xref: /openbmc/linux/arch/x86/kvm/x86.c (revision afbcf7ab8d1bc8c2d04792f6d9e786e0adeb328d)
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 #include <trace/events/kvm.h>
41 #undef TRACE_INCLUDE_FILE
42 #define CREATE_TRACE_POINTS
43 #include "trace.h"
44 
45 #include <asm/uaccess.h>
46 #include <asm/msr.h>
47 #include <asm/desc.h>
48 #include <asm/mtrr.h>
49 #include <asm/mce.h>
50 
51 #define MAX_IO_MSRS 256
52 #define CR0_RESERVED_BITS						\
53 	(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
54 			  | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
55 			  | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
56 #define CR4_RESERVED_BITS						\
57 	(~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
58 			  | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE	\
59 			  | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR	\
60 			  | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
61 
62 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
63 
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
66 
67 /* EFER defaults:
68  * - enable syscall per default because its emulated by KVM
69  * - enable LME and LMA per default on 64 bit KVM
70  */
71 #ifdef CONFIG_X86_64
72 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
73 #else
74 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
75 #endif
76 
77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
79 
80 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
81 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
82 				    struct kvm_cpuid_entry2 __user *entries);
83 
84 struct kvm_x86_ops *kvm_x86_ops;
85 EXPORT_SYMBOL_GPL(kvm_x86_ops);
86 
87 int ignore_msrs = 0;
88 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
89 
90 struct kvm_stats_debugfs_item debugfs_entries[] = {
91 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
92 	{ "pf_guest", VCPU_STAT(pf_guest) },
93 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
94 	{ "invlpg", VCPU_STAT(invlpg) },
95 	{ "exits", VCPU_STAT(exits) },
96 	{ "io_exits", VCPU_STAT(io_exits) },
97 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
98 	{ "signal_exits", VCPU_STAT(signal_exits) },
99 	{ "irq_window", VCPU_STAT(irq_window_exits) },
100 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
101 	{ "halt_exits", VCPU_STAT(halt_exits) },
102 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
103 	{ "hypercalls", VCPU_STAT(hypercalls) },
104 	{ "request_irq", VCPU_STAT(request_irq_exits) },
105 	{ "irq_exits", VCPU_STAT(irq_exits) },
106 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
107 	{ "efer_reload", VCPU_STAT(efer_reload) },
108 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
109 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
110 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
111 	{ "irq_injections", VCPU_STAT(irq_injections) },
112 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
113 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
114 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
115 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
116 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
117 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
118 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
119 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
120 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
121 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
122 	{ "largepages", VM_STAT(lpages) },
123 	{ NULL }
124 };
125 
126 unsigned long segment_base(u16 selector)
127 {
128 	struct descriptor_table gdt;
129 	struct desc_struct *d;
130 	unsigned long table_base;
131 	unsigned long v;
132 
133 	if (selector == 0)
134 		return 0;
135 
136 	kvm_get_gdt(&gdt);
137 	table_base = gdt.base;
138 
139 	if (selector & 4) {           /* from ldt */
140 		u16 ldt_selector = kvm_read_ldt();
141 
142 		table_base = segment_base(ldt_selector);
143 	}
144 	d = (struct desc_struct *)(table_base + (selector & ~7));
145 	v = get_desc_base(d);
146 #ifdef CONFIG_X86_64
147 	if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
148 		v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
149 #endif
150 	return v;
151 }
152 EXPORT_SYMBOL_GPL(segment_base);
153 
154 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
155 {
156 	if (irqchip_in_kernel(vcpu->kvm))
157 		return vcpu->arch.apic_base;
158 	else
159 		return vcpu->arch.apic_base;
160 }
161 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
162 
163 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
164 {
165 	/* TODO: reserve bits check */
166 	if (irqchip_in_kernel(vcpu->kvm))
167 		kvm_lapic_set_base(vcpu, data);
168 	else
169 		vcpu->arch.apic_base = data;
170 }
171 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
172 
173 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
174 {
175 	WARN_ON(vcpu->arch.exception.pending);
176 	vcpu->arch.exception.pending = true;
177 	vcpu->arch.exception.has_error_code = false;
178 	vcpu->arch.exception.nr = nr;
179 }
180 EXPORT_SYMBOL_GPL(kvm_queue_exception);
181 
182 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
183 			   u32 error_code)
184 {
185 	++vcpu->stat.pf_guest;
186 
187 	if (vcpu->arch.exception.pending) {
188 		switch(vcpu->arch.exception.nr) {
189 		case DF_VECTOR:
190 			/* triple fault -> shutdown */
191 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
192 			return;
193 		case PF_VECTOR:
194 			vcpu->arch.exception.nr = DF_VECTOR;
195 			vcpu->arch.exception.error_code = 0;
196 			return;
197 		default:
198 			/* replace previous exception with a new one in a hope
199 			   that instruction re-execution will regenerate lost
200 			   exception */
201 			vcpu->arch.exception.pending = false;
202 			break;
203 		}
204 	}
205 	vcpu->arch.cr2 = addr;
206 	kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
207 }
208 
209 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
210 {
211 	vcpu->arch.nmi_pending = 1;
212 }
213 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
214 
215 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
216 {
217 	WARN_ON(vcpu->arch.exception.pending);
218 	vcpu->arch.exception.pending = true;
219 	vcpu->arch.exception.has_error_code = true;
220 	vcpu->arch.exception.nr = nr;
221 	vcpu->arch.exception.error_code = error_code;
222 }
223 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
224 
225 /*
226  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
227  * a #GP and return false.
228  */
229 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
230 {
231 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
232 		return true;
233 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
234 	return false;
235 }
236 EXPORT_SYMBOL_GPL(kvm_require_cpl);
237 
238 /*
239  * Load the pae pdptrs.  Return true is they are all valid.
240  */
241 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
242 {
243 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
244 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
245 	int i;
246 	int ret;
247 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
248 
249 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
250 				  offset * sizeof(u64), sizeof(pdpte));
251 	if (ret < 0) {
252 		ret = 0;
253 		goto out;
254 	}
255 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
256 		if (is_present_gpte(pdpte[i]) &&
257 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
258 			ret = 0;
259 			goto out;
260 		}
261 	}
262 	ret = 1;
263 
264 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
265 	__set_bit(VCPU_EXREG_PDPTR,
266 		  (unsigned long *)&vcpu->arch.regs_avail);
267 	__set_bit(VCPU_EXREG_PDPTR,
268 		  (unsigned long *)&vcpu->arch.regs_dirty);
269 out:
270 
271 	return ret;
272 }
273 EXPORT_SYMBOL_GPL(load_pdptrs);
274 
275 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
276 {
277 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
278 	bool changed = true;
279 	int r;
280 
281 	if (is_long_mode(vcpu) || !is_pae(vcpu))
282 		return false;
283 
284 	if (!test_bit(VCPU_EXREG_PDPTR,
285 		      (unsigned long *)&vcpu->arch.regs_avail))
286 		return true;
287 
288 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
289 	if (r < 0)
290 		goto out;
291 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
292 out:
293 
294 	return changed;
295 }
296 
297 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
298 {
299 	if (cr0 & CR0_RESERVED_BITS) {
300 		printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
301 		       cr0, vcpu->arch.cr0);
302 		kvm_inject_gp(vcpu, 0);
303 		return;
304 	}
305 
306 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
307 		printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
308 		kvm_inject_gp(vcpu, 0);
309 		return;
310 	}
311 
312 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
313 		printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
314 		       "and a clear PE flag\n");
315 		kvm_inject_gp(vcpu, 0);
316 		return;
317 	}
318 
319 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
320 #ifdef CONFIG_X86_64
321 		if ((vcpu->arch.shadow_efer & EFER_LME)) {
322 			int cs_db, cs_l;
323 
324 			if (!is_pae(vcpu)) {
325 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
326 				       "in long mode while PAE is disabled\n");
327 				kvm_inject_gp(vcpu, 0);
328 				return;
329 			}
330 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
331 			if (cs_l) {
332 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
333 				       "in long mode while CS.L == 1\n");
334 				kvm_inject_gp(vcpu, 0);
335 				return;
336 
337 			}
338 		} else
339 #endif
340 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
341 			printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
342 			       "reserved bits\n");
343 			kvm_inject_gp(vcpu, 0);
344 			return;
345 		}
346 
347 	}
348 
349 	kvm_x86_ops->set_cr0(vcpu, cr0);
350 	vcpu->arch.cr0 = cr0;
351 
352 	kvm_mmu_reset_context(vcpu);
353 	return;
354 }
355 EXPORT_SYMBOL_GPL(kvm_set_cr0);
356 
357 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
358 {
359 	kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
360 }
361 EXPORT_SYMBOL_GPL(kvm_lmsw);
362 
363 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
364 {
365 	unsigned long old_cr4 = vcpu->arch.cr4;
366 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
367 
368 	if (cr4 & CR4_RESERVED_BITS) {
369 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
370 		kvm_inject_gp(vcpu, 0);
371 		return;
372 	}
373 
374 	if (is_long_mode(vcpu)) {
375 		if (!(cr4 & X86_CR4_PAE)) {
376 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
377 			       "in long mode\n");
378 			kvm_inject_gp(vcpu, 0);
379 			return;
380 		}
381 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
382 		   && ((cr4 ^ old_cr4) & pdptr_bits)
383 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
384 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
385 		kvm_inject_gp(vcpu, 0);
386 		return;
387 	}
388 
389 	if (cr4 & X86_CR4_VMXE) {
390 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
391 		kvm_inject_gp(vcpu, 0);
392 		return;
393 	}
394 	kvm_x86_ops->set_cr4(vcpu, cr4);
395 	vcpu->arch.cr4 = cr4;
396 	vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
397 	kvm_mmu_reset_context(vcpu);
398 }
399 EXPORT_SYMBOL_GPL(kvm_set_cr4);
400 
401 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
402 {
403 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
404 		kvm_mmu_sync_roots(vcpu);
405 		kvm_mmu_flush_tlb(vcpu);
406 		return;
407 	}
408 
409 	if (is_long_mode(vcpu)) {
410 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
411 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
412 			kvm_inject_gp(vcpu, 0);
413 			return;
414 		}
415 	} else {
416 		if (is_pae(vcpu)) {
417 			if (cr3 & CR3_PAE_RESERVED_BITS) {
418 				printk(KERN_DEBUG
419 				       "set_cr3: #GP, reserved bits\n");
420 				kvm_inject_gp(vcpu, 0);
421 				return;
422 			}
423 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
424 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
425 				       "reserved bits\n");
426 				kvm_inject_gp(vcpu, 0);
427 				return;
428 			}
429 		}
430 		/*
431 		 * We don't check reserved bits in nonpae mode, because
432 		 * this isn't enforced, and VMware depends on this.
433 		 */
434 	}
435 
436 	/*
437 	 * Does the new cr3 value map to physical memory? (Note, we
438 	 * catch an invalid cr3 even in real-mode, because it would
439 	 * cause trouble later on when we turn on paging anyway.)
440 	 *
441 	 * A real CPU would silently accept an invalid cr3 and would
442 	 * attempt to use it - with largely undefined (and often hard
443 	 * to debug) behavior on the guest side.
444 	 */
445 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
446 		kvm_inject_gp(vcpu, 0);
447 	else {
448 		vcpu->arch.cr3 = cr3;
449 		vcpu->arch.mmu.new_cr3(vcpu);
450 	}
451 }
452 EXPORT_SYMBOL_GPL(kvm_set_cr3);
453 
454 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
455 {
456 	if (cr8 & CR8_RESERVED_BITS) {
457 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
458 		kvm_inject_gp(vcpu, 0);
459 		return;
460 	}
461 	if (irqchip_in_kernel(vcpu->kvm))
462 		kvm_lapic_set_tpr(vcpu, cr8);
463 	else
464 		vcpu->arch.cr8 = cr8;
465 }
466 EXPORT_SYMBOL_GPL(kvm_set_cr8);
467 
468 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
469 {
470 	if (irqchip_in_kernel(vcpu->kvm))
471 		return kvm_lapic_get_cr8(vcpu);
472 	else
473 		return vcpu->arch.cr8;
474 }
475 EXPORT_SYMBOL_GPL(kvm_get_cr8);
476 
477 static inline u32 bit(int bitno)
478 {
479 	return 1 << (bitno & 31);
480 }
481 
482 /*
483  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
484  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
485  *
486  * This list is modified at module load time to reflect the
487  * capabilities of the host cpu. This capabilities test skips MSRs that are
488  * kvm-specific. Those are put in the beginning of the list.
489  */
490 
491 #define KVM_SAVE_MSRS_BEGIN	2
492 static u32 msrs_to_save[] = {
493 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
494 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
495 	MSR_K6_STAR,
496 #ifdef CONFIG_X86_64
497 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
498 #endif
499 	MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
500 };
501 
502 static unsigned num_msrs_to_save;
503 
504 static u32 emulated_msrs[] = {
505 	MSR_IA32_MISC_ENABLE,
506 };
507 
508 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
509 {
510 	if (efer & efer_reserved_bits) {
511 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
512 		       efer);
513 		kvm_inject_gp(vcpu, 0);
514 		return;
515 	}
516 
517 	if (is_paging(vcpu)
518 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
519 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
520 		kvm_inject_gp(vcpu, 0);
521 		return;
522 	}
523 
524 	if (efer & EFER_FFXSR) {
525 		struct kvm_cpuid_entry2 *feat;
526 
527 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
528 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
529 			printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
530 			kvm_inject_gp(vcpu, 0);
531 			return;
532 		}
533 	}
534 
535 	if (efer & EFER_SVME) {
536 		struct kvm_cpuid_entry2 *feat;
537 
538 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
539 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
540 			printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
541 			kvm_inject_gp(vcpu, 0);
542 			return;
543 		}
544 	}
545 
546 	kvm_x86_ops->set_efer(vcpu, efer);
547 
548 	efer &= ~EFER_LMA;
549 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
550 
551 	vcpu->arch.shadow_efer = efer;
552 
553 	vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
554 	kvm_mmu_reset_context(vcpu);
555 }
556 
557 void kvm_enable_efer_bits(u64 mask)
558 {
559        efer_reserved_bits &= ~mask;
560 }
561 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
562 
563 
564 /*
565  * Writes msr value into into the appropriate "register".
566  * Returns 0 on success, non-0 otherwise.
567  * Assumes vcpu_load() was already called.
568  */
569 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
570 {
571 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
572 }
573 
574 /*
575  * Adapt set_msr() to msr_io()'s calling convention
576  */
577 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
578 {
579 	return kvm_set_msr(vcpu, index, *data);
580 }
581 
582 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
583 {
584 	static int version;
585 	struct pvclock_wall_clock wc;
586 	struct timespec now, sys, boot;
587 
588 	if (!wall_clock)
589 		return;
590 
591 	version++;
592 
593 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
594 
595 	/*
596 	 * The guest calculates current wall clock time by adding
597 	 * system time (updated by kvm_write_guest_time below) to the
598 	 * wall clock specified here.  guest system time equals host
599 	 * system time for us, thus we must fill in host boot time here.
600 	 */
601 	now = current_kernel_time();
602 	ktime_get_ts(&sys);
603 	boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
604 
605 	wc.sec = boot.tv_sec;
606 	wc.nsec = boot.tv_nsec;
607 	wc.version = version;
608 
609 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
610 
611 	version++;
612 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
613 }
614 
615 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
616 {
617 	uint32_t quotient, remainder;
618 
619 	/* Don't try to replace with do_div(), this one calculates
620 	 * "(dividend << 32) / divisor" */
621 	__asm__ ( "divl %4"
622 		  : "=a" (quotient), "=d" (remainder)
623 		  : "0" (0), "1" (dividend), "r" (divisor) );
624 	return quotient;
625 }
626 
627 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
628 {
629 	uint64_t nsecs = 1000000000LL;
630 	int32_t  shift = 0;
631 	uint64_t tps64;
632 	uint32_t tps32;
633 
634 	tps64 = tsc_khz * 1000LL;
635 	while (tps64 > nsecs*2) {
636 		tps64 >>= 1;
637 		shift--;
638 	}
639 
640 	tps32 = (uint32_t)tps64;
641 	while (tps32 <= (uint32_t)nsecs) {
642 		tps32 <<= 1;
643 		shift++;
644 	}
645 
646 	hv_clock->tsc_shift = shift;
647 	hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
648 
649 	pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
650 		 __func__, tsc_khz, hv_clock->tsc_shift,
651 		 hv_clock->tsc_to_system_mul);
652 }
653 
654 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
655 
656 static void kvm_write_guest_time(struct kvm_vcpu *v)
657 {
658 	struct timespec ts;
659 	unsigned long flags;
660 	struct kvm_vcpu_arch *vcpu = &v->arch;
661 	void *shared_kaddr;
662 	unsigned long this_tsc_khz;
663 
664 	if ((!vcpu->time_page))
665 		return;
666 
667 	this_tsc_khz = get_cpu_var(cpu_tsc_khz);
668 	if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
669 		kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
670 		vcpu->hv_clock_tsc_khz = this_tsc_khz;
671 	}
672 	put_cpu_var(cpu_tsc_khz);
673 
674 	/* Keep irq disabled to prevent changes to the clock */
675 	local_irq_save(flags);
676 	kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
677 	ktime_get_ts(&ts);
678 	local_irq_restore(flags);
679 
680 	/* With all the info we got, fill in the values */
681 
682 	vcpu->hv_clock.system_time = ts.tv_nsec +
683 				     (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
684 
685 	/*
686 	 * The interface expects us to write an even number signaling that the
687 	 * update is finished. Since the guest won't see the intermediate
688 	 * state, we just increase by 2 at the end.
689 	 */
690 	vcpu->hv_clock.version += 2;
691 
692 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
693 
694 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
695 	       sizeof(vcpu->hv_clock));
696 
697 	kunmap_atomic(shared_kaddr, KM_USER0);
698 
699 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
700 }
701 
702 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
703 {
704 	struct kvm_vcpu_arch *vcpu = &v->arch;
705 
706 	if (!vcpu->time_page)
707 		return 0;
708 	set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
709 	return 1;
710 }
711 
712 static bool msr_mtrr_valid(unsigned msr)
713 {
714 	switch (msr) {
715 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
716 	case MSR_MTRRfix64K_00000:
717 	case MSR_MTRRfix16K_80000:
718 	case MSR_MTRRfix16K_A0000:
719 	case MSR_MTRRfix4K_C0000:
720 	case MSR_MTRRfix4K_C8000:
721 	case MSR_MTRRfix4K_D0000:
722 	case MSR_MTRRfix4K_D8000:
723 	case MSR_MTRRfix4K_E0000:
724 	case MSR_MTRRfix4K_E8000:
725 	case MSR_MTRRfix4K_F0000:
726 	case MSR_MTRRfix4K_F8000:
727 	case MSR_MTRRdefType:
728 	case MSR_IA32_CR_PAT:
729 		return true;
730 	case 0x2f8:
731 		return true;
732 	}
733 	return false;
734 }
735 
736 static bool valid_pat_type(unsigned t)
737 {
738 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
739 }
740 
741 static bool valid_mtrr_type(unsigned t)
742 {
743 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
744 }
745 
746 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
747 {
748 	int i;
749 
750 	if (!msr_mtrr_valid(msr))
751 		return false;
752 
753 	if (msr == MSR_IA32_CR_PAT) {
754 		for (i = 0; i < 8; i++)
755 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
756 				return false;
757 		return true;
758 	} else if (msr == MSR_MTRRdefType) {
759 		if (data & ~0xcff)
760 			return false;
761 		return valid_mtrr_type(data & 0xff);
762 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
763 		for (i = 0; i < 8 ; i++)
764 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
765 				return false;
766 		return true;
767 	}
768 
769 	/* variable MTRRs */
770 	return valid_mtrr_type(data & 0xff);
771 }
772 
773 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
774 {
775 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
776 
777 	if (!mtrr_valid(vcpu, msr, data))
778 		return 1;
779 
780 	if (msr == MSR_MTRRdefType) {
781 		vcpu->arch.mtrr_state.def_type = data;
782 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
783 	} else if (msr == MSR_MTRRfix64K_00000)
784 		p[0] = data;
785 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
786 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
787 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
788 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
789 	else if (msr == MSR_IA32_CR_PAT)
790 		vcpu->arch.pat = data;
791 	else {	/* Variable MTRRs */
792 		int idx, is_mtrr_mask;
793 		u64 *pt;
794 
795 		idx = (msr - 0x200) / 2;
796 		is_mtrr_mask = msr - 0x200 - 2 * idx;
797 		if (!is_mtrr_mask)
798 			pt =
799 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
800 		else
801 			pt =
802 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
803 		*pt = data;
804 	}
805 
806 	kvm_mmu_reset_context(vcpu);
807 	return 0;
808 }
809 
810 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
811 {
812 	u64 mcg_cap = vcpu->arch.mcg_cap;
813 	unsigned bank_num = mcg_cap & 0xff;
814 
815 	switch (msr) {
816 	case MSR_IA32_MCG_STATUS:
817 		vcpu->arch.mcg_status = data;
818 		break;
819 	case MSR_IA32_MCG_CTL:
820 		if (!(mcg_cap & MCG_CTL_P))
821 			return 1;
822 		if (data != 0 && data != ~(u64)0)
823 			return -1;
824 		vcpu->arch.mcg_ctl = data;
825 		break;
826 	default:
827 		if (msr >= MSR_IA32_MC0_CTL &&
828 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
829 			u32 offset = msr - MSR_IA32_MC0_CTL;
830 			/* only 0 or all 1s can be written to IA32_MCi_CTL */
831 			if ((offset & 0x3) == 0 &&
832 			    data != 0 && data != ~(u64)0)
833 				return -1;
834 			vcpu->arch.mce_banks[offset] = data;
835 			break;
836 		}
837 		return 1;
838 	}
839 	return 0;
840 }
841 
842 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
843 {
844 	struct kvm *kvm = vcpu->kvm;
845 	int lm = is_long_mode(vcpu);
846 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
847 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
848 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
849 		: kvm->arch.xen_hvm_config.blob_size_32;
850 	u32 page_num = data & ~PAGE_MASK;
851 	u64 page_addr = data & PAGE_MASK;
852 	u8 *page;
853 	int r;
854 
855 	r = -E2BIG;
856 	if (page_num >= blob_size)
857 		goto out;
858 	r = -ENOMEM;
859 	page = kzalloc(PAGE_SIZE, GFP_KERNEL);
860 	if (!page)
861 		goto out;
862 	r = -EFAULT;
863 	if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
864 		goto out_free;
865 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
866 		goto out_free;
867 	r = 0;
868 out_free:
869 	kfree(page);
870 out:
871 	return r;
872 }
873 
874 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
875 {
876 	switch (msr) {
877 	case MSR_EFER:
878 		set_efer(vcpu, data);
879 		break;
880 	case MSR_K7_HWCR:
881 		data &= ~(u64)0x40;	/* ignore flush filter disable */
882 		if (data != 0) {
883 			pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
884 				data);
885 			return 1;
886 		}
887 		break;
888 	case MSR_FAM10H_MMIO_CONF_BASE:
889 		if (data != 0) {
890 			pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
891 				"0x%llx\n", data);
892 			return 1;
893 		}
894 		break;
895 	case MSR_AMD64_NB_CFG:
896 		break;
897 	case MSR_IA32_DEBUGCTLMSR:
898 		if (!data) {
899 			/* We support the non-activated case already */
900 			break;
901 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
902 			/* Values other than LBR and BTF are vendor-specific,
903 			   thus reserved and should throw a #GP */
904 			return 1;
905 		}
906 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
907 			__func__, data);
908 		break;
909 	case MSR_IA32_UCODE_REV:
910 	case MSR_IA32_UCODE_WRITE:
911 	case MSR_VM_HSAVE_PA:
912 	case MSR_AMD64_PATCH_LOADER:
913 		break;
914 	case 0x200 ... 0x2ff:
915 		return set_msr_mtrr(vcpu, msr, data);
916 	case MSR_IA32_APICBASE:
917 		kvm_set_apic_base(vcpu, data);
918 		break;
919 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
920 		return kvm_x2apic_msr_write(vcpu, msr, data);
921 	case MSR_IA32_MISC_ENABLE:
922 		vcpu->arch.ia32_misc_enable_msr = data;
923 		break;
924 	case MSR_KVM_WALL_CLOCK:
925 		vcpu->kvm->arch.wall_clock = data;
926 		kvm_write_wall_clock(vcpu->kvm, data);
927 		break;
928 	case MSR_KVM_SYSTEM_TIME: {
929 		if (vcpu->arch.time_page) {
930 			kvm_release_page_dirty(vcpu->arch.time_page);
931 			vcpu->arch.time_page = NULL;
932 		}
933 
934 		vcpu->arch.time = data;
935 
936 		/* we verify if the enable bit is set... */
937 		if (!(data & 1))
938 			break;
939 
940 		/* ...but clean it before doing the actual write */
941 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
942 
943 		vcpu->arch.time_page =
944 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
945 
946 		if (is_error_page(vcpu->arch.time_page)) {
947 			kvm_release_page_clean(vcpu->arch.time_page);
948 			vcpu->arch.time_page = NULL;
949 		}
950 
951 		kvm_request_guest_time_update(vcpu);
952 		break;
953 	}
954 	case MSR_IA32_MCG_CTL:
955 	case MSR_IA32_MCG_STATUS:
956 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
957 		return set_msr_mce(vcpu, msr, data);
958 
959 	/* Performance counters are not protected by a CPUID bit,
960 	 * so we should check all of them in the generic path for the sake of
961 	 * cross vendor migration.
962 	 * Writing a zero into the event select MSRs disables them,
963 	 * which we perfectly emulate ;-). Any other value should be at least
964 	 * reported, some guests depend on them.
965 	 */
966 	case MSR_P6_EVNTSEL0:
967 	case MSR_P6_EVNTSEL1:
968 	case MSR_K7_EVNTSEL0:
969 	case MSR_K7_EVNTSEL1:
970 	case MSR_K7_EVNTSEL2:
971 	case MSR_K7_EVNTSEL3:
972 		if (data != 0)
973 			pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
974 				"0x%x data 0x%llx\n", msr, data);
975 		break;
976 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
977 	 * so we ignore writes to make it happy.
978 	 */
979 	case MSR_P6_PERFCTR0:
980 	case MSR_P6_PERFCTR1:
981 	case MSR_K7_PERFCTR0:
982 	case MSR_K7_PERFCTR1:
983 	case MSR_K7_PERFCTR2:
984 	case MSR_K7_PERFCTR3:
985 		pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
986 			"0x%x data 0x%llx\n", msr, data);
987 		break;
988 	default:
989 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
990 			return xen_hvm_config(vcpu, data);
991 		if (!ignore_msrs) {
992 			pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
993 				msr, data);
994 			return 1;
995 		} else {
996 			pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
997 				msr, data);
998 			break;
999 		}
1000 	}
1001 	return 0;
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1004 
1005 
1006 /*
1007  * Reads an msr value (of 'msr_index') into 'pdata'.
1008  * Returns 0 on success, non-0 otherwise.
1009  * Assumes vcpu_load() was already called.
1010  */
1011 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1012 {
1013 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1014 }
1015 
1016 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1017 {
1018 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1019 
1020 	if (!msr_mtrr_valid(msr))
1021 		return 1;
1022 
1023 	if (msr == MSR_MTRRdefType)
1024 		*pdata = vcpu->arch.mtrr_state.def_type +
1025 			 (vcpu->arch.mtrr_state.enabled << 10);
1026 	else if (msr == MSR_MTRRfix64K_00000)
1027 		*pdata = p[0];
1028 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1029 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1030 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1031 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1032 	else if (msr == MSR_IA32_CR_PAT)
1033 		*pdata = vcpu->arch.pat;
1034 	else {	/* Variable MTRRs */
1035 		int idx, is_mtrr_mask;
1036 		u64 *pt;
1037 
1038 		idx = (msr - 0x200) / 2;
1039 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1040 		if (!is_mtrr_mask)
1041 			pt =
1042 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1043 		else
1044 			pt =
1045 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1046 		*pdata = *pt;
1047 	}
1048 
1049 	return 0;
1050 }
1051 
1052 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1053 {
1054 	u64 data;
1055 	u64 mcg_cap = vcpu->arch.mcg_cap;
1056 	unsigned bank_num = mcg_cap & 0xff;
1057 
1058 	switch (msr) {
1059 	case MSR_IA32_P5_MC_ADDR:
1060 	case MSR_IA32_P5_MC_TYPE:
1061 		data = 0;
1062 		break;
1063 	case MSR_IA32_MCG_CAP:
1064 		data = vcpu->arch.mcg_cap;
1065 		break;
1066 	case MSR_IA32_MCG_CTL:
1067 		if (!(mcg_cap & MCG_CTL_P))
1068 			return 1;
1069 		data = vcpu->arch.mcg_ctl;
1070 		break;
1071 	case MSR_IA32_MCG_STATUS:
1072 		data = vcpu->arch.mcg_status;
1073 		break;
1074 	default:
1075 		if (msr >= MSR_IA32_MC0_CTL &&
1076 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1077 			u32 offset = msr - MSR_IA32_MC0_CTL;
1078 			data = vcpu->arch.mce_banks[offset];
1079 			break;
1080 		}
1081 		return 1;
1082 	}
1083 	*pdata = data;
1084 	return 0;
1085 }
1086 
1087 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1088 {
1089 	u64 data;
1090 
1091 	switch (msr) {
1092 	case MSR_IA32_PLATFORM_ID:
1093 	case MSR_IA32_UCODE_REV:
1094 	case MSR_IA32_EBL_CR_POWERON:
1095 	case MSR_IA32_DEBUGCTLMSR:
1096 	case MSR_IA32_LASTBRANCHFROMIP:
1097 	case MSR_IA32_LASTBRANCHTOIP:
1098 	case MSR_IA32_LASTINTFROMIP:
1099 	case MSR_IA32_LASTINTTOIP:
1100 	case MSR_K8_SYSCFG:
1101 	case MSR_K7_HWCR:
1102 	case MSR_VM_HSAVE_PA:
1103 	case MSR_P6_PERFCTR0:
1104 	case MSR_P6_PERFCTR1:
1105 	case MSR_P6_EVNTSEL0:
1106 	case MSR_P6_EVNTSEL1:
1107 	case MSR_K7_EVNTSEL0:
1108 	case MSR_K7_PERFCTR0:
1109 	case MSR_K8_INT_PENDING_MSG:
1110 	case MSR_AMD64_NB_CFG:
1111 	case MSR_FAM10H_MMIO_CONF_BASE:
1112 		data = 0;
1113 		break;
1114 	case MSR_MTRRcap:
1115 		data = 0x500 | KVM_NR_VAR_MTRR;
1116 		break;
1117 	case 0x200 ... 0x2ff:
1118 		return get_msr_mtrr(vcpu, msr, pdata);
1119 	case 0xcd: /* fsb frequency */
1120 		data = 3;
1121 		break;
1122 	case MSR_IA32_APICBASE:
1123 		data = kvm_get_apic_base(vcpu);
1124 		break;
1125 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1126 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
1127 		break;
1128 	case MSR_IA32_MISC_ENABLE:
1129 		data = vcpu->arch.ia32_misc_enable_msr;
1130 		break;
1131 	case MSR_IA32_PERF_STATUS:
1132 		/* TSC increment by tick */
1133 		data = 1000ULL;
1134 		/* CPU multiplier */
1135 		data |= (((uint64_t)4ULL) << 40);
1136 		break;
1137 	case MSR_EFER:
1138 		data = vcpu->arch.shadow_efer;
1139 		break;
1140 	case MSR_KVM_WALL_CLOCK:
1141 		data = vcpu->kvm->arch.wall_clock;
1142 		break;
1143 	case MSR_KVM_SYSTEM_TIME:
1144 		data = vcpu->arch.time;
1145 		break;
1146 	case MSR_IA32_P5_MC_ADDR:
1147 	case MSR_IA32_P5_MC_TYPE:
1148 	case MSR_IA32_MCG_CAP:
1149 	case MSR_IA32_MCG_CTL:
1150 	case MSR_IA32_MCG_STATUS:
1151 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1152 		return get_msr_mce(vcpu, msr, pdata);
1153 	default:
1154 		if (!ignore_msrs) {
1155 			pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1156 			return 1;
1157 		} else {
1158 			pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1159 			data = 0;
1160 		}
1161 		break;
1162 	}
1163 	*pdata = data;
1164 	return 0;
1165 }
1166 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1167 
1168 /*
1169  * Read or write a bunch of msrs. All parameters are kernel addresses.
1170  *
1171  * @return number of msrs set successfully.
1172  */
1173 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1174 		    struct kvm_msr_entry *entries,
1175 		    int (*do_msr)(struct kvm_vcpu *vcpu,
1176 				  unsigned index, u64 *data))
1177 {
1178 	int i;
1179 
1180 	vcpu_load(vcpu);
1181 
1182 	down_read(&vcpu->kvm->slots_lock);
1183 	for (i = 0; i < msrs->nmsrs; ++i)
1184 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
1185 			break;
1186 	up_read(&vcpu->kvm->slots_lock);
1187 
1188 	vcpu_put(vcpu);
1189 
1190 	return i;
1191 }
1192 
1193 /*
1194  * Read or write a bunch of msrs. Parameters are user addresses.
1195  *
1196  * @return number of msrs set successfully.
1197  */
1198 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1199 		  int (*do_msr)(struct kvm_vcpu *vcpu,
1200 				unsigned index, u64 *data),
1201 		  int writeback)
1202 {
1203 	struct kvm_msrs msrs;
1204 	struct kvm_msr_entry *entries;
1205 	int r, n;
1206 	unsigned size;
1207 
1208 	r = -EFAULT;
1209 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1210 		goto out;
1211 
1212 	r = -E2BIG;
1213 	if (msrs.nmsrs >= MAX_IO_MSRS)
1214 		goto out;
1215 
1216 	r = -ENOMEM;
1217 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1218 	entries = vmalloc(size);
1219 	if (!entries)
1220 		goto out;
1221 
1222 	r = -EFAULT;
1223 	if (copy_from_user(entries, user_msrs->entries, size))
1224 		goto out_free;
1225 
1226 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1227 	if (r < 0)
1228 		goto out_free;
1229 
1230 	r = -EFAULT;
1231 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
1232 		goto out_free;
1233 
1234 	r = n;
1235 
1236 out_free:
1237 	vfree(entries);
1238 out:
1239 	return r;
1240 }
1241 
1242 int kvm_dev_ioctl_check_extension(long ext)
1243 {
1244 	int r;
1245 
1246 	switch (ext) {
1247 	case KVM_CAP_IRQCHIP:
1248 	case KVM_CAP_HLT:
1249 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1250 	case KVM_CAP_SET_TSS_ADDR:
1251 	case KVM_CAP_EXT_CPUID:
1252 	case KVM_CAP_CLOCKSOURCE:
1253 	case KVM_CAP_PIT:
1254 	case KVM_CAP_NOP_IO_DELAY:
1255 	case KVM_CAP_MP_STATE:
1256 	case KVM_CAP_SYNC_MMU:
1257 	case KVM_CAP_REINJECT_CONTROL:
1258 	case KVM_CAP_IRQ_INJECT_STATUS:
1259 	case KVM_CAP_ASSIGN_DEV_IRQ:
1260 	case KVM_CAP_IRQFD:
1261 	case KVM_CAP_IOEVENTFD:
1262 	case KVM_CAP_PIT2:
1263 	case KVM_CAP_PIT_STATE2:
1264 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1265 	case KVM_CAP_XEN_HVM:
1266 	case KVM_CAP_ADJUST_CLOCK:
1267 		r = 1;
1268 		break;
1269 	case KVM_CAP_COALESCED_MMIO:
1270 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1271 		break;
1272 	case KVM_CAP_VAPIC:
1273 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1274 		break;
1275 	case KVM_CAP_NR_VCPUS:
1276 		r = KVM_MAX_VCPUS;
1277 		break;
1278 	case KVM_CAP_NR_MEMSLOTS:
1279 		r = KVM_MEMORY_SLOTS;
1280 		break;
1281 	case KVM_CAP_PV_MMU:	/* obsolete */
1282 		r = 0;
1283 		break;
1284 	case KVM_CAP_IOMMU:
1285 		r = iommu_found();
1286 		break;
1287 	case KVM_CAP_MCE:
1288 		r = KVM_MAX_MCE_BANKS;
1289 		break;
1290 	default:
1291 		r = 0;
1292 		break;
1293 	}
1294 	return r;
1295 
1296 }
1297 
1298 long kvm_arch_dev_ioctl(struct file *filp,
1299 			unsigned int ioctl, unsigned long arg)
1300 {
1301 	void __user *argp = (void __user *)arg;
1302 	long r;
1303 
1304 	switch (ioctl) {
1305 	case KVM_GET_MSR_INDEX_LIST: {
1306 		struct kvm_msr_list __user *user_msr_list = argp;
1307 		struct kvm_msr_list msr_list;
1308 		unsigned n;
1309 
1310 		r = -EFAULT;
1311 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1312 			goto out;
1313 		n = msr_list.nmsrs;
1314 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1315 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1316 			goto out;
1317 		r = -E2BIG;
1318 		if (n < msr_list.nmsrs)
1319 			goto out;
1320 		r = -EFAULT;
1321 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1322 				 num_msrs_to_save * sizeof(u32)))
1323 			goto out;
1324 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1325 				 &emulated_msrs,
1326 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1327 			goto out;
1328 		r = 0;
1329 		break;
1330 	}
1331 	case KVM_GET_SUPPORTED_CPUID: {
1332 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1333 		struct kvm_cpuid2 cpuid;
1334 
1335 		r = -EFAULT;
1336 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1337 			goto out;
1338 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1339 						      cpuid_arg->entries);
1340 		if (r)
1341 			goto out;
1342 
1343 		r = -EFAULT;
1344 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1345 			goto out;
1346 		r = 0;
1347 		break;
1348 	}
1349 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1350 		u64 mce_cap;
1351 
1352 		mce_cap = KVM_MCE_CAP_SUPPORTED;
1353 		r = -EFAULT;
1354 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1355 			goto out;
1356 		r = 0;
1357 		break;
1358 	}
1359 	default:
1360 		r = -EINVAL;
1361 	}
1362 out:
1363 	return r;
1364 }
1365 
1366 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1367 {
1368 	kvm_x86_ops->vcpu_load(vcpu, cpu);
1369 	if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1370 		unsigned long khz = cpufreq_quick_get(cpu);
1371 		if (!khz)
1372 			khz = tsc_khz;
1373 		per_cpu(cpu_tsc_khz, cpu) = khz;
1374 	}
1375 	kvm_request_guest_time_update(vcpu);
1376 }
1377 
1378 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1379 {
1380 	kvm_x86_ops->vcpu_put(vcpu);
1381 	kvm_put_guest_fpu(vcpu);
1382 }
1383 
1384 static int is_efer_nx(void)
1385 {
1386 	unsigned long long efer = 0;
1387 
1388 	rdmsrl_safe(MSR_EFER, &efer);
1389 	return efer & EFER_NX;
1390 }
1391 
1392 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1393 {
1394 	int i;
1395 	struct kvm_cpuid_entry2 *e, *entry;
1396 
1397 	entry = NULL;
1398 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1399 		e = &vcpu->arch.cpuid_entries[i];
1400 		if (e->function == 0x80000001) {
1401 			entry = e;
1402 			break;
1403 		}
1404 	}
1405 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1406 		entry->edx &= ~(1 << 20);
1407 		printk(KERN_INFO "kvm: guest NX capability removed\n");
1408 	}
1409 }
1410 
1411 /* when an old userspace process fills a new kernel module */
1412 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1413 				    struct kvm_cpuid *cpuid,
1414 				    struct kvm_cpuid_entry __user *entries)
1415 {
1416 	int r, i;
1417 	struct kvm_cpuid_entry *cpuid_entries;
1418 
1419 	r = -E2BIG;
1420 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1421 		goto out;
1422 	r = -ENOMEM;
1423 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1424 	if (!cpuid_entries)
1425 		goto out;
1426 	r = -EFAULT;
1427 	if (copy_from_user(cpuid_entries, entries,
1428 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1429 		goto out_free;
1430 	for (i = 0; i < cpuid->nent; i++) {
1431 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1432 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1433 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1434 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1435 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1436 		vcpu->arch.cpuid_entries[i].index = 0;
1437 		vcpu->arch.cpuid_entries[i].flags = 0;
1438 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
1439 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
1440 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
1441 	}
1442 	vcpu->arch.cpuid_nent = cpuid->nent;
1443 	cpuid_fix_nx_cap(vcpu);
1444 	r = 0;
1445 	kvm_apic_set_version(vcpu);
1446 
1447 out_free:
1448 	vfree(cpuid_entries);
1449 out:
1450 	return r;
1451 }
1452 
1453 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1454 				     struct kvm_cpuid2 *cpuid,
1455 				     struct kvm_cpuid_entry2 __user *entries)
1456 {
1457 	int r;
1458 
1459 	r = -E2BIG;
1460 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1461 		goto out;
1462 	r = -EFAULT;
1463 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1464 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1465 		goto out;
1466 	vcpu->arch.cpuid_nent = cpuid->nent;
1467 	kvm_apic_set_version(vcpu);
1468 	return 0;
1469 
1470 out:
1471 	return r;
1472 }
1473 
1474 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1475 				     struct kvm_cpuid2 *cpuid,
1476 				     struct kvm_cpuid_entry2 __user *entries)
1477 {
1478 	int r;
1479 
1480 	r = -E2BIG;
1481 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1482 		goto out;
1483 	r = -EFAULT;
1484 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1485 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1486 		goto out;
1487 	return 0;
1488 
1489 out:
1490 	cpuid->nent = vcpu->arch.cpuid_nent;
1491 	return r;
1492 }
1493 
1494 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1495 			   u32 index)
1496 {
1497 	entry->function = function;
1498 	entry->index = index;
1499 	cpuid_count(entry->function, entry->index,
1500 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1501 	entry->flags = 0;
1502 }
1503 
1504 #define F(x) bit(X86_FEATURE_##x)
1505 
1506 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1507 			 u32 index, int *nent, int maxnent)
1508 {
1509 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1510 	unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1511 #ifdef CONFIG_X86_64
1512 	unsigned f_lm = F(LM);
1513 #else
1514 	unsigned f_lm = 0;
1515 #endif
1516 
1517 	/* cpuid 1.edx */
1518 	const u32 kvm_supported_word0_x86_features =
1519 		F(FPU) | F(VME) | F(DE) | F(PSE) |
1520 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1521 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1522 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1523 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1524 		0 /* Reserved, DS, ACPI */ | F(MMX) |
1525 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1526 		0 /* HTT, TM, Reserved, PBE */;
1527 	/* cpuid 0x80000001.edx */
1528 	const u32 kvm_supported_word1_x86_features =
1529 		F(FPU) | F(VME) | F(DE) | F(PSE) |
1530 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1531 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1532 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1533 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
1534 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1535 		F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1536 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1537 	/* cpuid 1.ecx */
1538 	const u32 kvm_supported_word4_x86_features =
1539 		F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1540 		0 /* DS-CPL, VMX, SMX, EST */ |
1541 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1542 		0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1543 		0 /* Reserved, DCA */ | F(XMM4_1) |
1544 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1545 		0 /* Reserved, XSAVE, OSXSAVE */;
1546 	/* cpuid 0x80000001.ecx */
1547 	const u32 kvm_supported_word6_x86_features =
1548 		F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1549 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1550 		F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1551 		0 /* SKINIT */ | 0 /* WDT */;
1552 
1553 	/* all calls to cpuid_count() should be made on the same cpu */
1554 	get_cpu();
1555 	do_cpuid_1_ent(entry, function, index);
1556 	++*nent;
1557 
1558 	switch (function) {
1559 	case 0:
1560 		entry->eax = min(entry->eax, (u32)0xb);
1561 		break;
1562 	case 1:
1563 		entry->edx &= kvm_supported_word0_x86_features;
1564 		entry->ecx &= kvm_supported_word4_x86_features;
1565 		/* we support x2apic emulation even if host does not support
1566 		 * it since we emulate x2apic in software */
1567 		entry->ecx |= F(X2APIC);
1568 		break;
1569 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1570 	 * may return different values. This forces us to get_cpu() before
1571 	 * issuing the first command, and also to emulate this annoying behavior
1572 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1573 	case 2: {
1574 		int t, times = entry->eax & 0xff;
1575 
1576 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1577 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1578 		for (t = 1; t < times && *nent < maxnent; ++t) {
1579 			do_cpuid_1_ent(&entry[t], function, 0);
1580 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1581 			++*nent;
1582 		}
1583 		break;
1584 	}
1585 	/* function 4 and 0xb have additional index. */
1586 	case 4: {
1587 		int i, cache_type;
1588 
1589 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1590 		/* read more entries until cache_type is zero */
1591 		for (i = 1; *nent < maxnent; ++i) {
1592 			cache_type = entry[i - 1].eax & 0x1f;
1593 			if (!cache_type)
1594 				break;
1595 			do_cpuid_1_ent(&entry[i], function, i);
1596 			entry[i].flags |=
1597 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1598 			++*nent;
1599 		}
1600 		break;
1601 	}
1602 	case 0xb: {
1603 		int i, level_type;
1604 
1605 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1606 		/* read more entries until level_type is zero */
1607 		for (i = 1; *nent < maxnent; ++i) {
1608 			level_type = entry[i - 1].ecx & 0xff00;
1609 			if (!level_type)
1610 				break;
1611 			do_cpuid_1_ent(&entry[i], function, i);
1612 			entry[i].flags |=
1613 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1614 			++*nent;
1615 		}
1616 		break;
1617 	}
1618 	case 0x80000000:
1619 		entry->eax = min(entry->eax, 0x8000001a);
1620 		break;
1621 	case 0x80000001:
1622 		entry->edx &= kvm_supported_word1_x86_features;
1623 		entry->ecx &= kvm_supported_word6_x86_features;
1624 		break;
1625 	}
1626 	put_cpu();
1627 }
1628 
1629 #undef F
1630 
1631 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1632 				     struct kvm_cpuid_entry2 __user *entries)
1633 {
1634 	struct kvm_cpuid_entry2 *cpuid_entries;
1635 	int limit, nent = 0, r = -E2BIG;
1636 	u32 func;
1637 
1638 	if (cpuid->nent < 1)
1639 		goto out;
1640 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1641 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1642 	r = -ENOMEM;
1643 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1644 	if (!cpuid_entries)
1645 		goto out;
1646 
1647 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1648 	limit = cpuid_entries[0].eax;
1649 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1650 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1651 			     &nent, cpuid->nent);
1652 	r = -E2BIG;
1653 	if (nent >= cpuid->nent)
1654 		goto out_free;
1655 
1656 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1657 	limit = cpuid_entries[nent - 1].eax;
1658 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1659 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1660 			     &nent, cpuid->nent);
1661 	r = -E2BIG;
1662 	if (nent >= cpuid->nent)
1663 		goto out_free;
1664 
1665 	r = -EFAULT;
1666 	if (copy_to_user(entries, cpuid_entries,
1667 			 nent * sizeof(struct kvm_cpuid_entry2)))
1668 		goto out_free;
1669 	cpuid->nent = nent;
1670 	r = 0;
1671 
1672 out_free:
1673 	vfree(cpuid_entries);
1674 out:
1675 	return r;
1676 }
1677 
1678 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1679 				    struct kvm_lapic_state *s)
1680 {
1681 	vcpu_load(vcpu);
1682 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1683 	vcpu_put(vcpu);
1684 
1685 	return 0;
1686 }
1687 
1688 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1689 				    struct kvm_lapic_state *s)
1690 {
1691 	vcpu_load(vcpu);
1692 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1693 	kvm_apic_post_state_restore(vcpu);
1694 	update_cr8_intercept(vcpu);
1695 	vcpu_put(vcpu);
1696 
1697 	return 0;
1698 }
1699 
1700 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1701 				    struct kvm_interrupt *irq)
1702 {
1703 	if (irq->irq < 0 || irq->irq >= 256)
1704 		return -EINVAL;
1705 	if (irqchip_in_kernel(vcpu->kvm))
1706 		return -ENXIO;
1707 	vcpu_load(vcpu);
1708 
1709 	kvm_queue_interrupt(vcpu, irq->irq, false);
1710 
1711 	vcpu_put(vcpu);
1712 
1713 	return 0;
1714 }
1715 
1716 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1717 {
1718 	vcpu_load(vcpu);
1719 	kvm_inject_nmi(vcpu);
1720 	vcpu_put(vcpu);
1721 
1722 	return 0;
1723 }
1724 
1725 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1726 					   struct kvm_tpr_access_ctl *tac)
1727 {
1728 	if (tac->flags)
1729 		return -EINVAL;
1730 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1731 	return 0;
1732 }
1733 
1734 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1735 					u64 mcg_cap)
1736 {
1737 	int r;
1738 	unsigned bank_num = mcg_cap & 0xff, bank;
1739 
1740 	r = -EINVAL;
1741 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1742 		goto out;
1743 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1744 		goto out;
1745 	r = 0;
1746 	vcpu->arch.mcg_cap = mcg_cap;
1747 	/* Init IA32_MCG_CTL to all 1s */
1748 	if (mcg_cap & MCG_CTL_P)
1749 		vcpu->arch.mcg_ctl = ~(u64)0;
1750 	/* Init IA32_MCi_CTL to all 1s */
1751 	for (bank = 0; bank < bank_num; bank++)
1752 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1753 out:
1754 	return r;
1755 }
1756 
1757 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1758 				      struct kvm_x86_mce *mce)
1759 {
1760 	u64 mcg_cap = vcpu->arch.mcg_cap;
1761 	unsigned bank_num = mcg_cap & 0xff;
1762 	u64 *banks = vcpu->arch.mce_banks;
1763 
1764 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1765 		return -EINVAL;
1766 	/*
1767 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1768 	 * reporting is disabled
1769 	 */
1770 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1771 	    vcpu->arch.mcg_ctl != ~(u64)0)
1772 		return 0;
1773 	banks += 4 * mce->bank;
1774 	/*
1775 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1776 	 * reporting is disabled for the bank
1777 	 */
1778 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1779 		return 0;
1780 	if (mce->status & MCI_STATUS_UC) {
1781 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1782 		    !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1783 			printk(KERN_DEBUG "kvm: set_mce: "
1784 			       "injects mce exception while "
1785 			       "previous one is in progress!\n");
1786 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1787 			return 0;
1788 		}
1789 		if (banks[1] & MCI_STATUS_VAL)
1790 			mce->status |= MCI_STATUS_OVER;
1791 		banks[2] = mce->addr;
1792 		banks[3] = mce->misc;
1793 		vcpu->arch.mcg_status = mce->mcg_status;
1794 		banks[1] = mce->status;
1795 		kvm_queue_exception(vcpu, MC_VECTOR);
1796 	} else if (!(banks[1] & MCI_STATUS_VAL)
1797 		   || !(banks[1] & MCI_STATUS_UC)) {
1798 		if (banks[1] & MCI_STATUS_VAL)
1799 			mce->status |= MCI_STATUS_OVER;
1800 		banks[2] = mce->addr;
1801 		banks[3] = mce->misc;
1802 		banks[1] = mce->status;
1803 	} else
1804 		banks[1] |= MCI_STATUS_OVER;
1805 	return 0;
1806 }
1807 
1808 long kvm_arch_vcpu_ioctl(struct file *filp,
1809 			 unsigned int ioctl, unsigned long arg)
1810 {
1811 	struct kvm_vcpu *vcpu = filp->private_data;
1812 	void __user *argp = (void __user *)arg;
1813 	int r;
1814 	struct kvm_lapic_state *lapic = NULL;
1815 
1816 	switch (ioctl) {
1817 	case KVM_GET_LAPIC: {
1818 		lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1819 
1820 		r = -ENOMEM;
1821 		if (!lapic)
1822 			goto out;
1823 		r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1824 		if (r)
1825 			goto out;
1826 		r = -EFAULT;
1827 		if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1828 			goto out;
1829 		r = 0;
1830 		break;
1831 	}
1832 	case KVM_SET_LAPIC: {
1833 		lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1834 		r = -ENOMEM;
1835 		if (!lapic)
1836 			goto out;
1837 		r = -EFAULT;
1838 		if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1839 			goto out;
1840 		r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1841 		if (r)
1842 			goto out;
1843 		r = 0;
1844 		break;
1845 	}
1846 	case KVM_INTERRUPT: {
1847 		struct kvm_interrupt irq;
1848 
1849 		r = -EFAULT;
1850 		if (copy_from_user(&irq, argp, sizeof irq))
1851 			goto out;
1852 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1853 		if (r)
1854 			goto out;
1855 		r = 0;
1856 		break;
1857 	}
1858 	case KVM_NMI: {
1859 		r = kvm_vcpu_ioctl_nmi(vcpu);
1860 		if (r)
1861 			goto out;
1862 		r = 0;
1863 		break;
1864 	}
1865 	case KVM_SET_CPUID: {
1866 		struct kvm_cpuid __user *cpuid_arg = argp;
1867 		struct kvm_cpuid cpuid;
1868 
1869 		r = -EFAULT;
1870 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1871 			goto out;
1872 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1873 		if (r)
1874 			goto out;
1875 		break;
1876 	}
1877 	case KVM_SET_CPUID2: {
1878 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1879 		struct kvm_cpuid2 cpuid;
1880 
1881 		r = -EFAULT;
1882 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1883 			goto out;
1884 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1885 					      cpuid_arg->entries);
1886 		if (r)
1887 			goto out;
1888 		break;
1889 	}
1890 	case KVM_GET_CPUID2: {
1891 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1892 		struct kvm_cpuid2 cpuid;
1893 
1894 		r = -EFAULT;
1895 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1896 			goto out;
1897 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1898 					      cpuid_arg->entries);
1899 		if (r)
1900 			goto out;
1901 		r = -EFAULT;
1902 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1903 			goto out;
1904 		r = 0;
1905 		break;
1906 	}
1907 	case KVM_GET_MSRS:
1908 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1909 		break;
1910 	case KVM_SET_MSRS:
1911 		r = msr_io(vcpu, argp, do_set_msr, 0);
1912 		break;
1913 	case KVM_TPR_ACCESS_REPORTING: {
1914 		struct kvm_tpr_access_ctl tac;
1915 
1916 		r = -EFAULT;
1917 		if (copy_from_user(&tac, argp, sizeof tac))
1918 			goto out;
1919 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1920 		if (r)
1921 			goto out;
1922 		r = -EFAULT;
1923 		if (copy_to_user(argp, &tac, sizeof tac))
1924 			goto out;
1925 		r = 0;
1926 		break;
1927 	};
1928 	case KVM_SET_VAPIC_ADDR: {
1929 		struct kvm_vapic_addr va;
1930 
1931 		r = -EINVAL;
1932 		if (!irqchip_in_kernel(vcpu->kvm))
1933 			goto out;
1934 		r = -EFAULT;
1935 		if (copy_from_user(&va, argp, sizeof va))
1936 			goto out;
1937 		r = 0;
1938 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1939 		break;
1940 	}
1941 	case KVM_X86_SETUP_MCE: {
1942 		u64 mcg_cap;
1943 
1944 		r = -EFAULT;
1945 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1946 			goto out;
1947 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1948 		break;
1949 	}
1950 	case KVM_X86_SET_MCE: {
1951 		struct kvm_x86_mce mce;
1952 
1953 		r = -EFAULT;
1954 		if (copy_from_user(&mce, argp, sizeof mce))
1955 			goto out;
1956 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1957 		break;
1958 	}
1959 	default:
1960 		r = -EINVAL;
1961 	}
1962 out:
1963 	kfree(lapic);
1964 	return r;
1965 }
1966 
1967 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1968 {
1969 	int ret;
1970 
1971 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1972 		return -1;
1973 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1974 	return ret;
1975 }
1976 
1977 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
1978 					      u64 ident_addr)
1979 {
1980 	kvm->arch.ept_identity_map_addr = ident_addr;
1981 	return 0;
1982 }
1983 
1984 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1985 					  u32 kvm_nr_mmu_pages)
1986 {
1987 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1988 		return -EINVAL;
1989 
1990 	down_write(&kvm->slots_lock);
1991 	spin_lock(&kvm->mmu_lock);
1992 
1993 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1994 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1995 
1996 	spin_unlock(&kvm->mmu_lock);
1997 	up_write(&kvm->slots_lock);
1998 	return 0;
1999 }
2000 
2001 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2002 {
2003 	return kvm->arch.n_alloc_mmu_pages;
2004 }
2005 
2006 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2007 {
2008 	int i;
2009 	struct kvm_mem_alias *alias;
2010 
2011 	for (i = 0; i < kvm->arch.naliases; ++i) {
2012 		alias = &kvm->arch.aliases[i];
2013 		if (gfn >= alias->base_gfn
2014 		    && gfn < alias->base_gfn + alias->npages)
2015 			return alias->target_gfn + gfn - alias->base_gfn;
2016 	}
2017 	return gfn;
2018 }
2019 
2020 /*
2021  * Set a new alias region.  Aliases map a portion of physical memory into
2022  * another portion.  This is useful for memory windows, for example the PC
2023  * VGA region.
2024  */
2025 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2026 					 struct kvm_memory_alias *alias)
2027 {
2028 	int r, n;
2029 	struct kvm_mem_alias *p;
2030 
2031 	r = -EINVAL;
2032 	/* General sanity checks */
2033 	if (alias->memory_size & (PAGE_SIZE - 1))
2034 		goto out;
2035 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2036 		goto out;
2037 	if (alias->slot >= KVM_ALIAS_SLOTS)
2038 		goto out;
2039 	if (alias->guest_phys_addr + alias->memory_size
2040 	    < alias->guest_phys_addr)
2041 		goto out;
2042 	if (alias->target_phys_addr + alias->memory_size
2043 	    < alias->target_phys_addr)
2044 		goto out;
2045 
2046 	down_write(&kvm->slots_lock);
2047 	spin_lock(&kvm->mmu_lock);
2048 
2049 	p = &kvm->arch.aliases[alias->slot];
2050 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2051 	p->npages = alias->memory_size >> PAGE_SHIFT;
2052 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2053 
2054 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2055 		if (kvm->arch.aliases[n - 1].npages)
2056 			break;
2057 	kvm->arch.naliases = n;
2058 
2059 	spin_unlock(&kvm->mmu_lock);
2060 	kvm_mmu_zap_all(kvm);
2061 
2062 	up_write(&kvm->slots_lock);
2063 
2064 	return 0;
2065 
2066 out:
2067 	return r;
2068 }
2069 
2070 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2071 {
2072 	int r;
2073 
2074 	r = 0;
2075 	switch (chip->chip_id) {
2076 	case KVM_IRQCHIP_PIC_MASTER:
2077 		memcpy(&chip->chip.pic,
2078 			&pic_irqchip(kvm)->pics[0],
2079 			sizeof(struct kvm_pic_state));
2080 		break;
2081 	case KVM_IRQCHIP_PIC_SLAVE:
2082 		memcpy(&chip->chip.pic,
2083 			&pic_irqchip(kvm)->pics[1],
2084 			sizeof(struct kvm_pic_state));
2085 		break;
2086 	case KVM_IRQCHIP_IOAPIC:
2087 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2088 		break;
2089 	default:
2090 		r = -EINVAL;
2091 		break;
2092 	}
2093 	return r;
2094 }
2095 
2096 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2097 {
2098 	int r;
2099 
2100 	r = 0;
2101 	switch (chip->chip_id) {
2102 	case KVM_IRQCHIP_PIC_MASTER:
2103 		spin_lock(&pic_irqchip(kvm)->lock);
2104 		memcpy(&pic_irqchip(kvm)->pics[0],
2105 			&chip->chip.pic,
2106 			sizeof(struct kvm_pic_state));
2107 		spin_unlock(&pic_irqchip(kvm)->lock);
2108 		break;
2109 	case KVM_IRQCHIP_PIC_SLAVE:
2110 		spin_lock(&pic_irqchip(kvm)->lock);
2111 		memcpy(&pic_irqchip(kvm)->pics[1],
2112 			&chip->chip.pic,
2113 			sizeof(struct kvm_pic_state));
2114 		spin_unlock(&pic_irqchip(kvm)->lock);
2115 		break;
2116 	case KVM_IRQCHIP_IOAPIC:
2117 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2118 		break;
2119 	default:
2120 		r = -EINVAL;
2121 		break;
2122 	}
2123 	kvm_pic_update_irq(pic_irqchip(kvm));
2124 	return r;
2125 }
2126 
2127 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2128 {
2129 	int r = 0;
2130 
2131 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2132 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2133 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2134 	return r;
2135 }
2136 
2137 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2138 {
2139 	int r = 0;
2140 
2141 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2142 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2143 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2144 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2145 	return r;
2146 }
2147 
2148 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2149 {
2150 	int r = 0;
2151 
2152 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2153 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2154 		sizeof(ps->channels));
2155 	ps->flags = kvm->arch.vpit->pit_state.flags;
2156 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2157 	return r;
2158 }
2159 
2160 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2161 {
2162 	int r = 0, start = 0;
2163 	u32 prev_legacy, cur_legacy;
2164 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2165 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2166 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2167 	if (!prev_legacy && cur_legacy)
2168 		start = 1;
2169 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2170 	       sizeof(kvm->arch.vpit->pit_state.channels));
2171 	kvm->arch.vpit->pit_state.flags = ps->flags;
2172 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2173 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2174 	return r;
2175 }
2176 
2177 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2178 				 struct kvm_reinject_control *control)
2179 {
2180 	if (!kvm->arch.vpit)
2181 		return -ENXIO;
2182 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2183 	kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2184 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2185 	return 0;
2186 }
2187 
2188 /*
2189  * Get (and clear) the dirty memory log for a memory slot.
2190  */
2191 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2192 				      struct kvm_dirty_log *log)
2193 {
2194 	int r;
2195 	int n;
2196 	struct kvm_memory_slot *memslot;
2197 	int is_dirty = 0;
2198 
2199 	down_write(&kvm->slots_lock);
2200 
2201 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
2202 	if (r)
2203 		goto out;
2204 
2205 	/* If nothing is dirty, don't bother messing with page tables. */
2206 	if (is_dirty) {
2207 		spin_lock(&kvm->mmu_lock);
2208 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
2209 		spin_unlock(&kvm->mmu_lock);
2210 		memslot = &kvm->memslots[log->slot];
2211 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2212 		memset(memslot->dirty_bitmap, 0, n);
2213 	}
2214 	r = 0;
2215 out:
2216 	up_write(&kvm->slots_lock);
2217 	return r;
2218 }
2219 
2220 long kvm_arch_vm_ioctl(struct file *filp,
2221 		       unsigned int ioctl, unsigned long arg)
2222 {
2223 	struct kvm *kvm = filp->private_data;
2224 	void __user *argp = (void __user *)arg;
2225 	int r = -ENOTTY;
2226 	/*
2227 	 * This union makes it completely explicit to gcc-3.x
2228 	 * that these two variables' stack usage should be
2229 	 * combined, not added together.
2230 	 */
2231 	union {
2232 		struct kvm_pit_state ps;
2233 		struct kvm_pit_state2 ps2;
2234 		struct kvm_memory_alias alias;
2235 		struct kvm_pit_config pit_config;
2236 	} u;
2237 
2238 	switch (ioctl) {
2239 	case KVM_SET_TSS_ADDR:
2240 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2241 		if (r < 0)
2242 			goto out;
2243 		break;
2244 	case KVM_SET_IDENTITY_MAP_ADDR: {
2245 		u64 ident_addr;
2246 
2247 		r = -EFAULT;
2248 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2249 			goto out;
2250 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2251 		if (r < 0)
2252 			goto out;
2253 		break;
2254 	}
2255 	case KVM_SET_MEMORY_REGION: {
2256 		struct kvm_memory_region kvm_mem;
2257 		struct kvm_userspace_memory_region kvm_userspace_mem;
2258 
2259 		r = -EFAULT;
2260 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2261 			goto out;
2262 		kvm_userspace_mem.slot = kvm_mem.slot;
2263 		kvm_userspace_mem.flags = kvm_mem.flags;
2264 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2265 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2266 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2267 		if (r)
2268 			goto out;
2269 		break;
2270 	}
2271 	case KVM_SET_NR_MMU_PAGES:
2272 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2273 		if (r)
2274 			goto out;
2275 		break;
2276 	case KVM_GET_NR_MMU_PAGES:
2277 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2278 		break;
2279 	case KVM_SET_MEMORY_ALIAS:
2280 		r = -EFAULT;
2281 		if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2282 			goto out;
2283 		r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2284 		if (r)
2285 			goto out;
2286 		break;
2287 	case KVM_CREATE_IRQCHIP:
2288 		r = -ENOMEM;
2289 		kvm->arch.vpic = kvm_create_pic(kvm);
2290 		if (kvm->arch.vpic) {
2291 			r = kvm_ioapic_init(kvm);
2292 			if (r) {
2293 				kfree(kvm->arch.vpic);
2294 				kvm->arch.vpic = NULL;
2295 				goto out;
2296 			}
2297 		} else
2298 			goto out;
2299 		r = kvm_setup_default_irq_routing(kvm);
2300 		if (r) {
2301 			kfree(kvm->arch.vpic);
2302 			kfree(kvm->arch.vioapic);
2303 			goto out;
2304 		}
2305 		break;
2306 	case KVM_CREATE_PIT:
2307 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2308 		goto create_pit;
2309 	case KVM_CREATE_PIT2:
2310 		r = -EFAULT;
2311 		if (copy_from_user(&u.pit_config, argp,
2312 				   sizeof(struct kvm_pit_config)))
2313 			goto out;
2314 	create_pit:
2315 		down_write(&kvm->slots_lock);
2316 		r = -EEXIST;
2317 		if (kvm->arch.vpit)
2318 			goto create_pit_unlock;
2319 		r = -ENOMEM;
2320 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2321 		if (kvm->arch.vpit)
2322 			r = 0;
2323 	create_pit_unlock:
2324 		up_write(&kvm->slots_lock);
2325 		break;
2326 	case KVM_IRQ_LINE_STATUS:
2327 	case KVM_IRQ_LINE: {
2328 		struct kvm_irq_level irq_event;
2329 
2330 		r = -EFAULT;
2331 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
2332 			goto out;
2333 		if (irqchip_in_kernel(kvm)) {
2334 			__s32 status;
2335 			status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2336 					irq_event.irq, irq_event.level);
2337 			if (ioctl == KVM_IRQ_LINE_STATUS) {
2338 				irq_event.status = status;
2339 				if (copy_to_user(argp, &irq_event,
2340 							sizeof irq_event))
2341 					goto out;
2342 			}
2343 			r = 0;
2344 		}
2345 		break;
2346 	}
2347 	case KVM_GET_IRQCHIP: {
2348 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2349 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2350 
2351 		r = -ENOMEM;
2352 		if (!chip)
2353 			goto out;
2354 		r = -EFAULT;
2355 		if (copy_from_user(chip, argp, sizeof *chip))
2356 			goto get_irqchip_out;
2357 		r = -ENXIO;
2358 		if (!irqchip_in_kernel(kvm))
2359 			goto get_irqchip_out;
2360 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2361 		if (r)
2362 			goto get_irqchip_out;
2363 		r = -EFAULT;
2364 		if (copy_to_user(argp, chip, sizeof *chip))
2365 			goto get_irqchip_out;
2366 		r = 0;
2367 	get_irqchip_out:
2368 		kfree(chip);
2369 		if (r)
2370 			goto out;
2371 		break;
2372 	}
2373 	case KVM_SET_IRQCHIP: {
2374 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2375 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2376 
2377 		r = -ENOMEM;
2378 		if (!chip)
2379 			goto out;
2380 		r = -EFAULT;
2381 		if (copy_from_user(chip, argp, sizeof *chip))
2382 			goto set_irqchip_out;
2383 		r = -ENXIO;
2384 		if (!irqchip_in_kernel(kvm))
2385 			goto set_irqchip_out;
2386 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2387 		if (r)
2388 			goto set_irqchip_out;
2389 		r = 0;
2390 	set_irqchip_out:
2391 		kfree(chip);
2392 		if (r)
2393 			goto out;
2394 		break;
2395 	}
2396 	case KVM_GET_PIT: {
2397 		r = -EFAULT;
2398 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2399 			goto out;
2400 		r = -ENXIO;
2401 		if (!kvm->arch.vpit)
2402 			goto out;
2403 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2404 		if (r)
2405 			goto out;
2406 		r = -EFAULT;
2407 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2408 			goto out;
2409 		r = 0;
2410 		break;
2411 	}
2412 	case KVM_SET_PIT: {
2413 		r = -EFAULT;
2414 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
2415 			goto out;
2416 		r = -ENXIO;
2417 		if (!kvm->arch.vpit)
2418 			goto out;
2419 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2420 		if (r)
2421 			goto out;
2422 		r = 0;
2423 		break;
2424 	}
2425 	case KVM_GET_PIT2: {
2426 		r = -ENXIO;
2427 		if (!kvm->arch.vpit)
2428 			goto out;
2429 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2430 		if (r)
2431 			goto out;
2432 		r = -EFAULT;
2433 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2434 			goto out;
2435 		r = 0;
2436 		break;
2437 	}
2438 	case KVM_SET_PIT2: {
2439 		r = -EFAULT;
2440 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2441 			goto out;
2442 		r = -ENXIO;
2443 		if (!kvm->arch.vpit)
2444 			goto out;
2445 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2446 		if (r)
2447 			goto out;
2448 		r = 0;
2449 		break;
2450 	}
2451 	case KVM_REINJECT_CONTROL: {
2452 		struct kvm_reinject_control control;
2453 		r =  -EFAULT;
2454 		if (copy_from_user(&control, argp, sizeof(control)))
2455 			goto out;
2456 		r = kvm_vm_ioctl_reinject(kvm, &control);
2457 		if (r)
2458 			goto out;
2459 		r = 0;
2460 		break;
2461 	}
2462 	case KVM_XEN_HVM_CONFIG: {
2463 		r = -EFAULT;
2464 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2465 				   sizeof(struct kvm_xen_hvm_config)))
2466 			goto out;
2467 		r = -EINVAL;
2468 		if (kvm->arch.xen_hvm_config.flags)
2469 			goto out;
2470 		r = 0;
2471 		break;
2472 	}
2473 	case KVM_SET_CLOCK: {
2474 		struct timespec now;
2475 		struct kvm_clock_data user_ns;
2476 		u64 now_ns;
2477 		s64 delta;
2478 
2479 		r = -EFAULT;
2480 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2481 			goto out;
2482 
2483 		r = -EINVAL;
2484 		if (user_ns.flags)
2485 			goto out;
2486 
2487 		r = 0;
2488 		ktime_get_ts(&now);
2489 		now_ns = timespec_to_ns(&now);
2490 		delta = user_ns.clock - now_ns;
2491 		kvm->arch.kvmclock_offset = delta;
2492 		break;
2493 	}
2494 	case KVM_GET_CLOCK: {
2495 		struct timespec now;
2496 		struct kvm_clock_data user_ns;
2497 		u64 now_ns;
2498 
2499 		ktime_get_ts(&now);
2500 		now_ns = timespec_to_ns(&now);
2501 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2502 		user_ns.flags = 0;
2503 
2504 		r = -EFAULT;
2505 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2506 			goto out;
2507 		r = 0;
2508 		break;
2509 	}
2510 
2511 	default:
2512 		;
2513 	}
2514 out:
2515 	return r;
2516 }
2517 
2518 static void kvm_init_msr_list(void)
2519 {
2520 	u32 dummy[2];
2521 	unsigned i, j;
2522 
2523 	/* skip the first msrs in the list. KVM-specific */
2524 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2525 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2526 			continue;
2527 		if (j < i)
2528 			msrs_to_save[j] = msrs_to_save[i];
2529 		j++;
2530 	}
2531 	num_msrs_to_save = j;
2532 }
2533 
2534 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2535 			   const void *v)
2536 {
2537 	if (vcpu->arch.apic &&
2538 	    !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2539 		return 0;
2540 
2541 	return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2542 }
2543 
2544 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2545 {
2546 	if (vcpu->arch.apic &&
2547 	    !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2548 		return 0;
2549 
2550 	return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2551 }
2552 
2553 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2554 			       struct kvm_vcpu *vcpu)
2555 {
2556 	void *data = val;
2557 	int r = X86EMUL_CONTINUE;
2558 
2559 	while (bytes) {
2560 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2561 		unsigned offset = addr & (PAGE_SIZE-1);
2562 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2563 		int ret;
2564 
2565 		if (gpa == UNMAPPED_GVA) {
2566 			r = X86EMUL_PROPAGATE_FAULT;
2567 			goto out;
2568 		}
2569 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2570 		if (ret < 0) {
2571 			r = X86EMUL_UNHANDLEABLE;
2572 			goto out;
2573 		}
2574 
2575 		bytes -= toread;
2576 		data += toread;
2577 		addr += toread;
2578 	}
2579 out:
2580 	return r;
2581 }
2582 
2583 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2584 				struct kvm_vcpu *vcpu)
2585 {
2586 	void *data = val;
2587 	int r = X86EMUL_CONTINUE;
2588 
2589 	while (bytes) {
2590 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2591 		unsigned offset = addr & (PAGE_SIZE-1);
2592 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2593 		int ret;
2594 
2595 		if (gpa == UNMAPPED_GVA) {
2596 			r = X86EMUL_PROPAGATE_FAULT;
2597 			goto out;
2598 		}
2599 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2600 		if (ret < 0) {
2601 			r = X86EMUL_UNHANDLEABLE;
2602 			goto out;
2603 		}
2604 
2605 		bytes -= towrite;
2606 		data += towrite;
2607 		addr += towrite;
2608 	}
2609 out:
2610 	return r;
2611 }
2612 
2613 
2614 static int emulator_read_emulated(unsigned long addr,
2615 				  void *val,
2616 				  unsigned int bytes,
2617 				  struct kvm_vcpu *vcpu)
2618 {
2619 	gpa_t                 gpa;
2620 
2621 	if (vcpu->mmio_read_completed) {
2622 		memcpy(val, vcpu->mmio_data, bytes);
2623 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2624 			       vcpu->mmio_phys_addr, *(u64 *)val);
2625 		vcpu->mmio_read_completed = 0;
2626 		return X86EMUL_CONTINUE;
2627 	}
2628 
2629 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2630 
2631 	/* For APIC access vmexit */
2632 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2633 		goto mmio;
2634 
2635 	if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2636 				== X86EMUL_CONTINUE)
2637 		return X86EMUL_CONTINUE;
2638 	if (gpa == UNMAPPED_GVA)
2639 		return X86EMUL_PROPAGATE_FAULT;
2640 
2641 mmio:
2642 	/*
2643 	 * Is this MMIO handled locally?
2644 	 */
2645 	if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2646 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2647 		return X86EMUL_CONTINUE;
2648 	}
2649 
2650 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2651 
2652 	vcpu->mmio_needed = 1;
2653 	vcpu->mmio_phys_addr = gpa;
2654 	vcpu->mmio_size = bytes;
2655 	vcpu->mmio_is_write = 0;
2656 
2657 	return X86EMUL_UNHANDLEABLE;
2658 }
2659 
2660 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2661 			  const void *val, int bytes)
2662 {
2663 	int ret;
2664 
2665 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2666 	if (ret < 0)
2667 		return 0;
2668 	kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2669 	return 1;
2670 }
2671 
2672 static int emulator_write_emulated_onepage(unsigned long addr,
2673 					   const void *val,
2674 					   unsigned int bytes,
2675 					   struct kvm_vcpu *vcpu)
2676 {
2677 	gpa_t                 gpa;
2678 
2679 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2680 
2681 	if (gpa == UNMAPPED_GVA) {
2682 		kvm_inject_page_fault(vcpu, addr, 2);
2683 		return X86EMUL_PROPAGATE_FAULT;
2684 	}
2685 
2686 	/* For APIC access vmexit */
2687 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2688 		goto mmio;
2689 
2690 	if (emulator_write_phys(vcpu, gpa, val, bytes))
2691 		return X86EMUL_CONTINUE;
2692 
2693 mmio:
2694 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2695 	/*
2696 	 * Is this MMIO handled locally?
2697 	 */
2698 	if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2699 		return X86EMUL_CONTINUE;
2700 
2701 	vcpu->mmio_needed = 1;
2702 	vcpu->mmio_phys_addr = gpa;
2703 	vcpu->mmio_size = bytes;
2704 	vcpu->mmio_is_write = 1;
2705 	memcpy(vcpu->mmio_data, val, bytes);
2706 
2707 	return X86EMUL_CONTINUE;
2708 }
2709 
2710 int emulator_write_emulated(unsigned long addr,
2711 				   const void *val,
2712 				   unsigned int bytes,
2713 				   struct kvm_vcpu *vcpu)
2714 {
2715 	/* Crossing a page boundary? */
2716 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2717 		int rc, now;
2718 
2719 		now = -addr & ~PAGE_MASK;
2720 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2721 		if (rc != X86EMUL_CONTINUE)
2722 			return rc;
2723 		addr += now;
2724 		val += now;
2725 		bytes -= now;
2726 	}
2727 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2728 }
2729 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2730 
2731 static int emulator_cmpxchg_emulated(unsigned long addr,
2732 				     const void *old,
2733 				     const void *new,
2734 				     unsigned int bytes,
2735 				     struct kvm_vcpu *vcpu)
2736 {
2737 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2738 #ifndef CONFIG_X86_64
2739 	/* guests cmpxchg8b have to be emulated atomically */
2740 	if (bytes == 8) {
2741 		gpa_t gpa;
2742 		struct page *page;
2743 		char *kaddr;
2744 		u64 val;
2745 
2746 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2747 
2748 		if (gpa == UNMAPPED_GVA ||
2749 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2750 			goto emul_write;
2751 
2752 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2753 			goto emul_write;
2754 
2755 		val = *(u64 *)new;
2756 
2757 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2758 
2759 		kaddr = kmap_atomic(page, KM_USER0);
2760 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2761 		kunmap_atomic(kaddr, KM_USER0);
2762 		kvm_release_page_dirty(page);
2763 	}
2764 emul_write:
2765 #endif
2766 
2767 	return emulator_write_emulated(addr, new, bytes, vcpu);
2768 }
2769 
2770 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2771 {
2772 	return kvm_x86_ops->get_segment_base(vcpu, seg);
2773 }
2774 
2775 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2776 {
2777 	kvm_mmu_invlpg(vcpu, address);
2778 	return X86EMUL_CONTINUE;
2779 }
2780 
2781 int emulate_clts(struct kvm_vcpu *vcpu)
2782 {
2783 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2784 	return X86EMUL_CONTINUE;
2785 }
2786 
2787 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2788 {
2789 	struct kvm_vcpu *vcpu = ctxt->vcpu;
2790 
2791 	switch (dr) {
2792 	case 0 ... 3:
2793 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
2794 		return X86EMUL_CONTINUE;
2795 	default:
2796 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2797 		return X86EMUL_UNHANDLEABLE;
2798 	}
2799 }
2800 
2801 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2802 {
2803 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2804 	int exception;
2805 
2806 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2807 	if (exception) {
2808 		/* FIXME: better handling */
2809 		return X86EMUL_UNHANDLEABLE;
2810 	}
2811 	return X86EMUL_CONTINUE;
2812 }
2813 
2814 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2815 {
2816 	u8 opcodes[4];
2817 	unsigned long rip = kvm_rip_read(vcpu);
2818 	unsigned long rip_linear;
2819 
2820 	if (!printk_ratelimit())
2821 		return;
2822 
2823 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2824 
2825 	kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2826 
2827 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2828 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2829 }
2830 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2831 
2832 static struct x86_emulate_ops emulate_ops = {
2833 	.read_std            = kvm_read_guest_virt,
2834 	.read_emulated       = emulator_read_emulated,
2835 	.write_emulated      = emulator_write_emulated,
2836 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2837 };
2838 
2839 static void cache_all_regs(struct kvm_vcpu *vcpu)
2840 {
2841 	kvm_register_read(vcpu, VCPU_REGS_RAX);
2842 	kvm_register_read(vcpu, VCPU_REGS_RSP);
2843 	kvm_register_read(vcpu, VCPU_REGS_RIP);
2844 	vcpu->arch.regs_dirty = ~0;
2845 }
2846 
2847 int emulate_instruction(struct kvm_vcpu *vcpu,
2848 			unsigned long cr2,
2849 			u16 error_code,
2850 			int emulation_type)
2851 {
2852 	int r, shadow_mask;
2853 	struct decode_cache *c;
2854 	struct kvm_run *run = vcpu->run;
2855 
2856 	kvm_clear_exception_queue(vcpu);
2857 	vcpu->arch.mmio_fault_cr2 = cr2;
2858 	/*
2859 	 * TODO: fix emulate.c to use guest_read/write_register
2860 	 * instead of direct ->regs accesses, can save hundred cycles
2861 	 * on Intel for instructions that don't read/change RSP, for
2862 	 * for example.
2863 	 */
2864 	cache_all_regs(vcpu);
2865 
2866 	vcpu->mmio_is_write = 0;
2867 	vcpu->arch.pio.string = 0;
2868 
2869 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2870 		int cs_db, cs_l;
2871 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2872 
2873 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2874 		vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
2875 		vcpu->arch.emulate_ctxt.mode =
2876 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2877 			? X86EMUL_MODE_REAL : cs_l
2878 			? X86EMUL_MODE_PROT64 :	cs_db
2879 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2880 
2881 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2882 
2883 		/* Only allow emulation of specific instructions on #UD
2884 		 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2885 		c = &vcpu->arch.emulate_ctxt.decode;
2886 		if (emulation_type & EMULTYPE_TRAP_UD) {
2887 			if (!c->twobyte)
2888 				return EMULATE_FAIL;
2889 			switch (c->b) {
2890 			case 0x01: /* VMMCALL */
2891 				if (c->modrm_mod != 3 || c->modrm_rm != 1)
2892 					return EMULATE_FAIL;
2893 				break;
2894 			case 0x34: /* sysenter */
2895 			case 0x35: /* sysexit */
2896 				if (c->modrm_mod != 0 || c->modrm_rm != 0)
2897 					return EMULATE_FAIL;
2898 				break;
2899 			case 0x05: /* syscall */
2900 				if (c->modrm_mod != 0 || c->modrm_rm != 0)
2901 					return EMULATE_FAIL;
2902 				break;
2903 			default:
2904 				return EMULATE_FAIL;
2905 			}
2906 
2907 			if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2908 				return EMULATE_FAIL;
2909 		}
2910 
2911 		++vcpu->stat.insn_emulation;
2912 		if (r)  {
2913 			++vcpu->stat.insn_emulation_fail;
2914 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2915 				return EMULATE_DONE;
2916 			return EMULATE_FAIL;
2917 		}
2918 	}
2919 
2920 	if (emulation_type & EMULTYPE_SKIP) {
2921 		kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2922 		return EMULATE_DONE;
2923 	}
2924 
2925 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2926 	shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2927 
2928 	if (r == 0)
2929 		kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2930 
2931 	if (vcpu->arch.pio.string)
2932 		return EMULATE_DO_MMIO;
2933 
2934 	if ((r || vcpu->mmio_is_write) && run) {
2935 		run->exit_reason = KVM_EXIT_MMIO;
2936 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2937 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2938 		run->mmio.len = vcpu->mmio_size;
2939 		run->mmio.is_write = vcpu->mmio_is_write;
2940 	}
2941 
2942 	if (r) {
2943 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2944 			return EMULATE_DONE;
2945 		if (!vcpu->mmio_needed) {
2946 			kvm_report_emulation_failure(vcpu, "mmio");
2947 			return EMULATE_FAIL;
2948 		}
2949 		return EMULATE_DO_MMIO;
2950 	}
2951 
2952 	kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2953 
2954 	if (vcpu->mmio_is_write) {
2955 		vcpu->mmio_needed = 0;
2956 		return EMULATE_DO_MMIO;
2957 	}
2958 
2959 	return EMULATE_DONE;
2960 }
2961 EXPORT_SYMBOL_GPL(emulate_instruction);
2962 
2963 static int pio_copy_data(struct kvm_vcpu *vcpu)
2964 {
2965 	void *p = vcpu->arch.pio_data;
2966 	gva_t q = vcpu->arch.pio.guest_gva;
2967 	unsigned bytes;
2968 	int ret;
2969 
2970 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2971 	if (vcpu->arch.pio.in)
2972 		ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2973 	else
2974 		ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2975 	return ret;
2976 }
2977 
2978 int complete_pio(struct kvm_vcpu *vcpu)
2979 {
2980 	struct kvm_pio_request *io = &vcpu->arch.pio;
2981 	long delta;
2982 	int r;
2983 	unsigned long val;
2984 
2985 	if (!io->string) {
2986 		if (io->in) {
2987 			val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2988 			memcpy(&val, vcpu->arch.pio_data, io->size);
2989 			kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2990 		}
2991 	} else {
2992 		if (io->in) {
2993 			r = pio_copy_data(vcpu);
2994 			if (r)
2995 				return r;
2996 		}
2997 
2998 		delta = 1;
2999 		if (io->rep) {
3000 			delta *= io->cur_count;
3001 			/*
3002 			 * The size of the register should really depend on
3003 			 * current address size.
3004 			 */
3005 			val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3006 			val -= delta;
3007 			kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3008 		}
3009 		if (io->down)
3010 			delta = -delta;
3011 		delta *= io->size;
3012 		if (io->in) {
3013 			val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3014 			val += delta;
3015 			kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3016 		} else {
3017 			val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3018 			val += delta;
3019 			kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3020 		}
3021 	}
3022 
3023 	io->count -= io->cur_count;
3024 	io->cur_count = 0;
3025 
3026 	return 0;
3027 }
3028 
3029 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3030 {
3031 	/* TODO: String I/O for in kernel device */
3032 	int r;
3033 
3034 	if (vcpu->arch.pio.in)
3035 		r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3036 				    vcpu->arch.pio.size, pd);
3037 	else
3038 		r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3039 				     vcpu->arch.pio.size, pd);
3040 	return r;
3041 }
3042 
3043 static int pio_string_write(struct kvm_vcpu *vcpu)
3044 {
3045 	struct kvm_pio_request *io = &vcpu->arch.pio;
3046 	void *pd = vcpu->arch.pio_data;
3047 	int i, r = 0;
3048 
3049 	for (i = 0; i < io->cur_count; i++) {
3050 		if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
3051 				     io->port, io->size, pd)) {
3052 			r = -EOPNOTSUPP;
3053 			break;
3054 		}
3055 		pd += io->size;
3056 	}
3057 	return r;
3058 }
3059 
3060 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3061 {
3062 	unsigned long val;
3063 
3064 	vcpu->run->exit_reason = KVM_EXIT_IO;
3065 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3066 	vcpu->run->io.size = vcpu->arch.pio.size = size;
3067 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3068 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3069 	vcpu->run->io.port = vcpu->arch.pio.port = port;
3070 	vcpu->arch.pio.in = in;
3071 	vcpu->arch.pio.string = 0;
3072 	vcpu->arch.pio.down = 0;
3073 	vcpu->arch.pio.rep = 0;
3074 
3075 	trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3076 		      size, 1);
3077 
3078 	val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3079 	memcpy(vcpu->arch.pio_data, &val, 4);
3080 
3081 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3082 		complete_pio(vcpu);
3083 		return 1;
3084 	}
3085 	return 0;
3086 }
3087 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3088 
3089 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3090 		  int size, unsigned long count, int down,
3091 		  gva_t address, int rep, unsigned port)
3092 {
3093 	unsigned now, in_page;
3094 	int ret = 0;
3095 
3096 	vcpu->run->exit_reason = KVM_EXIT_IO;
3097 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3098 	vcpu->run->io.size = vcpu->arch.pio.size = size;
3099 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3100 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3101 	vcpu->run->io.port = vcpu->arch.pio.port = port;
3102 	vcpu->arch.pio.in = in;
3103 	vcpu->arch.pio.string = 1;
3104 	vcpu->arch.pio.down = down;
3105 	vcpu->arch.pio.rep = rep;
3106 
3107 	trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3108 		      size, count);
3109 
3110 	if (!count) {
3111 		kvm_x86_ops->skip_emulated_instruction(vcpu);
3112 		return 1;
3113 	}
3114 
3115 	if (!down)
3116 		in_page = PAGE_SIZE - offset_in_page(address);
3117 	else
3118 		in_page = offset_in_page(address) + size;
3119 	now = min(count, (unsigned long)in_page / size);
3120 	if (!now)
3121 		now = 1;
3122 	if (down) {
3123 		/*
3124 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3125 		 */
3126 		pr_unimpl(vcpu, "guest string pio down\n");
3127 		kvm_inject_gp(vcpu, 0);
3128 		return 1;
3129 	}
3130 	vcpu->run->io.count = now;
3131 	vcpu->arch.pio.cur_count = now;
3132 
3133 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3134 		kvm_x86_ops->skip_emulated_instruction(vcpu);
3135 
3136 	vcpu->arch.pio.guest_gva = address;
3137 
3138 	if (!vcpu->arch.pio.in) {
3139 		/* string PIO write */
3140 		ret = pio_copy_data(vcpu);
3141 		if (ret == X86EMUL_PROPAGATE_FAULT) {
3142 			kvm_inject_gp(vcpu, 0);
3143 			return 1;
3144 		}
3145 		if (ret == 0 && !pio_string_write(vcpu)) {
3146 			complete_pio(vcpu);
3147 			if (vcpu->arch.pio.count == 0)
3148 				ret = 1;
3149 		}
3150 	}
3151 	/* no string PIO read support yet */
3152 
3153 	return ret;
3154 }
3155 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3156 
3157 static void bounce_off(void *info)
3158 {
3159 	/* nothing */
3160 }
3161 
3162 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3163 				     void *data)
3164 {
3165 	struct cpufreq_freqs *freq = data;
3166 	struct kvm *kvm;
3167 	struct kvm_vcpu *vcpu;
3168 	int i, send_ipi = 0;
3169 
3170 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3171 		return 0;
3172 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3173 		return 0;
3174 	per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3175 
3176 	spin_lock(&kvm_lock);
3177 	list_for_each_entry(kvm, &vm_list, vm_list) {
3178 		kvm_for_each_vcpu(i, vcpu, kvm) {
3179 			if (vcpu->cpu != freq->cpu)
3180 				continue;
3181 			if (!kvm_request_guest_time_update(vcpu))
3182 				continue;
3183 			if (vcpu->cpu != smp_processor_id())
3184 				send_ipi++;
3185 		}
3186 	}
3187 	spin_unlock(&kvm_lock);
3188 
3189 	if (freq->old < freq->new && send_ipi) {
3190 		/*
3191 		 * We upscale the frequency.  Must make the guest
3192 		 * doesn't see old kvmclock values while running with
3193 		 * the new frequency, otherwise we risk the guest sees
3194 		 * time go backwards.
3195 		 *
3196 		 * In case we update the frequency for another cpu
3197 		 * (which might be in guest context) send an interrupt
3198 		 * to kick the cpu out of guest context.  Next time
3199 		 * guest context is entered kvmclock will be updated,
3200 		 * so the guest will not see stale values.
3201 		 */
3202 		smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3203 	}
3204 	return 0;
3205 }
3206 
3207 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3208         .notifier_call  = kvmclock_cpufreq_notifier
3209 };
3210 
3211 static void kvm_timer_init(void)
3212 {
3213 	int cpu;
3214 
3215 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3216 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3217 					  CPUFREQ_TRANSITION_NOTIFIER);
3218 		for_each_online_cpu(cpu) {
3219 			unsigned long khz = cpufreq_get(cpu);
3220 			if (!khz)
3221 				khz = tsc_khz;
3222 			per_cpu(cpu_tsc_khz, cpu) = khz;
3223 		}
3224 	} else {
3225 		for_each_possible_cpu(cpu)
3226 			per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3227 	}
3228 }
3229 
3230 int kvm_arch_init(void *opaque)
3231 {
3232 	int r;
3233 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3234 
3235 	if (kvm_x86_ops) {
3236 		printk(KERN_ERR "kvm: already loaded the other module\n");
3237 		r = -EEXIST;
3238 		goto out;
3239 	}
3240 
3241 	if (!ops->cpu_has_kvm_support()) {
3242 		printk(KERN_ERR "kvm: no hardware support\n");
3243 		r = -EOPNOTSUPP;
3244 		goto out;
3245 	}
3246 	if (ops->disabled_by_bios()) {
3247 		printk(KERN_ERR "kvm: disabled by bios\n");
3248 		r = -EOPNOTSUPP;
3249 		goto out;
3250 	}
3251 
3252 	r = kvm_mmu_module_init();
3253 	if (r)
3254 		goto out;
3255 
3256 	kvm_init_msr_list();
3257 
3258 	kvm_x86_ops = ops;
3259 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3260 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3261 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3262 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
3263 
3264 	kvm_timer_init();
3265 
3266 	return 0;
3267 
3268 out:
3269 	return r;
3270 }
3271 
3272 void kvm_arch_exit(void)
3273 {
3274 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3275 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3276 					    CPUFREQ_TRANSITION_NOTIFIER);
3277 	kvm_x86_ops = NULL;
3278 	kvm_mmu_module_exit();
3279 }
3280 
3281 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3282 {
3283 	++vcpu->stat.halt_exits;
3284 	if (irqchip_in_kernel(vcpu->kvm)) {
3285 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3286 		return 1;
3287 	} else {
3288 		vcpu->run->exit_reason = KVM_EXIT_HLT;
3289 		return 0;
3290 	}
3291 }
3292 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3293 
3294 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3295 			   unsigned long a1)
3296 {
3297 	if (is_long_mode(vcpu))
3298 		return a0;
3299 	else
3300 		return a0 | ((gpa_t)a1 << 32);
3301 }
3302 
3303 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3304 {
3305 	unsigned long nr, a0, a1, a2, a3, ret;
3306 	int r = 1;
3307 
3308 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3309 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3310 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3311 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3312 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3313 
3314 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
3315 
3316 	if (!is_long_mode(vcpu)) {
3317 		nr &= 0xFFFFFFFF;
3318 		a0 &= 0xFFFFFFFF;
3319 		a1 &= 0xFFFFFFFF;
3320 		a2 &= 0xFFFFFFFF;
3321 		a3 &= 0xFFFFFFFF;
3322 	}
3323 
3324 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3325 		ret = -KVM_EPERM;
3326 		goto out;
3327 	}
3328 
3329 	switch (nr) {
3330 	case KVM_HC_VAPIC_POLL_IRQ:
3331 		ret = 0;
3332 		break;
3333 	case KVM_HC_MMU_OP:
3334 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3335 		break;
3336 	default:
3337 		ret = -KVM_ENOSYS;
3338 		break;
3339 	}
3340 out:
3341 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3342 	++vcpu->stat.hypercalls;
3343 	return r;
3344 }
3345 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3346 
3347 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3348 {
3349 	char instruction[3];
3350 	int ret = 0;
3351 	unsigned long rip = kvm_rip_read(vcpu);
3352 
3353 
3354 	/*
3355 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
3356 	 * to ensure that the updated hypercall appears atomically across all
3357 	 * VCPUs.
3358 	 */
3359 	kvm_mmu_zap_all(vcpu->kvm);
3360 
3361 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
3362 	if (emulator_write_emulated(rip, instruction, 3, vcpu)
3363 	    != X86EMUL_CONTINUE)
3364 		ret = -EFAULT;
3365 
3366 	return ret;
3367 }
3368 
3369 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3370 {
3371 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3372 }
3373 
3374 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3375 {
3376 	struct descriptor_table dt = { limit, base };
3377 
3378 	kvm_x86_ops->set_gdt(vcpu, &dt);
3379 }
3380 
3381 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3382 {
3383 	struct descriptor_table dt = { limit, base };
3384 
3385 	kvm_x86_ops->set_idt(vcpu, &dt);
3386 }
3387 
3388 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3389 		   unsigned long *rflags)
3390 {
3391 	kvm_lmsw(vcpu, msw);
3392 	*rflags = kvm_get_rflags(vcpu);
3393 }
3394 
3395 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3396 {
3397 	unsigned long value;
3398 
3399 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3400 	switch (cr) {
3401 	case 0:
3402 		value = vcpu->arch.cr0;
3403 		break;
3404 	case 2:
3405 		value = vcpu->arch.cr2;
3406 		break;
3407 	case 3:
3408 		value = vcpu->arch.cr3;
3409 		break;
3410 	case 4:
3411 		value = vcpu->arch.cr4;
3412 		break;
3413 	case 8:
3414 		value = kvm_get_cr8(vcpu);
3415 		break;
3416 	default:
3417 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3418 		return 0;
3419 	}
3420 
3421 	return value;
3422 }
3423 
3424 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3425 		     unsigned long *rflags)
3426 {
3427 	switch (cr) {
3428 	case 0:
3429 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3430 		*rflags = kvm_get_rflags(vcpu);
3431 		break;
3432 	case 2:
3433 		vcpu->arch.cr2 = val;
3434 		break;
3435 	case 3:
3436 		kvm_set_cr3(vcpu, val);
3437 		break;
3438 	case 4:
3439 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3440 		break;
3441 	case 8:
3442 		kvm_set_cr8(vcpu, val & 0xfUL);
3443 		break;
3444 	default:
3445 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3446 	}
3447 }
3448 
3449 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3450 {
3451 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3452 	int j, nent = vcpu->arch.cpuid_nent;
3453 
3454 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3455 	/* when no next entry is found, the current entry[i] is reselected */
3456 	for (j = i + 1; ; j = (j + 1) % nent) {
3457 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3458 		if (ej->function == e->function) {
3459 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3460 			return j;
3461 		}
3462 	}
3463 	return 0; /* silence gcc, even though control never reaches here */
3464 }
3465 
3466 /* find an entry with matching function, matching index (if needed), and that
3467  * should be read next (if it's stateful) */
3468 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3469 	u32 function, u32 index)
3470 {
3471 	if (e->function != function)
3472 		return 0;
3473 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3474 		return 0;
3475 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3476 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3477 		return 0;
3478 	return 1;
3479 }
3480 
3481 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3482 					      u32 function, u32 index)
3483 {
3484 	int i;
3485 	struct kvm_cpuid_entry2 *best = NULL;
3486 
3487 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3488 		struct kvm_cpuid_entry2 *e;
3489 
3490 		e = &vcpu->arch.cpuid_entries[i];
3491 		if (is_matching_cpuid_entry(e, function, index)) {
3492 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3493 				move_to_next_stateful_cpuid_entry(vcpu, i);
3494 			best = e;
3495 			break;
3496 		}
3497 		/*
3498 		 * Both basic or both extended?
3499 		 */
3500 		if (((e->function ^ function) & 0x80000000) == 0)
3501 			if (!best || e->function > best->function)
3502 				best = e;
3503 	}
3504 	return best;
3505 }
3506 
3507 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3508 {
3509 	struct kvm_cpuid_entry2 *best;
3510 
3511 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3512 	if (best)
3513 		return best->eax & 0xff;
3514 	return 36;
3515 }
3516 
3517 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3518 {
3519 	u32 function, index;
3520 	struct kvm_cpuid_entry2 *best;
3521 
3522 	function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3523 	index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3524 	kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3525 	kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3526 	kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3527 	kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3528 	best = kvm_find_cpuid_entry(vcpu, function, index);
3529 	if (best) {
3530 		kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3531 		kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3532 		kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3533 		kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3534 	}
3535 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3536 	trace_kvm_cpuid(function,
3537 			kvm_register_read(vcpu, VCPU_REGS_RAX),
3538 			kvm_register_read(vcpu, VCPU_REGS_RBX),
3539 			kvm_register_read(vcpu, VCPU_REGS_RCX),
3540 			kvm_register_read(vcpu, VCPU_REGS_RDX));
3541 }
3542 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3543 
3544 /*
3545  * Check if userspace requested an interrupt window, and that the
3546  * interrupt window is open.
3547  *
3548  * No need to exit to userspace if we already have an interrupt queued.
3549  */
3550 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3551 {
3552 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3553 		vcpu->run->request_interrupt_window &&
3554 		kvm_arch_interrupt_allowed(vcpu));
3555 }
3556 
3557 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3558 {
3559 	struct kvm_run *kvm_run = vcpu->run;
3560 
3561 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3562 	kvm_run->cr8 = kvm_get_cr8(vcpu);
3563 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
3564 	if (irqchip_in_kernel(vcpu->kvm))
3565 		kvm_run->ready_for_interrupt_injection = 1;
3566 	else
3567 		kvm_run->ready_for_interrupt_injection =
3568 			kvm_arch_interrupt_allowed(vcpu) &&
3569 			!kvm_cpu_has_interrupt(vcpu) &&
3570 			!kvm_event_needs_reinjection(vcpu);
3571 }
3572 
3573 static void vapic_enter(struct kvm_vcpu *vcpu)
3574 {
3575 	struct kvm_lapic *apic = vcpu->arch.apic;
3576 	struct page *page;
3577 
3578 	if (!apic || !apic->vapic_addr)
3579 		return;
3580 
3581 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3582 
3583 	vcpu->arch.apic->vapic_page = page;
3584 }
3585 
3586 static void vapic_exit(struct kvm_vcpu *vcpu)
3587 {
3588 	struct kvm_lapic *apic = vcpu->arch.apic;
3589 
3590 	if (!apic || !apic->vapic_addr)
3591 		return;
3592 
3593 	down_read(&vcpu->kvm->slots_lock);
3594 	kvm_release_page_dirty(apic->vapic_page);
3595 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3596 	up_read(&vcpu->kvm->slots_lock);
3597 }
3598 
3599 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3600 {
3601 	int max_irr, tpr;
3602 
3603 	if (!kvm_x86_ops->update_cr8_intercept)
3604 		return;
3605 
3606 	if (!vcpu->arch.apic)
3607 		return;
3608 
3609 	if (!vcpu->arch.apic->vapic_addr)
3610 		max_irr = kvm_lapic_find_highest_irr(vcpu);
3611 	else
3612 		max_irr = -1;
3613 
3614 	if (max_irr != -1)
3615 		max_irr >>= 4;
3616 
3617 	tpr = kvm_lapic_get_cr8(vcpu);
3618 
3619 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3620 }
3621 
3622 static void inject_pending_event(struct kvm_vcpu *vcpu)
3623 {
3624 	/* try to reinject previous events if any */
3625 	if (vcpu->arch.exception.pending) {
3626 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3627 					  vcpu->arch.exception.has_error_code,
3628 					  vcpu->arch.exception.error_code);
3629 		return;
3630 	}
3631 
3632 	if (vcpu->arch.nmi_injected) {
3633 		kvm_x86_ops->set_nmi(vcpu);
3634 		return;
3635 	}
3636 
3637 	if (vcpu->arch.interrupt.pending) {
3638 		kvm_x86_ops->set_irq(vcpu);
3639 		return;
3640 	}
3641 
3642 	/* try to inject new event if pending */
3643 	if (vcpu->arch.nmi_pending) {
3644 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
3645 			vcpu->arch.nmi_pending = false;
3646 			vcpu->arch.nmi_injected = true;
3647 			kvm_x86_ops->set_nmi(vcpu);
3648 		}
3649 	} else if (kvm_cpu_has_interrupt(vcpu)) {
3650 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3651 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3652 					    false);
3653 			kvm_x86_ops->set_irq(vcpu);
3654 		}
3655 	}
3656 }
3657 
3658 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3659 {
3660 	int r;
3661 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3662 		vcpu->run->request_interrupt_window;
3663 
3664 	if (vcpu->requests)
3665 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3666 			kvm_mmu_unload(vcpu);
3667 
3668 	r = kvm_mmu_reload(vcpu);
3669 	if (unlikely(r))
3670 		goto out;
3671 
3672 	if (vcpu->requests) {
3673 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3674 			__kvm_migrate_timers(vcpu);
3675 		if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3676 			kvm_write_guest_time(vcpu);
3677 		if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3678 			kvm_mmu_sync_roots(vcpu);
3679 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3680 			kvm_x86_ops->tlb_flush(vcpu);
3681 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3682 				       &vcpu->requests)) {
3683 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3684 			r = 0;
3685 			goto out;
3686 		}
3687 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3688 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3689 			r = 0;
3690 			goto out;
3691 		}
3692 	}
3693 
3694 	preempt_disable();
3695 
3696 	kvm_x86_ops->prepare_guest_switch(vcpu);
3697 	kvm_load_guest_fpu(vcpu);
3698 
3699 	local_irq_disable();
3700 
3701 	clear_bit(KVM_REQ_KICK, &vcpu->requests);
3702 	smp_mb__after_clear_bit();
3703 
3704 	if (vcpu->requests || need_resched() || signal_pending(current)) {
3705 		set_bit(KVM_REQ_KICK, &vcpu->requests);
3706 		local_irq_enable();
3707 		preempt_enable();
3708 		r = 1;
3709 		goto out;
3710 	}
3711 
3712 	inject_pending_event(vcpu);
3713 
3714 	/* enable NMI/IRQ window open exits if needed */
3715 	if (vcpu->arch.nmi_pending)
3716 		kvm_x86_ops->enable_nmi_window(vcpu);
3717 	else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3718 		kvm_x86_ops->enable_irq_window(vcpu);
3719 
3720 	if (kvm_lapic_enabled(vcpu)) {
3721 		update_cr8_intercept(vcpu);
3722 		kvm_lapic_sync_to_vapic(vcpu);
3723 	}
3724 
3725 	up_read(&vcpu->kvm->slots_lock);
3726 
3727 	kvm_guest_enter();
3728 
3729 	if (unlikely(vcpu->arch.switch_db_regs)) {
3730 		set_debugreg(0, 7);
3731 		set_debugreg(vcpu->arch.eff_db[0], 0);
3732 		set_debugreg(vcpu->arch.eff_db[1], 1);
3733 		set_debugreg(vcpu->arch.eff_db[2], 2);
3734 		set_debugreg(vcpu->arch.eff_db[3], 3);
3735 	}
3736 
3737 	trace_kvm_entry(vcpu->vcpu_id);
3738 	kvm_x86_ops->run(vcpu);
3739 
3740 	if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3741 		set_debugreg(current->thread.debugreg0, 0);
3742 		set_debugreg(current->thread.debugreg1, 1);
3743 		set_debugreg(current->thread.debugreg2, 2);
3744 		set_debugreg(current->thread.debugreg3, 3);
3745 		set_debugreg(current->thread.debugreg6, 6);
3746 		set_debugreg(current->thread.debugreg7, 7);
3747 	}
3748 
3749 	set_bit(KVM_REQ_KICK, &vcpu->requests);
3750 	local_irq_enable();
3751 
3752 	++vcpu->stat.exits;
3753 
3754 	/*
3755 	 * We must have an instruction between local_irq_enable() and
3756 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3757 	 * the interrupt shadow.  The stat.exits increment will do nicely.
3758 	 * But we need to prevent reordering, hence this barrier():
3759 	 */
3760 	barrier();
3761 
3762 	kvm_guest_exit();
3763 
3764 	preempt_enable();
3765 
3766 	down_read(&vcpu->kvm->slots_lock);
3767 
3768 	/*
3769 	 * Profile KVM exit RIPs:
3770 	 */
3771 	if (unlikely(prof_on == KVM_PROFILING)) {
3772 		unsigned long rip = kvm_rip_read(vcpu);
3773 		profile_hit(KVM_PROFILING, (void *)rip);
3774 	}
3775 
3776 
3777 	kvm_lapic_sync_from_vapic(vcpu);
3778 
3779 	r = kvm_x86_ops->handle_exit(vcpu);
3780 out:
3781 	return r;
3782 }
3783 
3784 
3785 static int __vcpu_run(struct kvm_vcpu *vcpu)
3786 {
3787 	int r;
3788 
3789 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3790 		pr_debug("vcpu %d received sipi with vector # %x\n",
3791 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3792 		kvm_lapic_reset(vcpu);
3793 		r = kvm_arch_vcpu_reset(vcpu);
3794 		if (r)
3795 			return r;
3796 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3797 	}
3798 
3799 	down_read(&vcpu->kvm->slots_lock);
3800 	vapic_enter(vcpu);
3801 
3802 	r = 1;
3803 	while (r > 0) {
3804 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3805 			r = vcpu_enter_guest(vcpu);
3806 		else {
3807 			up_read(&vcpu->kvm->slots_lock);
3808 			kvm_vcpu_block(vcpu);
3809 			down_read(&vcpu->kvm->slots_lock);
3810 			if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3811 			{
3812 				switch(vcpu->arch.mp_state) {
3813 				case KVM_MP_STATE_HALTED:
3814 					vcpu->arch.mp_state =
3815 						KVM_MP_STATE_RUNNABLE;
3816 				case KVM_MP_STATE_RUNNABLE:
3817 					break;
3818 				case KVM_MP_STATE_SIPI_RECEIVED:
3819 				default:
3820 					r = -EINTR;
3821 					break;
3822 				}
3823 			}
3824 		}
3825 
3826 		if (r <= 0)
3827 			break;
3828 
3829 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3830 		if (kvm_cpu_has_pending_timer(vcpu))
3831 			kvm_inject_pending_timer_irqs(vcpu);
3832 
3833 		if (dm_request_for_irq_injection(vcpu)) {
3834 			r = -EINTR;
3835 			vcpu->run->exit_reason = KVM_EXIT_INTR;
3836 			++vcpu->stat.request_irq_exits;
3837 		}
3838 		if (signal_pending(current)) {
3839 			r = -EINTR;
3840 			vcpu->run->exit_reason = KVM_EXIT_INTR;
3841 			++vcpu->stat.signal_exits;
3842 		}
3843 		if (need_resched()) {
3844 			up_read(&vcpu->kvm->slots_lock);
3845 			kvm_resched(vcpu);
3846 			down_read(&vcpu->kvm->slots_lock);
3847 		}
3848 	}
3849 
3850 	up_read(&vcpu->kvm->slots_lock);
3851 	post_kvm_run_save(vcpu);
3852 
3853 	vapic_exit(vcpu);
3854 
3855 	return r;
3856 }
3857 
3858 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3859 {
3860 	int r;
3861 	sigset_t sigsaved;
3862 
3863 	vcpu_load(vcpu);
3864 
3865 	if (vcpu->sigset_active)
3866 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3867 
3868 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3869 		kvm_vcpu_block(vcpu);
3870 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3871 		r = -EAGAIN;
3872 		goto out;
3873 	}
3874 
3875 	/* re-sync apic's tpr */
3876 	if (!irqchip_in_kernel(vcpu->kvm))
3877 		kvm_set_cr8(vcpu, kvm_run->cr8);
3878 
3879 	if (vcpu->arch.pio.cur_count) {
3880 		r = complete_pio(vcpu);
3881 		if (r)
3882 			goto out;
3883 	}
3884 	if (vcpu->mmio_needed) {
3885 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3886 		vcpu->mmio_read_completed = 1;
3887 		vcpu->mmio_needed = 0;
3888 
3889 		down_read(&vcpu->kvm->slots_lock);
3890 		r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
3891 					EMULTYPE_NO_DECODE);
3892 		up_read(&vcpu->kvm->slots_lock);
3893 		if (r == EMULATE_DO_MMIO) {
3894 			/*
3895 			 * Read-modify-write.  Back to userspace.
3896 			 */
3897 			r = 0;
3898 			goto out;
3899 		}
3900 	}
3901 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3902 		kvm_register_write(vcpu, VCPU_REGS_RAX,
3903 				     kvm_run->hypercall.ret);
3904 
3905 	r = __vcpu_run(vcpu);
3906 
3907 out:
3908 	if (vcpu->sigset_active)
3909 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3910 
3911 	vcpu_put(vcpu);
3912 	return r;
3913 }
3914 
3915 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3916 {
3917 	vcpu_load(vcpu);
3918 
3919 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3920 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3921 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3922 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3923 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3924 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3925 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3926 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3927 #ifdef CONFIG_X86_64
3928 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3929 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3930 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3931 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3932 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3933 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3934 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3935 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3936 #endif
3937 
3938 	regs->rip = kvm_rip_read(vcpu);
3939 	regs->rflags = kvm_get_rflags(vcpu);
3940 
3941 	vcpu_put(vcpu);
3942 
3943 	return 0;
3944 }
3945 
3946 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3947 {
3948 	vcpu_load(vcpu);
3949 
3950 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3951 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3952 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3953 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3954 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3955 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3956 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3957 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3958 #ifdef CONFIG_X86_64
3959 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3960 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3961 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3962 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3963 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3964 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3965 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3966 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3967 #endif
3968 
3969 	kvm_rip_write(vcpu, regs->rip);
3970 	kvm_set_rflags(vcpu, regs->rflags);
3971 
3972 	vcpu->arch.exception.pending = false;
3973 
3974 	vcpu_put(vcpu);
3975 
3976 	return 0;
3977 }
3978 
3979 void kvm_get_segment(struct kvm_vcpu *vcpu,
3980 		     struct kvm_segment *var, int seg)
3981 {
3982 	kvm_x86_ops->get_segment(vcpu, var, seg);
3983 }
3984 
3985 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3986 {
3987 	struct kvm_segment cs;
3988 
3989 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3990 	*db = cs.db;
3991 	*l = cs.l;
3992 }
3993 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3994 
3995 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3996 				  struct kvm_sregs *sregs)
3997 {
3998 	struct descriptor_table dt;
3999 
4000 	vcpu_load(vcpu);
4001 
4002 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4003 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4004 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4005 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4006 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4007 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4008 
4009 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4010 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4011 
4012 	kvm_x86_ops->get_idt(vcpu, &dt);
4013 	sregs->idt.limit = dt.limit;
4014 	sregs->idt.base = dt.base;
4015 	kvm_x86_ops->get_gdt(vcpu, &dt);
4016 	sregs->gdt.limit = dt.limit;
4017 	sregs->gdt.base = dt.base;
4018 
4019 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4020 	sregs->cr0 = vcpu->arch.cr0;
4021 	sregs->cr2 = vcpu->arch.cr2;
4022 	sregs->cr3 = vcpu->arch.cr3;
4023 	sregs->cr4 = vcpu->arch.cr4;
4024 	sregs->cr8 = kvm_get_cr8(vcpu);
4025 	sregs->efer = vcpu->arch.shadow_efer;
4026 	sregs->apic_base = kvm_get_apic_base(vcpu);
4027 
4028 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4029 
4030 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4031 		set_bit(vcpu->arch.interrupt.nr,
4032 			(unsigned long *)sregs->interrupt_bitmap);
4033 
4034 	vcpu_put(vcpu);
4035 
4036 	return 0;
4037 }
4038 
4039 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4040 				    struct kvm_mp_state *mp_state)
4041 {
4042 	vcpu_load(vcpu);
4043 	mp_state->mp_state = vcpu->arch.mp_state;
4044 	vcpu_put(vcpu);
4045 	return 0;
4046 }
4047 
4048 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4049 				    struct kvm_mp_state *mp_state)
4050 {
4051 	vcpu_load(vcpu);
4052 	vcpu->arch.mp_state = mp_state->mp_state;
4053 	vcpu_put(vcpu);
4054 	return 0;
4055 }
4056 
4057 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4058 			struct kvm_segment *var, int seg)
4059 {
4060 	kvm_x86_ops->set_segment(vcpu, var, seg);
4061 }
4062 
4063 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4064 				   struct kvm_segment *kvm_desct)
4065 {
4066 	kvm_desct->base = get_desc_base(seg_desc);
4067 	kvm_desct->limit = get_desc_limit(seg_desc);
4068 	if (seg_desc->g) {
4069 		kvm_desct->limit <<= 12;
4070 		kvm_desct->limit |= 0xfff;
4071 	}
4072 	kvm_desct->selector = selector;
4073 	kvm_desct->type = seg_desc->type;
4074 	kvm_desct->present = seg_desc->p;
4075 	kvm_desct->dpl = seg_desc->dpl;
4076 	kvm_desct->db = seg_desc->d;
4077 	kvm_desct->s = seg_desc->s;
4078 	kvm_desct->l = seg_desc->l;
4079 	kvm_desct->g = seg_desc->g;
4080 	kvm_desct->avl = seg_desc->avl;
4081 	if (!selector)
4082 		kvm_desct->unusable = 1;
4083 	else
4084 		kvm_desct->unusable = 0;
4085 	kvm_desct->padding = 0;
4086 }
4087 
4088 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4089 					  u16 selector,
4090 					  struct descriptor_table *dtable)
4091 {
4092 	if (selector & 1 << 2) {
4093 		struct kvm_segment kvm_seg;
4094 
4095 		kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4096 
4097 		if (kvm_seg.unusable)
4098 			dtable->limit = 0;
4099 		else
4100 			dtable->limit = kvm_seg.limit;
4101 		dtable->base = kvm_seg.base;
4102 	}
4103 	else
4104 		kvm_x86_ops->get_gdt(vcpu, dtable);
4105 }
4106 
4107 /* allowed just for 8 bytes segments */
4108 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4109 					 struct desc_struct *seg_desc)
4110 {
4111 	struct descriptor_table dtable;
4112 	u16 index = selector >> 3;
4113 
4114 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
4115 
4116 	if (dtable.limit < index * 8 + 7) {
4117 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4118 		return 1;
4119 	}
4120 	return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4121 }
4122 
4123 /* allowed just for 8 bytes segments */
4124 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4125 					 struct desc_struct *seg_desc)
4126 {
4127 	struct descriptor_table dtable;
4128 	u16 index = selector >> 3;
4129 
4130 	get_segment_descriptor_dtable(vcpu, selector, &dtable);
4131 
4132 	if (dtable.limit < index * 8 + 7)
4133 		return 1;
4134 	return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4135 }
4136 
4137 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4138 			     struct desc_struct *seg_desc)
4139 {
4140 	u32 base_addr = get_desc_base(seg_desc);
4141 
4142 	return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4143 }
4144 
4145 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4146 {
4147 	struct kvm_segment kvm_seg;
4148 
4149 	kvm_get_segment(vcpu, &kvm_seg, seg);
4150 	return kvm_seg.selector;
4151 }
4152 
4153 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4154 						u16 selector,
4155 						struct kvm_segment *kvm_seg)
4156 {
4157 	struct desc_struct seg_desc;
4158 
4159 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4160 		return 1;
4161 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4162 	return 0;
4163 }
4164 
4165 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4166 {
4167 	struct kvm_segment segvar = {
4168 		.base = selector << 4,
4169 		.limit = 0xffff,
4170 		.selector = selector,
4171 		.type = 3,
4172 		.present = 1,
4173 		.dpl = 3,
4174 		.db = 0,
4175 		.s = 1,
4176 		.l = 0,
4177 		.g = 0,
4178 		.avl = 0,
4179 		.unusable = 0,
4180 	};
4181 	kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4182 	return 0;
4183 }
4184 
4185 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4186 {
4187 	return (seg != VCPU_SREG_LDTR) &&
4188 		(seg != VCPU_SREG_TR) &&
4189 		(kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4190 }
4191 
4192 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4193 				int type_bits, int seg)
4194 {
4195 	struct kvm_segment kvm_seg;
4196 
4197 	if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4198 		return kvm_load_realmode_segment(vcpu, selector, seg);
4199 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4200 		return 1;
4201 	kvm_seg.type |= type_bits;
4202 
4203 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4204 	    seg != VCPU_SREG_LDTR)
4205 		if (!kvm_seg.s)
4206 			kvm_seg.unusable = 1;
4207 
4208 	kvm_set_segment(vcpu, &kvm_seg, seg);
4209 	return 0;
4210 }
4211 
4212 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4213 				struct tss_segment_32 *tss)
4214 {
4215 	tss->cr3 = vcpu->arch.cr3;
4216 	tss->eip = kvm_rip_read(vcpu);
4217 	tss->eflags = kvm_get_rflags(vcpu);
4218 	tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4219 	tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4220 	tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4221 	tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4222 	tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4223 	tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4224 	tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4225 	tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4226 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4227 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4228 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4229 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4230 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4231 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4232 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4233 }
4234 
4235 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4236 				  struct tss_segment_32 *tss)
4237 {
4238 	kvm_set_cr3(vcpu, tss->cr3);
4239 
4240 	kvm_rip_write(vcpu, tss->eip);
4241 	kvm_set_rflags(vcpu, tss->eflags | 2);
4242 
4243 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4244 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4245 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4246 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4247 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4248 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4249 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4250 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4251 
4252 	if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4253 		return 1;
4254 
4255 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4256 		return 1;
4257 
4258 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4259 		return 1;
4260 
4261 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4262 		return 1;
4263 
4264 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4265 		return 1;
4266 
4267 	if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4268 		return 1;
4269 
4270 	if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4271 		return 1;
4272 	return 0;
4273 }
4274 
4275 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4276 				struct tss_segment_16 *tss)
4277 {
4278 	tss->ip = kvm_rip_read(vcpu);
4279 	tss->flag = kvm_get_rflags(vcpu);
4280 	tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4281 	tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4282 	tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4283 	tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4284 	tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4285 	tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4286 	tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4287 	tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4288 
4289 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4290 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4291 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4292 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4293 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4294 }
4295 
4296 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4297 				 struct tss_segment_16 *tss)
4298 {
4299 	kvm_rip_write(vcpu, tss->ip);
4300 	kvm_set_rflags(vcpu, tss->flag | 2);
4301 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4302 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4303 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4304 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4305 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4306 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4307 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4308 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4309 
4310 	if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4311 		return 1;
4312 
4313 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4314 		return 1;
4315 
4316 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4317 		return 1;
4318 
4319 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4320 		return 1;
4321 
4322 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4323 		return 1;
4324 	return 0;
4325 }
4326 
4327 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4328 			      u16 old_tss_sel, u32 old_tss_base,
4329 			      struct desc_struct *nseg_desc)
4330 {
4331 	struct tss_segment_16 tss_segment_16;
4332 	int ret = 0;
4333 
4334 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4335 			   sizeof tss_segment_16))
4336 		goto out;
4337 
4338 	save_state_to_tss16(vcpu, &tss_segment_16);
4339 
4340 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4341 			    sizeof tss_segment_16))
4342 		goto out;
4343 
4344 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4345 			   &tss_segment_16, sizeof tss_segment_16))
4346 		goto out;
4347 
4348 	if (old_tss_sel != 0xffff) {
4349 		tss_segment_16.prev_task_link = old_tss_sel;
4350 
4351 		if (kvm_write_guest(vcpu->kvm,
4352 				    get_tss_base_addr(vcpu, nseg_desc),
4353 				    &tss_segment_16.prev_task_link,
4354 				    sizeof tss_segment_16.prev_task_link))
4355 			goto out;
4356 	}
4357 
4358 	if (load_state_from_tss16(vcpu, &tss_segment_16))
4359 		goto out;
4360 
4361 	ret = 1;
4362 out:
4363 	return ret;
4364 }
4365 
4366 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4367 		       u16 old_tss_sel, u32 old_tss_base,
4368 		       struct desc_struct *nseg_desc)
4369 {
4370 	struct tss_segment_32 tss_segment_32;
4371 	int ret = 0;
4372 
4373 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4374 			   sizeof tss_segment_32))
4375 		goto out;
4376 
4377 	save_state_to_tss32(vcpu, &tss_segment_32);
4378 
4379 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4380 			    sizeof tss_segment_32))
4381 		goto out;
4382 
4383 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4384 			   &tss_segment_32, sizeof tss_segment_32))
4385 		goto out;
4386 
4387 	if (old_tss_sel != 0xffff) {
4388 		tss_segment_32.prev_task_link = old_tss_sel;
4389 
4390 		if (kvm_write_guest(vcpu->kvm,
4391 				    get_tss_base_addr(vcpu, nseg_desc),
4392 				    &tss_segment_32.prev_task_link,
4393 				    sizeof tss_segment_32.prev_task_link))
4394 			goto out;
4395 	}
4396 
4397 	if (load_state_from_tss32(vcpu, &tss_segment_32))
4398 		goto out;
4399 
4400 	ret = 1;
4401 out:
4402 	return ret;
4403 }
4404 
4405 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4406 {
4407 	struct kvm_segment tr_seg;
4408 	struct desc_struct cseg_desc;
4409 	struct desc_struct nseg_desc;
4410 	int ret = 0;
4411 	u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4412 	u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4413 
4414 	old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4415 
4416 	/* FIXME: Handle errors. Failure to read either TSS or their
4417 	 * descriptors should generate a pagefault.
4418 	 */
4419 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4420 		goto out;
4421 
4422 	if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4423 		goto out;
4424 
4425 	if (reason != TASK_SWITCH_IRET) {
4426 		int cpl;
4427 
4428 		cpl = kvm_x86_ops->get_cpl(vcpu);
4429 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4430 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4431 			return 1;
4432 		}
4433 	}
4434 
4435 	if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4436 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4437 		return 1;
4438 	}
4439 
4440 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4441 		cseg_desc.type &= ~(1 << 1); //clear the B flag
4442 		save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4443 	}
4444 
4445 	if (reason == TASK_SWITCH_IRET) {
4446 		u32 eflags = kvm_get_rflags(vcpu);
4447 		kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4448 	}
4449 
4450 	/* set back link to prev task only if NT bit is set in eflags
4451 	   note that old_tss_sel is not used afetr this point */
4452 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4453 		old_tss_sel = 0xffff;
4454 
4455 	/* set back link to prev task only if NT bit is set in eflags
4456 	   note that old_tss_sel is not used afetr this point */
4457 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4458 		old_tss_sel = 0xffff;
4459 
4460 	if (nseg_desc.type & 8)
4461 		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4462 					 old_tss_base, &nseg_desc);
4463 	else
4464 		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4465 					 old_tss_base, &nseg_desc);
4466 
4467 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4468 		u32 eflags = kvm_get_rflags(vcpu);
4469 		kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4470 	}
4471 
4472 	if (reason != TASK_SWITCH_IRET) {
4473 		nseg_desc.type |= (1 << 1);
4474 		save_guest_segment_descriptor(vcpu, tss_selector,
4475 					      &nseg_desc);
4476 	}
4477 
4478 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4479 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4480 	tr_seg.type = 11;
4481 	kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4482 out:
4483 	return ret;
4484 }
4485 EXPORT_SYMBOL_GPL(kvm_task_switch);
4486 
4487 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4488 				  struct kvm_sregs *sregs)
4489 {
4490 	int mmu_reset_needed = 0;
4491 	int pending_vec, max_bits;
4492 	struct descriptor_table dt;
4493 
4494 	vcpu_load(vcpu);
4495 
4496 	dt.limit = sregs->idt.limit;
4497 	dt.base = sregs->idt.base;
4498 	kvm_x86_ops->set_idt(vcpu, &dt);
4499 	dt.limit = sregs->gdt.limit;
4500 	dt.base = sregs->gdt.base;
4501 	kvm_x86_ops->set_gdt(vcpu, &dt);
4502 
4503 	vcpu->arch.cr2 = sregs->cr2;
4504 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4505 	vcpu->arch.cr3 = sregs->cr3;
4506 
4507 	kvm_set_cr8(vcpu, sregs->cr8);
4508 
4509 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4510 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
4511 	kvm_set_apic_base(vcpu, sregs->apic_base);
4512 
4513 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4514 
4515 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4516 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4517 	vcpu->arch.cr0 = sregs->cr0;
4518 
4519 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4520 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4521 	if (!is_long_mode(vcpu) && is_pae(vcpu))
4522 		load_pdptrs(vcpu, vcpu->arch.cr3);
4523 
4524 	if (mmu_reset_needed)
4525 		kvm_mmu_reset_context(vcpu);
4526 
4527 	max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4528 	pending_vec = find_first_bit(
4529 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
4530 	if (pending_vec < max_bits) {
4531 		kvm_queue_interrupt(vcpu, pending_vec, false);
4532 		pr_debug("Set back pending irq %d\n", pending_vec);
4533 		if (irqchip_in_kernel(vcpu->kvm))
4534 			kvm_pic_clear_isr_ack(vcpu->kvm);
4535 	}
4536 
4537 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4538 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4539 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4540 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4541 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4542 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4543 
4544 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4545 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4546 
4547 	update_cr8_intercept(vcpu);
4548 
4549 	/* Older userspace won't unhalt the vcpu on reset. */
4550 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4551 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4552 	    !(vcpu->arch.cr0 & X86_CR0_PE))
4553 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4554 
4555 	vcpu_put(vcpu);
4556 
4557 	return 0;
4558 }
4559 
4560 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4561 					struct kvm_guest_debug *dbg)
4562 {
4563 	unsigned long rflags;
4564 	int i;
4565 
4566 	vcpu_load(vcpu);
4567 
4568 	/*
4569 	 * Read rflags as long as potentially injected trace flags are still
4570 	 * filtered out.
4571 	 */
4572 	rflags = kvm_get_rflags(vcpu);
4573 
4574 	vcpu->guest_debug = dbg->control;
4575 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4576 		vcpu->guest_debug = 0;
4577 
4578 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4579 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
4580 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4581 		vcpu->arch.switch_db_regs =
4582 			(dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4583 	} else {
4584 		for (i = 0; i < KVM_NR_DB_REGS; i++)
4585 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4586 		vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4587 	}
4588 
4589 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4590 		vcpu->arch.singlestep_cs =
4591 			get_segment_selector(vcpu, VCPU_SREG_CS);
4592 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4593 	}
4594 
4595 	/*
4596 	 * Trigger an rflags update that will inject or remove the trace
4597 	 * flags.
4598 	 */
4599 	kvm_set_rflags(vcpu, rflags);
4600 
4601 	kvm_x86_ops->set_guest_debug(vcpu, dbg);
4602 
4603 	if (vcpu->guest_debug & KVM_GUESTDBG_INJECT_DB)
4604 		kvm_queue_exception(vcpu, DB_VECTOR);
4605 	else if (vcpu->guest_debug & KVM_GUESTDBG_INJECT_BP)
4606 		kvm_queue_exception(vcpu, BP_VECTOR);
4607 
4608 	vcpu_put(vcpu);
4609 
4610 	return 0;
4611 }
4612 
4613 /*
4614  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4615  * we have asm/x86/processor.h
4616  */
4617 struct fxsave {
4618 	u16	cwd;
4619 	u16	swd;
4620 	u16	twd;
4621 	u16	fop;
4622 	u64	rip;
4623 	u64	rdp;
4624 	u32	mxcsr;
4625 	u32	mxcsr_mask;
4626 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
4627 #ifdef CONFIG_X86_64
4628 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
4629 #else
4630 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
4631 #endif
4632 };
4633 
4634 /*
4635  * Translate a guest virtual address to a guest physical address.
4636  */
4637 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4638 				    struct kvm_translation *tr)
4639 {
4640 	unsigned long vaddr = tr->linear_address;
4641 	gpa_t gpa;
4642 
4643 	vcpu_load(vcpu);
4644 	down_read(&vcpu->kvm->slots_lock);
4645 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4646 	up_read(&vcpu->kvm->slots_lock);
4647 	tr->physical_address = gpa;
4648 	tr->valid = gpa != UNMAPPED_GVA;
4649 	tr->writeable = 1;
4650 	tr->usermode = 0;
4651 	vcpu_put(vcpu);
4652 
4653 	return 0;
4654 }
4655 
4656 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4657 {
4658 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4659 
4660 	vcpu_load(vcpu);
4661 
4662 	memcpy(fpu->fpr, fxsave->st_space, 128);
4663 	fpu->fcw = fxsave->cwd;
4664 	fpu->fsw = fxsave->swd;
4665 	fpu->ftwx = fxsave->twd;
4666 	fpu->last_opcode = fxsave->fop;
4667 	fpu->last_ip = fxsave->rip;
4668 	fpu->last_dp = fxsave->rdp;
4669 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4670 
4671 	vcpu_put(vcpu);
4672 
4673 	return 0;
4674 }
4675 
4676 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4677 {
4678 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4679 
4680 	vcpu_load(vcpu);
4681 
4682 	memcpy(fxsave->st_space, fpu->fpr, 128);
4683 	fxsave->cwd = fpu->fcw;
4684 	fxsave->swd = fpu->fsw;
4685 	fxsave->twd = fpu->ftwx;
4686 	fxsave->fop = fpu->last_opcode;
4687 	fxsave->rip = fpu->last_ip;
4688 	fxsave->rdp = fpu->last_dp;
4689 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4690 
4691 	vcpu_put(vcpu);
4692 
4693 	return 0;
4694 }
4695 
4696 void fx_init(struct kvm_vcpu *vcpu)
4697 {
4698 	unsigned after_mxcsr_mask;
4699 
4700 	/*
4701 	 * Touch the fpu the first time in non atomic context as if
4702 	 * this is the first fpu instruction the exception handler
4703 	 * will fire before the instruction returns and it'll have to
4704 	 * allocate ram with GFP_KERNEL.
4705 	 */
4706 	if (!used_math())
4707 		kvm_fx_save(&vcpu->arch.host_fx_image);
4708 
4709 	/* Initialize guest FPU by resetting ours and saving into guest's */
4710 	preempt_disable();
4711 	kvm_fx_save(&vcpu->arch.host_fx_image);
4712 	kvm_fx_finit();
4713 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4714 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4715 	preempt_enable();
4716 
4717 	vcpu->arch.cr0 |= X86_CR0_ET;
4718 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4719 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4720 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4721 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4722 }
4723 EXPORT_SYMBOL_GPL(fx_init);
4724 
4725 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4726 {
4727 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4728 		return;
4729 
4730 	vcpu->guest_fpu_loaded = 1;
4731 	kvm_fx_save(&vcpu->arch.host_fx_image);
4732 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
4733 }
4734 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4735 
4736 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4737 {
4738 	if (!vcpu->guest_fpu_loaded)
4739 		return;
4740 
4741 	vcpu->guest_fpu_loaded = 0;
4742 	kvm_fx_save(&vcpu->arch.guest_fx_image);
4743 	kvm_fx_restore(&vcpu->arch.host_fx_image);
4744 	++vcpu->stat.fpu_reload;
4745 }
4746 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4747 
4748 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4749 {
4750 	if (vcpu->arch.time_page) {
4751 		kvm_release_page_dirty(vcpu->arch.time_page);
4752 		vcpu->arch.time_page = NULL;
4753 	}
4754 
4755 	kvm_x86_ops->vcpu_free(vcpu);
4756 }
4757 
4758 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4759 						unsigned int id)
4760 {
4761 	return kvm_x86_ops->vcpu_create(kvm, id);
4762 }
4763 
4764 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4765 {
4766 	int r;
4767 
4768 	/* We do fxsave: this must be aligned. */
4769 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4770 
4771 	vcpu->arch.mtrr_state.have_fixed = 1;
4772 	vcpu_load(vcpu);
4773 	r = kvm_arch_vcpu_reset(vcpu);
4774 	if (r == 0)
4775 		r = kvm_mmu_setup(vcpu);
4776 	vcpu_put(vcpu);
4777 	if (r < 0)
4778 		goto free_vcpu;
4779 
4780 	return 0;
4781 free_vcpu:
4782 	kvm_x86_ops->vcpu_free(vcpu);
4783 	return r;
4784 }
4785 
4786 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4787 {
4788 	vcpu_load(vcpu);
4789 	kvm_mmu_unload(vcpu);
4790 	vcpu_put(vcpu);
4791 
4792 	kvm_x86_ops->vcpu_free(vcpu);
4793 }
4794 
4795 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4796 {
4797 	vcpu->arch.nmi_pending = false;
4798 	vcpu->arch.nmi_injected = false;
4799 
4800 	vcpu->arch.switch_db_regs = 0;
4801 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4802 	vcpu->arch.dr6 = DR6_FIXED_1;
4803 	vcpu->arch.dr7 = DR7_FIXED_1;
4804 
4805 	return kvm_x86_ops->vcpu_reset(vcpu);
4806 }
4807 
4808 int kvm_arch_hardware_enable(void *garbage)
4809 {
4810 	/*
4811 	 * Since this may be called from a hotplug notifcation,
4812 	 * we can't get the CPU frequency directly.
4813 	 */
4814 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4815 		int cpu = raw_smp_processor_id();
4816 		per_cpu(cpu_tsc_khz, cpu) = 0;
4817 	}
4818 	return kvm_x86_ops->hardware_enable(garbage);
4819 }
4820 
4821 void kvm_arch_hardware_disable(void *garbage)
4822 {
4823 	kvm_x86_ops->hardware_disable(garbage);
4824 }
4825 
4826 int kvm_arch_hardware_setup(void)
4827 {
4828 	return kvm_x86_ops->hardware_setup();
4829 }
4830 
4831 void kvm_arch_hardware_unsetup(void)
4832 {
4833 	kvm_x86_ops->hardware_unsetup();
4834 }
4835 
4836 void kvm_arch_check_processor_compat(void *rtn)
4837 {
4838 	kvm_x86_ops->check_processor_compatibility(rtn);
4839 }
4840 
4841 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4842 {
4843 	struct page *page;
4844 	struct kvm *kvm;
4845 	int r;
4846 
4847 	BUG_ON(vcpu->kvm == NULL);
4848 	kvm = vcpu->kvm;
4849 
4850 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4851 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4852 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4853 	else
4854 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4855 
4856 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4857 	if (!page) {
4858 		r = -ENOMEM;
4859 		goto fail;
4860 	}
4861 	vcpu->arch.pio_data = page_address(page);
4862 
4863 	r = kvm_mmu_create(vcpu);
4864 	if (r < 0)
4865 		goto fail_free_pio_data;
4866 
4867 	if (irqchip_in_kernel(kvm)) {
4868 		r = kvm_create_lapic(vcpu);
4869 		if (r < 0)
4870 			goto fail_mmu_destroy;
4871 	}
4872 
4873 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4874 				       GFP_KERNEL);
4875 	if (!vcpu->arch.mce_banks) {
4876 		r = -ENOMEM;
4877 		goto fail_mmu_destroy;
4878 	}
4879 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4880 
4881 	return 0;
4882 
4883 fail_mmu_destroy:
4884 	kvm_mmu_destroy(vcpu);
4885 fail_free_pio_data:
4886 	free_page((unsigned long)vcpu->arch.pio_data);
4887 fail:
4888 	return r;
4889 }
4890 
4891 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4892 {
4893 	kvm_free_lapic(vcpu);
4894 	down_read(&vcpu->kvm->slots_lock);
4895 	kvm_mmu_destroy(vcpu);
4896 	up_read(&vcpu->kvm->slots_lock);
4897 	free_page((unsigned long)vcpu->arch.pio_data);
4898 }
4899 
4900 struct  kvm *kvm_arch_create_vm(void)
4901 {
4902 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4903 
4904 	if (!kvm)
4905 		return ERR_PTR(-ENOMEM);
4906 
4907 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4908 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4909 
4910 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4911 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4912 
4913 	rdtscll(kvm->arch.vm_init_tsc);
4914 
4915 	return kvm;
4916 }
4917 
4918 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4919 {
4920 	vcpu_load(vcpu);
4921 	kvm_mmu_unload(vcpu);
4922 	vcpu_put(vcpu);
4923 }
4924 
4925 static void kvm_free_vcpus(struct kvm *kvm)
4926 {
4927 	unsigned int i;
4928 	struct kvm_vcpu *vcpu;
4929 
4930 	/*
4931 	 * Unpin any mmu pages first.
4932 	 */
4933 	kvm_for_each_vcpu(i, vcpu, kvm)
4934 		kvm_unload_vcpu_mmu(vcpu);
4935 	kvm_for_each_vcpu(i, vcpu, kvm)
4936 		kvm_arch_vcpu_free(vcpu);
4937 
4938 	mutex_lock(&kvm->lock);
4939 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4940 		kvm->vcpus[i] = NULL;
4941 
4942 	atomic_set(&kvm->online_vcpus, 0);
4943 	mutex_unlock(&kvm->lock);
4944 }
4945 
4946 void kvm_arch_sync_events(struct kvm *kvm)
4947 {
4948 	kvm_free_all_assigned_devices(kvm);
4949 }
4950 
4951 void kvm_arch_destroy_vm(struct kvm *kvm)
4952 {
4953 	kvm_iommu_unmap_guest(kvm);
4954 	kvm_free_pit(kvm);
4955 	kfree(kvm->arch.vpic);
4956 	kfree(kvm->arch.vioapic);
4957 	kvm_free_vcpus(kvm);
4958 	kvm_free_physmem(kvm);
4959 	if (kvm->arch.apic_access_page)
4960 		put_page(kvm->arch.apic_access_page);
4961 	if (kvm->arch.ept_identity_pagetable)
4962 		put_page(kvm->arch.ept_identity_pagetable);
4963 	kfree(kvm);
4964 }
4965 
4966 int kvm_arch_set_memory_region(struct kvm *kvm,
4967 				struct kvm_userspace_memory_region *mem,
4968 				struct kvm_memory_slot old,
4969 				int user_alloc)
4970 {
4971 	int npages = mem->memory_size >> PAGE_SHIFT;
4972 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4973 
4974 	/*To keep backward compatibility with older userspace,
4975 	 *x86 needs to hanlde !user_alloc case.
4976 	 */
4977 	if (!user_alloc) {
4978 		if (npages && !old.rmap) {
4979 			unsigned long userspace_addr;
4980 
4981 			down_write(&current->mm->mmap_sem);
4982 			userspace_addr = do_mmap(NULL, 0,
4983 						 npages * PAGE_SIZE,
4984 						 PROT_READ | PROT_WRITE,
4985 						 MAP_PRIVATE | MAP_ANONYMOUS,
4986 						 0);
4987 			up_write(&current->mm->mmap_sem);
4988 
4989 			if (IS_ERR((void *)userspace_addr))
4990 				return PTR_ERR((void *)userspace_addr);
4991 
4992 			/* set userspace_addr atomically for kvm_hva_to_rmapp */
4993 			spin_lock(&kvm->mmu_lock);
4994 			memslot->userspace_addr = userspace_addr;
4995 			spin_unlock(&kvm->mmu_lock);
4996 		} else {
4997 			if (!old.user_alloc && old.rmap) {
4998 				int ret;
4999 
5000 				down_write(&current->mm->mmap_sem);
5001 				ret = do_munmap(current->mm, old.userspace_addr,
5002 						old.npages * PAGE_SIZE);
5003 				up_write(&current->mm->mmap_sem);
5004 				if (ret < 0)
5005 					printk(KERN_WARNING
5006 				       "kvm_vm_ioctl_set_memory_region: "
5007 				       "failed to munmap memory\n");
5008 			}
5009 		}
5010 	}
5011 
5012 	spin_lock(&kvm->mmu_lock);
5013 	if (!kvm->arch.n_requested_mmu_pages) {
5014 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5015 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5016 	}
5017 
5018 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5019 	spin_unlock(&kvm->mmu_lock);
5020 
5021 	return 0;
5022 }
5023 
5024 void kvm_arch_flush_shadow(struct kvm *kvm)
5025 {
5026 	kvm_mmu_zap_all(kvm);
5027 	kvm_reload_remote_mmus(kvm);
5028 }
5029 
5030 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5031 {
5032 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5033 		|| vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5034 		|| vcpu->arch.nmi_pending ||
5035 		(kvm_arch_interrupt_allowed(vcpu) &&
5036 		 kvm_cpu_has_interrupt(vcpu));
5037 }
5038 
5039 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5040 {
5041 	int me;
5042 	int cpu = vcpu->cpu;
5043 
5044 	if (waitqueue_active(&vcpu->wq)) {
5045 		wake_up_interruptible(&vcpu->wq);
5046 		++vcpu->stat.halt_wakeup;
5047 	}
5048 
5049 	me = get_cpu();
5050 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5051 		if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5052 			smp_send_reschedule(cpu);
5053 	put_cpu();
5054 }
5055 
5056 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5057 {
5058 	return kvm_x86_ops->interrupt_allowed(vcpu);
5059 }
5060 
5061 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5062 {
5063 	unsigned long rflags;
5064 
5065 	rflags = kvm_x86_ops->get_rflags(vcpu);
5066 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5067 		rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5068 	return rflags;
5069 }
5070 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5071 
5072 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5073 {
5074 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5075 	    vcpu->arch.singlestep_cs ==
5076 			get_segment_selector(vcpu, VCPU_SREG_CS) &&
5077 	    vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5078 		rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5079 	kvm_x86_ops->set_rflags(vcpu, rflags);
5080 }
5081 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5082 
5083 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5084 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5085 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5086 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5087 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5088 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5089 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5090 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5091 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5092 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5093 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5094