xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
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/intel-iommu.h>
38 
39 #include <asm/uaccess.h>
40 #include <asm/msr.h>
41 #include <asm/desc.h>
42 
43 #define MAX_IO_MSRS 256
44 #define CR0_RESERVED_BITS						\
45 	(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
46 			  | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
47 			  | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
48 #define CR4_RESERVED_BITS						\
49 	(~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
50 			  | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE	\
51 			  | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR	\
52 			  | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
53 
54 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
55 /* EFER defaults:
56  * - enable syscall per default because its emulated by KVM
57  * - enable LME and LMA per default on 64 bit KVM
58  */
59 #ifdef CONFIG_X86_64
60 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
61 #else
62 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
63 #endif
64 
65 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
66 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
67 
68 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
69 				    struct kvm_cpuid_entry2 __user *entries);
70 
71 struct kvm_x86_ops *kvm_x86_ops;
72 EXPORT_SYMBOL_GPL(kvm_x86_ops);
73 
74 struct kvm_stats_debugfs_item debugfs_entries[] = {
75 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
76 	{ "pf_guest", VCPU_STAT(pf_guest) },
77 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
78 	{ "invlpg", VCPU_STAT(invlpg) },
79 	{ "exits", VCPU_STAT(exits) },
80 	{ "io_exits", VCPU_STAT(io_exits) },
81 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
82 	{ "signal_exits", VCPU_STAT(signal_exits) },
83 	{ "irq_window", VCPU_STAT(irq_window_exits) },
84 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
85 	{ "halt_exits", VCPU_STAT(halt_exits) },
86 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
87 	{ "hypercalls", VCPU_STAT(hypercalls) },
88 	{ "request_irq", VCPU_STAT(request_irq_exits) },
89 	{ "irq_exits", VCPU_STAT(irq_exits) },
90 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
91 	{ "efer_reload", VCPU_STAT(efer_reload) },
92 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
93 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
94 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
95 	{ "irq_injections", VCPU_STAT(irq_injections) },
96 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
97 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
98 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
99 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
100 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
101 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
102 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
103 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
104 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
105 	{ "largepages", VM_STAT(lpages) },
106 	{ NULL }
107 };
108 
109 unsigned long segment_base(u16 selector)
110 {
111 	struct descriptor_table gdt;
112 	struct desc_struct *d;
113 	unsigned long table_base;
114 	unsigned long v;
115 
116 	if (selector == 0)
117 		return 0;
118 
119 	asm("sgdt %0" : "=m"(gdt));
120 	table_base = gdt.base;
121 
122 	if (selector & 4) {           /* from ldt */
123 		u16 ldt_selector;
124 
125 		asm("sldt %0" : "=g"(ldt_selector));
126 		table_base = segment_base(ldt_selector);
127 	}
128 	d = (struct desc_struct *)(table_base + (selector & ~7));
129 	v = d->base0 | ((unsigned long)d->base1 << 16) |
130 		((unsigned long)d->base2 << 24);
131 #ifdef CONFIG_X86_64
132 	if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
133 		v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
134 #endif
135 	return v;
136 }
137 EXPORT_SYMBOL_GPL(segment_base);
138 
139 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
140 {
141 	if (irqchip_in_kernel(vcpu->kvm))
142 		return vcpu->arch.apic_base;
143 	else
144 		return vcpu->arch.apic_base;
145 }
146 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
147 
148 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
149 {
150 	/* TODO: reserve bits check */
151 	if (irqchip_in_kernel(vcpu->kvm))
152 		kvm_lapic_set_base(vcpu, data);
153 	else
154 		vcpu->arch.apic_base = data;
155 }
156 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
157 
158 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
159 {
160 	WARN_ON(vcpu->arch.exception.pending);
161 	vcpu->arch.exception.pending = true;
162 	vcpu->arch.exception.has_error_code = false;
163 	vcpu->arch.exception.nr = nr;
164 }
165 EXPORT_SYMBOL_GPL(kvm_queue_exception);
166 
167 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
168 			   u32 error_code)
169 {
170 	++vcpu->stat.pf_guest;
171 	if (vcpu->arch.exception.pending) {
172 		if (vcpu->arch.exception.nr == PF_VECTOR) {
173 			printk(KERN_DEBUG "kvm: inject_page_fault:"
174 					" double fault 0x%lx\n", addr);
175 			vcpu->arch.exception.nr = DF_VECTOR;
176 			vcpu->arch.exception.error_code = 0;
177 		} else if (vcpu->arch.exception.nr == DF_VECTOR) {
178 			/* triple fault -> shutdown */
179 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
180 		}
181 		return;
182 	}
183 	vcpu->arch.cr2 = addr;
184 	kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
185 }
186 
187 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
188 {
189 	vcpu->arch.nmi_pending = 1;
190 }
191 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
192 
193 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
194 {
195 	WARN_ON(vcpu->arch.exception.pending);
196 	vcpu->arch.exception.pending = true;
197 	vcpu->arch.exception.has_error_code = true;
198 	vcpu->arch.exception.nr = nr;
199 	vcpu->arch.exception.error_code = error_code;
200 }
201 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
202 
203 static void __queue_exception(struct kvm_vcpu *vcpu)
204 {
205 	kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
206 				     vcpu->arch.exception.has_error_code,
207 				     vcpu->arch.exception.error_code);
208 }
209 
210 /*
211  * Load the pae pdptrs.  Return true is they are all valid.
212  */
213 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
214 {
215 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
216 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
217 	int i;
218 	int ret;
219 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
220 
221 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
222 				  offset * sizeof(u64), sizeof(pdpte));
223 	if (ret < 0) {
224 		ret = 0;
225 		goto out;
226 	}
227 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
228 		if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
229 			ret = 0;
230 			goto out;
231 		}
232 	}
233 	ret = 1;
234 
235 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
236 out:
237 
238 	return ret;
239 }
240 EXPORT_SYMBOL_GPL(load_pdptrs);
241 
242 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
243 {
244 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
245 	bool changed = true;
246 	int r;
247 
248 	if (is_long_mode(vcpu) || !is_pae(vcpu))
249 		return false;
250 
251 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
252 	if (r < 0)
253 		goto out;
254 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
255 out:
256 
257 	return changed;
258 }
259 
260 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
261 {
262 	if (cr0 & CR0_RESERVED_BITS) {
263 		printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
264 		       cr0, vcpu->arch.cr0);
265 		kvm_inject_gp(vcpu, 0);
266 		return;
267 	}
268 
269 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
270 		printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
271 		kvm_inject_gp(vcpu, 0);
272 		return;
273 	}
274 
275 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
276 		printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
277 		       "and a clear PE flag\n");
278 		kvm_inject_gp(vcpu, 0);
279 		return;
280 	}
281 
282 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
283 #ifdef CONFIG_X86_64
284 		if ((vcpu->arch.shadow_efer & EFER_LME)) {
285 			int cs_db, cs_l;
286 
287 			if (!is_pae(vcpu)) {
288 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
289 				       "in long mode while PAE is disabled\n");
290 				kvm_inject_gp(vcpu, 0);
291 				return;
292 			}
293 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
294 			if (cs_l) {
295 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
296 				       "in long mode while CS.L == 1\n");
297 				kvm_inject_gp(vcpu, 0);
298 				return;
299 
300 			}
301 		} else
302 #endif
303 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
304 			printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
305 			       "reserved bits\n");
306 			kvm_inject_gp(vcpu, 0);
307 			return;
308 		}
309 
310 	}
311 
312 	kvm_x86_ops->set_cr0(vcpu, cr0);
313 	vcpu->arch.cr0 = cr0;
314 
315 	kvm_mmu_reset_context(vcpu);
316 	return;
317 }
318 EXPORT_SYMBOL_GPL(kvm_set_cr0);
319 
320 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
321 {
322 	kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
323 	KVMTRACE_1D(LMSW, vcpu,
324 		    (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
325 		    handler);
326 }
327 EXPORT_SYMBOL_GPL(kvm_lmsw);
328 
329 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
330 {
331 	if (cr4 & CR4_RESERVED_BITS) {
332 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
333 		kvm_inject_gp(vcpu, 0);
334 		return;
335 	}
336 
337 	if (is_long_mode(vcpu)) {
338 		if (!(cr4 & X86_CR4_PAE)) {
339 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
340 			       "in long mode\n");
341 			kvm_inject_gp(vcpu, 0);
342 			return;
343 		}
344 	} else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
345 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
346 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
347 		kvm_inject_gp(vcpu, 0);
348 		return;
349 	}
350 
351 	if (cr4 & X86_CR4_VMXE) {
352 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
353 		kvm_inject_gp(vcpu, 0);
354 		return;
355 	}
356 	kvm_x86_ops->set_cr4(vcpu, cr4);
357 	vcpu->arch.cr4 = cr4;
358 	kvm_mmu_reset_context(vcpu);
359 }
360 EXPORT_SYMBOL_GPL(kvm_set_cr4);
361 
362 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
363 {
364 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
365 		kvm_mmu_sync_roots(vcpu);
366 		kvm_mmu_flush_tlb(vcpu);
367 		return;
368 	}
369 
370 	if (is_long_mode(vcpu)) {
371 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
372 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
373 			kvm_inject_gp(vcpu, 0);
374 			return;
375 		}
376 	} else {
377 		if (is_pae(vcpu)) {
378 			if (cr3 & CR3_PAE_RESERVED_BITS) {
379 				printk(KERN_DEBUG
380 				       "set_cr3: #GP, reserved bits\n");
381 				kvm_inject_gp(vcpu, 0);
382 				return;
383 			}
384 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
385 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
386 				       "reserved bits\n");
387 				kvm_inject_gp(vcpu, 0);
388 				return;
389 			}
390 		}
391 		/*
392 		 * We don't check reserved bits in nonpae mode, because
393 		 * this isn't enforced, and VMware depends on this.
394 		 */
395 	}
396 
397 	/*
398 	 * Does the new cr3 value map to physical memory? (Note, we
399 	 * catch an invalid cr3 even in real-mode, because it would
400 	 * cause trouble later on when we turn on paging anyway.)
401 	 *
402 	 * A real CPU would silently accept an invalid cr3 and would
403 	 * attempt to use it - with largely undefined (and often hard
404 	 * to debug) behavior on the guest side.
405 	 */
406 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
407 		kvm_inject_gp(vcpu, 0);
408 	else {
409 		vcpu->arch.cr3 = cr3;
410 		vcpu->arch.mmu.new_cr3(vcpu);
411 	}
412 }
413 EXPORT_SYMBOL_GPL(kvm_set_cr3);
414 
415 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
416 {
417 	if (cr8 & CR8_RESERVED_BITS) {
418 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
419 		kvm_inject_gp(vcpu, 0);
420 		return;
421 	}
422 	if (irqchip_in_kernel(vcpu->kvm))
423 		kvm_lapic_set_tpr(vcpu, cr8);
424 	else
425 		vcpu->arch.cr8 = cr8;
426 }
427 EXPORT_SYMBOL_GPL(kvm_set_cr8);
428 
429 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
430 {
431 	if (irqchip_in_kernel(vcpu->kvm))
432 		return kvm_lapic_get_cr8(vcpu);
433 	else
434 		return vcpu->arch.cr8;
435 }
436 EXPORT_SYMBOL_GPL(kvm_get_cr8);
437 
438 /*
439  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
440  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
441  *
442  * This list is modified at module load time to reflect the
443  * capabilities of the host cpu.
444  */
445 static u32 msrs_to_save[] = {
446 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
447 	MSR_K6_STAR,
448 #ifdef CONFIG_X86_64
449 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
450 #endif
451 	MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
452 	MSR_IA32_PERF_STATUS,
453 };
454 
455 static unsigned num_msrs_to_save;
456 
457 static u32 emulated_msrs[] = {
458 	MSR_IA32_MISC_ENABLE,
459 };
460 
461 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
462 {
463 	if (efer & efer_reserved_bits) {
464 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
465 		       efer);
466 		kvm_inject_gp(vcpu, 0);
467 		return;
468 	}
469 
470 	if (is_paging(vcpu)
471 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
472 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
473 		kvm_inject_gp(vcpu, 0);
474 		return;
475 	}
476 
477 	kvm_x86_ops->set_efer(vcpu, efer);
478 
479 	efer &= ~EFER_LMA;
480 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
481 
482 	vcpu->arch.shadow_efer = efer;
483 }
484 
485 void kvm_enable_efer_bits(u64 mask)
486 {
487        efer_reserved_bits &= ~mask;
488 }
489 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
490 
491 
492 /*
493  * Writes msr value into into the appropriate "register".
494  * Returns 0 on success, non-0 otherwise.
495  * Assumes vcpu_load() was already called.
496  */
497 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
498 {
499 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
500 }
501 
502 /*
503  * Adapt set_msr() to msr_io()'s calling convention
504  */
505 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
506 {
507 	return kvm_set_msr(vcpu, index, *data);
508 }
509 
510 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
511 {
512 	static int version;
513 	struct pvclock_wall_clock wc;
514 	struct timespec now, sys, boot;
515 
516 	if (!wall_clock)
517 		return;
518 
519 	version++;
520 
521 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
522 
523 	/*
524 	 * The guest calculates current wall clock time by adding
525 	 * system time (updated by kvm_write_guest_time below) to the
526 	 * wall clock specified here.  guest system time equals host
527 	 * system time for us, thus we must fill in host boot time here.
528 	 */
529 	now = current_kernel_time();
530 	ktime_get_ts(&sys);
531 	boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
532 
533 	wc.sec = boot.tv_sec;
534 	wc.nsec = boot.tv_nsec;
535 	wc.version = version;
536 
537 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
538 
539 	version++;
540 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
541 }
542 
543 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
544 {
545 	uint32_t quotient, remainder;
546 
547 	/* Don't try to replace with do_div(), this one calculates
548 	 * "(dividend << 32) / divisor" */
549 	__asm__ ( "divl %4"
550 		  : "=a" (quotient), "=d" (remainder)
551 		  : "0" (0), "1" (dividend), "r" (divisor) );
552 	return quotient;
553 }
554 
555 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
556 {
557 	uint64_t nsecs = 1000000000LL;
558 	int32_t  shift = 0;
559 	uint64_t tps64;
560 	uint32_t tps32;
561 
562 	tps64 = tsc_khz * 1000LL;
563 	while (tps64 > nsecs*2) {
564 		tps64 >>= 1;
565 		shift--;
566 	}
567 
568 	tps32 = (uint32_t)tps64;
569 	while (tps32 <= (uint32_t)nsecs) {
570 		tps32 <<= 1;
571 		shift++;
572 	}
573 
574 	hv_clock->tsc_shift = shift;
575 	hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
576 
577 	pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
578 		 __func__, tsc_khz, hv_clock->tsc_shift,
579 		 hv_clock->tsc_to_system_mul);
580 }
581 
582 static void kvm_write_guest_time(struct kvm_vcpu *v)
583 {
584 	struct timespec ts;
585 	unsigned long flags;
586 	struct kvm_vcpu_arch *vcpu = &v->arch;
587 	void *shared_kaddr;
588 
589 	if ((!vcpu->time_page))
590 		return;
591 
592 	if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
593 		kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
594 		vcpu->hv_clock_tsc_khz = tsc_khz;
595 	}
596 
597 	/* Keep irq disabled to prevent changes to the clock */
598 	local_irq_save(flags);
599 	kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
600 			  &vcpu->hv_clock.tsc_timestamp);
601 	ktime_get_ts(&ts);
602 	local_irq_restore(flags);
603 
604 	/* With all the info we got, fill in the values */
605 
606 	vcpu->hv_clock.system_time = ts.tv_nsec +
607 				     (NSEC_PER_SEC * (u64)ts.tv_sec);
608 	/*
609 	 * The interface expects us to write an even number signaling that the
610 	 * update is finished. Since the guest won't see the intermediate
611 	 * state, we just increase by 2 at the end.
612 	 */
613 	vcpu->hv_clock.version += 2;
614 
615 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
616 
617 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
618 	       sizeof(vcpu->hv_clock));
619 
620 	kunmap_atomic(shared_kaddr, KM_USER0);
621 
622 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
623 }
624 
625 static bool msr_mtrr_valid(unsigned msr)
626 {
627 	switch (msr) {
628 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
629 	case MSR_MTRRfix64K_00000:
630 	case MSR_MTRRfix16K_80000:
631 	case MSR_MTRRfix16K_A0000:
632 	case MSR_MTRRfix4K_C0000:
633 	case MSR_MTRRfix4K_C8000:
634 	case MSR_MTRRfix4K_D0000:
635 	case MSR_MTRRfix4K_D8000:
636 	case MSR_MTRRfix4K_E0000:
637 	case MSR_MTRRfix4K_E8000:
638 	case MSR_MTRRfix4K_F0000:
639 	case MSR_MTRRfix4K_F8000:
640 	case MSR_MTRRdefType:
641 	case MSR_IA32_CR_PAT:
642 		return true;
643 	case 0x2f8:
644 		return true;
645 	}
646 	return false;
647 }
648 
649 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
650 {
651 	if (!msr_mtrr_valid(msr))
652 		return 1;
653 
654 	vcpu->arch.mtrr[msr - 0x200] = data;
655 	return 0;
656 }
657 
658 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
659 {
660 	switch (msr) {
661 	case MSR_EFER:
662 		set_efer(vcpu, data);
663 		break;
664 	case MSR_IA32_MC0_STATUS:
665 		pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
666 		       __func__, data);
667 		break;
668 	case MSR_IA32_MCG_STATUS:
669 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
670 			__func__, data);
671 		break;
672 	case MSR_IA32_MCG_CTL:
673 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
674 			__func__, data);
675 		break;
676 	case MSR_IA32_DEBUGCTLMSR:
677 		if (!data) {
678 			/* We support the non-activated case already */
679 			break;
680 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
681 			/* Values other than LBR and BTF are vendor-specific,
682 			   thus reserved and should throw a #GP */
683 			return 1;
684 		}
685 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
686 			__func__, data);
687 		break;
688 	case MSR_IA32_UCODE_REV:
689 	case MSR_IA32_UCODE_WRITE:
690 		break;
691 	case 0x200 ... 0x2ff:
692 		return set_msr_mtrr(vcpu, msr, data);
693 	case MSR_IA32_APICBASE:
694 		kvm_set_apic_base(vcpu, data);
695 		break;
696 	case MSR_IA32_MISC_ENABLE:
697 		vcpu->arch.ia32_misc_enable_msr = data;
698 		break;
699 	case MSR_KVM_WALL_CLOCK:
700 		vcpu->kvm->arch.wall_clock = data;
701 		kvm_write_wall_clock(vcpu->kvm, data);
702 		break;
703 	case MSR_KVM_SYSTEM_TIME: {
704 		if (vcpu->arch.time_page) {
705 			kvm_release_page_dirty(vcpu->arch.time_page);
706 			vcpu->arch.time_page = NULL;
707 		}
708 
709 		vcpu->arch.time = data;
710 
711 		/* we verify if the enable bit is set... */
712 		if (!(data & 1))
713 			break;
714 
715 		/* ...but clean it before doing the actual write */
716 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
717 
718 		vcpu->arch.time_page =
719 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
720 
721 		if (is_error_page(vcpu->arch.time_page)) {
722 			kvm_release_page_clean(vcpu->arch.time_page);
723 			vcpu->arch.time_page = NULL;
724 		}
725 
726 		kvm_write_guest_time(vcpu);
727 		break;
728 	}
729 	default:
730 		pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
731 		return 1;
732 	}
733 	return 0;
734 }
735 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
736 
737 
738 /*
739  * Reads an msr value (of 'msr_index') into 'pdata'.
740  * Returns 0 on success, non-0 otherwise.
741  * Assumes vcpu_load() was already called.
742  */
743 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
744 {
745 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
746 }
747 
748 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
749 {
750 	if (!msr_mtrr_valid(msr))
751 		return 1;
752 
753 	*pdata = vcpu->arch.mtrr[msr - 0x200];
754 	return 0;
755 }
756 
757 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
758 {
759 	u64 data;
760 
761 	switch (msr) {
762 	case 0xc0010010: /* SYSCFG */
763 	case 0xc0010015: /* HWCR */
764 	case MSR_IA32_PLATFORM_ID:
765 	case MSR_IA32_P5_MC_ADDR:
766 	case MSR_IA32_P5_MC_TYPE:
767 	case MSR_IA32_MC0_CTL:
768 	case MSR_IA32_MCG_STATUS:
769 	case MSR_IA32_MCG_CAP:
770 	case MSR_IA32_MCG_CTL:
771 	case MSR_IA32_MC0_MISC:
772 	case MSR_IA32_MC0_MISC+4:
773 	case MSR_IA32_MC0_MISC+8:
774 	case MSR_IA32_MC0_MISC+12:
775 	case MSR_IA32_MC0_MISC+16:
776 	case MSR_IA32_MC0_MISC+20:
777 	case MSR_IA32_UCODE_REV:
778 	case MSR_IA32_EBL_CR_POWERON:
779 	case MSR_IA32_DEBUGCTLMSR:
780 	case MSR_IA32_LASTBRANCHFROMIP:
781 	case MSR_IA32_LASTBRANCHTOIP:
782 	case MSR_IA32_LASTINTFROMIP:
783 	case MSR_IA32_LASTINTTOIP:
784 		data = 0;
785 		break;
786 	case MSR_MTRRcap:
787 		data = 0x500 | KVM_NR_VAR_MTRR;
788 		break;
789 	case 0x200 ... 0x2ff:
790 		return get_msr_mtrr(vcpu, msr, pdata);
791 	case 0xcd: /* fsb frequency */
792 		data = 3;
793 		break;
794 	case MSR_IA32_APICBASE:
795 		data = kvm_get_apic_base(vcpu);
796 		break;
797 	case MSR_IA32_MISC_ENABLE:
798 		data = vcpu->arch.ia32_misc_enable_msr;
799 		break;
800 	case MSR_IA32_PERF_STATUS:
801 		/* TSC increment by tick */
802 		data = 1000ULL;
803 		/* CPU multiplier */
804 		data |= (((uint64_t)4ULL) << 40);
805 		break;
806 	case MSR_EFER:
807 		data = vcpu->arch.shadow_efer;
808 		break;
809 	case MSR_KVM_WALL_CLOCK:
810 		data = vcpu->kvm->arch.wall_clock;
811 		break;
812 	case MSR_KVM_SYSTEM_TIME:
813 		data = vcpu->arch.time;
814 		break;
815 	default:
816 		pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
817 		return 1;
818 	}
819 	*pdata = data;
820 	return 0;
821 }
822 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
823 
824 /*
825  * Read or write a bunch of msrs. All parameters are kernel addresses.
826  *
827  * @return number of msrs set successfully.
828  */
829 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
830 		    struct kvm_msr_entry *entries,
831 		    int (*do_msr)(struct kvm_vcpu *vcpu,
832 				  unsigned index, u64 *data))
833 {
834 	int i;
835 
836 	vcpu_load(vcpu);
837 
838 	down_read(&vcpu->kvm->slots_lock);
839 	for (i = 0; i < msrs->nmsrs; ++i)
840 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
841 			break;
842 	up_read(&vcpu->kvm->slots_lock);
843 
844 	vcpu_put(vcpu);
845 
846 	return i;
847 }
848 
849 /*
850  * Read or write a bunch of msrs. Parameters are user addresses.
851  *
852  * @return number of msrs set successfully.
853  */
854 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
855 		  int (*do_msr)(struct kvm_vcpu *vcpu,
856 				unsigned index, u64 *data),
857 		  int writeback)
858 {
859 	struct kvm_msrs msrs;
860 	struct kvm_msr_entry *entries;
861 	int r, n;
862 	unsigned size;
863 
864 	r = -EFAULT;
865 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
866 		goto out;
867 
868 	r = -E2BIG;
869 	if (msrs.nmsrs >= MAX_IO_MSRS)
870 		goto out;
871 
872 	r = -ENOMEM;
873 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
874 	entries = vmalloc(size);
875 	if (!entries)
876 		goto out;
877 
878 	r = -EFAULT;
879 	if (copy_from_user(entries, user_msrs->entries, size))
880 		goto out_free;
881 
882 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
883 	if (r < 0)
884 		goto out_free;
885 
886 	r = -EFAULT;
887 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
888 		goto out_free;
889 
890 	r = n;
891 
892 out_free:
893 	vfree(entries);
894 out:
895 	return r;
896 }
897 
898 int kvm_dev_ioctl_check_extension(long ext)
899 {
900 	int r;
901 
902 	switch (ext) {
903 	case KVM_CAP_IRQCHIP:
904 	case KVM_CAP_HLT:
905 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
906 	case KVM_CAP_USER_MEMORY:
907 	case KVM_CAP_SET_TSS_ADDR:
908 	case KVM_CAP_EXT_CPUID:
909 	case KVM_CAP_CLOCKSOURCE:
910 	case KVM_CAP_PIT:
911 	case KVM_CAP_NOP_IO_DELAY:
912 	case KVM_CAP_MP_STATE:
913 	case KVM_CAP_SYNC_MMU:
914 		r = 1;
915 		break;
916 	case KVM_CAP_COALESCED_MMIO:
917 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
918 		break;
919 	case KVM_CAP_VAPIC:
920 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
921 		break;
922 	case KVM_CAP_NR_VCPUS:
923 		r = KVM_MAX_VCPUS;
924 		break;
925 	case KVM_CAP_NR_MEMSLOTS:
926 		r = KVM_MEMORY_SLOTS;
927 		break;
928 	case KVM_CAP_PV_MMU:
929 		r = !tdp_enabled;
930 		break;
931 	case KVM_CAP_IOMMU:
932 		r = intel_iommu_found();
933 		break;
934 	default:
935 		r = 0;
936 		break;
937 	}
938 	return r;
939 
940 }
941 
942 long kvm_arch_dev_ioctl(struct file *filp,
943 			unsigned int ioctl, unsigned long arg)
944 {
945 	void __user *argp = (void __user *)arg;
946 	long r;
947 
948 	switch (ioctl) {
949 	case KVM_GET_MSR_INDEX_LIST: {
950 		struct kvm_msr_list __user *user_msr_list = argp;
951 		struct kvm_msr_list msr_list;
952 		unsigned n;
953 
954 		r = -EFAULT;
955 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
956 			goto out;
957 		n = msr_list.nmsrs;
958 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
959 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
960 			goto out;
961 		r = -E2BIG;
962 		if (n < num_msrs_to_save)
963 			goto out;
964 		r = -EFAULT;
965 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
966 				 num_msrs_to_save * sizeof(u32)))
967 			goto out;
968 		if (copy_to_user(user_msr_list->indices
969 				 + num_msrs_to_save * sizeof(u32),
970 				 &emulated_msrs,
971 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
972 			goto out;
973 		r = 0;
974 		break;
975 	}
976 	case KVM_GET_SUPPORTED_CPUID: {
977 		struct kvm_cpuid2 __user *cpuid_arg = argp;
978 		struct kvm_cpuid2 cpuid;
979 
980 		r = -EFAULT;
981 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
982 			goto out;
983 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
984 			cpuid_arg->entries);
985 		if (r)
986 			goto out;
987 
988 		r = -EFAULT;
989 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
990 			goto out;
991 		r = 0;
992 		break;
993 	}
994 	default:
995 		r = -EINVAL;
996 	}
997 out:
998 	return r;
999 }
1000 
1001 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1002 {
1003 	kvm_x86_ops->vcpu_load(vcpu, cpu);
1004 	kvm_write_guest_time(vcpu);
1005 }
1006 
1007 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1008 {
1009 	kvm_x86_ops->vcpu_put(vcpu);
1010 	kvm_put_guest_fpu(vcpu);
1011 }
1012 
1013 static int is_efer_nx(void)
1014 {
1015 	u64 efer;
1016 
1017 	rdmsrl(MSR_EFER, efer);
1018 	return efer & EFER_NX;
1019 }
1020 
1021 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1022 {
1023 	int i;
1024 	struct kvm_cpuid_entry2 *e, *entry;
1025 
1026 	entry = NULL;
1027 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1028 		e = &vcpu->arch.cpuid_entries[i];
1029 		if (e->function == 0x80000001) {
1030 			entry = e;
1031 			break;
1032 		}
1033 	}
1034 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1035 		entry->edx &= ~(1 << 20);
1036 		printk(KERN_INFO "kvm: guest NX capability removed\n");
1037 	}
1038 }
1039 
1040 /* when an old userspace process fills a new kernel module */
1041 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1042 				    struct kvm_cpuid *cpuid,
1043 				    struct kvm_cpuid_entry __user *entries)
1044 {
1045 	int r, i;
1046 	struct kvm_cpuid_entry *cpuid_entries;
1047 
1048 	r = -E2BIG;
1049 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1050 		goto out;
1051 	r = -ENOMEM;
1052 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1053 	if (!cpuid_entries)
1054 		goto out;
1055 	r = -EFAULT;
1056 	if (copy_from_user(cpuid_entries, entries,
1057 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1058 		goto out_free;
1059 	for (i = 0; i < cpuid->nent; i++) {
1060 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1061 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1062 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1063 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1064 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1065 		vcpu->arch.cpuid_entries[i].index = 0;
1066 		vcpu->arch.cpuid_entries[i].flags = 0;
1067 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
1068 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
1069 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
1070 	}
1071 	vcpu->arch.cpuid_nent = cpuid->nent;
1072 	cpuid_fix_nx_cap(vcpu);
1073 	r = 0;
1074 
1075 out_free:
1076 	vfree(cpuid_entries);
1077 out:
1078 	return r;
1079 }
1080 
1081 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1082 				    struct kvm_cpuid2 *cpuid,
1083 				    struct kvm_cpuid_entry2 __user *entries)
1084 {
1085 	int r;
1086 
1087 	r = -E2BIG;
1088 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1089 		goto out;
1090 	r = -EFAULT;
1091 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1092 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1093 		goto out;
1094 	vcpu->arch.cpuid_nent = cpuid->nent;
1095 	return 0;
1096 
1097 out:
1098 	return r;
1099 }
1100 
1101 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1102 				    struct kvm_cpuid2 *cpuid,
1103 				    struct kvm_cpuid_entry2 __user *entries)
1104 {
1105 	int r;
1106 
1107 	r = -E2BIG;
1108 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1109 		goto out;
1110 	r = -EFAULT;
1111 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1112 			   vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1113 		goto out;
1114 	return 0;
1115 
1116 out:
1117 	cpuid->nent = vcpu->arch.cpuid_nent;
1118 	return r;
1119 }
1120 
1121 static inline u32 bit(int bitno)
1122 {
1123 	return 1 << (bitno & 31);
1124 }
1125 
1126 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1127 			  u32 index)
1128 {
1129 	entry->function = function;
1130 	entry->index = index;
1131 	cpuid_count(entry->function, entry->index,
1132 		&entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1133 	entry->flags = 0;
1134 }
1135 
1136 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1137 			 u32 index, int *nent, int maxnent)
1138 {
1139 	const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1140 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1141 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1142 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1143 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1144 		bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1145 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1146 		bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1147 		bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1148 		bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1149 	const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1150 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1151 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1152 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1153 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1154 		bit(X86_FEATURE_PGE) |
1155 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1156 		bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1157 		bit(X86_FEATURE_SYSCALL) |
1158 		(bit(X86_FEATURE_NX) && is_efer_nx()) |
1159 #ifdef CONFIG_X86_64
1160 		bit(X86_FEATURE_LM) |
1161 #endif
1162 		bit(X86_FEATURE_MMXEXT) |
1163 		bit(X86_FEATURE_3DNOWEXT) |
1164 		bit(X86_FEATURE_3DNOW);
1165 	const u32 kvm_supported_word3_x86_features =
1166 		bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1167 	const u32 kvm_supported_word6_x86_features =
1168 		bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1169 
1170 	/* all func 2 cpuid_count() should be called on the same cpu */
1171 	get_cpu();
1172 	do_cpuid_1_ent(entry, function, index);
1173 	++*nent;
1174 
1175 	switch (function) {
1176 	case 0:
1177 		entry->eax = min(entry->eax, (u32)0xb);
1178 		break;
1179 	case 1:
1180 		entry->edx &= kvm_supported_word0_x86_features;
1181 		entry->ecx &= kvm_supported_word3_x86_features;
1182 		break;
1183 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1184 	 * may return different values. This forces us to get_cpu() before
1185 	 * issuing the first command, and also to emulate this annoying behavior
1186 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1187 	case 2: {
1188 		int t, times = entry->eax & 0xff;
1189 
1190 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1191 		for (t = 1; t < times && *nent < maxnent; ++t) {
1192 			do_cpuid_1_ent(&entry[t], function, 0);
1193 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1194 			++*nent;
1195 		}
1196 		break;
1197 	}
1198 	/* function 4 and 0xb have additional index. */
1199 	case 4: {
1200 		int i, cache_type;
1201 
1202 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1203 		/* read more entries until cache_type is zero */
1204 		for (i = 1; *nent < maxnent; ++i) {
1205 			cache_type = entry[i - 1].eax & 0x1f;
1206 			if (!cache_type)
1207 				break;
1208 			do_cpuid_1_ent(&entry[i], function, i);
1209 			entry[i].flags |=
1210 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1211 			++*nent;
1212 		}
1213 		break;
1214 	}
1215 	case 0xb: {
1216 		int i, level_type;
1217 
1218 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1219 		/* read more entries until level_type is zero */
1220 		for (i = 1; *nent < maxnent; ++i) {
1221 			level_type = entry[i - 1].ecx & 0xff;
1222 			if (!level_type)
1223 				break;
1224 			do_cpuid_1_ent(&entry[i], function, i);
1225 			entry[i].flags |=
1226 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1227 			++*nent;
1228 		}
1229 		break;
1230 	}
1231 	case 0x80000000:
1232 		entry->eax = min(entry->eax, 0x8000001a);
1233 		break;
1234 	case 0x80000001:
1235 		entry->edx &= kvm_supported_word1_x86_features;
1236 		entry->ecx &= kvm_supported_word6_x86_features;
1237 		break;
1238 	}
1239 	put_cpu();
1240 }
1241 
1242 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1243 				    struct kvm_cpuid_entry2 __user *entries)
1244 {
1245 	struct kvm_cpuid_entry2 *cpuid_entries;
1246 	int limit, nent = 0, r = -E2BIG;
1247 	u32 func;
1248 
1249 	if (cpuid->nent < 1)
1250 		goto out;
1251 	r = -ENOMEM;
1252 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1253 	if (!cpuid_entries)
1254 		goto out;
1255 
1256 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1257 	limit = cpuid_entries[0].eax;
1258 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1259 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1260 				&nent, cpuid->nent);
1261 	r = -E2BIG;
1262 	if (nent >= cpuid->nent)
1263 		goto out_free;
1264 
1265 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1266 	limit = cpuid_entries[nent - 1].eax;
1267 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1268 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1269 			       &nent, cpuid->nent);
1270 	r = -EFAULT;
1271 	if (copy_to_user(entries, cpuid_entries,
1272 			nent * sizeof(struct kvm_cpuid_entry2)))
1273 		goto out_free;
1274 	cpuid->nent = nent;
1275 	r = 0;
1276 
1277 out_free:
1278 	vfree(cpuid_entries);
1279 out:
1280 	return r;
1281 }
1282 
1283 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1284 				    struct kvm_lapic_state *s)
1285 {
1286 	vcpu_load(vcpu);
1287 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1288 	vcpu_put(vcpu);
1289 
1290 	return 0;
1291 }
1292 
1293 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1294 				    struct kvm_lapic_state *s)
1295 {
1296 	vcpu_load(vcpu);
1297 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1298 	kvm_apic_post_state_restore(vcpu);
1299 	vcpu_put(vcpu);
1300 
1301 	return 0;
1302 }
1303 
1304 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1305 				    struct kvm_interrupt *irq)
1306 {
1307 	if (irq->irq < 0 || irq->irq >= 256)
1308 		return -EINVAL;
1309 	if (irqchip_in_kernel(vcpu->kvm))
1310 		return -ENXIO;
1311 	vcpu_load(vcpu);
1312 
1313 	set_bit(irq->irq, vcpu->arch.irq_pending);
1314 	set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1315 
1316 	vcpu_put(vcpu);
1317 
1318 	return 0;
1319 }
1320 
1321 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1322 					   struct kvm_tpr_access_ctl *tac)
1323 {
1324 	if (tac->flags)
1325 		return -EINVAL;
1326 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1327 	return 0;
1328 }
1329 
1330 long kvm_arch_vcpu_ioctl(struct file *filp,
1331 			 unsigned int ioctl, unsigned long arg)
1332 {
1333 	struct kvm_vcpu *vcpu = filp->private_data;
1334 	void __user *argp = (void __user *)arg;
1335 	int r;
1336 	struct kvm_lapic_state *lapic = NULL;
1337 
1338 	switch (ioctl) {
1339 	case KVM_GET_LAPIC: {
1340 		lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1341 
1342 		r = -ENOMEM;
1343 		if (!lapic)
1344 			goto out;
1345 		r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1346 		if (r)
1347 			goto out;
1348 		r = -EFAULT;
1349 		if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1350 			goto out;
1351 		r = 0;
1352 		break;
1353 	}
1354 	case KVM_SET_LAPIC: {
1355 		lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1356 		r = -ENOMEM;
1357 		if (!lapic)
1358 			goto out;
1359 		r = -EFAULT;
1360 		if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1361 			goto out;
1362 		r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1363 		if (r)
1364 			goto out;
1365 		r = 0;
1366 		break;
1367 	}
1368 	case KVM_INTERRUPT: {
1369 		struct kvm_interrupt irq;
1370 
1371 		r = -EFAULT;
1372 		if (copy_from_user(&irq, argp, sizeof irq))
1373 			goto out;
1374 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1375 		if (r)
1376 			goto out;
1377 		r = 0;
1378 		break;
1379 	}
1380 	case KVM_SET_CPUID: {
1381 		struct kvm_cpuid __user *cpuid_arg = argp;
1382 		struct kvm_cpuid cpuid;
1383 
1384 		r = -EFAULT;
1385 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1386 			goto out;
1387 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1388 		if (r)
1389 			goto out;
1390 		break;
1391 	}
1392 	case KVM_SET_CPUID2: {
1393 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1394 		struct kvm_cpuid2 cpuid;
1395 
1396 		r = -EFAULT;
1397 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1398 			goto out;
1399 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1400 				cpuid_arg->entries);
1401 		if (r)
1402 			goto out;
1403 		break;
1404 	}
1405 	case KVM_GET_CPUID2: {
1406 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1407 		struct kvm_cpuid2 cpuid;
1408 
1409 		r = -EFAULT;
1410 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1411 			goto out;
1412 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1413 				cpuid_arg->entries);
1414 		if (r)
1415 			goto out;
1416 		r = -EFAULT;
1417 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1418 			goto out;
1419 		r = 0;
1420 		break;
1421 	}
1422 	case KVM_GET_MSRS:
1423 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1424 		break;
1425 	case KVM_SET_MSRS:
1426 		r = msr_io(vcpu, argp, do_set_msr, 0);
1427 		break;
1428 	case KVM_TPR_ACCESS_REPORTING: {
1429 		struct kvm_tpr_access_ctl tac;
1430 
1431 		r = -EFAULT;
1432 		if (copy_from_user(&tac, argp, sizeof tac))
1433 			goto out;
1434 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1435 		if (r)
1436 			goto out;
1437 		r = -EFAULT;
1438 		if (copy_to_user(argp, &tac, sizeof tac))
1439 			goto out;
1440 		r = 0;
1441 		break;
1442 	};
1443 	case KVM_SET_VAPIC_ADDR: {
1444 		struct kvm_vapic_addr va;
1445 
1446 		r = -EINVAL;
1447 		if (!irqchip_in_kernel(vcpu->kvm))
1448 			goto out;
1449 		r = -EFAULT;
1450 		if (copy_from_user(&va, argp, sizeof va))
1451 			goto out;
1452 		r = 0;
1453 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1454 		break;
1455 	}
1456 	default:
1457 		r = -EINVAL;
1458 	}
1459 out:
1460 	if (lapic)
1461 		kfree(lapic);
1462 	return r;
1463 }
1464 
1465 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1466 {
1467 	int ret;
1468 
1469 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1470 		return -1;
1471 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1472 	return ret;
1473 }
1474 
1475 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1476 					  u32 kvm_nr_mmu_pages)
1477 {
1478 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1479 		return -EINVAL;
1480 
1481 	down_write(&kvm->slots_lock);
1482 
1483 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1484 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1485 
1486 	up_write(&kvm->slots_lock);
1487 	return 0;
1488 }
1489 
1490 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1491 {
1492 	return kvm->arch.n_alloc_mmu_pages;
1493 }
1494 
1495 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1496 {
1497 	int i;
1498 	struct kvm_mem_alias *alias;
1499 
1500 	for (i = 0; i < kvm->arch.naliases; ++i) {
1501 		alias = &kvm->arch.aliases[i];
1502 		if (gfn >= alias->base_gfn
1503 		    && gfn < alias->base_gfn + alias->npages)
1504 			return alias->target_gfn + gfn - alias->base_gfn;
1505 	}
1506 	return gfn;
1507 }
1508 
1509 /*
1510  * Set a new alias region.  Aliases map a portion of physical memory into
1511  * another portion.  This is useful for memory windows, for example the PC
1512  * VGA region.
1513  */
1514 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1515 					 struct kvm_memory_alias *alias)
1516 {
1517 	int r, n;
1518 	struct kvm_mem_alias *p;
1519 
1520 	r = -EINVAL;
1521 	/* General sanity checks */
1522 	if (alias->memory_size & (PAGE_SIZE - 1))
1523 		goto out;
1524 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1525 		goto out;
1526 	if (alias->slot >= KVM_ALIAS_SLOTS)
1527 		goto out;
1528 	if (alias->guest_phys_addr + alias->memory_size
1529 	    < alias->guest_phys_addr)
1530 		goto out;
1531 	if (alias->target_phys_addr + alias->memory_size
1532 	    < alias->target_phys_addr)
1533 		goto out;
1534 
1535 	down_write(&kvm->slots_lock);
1536 	spin_lock(&kvm->mmu_lock);
1537 
1538 	p = &kvm->arch.aliases[alias->slot];
1539 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1540 	p->npages = alias->memory_size >> PAGE_SHIFT;
1541 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1542 
1543 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1544 		if (kvm->arch.aliases[n - 1].npages)
1545 			break;
1546 	kvm->arch.naliases = n;
1547 
1548 	spin_unlock(&kvm->mmu_lock);
1549 	kvm_mmu_zap_all(kvm);
1550 
1551 	up_write(&kvm->slots_lock);
1552 
1553 	return 0;
1554 
1555 out:
1556 	return r;
1557 }
1558 
1559 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1560 {
1561 	int r;
1562 
1563 	r = 0;
1564 	switch (chip->chip_id) {
1565 	case KVM_IRQCHIP_PIC_MASTER:
1566 		memcpy(&chip->chip.pic,
1567 			&pic_irqchip(kvm)->pics[0],
1568 			sizeof(struct kvm_pic_state));
1569 		break;
1570 	case KVM_IRQCHIP_PIC_SLAVE:
1571 		memcpy(&chip->chip.pic,
1572 			&pic_irqchip(kvm)->pics[1],
1573 			sizeof(struct kvm_pic_state));
1574 		break;
1575 	case KVM_IRQCHIP_IOAPIC:
1576 		memcpy(&chip->chip.ioapic,
1577 			ioapic_irqchip(kvm),
1578 			sizeof(struct kvm_ioapic_state));
1579 		break;
1580 	default:
1581 		r = -EINVAL;
1582 		break;
1583 	}
1584 	return r;
1585 }
1586 
1587 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1588 {
1589 	int r;
1590 
1591 	r = 0;
1592 	switch (chip->chip_id) {
1593 	case KVM_IRQCHIP_PIC_MASTER:
1594 		memcpy(&pic_irqchip(kvm)->pics[0],
1595 			&chip->chip.pic,
1596 			sizeof(struct kvm_pic_state));
1597 		break;
1598 	case KVM_IRQCHIP_PIC_SLAVE:
1599 		memcpy(&pic_irqchip(kvm)->pics[1],
1600 			&chip->chip.pic,
1601 			sizeof(struct kvm_pic_state));
1602 		break;
1603 	case KVM_IRQCHIP_IOAPIC:
1604 		memcpy(ioapic_irqchip(kvm),
1605 			&chip->chip.ioapic,
1606 			sizeof(struct kvm_ioapic_state));
1607 		break;
1608 	default:
1609 		r = -EINVAL;
1610 		break;
1611 	}
1612 	kvm_pic_update_irq(pic_irqchip(kvm));
1613 	return r;
1614 }
1615 
1616 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1617 {
1618 	int r = 0;
1619 
1620 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1621 	return r;
1622 }
1623 
1624 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1625 {
1626 	int r = 0;
1627 
1628 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1629 	kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1630 	return r;
1631 }
1632 
1633 /*
1634  * Get (and clear) the dirty memory log for a memory slot.
1635  */
1636 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1637 				      struct kvm_dirty_log *log)
1638 {
1639 	int r;
1640 	int n;
1641 	struct kvm_memory_slot *memslot;
1642 	int is_dirty = 0;
1643 
1644 	down_write(&kvm->slots_lock);
1645 
1646 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1647 	if (r)
1648 		goto out;
1649 
1650 	/* If nothing is dirty, don't bother messing with page tables. */
1651 	if (is_dirty) {
1652 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
1653 		kvm_flush_remote_tlbs(kvm);
1654 		memslot = &kvm->memslots[log->slot];
1655 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1656 		memset(memslot->dirty_bitmap, 0, n);
1657 	}
1658 	r = 0;
1659 out:
1660 	up_write(&kvm->slots_lock);
1661 	return r;
1662 }
1663 
1664 long kvm_arch_vm_ioctl(struct file *filp,
1665 		       unsigned int ioctl, unsigned long arg)
1666 {
1667 	struct kvm *kvm = filp->private_data;
1668 	void __user *argp = (void __user *)arg;
1669 	int r = -EINVAL;
1670 	/*
1671 	 * This union makes it completely explicit to gcc-3.x
1672 	 * that these two variables' stack usage should be
1673 	 * combined, not added together.
1674 	 */
1675 	union {
1676 		struct kvm_pit_state ps;
1677 		struct kvm_memory_alias alias;
1678 	} u;
1679 
1680 	switch (ioctl) {
1681 	case KVM_SET_TSS_ADDR:
1682 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1683 		if (r < 0)
1684 			goto out;
1685 		break;
1686 	case KVM_SET_MEMORY_REGION: {
1687 		struct kvm_memory_region kvm_mem;
1688 		struct kvm_userspace_memory_region kvm_userspace_mem;
1689 
1690 		r = -EFAULT;
1691 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1692 			goto out;
1693 		kvm_userspace_mem.slot = kvm_mem.slot;
1694 		kvm_userspace_mem.flags = kvm_mem.flags;
1695 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1696 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1697 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1698 		if (r)
1699 			goto out;
1700 		break;
1701 	}
1702 	case KVM_SET_NR_MMU_PAGES:
1703 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1704 		if (r)
1705 			goto out;
1706 		break;
1707 	case KVM_GET_NR_MMU_PAGES:
1708 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1709 		break;
1710 	case KVM_SET_MEMORY_ALIAS:
1711 		r = -EFAULT;
1712 		if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1713 			goto out;
1714 		r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1715 		if (r)
1716 			goto out;
1717 		break;
1718 	case KVM_CREATE_IRQCHIP:
1719 		r = -ENOMEM;
1720 		kvm->arch.vpic = kvm_create_pic(kvm);
1721 		if (kvm->arch.vpic) {
1722 			r = kvm_ioapic_init(kvm);
1723 			if (r) {
1724 				kfree(kvm->arch.vpic);
1725 				kvm->arch.vpic = NULL;
1726 				goto out;
1727 			}
1728 		} else
1729 			goto out;
1730 		break;
1731 	case KVM_CREATE_PIT:
1732 		r = -ENOMEM;
1733 		kvm->arch.vpit = kvm_create_pit(kvm);
1734 		if (kvm->arch.vpit)
1735 			r = 0;
1736 		break;
1737 	case KVM_IRQ_LINE: {
1738 		struct kvm_irq_level irq_event;
1739 
1740 		r = -EFAULT;
1741 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
1742 			goto out;
1743 		if (irqchip_in_kernel(kvm)) {
1744 			mutex_lock(&kvm->lock);
1745 			kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1746 				    irq_event.irq, irq_event.level);
1747 			mutex_unlock(&kvm->lock);
1748 			r = 0;
1749 		}
1750 		break;
1751 	}
1752 	case KVM_GET_IRQCHIP: {
1753 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1754 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1755 
1756 		r = -ENOMEM;
1757 		if (!chip)
1758 			goto out;
1759 		r = -EFAULT;
1760 		if (copy_from_user(chip, argp, sizeof *chip))
1761 			goto get_irqchip_out;
1762 		r = -ENXIO;
1763 		if (!irqchip_in_kernel(kvm))
1764 			goto get_irqchip_out;
1765 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1766 		if (r)
1767 			goto get_irqchip_out;
1768 		r = -EFAULT;
1769 		if (copy_to_user(argp, chip, sizeof *chip))
1770 			goto get_irqchip_out;
1771 		r = 0;
1772 	get_irqchip_out:
1773 		kfree(chip);
1774 		if (r)
1775 			goto out;
1776 		break;
1777 	}
1778 	case KVM_SET_IRQCHIP: {
1779 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1780 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1781 
1782 		r = -ENOMEM;
1783 		if (!chip)
1784 			goto out;
1785 		r = -EFAULT;
1786 		if (copy_from_user(chip, argp, sizeof *chip))
1787 			goto set_irqchip_out;
1788 		r = -ENXIO;
1789 		if (!irqchip_in_kernel(kvm))
1790 			goto set_irqchip_out;
1791 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1792 		if (r)
1793 			goto set_irqchip_out;
1794 		r = 0;
1795 	set_irqchip_out:
1796 		kfree(chip);
1797 		if (r)
1798 			goto out;
1799 		break;
1800 	}
1801 	case KVM_GET_PIT: {
1802 		r = -EFAULT;
1803 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1804 			goto out;
1805 		r = -ENXIO;
1806 		if (!kvm->arch.vpit)
1807 			goto out;
1808 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1809 		if (r)
1810 			goto out;
1811 		r = -EFAULT;
1812 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1813 			goto out;
1814 		r = 0;
1815 		break;
1816 	}
1817 	case KVM_SET_PIT: {
1818 		r = -EFAULT;
1819 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
1820 			goto out;
1821 		r = -ENXIO;
1822 		if (!kvm->arch.vpit)
1823 			goto out;
1824 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1825 		if (r)
1826 			goto out;
1827 		r = 0;
1828 		break;
1829 	}
1830 	default:
1831 		;
1832 	}
1833 out:
1834 	return r;
1835 }
1836 
1837 static void kvm_init_msr_list(void)
1838 {
1839 	u32 dummy[2];
1840 	unsigned i, j;
1841 
1842 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1843 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1844 			continue;
1845 		if (j < i)
1846 			msrs_to_save[j] = msrs_to_save[i];
1847 		j++;
1848 	}
1849 	num_msrs_to_save = j;
1850 }
1851 
1852 /*
1853  * Only apic need an MMIO device hook, so shortcut now..
1854  */
1855 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1856 						gpa_t addr, int len,
1857 						int is_write)
1858 {
1859 	struct kvm_io_device *dev;
1860 
1861 	if (vcpu->arch.apic) {
1862 		dev = &vcpu->arch.apic->dev;
1863 		if (dev->in_range(dev, addr, len, is_write))
1864 			return dev;
1865 	}
1866 	return NULL;
1867 }
1868 
1869 
1870 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1871 						gpa_t addr, int len,
1872 						int is_write)
1873 {
1874 	struct kvm_io_device *dev;
1875 
1876 	dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1877 	if (dev == NULL)
1878 		dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1879 					  is_write);
1880 	return dev;
1881 }
1882 
1883 int emulator_read_std(unsigned long addr,
1884 			     void *val,
1885 			     unsigned int bytes,
1886 			     struct kvm_vcpu *vcpu)
1887 {
1888 	void *data = val;
1889 	int r = X86EMUL_CONTINUE;
1890 
1891 	while (bytes) {
1892 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1893 		unsigned offset = addr & (PAGE_SIZE-1);
1894 		unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1895 		int ret;
1896 
1897 		if (gpa == UNMAPPED_GVA) {
1898 			r = X86EMUL_PROPAGATE_FAULT;
1899 			goto out;
1900 		}
1901 		ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1902 		if (ret < 0) {
1903 			r = X86EMUL_UNHANDLEABLE;
1904 			goto out;
1905 		}
1906 
1907 		bytes -= tocopy;
1908 		data += tocopy;
1909 		addr += tocopy;
1910 	}
1911 out:
1912 	return r;
1913 }
1914 EXPORT_SYMBOL_GPL(emulator_read_std);
1915 
1916 static int emulator_read_emulated(unsigned long addr,
1917 				  void *val,
1918 				  unsigned int bytes,
1919 				  struct kvm_vcpu *vcpu)
1920 {
1921 	struct kvm_io_device *mmio_dev;
1922 	gpa_t                 gpa;
1923 
1924 	if (vcpu->mmio_read_completed) {
1925 		memcpy(val, vcpu->mmio_data, bytes);
1926 		vcpu->mmio_read_completed = 0;
1927 		return X86EMUL_CONTINUE;
1928 	}
1929 
1930 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1931 
1932 	/* For APIC access vmexit */
1933 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1934 		goto mmio;
1935 
1936 	if (emulator_read_std(addr, val, bytes, vcpu)
1937 			== X86EMUL_CONTINUE)
1938 		return X86EMUL_CONTINUE;
1939 	if (gpa == UNMAPPED_GVA)
1940 		return X86EMUL_PROPAGATE_FAULT;
1941 
1942 mmio:
1943 	/*
1944 	 * Is this MMIO handled locally?
1945 	 */
1946 	mutex_lock(&vcpu->kvm->lock);
1947 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1948 	if (mmio_dev) {
1949 		kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1950 		mutex_unlock(&vcpu->kvm->lock);
1951 		return X86EMUL_CONTINUE;
1952 	}
1953 	mutex_unlock(&vcpu->kvm->lock);
1954 
1955 	vcpu->mmio_needed = 1;
1956 	vcpu->mmio_phys_addr = gpa;
1957 	vcpu->mmio_size = bytes;
1958 	vcpu->mmio_is_write = 0;
1959 
1960 	return X86EMUL_UNHANDLEABLE;
1961 }
1962 
1963 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1964 			  const void *val, int bytes)
1965 {
1966 	int ret;
1967 
1968 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1969 	if (ret < 0)
1970 		return 0;
1971 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1972 	return 1;
1973 }
1974 
1975 static int emulator_write_emulated_onepage(unsigned long addr,
1976 					   const void *val,
1977 					   unsigned int bytes,
1978 					   struct kvm_vcpu *vcpu)
1979 {
1980 	struct kvm_io_device *mmio_dev;
1981 	gpa_t                 gpa;
1982 
1983 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1984 
1985 	if (gpa == UNMAPPED_GVA) {
1986 		kvm_inject_page_fault(vcpu, addr, 2);
1987 		return X86EMUL_PROPAGATE_FAULT;
1988 	}
1989 
1990 	/* For APIC access vmexit */
1991 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1992 		goto mmio;
1993 
1994 	if (emulator_write_phys(vcpu, gpa, val, bytes))
1995 		return X86EMUL_CONTINUE;
1996 
1997 mmio:
1998 	/*
1999 	 * Is this MMIO handled locally?
2000 	 */
2001 	mutex_lock(&vcpu->kvm->lock);
2002 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2003 	if (mmio_dev) {
2004 		kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2005 		mutex_unlock(&vcpu->kvm->lock);
2006 		return X86EMUL_CONTINUE;
2007 	}
2008 	mutex_unlock(&vcpu->kvm->lock);
2009 
2010 	vcpu->mmio_needed = 1;
2011 	vcpu->mmio_phys_addr = gpa;
2012 	vcpu->mmio_size = bytes;
2013 	vcpu->mmio_is_write = 1;
2014 	memcpy(vcpu->mmio_data, val, bytes);
2015 
2016 	return X86EMUL_CONTINUE;
2017 }
2018 
2019 int emulator_write_emulated(unsigned long addr,
2020 				   const void *val,
2021 				   unsigned int bytes,
2022 				   struct kvm_vcpu *vcpu)
2023 {
2024 	/* Crossing a page boundary? */
2025 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2026 		int rc, now;
2027 
2028 		now = -addr & ~PAGE_MASK;
2029 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2030 		if (rc != X86EMUL_CONTINUE)
2031 			return rc;
2032 		addr += now;
2033 		val += now;
2034 		bytes -= now;
2035 	}
2036 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2037 }
2038 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2039 
2040 static int emulator_cmpxchg_emulated(unsigned long addr,
2041 				     const void *old,
2042 				     const void *new,
2043 				     unsigned int bytes,
2044 				     struct kvm_vcpu *vcpu)
2045 {
2046 	static int reported;
2047 
2048 	if (!reported) {
2049 		reported = 1;
2050 		printk(KERN_WARNING "kvm: emulating exchange as write\n");
2051 	}
2052 #ifndef CONFIG_X86_64
2053 	/* guests cmpxchg8b have to be emulated atomically */
2054 	if (bytes == 8) {
2055 		gpa_t gpa;
2056 		struct page *page;
2057 		char *kaddr;
2058 		u64 val;
2059 
2060 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2061 
2062 		if (gpa == UNMAPPED_GVA ||
2063 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2064 			goto emul_write;
2065 
2066 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2067 			goto emul_write;
2068 
2069 		val = *(u64 *)new;
2070 
2071 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2072 
2073 		kaddr = kmap_atomic(page, KM_USER0);
2074 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2075 		kunmap_atomic(kaddr, KM_USER0);
2076 		kvm_release_page_dirty(page);
2077 	}
2078 emul_write:
2079 #endif
2080 
2081 	return emulator_write_emulated(addr, new, bytes, vcpu);
2082 }
2083 
2084 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2085 {
2086 	return kvm_x86_ops->get_segment_base(vcpu, seg);
2087 }
2088 
2089 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2090 {
2091 	kvm_mmu_invlpg(vcpu, address);
2092 	return X86EMUL_CONTINUE;
2093 }
2094 
2095 int emulate_clts(struct kvm_vcpu *vcpu)
2096 {
2097 	KVMTRACE_0D(CLTS, vcpu, handler);
2098 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2099 	return X86EMUL_CONTINUE;
2100 }
2101 
2102 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2103 {
2104 	struct kvm_vcpu *vcpu = ctxt->vcpu;
2105 
2106 	switch (dr) {
2107 	case 0 ... 3:
2108 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
2109 		return X86EMUL_CONTINUE;
2110 	default:
2111 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2112 		return X86EMUL_UNHANDLEABLE;
2113 	}
2114 }
2115 
2116 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2117 {
2118 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2119 	int exception;
2120 
2121 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2122 	if (exception) {
2123 		/* FIXME: better handling */
2124 		return X86EMUL_UNHANDLEABLE;
2125 	}
2126 	return X86EMUL_CONTINUE;
2127 }
2128 
2129 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2130 {
2131 	u8 opcodes[4];
2132 	unsigned long rip = kvm_rip_read(vcpu);
2133 	unsigned long rip_linear;
2134 
2135 	if (!printk_ratelimit())
2136 		return;
2137 
2138 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2139 
2140 	emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2141 
2142 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2143 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2144 }
2145 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2146 
2147 static struct x86_emulate_ops emulate_ops = {
2148 	.read_std            = emulator_read_std,
2149 	.read_emulated       = emulator_read_emulated,
2150 	.write_emulated      = emulator_write_emulated,
2151 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2152 };
2153 
2154 static void cache_all_regs(struct kvm_vcpu *vcpu)
2155 {
2156 	kvm_register_read(vcpu, VCPU_REGS_RAX);
2157 	kvm_register_read(vcpu, VCPU_REGS_RSP);
2158 	kvm_register_read(vcpu, VCPU_REGS_RIP);
2159 	vcpu->arch.regs_dirty = ~0;
2160 }
2161 
2162 int emulate_instruction(struct kvm_vcpu *vcpu,
2163 			struct kvm_run *run,
2164 			unsigned long cr2,
2165 			u16 error_code,
2166 			int emulation_type)
2167 {
2168 	int r;
2169 	struct decode_cache *c;
2170 
2171 	kvm_clear_exception_queue(vcpu);
2172 	vcpu->arch.mmio_fault_cr2 = cr2;
2173 	/*
2174 	 * TODO: fix x86_emulate.c to use guest_read/write_register
2175 	 * instead of direct ->regs accesses, can save hundred cycles
2176 	 * on Intel for instructions that don't read/change RSP, for
2177 	 * for example.
2178 	 */
2179 	cache_all_regs(vcpu);
2180 
2181 	vcpu->mmio_is_write = 0;
2182 	vcpu->arch.pio.string = 0;
2183 
2184 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2185 		int cs_db, cs_l;
2186 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2187 
2188 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2189 		vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2190 		vcpu->arch.emulate_ctxt.mode =
2191 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2192 			? X86EMUL_MODE_REAL : cs_l
2193 			? X86EMUL_MODE_PROT64 :	cs_db
2194 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2195 
2196 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2197 
2198 		/* Reject the instructions other than VMCALL/VMMCALL when
2199 		 * try to emulate invalid opcode */
2200 		c = &vcpu->arch.emulate_ctxt.decode;
2201 		if ((emulation_type & EMULTYPE_TRAP_UD) &&
2202 		    (!(c->twobyte && c->b == 0x01 &&
2203 		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2204 		       c->modrm_mod == 3 && c->modrm_rm == 1)))
2205 			return EMULATE_FAIL;
2206 
2207 		++vcpu->stat.insn_emulation;
2208 		if (r)  {
2209 			++vcpu->stat.insn_emulation_fail;
2210 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2211 				return EMULATE_DONE;
2212 			return EMULATE_FAIL;
2213 		}
2214 	}
2215 
2216 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2217 
2218 	if (vcpu->arch.pio.string)
2219 		return EMULATE_DO_MMIO;
2220 
2221 	if ((r || vcpu->mmio_is_write) && run) {
2222 		run->exit_reason = KVM_EXIT_MMIO;
2223 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2224 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2225 		run->mmio.len = vcpu->mmio_size;
2226 		run->mmio.is_write = vcpu->mmio_is_write;
2227 	}
2228 
2229 	if (r) {
2230 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2231 			return EMULATE_DONE;
2232 		if (!vcpu->mmio_needed) {
2233 			kvm_report_emulation_failure(vcpu, "mmio");
2234 			return EMULATE_FAIL;
2235 		}
2236 		return EMULATE_DO_MMIO;
2237 	}
2238 
2239 	kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2240 
2241 	if (vcpu->mmio_is_write) {
2242 		vcpu->mmio_needed = 0;
2243 		return EMULATE_DO_MMIO;
2244 	}
2245 
2246 	return EMULATE_DONE;
2247 }
2248 EXPORT_SYMBOL_GPL(emulate_instruction);
2249 
2250 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2251 {
2252 	int i;
2253 
2254 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2255 		if (vcpu->arch.pio.guest_pages[i]) {
2256 			kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2257 			vcpu->arch.pio.guest_pages[i] = NULL;
2258 		}
2259 }
2260 
2261 static int pio_copy_data(struct kvm_vcpu *vcpu)
2262 {
2263 	void *p = vcpu->arch.pio_data;
2264 	void *q;
2265 	unsigned bytes;
2266 	int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2267 
2268 	q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2269 		 PAGE_KERNEL);
2270 	if (!q) {
2271 		free_pio_guest_pages(vcpu);
2272 		return -ENOMEM;
2273 	}
2274 	q += vcpu->arch.pio.guest_page_offset;
2275 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2276 	if (vcpu->arch.pio.in)
2277 		memcpy(q, p, bytes);
2278 	else
2279 		memcpy(p, q, bytes);
2280 	q -= vcpu->arch.pio.guest_page_offset;
2281 	vunmap(q);
2282 	free_pio_guest_pages(vcpu);
2283 	return 0;
2284 }
2285 
2286 int complete_pio(struct kvm_vcpu *vcpu)
2287 {
2288 	struct kvm_pio_request *io = &vcpu->arch.pio;
2289 	long delta;
2290 	int r;
2291 	unsigned long val;
2292 
2293 	if (!io->string) {
2294 		if (io->in) {
2295 			val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2296 			memcpy(&val, vcpu->arch.pio_data, io->size);
2297 			kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2298 		}
2299 	} else {
2300 		if (io->in) {
2301 			r = pio_copy_data(vcpu);
2302 			if (r)
2303 				return r;
2304 		}
2305 
2306 		delta = 1;
2307 		if (io->rep) {
2308 			delta *= io->cur_count;
2309 			/*
2310 			 * The size of the register should really depend on
2311 			 * current address size.
2312 			 */
2313 			val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2314 			val -= delta;
2315 			kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2316 		}
2317 		if (io->down)
2318 			delta = -delta;
2319 		delta *= io->size;
2320 		if (io->in) {
2321 			val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2322 			val += delta;
2323 			kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2324 		} else {
2325 			val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2326 			val += delta;
2327 			kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2328 		}
2329 	}
2330 
2331 	io->count -= io->cur_count;
2332 	io->cur_count = 0;
2333 
2334 	return 0;
2335 }
2336 
2337 static void kernel_pio(struct kvm_io_device *pio_dev,
2338 		       struct kvm_vcpu *vcpu,
2339 		       void *pd)
2340 {
2341 	/* TODO: String I/O for in kernel device */
2342 
2343 	mutex_lock(&vcpu->kvm->lock);
2344 	if (vcpu->arch.pio.in)
2345 		kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2346 				  vcpu->arch.pio.size,
2347 				  pd);
2348 	else
2349 		kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2350 				   vcpu->arch.pio.size,
2351 				   pd);
2352 	mutex_unlock(&vcpu->kvm->lock);
2353 }
2354 
2355 static void pio_string_write(struct kvm_io_device *pio_dev,
2356 			     struct kvm_vcpu *vcpu)
2357 {
2358 	struct kvm_pio_request *io = &vcpu->arch.pio;
2359 	void *pd = vcpu->arch.pio_data;
2360 	int i;
2361 
2362 	mutex_lock(&vcpu->kvm->lock);
2363 	for (i = 0; i < io->cur_count; i++) {
2364 		kvm_iodevice_write(pio_dev, io->port,
2365 				   io->size,
2366 				   pd);
2367 		pd += io->size;
2368 	}
2369 	mutex_unlock(&vcpu->kvm->lock);
2370 }
2371 
2372 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2373 					       gpa_t addr, int len,
2374 					       int is_write)
2375 {
2376 	return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2377 }
2378 
2379 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2380 		  int size, unsigned port)
2381 {
2382 	struct kvm_io_device *pio_dev;
2383 	unsigned long val;
2384 
2385 	vcpu->run->exit_reason = KVM_EXIT_IO;
2386 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2387 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2388 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2389 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2390 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2391 	vcpu->arch.pio.in = in;
2392 	vcpu->arch.pio.string = 0;
2393 	vcpu->arch.pio.down = 0;
2394 	vcpu->arch.pio.guest_page_offset = 0;
2395 	vcpu->arch.pio.rep = 0;
2396 
2397 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2398 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2399 			    handler);
2400 	else
2401 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2402 			    handler);
2403 
2404 	val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2405 	memcpy(vcpu->arch.pio_data, &val, 4);
2406 
2407 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2408 
2409 	pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2410 	if (pio_dev) {
2411 		kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2412 		complete_pio(vcpu);
2413 		return 1;
2414 	}
2415 	return 0;
2416 }
2417 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2418 
2419 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2420 		  int size, unsigned long count, int down,
2421 		  gva_t address, int rep, unsigned port)
2422 {
2423 	unsigned now, in_page;
2424 	int i, ret = 0;
2425 	int nr_pages = 1;
2426 	struct page *page;
2427 	struct kvm_io_device *pio_dev;
2428 
2429 	vcpu->run->exit_reason = KVM_EXIT_IO;
2430 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2431 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2432 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2433 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2434 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2435 	vcpu->arch.pio.in = in;
2436 	vcpu->arch.pio.string = 1;
2437 	vcpu->arch.pio.down = down;
2438 	vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2439 	vcpu->arch.pio.rep = rep;
2440 
2441 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2442 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2443 			    handler);
2444 	else
2445 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2446 			    handler);
2447 
2448 	if (!count) {
2449 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2450 		return 1;
2451 	}
2452 
2453 	if (!down)
2454 		in_page = PAGE_SIZE - offset_in_page(address);
2455 	else
2456 		in_page = offset_in_page(address) + size;
2457 	now = min(count, (unsigned long)in_page / size);
2458 	if (!now) {
2459 		/*
2460 		 * String I/O straddles page boundary.  Pin two guest pages
2461 		 * so that we satisfy atomicity constraints.  Do just one
2462 		 * transaction to avoid complexity.
2463 		 */
2464 		nr_pages = 2;
2465 		now = 1;
2466 	}
2467 	if (down) {
2468 		/*
2469 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2470 		 */
2471 		pr_unimpl(vcpu, "guest string pio down\n");
2472 		kvm_inject_gp(vcpu, 0);
2473 		return 1;
2474 	}
2475 	vcpu->run->io.count = now;
2476 	vcpu->arch.pio.cur_count = now;
2477 
2478 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2479 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2480 
2481 	for (i = 0; i < nr_pages; ++i) {
2482 		page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2483 		vcpu->arch.pio.guest_pages[i] = page;
2484 		if (!page) {
2485 			kvm_inject_gp(vcpu, 0);
2486 			free_pio_guest_pages(vcpu);
2487 			return 1;
2488 		}
2489 	}
2490 
2491 	pio_dev = vcpu_find_pio_dev(vcpu, port,
2492 				    vcpu->arch.pio.cur_count,
2493 				    !vcpu->arch.pio.in);
2494 	if (!vcpu->arch.pio.in) {
2495 		/* string PIO write */
2496 		ret = pio_copy_data(vcpu);
2497 		if (ret >= 0 && pio_dev) {
2498 			pio_string_write(pio_dev, vcpu);
2499 			complete_pio(vcpu);
2500 			if (vcpu->arch.pio.count == 0)
2501 				ret = 1;
2502 		}
2503 	} else if (pio_dev)
2504 		pr_unimpl(vcpu, "no string pio read support yet, "
2505 		       "port %x size %d count %ld\n",
2506 			port, size, count);
2507 
2508 	return ret;
2509 }
2510 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2511 
2512 int kvm_arch_init(void *opaque)
2513 {
2514 	int r;
2515 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2516 
2517 	if (kvm_x86_ops) {
2518 		printk(KERN_ERR "kvm: already loaded the other module\n");
2519 		r = -EEXIST;
2520 		goto out;
2521 	}
2522 
2523 	if (!ops->cpu_has_kvm_support()) {
2524 		printk(KERN_ERR "kvm: no hardware support\n");
2525 		r = -EOPNOTSUPP;
2526 		goto out;
2527 	}
2528 	if (ops->disabled_by_bios()) {
2529 		printk(KERN_ERR "kvm: disabled by bios\n");
2530 		r = -EOPNOTSUPP;
2531 		goto out;
2532 	}
2533 
2534 	r = kvm_mmu_module_init();
2535 	if (r)
2536 		goto out;
2537 
2538 	kvm_init_msr_list();
2539 
2540 	kvm_x86_ops = ops;
2541 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2542 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2543 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2544 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
2545 	return 0;
2546 
2547 out:
2548 	return r;
2549 }
2550 
2551 void kvm_arch_exit(void)
2552 {
2553 	kvm_x86_ops = NULL;
2554 	kvm_mmu_module_exit();
2555 }
2556 
2557 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2558 {
2559 	++vcpu->stat.halt_exits;
2560 	KVMTRACE_0D(HLT, vcpu, handler);
2561 	if (irqchip_in_kernel(vcpu->kvm)) {
2562 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2563 		return 1;
2564 	} else {
2565 		vcpu->run->exit_reason = KVM_EXIT_HLT;
2566 		return 0;
2567 	}
2568 }
2569 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2570 
2571 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2572 			   unsigned long a1)
2573 {
2574 	if (is_long_mode(vcpu))
2575 		return a0;
2576 	else
2577 		return a0 | ((gpa_t)a1 << 32);
2578 }
2579 
2580 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2581 {
2582 	unsigned long nr, a0, a1, a2, a3, ret;
2583 	int r = 1;
2584 
2585 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2586 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2587 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2588 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2589 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2590 
2591 	KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2592 
2593 	if (!is_long_mode(vcpu)) {
2594 		nr &= 0xFFFFFFFF;
2595 		a0 &= 0xFFFFFFFF;
2596 		a1 &= 0xFFFFFFFF;
2597 		a2 &= 0xFFFFFFFF;
2598 		a3 &= 0xFFFFFFFF;
2599 	}
2600 
2601 	switch (nr) {
2602 	case KVM_HC_VAPIC_POLL_IRQ:
2603 		ret = 0;
2604 		break;
2605 	case KVM_HC_MMU_OP:
2606 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2607 		break;
2608 	default:
2609 		ret = -KVM_ENOSYS;
2610 		break;
2611 	}
2612 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2613 	++vcpu->stat.hypercalls;
2614 	return r;
2615 }
2616 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2617 
2618 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2619 {
2620 	char instruction[3];
2621 	int ret = 0;
2622 	unsigned long rip = kvm_rip_read(vcpu);
2623 
2624 
2625 	/*
2626 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
2627 	 * to ensure that the updated hypercall appears atomically across all
2628 	 * VCPUs.
2629 	 */
2630 	kvm_mmu_zap_all(vcpu->kvm);
2631 
2632 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
2633 	if (emulator_write_emulated(rip, instruction, 3, vcpu)
2634 	    != X86EMUL_CONTINUE)
2635 		ret = -EFAULT;
2636 
2637 	return ret;
2638 }
2639 
2640 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2641 {
2642 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2643 }
2644 
2645 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2646 {
2647 	struct descriptor_table dt = { limit, base };
2648 
2649 	kvm_x86_ops->set_gdt(vcpu, &dt);
2650 }
2651 
2652 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2653 {
2654 	struct descriptor_table dt = { limit, base };
2655 
2656 	kvm_x86_ops->set_idt(vcpu, &dt);
2657 }
2658 
2659 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2660 		   unsigned long *rflags)
2661 {
2662 	kvm_lmsw(vcpu, msw);
2663 	*rflags = kvm_x86_ops->get_rflags(vcpu);
2664 }
2665 
2666 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2667 {
2668 	unsigned long value;
2669 
2670 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2671 	switch (cr) {
2672 	case 0:
2673 		value = vcpu->arch.cr0;
2674 		break;
2675 	case 2:
2676 		value = vcpu->arch.cr2;
2677 		break;
2678 	case 3:
2679 		value = vcpu->arch.cr3;
2680 		break;
2681 	case 4:
2682 		value = vcpu->arch.cr4;
2683 		break;
2684 	case 8:
2685 		value = kvm_get_cr8(vcpu);
2686 		break;
2687 	default:
2688 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2689 		return 0;
2690 	}
2691 	KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2692 		    (u32)((u64)value >> 32), handler);
2693 
2694 	return value;
2695 }
2696 
2697 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2698 		     unsigned long *rflags)
2699 {
2700 	KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2701 		    (u32)((u64)val >> 32), handler);
2702 
2703 	switch (cr) {
2704 	case 0:
2705 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2706 		*rflags = kvm_x86_ops->get_rflags(vcpu);
2707 		break;
2708 	case 2:
2709 		vcpu->arch.cr2 = val;
2710 		break;
2711 	case 3:
2712 		kvm_set_cr3(vcpu, val);
2713 		break;
2714 	case 4:
2715 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2716 		break;
2717 	case 8:
2718 		kvm_set_cr8(vcpu, val & 0xfUL);
2719 		break;
2720 	default:
2721 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2722 	}
2723 }
2724 
2725 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2726 {
2727 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2728 	int j, nent = vcpu->arch.cpuid_nent;
2729 
2730 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2731 	/* when no next entry is found, the current entry[i] is reselected */
2732 	for (j = i + 1; j == i; j = (j + 1) % nent) {
2733 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2734 		if (ej->function == e->function) {
2735 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2736 			return j;
2737 		}
2738 	}
2739 	return 0; /* silence gcc, even though control never reaches here */
2740 }
2741 
2742 /* find an entry with matching function, matching index (if needed), and that
2743  * should be read next (if it's stateful) */
2744 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2745 	u32 function, u32 index)
2746 {
2747 	if (e->function != function)
2748 		return 0;
2749 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2750 		return 0;
2751 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2752 		!(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2753 		return 0;
2754 	return 1;
2755 }
2756 
2757 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2758 {
2759 	int i;
2760 	u32 function, index;
2761 	struct kvm_cpuid_entry2 *e, *best;
2762 
2763 	function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2764 	index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2765 	kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2766 	kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2767 	kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2768 	kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2769 	best = NULL;
2770 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2771 		e = &vcpu->arch.cpuid_entries[i];
2772 		if (is_matching_cpuid_entry(e, function, index)) {
2773 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2774 				move_to_next_stateful_cpuid_entry(vcpu, i);
2775 			best = e;
2776 			break;
2777 		}
2778 		/*
2779 		 * Both basic or both extended?
2780 		 */
2781 		if (((e->function ^ function) & 0x80000000) == 0)
2782 			if (!best || e->function > best->function)
2783 				best = e;
2784 	}
2785 	if (best) {
2786 		kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2787 		kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2788 		kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2789 		kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2790 	}
2791 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2792 	KVMTRACE_5D(CPUID, vcpu, function,
2793 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2794 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2795 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2796 		    (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2797 }
2798 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2799 
2800 /*
2801  * Check if userspace requested an interrupt window, and that the
2802  * interrupt window is open.
2803  *
2804  * No need to exit to userspace if we already have an interrupt queued.
2805  */
2806 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2807 					  struct kvm_run *kvm_run)
2808 {
2809 	return (!vcpu->arch.irq_summary &&
2810 		kvm_run->request_interrupt_window &&
2811 		vcpu->arch.interrupt_window_open &&
2812 		(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2813 }
2814 
2815 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2816 			      struct kvm_run *kvm_run)
2817 {
2818 	kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2819 	kvm_run->cr8 = kvm_get_cr8(vcpu);
2820 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
2821 	if (irqchip_in_kernel(vcpu->kvm))
2822 		kvm_run->ready_for_interrupt_injection = 1;
2823 	else
2824 		kvm_run->ready_for_interrupt_injection =
2825 					(vcpu->arch.interrupt_window_open &&
2826 					 vcpu->arch.irq_summary == 0);
2827 }
2828 
2829 static void vapic_enter(struct kvm_vcpu *vcpu)
2830 {
2831 	struct kvm_lapic *apic = vcpu->arch.apic;
2832 	struct page *page;
2833 
2834 	if (!apic || !apic->vapic_addr)
2835 		return;
2836 
2837 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2838 
2839 	vcpu->arch.apic->vapic_page = page;
2840 }
2841 
2842 static void vapic_exit(struct kvm_vcpu *vcpu)
2843 {
2844 	struct kvm_lapic *apic = vcpu->arch.apic;
2845 
2846 	if (!apic || !apic->vapic_addr)
2847 		return;
2848 
2849 	down_read(&vcpu->kvm->slots_lock);
2850 	kvm_release_page_dirty(apic->vapic_page);
2851 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2852 	up_read(&vcpu->kvm->slots_lock);
2853 }
2854 
2855 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2856 {
2857 	int r;
2858 
2859 	if (vcpu->requests)
2860 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2861 			kvm_mmu_unload(vcpu);
2862 
2863 	r = kvm_mmu_reload(vcpu);
2864 	if (unlikely(r))
2865 		goto out;
2866 
2867 	if (vcpu->requests) {
2868 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2869 			__kvm_migrate_timers(vcpu);
2870 		if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2871 			kvm_mmu_sync_roots(vcpu);
2872 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2873 			kvm_x86_ops->tlb_flush(vcpu);
2874 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2875 				       &vcpu->requests)) {
2876 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2877 			r = 0;
2878 			goto out;
2879 		}
2880 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2881 			kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2882 			r = 0;
2883 			goto out;
2884 		}
2885 	}
2886 
2887 	clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2888 	kvm_inject_pending_timer_irqs(vcpu);
2889 
2890 	preempt_disable();
2891 
2892 	kvm_x86_ops->prepare_guest_switch(vcpu);
2893 	kvm_load_guest_fpu(vcpu);
2894 
2895 	local_irq_disable();
2896 
2897 	if (vcpu->requests || need_resched() || signal_pending(current)) {
2898 		local_irq_enable();
2899 		preempt_enable();
2900 		r = 1;
2901 		goto out;
2902 	}
2903 
2904 	if (vcpu->guest_debug.enabled)
2905 		kvm_x86_ops->guest_debug_pre(vcpu);
2906 
2907 	vcpu->guest_mode = 1;
2908 	/*
2909 	 * Make sure that guest_mode assignment won't happen after
2910 	 * testing the pending IRQ vector bitmap.
2911 	 */
2912 	smp_wmb();
2913 
2914 	if (vcpu->arch.exception.pending)
2915 		__queue_exception(vcpu);
2916 	else if (irqchip_in_kernel(vcpu->kvm))
2917 		kvm_x86_ops->inject_pending_irq(vcpu);
2918 	else
2919 		kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2920 
2921 	kvm_lapic_sync_to_vapic(vcpu);
2922 
2923 	up_read(&vcpu->kvm->slots_lock);
2924 
2925 	kvm_guest_enter();
2926 
2927 
2928 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2929 	kvm_x86_ops->run(vcpu, kvm_run);
2930 
2931 	vcpu->guest_mode = 0;
2932 	local_irq_enable();
2933 
2934 	++vcpu->stat.exits;
2935 
2936 	/*
2937 	 * We must have an instruction between local_irq_enable() and
2938 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2939 	 * the interrupt shadow.  The stat.exits increment will do nicely.
2940 	 * But we need to prevent reordering, hence this barrier():
2941 	 */
2942 	barrier();
2943 
2944 	kvm_guest_exit();
2945 
2946 	preempt_enable();
2947 
2948 	down_read(&vcpu->kvm->slots_lock);
2949 
2950 	/*
2951 	 * Profile KVM exit RIPs:
2952 	 */
2953 	if (unlikely(prof_on == KVM_PROFILING)) {
2954 		unsigned long rip = kvm_rip_read(vcpu);
2955 		profile_hit(KVM_PROFILING, (void *)rip);
2956 	}
2957 
2958 	if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2959 		vcpu->arch.exception.pending = false;
2960 
2961 	kvm_lapic_sync_from_vapic(vcpu);
2962 
2963 	r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2964 out:
2965 	return r;
2966 }
2967 
2968 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2969 {
2970 	int r;
2971 
2972 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2973 		pr_debug("vcpu %d received sipi with vector # %x\n",
2974 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2975 		kvm_lapic_reset(vcpu);
2976 		r = kvm_x86_ops->vcpu_reset(vcpu);
2977 		if (r)
2978 			return r;
2979 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2980 	}
2981 
2982 	down_read(&vcpu->kvm->slots_lock);
2983 	vapic_enter(vcpu);
2984 
2985 	r = 1;
2986 	while (r > 0) {
2987 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
2988 			r = vcpu_enter_guest(vcpu, kvm_run);
2989 		else {
2990 			up_read(&vcpu->kvm->slots_lock);
2991 			kvm_vcpu_block(vcpu);
2992 			down_read(&vcpu->kvm->slots_lock);
2993 			if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
2994 				if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
2995 					vcpu->arch.mp_state =
2996 							KVM_MP_STATE_RUNNABLE;
2997 			if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2998 				r = -EINTR;
2999 		}
3000 
3001 		if (r > 0) {
3002 			if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3003 				r = -EINTR;
3004 				kvm_run->exit_reason = KVM_EXIT_INTR;
3005 				++vcpu->stat.request_irq_exits;
3006 			}
3007 			if (signal_pending(current)) {
3008 				r = -EINTR;
3009 				kvm_run->exit_reason = KVM_EXIT_INTR;
3010 				++vcpu->stat.signal_exits;
3011 			}
3012 			if (need_resched()) {
3013 				up_read(&vcpu->kvm->slots_lock);
3014 				kvm_resched(vcpu);
3015 				down_read(&vcpu->kvm->slots_lock);
3016 			}
3017 		}
3018 	}
3019 
3020 	up_read(&vcpu->kvm->slots_lock);
3021 	post_kvm_run_save(vcpu, kvm_run);
3022 
3023 	vapic_exit(vcpu);
3024 
3025 	return r;
3026 }
3027 
3028 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3029 {
3030 	int r;
3031 	sigset_t sigsaved;
3032 
3033 	vcpu_load(vcpu);
3034 
3035 	if (vcpu->sigset_active)
3036 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3037 
3038 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3039 		kvm_vcpu_block(vcpu);
3040 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3041 		r = -EAGAIN;
3042 		goto out;
3043 	}
3044 
3045 	/* re-sync apic's tpr */
3046 	if (!irqchip_in_kernel(vcpu->kvm))
3047 		kvm_set_cr8(vcpu, kvm_run->cr8);
3048 
3049 	if (vcpu->arch.pio.cur_count) {
3050 		r = complete_pio(vcpu);
3051 		if (r)
3052 			goto out;
3053 	}
3054 #if CONFIG_HAS_IOMEM
3055 	if (vcpu->mmio_needed) {
3056 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3057 		vcpu->mmio_read_completed = 1;
3058 		vcpu->mmio_needed = 0;
3059 
3060 		down_read(&vcpu->kvm->slots_lock);
3061 		r = emulate_instruction(vcpu, kvm_run,
3062 					vcpu->arch.mmio_fault_cr2, 0,
3063 					EMULTYPE_NO_DECODE);
3064 		up_read(&vcpu->kvm->slots_lock);
3065 		if (r == EMULATE_DO_MMIO) {
3066 			/*
3067 			 * Read-modify-write.  Back to userspace.
3068 			 */
3069 			r = 0;
3070 			goto out;
3071 		}
3072 	}
3073 #endif
3074 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3075 		kvm_register_write(vcpu, VCPU_REGS_RAX,
3076 				     kvm_run->hypercall.ret);
3077 
3078 	r = __vcpu_run(vcpu, kvm_run);
3079 
3080 out:
3081 	if (vcpu->sigset_active)
3082 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3083 
3084 	vcpu_put(vcpu);
3085 	return r;
3086 }
3087 
3088 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3089 {
3090 	vcpu_load(vcpu);
3091 
3092 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3093 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3094 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3095 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3096 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3097 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3098 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3099 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3100 #ifdef CONFIG_X86_64
3101 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3102 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3103 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3104 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3105 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3106 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3107 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3108 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3109 #endif
3110 
3111 	regs->rip = kvm_rip_read(vcpu);
3112 	regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3113 
3114 	/*
3115 	 * Don't leak debug flags in case they were set for guest debugging
3116 	 */
3117 	if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3118 		regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3119 
3120 	vcpu_put(vcpu);
3121 
3122 	return 0;
3123 }
3124 
3125 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3126 {
3127 	vcpu_load(vcpu);
3128 
3129 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3130 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3131 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3132 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3133 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3134 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3135 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3136 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3137 #ifdef CONFIG_X86_64
3138 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3139 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3140 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3141 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3142 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3143 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3144 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3145 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3146 
3147 #endif
3148 
3149 	kvm_rip_write(vcpu, regs->rip);
3150 	kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3151 
3152 
3153 	vcpu->arch.exception.pending = false;
3154 
3155 	vcpu_put(vcpu);
3156 
3157 	return 0;
3158 }
3159 
3160 void kvm_get_segment(struct kvm_vcpu *vcpu,
3161 		     struct kvm_segment *var, int seg)
3162 {
3163 	kvm_x86_ops->get_segment(vcpu, var, seg);
3164 }
3165 
3166 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3167 {
3168 	struct kvm_segment cs;
3169 
3170 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3171 	*db = cs.db;
3172 	*l = cs.l;
3173 }
3174 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3175 
3176 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3177 				  struct kvm_sregs *sregs)
3178 {
3179 	struct descriptor_table dt;
3180 	int pending_vec;
3181 
3182 	vcpu_load(vcpu);
3183 
3184 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3185 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3186 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3187 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3188 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3189 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3190 
3191 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3192 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3193 
3194 	kvm_x86_ops->get_idt(vcpu, &dt);
3195 	sregs->idt.limit = dt.limit;
3196 	sregs->idt.base = dt.base;
3197 	kvm_x86_ops->get_gdt(vcpu, &dt);
3198 	sregs->gdt.limit = dt.limit;
3199 	sregs->gdt.base = dt.base;
3200 
3201 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3202 	sregs->cr0 = vcpu->arch.cr0;
3203 	sregs->cr2 = vcpu->arch.cr2;
3204 	sregs->cr3 = vcpu->arch.cr3;
3205 	sregs->cr4 = vcpu->arch.cr4;
3206 	sregs->cr8 = kvm_get_cr8(vcpu);
3207 	sregs->efer = vcpu->arch.shadow_efer;
3208 	sregs->apic_base = kvm_get_apic_base(vcpu);
3209 
3210 	if (irqchip_in_kernel(vcpu->kvm)) {
3211 		memset(sregs->interrupt_bitmap, 0,
3212 		       sizeof sregs->interrupt_bitmap);
3213 		pending_vec = kvm_x86_ops->get_irq(vcpu);
3214 		if (pending_vec >= 0)
3215 			set_bit(pending_vec,
3216 				(unsigned long *)sregs->interrupt_bitmap);
3217 	} else
3218 		memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3219 		       sizeof sregs->interrupt_bitmap);
3220 
3221 	vcpu_put(vcpu);
3222 
3223 	return 0;
3224 }
3225 
3226 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3227 				    struct kvm_mp_state *mp_state)
3228 {
3229 	vcpu_load(vcpu);
3230 	mp_state->mp_state = vcpu->arch.mp_state;
3231 	vcpu_put(vcpu);
3232 	return 0;
3233 }
3234 
3235 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3236 				    struct kvm_mp_state *mp_state)
3237 {
3238 	vcpu_load(vcpu);
3239 	vcpu->arch.mp_state = mp_state->mp_state;
3240 	vcpu_put(vcpu);
3241 	return 0;
3242 }
3243 
3244 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3245 			struct kvm_segment *var, int seg)
3246 {
3247 	kvm_x86_ops->set_segment(vcpu, var, seg);
3248 }
3249 
3250 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3251 				   struct kvm_segment *kvm_desct)
3252 {
3253 	kvm_desct->base = seg_desc->base0;
3254 	kvm_desct->base |= seg_desc->base1 << 16;
3255 	kvm_desct->base |= seg_desc->base2 << 24;
3256 	kvm_desct->limit = seg_desc->limit0;
3257 	kvm_desct->limit |= seg_desc->limit << 16;
3258 	if (seg_desc->g) {
3259 		kvm_desct->limit <<= 12;
3260 		kvm_desct->limit |= 0xfff;
3261 	}
3262 	kvm_desct->selector = selector;
3263 	kvm_desct->type = seg_desc->type;
3264 	kvm_desct->present = seg_desc->p;
3265 	kvm_desct->dpl = seg_desc->dpl;
3266 	kvm_desct->db = seg_desc->d;
3267 	kvm_desct->s = seg_desc->s;
3268 	kvm_desct->l = seg_desc->l;
3269 	kvm_desct->g = seg_desc->g;
3270 	kvm_desct->avl = seg_desc->avl;
3271 	if (!selector)
3272 		kvm_desct->unusable = 1;
3273 	else
3274 		kvm_desct->unusable = 0;
3275 	kvm_desct->padding = 0;
3276 }
3277 
3278 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3279 					   u16 selector,
3280 					   struct descriptor_table *dtable)
3281 {
3282 	if (selector & 1 << 2) {
3283 		struct kvm_segment kvm_seg;
3284 
3285 		kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3286 
3287 		if (kvm_seg.unusable)
3288 			dtable->limit = 0;
3289 		else
3290 			dtable->limit = kvm_seg.limit;
3291 		dtable->base = kvm_seg.base;
3292 	}
3293 	else
3294 		kvm_x86_ops->get_gdt(vcpu, dtable);
3295 }
3296 
3297 /* allowed just for 8 bytes segments */
3298 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3299 					 struct desc_struct *seg_desc)
3300 {
3301 	gpa_t gpa;
3302 	struct descriptor_table dtable;
3303 	u16 index = selector >> 3;
3304 
3305 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3306 
3307 	if (dtable.limit < index * 8 + 7) {
3308 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3309 		return 1;
3310 	}
3311 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3312 	gpa += index * 8;
3313 	return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3314 }
3315 
3316 /* allowed just for 8 bytes segments */
3317 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3318 					 struct desc_struct *seg_desc)
3319 {
3320 	gpa_t gpa;
3321 	struct descriptor_table dtable;
3322 	u16 index = selector >> 3;
3323 
3324 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3325 
3326 	if (dtable.limit < index * 8 + 7)
3327 		return 1;
3328 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3329 	gpa += index * 8;
3330 	return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3331 }
3332 
3333 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3334 			     struct desc_struct *seg_desc)
3335 {
3336 	u32 base_addr;
3337 
3338 	base_addr = seg_desc->base0;
3339 	base_addr |= (seg_desc->base1 << 16);
3340 	base_addr |= (seg_desc->base2 << 24);
3341 
3342 	return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3343 }
3344 
3345 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3346 {
3347 	struct kvm_segment kvm_seg;
3348 
3349 	kvm_get_segment(vcpu, &kvm_seg, seg);
3350 	return kvm_seg.selector;
3351 }
3352 
3353 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3354 						u16 selector,
3355 						struct kvm_segment *kvm_seg)
3356 {
3357 	struct desc_struct seg_desc;
3358 
3359 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3360 		return 1;
3361 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3362 	return 0;
3363 }
3364 
3365 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3366 {
3367 	struct kvm_segment segvar = {
3368 		.base = selector << 4,
3369 		.limit = 0xffff,
3370 		.selector = selector,
3371 		.type = 3,
3372 		.present = 1,
3373 		.dpl = 3,
3374 		.db = 0,
3375 		.s = 1,
3376 		.l = 0,
3377 		.g = 0,
3378 		.avl = 0,
3379 		.unusable = 0,
3380 	};
3381 	kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3382 	return 0;
3383 }
3384 
3385 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3386 				int type_bits, int seg)
3387 {
3388 	struct kvm_segment kvm_seg;
3389 
3390 	if (!(vcpu->arch.cr0 & X86_CR0_PE))
3391 		return kvm_load_realmode_segment(vcpu, selector, seg);
3392 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3393 		return 1;
3394 	kvm_seg.type |= type_bits;
3395 
3396 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3397 	    seg != VCPU_SREG_LDTR)
3398 		if (!kvm_seg.s)
3399 			kvm_seg.unusable = 1;
3400 
3401 	kvm_set_segment(vcpu, &kvm_seg, seg);
3402 	return 0;
3403 }
3404 
3405 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3406 				struct tss_segment_32 *tss)
3407 {
3408 	tss->cr3 = vcpu->arch.cr3;
3409 	tss->eip = kvm_rip_read(vcpu);
3410 	tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3411 	tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3412 	tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3413 	tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3414 	tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3415 	tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3416 	tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3417 	tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3418 	tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3419 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3420 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3421 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3422 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3423 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3424 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3425 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3426 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3427 }
3428 
3429 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3430 				  struct tss_segment_32 *tss)
3431 {
3432 	kvm_set_cr3(vcpu, tss->cr3);
3433 
3434 	kvm_rip_write(vcpu, tss->eip);
3435 	kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3436 
3437 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3438 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3439 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3440 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3441 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3442 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3443 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3444 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3445 
3446 	if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3447 		return 1;
3448 
3449 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3450 		return 1;
3451 
3452 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3453 		return 1;
3454 
3455 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3456 		return 1;
3457 
3458 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3459 		return 1;
3460 
3461 	if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3462 		return 1;
3463 
3464 	if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3465 		return 1;
3466 	return 0;
3467 }
3468 
3469 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3470 				struct tss_segment_16 *tss)
3471 {
3472 	tss->ip = kvm_rip_read(vcpu);
3473 	tss->flag = kvm_x86_ops->get_rflags(vcpu);
3474 	tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3475 	tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3476 	tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3477 	tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3478 	tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3479 	tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3480 	tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3481 	tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3482 
3483 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3484 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3485 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3486 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3487 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3488 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3489 }
3490 
3491 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3492 				 struct tss_segment_16 *tss)
3493 {
3494 	kvm_rip_write(vcpu, tss->ip);
3495 	kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3496 	kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3497 	kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3498 	kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3499 	kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3500 	kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3501 	kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3502 	kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3503 	kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3504 
3505 	if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3506 		return 1;
3507 
3508 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3509 		return 1;
3510 
3511 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3512 		return 1;
3513 
3514 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3515 		return 1;
3516 
3517 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3518 		return 1;
3519 	return 0;
3520 }
3521 
3522 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3523 		       u32 old_tss_base,
3524 		       struct desc_struct *nseg_desc)
3525 {
3526 	struct tss_segment_16 tss_segment_16;
3527 	int ret = 0;
3528 
3529 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3530 			   sizeof tss_segment_16))
3531 		goto out;
3532 
3533 	save_state_to_tss16(vcpu, &tss_segment_16);
3534 
3535 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3536 			    sizeof tss_segment_16))
3537 		goto out;
3538 
3539 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3540 			   &tss_segment_16, sizeof tss_segment_16))
3541 		goto out;
3542 
3543 	if (load_state_from_tss16(vcpu, &tss_segment_16))
3544 		goto out;
3545 
3546 	ret = 1;
3547 out:
3548 	return ret;
3549 }
3550 
3551 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3552 		       u32 old_tss_base,
3553 		       struct desc_struct *nseg_desc)
3554 {
3555 	struct tss_segment_32 tss_segment_32;
3556 	int ret = 0;
3557 
3558 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3559 			   sizeof tss_segment_32))
3560 		goto out;
3561 
3562 	save_state_to_tss32(vcpu, &tss_segment_32);
3563 
3564 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3565 			    sizeof tss_segment_32))
3566 		goto out;
3567 
3568 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3569 			   &tss_segment_32, sizeof tss_segment_32))
3570 		goto out;
3571 
3572 	if (load_state_from_tss32(vcpu, &tss_segment_32))
3573 		goto out;
3574 
3575 	ret = 1;
3576 out:
3577 	return ret;
3578 }
3579 
3580 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3581 {
3582 	struct kvm_segment tr_seg;
3583 	struct desc_struct cseg_desc;
3584 	struct desc_struct nseg_desc;
3585 	int ret = 0;
3586 	u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3587 	u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3588 
3589 	old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3590 
3591 	/* FIXME: Handle errors. Failure to read either TSS or their
3592 	 * descriptors should generate a pagefault.
3593 	 */
3594 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3595 		goto out;
3596 
3597 	if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3598 		goto out;
3599 
3600 	if (reason != TASK_SWITCH_IRET) {
3601 		int cpl;
3602 
3603 		cpl = kvm_x86_ops->get_cpl(vcpu);
3604 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3605 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3606 			return 1;
3607 		}
3608 	}
3609 
3610 	if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3611 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3612 		return 1;
3613 	}
3614 
3615 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3616 		cseg_desc.type &= ~(1 << 1); //clear the B flag
3617 		save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3618 	}
3619 
3620 	if (reason == TASK_SWITCH_IRET) {
3621 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3622 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3623 	}
3624 
3625 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3626 
3627 	if (nseg_desc.type & 8)
3628 		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3629 					 &nseg_desc);
3630 	else
3631 		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3632 					 &nseg_desc);
3633 
3634 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3635 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3636 		kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3637 	}
3638 
3639 	if (reason != TASK_SWITCH_IRET) {
3640 		nseg_desc.type |= (1 << 1);
3641 		save_guest_segment_descriptor(vcpu, tss_selector,
3642 					      &nseg_desc);
3643 	}
3644 
3645 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3646 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3647 	tr_seg.type = 11;
3648 	kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3649 out:
3650 	return ret;
3651 }
3652 EXPORT_SYMBOL_GPL(kvm_task_switch);
3653 
3654 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3655 				  struct kvm_sregs *sregs)
3656 {
3657 	int mmu_reset_needed = 0;
3658 	int i, pending_vec, max_bits;
3659 	struct descriptor_table dt;
3660 
3661 	vcpu_load(vcpu);
3662 
3663 	dt.limit = sregs->idt.limit;
3664 	dt.base = sregs->idt.base;
3665 	kvm_x86_ops->set_idt(vcpu, &dt);
3666 	dt.limit = sregs->gdt.limit;
3667 	dt.base = sregs->gdt.base;
3668 	kvm_x86_ops->set_gdt(vcpu, &dt);
3669 
3670 	vcpu->arch.cr2 = sregs->cr2;
3671 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3672 	vcpu->arch.cr3 = sregs->cr3;
3673 
3674 	kvm_set_cr8(vcpu, sregs->cr8);
3675 
3676 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3677 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
3678 	kvm_set_apic_base(vcpu, sregs->apic_base);
3679 
3680 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3681 
3682 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3683 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3684 	vcpu->arch.cr0 = sregs->cr0;
3685 
3686 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3687 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3688 	if (!is_long_mode(vcpu) && is_pae(vcpu))
3689 		load_pdptrs(vcpu, vcpu->arch.cr3);
3690 
3691 	if (mmu_reset_needed)
3692 		kvm_mmu_reset_context(vcpu);
3693 
3694 	if (!irqchip_in_kernel(vcpu->kvm)) {
3695 		memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3696 		       sizeof vcpu->arch.irq_pending);
3697 		vcpu->arch.irq_summary = 0;
3698 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3699 			if (vcpu->arch.irq_pending[i])
3700 				__set_bit(i, &vcpu->arch.irq_summary);
3701 	} else {
3702 		max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3703 		pending_vec = find_first_bit(
3704 			(const unsigned long *)sregs->interrupt_bitmap,
3705 			max_bits);
3706 		/* Only pending external irq is handled here */
3707 		if (pending_vec < max_bits) {
3708 			kvm_x86_ops->set_irq(vcpu, pending_vec);
3709 			pr_debug("Set back pending irq %d\n",
3710 				 pending_vec);
3711 		}
3712 		kvm_pic_clear_isr_ack(vcpu->kvm);
3713 	}
3714 
3715 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3716 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3717 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3718 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3719 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3720 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3721 
3722 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3723 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3724 
3725 	/* Older userspace won't unhalt the vcpu on reset. */
3726 	if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3727 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3728 	    !(vcpu->arch.cr0 & X86_CR0_PE))
3729 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3730 
3731 	vcpu_put(vcpu);
3732 
3733 	return 0;
3734 }
3735 
3736 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3737 				    struct kvm_debug_guest *dbg)
3738 {
3739 	int r;
3740 
3741 	vcpu_load(vcpu);
3742 
3743 	r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3744 
3745 	vcpu_put(vcpu);
3746 
3747 	return r;
3748 }
3749 
3750 /*
3751  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3752  * we have asm/x86/processor.h
3753  */
3754 struct fxsave {
3755 	u16	cwd;
3756 	u16	swd;
3757 	u16	twd;
3758 	u16	fop;
3759 	u64	rip;
3760 	u64	rdp;
3761 	u32	mxcsr;
3762 	u32	mxcsr_mask;
3763 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
3764 #ifdef CONFIG_X86_64
3765 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
3766 #else
3767 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
3768 #endif
3769 };
3770 
3771 /*
3772  * Translate a guest virtual address to a guest physical address.
3773  */
3774 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3775 				    struct kvm_translation *tr)
3776 {
3777 	unsigned long vaddr = tr->linear_address;
3778 	gpa_t gpa;
3779 
3780 	vcpu_load(vcpu);
3781 	down_read(&vcpu->kvm->slots_lock);
3782 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3783 	up_read(&vcpu->kvm->slots_lock);
3784 	tr->physical_address = gpa;
3785 	tr->valid = gpa != UNMAPPED_GVA;
3786 	tr->writeable = 1;
3787 	tr->usermode = 0;
3788 	vcpu_put(vcpu);
3789 
3790 	return 0;
3791 }
3792 
3793 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3794 {
3795 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3796 
3797 	vcpu_load(vcpu);
3798 
3799 	memcpy(fpu->fpr, fxsave->st_space, 128);
3800 	fpu->fcw = fxsave->cwd;
3801 	fpu->fsw = fxsave->swd;
3802 	fpu->ftwx = fxsave->twd;
3803 	fpu->last_opcode = fxsave->fop;
3804 	fpu->last_ip = fxsave->rip;
3805 	fpu->last_dp = fxsave->rdp;
3806 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3807 
3808 	vcpu_put(vcpu);
3809 
3810 	return 0;
3811 }
3812 
3813 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3814 {
3815 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3816 
3817 	vcpu_load(vcpu);
3818 
3819 	memcpy(fxsave->st_space, fpu->fpr, 128);
3820 	fxsave->cwd = fpu->fcw;
3821 	fxsave->swd = fpu->fsw;
3822 	fxsave->twd = fpu->ftwx;
3823 	fxsave->fop = fpu->last_opcode;
3824 	fxsave->rip = fpu->last_ip;
3825 	fxsave->rdp = fpu->last_dp;
3826 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3827 
3828 	vcpu_put(vcpu);
3829 
3830 	return 0;
3831 }
3832 
3833 void fx_init(struct kvm_vcpu *vcpu)
3834 {
3835 	unsigned after_mxcsr_mask;
3836 
3837 	/*
3838 	 * Touch the fpu the first time in non atomic context as if
3839 	 * this is the first fpu instruction the exception handler
3840 	 * will fire before the instruction returns and it'll have to
3841 	 * allocate ram with GFP_KERNEL.
3842 	 */
3843 	if (!used_math())
3844 		kvm_fx_save(&vcpu->arch.host_fx_image);
3845 
3846 	/* Initialize guest FPU by resetting ours and saving into guest's */
3847 	preempt_disable();
3848 	kvm_fx_save(&vcpu->arch.host_fx_image);
3849 	kvm_fx_finit();
3850 	kvm_fx_save(&vcpu->arch.guest_fx_image);
3851 	kvm_fx_restore(&vcpu->arch.host_fx_image);
3852 	preempt_enable();
3853 
3854 	vcpu->arch.cr0 |= X86_CR0_ET;
3855 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3856 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3857 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3858 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3859 }
3860 EXPORT_SYMBOL_GPL(fx_init);
3861 
3862 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3863 {
3864 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3865 		return;
3866 
3867 	vcpu->guest_fpu_loaded = 1;
3868 	kvm_fx_save(&vcpu->arch.host_fx_image);
3869 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
3870 }
3871 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3872 
3873 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3874 {
3875 	if (!vcpu->guest_fpu_loaded)
3876 		return;
3877 
3878 	vcpu->guest_fpu_loaded = 0;
3879 	kvm_fx_save(&vcpu->arch.guest_fx_image);
3880 	kvm_fx_restore(&vcpu->arch.host_fx_image);
3881 	++vcpu->stat.fpu_reload;
3882 }
3883 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3884 
3885 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3886 {
3887 	kvm_x86_ops->vcpu_free(vcpu);
3888 }
3889 
3890 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3891 						unsigned int id)
3892 {
3893 	return kvm_x86_ops->vcpu_create(kvm, id);
3894 }
3895 
3896 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3897 {
3898 	int r;
3899 
3900 	/* We do fxsave: this must be aligned. */
3901 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3902 
3903 	vcpu_load(vcpu);
3904 	r = kvm_arch_vcpu_reset(vcpu);
3905 	if (r == 0)
3906 		r = kvm_mmu_setup(vcpu);
3907 	vcpu_put(vcpu);
3908 	if (r < 0)
3909 		goto free_vcpu;
3910 
3911 	return 0;
3912 free_vcpu:
3913 	kvm_x86_ops->vcpu_free(vcpu);
3914 	return r;
3915 }
3916 
3917 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3918 {
3919 	vcpu_load(vcpu);
3920 	kvm_mmu_unload(vcpu);
3921 	vcpu_put(vcpu);
3922 
3923 	kvm_x86_ops->vcpu_free(vcpu);
3924 }
3925 
3926 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3927 {
3928 	return kvm_x86_ops->vcpu_reset(vcpu);
3929 }
3930 
3931 void kvm_arch_hardware_enable(void *garbage)
3932 {
3933 	kvm_x86_ops->hardware_enable(garbage);
3934 }
3935 
3936 void kvm_arch_hardware_disable(void *garbage)
3937 {
3938 	kvm_x86_ops->hardware_disable(garbage);
3939 }
3940 
3941 int kvm_arch_hardware_setup(void)
3942 {
3943 	return kvm_x86_ops->hardware_setup();
3944 }
3945 
3946 void kvm_arch_hardware_unsetup(void)
3947 {
3948 	kvm_x86_ops->hardware_unsetup();
3949 }
3950 
3951 void kvm_arch_check_processor_compat(void *rtn)
3952 {
3953 	kvm_x86_ops->check_processor_compatibility(rtn);
3954 }
3955 
3956 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3957 {
3958 	struct page *page;
3959 	struct kvm *kvm;
3960 	int r;
3961 
3962 	BUG_ON(vcpu->kvm == NULL);
3963 	kvm = vcpu->kvm;
3964 
3965 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3966 	if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3967 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3968 	else
3969 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3970 
3971 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3972 	if (!page) {
3973 		r = -ENOMEM;
3974 		goto fail;
3975 	}
3976 	vcpu->arch.pio_data = page_address(page);
3977 
3978 	r = kvm_mmu_create(vcpu);
3979 	if (r < 0)
3980 		goto fail_free_pio_data;
3981 
3982 	if (irqchip_in_kernel(kvm)) {
3983 		r = kvm_create_lapic(vcpu);
3984 		if (r < 0)
3985 			goto fail_mmu_destroy;
3986 	}
3987 
3988 	return 0;
3989 
3990 fail_mmu_destroy:
3991 	kvm_mmu_destroy(vcpu);
3992 fail_free_pio_data:
3993 	free_page((unsigned long)vcpu->arch.pio_data);
3994 fail:
3995 	return r;
3996 }
3997 
3998 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3999 {
4000 	kvm_free_lapic(vcpu);
4001 	down_read(&vcpu->kvm->slots_lock);
4002 	kvm_mmu_destroy(vcpu);
4003 	up_read(&vcpu->kvm->slots_lock);
4004 	free_page((unsigned long)vcpu->arch.pio_data);
4005 }
4006 
4007 struct  kvm *kvm_arch_create_vm(void)
4008 {
4009 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4010 
4011 	if (!kvm)
4012 		return ERR_PTR(-ENOMEM);
4013 
4014 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4015 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4016 
4017 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4018 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4019 
4020 	return kvm;
4021 }
4022 
4023 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4024 {
4025 	vcpu_load(vcpu);
4026 	kvm_mmu_unload(vcpu);
4027 	vcpu_put(vcpu);
4028 }
4029 
4030 static void kvm_free_vcpus(struct kvm *kvm)
4031 {
4032 	unsigned int i;
4033 
4034 	/*
4035 	 * Unpin any mmu pages first.
4036 	 */
4037 	for (i = 0; i < KVM_MAX_VCPUS; ++i)
4038 		if (kvm->vcpus[i])
4039 			kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4040 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4041 		if (kvm->vcpus[i]) {
4042 			kvm_arch_vcpu_free(kvm->vcpus[i]);
4043 			kvm->vcpus[i] = NULL;
4044 		}
4045 	}
4046 
4047 }
4048 
4049 void kvm_arch_destroy_vm(struct kvm *kvm)
4050 {
4051 	kvm_iommu_unmap_guest(kvm);
4052 	kvm_free_all_assigned_devices(kvm);
4053 	kvm_free_pit(kvm);
4054 	kfree(kvm->arch.vpic);
4055 	kfree(kvm->arch.vioapic);
4056 	kvm_free_vcpus(kvm);
4057 	kvm_free_physmem(kvm);
4058 	if (kvm->arch.apic_access_page)
4059 		put_page(kvm->arch.apic_access_page);
4060 	if (kvm->arch.ept_identity_pagetable)
4061 		put_page(kvm->arch.ept_identity_pagetable);
4062 	kfree(kvm);
4063 }
4064 
4065 int kvm_arch_set_memory_region(struct kvm *kvm,
4066 				struct kvm_userspace_memory_region *mem,
4067 				struct kvm_memory_slot old,
4068 				int user_alloc)
4069 {
4070 	int npages = mem->memory_size >> PAGE_SHIFT;
4071 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4072 
4073 	/*To keep backward compatibility with older userspace,
4074 	 *x86 needs to hanlde !user_alloc case.
4075 	 */
4076 	if (!user_alloc) {
4077 		if (npages && !old.rmap) {
4078 			unsigned long userspace_addr;
4079 
4080 			down_write(&current->mm->mmap_sem);
4081 			userspace_addr = do_mmap(NULL, 0,
4082 						 npages * PAGE_SIZE,
4083 						 PROT_READ | PROT_WRITE,
4084 						 MAP_PRIVATE | MAP_ANONYMOUS,
4085 						 0);
4086 			up_write(&current->mm->mmap_sem);
4087 
4088 			if (IS_ERR((void *)userspace_addr))
4089 				return PTR_ERR((void *)userspace_addr);
4090 
4091 			/* set userspace_addr atomically for kvm_hva_to_rmapp */
4092 			spin_lock(&kvm->mmu_lock);
4093 			memslot->userspace_addr = userspace_addr;
4094 			spin_unlock(&kvm->mmu_lock);
4095 		} else {
4096 			if (!old.user_alloc && old.rmap) {
4097 				int ret;
4098 
4099 				down_write(&current->mm->mmap_sem);
4100 				ret = do_munmap(current->mm, old.userspace_addr,
4101 						old.npages * PAGE_SIZE);
4102 				up_write(&current->mm->mmap_sem);
4103 				if (ret < 0)
4104 					printk(KERN_WARNING
4105 				       "kvm_vm_ioctl_set_memory_region: "
4106 				       "failed to munmap memory\n");
4107 			}
4108 		}
4109 	}
4110 
4111 	if (!kvm->arch.n_requested_mmu_pages) {
4112 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4113 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4114 	}
4115 
4116 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4117 	kvm_flush_remote_tlbs(kvm);
4118 
4119 	return 0;
4120 }
4121 
4122 void kvm_arch_flush_shadow(struct kvm *kvm)
4123 {
4124 	kvm_mmu_zap_all(kvm);
4125 }
4126 
4127 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4128 {
4129 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4130 	       || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4131 }
4132 
4133 static void vcpu_kick_intr(void *info)
4134 {
4135 #ifdef DEBUG
4136 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4137 	printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4138 #endif
4139 }
4140 
4141 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4142 {
4143 	int ipi_pcpu = vcpu->cpu;
4144 	int cpu = get_cpu();
4145 
4146 	if (waitqueue_active(&vcpu->wq)) {
4147 		wake_up_interruptible(&vcpu->wq);
4148 		++vcpu->stat.halt_wakeup;
4149 	}
4150 	/*
4151 	 * We may be called synchronously with irqs disabled in guest mode,
4152 	 * So need not to call smp_call_function_single() in that case.
4153 	 */
4154 	if (vcpu->guest_mode && vcpu->cpu != cpu)
4155 		smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4156 	put_cpu();
4157 }
4158