4182 
4183 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4184 {
4185 	int r;
4186 
4187 	/* We do fxsave: this must be aligned. */
4188 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4189 
4190 	vcpu->arch.mtrr_state.have_fixed = 1;
4191 	vcpu_load(vcpu);
4192 	r = kvm_arch_vcpu_reset(vcpu);
4193 	if (r == 0)
4194 		r = kvm_mmu_setup(vcpu);
4195 	vcpu_put(vcpu);
4196 	if (r < 0)
4197 		goto free_vcpu;
4198 
4199 	return 0;
4200 free_vcpu:
4201 	kvm_x86_ops->vcpu_free(vcpu);
4202 	return r;
4203 }
4204 
4205 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4206 {
4207 	vcpu_load(vcpu);
4208 	kvm_mmu_unload(vcpu);
4209 	vcpu_put(vcpu);
4210 
4211 	kvm_x86_ops->vcpu_free(vcpu);
4212 }
4213 
4214 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4215 {
4216 	vcpu->arch.nmi_pending = false;
4217 	vcpu->arch.nmi_injected = false;
4218 
4219 	vcpu->arch.switch_db_regs = 0;
4220 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4221 	vcpu->arch.dr6 = DR6_FIXED_1;
4222 	vcpu->arch.dr7 = DR7_FIXED_1;
4223 
4224 	return kvm_x86_ops->vcpu_reset(vcpu);
4225 }
4226 
4227 void kvm_arch_hardware_enable(void *garbage)
4228 {
4229 	kvm_x86_ops->hardware_enable(garbage);
4230 }
4231 
4232 void kvm_arch_hardware_disable(void *garbage)
4233 {
4234 	kvm_x86_ops->hardware_disable(garbage);
4235 }
4236 
4237 int kvm_arch_hardware_setup(void)
4238 {
4239 	return kvm_x86_ops->hardware_setup();
4240 }
4241 
4242 void kvm_arch_hardware_unsetup(void)
4243 {
4244 	kvm_x86_ops->hardware_unsetup();
4245 }
4246 
4247 void kvm_arch_check_processor_compat(void *rtn)
4248 {
4249 	kvm_x86_ops->check_processor_compatibility(rtn);
4250 }
4251 
4252 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4253 {
4254 	struct page *page;
4255 	struct kvm *kvm;
4256 	int r;
4257 
4258 	BUG_ON(vcpu->kvm == NULL);
4259 	kvm = vcpu->kvm;
4260 
4261 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4262 	if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4263 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4264 	else
4265 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4266 
4267 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4268 	if (!page) {
4269 		r = -ENOMEM;
4270 		goto fail;
4271 	}
4272 	vcpu->arch.pio_data = page_address(page);
4273 
4274 	r = kvm_mmu_create(vcpu);
4275 	if (r < 0)
4276 		goto fail_free_pio_data;
4277 
4278 	if (irqchip_in_kernel(kvm)) {
4279 		r = kvm_create_lapic(vcpu);
4280 		if (r < 0)
4281 			goto fail_mmu_destroy;
4282 	}
4283 
4284 	return 0;
4285 
4286 fail_mmu_destroy:
4287 	kvm_mmu_destroy(vcpu);
4288 fail_free_pio_data:
4289 	free_page((unsigned long)vcpu->arch.pio_data);
4290 fail:
4291 	return r;
4292 }
4293 
4294 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4295 {
4296 	kvm_free_lapic(vcpu);
4297 	down_read(&vcpu->kvm->slots_lock);
4298 	kvm_mmu_destroy(vcpu);
4299 	up_read(&vcpu->kvm->slots_lock);
4300 	free_page((unsigned long)vcpu->arch.pio_data);
4301 }
4302 
4303 struct  kvm *kvm_arch_create_vm(void)
4304 {
4305 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4306 
4307 	if (!kvm)
4308 		return ERR_PTR(-ENOMEM);
4309 
4310 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4311 	INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4312 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4313 
4314 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4315 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4316 
4317 	rdtscll(kvm->arch.vm_init_tsc);
4318 
4319 	return kvm;
4320 }
4321 
4322 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4323 {
4324 	vcpu_load(vcpu);
4325 	kvm_mmu_unload(vcpu);
4326 	vcpu_put(vcpu);
4327 }
4328 
4329 static void kvm_free_vcpus(struct kvm *kvm)
4330 {
4331 	unsigned int i;
4332 
4333 	/*
4334 	 * Unpin any mmu pages first.
4335 	 */
4336 	for (i = 0; i < KVM_MAX_VCPUS; ++i)
4337 		if (kvm->vcpus[i])
4338 			kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4339 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4340 		if (kvm->vcpus[i]) {
4341 			kvm_arch_vcpu_free(kvm->vcpus[i]);
4342 			kvm->vcpus[i] = NULL;
4343 		}
4344 	}
4345 
4346 }
4347 
4348 void kvm_arch_sync_events(struct kvm *kvm)
4349 {
4350 	kvm_free_all_assigned_devices(kvm);
4351 }
4352 
4353 void kvm_arch_destroy_vm(struct kvm *kvm)
4354 {
4355 	kvm_iommu_unmap_guest(kvm);
4356 	kvm_free_pit(kvm);
4357 	kfree(kvm->arch.vpic);
4358 	kfree(kvm->arch.vioapic);
4359 	kvm_free_vcpus(kvm);
4360 	kvm_free_physmem(kvm);
4361 	if (kvm->arch.apic_access_page)
4362 		put_page(kvm->arch.apic_access_page);
4363 	if (kvm->arch.ept_identity_pagetable)
4364 		put_page(kvm->arch.ept_identity_pagetable);
4365 	kfree(kvm);
4366 }
4367 
4368 int kvm_arch_set_memory_region(struct kvm *kvm,
4369 				struct kvm_userspace_memory_region *mem,
4370 				struct kvm_memory_slot old,
4371 				int user_alloc)
4372 {
4373 	int npages = mem->memory_size >> PAGE_SHIFT;
4374 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4375 
4376 	/*To keep backward compatibility with older userspace,
4377 	 *x86 needs to hanlde !user_alloc case.
4378 	 */
4379 	if (!user_alloc) {
4380 		if (npages && !old.rmap) {
4381 			unsigned long userspace_addr;
4382 
4383 			down_write(&current->mm->mmap_sem);
4384 			userspace_addr = do_mmap(NULL, 0,
4385 						 npages * PAGE_SIZE,
4386 						 PROT_READ | PROT_WRITE,
4387 						 MAP_PRIVATE | MAP_ANONYMOUS,
4388 						 0);
4389 			up_write(&current->mm->mmap_sem);
4390 
4391 			if (IS_ERR((void *)userspace_addr))
4392 				return PTR_ERR((void *)userspace_addr);
4393 
4394 			/* set userspace_addr atomically for kvm_hva_to_rmapp */
4395 			spin_lock(&kvm->mmu_lock);
4396 			memslot->userspace_addr = userspace_addr;
4397 			spin_unlock(&kvm->mmu_lock);
4398 		} else {
4399 			if (!old.user_alloc && old.rmap) {
4400 				int ret;
4401 
4402 				down_write(&current->mm->mmap_sem);
4403 				ret = do_munmap(current->mm, old.userspace_addr,
4404 						old.npages * PAGE_SIZE);
4405 				up_write(&current->mm->mmap_sem);
4406 				if (ret < 0)
4407 					printk(KERN_WARNING
4408 				       "kvm_vm_ioctl_set_memory_region: "
4409 				       "failed to munmap memory\n");
4410 			}
4411 		}
4412 	}
4413 
4414 	if (!kvm->arch.n_requested_mmu_pages) {
4415 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4416 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4417 	}
4418 
4419 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4420 	kvm_flush_remote_tlbs(kvm);
4421 
4422 	return 0;
4423 }
4424 
4425 void kvm_arch_flush_shadow(struct kvm *kvm)
4426 {
4427 	kvm_mmu_zap_all(kvm);
4428 }
4429 
4430 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4431 {
4432 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4433 	       || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4434 	       || vcpu->arch.nmi_pending;
4435 }
4436 
4437 static void vcpu_kick_intr(void *info)
4438 {
4439 #ifdef DEBUG
4440 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4441 	printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4442 #endif
4443 }
4444 
4445 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4446 {
4447 	int ipi_pcpu = vcpu->cpu;
4448 	int cpu = get_cpu();
4449 
4450 	if (waitqueue_active(&vcpu->wq)) {
4451 		wake_up_interruptible(&vcpu->wq);
4452 		++vcpu->stat.halt_wakeup;
4453 	}
4454 	/*
4455 	 * We may be called synchronously with irqs disabled in guest mode,
4456 	 * So need not to call smp_call_function_single() in that case.
4457 	 */
4458 	if (vcpu->guest_mode && vcpu->cpu != cpu)
4459 		smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4460 	put_cpu();
4461 }
4462