xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 22246614)
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  *
8  * Authors:
9  *   Avi Kivity   <avi@qumranet.com>
10  *   Yaniv Kamay  <yaniv@qumranet.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2.  See
13  * the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include <linux/kvm_host.h>
18 #include "irq.h"
19 #include "mmu.h"
20 #include "i8254.h"
21 #include "tss.h"
22 
23 #include <linux/clocksource.h>
24 #include <linux/kvm.h>
25 #include <linux/fs.h>
26 #include <linux/vmalloc.h>
27 #include <linux/module.h>
28 #include <linux/mman.h>
29 #include <linux/highmem.h>
30 
31 #include <asm/uaccess.h>
32 #include <asm/msr.h>
33 #include <asm/desc.h>
34 
35 #define MAX_IO_MSRS 256
36 #define CR0_RESERVED_BITS						\
37 	(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
38 			  | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
39 			  | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
40 #define CR4_RESERVED_BITS						\
41 	(~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
42 			  | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE	\
43 			  | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR	\
44 			  | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
45 
46 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
47 /* EFER defaults:
48  * - enable syscall per default because its emulated by KVM
49  * - enable LME and LMA per default on 64 bit KVM
50  */
51 #ifdef CONFIG_X86_64
52 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
53 #else
54 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
55 #endif
56 
57 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59 
60 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
61 				    struct kvm_cpuid_entry2 __user *entries);
62 
63 struct kvm_x86_ops *kvm_x86_ops;
64 
65 struct kvm_stats_debugfs_item debugfs_entries[] = {
66 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
67 	{ "pf_guest", VCPU_STAT(pf_guest) },
68 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
69 	{ "invlpg", VCPU_STAT(invlpg) },
70 	{ "exits", VCPU_STAT(exits) },
71 	{ "io_exits", VCPU_STAT(io_exits) },
72 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
73 	{ "signal_exits", VCPU_STAT(signal_exits) },
74 	{ "irq_window", VCPU_STAT(irq_window_exits) },
75 	{ "halt_exits", VCPU_STAT(halt_exits) },
76 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
77 	{ "hypercalls", VCPU_STAT(hypercalls) },
78 	{ "request_irq", VCPU_STAT(request_irq_exits) },
79 	{ "irq_exits", VCPU_STAT(irq_exits) },
80 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
81 	{ "efer_reload", VCPU_STAT(efer_reload) },
82 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
83 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
84 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
85 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
86 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
87 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
88 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
89 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
90 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
91 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
92 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
93 	{ "largepages", VM_STAT(lpages) },
94 	{ NULL }
95 };
96 
97 
98 unsigned long segment_base(u16 selector)
99 {
100 	struct descriptor_table gdt;
101 	struct desc_struct *d;
102 	unsigned long table_base;
103 	unsigned long v;
104 
105 	if (selector == 0)
106 		return 0;
107 
108 	asm("sgdt %0" : "=m"(gdt));
109 	table_base = gdt.base;
110 
111 	if (selector & 4) {           /* from ldt */
112 		u16 ldt_selector;
113 
114 		asm("sldt %0" : "=g"(ldt_selector));
115 		table_base = segment_base(ldt_selector);
116 	}
117 	d = (struct desc_struct *)(table_base + (selector & ~7));
118 	v = d->base0 | ((unsigned long)d->base1 << 16) |
119 		((unsigned long)d->base2 << 24);
120 #ifdef CONFIG_X86_64
121 	if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
122 		v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
123 #endif
124 	return v;
125 }
126 EXPORT_SYMBOL_GPL(segment_base);
127 
128 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
129 {
130 	if (irqchip_in_kernel(vcpu->kvm))
131 		return vcpu->arch.apic_base;
132 	else
133 		return vcpu->arch.apic_base;
134 }
135 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
136 
137 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
138 {
139 	/* TODO: reserve bits check */
140 	if (irqchip_in_kernel(vcpu->kvm))
141 		kvm_lapic_set_base(vcpu, data);
142 	else
143 		vcpu->arch.apic_base = data;
144 }
145 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
146 
147 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
148 {
149 	WARN_ON(vcpu->arch.exception.pending);
150 	vcpu->arch.exception.pending = true;
151 	vcpu->arch.exception.has_error_code = false;
152 	vcpu->arch.exception.nr = nr;
153 }
154 EXPORT_SYMBOL_GPL(kvm_queue_exception);
155 
156 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
157 			   u32 error_code)
158 {
159 	++vcpu->stat.pf_guest;
160 	if (vcpu->arch.exception.pending) {
161 		if (vcpu->arch.exception.nr == PF_VECTOR) {
162 			printk(KERN_DEBUG "kvm: inject_page_fault:"
163 					" double fault 0x%lx\n", addr);
164 			vcpu->arch.exception.nr = DF_VECTOR;
165 			vcpu->arch.exception.error_code = 0;
166 		} else if (vcpu->arch.exception.nr == DF_VECTOR) {
167 			/* triple fault -> shutdown */
168 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
169 		}
170 		return;
171 	}
172 	vcpu->arch.cr2 = addr;
173 	kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
174 }
175 
176 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
177 {
178 	WARN_ON(vcpu->arch.exception.pending);
179 	vcpu->arch.exception.pending = true;
180 	vcpu->arch.exception.has_error_code = true;
181 	vcpu->arch.exception.nr = nr;
182 	vcpu->arch.exception.error_code = error_code;
183 }
184 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
185 
186 static void __queue_exception(struct kvm_vcpu *vcpu)
187 {
188 	kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
189 				     vcpu->arch.exception.has_error_code,
190 				     vcpu->arch.exception.error_code);
191 }
192 
193 /*
194  * Load the pae pdptrs.  Return true is they are all valid.
195  */
196 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
197 {
198 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
199 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
200 	int i;
201 	int ret;
202 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
203 
204 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
205 				  offset * sizeof(u64), sizeof(pdpte));
206 	if (ret < 0) {
207 		ret = 0;
208 		goto out;
209 	}
210 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
211 		if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
212 			ret = 0;
213 			goto out;
214 		}
215 	}
216 	ret = 1;
217 
218 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
219 out:
220 
221 	return ret;
222 }
223 EXPORT_SYMBOL_GPL(load_pdptrs);
224 
225 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
226 {
227 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
228 	bool changed = true;
229 	int r;
230 
231 	if (is_long_mode(vcpu) || !is_pae(vcpu))
232 		return false;
233 
234 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
235 	if (r < 0)
236 		goto out;
237 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
238 out:
239 
240 	return changed;
241 }
242 
243 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
244 {
245 	if (cr0 & CR0_RESERVED_BITS) {
246 		printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
247 		       cr0, vcpu->arch.cr0);
248 		kvm_inject_gp(vcpu, 0);
249 		return;
250 	}
251 
252 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
253 		printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
254 		kvm_inject_gp(vcpu, 0);
255 		return;
256 	}
257 
258 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
259 		printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
260 		       "and a clear PE flag\n");
261 		kvm_inject_gp(vcpu, 0);
262 		return;
263 	}
264 
265 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
266 #ifdef CONFIG_X86_64
267 		if ((vcpu->arch.shadow_efer & EFER_LME)) {
268 			int cs_db, cs_l;
269 
270 			if (!is_pae(vcpu)) {
271 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
272 				       "in long mode while PAE is disabled\n");
273 				kvm_inject_gp(vcpu, 0);
274 				return;
275 			}
276 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
277 			if (cs_l) {
278 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
279 				       "in long mode while CS.L == 1\n");
280 				kvm_inject_gp(vcpu, 0);
281 				return;
282 
283 			}
284 		} else
285 #endif
286 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
287 			printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
288 			       "reserved bits\n");
289 			kvm_inject_gp(vcpu, 0);
290 			return;
291 		}
292 
293 	}
294 
295 	kvm_x86_ops->set_cr0(vcpu, cr0);
296 	vcpu->arch.cr0 = cr0;
297 
298 	kvm_mmu_reset_context(vcpu);
299 	return;
300 }
301 EXPORT_SYMBOL_GPL(kvm_set_cr0);
302 
303 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
304 {
305 	kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
306 	KVMTRACE_1D(LMSW, vcpu,
307 		    (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
308 		    handler);
309 }
310 EXPORT_SYMBOL_GPL(kvm_lmsw);
311 
312 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
313 {
314 	if (cr4 & CR4_RESERVED_BITS) {
315 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
316 		kvm_inject_gp(vcpu, 0);
317 		return;
318 	}
319 
320 	if (is_long_mode(vcpu)) {
321 		if (!(cr4 & X86_CR4_PAE)) {
322 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
323 			       "in long mode\n");
324 			kvm_inject_gp(vcpu, 0);
325 			return;
326 		}
327 	} else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
328 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
329 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
330 		kvm_inject_gp(vcpu, 0);
331 		return;
332 	}
333 
334 	if (cr4 & X86_CR4_VMXE) {
335 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
336 		kvm_inject_gp(vcpu, 0);
337 		return;
338 	}
339 	kvm_x86_ops->set_cr4(vcpu, cr4);
340 	vcpu->arch.cr4 = cr4;
341 	kvm_mmu_reset_context(vcpu);
342 }
343 EXPORT_SYMBOL_GPL(kvm_set_cr4);
344 
345 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
346 {
347 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
348 		kvm_mmu_flush_tlb(vcpu);
349 		return;
350 	}
351 
352 	if (is_long_mode(vcpu)) {
353 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
354 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
355 			kvm_inject_gp(vcpu, 0);
356 			return;
357 		}
358 	} else {
359 		if (is_pae(vcpu)) {
360 			if (cr3 & CR3_PAE_RESERVED_BITS) {
361 				printk(KERN_DEBUG
362 				       "set_cr3: #GP, reserved bits\n");
363 				kvm_inject_gp(vcpu, 0);
364 				return;
365 			}
366 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
367 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
368 				       "reserved bits\n");
369 				kvm_inject_gp(vcpu, 0);
370 				return;
371 			}
372 		}
373 		/*
374 		 * We don't check reserved bits in nonpae mode, because
375 		 * this isn't enforced, and VMware depends on this.
376 		 */
377 	}
378 
379 	/*
380 	 * Does the new cr3 value map to physical memory? (Note, we
381 	 * catch an invalid cr3 even in real-mode, because it would
382 	 * cause trouble later on when we turn on paging anyway.)
383 	 *
384 	 * A real CPU would silently accept an invalid cr3 and would
385 	 * attempt to use it - with largely undefined (and often hard
386 	 * to debug) behavior on the guest side.
387 	 */
388 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
389 		kvm_inject_gp(vcpu, 0);
390 	else {
391 		vcpu->arch.cr3 = cr3;
392 		vcpu->arch.mmu.new_cr3(vcpu);
393 	}
394 }
395 EXPORT_SYMBOL_GPL(kvm_set_cr3);
396 
397 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
398 {
399 	if (cr8 & CR8_RESERVED_BITS) {
400 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
401 		kvm_inject_gp(vcpu, 0);
402 		return;
403 	}
404 	if (irqchip_in_kernel(vcpu->kvm))
405 		kvm_lapic_set_tpr(vcpu, cr8);
406 	else
407 		vcpu->arch.cr8 = cr8;
408 }
409 EXPORT_SYMBOL_GPL(kvm_set_cr8);
410 
411 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
412 {
413 	if (irqchip_in_kernel(vcpu->kvm))
414 		return kvm_lapic_get_cr8(vcpu);
415 	else
416 		return vcpu->arch.cr8;
417 }
418 EXPORT_SYMBOL_GPL(kvm_get_cr8);
419 
420 /*
421  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
422  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
423  *
424  * This list is modified at module load time to reflect the
425  * capabilities of the host cpu.
426  */
427 static u32 msrs_to_save[] = {
428 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
429 	MSR_K6_STAR,
430 #ifdef CONFIG_X86_64
431 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
432 #endif
433 	MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
434 	MSR_IA32_PERF_STATUS,
435 };
436 
437 static unsigned num_msrs_to_save;
438 
439 static u32 emulated_msrs[] = {
440 	MSR_IA32_MISC_ENABLE,
441 };
442 
443 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
444 {
445 	if (efer & efer_reserved_bits) {
446 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
447 		       efer);
448 		kvm_inject_gp(vcpu, 0);
449 		return;
450 	}
451 
452 	if (is_paging(vcpu)
453 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
454 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
455 		kvm_inject_gp(vcpu, 0);
456 		return;
457 	}
458 
459 	kvm_x86_ops->set_efer(vcpu, efer);
460 
461 	efer &= ~EFER_LMA;
462 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
463 
464 	vcpu->arch.shadow_efer = efer;
465 }
466 
467 void kvm_enable_efer_bits(u64 mask)
468 {
469        efer_reserved_bits &= ~mask;
470 }
471 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
472 
473 
474 /*
475  * Writes msr value into into the appropriate "register".
476  * Returns 0 on success, non-0 otherwise.
477  * Assumes vcpu_load() was already called.
478  */
479 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
480 {
481 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
482 }
483 
484 /*
485  * Adapt set_msr() to msr_io()'s calling convention
486  */
487 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
488 {
489 	return kvm_set_msr(vcpu, index, *data);
490 }
491 
492 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
493 {
494 	static int version;
495 	struct kvm_wall_clock wc;
496 	struct timespec wc_ts;
497 
498 	if (!wall_clock)
499 		return;
500 
501 	version++;
502 
503 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
504 
505 	wc_ts = current_kernel_time();
506 	wc.wc_sec = wc_ts.tv_sec;
507 	wc.wc_nsec = wc_ts.tv_nsec;
508 	wc.wc_version = version;
509 
510 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
511 
512 	version++;
513 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
514 }
515 
516 static void kvm_write_guest_time(struct kvm_vcpu *v)
517 {
518 	struct timespec ts;
519 	unsigned long flags;
520 	struct kvm_vcpu_arch *vcpu = &v->arch;
521 	void *shared_kaddr;
522 
523 	if ((!vcpu->time_page))
524 		return;
525 
526 	/* Keep irq disabled to prevent changes to the clock */
527 	local_irq_save(flags);
528 	kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
529 			  &vcpu->hv_clock.tsc_timestamp);
530 	ktime_get_ts(&ts);
531 	local_irq_restore(flags);
532 
533 	/* With all the info we got, fill in the values */
534 
535 	vcpu->hv_clock.system_time = ts.tv_nsec +
536 				     (NSEC_PER_SEC * (u64)ts.tv_sec);
537 	/*
538 	 * The interface expects us to write an even number signaling that the
539 	 * update is finished. Since the guest won't see the intermediate
540 	 * state, we just write "2" at the end
541 	 */
542 	vcpu->hv_clock.version = 2;
543 
544 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
545 
546 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
547 		sizeof(vcpu->hv_clock));
548 
549 	kunmap_atomic(shared_kaddr, KM_USER0);
550 
551 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
552 }
553 
554 
555 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
556 {
557 	switch (msr) {
558 	case MSR_EFER:
559 		set_efer(vcpu, data);
560 		break;
561 	case MSR_IA32_MC0_STATUS:
562 		pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
563 		       __func__, data);
564 		break;
565 	case MSR_IA32_MCG_STATUS:
566 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
567 			__func__, data);
568 		break;
569 	case MSR_IA32_MCG_CTL:
570 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
571 			__func__, data);
572 		break;
573 	case MSR_IA32_UCODE_REV:
574 	case MSR_IA32_UCODE_WRITE:
575 	case 0x200 ... 0x2ff: /* MTRRs */
576 		break;
577 	case MSR_IA32_APICBASE:
578 		kvm_set_apic_base(vcpu, data);
579 		break;
580 	case MSR_IA32_MISC_ENABLE:
581 		vcpu->arch.ia32_misc_enable_msr = data;
582 		break;
583 	case MSR_KVM_WALL_CLOCK:
584 		vcpu->kvm->arch.wall_clock = data;
585 		kvm_write_wall_clock(vcpu->kvm, data);
586 		break;
587 	case MSR_KVM_SYSTEM_TIME: {
588 		if (vcpu->arch.time_page) {
589 			kvm_release_page_dirty(vcpu->arch.time_page);
590 			vcpu->arch.time_page = NULL;
591 		}
592 
593 		vcpu->arch.time = data;
594 
595 		/* we verify if the enable bit is set... */
596 		if (!(data & 1))
597 			break;
598 
599 		/* ...but clean it before doing the actual write */
600 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
601 
602 		vcpu->arch.hv_clock.tsc_to_system_mul =
603 					clocksource_khz2mult(tsc_khz, 22);
604 		vcpu->arch.hv_clock.tsc_shift = 22;
605 
606 		down_read(&current->mm->mmap_sem);
607 		vcpu->arch.time_page =
608 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
609 		up_read(&current->mm->mmap_sem);
610 
611 		if (is_error_page(vcpu->arch.time_page)) {
612 			kvm_release_page_clean(vcpu->arch.time_page);
613 			vcpu->arch.time_page = NULL;
614 		}
615 
616 		kvm_write_guest_time(vcpu);
617 		break;
618 	}
619 	default:
620 		pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
621 		return 1;
622 	}
623 	return 0;
624 }
625 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
626 
627 
628 /*
629  * Reads an msr value (of 'msr_index') into 'pdata'.
630  * Returns 0 on success, non-0 otherwise.
631  * Assumes vcpu_load() was already called.
632  */
633 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
634 {
635 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
636 }
637 
638 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
639 {
640 	u64 data;
641 
642 	switch (msr) {
643 	case 0xc0010010: /* SYSCFG */
644 	case 0xc0010015: /* HWCR */
645 	case MSR_IA32_PLATFORM_ID:
646 	case MSR_IA32_P5_MC_ADDR:
647 	case MSR_IA32_P5_MC_TYPE:
648 	case MSR_IA32_MC0_CTL:
649 	case MSR_IA32_MCG_STATUS:
650 	case MSR_IA32_MCG_CAP:
651 	case MSR_IA32_MCG_CTL:
652 	case MSR_IA32_MC0_MISC:
653 	case MSR_IA32_MC0_MISC+4:
654 	case MSR_IA32_MC0_MISC+8:
655 	case MSR_IA32_MC0_MISC+12:
656 	case MSR_IA32_MC0_MISC+16:
657 	case MSR_IA32_UCODE_REV:
658 	case MSR_IA32_EBL_CR_POWERON:
659 		/* MTRR registers */
660 	case 0xfe:
661 	case 0x200 ... 0x2ff:
662 		data = 0;
663 		break;
664 	case 0xcd: /* fsb frequency */
665 		data = 3;
666 		break;
667 	case MSR_IA32_APICBASE:
668 		data = kvm_get_apic_base(vcpu);
669 		break;
670 	case MSR_IA32_MISC_ENABLE:
671 		data = vcpu->arch.ia32_misc_enable_msr;
672 		break;
673 	case MSR_IA32_PERF_STATUS:
674 		/* TSC increment by tick */
675 		data = 1000ULL;
676 		/* CPU multiplier */
677 		data |= (((uint64_t)4ULL) << 40);
678 		break;
679 	case MSR_EFER:
680 		data = vcpu->arch.shadow_efer;
681 		break;
682 	case MSR_KVM_WALL_CLOCK:
683 		data = vcpu->kvm->arch.wall_clock;
684 		break;
685 	case MSR_KVM_SYSTEM_TIME:
686 		data = vcpu->arch.time;
687 		break;
688 	default:
689 		pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
690 		return 1;
691 	}
692 	*pdata = data;
693 	return 0;
694 }
695 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
696 
697 /*
698  * Read or write a bunch of msrs. All parameters are kernel addresses.
699  *
700  * @return number of msrs set successfully.
701  */
702 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
703 		    struct kvm_msr_entry *entries,
704 		    int (*do_msr)(struct kvm_vcpu *vcpu,
705 				  unsigned index, u64 *data))
706 {
707 	int i;
708 
709 	vcpu_load(vcpu);
710 
711 	down_read(&vcpu->kvm->slots_lock);
712 	for (i = 0; i < msrs->nmsrs; ++i)
713 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
714 			break;
715 	up_read(&vcpu->kvm->slots_lock);
716 
717 	vcpu_put(vcpu);
718 
719 	return i;
720 }
721 
722 /*
723  * Read or write a bunch of msrs. Parameters are user addresses.
724  *
725  * @return number of msrs set successfully.
726  */
727 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
728 		  int (*do_msr)(struct kvm_vcpu *vcpu,
729 				unsigned index, u64 *data),
730 		  int writeback)
731 {
732 	struct kvm_msrs msrs;
733 	struct kvm_msr_entry *entries;
734 	int r, n;
735 	unsigned size;
736 
737 	r = -EFAULT;
738 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
739 		goto out;
740 
741 	r = -E2BIG;
742 	if (msrs.nmsrs >= MAX_IO_MSRS)
743 		goto out;
744 
745 	r = -ENOMEM;
746 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
747 	entries = vmalloc(size);
748 	if (!entries)
749 		goto out;
750 
751 	r = -EFAULT;
752 	if (copy_from_user(entries, user_msrs->entries, size))
753 		goto out_free;
754 
755 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
756 	if (r < 0)
757 		goto out_free;
758 
759 	r = -EFAULT;
760 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
761 		goto out_free;
762 
763 	r = n;
764 
765 out_free:
766 	vfree(entries);
767 out:
768 	return r;
769 }
770 
771 /*
772  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
773  * cached on it.
774  */
775 void decache_vcpus_on_cpu(int cpu)
776 {
777 	struct kvm *vm;
778 	struct kvm_vcpu *vcpu;
779 	int i;
780 
781 	spin_lock(&kvm_lock);
782 	list_for_each_entry(vm, &vm_list, vm_list)
783 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
784 			vcpu = vm->vcpus[i];
785 			if (!vcpu)
786 				continue;
787 			/*
788 			 * If the vcpu is locked, then it is running on some
789 			 * other cpu and therefore it is not cached on the
790 			 * cpu in question.
791 			 *
792 			 * If it's not locked, check the last cpu it executed
793 			 * on.
794 			 */
795 			if (mutex_trylock(&vcpu->mutex)) {
796 				if (vcpu->cpu == cpu) {
797 					kvm_x86_ops->vcpu_decache(vcpu);
798 					vcpu->cpu = -1;
799 				}
800 				mutex_unlock(&vcpu->mutex);
801 			}
802 		}
803 	spin_unlock(&kvm_lock);
804 }
805 
806 int kvm_dev_ioctl_check_extension(long ext)
807 {
808 	int r;
809 
810 	switch (ext) {
811 	case KVM_CAP_IRQCHIP:
812 	case KVM_CAP_HLT:
813 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
814 	case KVM_CAP_USER_MEMORY:
815 	case KVM_CAP_SET_TSS_ADDR:
816 	case KVM_CAP_EXT_CPUID:
817 	case KVM_CAP_CLOCKSOURCE:
818 	case KVM_CAP_PIT:
819 	case KVM_CAP_NOP_IO_DELAY:
820 	case KVM_CAP_MP_STATE:
821 		r = 1;
822 		break;
823 	case KVM_CAP_VAPIC:
824 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
825 		break;
826 	case KVM_CAP_NR_VCPUS:
827 		r = KVM_MAX_VCPUS;
828 		break;
829 	case KVM_CAP_NR_MEMSLOTS:
830 		r = KVM_MEMORY_SLOTS;
831 		break;
832 	case KVM_CAP_PV_MMU:
833 		r = !tdp_enabled;
834 		break;
835 	default:
836 		r = 0;
837 		break;
838 	}
839 	return r;
840 
841 }
842 
843 long kvm_arch_dev_ioctl(struct file *filp,
844 			unsigned int ioctl, unsigned long arg)
845 {
846 	void __user *argp = (void __user *)arg;
847 	long r;
848 
849 	switch (ioctl) {
850 	case KVM_GET_MSR_INDEX_LIST: {
851 		struct kvm_msr_list __user *user_msr_list = argp;
852 		struct kvm_msr_list msr_list;
853 		unsigned n;
854 
855 		r = -EFAULT;
856 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
857 			goto out;
858 		n = msr_list.nmsrs;
859 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
860 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
861 			goto out;
862 		r = -E2BIG;
863 		if (n < num_msrs_to_save)
864 			goto out;
865 		r = -EFAULT;
866 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
867 				 num_msrs_to_save * sizeof(u32)))
868 			goto out;
869 		if (copy_to_user(user_msr_list->indices
870 				 + num_msrs_to_save * sizeof(u32),
871 				 &emulated_msrs,
872 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
873 			goto out;
874 		r = 0;
875 		break;
876 	}
877 	case KVM_GET_SUPPORTED_CPUID: {
878 		struct kvm_cpuid2 __user *cpuid_arg = argp;
879 		struct kvm_cpuid2 cpuid;
880 
881 		r = -EFAULT;
882 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
883 			goto out;
884 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
885 			cpuid_arg->entries);
886 		if (r)
887 			goto out;
888 
889 		r = -EFAULT;
890 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
891 			goto out;
892 		r = 0;
893 		break;
894 	}
895 	default:
896 		r = -EINVAL;
897 	}
898 out:
899 	return r;
900 }
901 
902 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
903 {
904 	kvm_x86_ops->vcpu_load(vcpu, cpu);
905 	kvm_write_guest_time(vcpu);
906 }
907 
908 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
909 {
910 	kvm_x86_ops->vcpu_put(vcpu);
911 	kvm_put_guest_fpu(vcpu);
912 }
913 
914 static int is_efer_nx(void)
915 {
916 	u64 efer;
917 
918 	rdmsrl(MSR_EFER, efer);
919 	return efer & EFER_NX;
920 }
921 
922 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
923 {
924 	int i;
925 	struct kvm_cpuid_entry2 *e, *entry;
926 
927 	entry = NULL;
928 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
929 		e = &vcpu->arch.cpuid_entries[i];
930 		if (e->function == 0x80000001) {
931 			entry = e;
932 			break;
933 		}
934 	}
935 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
936 		entry->edx &= ~(1 << 20);
937 		printk(KERN_INFO "kvm: guest NX capability removed\n");
938 	}
939 }
940 
941 /* when an old userspace process fills a new kernel module */
942 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
943 				    struct kvm_cpuid *cpuid,
944 				    struct kvm_cpuid_entry __user *entries)
945 {
946 	int r, i;
947 	struct kvm_cpuid_entry *cpuid_entries;
948 
949 	r = -E2BIG;
950 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
951 		goto out;
952 	r = -ENOMEM;
953 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
954 	if (!cpuid_entries)
955 		goto out;
956 	r = -EFAULT;
957 	if (copy_from_user(cpuid_entries, entries,
958 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
959 		goto out_free;
960 	for (i = 0; i < cpuid->nent; i++) {
961 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
962 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
963 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
964 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
965 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
966 		vcpu->arch.cpuid_entries[i].index = 0;
967 		vcpu->arch.cpuid_entries[i].flags = 0;
968 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
969 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
970 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
971 	}
972 	vcpu->arch.cpuid_nent = cpuid->nent;
973 	cpuid_fix_nx_cap(vcpu);
974 	r = 0;
975 
976 out_free:
977 	vfree(cpuid_entries);
978 out:
979 	return r;
980 }
981 
982 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
983 				    struct kvm_cpuid2 *cpuid,
984 				    struct kvm_cpuid_entry2 __user *entries)
985 {
986 	int r;
987 
988 	r = -E2BIG;
989 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
990 		goto out;
991 	r = -EFAULT;
992 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
993 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
994 		goto out;
995 	vcpu->arch.cpuid_nent = cpuid->nent;
996 	return 0;
997 
998 out:
999 	return r;
1000 }
1001 
1002 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1003 				    struct kvm_cpuid2 *cpuid,
1004 				    struct kvm_cpuid_entry2 __user *entries)
1005 {
1006 	int r;
1007 
1008 	r = -E2BIG;
1009 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1010 		goto out;
1011 	r = -EFAULT;
1012 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1013 			   vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1014 		goto out;
1015 	return 0;
1016 
1017 out:
1018 	cpuid->nent = vcpu->arch.cpuid_nent;
1019 	return r;
1020 }
1021 
1022 static inline u32 bit(int bitno)
1023 {
1024 	return 1 << (bitno & 31);
1025 }
1026 
1027 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1028 			  u32 index)
1029 {
1030 	entry->function = function;
1031 	entry->index = index;
1032 	cpuid_count(entry->function, entry->index,
1033 		&entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1034 	entry->flags = 0;
1035 }
1036 
1037 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1038 			 u32 index, int *nent, int maxnent)
1039 {
1040 	const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1041 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1042 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1043 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1044 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1045 		bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1046 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1047 		bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1048 		bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1049 		bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1050 	const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1051 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1052 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1053 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1054 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1055 		bit(X86_FEATURE_PGE) |
1056 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1057 		bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1058 		bit(X86_FEATURE_SYSCALL) |
1059 		(bit(X86_FEATURE_NX) && is_efer_nx()) |
1060 #ifdef CONFIG_X86_64
1061 		bit(X86_FEATURE_LM) |
1062 #endif
1063 		bit(X86_FEATURE_MMXEXT) |
1064 		bit(X86_FEATURE_3DNOWEXT) |
1065 		bit(X86_FEATURE_3DNOW);
1066 	const u32 kvm_supported_word3_x86_features =
1067 		bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1068 	const u32 kvm_supported_word6_x86_features =
1069 		bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1070 
1071 	/* all func 2 cpuid_count() should be called on the same cpu */
1072 	get_cpu();
1073 	do_cpuid_1_ent(entry, function, index);
1074 	++*nent;
1075 
1076 	switch (function) {
1077 	case 0:
1078 		entry->eax = min(entry->eax, (u32)0xb);
1079 		break;
1080 	case 1:
1081 		entry->edx &= kvm_supported_word0_x86_features;
1082 		entry->ecx &= kvm_supported_word3_x86_features;
1083 		break;
1084 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1085 	 * may return different values. This forces us to get_cpu() before
1086 	 * issuing the first command, and also to emulate this annoying behavior
1087 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1088 	case 2: {
1089 		int t, times = entry->eax & 0xff;
1090 
1091 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1092 		for (t = 1; t < times && *nent < maxnent; ++t) {
1093 			do_cpuid_1_ent(&entry[t], function, 0);
1094 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1095 			++*nent;
1096 		}
1097 		break;
1098 	}
1099 	/* function 4 and 0xb have additional index. */
1100 	case 4: {
1101 		int i, cache_type;
1102 
1103 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1104 		/* read more entries until cache_type is zero */
1105 		for (i = 1; *nent < maxnent; ++i) {
1106 			cache_type = entry[i - 1].eax & 0x1f;
1107 			if (!cache_type)
1108 				break;
1109 			do_cpuid_1_ent(&entry[i], function, i);
1110 			entry[i].flags |=
1111 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1112 			++*nent;
1113 		}
1114 		break;
1115 	}
1116 	case 0xb: {
1117 		int i, level_type;
1118 
1119 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1120 		/* read more entries until level_type is zero */
1121 		for (i = 1; *nent < maxnent; ++i) {
1122 			level_type = entry[i - 1].ecx & 0xff;
1123 			if (!level_type)
1124 				break;
1125 			do_cpuid_1_ent(&entry[i], function, i);
1126 			entry[i].flags |=
1127 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1128 			++*nent;
1129 		}
1130 		break;
1131 	}
1132 	case 0x80000000:
1133 		entry->eax = min(entry->eax, 0x8000001a);
1134 		break;
1135 	case 0x80000001:
1136 		entry->edx &= kvm_supported_word1_x86_features;
1137 		entry->ecx &= kvm_supported_word6_x86_features;
1138 		break;
1139 	}
1140 	put_cpu();
1141 }
1142 
1143 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1144 				    struct kvm_cpuid_entry2 __user *entries)
1145 {
1146 	struct kvm_cpuid_entry2 *cpuid_entries;
1147 	int limit, nent = 0, r = -E2BIG;
1148 	u32 func;
1149 
1150 	if (cpuid->nent < 1)
1151 		goto out;
1152 	r = -ENOMEM;
1153 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1154 	if (!cpuid_entries)
1155 		goto out;
1156 
1157 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1158 	limit = cpuid_entries[0].eax;
1159 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1160 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1161 				&nent, cpuid->nent);
1162 	r = -E2BIG;
1163 	if (nent >= cpuid->nent)
1164 		goto out_free;
1165 
1166 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1167 	limit = cpuid_entries[nent - 1].eax;
1168 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1169 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1170 			       &nent, cpuid->nent);
1171 	r = -EFAULT;
1172 	if (copy_to_user(entries, cpuid_entries,
1173 			nent * sizeof(struct kvm_cpuid_entry2)))
1174 		goto out_free;
1175 	cpuid->nent = nent;
1176 	r = 0;
1177 
1178 out_free:
1179 	vfree(cpuid_entries);
1180 out:
1181 	return r;
1182 }
1183 
1184 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1185 				    struct kvm_lapic_state *s)
1186 {
1187 	vcpu_load(vcpu);
1188 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1189 	vcpu_put(vcpu);
1190 
1191 	return 0;
1192 }
1193 
1194 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1195 				    struct kvm_lapic_state *s)
1196 {
1197 	vcpu_load(vcpu);
1198 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1199 	kvm_apic_post_state_restore(vcpu);
1200 	vcpu_put(vcpu);
1201 
1202 	return 0;
1203 }
1204 
1205 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1206 				    struct kvm_interrupt *irq)
1207 {
1208 	if (irq->irq < 0 || irq->irq >= 256)
1209 		return -EINVAL;
1210 	if (irqchip_in_kernel(vcpu->kvm))
1211 		return -ENXIO;
1212 	vcpu_load(vcpu);
1213 
1214 	set_bit(irq->irq, vcpu->arch.irq_pending);
1215 	set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1216 
1217 	vcpu_put(vcpu);
1218 
1219 	return 0;
1220 }
1221 
1222 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1223 					   struct kvm_tpr_access_ctl *tac)
1224 {
1225 	if (tac->flags)
1226 		return -EINVAL;
1227 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1228 	return 0;
1229 }
1230 
1231 long kvm_arch_vcpu_ioctl(struct file *filp,
1232 			 unsigned int ioctl, unsigned long arg)
1233 {
1234 	struct kvm_vcpu *vcpu = filp->private_data;
1235 	void __user *argp = (void __user *)arg;
1236 	int r;
1237 
1238 	switch (ioctl) {
1239 	case KVM_GET_LAPIC: {
1240 		struct kvm_lapic_state lapic;
1241 
1242 		memset(&lapic, 0, sizeof lapic);
1243 		r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1244 		if (r)
1245 			goto out;
1246 		r = -EFAULT;
1247 		if (copy_to_user(argp, &lapic, sizeof lapic))
1248 			goto out;
1249 		r = 0;
1250 		break;
1251 	}
1252 	case KVM_SET_LAPIC: {
1253 		struct kvm_lapic_state lapic;
1254 
1255 		r = -EFAULT;
1256 		if (copy_from_user(&lapic, argp, sizeof lapic))
1257 			goto out;
1258 		r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1259 		if (r)
1260 			goto out;
1261 		r = 0;
1262 		break;
1263 	}
1264 	case KVM_INTERRUPT: {
1265 		struct kvm_interrupt irq;
1266 
1267 		r = -EFAULT;
1268 		if (copy_from_user(&irq, argp, sizeof irq))
1269 			goto out;
1270 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1271 		if (r)
1272 			goto out;
1273 		r = 0;
1274 		break;
1275 	}
1276 	case KVM_SET_CPUID: {
1277 		struct kvm_cpuid __user *cpuid_arg = argp;
1278 		struct kvm_cpuid cpuid;
1279 
1280 		r = -EFAULT;
1281 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1282 			goto out;
1283 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1284 		if (r)
1285 			goto out;
1286 		break;
1287 	}
1288 	case KVM_SET_CPUID2: {
1289 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1290 		struct kvm_cpuid2 cpuid;
1291 
1292 		r = -EFAULT;
1293 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1294 			goto out;
1295 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1296 				cpuid_arg->entries);
1297 		if (r)
1298 			goto out;
1299 		break;
1300 	}
1301 	case KVM_GET_CPUID2: {
1302 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1303 		struct kvm_cpuid2 cpuid;
1304 
1305 		r = -EFAULT;
1306 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1307 			goto out;
1308 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1309 				cpuid_arg->entries);
1310 		if (r)
1311 			goto out;
1312 		r = -EFAULT;
1313 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1314 			goto out;
1315 		r = 0;
1316 		break;
1317 	}
1318 	case KVM_GET_MSRS:
1319 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1320 		break;
1321 	case KVM_SET_MSRS:
1322 		r = msr_io(vcpu, argp, do_set_msr, 0);
1323 		break;
1324 	case KVM_TPR_ACCESS_REPORTING: {
1325 		struct kvm_tpr_access_ctl tac;
1326 
1327 		r = -EFAULT;
1328 		if (copy_from_user(&tac, argp, sizeof tac))
1329 			goto out;
1330 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1331 		if (r)
1332 			goto out;
1333 		r = -EFAULT;
1334 		if (copy_to_user(argp, &tac, sizeof tac))
1335 			goto out;
1336 		r = 0;
1337 		break;
1338 	};
1339 	case KVM_SET_VAPIC_ADDR: {
1340 		struct kvm_vapic_addr va;
1341 
1342 		r = -EINVAL;
1343 		if (!irqchip_in_kernel(vcpu->kvm))
1344 			goto out;
1345 		r = -EFAULT;
1346 		if (copy_from_user(&va, argp, sizeof va))
1347 			goto out;
1348 		r = 0;
1349 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1350 		break;
1351 	}
1352 	default:
1353 		r = -EINVAL;
1354 	}
1355 out:
1356 	return r;
1357 }
1358 
1359 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1360 {
1361 	int ret;
1362 
1363 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1364 		return -1;
1365 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1366 	return ret;
1367 }
1368 
1369 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1370 					  u32 kvm_nr_mmu_pages)
1371 {
1372 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1373 		return -EINVAL;
1374 
1375 	down_write(&kvm->slots_lock);
1376 
1377 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1378 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1379 
1380 	up_write(&kvm->slots_lock);
1381 	return 0;
1382 }
1383 
1384 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1385 {
1386 	return kvm->arch.n_alloc_mmu_pages;
1387 }
1388 
1389 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1390 {
1391 	int i;
1392 	struct kvm_mem_alias *alias;
1393 
1394 	for (i = 0; i < kvm->arch.naliases; ++i) {
1395 		alias = &kvm->arch.aliases[i];
1396 		if (gfn >= alias->base_gfn
1397 		    && gfn < alias->base_gfn + alias->npages)
1398 			return alias->target_gfn + gfn - alias->base_gfn;
1399 	}
1400 	return gfn;
1401 }
1402 
1403 /*
1404  * Set a new alias region.  Aliases map a portion of physical memory into
1405  * another portion.  This is useful for memory windows, for example the PC
1406  * VGA region.
1407  */
1408 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1409 					 struct kvm_memory_alias *alias)
1410 {
1411 	int r, n;
1412 	struct kvm_mem_alias *p;
1413 
1414 	r = -EINVAL;
1415 	/* General sanity checks */
1416 	if (alias->memory_size & (PAGE_SIZE - 1))
1417 		goto out;
1418 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1419 		goto out;
1420 	if (alias->slot >= KVM_ALIAS_SLOTS)
1421 		goto out;
1422 	if (alias->guest_phys_addr + alias->memory_size
1423 	    < alias->guest_phys_addr)
1424 		goto out;
1425 	if (alias->target_phys_addr + alias->memory_size
1426 	    < alias->target_phys_addr)
1427 		goto out;
1428 
1429 	down_write(&kvm->slots_lock);
1430 
1431 	p = &kvm->arch.aliases[alias->slot];
1432 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1433 	p->npages = alias->memory_size >> PAGE_SHIFT;
1434 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1435 
1436 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1437 		if (kvm->arch.aliases[n - 1].npages)
1438 			break;
1439 	kvm->arch.naliases = n;
1440 
1441 	kvm_mmu_zap_all(kvm);
1442 
1443 	up_write(&kvm->slots_lock);
1444 
1445 	return 0;
1446 
1447 out:
1448 	return r;
1449 }
1450 
1451 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1452 {
1453 	int r;
1454 
1455 	r = 0;
1456 	switch (chip->chip_id) {
1457 	case KVM_IRQCHIP_PIC_MASTER:
1458 		memcpy(&chip->chip.pic,
1459 			&pic_irqchip(kvm)->pics[0],
1460 			sizeof(struct kvm_pic_state));
1461 		break;
1462 	case KVM_IRQCHIP_PIC_SLAVE:
1463 		memcpy(&chip->chip.pic,
1464 			&pic_irqchip(kvm)->pics[1],
1465 			sizeof(struct kvm_pic_state));
1466 		break;
1467 	case KVM_IRQCHIP_IOAPIC:
1468 		memcpy(&chip->chip.ioapic,
1469 			ioapic_irqchip(kvm),
1470 			sizeof(struct kvm_ioapic_state));
1471 		break;
1472 	default:
1473 		r = -EINVAL;
1474 		break;
1475 	}
1476 	return r;
1477 }
1478 
1479 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1480 {
1481 	int r;
1482 
1483 	r = 0;
1484 	switch (chip->chip_id) {
1485 	case KVM_IRQCHIP_PIC_MASTER:
1486 		memcpy(&pic_irqchip(kvm)->pics[0],
1487 			&chip->chip.pic,
1488 			sizeof(struct kvm_pic_state));
1489 		break;
1490 	case KVM_IRQCHIP_PIC_SLAVE:
1491 		memcpy(&pic_irqchip(kvm)->pics[1],
1492 			&chip->chip.pic,
1493 			sizeof(struct kvm_pic_state));
1494 		break;
1495 	case KVM_IRQCHIP_IOAPIC:
1496 		memcpy(ioapic_irqchip(kvm),
1497 			&chip->chip.ioapic,
1498 			sizeof(struct kvm_ioapic_state));
1499 		break;
1500 	default:
1501 		r = -EINVAL;
1502 		break;
1503 	}
1504 	kvm_pic_update_irq(pic_irqchip(kvm));
1505 	return r;
1506 }
1507 
1508 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1509 {
1510 	int r = 0;
1511 
1512 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1513 	return r;
1514 }
1515 
1516 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1517 {
1518 	int r = 0;
1519 
1520 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1521 	kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1522 	return r;
1523 }
1524 
1525 /*
1526  * Get (and clear) the dirty memory log for a memory slot.
1527  */
1528 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1529 				      struct kvm_dirty_log *log)
1530 {
1531 	int r;
1532 	int n;
1533 	struct kvm_memory_slot *memslot;
1534 	int is_dirty = 0;
1535 
1536 	down_write(&kvm->slots_lock);
1537 
1538 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1539 	if (r)
1540 		goto out;
1541 
1542 	/* If nothing is dirty, don't bother messing with page tables. */
1543 	if (is_dirty) {
1544 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
1545 		kvm_flush_remote_tlbs(kvm);
1546 		memslot = &kvm->memslots[log->slot];
1547 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1548 		memset(memslot->dirty_bitmap, 0, n);
1549 	}
1550 	r = 0;
1551 out:
1552 	up_write(&kvm->slots_lock);
1553 	return r;
1554 }
1555 
1556 long kvm_arch_vm_ioctl(struct file *filp,
1557 		       unsigned int ioctl, unsigned long arg)
1558 {
1559 	struct kvm *kvm = filp->private_data;
1560 	void __user *argp = (void __user *)arg;
1561 	int r = -EINVAL;
1562 
1563 	switch (ioctl) {
1564 	case KVM_SET_TSS_ADDR:
1565 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1566 		if (r < 0)
1567 			goto out;
1568 		break;
1569 	case KVM_SET_MEMORY_REGION: {
1570 		struct kvm_memory_region kvm_mem;
1571 		struct kvm_userspace_memory_region kvm_userspace_mem;
1572 
1573 		r = -EFAULT;
1574 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1575 			goto out;
1576 		kvm_userspace_mem.slot = kvm_mem.slot;
1577 		kvm_userspace_mem.flags = kvm_mem.flags;
1578 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1579 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1580 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1581 		if (r)
1582 			goto out;
1583 		break;
1584 	}
1585 	case KVM_SET_NR_MMU_PAGES:
1586 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1587 		if (r)
1588 			goto out;
1589 		break;
1590 	case KVM_GET_NR_MMU_PAGES:
1591 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1592 		break;
1593 	case KVM_SET_MEMORY_ALIAS: {
1594 		struct kvm_memory_alias alias;
1595 
1596 		r = -EFAULT;
1597 		if (copy_from_user(&alias, argp, sizeof alias))
1598 			goto out;
1599 		r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1600 		if (r)
1601 			goto out;
1602 		break;
1603 	}
1604 	case KVM_CREATE_IRQCHIP:
1605 		r = -ENOMEM;
1606 		kvm->arch.vpic = kvm_create_pic(kvm);
1607 		if (kvm->arch.vpic) {
1608 			r = kvm_ioapic_init(kvm);
1609 			if (r) {
1610 				kfree(kvm->arch.vpic);
1611 				kvm->arch.vpic = NULL;
1612 				goto out;
1613 			}
1614 		} else
1615 			goto out;
1616 		break;
1617 	case KVM_CREATE_PIT:
1618 		r = -ENOMEM;
1619 		kvm->arch.vpit = kvm_create_pit(kvm);
1620 		if (kvm->arch.vpit)
1621 			r = 0;
1622 		break;
1623 	case KVM_IRQ_LINE: {
1624 		struct kvm_irq_level irq_event;
1625 
1626 		r = -EFAULT;
1627 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
1628 			goto out;
1629 		if (irqchip_in_kernel(kvm)) {
1630 			mutex_lock(&kvm->lock);
1631 			if (irq_event.irq < 16)
1632 				kvm_pic_set_irq(pic_irqchip(kvm),
1633 					irq_event.irq,
1634 					irq_event.level);
1635 			kvm_ioapic_set_irq(kvm->arch.vioapic,
1636 					irq_event.irq,
1637 					irq_event.level);
1638 			mutex_unlock(&kvm->lock);
1639 			r = 0;
1640 		}
1641 		break;
1642 	}
1643 	case KVM_GET_IRQCHIP: {
1644 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1645 		struct kvm_irqchip chip;
1646 
1647 		r = -EFAULT;
1648 		if (copy_from_user(&chip, argp, sizeof chip))
1649 			goto out;
1650 		r = -ENXIO;
1651 		if (!irqchip_in_kernel(kvm))
1652 			goto out;
1653 		r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1654 		if (r)
1655 			goto out;
1656 		r = -EFAULT;
1657 		if (copy_to_user(argp, &chip, sizeof chip))
1658 			goto out;
1659 		r = 0;
1660 		break;
1661 	}
1662 	case KVM_SET_IRQCHIP: {
1663 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1664 		struct kvm_irqchip chip;
1665 
1666 		r = -EFAULT;
1667 		if (copy_from_user(&chip, argp, sizeof chip))
1668 			goto out;
1669 		r = -ENXIO;
1670 		if (!irqchip_in_kernel(kvm))
1671 			goto out;
1672 		r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1673 		if (r)
1674 			goto out;
1675 		r = 0;
1676 		break;
1677 	}
1678 	case KVM_GET_PIT: {
1679 		struct kvm_pit_state ps;
1680 		r = -EFAULT;
1681 		if (copy_from_user(&ps, argp, sizeof ps))
1682 			goto out;
1683 		r = -ENXIO;
1684 		if (!kvm->arch.vpit)
1685 			goto out;
1686 		r = kvm_vm_ioctl_get_pit(kvm, &ps);
1687 		if (r)
1688 			goto out;
1689 		r = -EFAULT;
1690 		if (copy_to_user(argp, &ps, sizeof ps))
1691 			goto out;
1692 		r = 0;
1693 		break;
1694 	}
1695 	case KVM_SET_PIT: {
1696 		struct kvm_pit_state ps;
1697 		r = -EFAULT;
1698 		if (copy_from_user(&ps, argp, sizeof ps))
1699 			goto out;
1700 		r = -ENXIO;
1701 		if (!kvm->arch.vpit)
1702 			goto out;
1703 		r = kvm_vm_ioctl_set_pit(kvm, &ps);
1704 		if (r)
1705 			goto out;
1706 		r = 0;
1707 		break;
1708 	}
1709 	default:
1710 		;
1711 	}
1712 out:
1713 	return r;
1714 }
1715 
1716 static void kvm_init_msr_list(void)
1717 {
1718 	u32 dummy[2];
1719 	unsigned i, j;
1720 
1721 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1722 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1723 			continue;
1724 		if (j < i)
1725 			msrs_to_save[j] = msrs_to_save[i];
1726 		j++;
1727 	}
1728 	num_msrs_to_save = j;
1729 }
1730 
1731 /*
1732  * Only apic need an MMIO device hook, so shortcut now..
1733  */
1734 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1735 						gpa_t addr)
1736 {
1737 	struct kvm_io_device *dev;
1738 
1739 	if (vcpu->arch.apic) {
1740 		dev = &vcpu->arch.apic->dev;
1741 		if (dev->in_range(dev, addr))
1742 			return dev;
1743 	}
1744 	return NULL;
1745 }
1746 
1747 
1748 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1749 						gpa_t addr)
1750 {
1751 	struct kvm_io_device *dev;
1752 
1753 	dev = vcpu_find_pervcpu_dev(vcpu, addr);
1754 	if (dev == NULL)
1755 		dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1756 	return dev;
1757 }
1758 
1759 int emulator_read_std(unsigned long addr,
1760 			     void *val,
1761 			     unsigned int bytes,
1762 			     struct kvm_vcpu *vcpu)
1763 {
1764 	void *data = val;
1765 	int r = X86EMUL_CONTINUE;
1766 
1767 	while (bytes) {
1768 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1769 		unsigned offset = addr & (PAGE_SIZE-1);
1770 		unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1771 		int ret;
1772 
1773 		if (gpa == UNMAPPED_GVA) {
1774 			r = X86EMUL_PROPAGATE_FAULT;
1775 			goto out;
1776 		}
1777 		ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1778 		if (ret < 0) {
1779 			r = X86EMUL_UNHANDLEABLE;
1780 			goto out;
1781 		}
1782 
1783 		bytes -= tocopy;
1784 		data += tocopy;
1785 		addr += tocopy;
1786 	}
1787 out:
1788 	return r;
1789 }
1790 EXPORT_SYMBOL_GPL(emulator_read_std);
1791 
1792 static int emulator_read_emulated(unsigned long addr,
1793 				  void *val,
1794 				  unsigned int bytes,
1795 				  struct kvm_vcpu *vcpu)
1796 {
1797 	struct kvm_io_device *mmio_dev;
1798 	gpa_t                 gpa;
1799 
1800 	if (vcpu->mmio_read_completed) {
1801 		memcpy(val, vcpu->mmio_data, bytes);
1802 		vcpu->mmio_read_completed = 0;
1803 		return X86EMUL_CONTINUE;
1804 	}
1805 
1806 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1807 
1808 	/* For APIC access vmexit */
1809 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1810 		goto mmio;
1811 
1812 	if (emulator_read_std(addr, val, bytes, vcpu)
1813 			== X86EMUL_CONTINUE)
1814 		return X86EMUL_CONTINUE;
1815 	if (gpa == UNMAPPED_GVA)
1816 		return X86EMUL_PROPAGATE_FAULT;
1817 
1818 mmio:
1819 	/*
1820 	 * Is this MMIO handled locally?
1821 	 */
1822 	mutex_lock(&vcpu->kvm->lock);
1823 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1824 	if (mmio_dev) {
1825 		kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1826 		mutex_unlock(&vcpu->kvm->lock);
1827 		return X86EMUL_CONTINUE;
1828 	}
1829 	mutex_unlock(&vcpu->kvm->lock);
1830 
1831 	vcpu->mmio_needed = 1;
1832 	vcpu->mmio_phys_addr = gpa;
1833 	vcpu->mmio_size = bytes;
1834 	vcpu->mmio_is_write = 0;
1835 
1836 	return X86EMUL_UNHANDLEABLE;
1837 }
1838 
1839 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1840 			  const void *val, int bytes)
1841 {
1842 	int ret;
1843 
1844 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1845 	if (ret < 0)
1846 		return 0;
1847 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1848 	return 1;
1849 }
1850 
1851 static int emulator_write_emulated_onepage(unsigned long addr,
1852 					   const void *val,
1853 					   unsigned int bytes,
1854 					   struct kvm_vcpu *vcpu)
1855 {
1856 	struct kvm_io_device *mmio_dev;
1857 	gpa_t                 gpa;
1858 
1859 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1860 
1861 	if (gpa == UNMAPPED_GVA) {
1862 		kvm_inject_page_fault(vcpu, addr, 2);
1863 		return X86EMUL_PROPAGATE_FAULT;
1864 	}
1865 
1866 	/* For APIC access vmexit */
1867 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1868 		goto mmio;
1869 
1870 	if (emulator_write_phys(vcpu, gpa, val, bytes))
1871 		return X86EMUL_CONTINUE;
1872 
1873 mmio:
1874 	/*
1875 	 * Is this MMIO handled locally?
1876 	 */
1877 	mutex_lock(&vcpu->kvm->lock);
1878 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1879 	if (mmio_dev) {
1880 		kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1881 		mutex_unlock(&vcpu->kvm->lock);
1882 		return X86EMUL_CONTINUE;
1883 	}
1884 	mutex_unlock(&vcpu->kvm->lock);
1885 
1886 	vcpu->mmio_needed = 1;
1887 	vcpu->mmio_phys_addr = gpa;
1888 	vcpu->mmio_size = bytes;
1889 	vcpu->mmio_is_write = 1;
1890 	memcpy(vcpu->mmio_data, val, bytes);
1891 
1892 	return X86EMUL_CONTINUE;
1893 }
1894 
1895 int emulator_write_emulated(unsigned long addr,
1896 				   const void *val,
1897 				   unsigned int bytes,
1898 				   struct kvm_vcpu *vcpu)
1899 {
1900 	/* Crossing a page boundary? */
1901 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1902 		int rc, now;
1903 
1904 		now = -addr & ~PAGE_MASK;
1905 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1906 		if (rc != X86EMUL_CONTINUE)
1907 			return rc;
1908 		addr += now;
1909 		val += now;
1910 		bytes -= now;
1911 	}
1912 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1913 }
1914 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1915 
1916 static int emulator_cmpxchg_emulated(unsigned long addr,
1917 				     const void *old,
1918 				     const void *new,
1919 				     unsigned int bytes,
1920 				     struct kvm_vcpu *vcpu)
1921 {
1922 	static int reported;
1923 
1924 	if (!reported) {
1925 		reported = 1;
1926 		printk(KERN_WARNING "kvm: emulating exchange as write\n");
1927 	}
1928 #ifndef CONFIG_X86_64
1929 	/* guests cmpxchg8b have to be emulated atomically */
1930 	if (bytes == 8) {
1931 		gpa_t gpa;
1932 		struct page *page;
1933 		char *kaddr;
1934 		u64 val;
1935 
1936 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1937 
1938 		if (gpa == UNMAPPED_GVA ||
1939 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1940 			goto emul_write;
1941 
1942 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1943 			goto emul_write;
1944 
1945 		val = *(u64 *)new;
1946 
1947 		down_read(&current->mm->mmap_sem);
1948 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1949 		up_read(&current->mm->mmap_sem);
1950 
1951 		kaddr = kmap_atomic(page, KM_USER0);
1952 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1953 		kunmap_atomic(kaddr, KM_USER0);
1954 		kvm_release_page_dirty(page);
1955 	}
1956 emul_write:
1957 #endif
1958 
1959 	return emulator_write_emulated(addr, new, bytes, vcpu);
1960 }
1961 
1962 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1963 {
1964 	return kvm_x86_ops->get_segment_base(vcpu, seg);
1965 }
1966 
1967 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1968 {
1969 	return X86EMUL_CONTINUE;
1970 }
1971 
1972 int emulate_clts(struct kvm_vcpu *vcpu)
1973 {
1974 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1975 	return X86EMUL_CONTINUE;
1976 }
1977 
1978 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1979 {
1980 	struct kvm_vcpu *vcpu = ctxt->vcpu;
1981 
1982 	switch (dr) {
1983 	case 0 ... 3:
1984 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
1985 		return X86EMUL_CONTINUE;
1986 	default:
1987 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
1988 		return X86EMUL_UNHANDLEABLE;
1989 	}
1990 }
1991 
1992 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1993 {
1994 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1995 	int exception;
1996 
1997 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1998 	if (exception) {
1999 		/* FIXME: better handling */
2000 		return X86EMUL_UNHANDLEABLE;
2001 	}
2002 	return X86EMUL_CONTINUE;
2003 }
2004 
2005 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2006 {
2007 	static int reported;
2008 	u8 opcodes[4];
2009 	unsigned long rip = vcpu->arch.rip;
2010 	unsigned long rip_linear;
2011 
2012 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2013 
2014 	if (reported)
2015 		return;
2016 
2017 	emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2018 
2019 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2020 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2021 	reported = 1;
2022 }
2023 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2024 
2025 static struct x86_emulate_ops emulate_ops = {
2026 	.read_std            = emulator_read_std,
2027 	.read_emulated       = emulator_read_emulated,
2028 	.write_emulated      = emulator_write_emulated,
2029 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2030 };
2031 
2032 int emulate_instruction(struct kvm_vcpu *vcpu,
2033 			struct kvm_run *run,
2034 			unsigned long cr2,
2035 			u16 error_code,
2036 			int emulation_type)
2037 {
2038 	int r;
2039 	struct decode_cache *c;
2040 
2041 	vcpu->arch.mmio_fault_cr2 = cr2;
2042 	kvm_x86_ops->cache_regs(vcpu);
2043 
2044 	vcpu->mmio_is_write = 0;
2045 	vcpu->arch.pio.string = 0;
2046 
2047 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2048 		int cs_db, cs_l;
2049 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2050 
2051 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2052 		vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2053 		vcpu->arch.emulate_ctxt.mode =
2054 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2055 			? X86EMUL_MODE_REAL : cs_l
2056 			? X86EMUL_MODE_PROT64 :	cs_db
2057 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2058 
2059 		if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2060 			vcpu->arch.emulate_ctxt.cs_base = 0;
2061 			vcpu->arch.emulate_ctxt.ds_base = 0;
2062 			vcpu->arch.emulate_ctxt.es_base = 0;
2063 			vcpu->arch.emulate_ctxt.ss_base = 0;
2064 		} else {
2065 			vcpu->arch.emulate_ctxt.cs_base =
2066 					get_segment_base(vcpu, VCPU_SREG_CS);
2067 			vcpu->arch.emulate_ctxt.ds_base =
2068 					get_segment_base(vcpu, VCPU_SREG_DS);
2069 			vcpu->arch.emulate_ctxt.es_base =
2070 					get_segment_base(vcpu, VCPU_SREG_ES);
2071 			vcpu->arch.emulate_ctxt.ss_base =
2072 					get_segment_base(vcpu, VCPU_SREG_SS);
2073 		}
2074 
2075 		vcpu->arch.emulate_ctxt.gs_base =
2076 					get_segment_base(vcpu, VCPU_SREG_GS);
2077 		vcpu->arch.emulate_ctxt.fs_base =
2078 					get_segment_base(vcpu, VCPU_SREG_FS);
2079 
2080 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2081 
2082 		/* Reject the instructions other than VMCALL/VMMCALL when
2083 		 * try to emulate invalid opcode */
2084 		c = &vcpu->arch.emulate_ctxt.decode;
2085 		if ((emulation_type & EMULTYPE_TRAP_UD) &&
2086 		    (!(c->twobyte && c->b == 0x01 &&
2087 		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2088 		       c->modrm_mod == 3 && c->modrm_rm == 1)))
2089 			return EMULATE_FAIL;
2090 
2091 		++vcpu->stat.insn_emulation;
2092 		if (r)  {
2093 			++vcpu->stat.insn_emulation_fail;
2094 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2095 				return EMULATE_DONE;
2096 			return EMULATE_FAIL;
2097 		}
2098 	}
2099 
2100 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2101 
2102 	if (vcpu->arch.pio.string)
2103 		return EMULATE_DO_MMIO;
2104 
2105 	if ((r || vcpu->mmio_is_write) && run) {
2106 		run->exit_reason = KVM_EXIT_MMIO;
2107 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2108 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2109 		run->mmio.len = vcpu->mmio_size;
2110 		run->mmio.is_write = vcpu->mmio_is_write;
2111 	}
2112 
2113 	if (r) {
2114 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2115 			return EMULATE_DONE;
2116 		if (!vcpu->mmio_needed) {
2117 			kvm_report_emulation_failure(vcpu, "mmio");
2118 			return EMULATE_FAIL;
2119 		}
2120 		return EMULATE_DO_MMIO;
2121 	}
2122 
2123 	kvm_x86_ops->decache_regs(vcpu);
2124 	kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2125 
2126 	if (vcpu->mmio_is_write) {
2127 		vcpu->mmio_needed = 0;
2128 		return EMULATE_DO_MMIO;
2129 	}
2130 
2131 	return EMULATE_DONE;
2132 }
2133 EXPORT_SYMBOL_GPL(emulate_instruction);
2134 
2135 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2136 {
2137 	int i;
2138 
2139 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2140 		if (vcpu->arch.pio.guest_pages[i]) {
2141 			kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2142 			vcpu->arch.pio.guest_pages[i] = NULL;
2143 		}
2144 }
2145 
2146 static int pio_copy_data(struct kvm_vcpu *vcpu)
2147 {
2148 	void *p = vcpu->arch.pio_data;
2149 	void *q;
2150 	unsigned bytes;
2151 	int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2152 
2153 	q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2154 		 PAGE_KERNEL);
2155 	if (!q) {
2156 		free_pio_guest_pages(vcpu);
2157 		return -ENOMEM;
2158 	}
2159 	q += vcpu->arch.pio.guest_page_offset;
2160 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2161 	if (vcpu->arch.pio.in)
2162 		memcpy(q, p, bytes);
2163 	else
2164 		memcpy(p, q, bytes);
2165 	q -= vcpu->arch.pio.guest_page_offset;
2166 	vunmap(q);
2167 	free_pio_guest_pages(vcpu);
2168 	return 0;
2169 }
2170 
2171 int complete_pio(struct kvm_vcpu *vcpu)
2172 {
2173 	struct kvm_pio_request *io = &vcpu->arch.pio;
2174 	long delta;
2175 	int r;
2176 
2177 	kvm_x86_ops->cache_regs(vcpu);
2178 
2179 	if (!io->string) {
2180 		if (io->in)
2181 			memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2182 			       io->size);
2183 	} else {
2184 		if (io->in) {
2185 			r = pio_copy_data(vcpu);
2186 			if (r) {
2187 				kvm_x86_ops->cache_regs(vcpu);
2188 				return r;
2189 			}
2190 		}
2191 
2192 		delta = 1;
2193 		if (io->rep) {
2194 			delta *= io->cur_count;
2195 			/*
2196 			 * The size of the register should really depend on
2197 			 * current address size.
2198 			 */
2199 			vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2200 		}
2201 		if (io->down)
2202 			delta = -delta;
2203 		delta *= io->size;
2204 		if (io->in)
2205 			vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2206 		else
2207 			vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2208 	}
2209 
2210 	kvm_x86_ops->decache_regs(vcpu);
2211 
2212 	io->count -= io->cur_count;
2213 	io->cur_count = 0;
2214 
2215 	return 0;
2216 }
2217 
2218 static void kernel_pio(struct kvm_io_device *pio_dev,
2219 		       struct kvm_vcpu *vcpu,
2220 		       void *pd)
2221 {
2222 	/* TODO: String I/O for in kernel device */
2223 
2224 	mutex_lock(&vcpu->kvm->lock);
2225 	if (vcpu->arch.pio.in)
2226 		kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2227 				  vcpu->arch.pio.size,
2228 				  pd);
2229 	else
2230 		kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2231 				   vcpu->arch.pio.size,
2232 				   pd);
2233 	mutex_unlock(&vcpu->kvm->lock);
2234 }
2235 
2236 static void pio_string_write(struct kvm_io_device *pio_dev,
2237 			     struct kvm_vcpu *vcpu)
2238 {
2239 	struct kvm_pio_request *io = &vcpu->arch.pio;
2240 	void *pd = vcpu->arch.pio_data;
2241 	int i;
2242 
2243 	mutex_lock(&vcpu->kvm->lock);
2244 	for (i = 0; i < io->cur_count; i++) {
2245 		kvm_iodevice_write(pio_dev, io->port,
2246 				   io->size,
2247 				   pd);
2248 		pd += io->size;
2249 	}
2250 	mutex_unlock(&vcpu->kvm->lock);
2251 }
2252 
2253 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2254 					       gpa_t addr)
2255 {
2256 	return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2257 }
2258 
2259 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2260 		  int size, unsigned port)
2261 {
2262 	struct kvm_io_device *pio_dev;
2263 
2264 	vcpu->run->exit_reason = KVM_EXIT_IO;
2265 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2266 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2267 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2268 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2269 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2270 	vcpu->arch.pio.in = in;
2271 	vcpu->arch.pio.string = 0;
2272 	vcpu->arch.pio.down = 0;
2273 	vcpu->arch.pio.guest_page_offset = 0;
2274 	vcpu->arch.pio.rep = 0;
2275 
2276 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2277 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2278 			    handler);
2279 	else
2280 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2281 			    handler);
2282 
2283 	kvm_x86_ops->cache_regs(vcpu);
2284 	memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2285 	kvm_x86_ops->decache_regs(vcpu);
2286 
2287 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2288 
2289 	pio_dev = vcpu_find_pio_dev(vcpu, port);
2290 	if (pio_dev) {
2291 		kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2292 		complete_pio(vcpu);
2293 		return 1;
2294 	}
2295 	return 0;
2296 }
2297 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2298 
2299 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2300 		  int size, unsigned long count, int down,
2301 		  gva_t address, int rep, unsigned port)
2302 {
2303 	unsigned now, in_page;
2304 	int i, ret = 0;
2305 	int nr_pages = 1;
2306 	struct page *page;
2307 	struct kvm_io_device *pio_dev;
2308 
2309 	vcpu->run->exit_reason = KVM_EXIT_IO;
2310 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2311 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2312 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2313 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2314 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2315 	vcpu->arch.pio.in = in;
2316 	vcpu->arch.pio.string = 1;
2317 	vcpu->arch.pio.down = down;
2318 	vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2319 	vcpu->arch.pio.rep = rep;
2320 
2321 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2322 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2323 			    handler);
2324 	else
2325 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2326 			    handler);
2327 
2328 	if (!count) {
2329 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2330 		return 1;
2331 	}
2332 
2333 	if (!down)
2334 		in_page = PAGE_SIZE - offset_in_page(address);
2335 	else
2336 		in_page = offset_in_page(address) + size;
2337 	now = min(count, (unsigned long)in_page / size);
2338 	if (!now) {
2339 		/*
2340 		 * String I/O straddles page boundary.  Pin two guest pages
2341 		 * so that we satisfy atomicity constraints.  Do just one
2342 		 * transaction to avoid complexity.
2343 		 */
2344 		nr_pages = 2;
2345 		now = 1;
2346 	}
2347 	if (down) {
2348 		/*
2349 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2350 		 */
2351 		pr_unimpl(vcpu, "guest string pio down\n");
2352 		kvm_inject_gp(vcpu, 0);
2353 		return 1;
2354 	}
2355 	vcpu->run->io.count = now;
2356 	vcpu->arch.pio.cur_count = now;
2357 
2358 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2359 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2360 
2361 	for (i = 0; i < nr_pages; ++i) {
2362 		page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2363 		vcpu->arch.pio.guest_pages[i] = page;
2364 		if (!page) {
2365 			kvm_inject_gp(vcpu, 0);
2366 			free_pio_guest_pages(vcpu);
2367 			return 1;
2368 		}
2369 	}
2370 
2371 	pio_dev = vcpu_find_pio_dev(vcpu, port);
2372 	if (!vcpu->arch.pio.in) {
2373 		/* string PIO write */
2374 		ret = pio_copy_data(vcpu);
2375 		if (ret >= 0 && pio_dev) {
2376 			pio_string_write(pio_dev, vcpu);
2377 			complete_pio(vcpu);
2378 			if (vcpu->arch.pio.count == 0)
2379 				ret = 1;
2380 		}
2381 	} else if (pio_dev)
2382 		pr_unimpl(vcpu, "no string pio read support yet, "
2383 		       "port %x size %d count %ld\n",
2384 			port, size, count);
2385 
2386 	return ret;
2387 }
2388 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2389 
2390 int kvm_arch_init(void *opaque)
2391 {
2392 	int r;
2393 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2394 
2395 	if (kvm_x86_ops) {
2396 		printk(KERN_ERR "kvm: already loaded the other module\n");
2397 		r = -EEXIST;
2398 		goto out;
2399 	}
2400 
2401 	if (!ops->cpu_has_kvm_support()) {
2402 		printk(KERN_ERR "kvm: no hardware support\n");
2403 		r = -EOPNOTSUPP;
2404 		goto out;
2405 	}
2406 	if (ops->disabled_by_bios()) {
2407 		printk(KERN_ERR "kvm: disabled by bios\n");
2408 		r = -EOPNOTSUPP;
2409 		goto out;
2410 	}
2411 
2412 	r = kvm_mmu_module_init();
2413 	if (r)
2414 		goto out;
2415 
2416 	kvm_init_msr_list();
2417 
2418 	kvm_x86_ops = ops;
2419 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2420 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2421 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2422 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
2423 	return 0;
2424 
2425 out:
2426 	return r;
2427 }
2428 
2429 void kvm_arch_exit(void)
2430 {
2431 	kvm_x86_ops = NULL;
2432 	kvm_mmu_module_exit();
2433 }
2434 
2435 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2436 {
2437 	++vcpu->stat.halt_exits;
2438 	KVMTRACE_0D(HLT, vcpu, handler);
2439 	if (irqchip_in_kernel(vcpu->kvm)) {
2440 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2441 		up_read(&vcpu->kvm->slots_lock);
2442 		kvm_vcpu_block(vcpu);
2443 		down_read(&vcpu->kvm->slots_lock);
2444 		if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2445 			return -EINTR;
2446 		return 1;
2447 	} else {
2448 		vcpu->run->exit_reason = KVM_EXIT_HLT;
2449 		return 0;
2450 	}
2451 }
2452 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2453 
2454 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2455 			   unsigned long a1)
2456 {
2457 	if (is_long_mode(vcpu))
2458 		return a0;
2459 	else
2460 		return a0 | ((gpa_t)a1 << 32);
2461 }
2462 
2463 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2464 {
2465 	unsigned long nr, a0, a1, a2, a3, ret;
2466 	int r = 1;
2467 
2468 	kvm_x86_ops->cache_regs(vcpu);
2469 
2470 	nr = vcpu->arch.regs[VCPU_REGS_RAX];
2471 	a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2472 	a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2473 	a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2474 	a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2475 
2476 	KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2477 
2478 	if (!is_long_mode(vcpu)) {
2479 		nr &= 0xFFFFFFFF;
2480 		a0 &= 0xFFFFFFFF;
2481 		a1 &= 0xFFFFFFFF;
2482 		a2 &= 0xFFFFFFFF;
2483 		a3 &= 0xFFFFFFFF;
2484 	}
2485 
2486 	switch (nr) {
2487 	case KVM_HC_VAPIC_POLL_IRQ:
2488 		ret = 0;
2489 		break;
2490 	case KVM_HC_MMU_OP:
2491 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2492 		break;
2493 	default:
2494 		ret = -KVM_ENOSYS;
2495 		break;
2496 	}
2497 	vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2498 	kvm_x86_ops->decache_regs(vcpu);
2499 	++vcpu->stat.hypercalls;
2500 	return r;
2501 }
2502 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2503 
2504 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2505 {
2506 	char instruction[3];
2507 	int ret = 0;
2508 
2509 
2510 	/*
2511 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
2512 	 * to ensure that the updated hypercall appears atomically across all
2513 	 * VCPUs.
2514 	 */
2515 	kvm_mmu_zap_all(vcpu->kvm);
2516 
2517 	kvm_x86_ops->cache_regs(vcpu);
2518 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
2519 	if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2520 	    != X86EMUL_CONTINUE)
2521 		ret = -EFAULT;
2522 
2523 	return ret;
2524 }
2525 
2526 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2527 {
2528 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2529 }
2530 
2531 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2532 {
2533 	struct descriptor_table dt = { limit, base };
2534 
2535 	kvm_x86_ops->set_gdt(vcpu, &dt);
2536 }
2537 
2538 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2539 {
2540 	struct descriptor_table dt = { limit, base };
2541 
2542 	kvm_x86_ops->set_idt(vcpu, &dt);
2543 }
2544 
2545 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2546 		   unsigned long *rflags)
2547 {
2548 	kvm_lmsw(vcpu, msw);
2549 	*rflags = kvm_x86_ops->get_rflags(vcpu);
2550 }
2551 
2552 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2553 {
2554 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2555 	switch (cr) {
2556 	case 0:
2557 		return vcpu->arch.cr0;
2558 	case 2:
2559 		return vcpu->arch.cr2;
2560 	case 3:
2561 		return vcpu->arch.cr3;
2562 	case 4:
2563 		return vcpu->arch.cr4;
2564 	case 8:
2565 		return kvm_get_cr8(vcpu);
2566 	default:
2567 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2568 		return 0;
2569 	}
2570 }
2571 
2572 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2573 		     unsigned long *rflags)
2574 {
2575 	switch (cr) {
2576 	case 0:
2577 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2578 		*rflags = kvm_x86_ops->get_rflags(vcpu);
2579 		break;
2580 	case 2:
2581 		vcpu->arch.cr2 = val;
2582 		break;
2583 	case 3:
2584 		kvm_set_cr3(vcpu, val);
2585 		break;
2586 	case 4:
2587 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2588 		break;
2589 	case 8:
2590 		kvm_set_cr8(vcpu, val & 0xfUL);
2591 		break;
2592 	default:
2593 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2594 	}
2595 }
2596 
2597 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2598 {
2599 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2600 	int j, nent = vcpu->arch.cpuid_nent;
2601 
2602 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2603 	/* when no next entry is found, the current entry[i] is reselected */
2604 	for (j = i + 1; j == i; j = (j + 1) % nent) {
2605 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2606 		if (ej->function == e->function) {
2607 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2608 			return j;
2609 		}
2610 	}
2611 	return 0; /* silence gcc, even though control never reaches here */
2612 }
2613 
2614 /* find an entry with matching function, matching index (if needed), and that
2615  * should be read next (if it's stateful) */
2616 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2617 	u32 function, u32 index)
2618 {
2619 	if (e->function != function)
2620 		return 0;
2621 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2622 		return 0;
2623 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2624 		!(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2625 		return 0;
2626 	return 1;
2627 }
2628 
2629 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2630 {
2631 	int i;
2632 	u32 function, index;
2633 	struct kvm_cpuid_entry2 *e, *best;
2634 
2635 	kvm_x86_ops->cache_regs(vcpu);
2636 	function = vcpu->arch.regs[VCPU_REGS_RAX];
2637 	index = vcpu->arch.regs[VCPU_REGS_RCX];
2638 	vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2639 	vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2640 	vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2641 	vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2642 	best = NULL;
2643 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2644 		e = &vcpu->arch.cpuid_entries[i];
2645 		if (is_matching_cpuid_entry(e, function, index)) {
2646 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2647 				move_to_next_stateful_cpuid_entry(vcpu, i);
2648 			best = e;
2649 			break;
2650 		}
2651 		/*
2652 		 * Both basic or both extended?
2653 		 */
2654 		if (((e->function ^ function) & 0x80000000) == 0)
2655 			if (!best || e->function > best->function)
2656 				best = e;
2657 	}
2658 	if (best) {
2659 		vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2660 		vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2661 		vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2662 		vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2663 	}
2664 	kvm_x86_ops->decache_regs(vcpu);
2665 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2666 	KVMTRACE_5D(CPUID, vcpu, function,
2667 		    (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2668 		    (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2669 		    (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2670 		    (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2671 }
2672 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2673 
2674 /*
2675  * Check if userspace requested an interrupt window, and that the
2676  * interrupt window is open.
2677  *
2678  * No need to exit to userspace if we already have an interrupt queued.
2679  */
2680 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2681 					  struct kvm_run *kvm_run)
2682 {
2683 	return (!vcpu->arch.irq_summary &&
2684 		kvm_run->request_interrupt_window &&
2685 		vcpu->arch.interrupt_window_open &&
2686 		(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2687 }
2688 
2689 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2690 			      struct kvm_run *kvm_run)
2691 {
2692 	kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2693 	kvm_run->cr8 = kvm_get_cr8(vcpu);
2694 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
2695 	if (irqchip_in_kernel(vcpu->kvm))
2696 		kvm_run->ready_for_interrupt_injection = 1;
2697 	else
2698 		kvm_run->ready_for_interrupt_injection =
2699 					(vcpu->arch.interrupt_window_open &&
2700 					 vcpu->arch.irq_summary == 0);
2701 }
2702 
2703 static void vapic_enter(struct kvm_vcpu *vcpu)
2704 {
2705 	struct kvm_lapic *apic = vcpu->arch.apic;
2706 	struct page *page;
2707 
2708 	if (!apic || !apic->vapic_addr)
2709 		return;
2710 
2711 	down_read(&current->mm->mmap_sem);
2712 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2713 	up_read(&current->mm->mmap_sem);
2714 
2715 	vcpu->arch.apic->vapic_page = page;
2716 }
2717 
2718 static void vapic_exit(struct kvm_vcpu *vcpu)
2719 {
2720 	struct kvm_lapic *apic = vcpu->arch.apic;
2721 
2722 	if (!apic || !apic->vapic_addr)
2723 		return;
2724 
2725 	kvm_release_page_dirty(apic->vapic_page);
2726 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2727 }
2728 
2729 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2730 {
2731 	int r;
2732 
2733 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2734 		pr_debug("vcpu %d received sipi with vector # %x\n",
2735 		       vcpu->vcpu_id, vcpu->arch.sipi_vector);
2736 		kvm_lapic_reset(vcpu);
2737 		r = kvm_x86_ops->vcpu_reset(vcpu);
2738 		if (r)
2739 			return r;
2740 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2741 	}
2742 
2743 	down_read(&vcpu->kvm->slots_lock);
2744 	vapic_enter(vcpu);
2745 
2746 preempted:
2747 	if (vcpu->guest_debug.enabled)
2748 		kvm_x86_ops->guest_debug_pre(vcpu);
2749 
2750 again:
2751 	if (vcpu->requests)
2752 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2753 			kvm_mmu_unload(vcpu);
2754 
2755 	r = kvm_mmu_reload(vcpu);
2756 	if (unlikely(r))
2757 		goto out;
2758 
2759 	if (vcpu->requests) {
2760 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2761 			__kvm_migrate_apic_timer(vcpu);
2762 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2763 				       &vcpu->requests)) {
2764 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2765 			r = 0;
2766 			goto out;
2767 		}
2768 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2769 			kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2770 			r = 0;
2771 			goto out;
2772 		}
2773 	}
2774 
2775 	kvm_inject_pending_timer_irqs(vcpu);
2776 
2777 	preempt_disable();
2778 
2779 	kvm_x86_ops->prepare_guest_switch(vcpu);
2780 	kvm_load_guest_fpu(vcpu);
2781 
2782 	local_irq_disable();
2783 
2784 	if (need_resched()) {
2785 		local_irq_enable();
2786 		preempt_enable();
2787 		r = 1;
2788 		goto out;
2789 	}
2790 
2791 	if (vcpu->requests)
2792 		if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
2793 			local_irq_enable();
2794 			preempt_enable();
2795 			r = 1;
2796 			goto out;
2797 		}
2798 
2799 	if (signal_pending(current)) {
2800 		local_irq_enable();
2801 		preempt_enable();
2802 		r = -EINTR;
2803 		kvm_run->exit_reason = KVM_EXIT_INTR;
2804 		++vcpu->stat.signal_exits;
2805 		goto out;
2806 	}
2807 
2808 	vcpu->guest_mode = 1;
2809 	/*
2810 	 * Make sure that guest_mode assignment won't happen after
2811 	 * testing the pending IRQ vector bitmap.
2812 	 */
2813 	smp_wmb();
2814 
2815 	if (vcpu->arch.exception.pending)
2816 		__queue_exception(vcpu);
2817 	else if (irqchip_in_kernel(vcpu->kvm))
2818 		kvm_x86_ops->inject_pending_irq(vcpu);
2819 	else
2820 		kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2821 
2822 	kvm_lapic_sync_to_vapic(vcpu);
2823 
2824 	up_read(&vcpu->kvm->slots_lock);
2825 
2826 	kvm_guest_enter();
2827 
2828 	if (vcpu->requests)
2829 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2830 			kvm_x86_ops->tlb_flush(vcpu);
2831 
2832 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2833 	kvm_x86_ops->run(vcpu, kvm_run);
2834 
2835 	vcpu->guest_mode = 0;
2836 	local_irq_enable();
2837 
2838 	++vcpu->stat.exits;
2839 
2840 	/*
2841 	 * We must have an instruction between local_irq_enable() and
2842 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2843 	 * the interrupt shadow.  The stat.exits increment will do nicely.
2844 	 * But we need to prevent reordering, hence this barrier():
2845 	 */
2846 	barrier();
2847 
2848 	kvm_guest_exit();
2849 
2850 	preempt_enable();
2851 
2852 	down_read(&vcpu->kvm->slots_lock);
2853 
2854 	/*
2855 	 * Profile KVM exit RIPs:
2856 	 */
2857 	if (unlikely(prof_on == KVM_PROFILING)) {
2858 		kvm_x86_ops->cache_regs(vcpu);
2859 		profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2860 	}
2861 
2862 	if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2863 		vcpu->arch.exception.pending = false;
2864 
2865 	kvm_lapic_sync_from_vapic(vcpu);
2866 
2867 	r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2868 
2869 	if (r > 0) {
2870 		if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2871 			r = -EINTR;
2872 			kvm_run->exit_reason = KVM_EXIT_INTR;
2873 			++vcpu->stat.request_irq_exits;
2874 			goto out;
2875 		}
2876 		if (!need_resched())
2877 			goto again;
2878 	}
2879 
2880 out:
2881 	up_read(&vcpu->kvm->slots_lock);
2882 	if (r > 0) {
2883 		kvm_resched(vcpu);
2884 		down_read(&vcpu->kvm->slots_lock);
2885 		goto preempted;
2886 	}
2887 
2888 	post_kvm_run_save(vcpu, kvm_run);
2889 
2890 	down_read(&vcpu->kvm->slots_lock);
2891 	vapic_exit(vcpu);
2892 	up_read(&vcpu->kvm->slots_lock);
2893 
2894 	return r;
2895 }
2896 
2897 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2898 {
2899 	int r;
2900 	sigset_t sigsaved;
2901 
2902 	vcpu_load(vcpu);
2903 
2904 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2905 		kvm_vcpu_block(vcpu);
2906 		vcpu_put(vcpu);
2907 		return -EAGAIN;
2908 	}
2909 
2910 	if (vcpu->sigset_active)
2911 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2912 
2913 	/* re-sync apic's tpr */
2914 	if (!irqchip_in_kernel(vcpu->kvm))
2915 		kvm_set_cr8(vcpu, kvm_run->cr8);
2916 
2917 	if (vcpu->arch.pio.cur_count) {
2918 		r = complete_pio(vcpu);
2919 		if (r)
2920 			goto out;
2921 	}
2922 #if CONFIG_HAS_IOMEM
2923 	if (vcpu->mmio_needed) {
2924 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2925 		vcpu->mmio_read_completed = 1;
2926 		vcpu->mmio_needed = 0;
2927 
2928 		down_read(&vcpu->kvm->slots_lock);
2929 		r = emulate_instruction(vcpu, kvm_run,
2930 					vcpu->arch.mmio_fault_cr2, 0,
2931 					EMULTYPE_NO_DECODE);
2932 		up_read(&vcpu->kvm->slots_lock);
2933 		if (r == EMULATE_DO_MMIO) {
2934 			/*
2935 			 * Read-modify-write.  Back to userspace.
2936 			 */
2937 			r = 0;
2938 			goto out;
2939 		}
2940 	}
2941 #endif
2942 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2943 		kvm_x86_ops->cache_regs(vcpu);
2944 		vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2945 		kvm_x86_ops->decache_regs(vcpu);
2946 	}
2947 
2948 	r = __vcpu_run(vcpu, kvm_run);
2949 
2950 out:
2951 	if (vcpu->sigset_active)
2952 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2953 
2954 	vcpu_put(vcpu);
2955 	return r;
2956 }
2957 
2958 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2959 {
2960 	vcpu_load(vcpu);
2961 
2962 	kvm_x86_ops->cache_regs(vcpu);
2963 
2964 	regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2965 	regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2966 	regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2967 	regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2968 	regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2969 	regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2970 	regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2971 	regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2972 #ifdef CONFIG_X86_64
2973 	regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2974 	regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2975 	regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2976 	regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2977 	regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2978 	regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2979 	regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2980 	regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2981 #endif
2982 
2983 	regs->rip = vcpu->arch.rip;
2984 	regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2985 
2986 	/*
2987 	 * Don't leak debug flags in case they were set for guest debugging
2988 	 */
2989 	if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2990 		regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2991 
2992 	vcpu_put(vcpu);
2993 
2994 	return 0;
2995 }
2996 
2997 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2998 {
2999 	vcpu_load(vcpu);
3000 
3001 	vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3002 	vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3003 	vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3004 	vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3005 	vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3006 	vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3007 	vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3008 	vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3009 #ifdef CONFIG_X86_64
3010 	vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3011 	vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3012 	vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3013 	vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3014 	vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3015 	vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3016 	vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3017 	vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3018 #endif
3019 
3020 	vcpu->arch.rip = regs->rip;
3021 	kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3022 
3023 	kvm_x86_ops->decache_regs(vcpu);
3024 
3025 	vcpu->arch.exception.pending = false;
3026 
3027 	vcpu_put(vcpu);
3028 
3029 	return 0;
3030 }
3031 
3032 static void get_segment(struct kvm_vcpu *vcpu,
3033 			struct kvm_segment *var, int seg)
3034 {
3035 	kvm_x86_ops->get_segment(vcpu, var, seg);
3036 }
3037 
3038 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3039 {
3040 	struct kvm_segment cs;
3041 
3042 	get_segment(vcpu, &cs, VCPU_SREG_CS);
3043 	*db = cs.db;
3044 	*l = cs.l;
3045 }
3046 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3047 
3048 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3049 				  struct kvm_sregs *sregs)
3050 {
3051 	struct descriptor_table dt;
3052 	int pending_vec;
3053 
3054 	vcpu_load(vcpu);
3055 
3056 	get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3057 	get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3058 	get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3059 	get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3060 	get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3061 	get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3062 
3063 	get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3064 	get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3065 
3066 	kvm_x86_ops->get_idt(vcpu, &dt);
3067 	sregs->idt.limit = dt.limit;
3068 	sregs->idt.base = dt.base;
3069 	kvm_x86_ops->get_gdt(vcpu, &dt);
3070 	sregs->gdt.limit = dt.limit;
3071 	sregs->gdt.base = dt.base;
3072 
3073 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3074 	sregs->cr0 = vcpu->arch.cr0;
3075 	sregs->cr2 = vcpu->arch.cr2;
3076 	sregs->cr3 = vcpu->arch.cr3;
3077 	sregs->cr4 = vcpu->arch.cr4;
3078 	sregs->cr8 = kvm_get_cr8(vcpu);
3079 	sregs->efer = vcpu->arch.shadow_efer;
3080 	sregs->apic_base = kvm_get_apic_base(vcpu);
3081 
3082 	if (irqchip_in_kernel(vcpu->kvm)) {
3083 		memset(sregs->interrupt_bitmap, 0,
3084 		       sizeof sregs->interrupt_bitmap);
3085 		pending_vec = kvm_x86_ops->get_irq(vcpu);
3086 		if (pending_vec >= 0)
3087 			set_bit(pending_vec,
3088 				(unsigned long *)sregs->interrupt_bitmap);
3089 	} else
3090 		memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3091 		       sizeof sregs->interrupt_bitmap);
3092 
3093 	vcpu_put(vcpu);
3094 
3095 	return 0;
3096 }
3097 
3098 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3099 				    struct kvm_mp_state *mp_state)
3100 {
3101 	vcpu_load(vcpu);
3102 	mp_state->mp_state = vcpu->arch.mp_state;
3103 	vcpu_put(vcpu);
3104 	return 0;
3105 }
3106 
3107 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3108 				    struct kvm_mp_state *mp_state)
3109 {
3110 	vcpu_load(vcpu);
3111 	vcpu->arch.mp_state = mp_state->mp_state;
3112 	vcpu_put(vcpu);
3113 	return 0;
3114 }
3115 
3116 static void set_segment(struct kvm_vcpu *vcpu,
3117 			struct kvm_segment *var, int seg)
3118 {
3119 	kvm_x86_ops->set_segment(vcpu, var, seg);
3120 }
3121 
3122 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3123 				   struct kvm_segment *kvm_desct)
3124 {
3125 	kvm_desct->base = seg_desc->base0;
3126 	kvm_desct->base |= seg_desc->base1 << 16;
3127 	kvm_desct->base |= seg_desc->base2 << 24;
3128 	kvm_desct->limit = seg_desc->limit0;
3129 	kvm_desct->limit |= seg_desc->limit << 16;
3130 	kvm_desct->selector = selector;
3131 	kvm_desct->type = seg_desc->type;
3132 	kvm_desct->present = seg_desc->p;
3133 	kvm_desct->dpl = seg_desc->dpl;
3134 	kvm_desct->db = seg_desc->d;
3135 	kvm_desct->s = seg_desc->s;
3136 	kvm_desct->l = seg_desc->l;
3137 	kvm_desct->g = seg_desc->g;
3138 	kvm_desct->avl = seg_desc->avl;
3139 	if (!selector)
3140 		kvm_desct->unusable = 1;
3141 	else
3142 		kvm_desct->unusable = 0;
3143 	kvm_desct->padding = 0;
3144 }
3145 
3146 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3147 					   u16 selector,
3148 					   struct descriptor_table *dtable)
3149 {
3150 	if (selector & 1 << 2) {
3151 		struct kvm_segment kvm_seg;
3152 
3153 		get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3154 
3155 		if (kvm_seg.unusable)
3156 			dtable->limit = 0;
3157 		else
3158 			dtable->limit = kvm_seg.limit;
3159 		dtable->base = kvm_seg.base;
3160 	}
3161 	else
3162 		kvm_x86_ops->get_gdt(vcpu, dtable);
3163 }
3164 
3165 /* allowed just for 8 bytes segments */
3166 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3167 					 struct desc_struct *seg_desc)
3168 {
3169 	struct descriptor_table dtable;
3170 	u16 index = selector >> 3;
3171 
3172 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3173 
3174 	if (dtable.limit < index * 8 + 7) {
3175 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3176 		return 1;
3177 	}
3178 	return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3179 }
3180 
3181 /* allowed just for 8 bytes segments */
3182 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3183 					 struct desc_struct *seg_desc)
3184 {
3185 	struct descriptor_table dtable;
3186 	u16 index = selector >> 3;
3187 
3188 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3189 
3190 	if (dtable.limit < index * 8 + 7)
3191 		return 1;
3192 	return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3193 }
3194 
3195 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3196 			     struct desc_struct *seg_desc)
3197 {
3198 	u32 base_addr;
3199 
3200 	base_addr = seg_desc->base0;
3201 	base_addr |= (seg_desc->base1 << 16);
3202 	base_addr |= (seg_desc->base2 << 24);
3203 
3204 	return base_addr;
3205 }
3206 
3207 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3208 			      struct desc_struct *seg_desc,
3209 			      struct tss_segment_32 *tss)
3210 {
3211 	u32 base_addr;
3212 
3213 	base_addr = get_tss_base_addr(vcpu, seg_desc);
3214 
3215 	return kvm_read_guest(vcpu->kvm, base_addr, tss,
3216 			      sizeof(struct tss_segment_32));
3217 }
3218 
3219 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3220 			      struct desc_struct *seg_desc,
3221 			      struct tss_segment_32 *tss)
3222 {
3223 	u32 base_addr;
3224 
3225 	base_addr = get_tss_base_addr(vcpu, seg_desc);
3226 
3227 	return kvm_write_guest(vcpu->kvm, base_addr, tss,
3228 			       sizeof(struct tss_segment_32));
3229 }
3230 
3231 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3232 			      struct desc_struct *seg_desc,
3233 			      struct tss_segment_16 *tss)
3234 {
3235 	u32 base_addr;
3236 
3237 	base_addr = get_tss_base_addr(vcpu, seg_desc);
3238 
3239 	return kvm_read_guest(vcpu->kvm, base_addr, tss,
3240 			      sizeof(struct tss_segment_16));
3241 }
3242 
3243 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3244 			      struct desc_struct *seg_desc,
3245 			      struct tss_segment_16 *tss)
3246 {
3247 	u32 base_addr;
3248 
3249 	base_addr = get_tss_base_addr(vcpu, seg_desc);
3250 
3251 	return kvm_write_guest(vcpu->kvm, base_addr, tss,
3252 			       sizeof(struct tss_segment_16));
3253 }
3254 
3255 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3256 {
3257 	struct kvm_segment kvm_seg;
3258 
3259 	get_segment(vcpu, &kvm_seg, seg);
3260 	return kvm_seg.selector;
3261 }
3262 
3263 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3264 						u16 selector,
3265 						struct kvm_segment *kvm_seg)
3266 {
3267 	struct desc_struct seg_desc;
3268 
3269 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3270 		return 1;
3271 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3272 	return 0;
3273 }
3274 
3275 static int load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3276 				   int type_bits, int seg)
3277 {
3278 	struct kvm_segment kvm_seg;
3279 
3280 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3281 		return 1;
3282 	kvm_seg.type |= type_bits;
3283 
3284 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3285 	    seg != VCPU_SREG_LDTR)
3286 		if (!kvm_seg.s)
3287 			kvm_seg.unusable = 1;
3288 
3289 	set_segment(vcpu, &kvm_seg, seg);
3290 	return 0;
3291 }
3292 
3293 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3294 				struct tss_segment_32 *tss)
3295 {
3296 	tss->cr3 = vcpu->arch.cr3;
3297 	tss->eip = vcpu->arch.rip;
3298 	tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3299 	tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3300 	tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3301 	tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3302 	tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3303 	tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3304 	tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3305 	tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3306 	tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3307 
3308 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3309 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3310 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3311 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3312 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3313 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3314 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3315 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3316 }
3317 
3318 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3319 				  struct tss_segment_32 *tss)
3320 {
3321 	kvm_set_cr3(vcpu, tss->cr3);
3322 
3323 	vcpu->arch.rip = tss->eip;
3324 	kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3325 
3326 	vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3327 	vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3328 	vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3329 	vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3330 	vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3331 	vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3332 	vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3333 	vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3334 
3335 	if (load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3336 		return 1;
3337 
3338 	if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3339 		return 1;
3340 
3341 	if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3342 		return 1;
3343 
3344 	if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3345 		return 1;
3346 
3347 	if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3348 		return 1;
3349 
3350 	if (load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3351 		return 1;
3352 
3353 	if (load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3354 		return 1;
3355 	return 0;
3356 }
3357 
3358 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3359 				struct tss_segment_16 *tss)
3360 {
3361 	tss->ip = vcpu->arch.rip;
3362 	tss->flag = kvm_x86_ops->get_rflags(vcpu);
3363 	tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3364 	tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3365 	tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3366 	tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3367 	tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3368 	tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3369 	tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3370 	tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3371 
3372 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3373 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3374 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3375 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3376 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3377 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3378 }
3379 
3380 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3381 				 struct tss_segment_16 *tss)
3382 {
3383 	vcpu->arch.rip = tss->ip;
3384 	kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3385 	vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3386 	vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3387 	vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3388 	vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3389 	vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3390 	vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3391 	vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3392 	vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3393 
3394 	if (load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3395 		return 1;
3396 
3397 	if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3398 		return 1;
3399 
3400 	if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3401 		return 1;
3402 
3403 	if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3404 		return 1;
3405 
3406 	if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3407 		return 1;
3408 	return 0;
3409 }
3410 
3411 int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3412 		       struct desc_struct *cseg_desc,
3413 		       struct desc_struct *nseg_desc)
3414 {
3415 	struct tss_segment_16 tss_segment_16;
3416 	int ret = 0;
3417 
3418 	if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3419 		goto out;
3420 
3421 	save_state_to_tss16(vcpu, &tss_segment_16);
3422 	save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3423 
3424 	if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3425 		goto out;
3426 	if (load_state_from_tss16(vcpu, &tss_segment_16))
3427 		goto out;
3428 
3429 	ret = 1;
3430 out:
3431 	return ret;
3432 }
3433 
3434 int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3435 		       struct desc_struct *cseg_desc,
3436 		       struct desc_struct *nseg_desc)
3437 {
3438 	struct tss_segment_32 tss_segment_32;
3439 	int ret = 0;
3440 
3441 	if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3442 		goto out;
3443 
3444 	save_state_to_tss32(vcpu, &tss_segment_32);
3445 	save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3446 
3447 	if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3448 		goto out;
3449 	if (load_state_from_tss32(vcpu, &tss_segment_32))
3450 		goto out;
3451 
3452 	ret = 1;
3453 out:
3454 	return ret;
3455 }
3456 
3457 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3458 {
3459 	struct kvm_segment tr_seg;
3460 	struct desc_struct cseg_desc;
3461 	struct desc_struct nseg_desc;
3462 	int ret = 0;
3463 
3464 	get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3465 
3466 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3467 		goto out;
3468 
3469 	if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3470 		goto out;
3471 
3472 
3473 	if (reason != TASK_SWITCH_IRET) {
3474 		int cpl;
3475 
3476 		cpl = kvm_x86_ops->get_cpl(vcpu);
3477 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3478 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3479 			return 1;
3480 		}
3481 	}
3482 
3483 	if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3484 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3485 		return 1;
3486 	}
3487 
3488 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3489 		cseg_desc.type &= ~(1 << 1); //clear the B flag
3490 		save_guest_segment_descriptor(vcpu, tr_seg.selector,
3491 					      &cseg_desc);
3492 	}
3493 
3494 	if (reason == TASK_SWITCH_IRET) {
3495 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3496 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3497 	}
3498 
3499 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3500 	kvm_x86_ops->cache_regs(vcpu);
3501 
3502 	if (nseg_desc.type & 8)
3503 		ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3504 					 &nseg_desc);
3505 	else
3506 		ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3507 					 &nseg_desc);
3508 
3509 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3510 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3511 		kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3512 	}
3513 
3514 	if (reason != TASK_SWITCH_IRET) {
3515 		nseg_desc.type |= (1 << 1);
3516 		save_guest_segment_descriptor(vcpu, tss_selector,
3517 					      &nseg_desc);
3518 	}
3519 
3520 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3521 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3522 	tr_seg.type = 11;
3523 	set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3524 out:
3525 	kvm_x86_ops->decache_regs(vcpu);
3526 	return ret;
3527 }
3528 EXPORT_SYMBOL_GPL(kvm_task_switch);
3529 
3530 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3531 				  struct kvm_sregs *sregs)
3532 {
3533 	int mmu_reset_needed = 0;
3534 	int i, pending_vec, max_bits;
3535 	struct descriptor_table dt;
3536 
3537 	vcpu_load(vcpu);
3538 
3539 	dt.limit = sregs->idt.limit;
3540 	dt.base = sregs->idt.base;
3541 	kvm_x86_ops->set_idt(vcpu, &dt);
3542 	dt.limit = sregs->gdt.limit;
3543 	dt.base = sregs->gdt.base;
3544 	kvm_x86_ops->set_gdt(vcpu, &dt);
3545 
3546 	vcpu->arch.cr2 = sregs->cr2;
3547 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3548 	vcpu->arch.cr3 = sregs->cr3;
3549 
3550 	kvm_set_cr8(vcpu, sregs->cr8);
3551 
3552 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3553 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
3554 	kvm_set_apic_base(vcpu, sregs->apic_base);
3555 
3556 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3557 
3558 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3559 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3560 	vcpu->arch.cr0 = sregs->cr0;
3561 
3562 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3563 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3564 	if (!is_long_mode(vcpu) && is_pae(vcpu))
3565 		load_pdptrs(vcpu, vcpu->arch.cr3);
3566 
3567 	if (mmu_reset_needed)
3568 		kvm_mmu_reset_context(vcpu);
3569 
3570 	if (!irqchip_in_kernel(vcpu->kvm)) {
3571 		memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3572 		       sizeof vcpu->arch.irq_pending);
3573 		vcpu->arch.irq_summary = 0;
3574 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3575 			if (vcpu->arch.irq_pending[i])
3576 				__set_bit(i, &vcpu->arch.irq_summary);
3577 	} else {
3578 		max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3579 		pending_vec = find_first_bit(
3580 			(const unsigned long *)sregs->interrupt_bitmap,
3581 			max_bits);
3582 		/* Only pending external irq is handled here */
3583 		if (pending_vec < max_bits) {
3584 			kvm_x86_ops->set_irq(vcpu, pending_vec);
3585 			pr_debug("Set back pending irq %d\n",
3586 				 pending_vec);
3587 		}
3588 	}
3589 
3590 	set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3591 	set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3592 	set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3593 	set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3594 	set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3595 	set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3596 
3597 	set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3598 	set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3599 
3600 	vcpu_put(vcpu);
3601 
3602 	return 0;
3603 }
3604 
3605 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3606 				    struct kvm_debug_guest *dbg)
3607 {
3608 	int r;
3609 
3610 	vcpu_load(vcpu);
3611 
3612 	r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3613 
3614 	vcpu_put(vcpu);
3615 
3616 	return r;
3617 }
3618 
3619 /*
3620  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3621  * we have asm/x86/processor.h
3622  */
3623 struct fxsave {
3624 	u16	cwd;
3625 	u16	swd;
3626 	u16	twd;
3627 	u16	fop;
3628 	u64	rip;
3629 	u64	rdp;
3630 	u32	mxcsr;
3631 	u32	mxcsr_mask;
3632 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
3633 #ifdef CONFIG_X86_64
3634 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
3635 #else
3636 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
3637 #endif
3638 };
3639 
3640 /*
3641  * Translate a guest virtual address to a guest physical address.
3642  */
3643 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3644 				    struct kvm_translation *tr)
3645 {
3646 	unsigned long vaddr = tr->linear_address;
3647 	gpa_t gpa;
3648 
3649 	vcpu_load(vcpu);
3650 	down_read(&vcpu->kvm->slots_lock);
3651 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3652 	up_read(&vcpu->kvm->slots_lock);
3653 	tr->physical_address = gpa;
3654 	tr->valid = gpa != UNMAPPED_GVA;
3655 	tr->writeable = 1;
3656 	tr->usermode = 0;
3657 	vcpu_put(vcpu);
3658 
3659 	return 0;
3660 }
3661 
3662 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3663 {
3664 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3665 
3666 	vcpu_load(vcpu);
3667 
3668 	memcpy(fpu->fpr, fxsave->st_space, 128);
3669 	fpu->fcw = fxsave->cwd;
3670 	fpu->fsw = fxsave->swd;
3671 	fpu->ftwx = fxsave->twd;
3672 	fpu->last_opcode = fxsave->fop;
3673 	fpu->last_ip = fxsave->rip;
3674 	fpu->last_dp = fxsave->rdp;
3675 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3676 
3677 	vcpu_put(vcpu);
3678 
3679 	return 0;
3680 }
3681 
3682 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3683 {
3684 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3685 
3686 	vcpu_load(vcpu);
3687 
3688 	memcpy(fxsave->st_space, fpu->fpr, 128);
3689 	fxsave->cwd = fpu->fcw;
3690 	fxsave->swd = fpu->fsw;
3691 	fxsave->twd = fpu->ftwx;
3692 	fxsave->fop = fpu->last_opcode;
3693 	fxsave->rip = fpu->last_ip;
3694 	fxsave->rdp = fpu->last_dp;
3695 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3696 
3697 	vcpu_put(vcpu);
3698 
3699 	return 0;
3700 }
3701 
3702 void fx_init(struct kvm_vcpu *vcpu)
3703 {
3704 	unsigned after_mxcsr_mask;
3705 
3706 	/*
3707 	 * Touch the fpu the first time in non atomic context as if
3708 	 * this is the first fpu instruction the exception handler
3709 	 * will fire before the instruction returns and it'll have to
3710 	 * allocate ram with GFP_KERNEL.
3711 	 */
3712 	if (!used_math())
3713 		fx_save(&vcpu->arch.host_fx_image);
3714 
3715 	/* Initialize guest FPU by resetting ours and saving into guest's */
3716 	preempt_disable();
3717 	fx_save(&vcpu->arch.host_fx_image);
3718 	fx_finit();
3719 	fx_save(&vcpu->arch.guest_fx_image);
3720 	fx_restore(&vcpu->arch.host_fx_image);
3721 	preempt_enable();
3722 
3723 	vcpu->arch.cr0 |= X86_CR0_ET;
3724 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3725 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3726 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3727 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3728 }
3729 EXPORT_SYMBOL_GPL(fx_init);
3730 
3731 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3732 {
3733 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3734 		return;
3735 
3736 	vcpu->guest_fpu_loaded = 1;
3737 	fx_save(&vcpu->arch.host_fx_image);
3738 	fx_restore(&vcpu->arch.guest_fx_image);
3739 }
3740 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3741 
3742 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3743 {
3744 	if (!vcpu->guest_fpu_loaded)
3745 		return;
3746 
3747 	vcpu->guest_fpu_loaded = 0;
3748 	fx_save(&vcpu->arch.guest_fx_image);
3749 	fx_restore(&vcpu->arch.host_fx_image);
3750 	++vcpu->stat.fpu_reload;
3751 }
3752 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3753 
3754 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3755 {
3756 	kvm_x86_ops->vcpu_free(vcpu);
3757 }
3758 
3759 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3760 						unsigned int id)
3761 {
3762 	return kvm_x86_ops->vcpu_create(kvm, id);
3763 }
3764 
3765 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3766 {
3767 	int r;
3768 
3769 	/* We do fxsave: this must be aligned. */
3770 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3771 
3772 	vcpu_load(vcpu);
3773 	r = kvm_arch_vcpu_reset(vcpu);
3774 	if (r == 0)
3775 		r = kvm_mmu_setup(vcpu);
3776 	vcpu_put(vcpu);
3777 	if (r < 0)
3778 		goto free_vcpu;
3779 
3780 	return 0;
3781 free_vcpu:
3782 	kvm_x86_ops->vcpu_free(vcpu);
3783 	return r;
3784 }
3785 
3786 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3787 {
3788 	vcpu_load(vcpu);
3789 	kvm_mmu_unload(vcpu);
3790 	vcpu_put(vcpu);
3791 
3792 	kvm_x86_ops->vcpu_free(vcpu);
3793 }
3794 
3795 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3796 {
3797 	return kvm_x86_ops->vcpu_reset(vcpu);
3798 }
3799 
3800 void kvm_arch_hardware_enable(void *garbage)
3801 {
3802 	kvm_x86_ops->hardware_enable(garbage);
3803 }
3804 
3805 void kvm_arch_hardware_disable(void *garbage)
3806 {
3807 	kvm_x86_ops->hardware_disable(garbage);
3808 }
3809 
3810 int kvm_arch_hardware_setup(void)
3811 {
3812 	return kvm_x86_ops->hardware_setup();
3813 }
3814 
3815 void kvm_arch_hardware_unsetup(void)
3816 {
3817 	kvm_x86_ops->hardware_unsetup();
3818 }
3819 
3820 void kvm_arch_check_processor_compat(void *rtn)
3821 {
3822 	kvm_x86_ops->check_processor_compatibility(rtn);
3823 }
3824 
3825 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3826 {
3827 	struct page *page;
3828 	struct kvm *kvm;
3829 	int r;
3830 
3831 	BUG_ON(vcpu->kvm == NULL);
3832 	kvm = vcpu->kvm;
3833 
3834 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3835 	if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3836 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3837 	else
3838 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3839 
3840 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3841 	if (!page) {
3842 		r = -ENOMEM;
3843 		goto fail;
3844 	}
3845 	vcpu->arch.pio_data = page_address(page);
3846 
3847 	r = kvm_mmu_create(vcpu);
3848 	if (r < 0)
3849 		goto fail_free_pio_data;
3850 
3851 	if (irqchip_in_kernel(kvm)) {
3852 		r = kvm_create_lapic(vcpu);
3853 		if (r < 0)
3854 			goto fail_mmu_destroy;
3855 	}
3856 
3857 	return 0;
3858 
3859 fail_mmu_destroy:
3860 	kvm_mmu_destroy(vcpu);
3861 fail_free_pio_data:
3862 	free_page((unsigned long)vcpu->arch.pio_data);
3863 fail:
3864 	return r;
3865 }
3866 
3867 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3868 {
3869 	kvm_free_lapic(vcpu);
3870 	down_read(&vcpu->kvm->slots_lock);
3871 	kvm_mmu_destroy(vcpu);
3872 	up_read(&vcpu->kvm->slots_lock);
3873 	free_page((unsigned long)vcpu->arch.pio_data);
3874 }
3875 
3876 struct  kvm *kvm_arch_create_vm(void)
3877 {
3878 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3879 
3880 	if (!kvm)
3881 		return ERR_PTR(-ENOMEM);
3882 
3883 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3884 
3885 	return kvm;
3886 }
3887 
3888 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3889 {
3890 	vcpu_load(vcpu);
3891 	kvm_mmu_unload(vcpu);
3892 	vcpu_put(vcpu);
3893 }
3894 
3895 static void kvm_free_vcpus(struct kvm *kvm)
3896 {
3897 	unsigned int i;
3898 
3899 	/*
3900 	 * Unpin any mmu pages first.
3901 	 */
3902 	for (i = 0; i < KVM_MAX_VCPUS; ++i)
3903 		if (kvm->vcpus[i])
3904 			kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3905 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3906 		if (kvm->vcpus[i]) {
3907 			kvm_arch_vcpu_free(kvm->vcpus[i]);
3908 			kvm->vcpus[i] = NULL;
3909 		}
3910 	}
3911 
3912 }
3913 
3914 void kvm_arch_destroy_vm(struct kvm *kvm)
3915 {
3916 	kvm_free_pit(kvm);
3917 	kfree(kvm->arch.vpic);
3918 	kfree(kvm->arch.vioapic);
3919 	kvm_free_vcpus(kvm);
3920 	kvm_free_physmem(kvm);
3921 	if (kvm->arch.apic_access_page)
3922 		put_page(kvm->arch.apic_access_page);
3923 	if (kvm->arch.ept_identity_pagetable)
3924 		put_page(kvm->arch.ept_identity_pagetable);
3925 	kfree(kvm);
3926 }
3927 
3928 int kvm_arch_set_memory_region(struct kvm *kvm,
3929 				struct kvm_userspace_memory_region *mem,
3930 				struct kvm_memory_slot old,
3931 				int user_alloc)
3932 {
3933 	int npages = mem->memory_size >> PAGE_SHIFT;
3934 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3935 
3936 	/*To keep backward compatibility with older userspace,
3937 	 *x86 needs to hanlde !user_alloc case.
3938 	 */
3939 	if (!user_alloc) {
3940 		if (npages && !old.rmap) {
3941 			down_write(&current->mm->mmap_sem);
3942 			memslot->userspace_addr = do_mmap(NULL, 0,
3943 						     npages * PAGE_SIZE,
3944 						     PROT_READ | PROT_WRITE,
3945 						     MAP_SHARED | MAP_ANONYMOUS,
3946 						     0);
3947 			up_write(&current->mm->mmap_sem);
3948 
3949 			if (IS_ERR((void *)memslot->userspace_addr))
3950 				return PTR_ERR((void *)memslot->userspace_addr);
3951 		} else {
3952 			if (!old.user_alloc && old.rmap) {
3953 				int ret;
3954 
3955 				down_write(&current->mm->mmap_sem);
3956 				ret = do_munmap(current->mm, old.userspace_addr,
3957 						old.npages * PAGE_SIZE);
3958 				up_write(&current->mm->mmap_sem);
3959 				if (ret < 0)
3960 					printk(KERN_WARNING
3961 				       "kvm_vm_ioctl_set_memory_region: "
3962 				       "failed to munmap memory\n");
3963 			}
3964 		}
3965 	}
3966 
3967 	if (!kvm->arch.n_requested_mmu_pages) {
3968 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3969 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3970 	}
3971 
3972 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3973 	kvm_flush_remote_tlbs(kvm);
3974 
3975 	return 0;
3976 }
3977 
3978 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3979 {
3980 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
3981 	       || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
3982 }
3983 
3984 static void vcpu_kick_intr(void *info)
3985 {
3986 #ifdef DEBUG
3987 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3988 	printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3989 #endif
3990 }
3991 
3992 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3993 {
3994 	int ipi_pcpu = vcpu->cpu;
3995 	int cpu = get_cpu();
3996 
3997 	if (waitqueue_active(&vcpu->wq)) {
3998 		wake_up_interruptible(&vcpu->wq);
3999 		++vcpu->stat.halt_wakeup;
4000 	}
4001 	/*
4002 	 * We may be called synchronously with irqs disabled in guest mode,
4003 	 * So need not to call smp_call_function_single() in that case.
4004 	 */
4005 	if (vcpu->guest_mode && vcpu->cpu != cpu)
4006 		smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
4007 	put_cpu();
4008 }
4009