xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 384740dc)
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 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
76 	{ "halt_exits", VCPU_STAT(halt_exits) },
77 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
78 	{ "hypercalls", VCPU_STAT(hypercalls) },
79 	{ "request_irq", VCPU_STAT(request_irq_exits) },
80 	{ "irq_exits", VCPU_STAT(irq_exits) },
81 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
82 	{ "efer_reload", VCPU_STAT(efer_reload) },
83 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
84 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
85 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
86 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
87 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
88 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
89 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
90 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
91 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
92 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
93 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
94 	{ "largepages", VM_STAT(lpages) },
95 	{ NULL }
96 };
97 
98 
99 unsigned long segment_base(u16 selector)
100 {
101 	struct descriptor_table gdt;
102 	struct desc_struct *d;
103 	unsigned long table_base;
104 	unsigned long v;
105 
106 	if (selector == 0)
107 		return 0;
108 
109 	asm("sgdt %0" : "=m"(gdt));
110 	table_base = gdt.base;
111 
112 	if (selector & 4) {           /* from ldt */
113 		u16 ldt_selector;
114 
115 		asm("sldt %0" : "=g"(ldt_selector));
116 		table_base = segment_base(ldt_selector);
117 	}
118 	d = (struct desc_struct *)(table_base + (selector & ~7));
119 	v = d->base0 | ((unsigned long)d->base1 << 16) |
120 		((unsigned long)d->base2 << 24);
121 #ifdef CONFIG_X86_64
122 	if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
123 		v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
124 #endif
125 	return v;
126 }
127 EXPORT_SYMBOL_GPL(segment_base);
128 
129 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
130 {
131 	if (irqchip_in_kernel(vcpu->kvm))
132 		return vcpu->arch.apic_base;
133 	else
134 		return vcpu->arch.apic_base;
135 }
136 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
137 
138 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
139 {
140 	/* TODO: reserve bits check */
141 	if (irqchip_in_kernel(vcpu->kvm))
142 		kvm_lapic_set_base(vcpu, data);
143 	else
144 		vcpu->arch.apic_base = data;
145 }
146 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
147 
148 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
149 {
150 	WARN_ON(vcpu->arch.exception.pending);
151 	vcpu->arch.exception.pending = true;
152 	vcpu->arch.exception.has_error_code = false;
153 	vcpu->arch.exception.nr = nr;
154 }
155 EXPORT_SYMBOL_GPL(kvm_queue_exception);
156 
157 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
158 			   u32 error_code)
159 {
160 	++vcpu->stat.pf_guest;
161 	if (vcpu->arch.exception.pending) {
162 		if (vcpu->arch.exception.nr == PF_VECTOR) {
163 			printk(KERN_DEBUG "kvm: inject_page_fault:"
164 					" double fault 0x%lx\n", addr);
165 			vcpu->arch.exception.nr = DF_VECTOR;
166 			vcpu->arch.exception.error_code = 0;
167 		} else if (vcpu->arch.exception.nr == DF_VECTOR) {
168 			/* triple fault -> shutdown */
169 			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
170 		}
171 		return;
172 	}
173 	vcpu->arch.cr2 = addr;
174 	kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
175 }
176 
177 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
178 {
179 	vcpu->arch.nmi_pending = 1;
180 }
181 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
182 
183 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
184 {
185 	WARN_ON(vcpu->arch.exception.pending);
186 	vcpu->arch.exception.pending = true;
187 	vcpu->arch.exception.has_error_code = true;
188 	vcpu->arch.exception.nr = nr;
189 	vcpu->arch.exception.error_code = error_code;
190 }
191 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
192 
193 static void __queue_exception(struct kvm_vcpu *vcpu)
194 {
195 	kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
196 				     vcpu->arch.exception.has_error_code,
197 				     vcpu->arch.exception.error_code);
198 }
199 
200 /*
201  * Load the pae pdptrs.  Return true is they are all valid.
202  */
203 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
204 {
205 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
206 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
207 	int i;
208 	int ret;
209 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
210 
211 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
212 				  offset * sizeof(u64), sizeof(pdpte));
213 	if (ret < 0) {
214 		ret = 0;
215 		goto out;
216 	}
217 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
218 		if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
219 			ret = 0;
220 			goto out;
221 		}
222 	}
223 	ret = 1;
224 
225 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
226 out:
227 
228 	return ret;
229 }
230 EXPORT_SYMBOL_GPL(load_pdptrs);
231 
232 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
233 {
234 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
235 	bool changed = true;
236 	int r;
237 
238 	if (is_long_mode(vcpu) || !is_pae(vcpu))
239 		return false;
240 
241 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
242 	if (r < 0)
243 		goto out;
244 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
245 out:
246 
247 	return changed;
248 }
249 
250 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
251 {
252 	if (cr0 & CR0_RESERVED_BITS) {
253 		printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
254 		       cr0, vcpu->arch.cr0);
255 		kvm_inject_gp(vcpu, 0);
256 		return;
257 	}
258 
259 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
260 		printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
261 		kvm_inject_gp(vcpu, 0);
262 		return;
263 	}
264 
265 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
266 		printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
267 		       "and a clear PE flag\n");
268 		kvm_inject_gp(vcpu, 0);
269 		return;
270 	}
271 
272 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
273 #ifdef CONFIG_X86_64
274 		if ((vcpu->arch.shadow_efer & EFER_LME)) {
275 			int cs_db, cs_l;
276 
277 			if (!is_pae(vcpu)) {
278 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
279 				       "in long mode while PAE is disabled\n");
280 				kvm_inject_gp(vcpu, 0);
281 				return;
282 			}
283 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
284 			if (cs_l) {
285 				printk(KERN_DEBUG "set_cr0: #GP, start paging "
286 				       "in long mode while CS.L == 1\n");
287 				kvm_inject_gp(vcpu, 0);
288 				return;
289 
290 			}
291 		} else
292 #endif
293 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
294 			printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
295 			       "reserved bits\n");
296 			kvm_inject_gp(vcpu, 0);
297 			return;
298 		}
299 
300 	}
301 
302 	kvm_x86_ops->set_cr0(vcpu, cr0);
303 	vcpu->arch.cr0 = cr0;
304 
305 	kvm_mmu_reset_context(vcpu);
306 	return;
307 }
308 EXPORT_SYMBOL_GPL(kvm_set_cr0);
309 
310 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
311 {
312 	kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
313 	KVMTRACE_1D(LMSW, vcpu,
314 		    (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
315 		    handler);
316 }
317 EXPORT_SYMBOL_GPL(kvm_lmsw);
318 
319 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
320 {
321 	if (cr4 & CR4_RESERVED_BITS) {
322 		printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
323 		kvm_inject_gp(vcpu, 0);
324 		return;
325 	}
326 
327 	if (is_long_mode(vcpu)) {
328 		if (!(cr4 & X86_CR4_PAE)) {
329 			printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
330 			       "in long mode\n");
331 			kvm_inject_gp(vcpu, 0);
332 			return;
333 		}
334 	} else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
335 		   && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
336 		printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
337 		kvm_inject_gp(vcpu, 0);
338 		return;
339 	}
340 
341 	if (cr4 & X86_CR4_VMXE) {
342 		printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
343 		kvm_inject_gp(vcpu, 0);
344 		return;
345 	}
346 	kvm_x86_ops->set_cr4(vcpu, cr4);
347 	vcpu->arch.cr4 = cr4;
348 	kvm_mmu_reset_context(vcpu);
349 }
350 EXPORT_SYMBOL_GPL(kvm_set_cr4);
351 
352 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
353 {
354 	if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
355 		kvm_mmu_flush_tlb(vcpu);
356 		return;
357 	}
358 
359 	if (is_long_mode(vcpu)) {
360 		if (cr3 & CR3_L_MODE_RESERVED_BITS) {
361 			printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
362 			kvm_inject_gp(vcpu, 0);
363 			return;
364 		}
365 	} else {
366 		if (is_pae(vcpu)) {
367 			if (cr3 & CR3_PAE_RESERVED_BITS) {
368 				printk(KERN_DEBUG
369 				       "set_cr3: #GP, reserved bits\n");
370 				kvm_inject_gp(vcpu, 0);
371 				return;
372 			}
373 			if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
374 				printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
375 				       "reserved bits\n");
376 				kvm_inject_gp(vcpu, 0);
377 				return;
378 			}
379 		}
380 		/*
381 		 * We don't check reserved bits in nonpae mode, because
382 		 * this isn't enforced, and VMware depends on this.
383 		 */
384 	}
385 
386 	/*
387 	 * Does the new cr3 value map to physical memory? (Note, we
388 	 * catch an invalid cr3 even in real-mode, because it would
389 	 * cause trouble later on when we turn on paging anyway.)
390 	 *
391 	 * A real CPU would silently accept an invalid cr3 and would
392 	 * attempt to use it - with largely undefined (and often hard
393 	 * to debug) behavior on the guest side.
394 	 */
395 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
396 		kvm_inject_gp(vcpu, 0);
397 	else {
398 		vcpu->arch.cr3 = cr3;
399 		vcpu->arch.mmu.new_cr3(vcpu);
400 	}
401 }
402 EXPORT_SYMBOL_GPL(kvm_set_cr3);
403 
404 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
405 {
406 	if (cr8 & CR8_RESERVED_BITS) {
407 		printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
408 		kvm_inject_gp(vcpu, 0);
409 		return;
410 	}
411 	if (irqchip_in_kernel(vcpu->kvm))
412 		kvm_lapic_set_tpr(vcpu, cr8);
413 	else
414 		vcpu->arch.cr8 = cr8;
415 }
416 EXPORT_SYMBOL_GPL(kvm_set_cr8);
417 
418 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
419 {
420 	if (irqchip_in_kernel(vcpu->kvm))
421 		return kvm_lapic_get_cr8(vcpu);
422 	else
423 		return vcpu->arch.cr8;
424 }
425 EXPORT_SYMBOL_GPL(kvm_get_cr8);
426 
427 /*
428  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
429  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
430  *
431  * This list is modified at module load time to reflect the
432  * capabilities of the host cpu.
433  */
434 static u32 msrs_to_save[] = {
435 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
436 	MSR_K6_STAR,
437 #ifdef CONFIG_X86_64
438 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
439 #endif
440 	MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
441 	MSR_IA32_PERF_STATUS,
442 };
443 
444 static unsigned num_msrs_to_save;
445 
446 static u32 emulated_msrs[] = {
447 	MSR_IA32_MISC_ENABLE,
448 };
449 
450 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
451 {
452 	if (efer & efer_reserved_bits) {
453 		printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
454 		       efer);
455 		kvm_inject_gp(vcpu, 0);
456 		return;
457 	}
458 
459 	if (is_paging(vcpu)
460 	    && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
461 		printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
462 		kvm_inject_gp(vcpu, 0);
463 		return;
464 	}
465 
466 	kvm_x86_ops->set_efer(vcpu, efer);
467 
468 	efer &= ~EFER_LMA;
469 	efer |= vcpu->arch.shadow_efer & EFER_LMA;
470 
471 	vcpu->arch.shadow_efer = efer;
472 }
473 
474 void kvm_enable_efer_bits(u64 mask)
475 {
476        efer_reserved_bits &= ~mask;
477 }
478 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
479 
480 
481 /*
482  * Writes msr value into into the appropriate "register".
483  * Returns 0 on success, non-0 otherwise.
484  * Assumes vcpu_load() was already called.
485  */
486 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
487 {
488 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
489 }
490 
491 /*
492  * Adapt set_msr() to msr_io()'s calling convention
493  */
494 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
495 {
496 	return kvm_set_msr(vcpu, index, *data);
497 }
498 
499 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
500 {
501 	static int version;
502 	struct pvclock_wall_clock wc;
503 	struct timespec now, sys, boot;
504 
505 	if (!wall_clock)
506 		return;
507 
508 	version++;
509 
510 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
511 
512 	/*
513 	 * The guest calculates current wall clock time by adding
514 	 * system time (updated by kvm_write_guest_time below) to the
515 	 * wall clock specified here.  guest system time equals host
516 	 * system time for us, thus we must fill in host boot time here.
517 	 */
518 	now = current_kernel_time();
519 	ktime_get_ts(&sys);
520 	boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
521 
522 	wc.sec = boot.tv_sec;
523 	wc.nsec = boot.tv_nsec;
524 	wc.version = version;
525 
526 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
527 
528 	version++;
529 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
530 }
531 
532 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
533 {
534 	uint32_t quotient, remainder;
535 
536 	/* Don't try to replace with do_div(), this one calculates
537 	 * "(dividend << 32) / divisor" */
538 	__asm__ ( "divl %4"
539 		  : "=a" (quotient), "=d" (remainder)
540 		  : "0" (0), "1" (dividend), "r" (divisor) );
541 	return quotient;
542 }
543 
544 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
545 {
546 	uint64_t nsecs = 1000000000LL;
547 	int32_t  shift = 0;
548 	uint64_t tps64;
549 	uint32_t tps32;
550 
551 	tps64 = tsc_khz * 1000LL;
552 	while (tps64 > nsecs*2) {
553 		tps64 >>= 1;
554 		shift--;
555 	}
556 
557 	tps32 = (uint32_t)tps64;
558 	while (tps32 <= (uint32_t)nsecs) {
559 		tps32 <<= 1;
560 		shift++;
561 	}
562 
563 	hv_clock->tsc_shift = shift;
564 	hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
565 
566 	pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
567 		 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
568 		 hv_clock->tsc_to_system_mul);
569 }
570 
571 static void kvm_write_guest_time(struct kvm_vcpu *v)
572 {
573 	struct timespec ts;
574 	unsigned long flags;
575 	struct kvm_vcpu_arch *vcpu = &v->arch;
576 	void *shared_kaddr;
577 
578 	if ((!vcpu->time_page))
579 		return;
580 
581 	if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
582 		kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
583 		vcpu->hv_clock_tsc_khz = tsc_khz;
584 	}
585 
586 	/* Keep irq disabled to prevent changes to the clock */
587 	local_irq_save(flags);
588 	kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
589 			  &vcpu->hv_clock.tsc_timestamp);
590 	ktime_get_ts(&ts);
591 	local_irq_restore(flags);
592 
593 	/* With all the info we got, fill in the values */
594 
595 	vcpu->hv_clock.system_time = ts.tv_nsec +
596 				     (NSEC_PER_SEC * (u64)ts.tv_sec);
597 	/*
598 	 * The interface expects us to write an even number signaling that the
599 	 * update is finished. Since the guest won't see the intermediate
600 	 * state, we just increase by 2 at the end.
601 	 */
602 	vcpu->hv_clock.version += 2;
603 
604 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
605 
606 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
607 	       sizeof(vcpu->hv_clock));
608 
609 	kunmap_atomic(shared_kaddr, KM_USER0);
610 
611 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
612 }
613 
614 static bool msr_mtrr_valid(unsigned msr)
615 {
616 	switch (msr) {
617 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
618 	case MSR_MTRRfix64K_00000:
619 	case MSR_MTRRfix16K_80000:
620 	case MSR_MTRRfix16K_A0000:
621 	case MSR_MTRRfix4K_C0000:
622 	case MSR_MTRRfix4K_C8000:
623 	case MSR_MTRRfix4K_D0000:
624 	case MSR_MTRRfix4K_D8000:
625 	case MSR_MTRRfix4K_E0000:
626 	case MSR_MTRRfix4K_E8000:
627 	case MSR_MTRRfix4K_F0000:
628 	case MSR_MTRRfix4K_F8000:
629 	case MSR_MTRRdefType:
630 	case MSR_IA32_CR_PAT:
631 		return true;
632 	case 0x2f8:
633 		return true;
634 	}
635 	return false;
636 }
637 
638 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
639 {
640 	if (!msr_mtrr_valid(msr))
641 		return 1;
642 
643 	vcpu->arch.mtrr[msr - 0x200] = data;
644 	return 0;
645 }
646 
647 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
648 {
649 	switch (msr) {
650 	case MSR_EFER:
651 		set_efer(vcpu, data);
652 		break;
653 	case MSR_IA32_MC0_STATUS:
654 		pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
655 		       __func__, data);
656 		break;
657 	case MSR_IA32_MCG_STATUS:
658 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
659 			__func__, data);
660 		break;
661 	case MSR_IA32_MCG_CTL:
662 		pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
663 			__func__, data);
664 		break;
665 	case MSR_IA32_UCODE_REV:
666 	case MSR_IA32_UCODE_WRITE:
667 		break;
668 	case 0x200 ... 0x2ff:
669 		return set_msr_mtrr(vcpu, msr, data);
670 	case MSR_IA32_APICBASE:
671 		kvm_set_apic_base(vcpu, data);
672 		break;
673 	case MSR_IA32_MISC_ENABLE:
674 		vcpu->arch.ia32_misc_enable_msr = data;
675 		break;
676 	case MSR_KVM_WALL_CLOCK:
677 		vcpu->kvm->arch.wall_clock = data;
678 		kvm_write_wall_clock(vcpu->kvm, data);
679 		break;
680 	case MSR_KVM_SYSTEM_TIME: {
681 		if (vcpu->arch.time_page) {
682 			kvm_release_page_dirty(vcpu->arch.time_page);
683 			vcpu->arch.time_page = NULL;
684 		}
685 
686 		vcpu->arch.time = data;
687 
688 		/* we verify if the enable bit is set... */
689 		if (!(data & 1))
690 			break;
691 
692 		/* ...but clean it before doing the actual write */
693 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
694 
695 		down_read(&current->mm->mmap_sem);
696 		vcpu->arch.time_page =
697 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
698 		up_read(&current->mm->mmap_sem);
699 
700 		if (is_error_page(vcpu->arch.time_page)) {
701 			kvm_release_page_clean(vcpu->arch.time_page);
702 			vcpu->arch.time_page = NULL;
703 		}
704 
705 		kvm_write_guest_time(vcpu);
706 		break;
707 	}
708 	default:
709 		pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
710 		return 1;
711 	}
712 	return 0;
713 }
714 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
715 
716 
717 /*
718  * Reads an msr value (of 'msr_index') into 'pdata'.
719  * Returns 0 on success, non-0 otherwise.
720  * Assumes vcpu_load() was already called.
721  */
722 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
723 {
724 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
725 }
726 
727 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
728 {
729 	if (!msr_mtrr_valid(msr))
730 		return 1;
731 
732 	*pdata = vcpu->arch.mtrr[msr - 0x200];
733 	return 0;
734 }
735 
736 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
737 {
738 	u64 data;
739 
740 	switch (msr) {
741 	case 0xc0010010: /* SYSCFG */
742 	case 0xc0010015: /* HWCR */
743 	case MSR_IA32_PLATFORM_ID:
744 	case MSR_IA32_P5_MC_ADDR:
745 	case MSR_IA32_P5_MC_TYPE:
746 	case MSR_IA32_MC0_CTL:
747 	case MSR_IA32_MCG_STATUS:
748 	case MSR_IA32_MCG_CAP:
749 	case MSR_IA32_MCG_CTL:
750 	case MSR_IA32_MC0_MISC:
751 	case MSR_IA32_MC0_MISC+4:
752 	case MSR_IA32_MC0_MISC+8:
753 	case MSR_IA32_MC0_MISC+12:
754 	case MSR_IA32_MC0_MISC+16:
755 	case MSR_IA32_UCODE_REV:
756 	case MSR_IA32_EBL_CR_POWERON:
757 		data = 0;
758 		break;
759 	case MSR_MTRRcap:
760 		data = 0x500 | KVM_NR_VAR_MTRR;
761 		break;
762 	case 0x200 ... 0x2ff:
763 		return get_msr_mtrr(vcpu, msr, pdata);
764 	case 0xcd: /* fsb frequency */
765 		data = 3;
766 		break;
767 	case MSR_IA32_APICBASE:
768 		data = kvm_get_apic_base(vcpu);
769 		break;
770 	case MSR_IA32_MISC_ENABLE:
771 		data = vcpu->arch.ia32_misc_enable_msr;
772 		break;
773 	case MSR_IA32_PERF_STATUS:
774 		/* TSC increment by tick */
775 		data = 1000ULL;
776 		/* CPU multiplier */
777 		data |= (((uint64_t)4ULL) << 40);
778 		break;
779 	case MSR_EFER:
780 		data = vcpu->arch.shadow_efer;
781 		break;
782 	case MSR_KVM_WALL_CLOCK:
783 		data = vcpu->kvm->arch.wall_clock;
784 		break;
785 	case MSR_KVM_SYSTEM_TIME:
786 		data = vcpu->arch.time;
787 		break;
788 	default:
789 		pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
790 		return 1;
791 	}
792 	*pdata = data;
793 	return 0;
794 }
795 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
796 
797 /*
798  * Read or write a bunch of msrs. All parameters are kernel addresses.
799  *
800  * @return number of msrs set successfully.
801  */
802 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
803 		    struct kvm_msr_entry *entries,
804 		    int (*do_msr)(struct kvm_vcpu *vcpu,
805 				  unsigned index, u64 *data))
806 {
807 	int i;
808 
809 	vcpu_load(vcpu);
810 
811 	down_read(&vcpu->kvm->slots_lock);
812 	for (i = 0; i < msrs->nmsrs; ++i)
813 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
814 			break;
815 	up_read(&vcpu->kvm->slots_lock);
816 
817 	vcpu_put(vcpu);
818 
819 	return i;
820 }
821 
822 /*
823  * Read or write a bunch of msrs. Parameters are user addresses.
824  *
825  * @return number of msrs set successfully.
826  */
827 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
828 		  int (*do_msr)(struct kvm_vcpu *vcpu,
829 				unsigned index, u64 *data),
830 		  int writeback)
831 {
832 	struct kvm_msrs msrs;
833 	struct kvm_msr_entry *entries;
834 	int r, n;
835 	unsigned size;
836 
837 	r = -EFAULT;
838 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
839 		goto out;
840 
841 	r = -E2BIG;
842 	if (msrs.nmsrs >= MAX_IO_MSRS)
843 		goto out;
844 
845 	r = -ENOMEM;
846 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
847 	entries = vmalloc(size);
848 	if (!entries)
849 		goto out;
850 
851 	r = -EFAULT;
852 	if (copy_from_user(entries, user_msrs->entries, size))
853 		goto out_free;
854 
855 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
856 	if (r < 0)
857 		goto out_free;
858 
859 	r = -EFAULT;
860 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
861 		goto out_free;
862 
863 	r = n;
864 
865 out_free:
866 	vfree(entries);
867 out:
868 	return r;
869 }
870 
871 int kvm_dev_ioctl_check_extension(long ext)
872 {
873 	int r;
874 
875 	switch (ext) {
876 	case KVM_CAP_IRQCHIP:
877 	case KVM_CAP_HLT:
878 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
879 	case KVM_CAP_USER_MEMORY:
880 	case KVM_CAP_SET_TSS_ADDR:
881 	case KVM_CAP_EXT_CPUID:
882 	case KVM_CAP_CLOCKSOURCE:
883 	case KVM_CAP_PIT:
884 	case KVM_CAP_NOP_IO_DELAY:
885 	case KVM_CAP_MP_STATE:
886 	case KVM_CAP_SYNC_MMU:
887 		r = 1;
888 		break;
889 	case KVM_CAP_COALESCED_MMIO:
890 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
891 		break;
892 	case KVM_CAP_VAPIC:
893 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
894 		break;
895 	case KVM_CAP_NR_VCPUS:
896 		r = KVM_MAX_VCPUS;
897 		break;
898 	case KVM_CAP_NR_MEMSLOTS:
899 		r = KVM_MEMORY_SLOTS;
900 		break;
901 	case KVM_CAP_PV_MMU:
902 		r = !tdp_enabled;
903 		break;
904 	default:
905 		r = 0;
906 		break;
907 	}
908 	return r;
909 
910 }
911 
912 long kvm_arch_dev_ioctl(struct file *filp,
913 			unsigned int ioctl, unsigned long arg)
914 {
915 	void __user *argp = (void __user *)arg;
916 	long r;
917 
918 	switch (ioctl) {
919 	case KVM_GET_MSR_INDEX_LIST: {
920 		struct kvm_msr_list __user *user_msr_list = argp;
921 		struct kvm_msr_list msr_list;
922 		unsigned n;
923 
924 		r = -EFAULT;
925 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
926 			goto out;
927 		n = msr_list.nmsrs;
928 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
929 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
930 			goto out;
931 		r = -E2BIG;
932 		if (n < num_msrs_to_save)
933 			goto out;
934 		r = -EFAULT;
935 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
936 				 num_msrs_to_save * sizeof(u32)))
937 			goto out;
938 		if (copy_to_user(user_msr_list->indices
939 				 + num_msrs_to_save * sizeof(u32),
940 				 &emulated_msrs,
941 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
942 			goto out;
943 		r = 0;
944 		break;
945 	}
946 	case KVM_GET_SUPPORTED_CPUID: {
947 		struct kvm_cpuid2 __user *cpuid_arg = argp;
948 		struct kvm_cpuid2 cpuid;
949 
950 		r = -EFAULT;
951 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
952 			goto out;
953 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
954 			cpuid_arg->entries);
955 		if (r)
956 			goto out;
957 
958 		r = -EFAULT;
959 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
960 			goto out;
961 		r = 0;
962 		break;
963 	}
964 	default:
965 		r = -EINVAL;
966 	}
967 out:
968 	return r;
969 }
970 
971 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
972 {
973 	kvm_x86_ops->vcpu_load(vcpu, cpu);
974 	kvm_write_guest_time(vcpu);
975 }
976 
977 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
978 {
979 	kvm_x86_ops->vcpu_put(vcpu);
980 	kvm_put_guest_fpu(vcpu);
981 }
982 
983 static int is_efer_nx(void)
984 {
985 	u64 efer;
986 
987 	rdmsrl(MSR_EFER, efer);
988 	return efer & EFER_NX;
989 }
990 
991 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
992 {
993 	int i;
994 	struct kvm_cpuid_entry2 *e, *entry;
995 
996 	entry = NULL;
997 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
998 		e = &vcpu->arch.cpuid_entries[i];
999 		if (e->function == 0x80000001) {
1000 			entry = e;
1001 			break;
1002 		}
1003 	}
1004 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1005 		entry->edx &= ~(1 << 20);
1006 		printk(KERN_INFO "kvm: guest NX capability removed\n");
1007 	}
1008 }
1009 
1010 /* when an old userspace process fills a new kernel module */
1011 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1012 				    struct kvm_cpuid *cpuid,
1013 				    struct kvm_cpuid_entry __user *entries)
1014 {
1015 	int r, i;
1016 	struct kvm_cpuid_entry *cpuid_entries;
1017 
1018 	r = -E2BIG;
1019 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1020 		goto out;
1021 	r = -ENOMEM;
1022 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1023 	if (!cpuid_entries)
1024 		goto out;
1025 	r = -EFAULT;
1026 	if (copy_from_user(cpuid_entries, entries,
1027 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1028 		goto out_free;
1029 	for (i = 0; i < cpuid->nent; i++) {
1030 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1031 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1032 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1033 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1034 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1035 		vcpu->arch.cpuid_entries[i].index = 0;
1036 		vcpu->arch.cpuid_entries[i].flags = 0;
1037 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
1038 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
1039 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
1040 	}
1041 	vcpu->arch.cpuid_nent = cpuid->nent;
1042 	cpuid_fix_nx_cap(vcpu);
1043 	r = 0;
1044 
1045 out_free:
1046 	vfree(cpuid_entries);
1047 out:
1048 	return r;
1049 }
1050 
1051 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1052 				    struct kvm_cpuid2 *cpuid,
1053 				    struct kvm_cpuid_entry2 __user *entries)
1054 {
1055 	int r;
1056 
1057 	r = -E2BIG;
1058 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1059 		goto out;
1060 	r = -EFAULT;
1061 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1062 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1063 		goto out;
1064 	vcpu->arch.cpuid_nent = cpuid->nent;
1065 	return 0;
1066 
1067 out:
1068 	return r;
1069 }
1070 
1071 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1072 				    struct kvm_cpuid2 *cpuid,
1073 				    struct kvm_cpuid_entry2 __user *entries)
1074 {
1075 	int r;
1076 
1077 	r = -E2BIG;
1078 	if (cpuid->nent < vcpu->arch.cpuid_nent)
1079 		goto out;
1080 	r = -EFAULT;
1081 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1082 			   vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1083 		goto out;
1084 	return 0;
1085 
1086 out:
1087 	cpuid->nent = vcpu->arch.cpuid_nent;
1088 	return r;
1089 }
1090 
1091 static inline u32 bit(int bitno)
1092 {
1093 	return 1 << (bitno & 31);
1094 }
1095 
1096 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1097 			  u32 index)
1098 {
1099 	entry->function = function;
1100 	entry->index = index;
1101 	cpuid_count(entry->function, entry->index,
1102 		&entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1103 	entry->flags = 0;
1104 }
1105 
1106 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1107 			 u32 index, int *nent, int maxnent)
1108 {
1109 	const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1110 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1111 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1112 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1113 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1114 		bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1115 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1116 		bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1117 		bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1118 		bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1119 	const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1120 		bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1121 		bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1122 		bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1123 		bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1124 		bit(X86_FEATURE_PGE) |
1125 		bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1126 		bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1127 		bit(X86_FEATURE_SYSCALL) |
1128 		(bit(X86_FEATURE_NX) && is_efer_nx()) |
1129 #ifdef CONFIG_X86_64
1130 		bit(X86_FEATURE_LM) |
1131 #endif
1132 		bit(X86_FEATURE_MMXEXT) |
1133 		bit(X86_FEATURE_3DNOWEXT) |
1134 		bit(X86_FEATURE_3DNOW);
1135 	const u32 kvm_supported_word3_x86_features =
1136 		bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1137 	const u32 kvm_supported_word6_x86_features =
1138 		bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1139 
1140 	/* all func 2 cpuid_count() should be called on the same cpu */
1141 	get_cpu();
1142 	do_cpuid_1_ent(entry, function, index);
1143 	++*nent;
1144 
1145 	switch (function) {
1146 	case 0:
1147 		entry->eax = min(entry->eax, (u32)0xb);
1148 		break;
1149 	case 1:
1150 		entry->edx &= kvm_supported_word0_x86_features;
1151 		entry->ecx &= kvm_supported_word3_x86_features;
1152 		break;
1153 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
1154 	 * may return different values. This forces us to get_cpu() before
1155 	 * issuing the first command, and also to emulate this annoying behavior
1156 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1157 	case 2: {
1158 		int t, times = entry->eax & 0xff;
1159 
1160 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1161 		for (t = 1; t < times && *nent < maxnent; ++t) {
1162 			do_cpuid_1_ent(&entry[t], function, 0);
1163 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1164 			++*nent;
1165 		}
1166 		break;
1167 	}
1168 	/* function 4 and 0xb have additional index. */
1169 	case 4: {
1170 		int i, cache_type;
1171 
1172 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1173 		/* read more entries until cache_type is zero */
1174 		for (i = 1; *nent < maxnent; ++i) {
1175 			cache_type = entry[i - 1].eax & 0x1f;
1176 			if (!cache_type)
1177 				break;
1178 			do_cpuid_1_ent(&entry[i], function, i);
1179 			entry[i].flags |=
1180 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1181 			++*nent;
1182 		}
1183 		break;
1184 	}
1185 	case 0xb: {
1186 		int i, level_type;
1187 
1188 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1189 		/* read more entries until level_type is zero */
1190 		for (i = 1; *nent < maxnent; ++i) {
1191 			level_type = entry[i - 1].ecx & 0xff;
1192 			if (!level_type)
1193 				break;
1194 			do_cpuid_1_ent(&entry[i], function, i);
1195 			entry[i].flags |=
1196 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1197 			++*nent;
1198 		}
1199 		break;
1200 	}
1201 	case 0x80000000:
1202 		entry->eax = min(entry->eax, 0x8000001a);
1203 		break;
1204 	case 0x80000001:
1205 		entry->edx &= kvm_supported_word1_x86_features;
1206 		entry->ecx &= kvm_supported_word6_x86_features;
1207 		break;
1208 	}
1209 	put_cpu();
1210 }
1211 
1212 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1213 				    struct kvm_cpuid_entry2 __user *entries)
1214 {
1215 	struct kvm_cpuid_entry2 *cpuid_entries;
1216 	int limit, nent = 0, r = -E2BIG;
1217 	u32 func;
1218 
1219 	if (cpuid->nent < 1)
1220 		goto out;
1221 	r = -ENOMEM;
1222 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1223 	if (!cpuid_entries)
1224 		goto out;
1225 
1226 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1227 	limit = cpuid_entries[0].eax;
1228 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1229 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1230 				&nent, cpuid->nent);
1231 	r = -E2BIG;
1232 	if (nent >= cpuid->nent)
1233 		goto out_free;
1234 
1235 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1236 	limit = cpuid_entries[nent - 1].eax;
1237 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1238 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
1239 			       &nent, cpuid->nent);
1240 	r = -EFAULT;
1241 	if (copy_to_user(entries, cpuid_entries,
1242 			nent * sizeof(struct kvm_cpuid_entry2)))
1243 		goto out_free;
1244 	cpuid->nent = nent;
1245 	r = 0;
1246 
1247 out_free:
1248 	vfree(cpuid_entries);
1249 out:
1250 	return r;
1251 }
1252 
1253 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1254 				    struct kvm_lapic_state *s)
1255 {
1256 	vcpu_load(vcpu);
1257 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1258 	vcpu_put(vcpu);
1259 
1260 	return 0;
1261 }
1262 
1263 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1264 				    struct kvm_lapic_state *s)
1265 {
1266 	vcpu_load(vcpu);
1267 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1268 	kvm_apic_post_state_restore(vcpu);
1269 	vcpu_put(vcpu);
1270 
1271 	return 0;
1272 }
1273 
1274 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1275 				    struct kvm_interrupt *irq)
1276 {
1277 	if (irq->irq < 0 || irq->irq >= 256)
1278 		return -EINVAL;
1279 	if (irqchip_in_kernel(vcpu->kvm))
1280 		return -ENXIO;
1281 	vcpu_load(vcpu);
1282 
1283 	set_bit(irq->irq, vcpu->arch.irq_pending);
1284 	set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1285 
1286 	vcpu_put(vcpu);
1287 
1288 	return 0;
1289 }
1290 
1291 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1292 					   struct kvm_tpr_access_ctl *tac)
1293 {
1294 	if (tac->flags)
1295 		return -EINVAL;
1296 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
1297 	return 0;
1298 }
1299 
1300 long kvm_arch_vcpu_ioctl(struct file *filp,
1301 			 unsigned int ioctl, unsigned long arg)
1302 {
1303 	struct kvm_vcpu *vcpu = filp->private_data;
1304 	void __user *argp = (void __user *)arg;
1305 	int r;
1306 
1307 	switch (ioctl) {
1308 	case KVM_GET_LAPIC: {
1309 		struct kvm_lapic_state lapic;
1310 
1311 		memset(&lapic, 0, sizeof lapic);
1312 		r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1313 		if (r)
1314 			goto out;
1315 		r = -EFAULT;
1316 		if (copy_to_user(argp, &lapic, sizeof lapic))
1317 			goto out;
1318 		r = 0;
1319 		break;
1320 	}
1321 	case KVM_SET_LAPIC: {
1322 		struct kvm_lapic_state lapic;
1323 
1324 		r = -EFAULT;
1325 		if (copy_from_user(&lapic, argp, sizeof lapic))
1326 			goto out;
1327 		r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1328 		if (r)
1329 			goto out;
1330 		r = 0;
1331 		break;
1332 	}
1333 	case KVM_INTERRUPT: {
1334 		struct kvm_interrupt irq;
1335 
1336 		r = -EFAULT;
1337 		if (copy_from_user(&irq, argp, sizeof irq))
1338 			goto out;
1339 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1340 		if (r)
1341 			goto out;
1342 		r = 0;
1343 		break;
1344 	}
1345 	case KVM_SET_CPUID: {
1346 		struct kvm_cpuid __user *cpuid_arg = argp;
1347 		struct kvm_cpuid cpuid;
1348 
1349 		r = -EFAULT;
1350 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1351 			goto out;
1352 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1353 		if (r)
1354 			goto out;
1355 		break;
1356 	}
1357 	case KVM_SET_CPUID2: {
1358 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1359 		struct kvm_cpuid2 cpuid;
1360 
1361 		r = -EFAULT;
1362 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1363 			goto out;
1364 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1365 				cpuid_arg->entries);
1366 		if (r)
1367 			goto out;
1368 		break;
1369 	}
1370 	case KVM_GET_CPUID2: {
1371 		struct kvm_cpuid2 __user *cpuid_arg = argp;
1372 		struct kvm_cpuid2 cpuid;
1373 
1374 		r = -EFAULT;
1375 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1376 			goto out;
1377 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1378 				cpuid_arg->entries);
1379 		if (r)
1380 			goto out;
1381 		r = -EFAULT;
1382 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1383 			goto out;
1384 		r = 0;
1385 		break;
1386 	}
1387 	case KVM_GET_MSRS:
1388 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
1389 		break;
1390 	case KVM_SET_MSRS:
1391 		r = msr_io(vcpu, argp, do_set_msr, 0);
1392 		break;
1393 	case KVM_TPR_ACCESS_REPORTING: {
1394 		struct kvm_tpr_access_ctl tac;
1395 
1396 		r = -EFAULT;
1397 		if (copy_from_user(&tac, argp, sizeof tac))
1398 			goto out;
1399 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1400 		if (r)
1401 			goto out;
1402 		r = -EFAULT;
1403 		if (copy_to_user(argp, &tac, sizeof tac))
1404 			goto out;
1405 		r = 0;
1406 		break;
1407 	};
1408 	case KVM_SET_VAPIC_ADDR: {
1409 		struct kvm_vapic_addr va;
1410 
1411 		r = -EINVAL;
1412 		if (!irqchip_in_kernel(vcpu->kvm))
1413 			goto out;
1414 		r = -EFAULT;
1415 		if (copy_from_user(&va, argp, sizeof va))
1416 			goto out;
1417 		r = 0;
1418 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1419 		break;
1420 	}
1421 	default:
1422 		r = -EINVAL;
1423 	}
1424 out:
1425 	return r;
1426 }
1427 
1428 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1429 {
1430 	int ret;
1431 
1432 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
1433 		return -1;
1434 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1435 	return ret;
1436 }
1437 
1438 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1439 					  u32 kvm_nr_mmu_pages)
1440 {
1441 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1442 		return -EINVAL;
1443 
1444 	down_write(&kvm->slots_lock);
1445 
1446 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1447 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1448 
1449 	up_write(&kvm->slots_lock);
1450 	return 0;
1451 }
1452 
1453 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1454 {
1455 	return kvm->arch.n_alloc_mmu_pages;
1456 }
1457 
1458 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1459 {
1460 	int i;
1461 	struct kvm_mem_alias *alias;
1462 
1463 	for (i = 0; i < kvm->arch.naliases; ++i) {
1464 		alias = &kvm->arch.aliases[i];
1465 		if (gfn >= alias->base_gfn
1466 		    && gfn < alias->base_gfn + alias->npages)
1467 			return alias->target_gfn + gfn - alias->base_gfn;
1468 	}
1469 	return gfn;
1470 }
1471 
1472 /*
1473  * Set a new alias region.  Aliases map a portion of physical memory into
1474  * another portion.  This is useful for memory windows, for example the PC
1475  * VGA region.
1476  */
1477 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1478 					 struct kvm_memory_alias *alias)
1479 {
1480 	int r, n;
1481 	struct kvm_mem_alias *p;
1482 
1483 	r = -EINVAL;
1484 	/* General sanity checks */
1485 	if (alias->memory_size & (PAGE_SIZE - 1))
1486 		goto out;
1487 	if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1488 		goto out;
1489 	if (alias->slot >= KVM_ALIAS_SLOTS)
1490 		goto out;
1491 	if (alias->guest_phys_addr + alias->memory_size
1492 	    < alias->guest_phys_addr)
1493 		goto out;
1494 	if (alias->target_phys_addr + alias->memory_size
1495 	    < alias->target_phys_addr)
1496 		goto out;
1497 
1498 	down_write(&kvm->slots_lock);
1499 	spin_lock(&kvm->mmu_lock);
1500 
1501 	p = &kvm->arch.aliases[alias->slot];
1502 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1503 	p->npages = alias->memory_size >> PAGE_SHIFT;
1504 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1505 
1506 	for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1507 		if (kvm->arch.aliases[n - 1].npages)
1508 			break;
1509 	kvm->arch.naliases = n;
1510 
1511 	spin_unlock(&kvm->mmu_lock);
1512 	kvm_mmu_zap_all(kvm);
1513 
1514 	up_write(&kvm->slots_lock);
1515 
1516 	return 0;
1517 
1518 out:
1519 	return r;
1520 }
1521 
1522 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1523 {
1524 	int r;
1525 
1526 	r = 0;
1527 	switch (chip->chip_id) {
1528 	case KVM_IRQCHIP_PIC_MASTER:
1529 		memcpy(&chip->chip.pic,
1530 			&pic_irqchip(kvm)->pics[0],
1531 			sizeof(struct kvm_pic_state));
1532 		break;
1533 	case KVM_IRQCHIP_PIC_SLAVE:
1534 		memcpy(&chip->chip.pic,
1535 			&pic_irqchip(kvm)->pics[1],
1536 			sizeof(struct kvm_pic_state));
1537 		break;
1538 	case KVM_IRQCHIP_IOAPIC:
1539 		memcpy(&chip->chip.ioapic,
1540 			ioapic_irqchip(kvm),
1541 			sizeof(struct kvm_ioapic_state));
1542 		break;
1543 	default:
1544 		r = -EINVAL;
1545 		break;
1546 	}
1547 	return r;
1548 }
1549 
1550 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1551 {
1552 	int r;
1553 
1554 	r = 0;
1555 	switch (chip->chip_id) {
1556 	case KVM_IRQCHIP_PIC_MASTER:
1557 		memcpy(&pic_irqchip(kvm)->pics[0],
1558 			&chip->chip.pic,
1559 			sizeof(struct kvm_pic_state));
1560 		break;
1561 	case KVM_IRQCHIP_PIC_SLAVE:
1562 		memcpy(&pic_irqchip(kvm)->pics[1],
1563 			&chip->chip.pic,
1564 			sizeof(struct kvm_pic_state));
1565 		break;
1566 	case KVM_IRQCHIP_IOAPIC:
1567 		memcpy(ioapic_irqchip(kvm),
1568 			&chip->chip.ioapic,
1569 			sizeof(struct kvm_ioapic_state));
1570 		break;
1571 	default:
1572 		r = -EINVAL;
1573 		break;
1574 	}
1575 	kvm_pic_update_irq(pic_irqchip(kvm));
1576 	return r;
1577 }
1578 
1579 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1580 {
1581 	int r = 0;
1582 
1583 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1584 	return r;
1585 }
1586 
1587 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1588 {
1589 	int r = 0;
1590 
1591 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1592 	kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1593 	return r;
1594 }
1595 
1596 /*
1597  * Get (and clear) the dirty memory log for a memory slot.
1598  */
1599 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1600 				      struct kvm_dirty_log *log)
1601 {
1602 	int r;
1603 	int n;
1604 	struct kvm_memory_slot *memslot;
1605 	int is_dirty = 0;
1606 
1607 	down_write(&kvm->slots_lock);
1608 
1609 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1610 	if (r)
1611 		goto out;
1612 
1613 	/* If nothing is dirty, don't bother messing with page tables. */
1614 	if (is_dirty) {
1615 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
1616 		kvm_flush_remote_tlbs(kvm);
1617 		memslot = &kvm->memslots[log->slot];
1618 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1619 		memset(memslot->dirty_bitmap, 0, n);
1620 	}
1621 	r = 0;
1622 out:
1623 	up_write(&kvm->slots_lock);
1624 	return r;
1625 }
1626 
1627 long kvm_arch_vm_ioctl(struct file *filp,
1628 		       unsigned int ioctl, unsigned long arg)
1629 {
1630 	struct kvm *kvm = filp->private_data;
1631 	void __user *argp = (void __user *)arg;
1632 	int r = -EINVAL;
1633 
1634 	switch (ioctl) {
1635 	case KVM_SET_TSS_ADDR:
1636 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1637 		if (r < 0)
1638 			goto out;
1639 		break;
1640 	case KVM_SET_MEMORY_REGION: {
1641 		struct kvm_memory_region kvm_mem;
1642 		struct kvm_userspace_memory_region kvm_userspace_mem;
1643 
1644 		r = -EFAULT;
1645 		if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1646 			goto out;
1647 		kvm_userspace_mem.slot = kvm_mem.slot;
1648 		kvm_userspace_mem.flags = kvm_mem.flags;
1649 		kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1650 		kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1651 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1652 		if (r)
1653 			goto out;
1654 		break;
1655 	}
1656 	case KVM_SET_NR_MMU_PAGES:
1657 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1658 		if (r)
1659 			goto out;
1660 		break;
1661 	case KVM_GET_NR_MMU_PAGES:
1662 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1663 		break;
1664 	case KVM_SET_MEMORY_ALIAS: {
1665 		struct kvm_memory_alias alias;
1666 
1667 		r = -EFAULT;
1668 		if (copy_from_user(&alias, argp, sizeof alias))
1669 			goto out;
1670 		r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1671 		if (r)
1672 			goto out;
1673 		break;
1674 	}
1675 	case KVM_CREATE_IRQCHIP:
1676 		r = -ENOMEM;
1677 		kvm->arch.vpic = kvm_create_pic(kvm);
1678 		if (kvm->arch.vpic) {
1679 			r = kvm_ioapic_init(kvm);
1680 			if (r) {
1681 				kfree(kvm->arch.vpic);
1682 				kvm->arch.vpic = NULL;
1683 				goto out;
1684 			}
1685 		} else
1686 			goto out;
1687 		break;
1688 	case KVM_CREATE_PIT:
1689 		r = -ENOMEM;
1690 		kvm->arch.vpit = kvm_create_pit(kvm);
1691 		if (kvm->arch.vpit)
1692 			r = 0;
1693 		break;
1694 	case KVM_IRQ_LINE: {
1695 		struct kvm_irq_level irq_event;
1696 
1697 		r = -EFAULT;
1698 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
1699 			goto out;
1700 		if (irqchip_in_kernel(kvm)) {
1701 			mutex_lock(&kvm->lock);
1702 			if (irq_event.irq < 16)
1703 				kvm_pic_set_irq(pic_irqchip(kvm),
1704 					irq_event.irq,
1705 					irq_event.level);
1706 			kvm_ioapic_set_irq(kvm->arch.vioapic,
1707 					irq_event.irq,
1708 					irq_event.level);
1709 			mutex_unlock(&kvm->lock);
1710 			r = 0;
1711 		}
1712 		break;
1713 	}
1714 	case KVM_GET_IRQCHIP: {
1715 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1716 		struct kvm_irqchip chip;
1717 
1718 		r = -EFAULT;
1719 		if (copy_from_user(&chip, argp, sizeof chip))
1720 			goto out;
1721 		r = -ENXIO;
1722 		if (!irqchip_in_kernel(kvm))
1723 			goto out;
1724 		r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1725 		if (r)
1726 			goto out;
1727 		r = -EFAULT;
1728 		if (copy_to_user(argp, &chip, sizeof chip))
1729 			goto out;
1730 		r = 0;
1731 		break;
1732 	}
1733 	case KVM_SET_IRQCHIP: {
1734 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1735 		struct kvm_irqchip chip;
1736 
1737 		r = -EFAULT;
1738 		if (copy_from_user(&chip, argp, sizeof chip))
1739 			goto out;
1740 		r = -ENXIO;
1741 		if (!irqchip_in_kernel(kvm))
1742 			goto out;
1743 		r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1744 		if (r)
1745 			goto out;
1746 		r = 0;
1747 		break;
1748 	}
1749 	case KVM_GET_PIT: {
1750 		struct kvm_pit_state ps;
1751 		r = -EFAULT;
1752 		if (copy_from_user(&ps, argp, sizeof ps))
1753 			goto out;
1754 		r = -ENXIO;
1755 		if (!kvm->arch.vpit)
1756 			goto out;
1757 		r = kvm_vm_ioctl_get_pit(kvm, &ps);
1758 		if (r)
1759 			goto out;
1760 		r = -EFAULT;
1761 		if (copy_to_user(argp, &ps, sizeof ps))
1762 			goto out;
1763 		r = 0;
1764 		break;
1765 	}
1766 	case KVM_SET_PIT: {
1767 		struct kvm_pit_state ps;
1768 		r = -EFAULT;
1769 		if (copy_from_user(&ps, argp, sizeof ps))
1770 			goto out;
1771 		r = -ENXIO;
1772 		if (!kvm->arch.vpit)
1773 			goto out;
1774 		r = kvm_vm_ioctl_set_pit(kvm, &ps);
1775 		if (r)
1776 			goto out;
1777 		r = 0;
1778 		break;
1779 	}
1780 	default:
1781 		;
1782 	}
1783 out:
1784 	return r;
1785 }
1786 
1787 static void kvm_init_msr_list(void)
1788 {
1789 	u32 dummy[2];
1790 	unsigned i, j;
1791 
1792 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1793 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1794 			continue;
1795 		if (j < i)
1796 			msrs_to_save[j] = msrs_to_save[i];
1797 		j++;
1798 	}
1799 	num_msrs_to_save = j;
1800 }
1801 
1802 /*
1803  * Only apic need an MMIO device hook, so shortcut now..
1804  */
1805 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1806 						gpa_t addr, int len,
1807 						int is_write)
1808 {
1809 	struct kvm_io_device *dev;
1810 
1811 	if (vcpu->arch.apic) {
1812 		dev = &vcpu->arch.apic->dev;
1813 		if (dev->in_range(dev, addr, len, is_write))
1814 			return dev;
1815 	}
1816 	return NULL;
1817 }
1818 
1819 
1820 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1821 						gpa_t addr, int len,
1822 						int is_write)
1823 {
1824 	struct kvm_io_device *dev;
1825 
1826 	dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1827 	if (dev == NULL)
1828 		dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1829 					  is_write);
1830 	return dev;
1831 }
1832 
1833 int emulator_read_std(unsigned long addr,
1834 			     void *val,
1835 			     unsigned int bytes,
1836 			     struct kvm_vcpu *vcpu)
1837 {
1838 	void *data = val;
1839 	int r = X86EMUL_CONTINUE;
1840 
1841 	while (bytes) {
1842 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1843 		unsigned offset = addr & (PAGE_SIZE-1);
1844 		unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1845 		int ret;
1846 
1847 		if (gpa == UNMAPPED_GVA) {
1848 			r = X86EMUL_PROPAGATE_FAULT;
1849 			goto out;
1850 		}
1851 		ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1852 		if (ret < 0) {
1853 			r = X86EMUL_UNHANDLEABLE;
1854 			goto out;
1855 		}
1856 
1857 		bytes -= tocopy;
1858 		data += tocopy;
1859 		addr += tocopy;
1860 	}
1861 out:
1862 	return r;
1863 }
1864 EXPORT_SYMBOL_GPL(emulator_read_std);
1865 
1866 static int emulator_read_emulated(unsigned long addr,
1867 				  void *val,
1868 				  unsigned int bytes,
1869 				  struct kvm_vcpu *vcpu)
1870 {
1871 	struct kvm_io_device *mmio_dev;
1872 	gpa_t                 gpa;
1873 
1874 	if (vcpu->mmio_read_completed) {
1875 		memcpy(val, vcpu->mmio_data, bytes);
1876 		vcpu->mmio_read_completed = 0;
1877 		return X86EMUL_CONTINUE;
1878 	}
1879 
1880 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1881 
1882 	/* For APIC access vmexit */
1883 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1884 		goto mmio;
1885 
1886 	if (emulator_read_std(addr, val, bytes, vcpu)
1887 			== X86EMUL_CONTINUE)
1888 		return X86EMUL_CONTINUE;
1889 	if (gpa == UNMAPPED_GVA)
1890 		return X86EMUL_PROPAGATE_FAULT;
1891 
1892 mmio:
1893 	/*
1894 	 * Is this MMIO handled locally?
1895 	 */
1896 	mutex_lock(&vcpu->kvm->lock);
1897 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1898 	if (mmio_dev) {
1899 		kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1900 		mutex_unlock(&vcpu->kvm->lock);
1901 		return X86EMUL_CONTINUE;
1902 	}
1903 	mutex_unlock(&vcpu->kvm->lock);
1904 
1905 	vcpu->mmio_needed = 1;
1906 	vcpu->mmio_phys_addr = gpa;
1907 	vcpu->mmio_size = bytes;
1908 	vcpu->mmio_is_write = 0;
1909 
1910 	return X86EMUL_UNHANDLEABLE;
1911 }
1912 
1913 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1914 			  const void *val, int bytes)
1915 {
1916 	int ret;
1917 
1918 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1919 	if (ret < 0)
1920 		return 0;
1921 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1922 	return 1;
1923 }
1924 
1925 static int emulator_write_emulated_onepage(unsigned long addr,
1926 					   const void *val,
1927 					   unsigned int bytes,
1928 					   struct kvm_vcpu *vcpu)
1929 {
1930 	struct kvm_io_device *mmio_dev;
1931 	gpa_t                 gpa;
1932 
1933 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1934 
1935 	if (gpa == UNMAPPED_GVA) {
1936 		kvm_inject_page_fault(vcpu, addr, 2);
1937 		return X86EMUL_PROPAGATE_FAULT;
1938 	}
1939 
1940 	/* For APIC access vmexit */
1941 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1942 		goto mmio;
1943 
1944 	if (emulator_write_phys(vcpu, gpa, val, bytes))
1945 		return X86EMUL_CONTINUE;
1946 
1947 mmio:
1948 	/*
1949 	 * Is this MMIO handled locally?
1950 	 */
1951 	mutex_lock(&vcpu->kvm->lock);
1952 	mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
1953 	if (mmio_dev) {
1954 		kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1955 		mutex_unlock(&vcpu->kvm->lock);
1956 		return X86EMUL_CONTINUE;
1957 	}
1958 	mutex_unlock(&vcpu->kvm->lock);
1959 
1960 	vcpu->mmio_needed = 1;
1961 	vcpu->mmio_phys_addr = gpa;
1962 	vcpu->mmio_size = bytes;
1963 	vcpu->mmio_is_write = 1;
1964 	memcpy(vcpu->mmio_data, val, bytes);
1965 
1966 	return X86EMUL_CONTINUE;
1967 }
1968 
1969 int emulator_write_emulated(unsigned long addr,
1970 				   const void *val,
1971 				   unsigned int bytes,
1972 				   struct kvm_vcpu *vcpu)
1973 {
1974 	/* Crossing a page boundary? */
1975 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1976 		int rc, now;
1977 
1978 		now = -addr & ~PAGE_MASK;
1979 		rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1980 		if (rc != X86EMUL_CONTINUE)
1981 			return rc;
1982 		addr += now;
1983 		val += now;
1984 		bytes -= now;
1985 	}
1986 	return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1987 }
1988 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1989 
1990 static int emulator_cmpxchg_emulated(unsigned long addr,
1991 				     const void *old,
1992 				     const void *new,
1993 				     unsigned int bytes,
1994 				     struct kvm_vcpu *vcpu)
1995 {
1996 	static int reported;
1997 
1998 	if (!reported) {
1999 		reported = 1;
2000 		printk(KERN_WARNING "kvm: emulating exchange as write\n");
2001 	}
2002 #ifndef CONFIG_X86_64
2003 	/* guests cmpxchg8b have to be emulated atomically */
2004 	if (bytes == 8) {
2005 		gpa_t gpa;
2006 		struct page *page;
2007 		char *kaddr;
2008 		u64 val;
2009 
2010 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2011 
2012 		if (gpa == UNMAPPED_GVA ||
2013 		   (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2014 			goto emul_write;
2015 
2016 		if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2017 			goto emul_write;
2018 
2019 		val = *(u64 *)new;
2020 
2021 		down_read(&current->mm->mmap_sem);
2022 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2023 		up_read(&current->mm->mmap_sem);
2024 
2025 		kaddr = kmap_atomic(page, KM_USER0);
2026 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2027 		kunmap_atomic(kaddr, KM_USER0);
2028 		kvm_release_page_dirty(page);
2029 	}
2030 emul_write:
2031 #endif
2032 
2033 	return emulator_write_emulated(addr, new, bytes, vcpu);
2034 }
2035 
2036 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2037 {
2038 	return kvm_x86_ops->get_segment_base(vcpu, seg);
2039 }
2040 
2041 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2042 {
2043 	return X86EMUL_CONTINUE;
2044 }
2045 
2046 int emulate_clts(struct kvm_vcpu *vcpu)
2047 {
2048 	KVMTRACE_0D(CLTS, vcpu, handler);
2049 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2050 	return X86EMUL_CONTINUE;
2051 }
2052 
2053 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2054 {
2055 	struct kvm_vcpu *vcpu = ctxt->vcpu;
2056 
2057 	switch (dr) {
2058 	case 0 ... 3:
2059 		*dest = kvm_x86_ops->get_dr(vcpu, dr);
2060 		return X86EMUL_CONTINUE;
2061 	default:
2062 		pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2063 		return X86EMUL_UNHANDLEABLE;
2064 	}
2065 }
2066 
2067 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2068 {
2069 	unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2070 	int exception;
2071 
2072 	kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2073 	if (exception) {
2074 		/* FIXME: better handling */
2075 		return X86EMUL_UNHANDLEABLE;
2076 	}
2077 	return X86EMUL_CONTINUE;
2078 }
2079 
2080 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2081 {
2082 	u8 opcodes[4];
2083 	unsigned long rip = vcpu->arch.rip;
2084 	unsigned long rip_linear;
2085 
2086 	if (!printk_ratelimit())
2087 		return;
2088 
2089 	rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2090 
2091 	emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2092 
2093 	printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2094 	       context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2095 }
2096 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2097 
2098 static struct x86_emulate_ops emulate_ops = {
2099 	.read_std            = emulator_read_std,
2100 	.read_emulated       = emulator_read_emulated,
2101 	.write_emulated      = emulator_write_emulated,
2102 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
2103 };
2104 
2105 int emulate_instruction(struct kvm_vcpu *vcpu,
2106 			struct kvm_run *run,
2107 			unsigned long cr2,
2108 			u16 error_code,
2109 			int emulation_type)
2110 {
2111 	int r;
2112 	struct decode_cache *c;
2113 
2114 	vcpu->arch.mmio_fault_cr2 = cr2;
2115 	kvm_x86_ops->cache_regs(vcpu);
2116 
2117 	vcpu->mmio_is_write = 0;
2118 	vcpu->arch.pio.string = 0;
2119 
2120 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2121 		int cs_db, cs_l;
2122 		kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2123 
2124 		vcpu->arch.emulate_ctxt.vcpu = vcpu;
2125 		vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2126 		vcpu->arch.emulate_ctxt.mode =
2127 			(vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2128 			? X86EMUL_MODE_REAL : cs_l
2129 			? X86EMUL_MODE_PROT64 :	cs_db
2130 			? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2131 
2132 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2133 
2134 		/* Reject the instructions other than VMCALL/VMMCALL when
2135 		 * try to emulate invalid opcode */
2136 		c = &vcpu->arch.emulate_ctxt.decode;
2137 		if ((emulation_type & EMULTYPE_TRAP_UD) &&
2138 		    (!(c->twobyte && c->b == 0x01 &&
2139 		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2140 		       c->modrm_mod == 3 && c->modrm_rm == 1)))
2141 			return EMULATE_FAIL;
2142 
2143 		++vcpu->stat.insn_emulation;
2144 		if (r)  {
2145 			++vcpu->stat.insn_emulation_fail;
2146 			if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2147 				return EMULATE_DONE;
2148 			return EMULATE_FAIL;
2149 		}
2150 	}
2151 
2152 	r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2153 
2154 	if (vcpu->arch.pio.string)
2155 		return EMULATE_DO_MMIO;
2156 
2157 	if ((r || vcpu->mmio_is_write) && run) {
2158 		run->exit_reason = KVM_EXIT_MMIO;
2159 		run->mmio.phys_addr = vcpu->mmio_phys_addr;
2160 		memcpy(run->mmio.data, vcpu->mmio_data, 8);
2161 		run->mmio.len = vcpu->mmio_size;
2162 		run->mmio.is_write = vcpu->mmio_is_write;
2163 	}
2164 
2165 	if (r) {
2166 		if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2167 			return EMULATE_DONE;
2168 		if (!vcpu->mmio_needed) {
2169 			kvm_report_emulation_failure(vcpu, "mmio");
2170 			return EMULATE_FAIL;
2171 		}
2172 		return EMULATE_DO_MMIO;
2173 	}
2174 
2175 	kvm_x86_ops->decache_regs(vcpu);
2176 	kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2177 
2178 	if (vcpu->mmio_is_write) {
2179 		vcpu->mmio_needed = 0;
2180 		return EMULATE_DO_MMIO;
2181 	}
2182 
2183 	return EMULATE_DONE;
2184 }
2185 EXPORT_SYMBOL_GPL(emulate_instruction);
2186 
2187 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2188 {
2189 	int i;
2190 
2191 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2192 		if (vcpu->arch.pio.guest_pages[i]) {
2193 			kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2194 			vcpu->arch.pio.guest_pages[i] = NULL;
2195 		}
2196 }
2197 
2198 static int pio_copy_data(struct kvm_vcpu *vcpu)
2199 {
2200 	void *p = vcpu->arch.pio_data;
2201 	void *q;
2202 	unsigned bytes;
2203 	int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2204 
2205 	q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2206 		 PAGE_KERNEL);
2207 	if (!q) {
2208 		free_pio_guest_pages(vcpu);
2209 		return -ENOMEM;
2210 	}
2211 	q += vcpu->arch.pio.guest_page_offset;
2212 	bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2213 	if (vcpu->arch.pio.in)
2214 		memcpy(q, p, bytes);
2215 	else
2216 		memcpy(p, q, bytes);
2217 	q -= vcpu->arch.pio.guest_page_offset;
2218 	vunmap(q);
2219 	free_pio_guest_pages(vcpu);
2220 	return 0;
2221 }
2222 
2223 int complete_pio(struct kvm_vcpu *vcpu)
2224 {
2225 	struct kvm_pio_request *io = &vcpu->arch.pio;
2226 	long delta;
2227 	int r;
2228 
2229 	kvm_x86_ops->cache_regs(vcpu);
2230 
2231 	if (!io->string) {
2232 		if (io->in)
2233 			memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2234 			       io->size);
2235 	} else {
2236 		if (io->in) {
2237 			r = pio_copy_data(vcpu);
2238 			if (r) {
2239 				kvm_x86_ops->cache_regs(vcpu);
2240 				return r;
2241 			}
2242 		}
2243 
2244 		delta = 1;
2245 		if (io->rep) {
2246 			delta *= io->cur_count;
2247 			/*
2248 			 * The size of the register should really depend on
2249 			 * current address size.
2250 			 */
2251 			vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2252 		}
2253 		if (io->down)
2254 			delta = -delta;
2255 		delta *= io->size;
2256 		if (io->in)
2257 			vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2258 		else
2259 			vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2260 	}
2261 
2262 	kvm_x86_ops->decache_regs(vcpu);
2263 
2264 	io->count -= io->cur_count;
2265 	io->cur_count = 0;
2266 
2267 	return 0;
2268 }
2269 
2270 static void kernel_pio(struct kvm_io_device *pio_dev,
2271 		       struct kvm_vcpu *vcpu,
2272 		       void *pd)
2273 {
2274 	/* TODO: String I/O for in kernel device */
2275 
2276 	mutex_lock(&vcpu->kvm->lock);
2277 	if (vcpu->arch.pio.in)
2278 		kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2279 				  vcpu->arch.pio.size,
2280 				  pd);
2281 	else
2282 		kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2283 				   vcpu->arch.pio.size,
2284 				   pd);
2285 	mutex_unlock(&vcpu->kvm->lock);
2286 }
2287 
2288 static void pio_string_write(struct kvm_io_device *pio_dev,
2289 			     struct kvm_vcpu *vcpu)
2290 {
2291 	struct kvm_pio_request *io = &vcpu->arch.pio;
2292 	void *pd = vcpu->arch.pio_data;
2293 	int i;
2294 
2295 	mutex_lock(&vcpu->kvm->lock);
2296 	for (i = 0; i < io->cur_count; i++) {
2297 		kvm_iodevice_write(pio_dev, io->port,
2298 				   io->size,
2299 				   pd);
2300 		pd += io->size;
2301 	}
2302 	mutex_unlock(&vcpu->kvm->lock);
2303 }
2304 
2305 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2306 					       gpa_t addr, int len,
2307 					       int is_write)
2308 {
2309 	return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2310 }
2311 
2312 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2313 		  int size, unsigned port)
2314 {
2315 	struct kvm_io_device *pio_dev;
2316 
2317 	vcpu->run->exit_reason = KVM_EXIT_IO;
2318 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2319 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2320 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2321 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2322 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2323 	vcpu->arch.pio.in = in;
2324 	vcpu->arch.pio.string = 0;
2325 	vcpu->arch.pio.down = 0;
2326 	vcpu->arch.pio.guest_page_offset = 0;
2327 	vcpu->arch.pio.rep = 0;
2328 
2329 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2330 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2331 			    handler);
2332 	else
2333 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2334 			    handler);
2335 
2336 	kvm_x86_ops->cache_regs(vcpu);
2337 	memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2338 
2339 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2340 
2341 	pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2342 	if (pio_dev) {
2343 		kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2344 		complete_pio(vcpu);
2345 		return 1;
2346 	}
2347 	return 0;
2348 }
2349 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2350 
2351 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2352 		  int size, unsigned long count, int down,
2353 		  gva_t address, int rep, unsigned port)
2354 {
2355 	unsigned now, in_page;
2356 	int i, ret = 0;
2357 	int nr_pages = 1;
2358 	struct page *page;
2359 	struct kvm_io_device *pio_dev;
2360 
2361 	vcpu->run->exit_reason = KVM_EXIT_IO;
2362 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2363 	vcpu->run->io.size = vcpu->arch.pio.size = size;
2364 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2365 	vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2366 	vcpu->run->io.port = vcpu->arch.pio.port = port;
2367 	vcpu->arch.pio.in = in;
2368 	vcpu->arch.pio.string = 1;
2369 	vcpu->arch.pio.down = down;
2370 	vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2371 	vcpu->arch.pio.rep = rep;
2372 
2373 	if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2374 		KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2375 			    handler);
2376 	else
2377 		KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2378 			    handler);
2379 
2380 	if (!count) {
2381 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2382 		return 1;
2383 	}
2384 
2385 	if (!down)
2386 		in_page = PAGE_SIZE - offset_in_page(address);
2387 	else
2388 		in_page = offset_in_page(address) + size;
2389 	now = min(count, (unsigned long)in_page / size);
2390 	if (!now) {
2391 		/*
2392 		 * String I/O straddles page boundary.  Pin two guest pages
2393 		 * so that we satisfy atomicity constraints.  Do just one
2394 		 * transaction to avoid complexity.
2395 		 */
2396 		nr_pages = 2;
2397 		now = 1;
2398 	}
2399 	if (down) {
2400 		/*
2401 		 * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2402 		 */
2403 		pr_unimpl(vcpu, "guest string pio down\n");
2404 		kvm_inject_gp(vcpu, 0);
2405 		return 1;
2406 	}
2407 	vcpu->run->io.count = now;
2408 	vcpu->arch.pio.cur_count = now;
2409 
2410 	if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2411 		kvm_x86_ops->skip_emulated_instruction(vcpu);
2412 
2413 	for (i = 0; i < nr_pages; ++i) {
2414 		page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2415 		vcpu->arch.pio.guest_pages[i] = page;
2416 		if (!page) {
2417 			kvm_inject_gp(vcpu, 0);
2418 			free_pio_guest_pages(vcpu);
2419 			return 1;
2420 		}
2421 	}
2422 
2423 	pio_dev = vcpu_find_pio_dev(vcpu, port,
2424 				    vcpu->arch.pio.cur_count,
2425 				    !vcpu->arch.pio.in);
2426 	if (!vcpu->arch.pio.in) {
2427 		/* string PIO write */
2428 		ret = pio_copy_data(vcpu);
2429 		if (ret >= 0 && pio_dev) {
2430 			pio_string_write(pio_dev, vcpu);
2431 			complete_pio(vcpu);
2432 			if (vcpu->arch.pio.count == 0)
2433 				ret = 1;
2434 		}
2435 	} else if (pio_dev)
2436 		pr_unimpl(vcpu, "no string pio read support yet, "
2437 		       "port %x size %d count %ld\n",
2438 			port, size, count);
2439 
2440 	return ret;
2441 }
2442 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2443 
2444 int kvm_arch_init(void *opaque)
2445 {
2446 	int r;
2447 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2448 
2449 	if (kvm_x86_ops) {
2450 		printk(KERN_ERR "kvm: already loaded the other module\n");
2451 		r = -EEXIST;
2452 		goto out;
2453 	}
2454 
2455 	if (!ops->cpu_has_kvm_support()) {
2456 		printk(KERN_ERR "kvm: no hardware support\n");
2457 		r = -EOPNOTSUPP;
2458 		goto out;
2459 	}
2460 	if (ops->disabled_by_bios()) {
2461 		printk(KERN_ERR "kvm: disabled by bios\n");
2462 		r = -EOPNOTSUPP;
2463 		goto out;
2464 	}
2465 
2466 	r = kvm_mmu_module_init();
2467 	if (r)
2468 		goto out;
2469 
2470 	kvm_init_msr_list();
2471 
2472 	kvm_x86_ops = ops;
2473 	kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2474 	kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2475 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2476 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
2477 	return 0;
2478 
2479 out:
2480 	return r;
2481 }
2482 
2483 void kvm_arch_exit(void)
2484 {
2485 	kvm_x86_ops = NULL;
2486 	kvm_mmu_module_exit();
2487 }
2488 
2489 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2490 {
2491 	++vcpu->stat.halt_exits;
2492 	KVMTRACE_0D(HLT, vcpu, handler);
2493 	if (irqchip_in_kernel(vcpu->kvm)) {
2494 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2495 		up_read(&vcpu->kvm->slots_lock);
2496 		kvm_vcpu_block(vcpu);
2497 		down_read(&vcpu->kvm->slots_lock);
2498 		if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2499 			return -EINTR;
2500 		return 1;
2501 	} else {
2502 		vcpu->run->exit_reason = KVM_EXIT_HLT;
2503 		return 0;
2504 	}
2505 }
2506 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2507 
2508 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2509 			   unsigned long a1)
2510 {
2511 	if (is_long_mode(vcpu))
2512 		return a0;
2513 	else
2514 		return a0 | ((gpa_t)a1 << 32);
2515 }
2516 
2517 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2518 {
2519 	unsigned long nr, a0, a1, a2, a3, ret;
2520 	int r = 1;
2521 
2522 	kvm_x86_ops->cache_regs(vcpu);
2523 
2524 	nr = vcpu->arch.regs[VCPU_REGS_RAX];
2525 	a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2526 	a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2527 	a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2528 	a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2529 
2530 	KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2531 
2532 	if (!is_long_mode(vcpu)) {
2533 		nr &= 0xFFFFFFFF;
2534 		a0 &= 0xFFFFFFFF;
2535 		a1 &= 0xFFFFFFFF;
2536 		a2 &= 0xFFFFFFFF;
2537 		a3 &= 0xFFFFFFFF;
2538 	}
2539 
2540 	switch (nr) {
2541 	case KVM_HC_VAPIC_POLL_IRQ:
2542 		ret = 0;
2543 		break;
2544 	case KVM_HC_MMU_OP:
2545 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2546 		break;
2547 	default:
2548 		ret = -KVM_ENOSYS;
2549 		break;
2550 	}
2551 	vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2552 	kvm_x86_ops->decache_regs(vcpu);
2553 	++vcpu->stat.hypercalls;
2554 	return r;
2555 }
2556 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2557 
2558 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2559 {
2560 	char instruction[3];
2561 	int ret = 0;
2562 
2563 
2564 	/*
2565 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
2566 	 * to ensure that the updated hypercall appears atomically across all
2567 	 * VCPUs.
2568 	 */
2569 	kvm_mmu_zap_all(vcpu->kvm);
2570 
2571 	kvm_x86_ops->cache_regs(vcpu);
2572 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
2573 	if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2574 	    != X86EMUL_CONTINUE)
2575 		ret = -EFAULT;
2576 
2577 	return ret;
2578 }
2579 
2580 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2581 {
2582 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2583 }
2584 
2585 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2586 {
2587 	struct descriptor_table dt = { limit, base };
2588 
2589 	kvm_x86_ops->set_gdt(vcpu, &dt);
2590 }
2591 
2592 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2593 {
2594 	struct descriptor_table dt = { limit, base };
2595 
2596 	kvm_x86_ops->set_idt(vcpu, &dt);
2597 }
2598 
2599 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2600 		   unsigned long *rflags)
2601 {
2602 	kvm_lmsw(vcpu, msw);
2603 	*rflags = kvm_x86_ops->get_rflags(vcpu);
2604 }
2605 
2606 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2607 {
2608 	unsigned long value;
2609 
2610 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2611 	switch (cr) {
2612 	case 0:
2613 		value = vcpu->arch.cr0;
2614 		break;
2615 	case 2:
2616 		value = vcpu->arch.cr2;
2617 		break;
2618 	case 3:
2619 		value = vcpu->arch.cr3;
2620 		break;
2621 	case 4:
2622 		value = vcpu->arch.cr4;
2623 		break;
2624 	case 8:
2625 		value = kvm_get_cr8(vcpu);
2626 		break;
2627 	default:
2628 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2629 		return 0;
2630 	}
2631 	KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2632 		    (u32)((u64)value >> 32), handler);
2633 
2634 	return value;
2635 }
2636 
2637 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2638 		     unsigned long *rflags)
2639 {
2640 	KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2641 		    (u32)((u64)val >> 32), handler);
2642 
2643 	switch (cr) {
2644 	case 0:
2645 		kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2646 		*rflags = kvm_x86_ops->get_rflags(vcpu);
2647 		break;
2648 	case 2:
2649 		vcpu->arch.cr2 = val;
2650 		break;
2651 	case 3:
2652 		kvm_set_cr3(vcpu, val);
2653 		break;
2654 	case 4:
2655 		kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2656 		break;
2657 	case 8:
2658 		kvm_set_cr8(vcpu, val & 0xfUL);
2659 		break;
2660 	default:
2661 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2662 	}
2663 }
2664 
2665 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2666 {
2667 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2668 	int j, nent = vcpu->arch.cpuid_nent;
2669 
2670 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2671 	/* when no next entry is found, the current entry[i] is reselected */
2672 	for (j = i + 1; j == i; j = (j + 1) % nent) {
2673 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2674 		if (ej->function == e->function) {
2675 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2676 			return j;
2677 		}
2678 	}
2679 	return 0; /* silence gcc, even though control never reaches here */
2680 }
2681 
2682 /* find an entry with matching function, matching index (if needed), and that
2683  * should be read next (if it's stateful) */
2684 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2685 	u32 function, u32 index)
2686 {
2687 	if (e->function != function)
2688 		return 0;
2689 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2690 		return 0;
2691 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2692 		!(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2693 		return 0;
2694 	return 1;
2695 }
2696 
2697 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2698 {
2699 	int i;
2700 	u32 function, index;
2701 	struct kvm_cpuid_entry2 *e, *best;
2702 
2703 	kvm_x86_ops->cache_regs(vcpu);
2704 	function = vcpu->arch.regs[VCPU_REGS_RAX];
2705 	index = vcpu->arch.regs[VCPU_REGS_RCX];
2706 	vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2707 	vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2708 	vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2709 	vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2710 	best = NULL;
2711 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2712 		e = &vcpu->arch.cpuid_entries[i];
2713 		if (is_matching_cpuid_entry(e, function, index)) {
2714 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2715 				move_to_next_stateful_cpuid_entry(vcpu, i);
2716 			best = e;
2717 			break;
2718 		}
2719 		/*
2720 		 * Both basic or both extended?
2721 		 */
2722 		if (((e->function ^ function) & 0x80000000) == 0)
2723 			if (!best || e->function > best->function)
2724 				best = e;
2725 	}
2726 	if (best) {
2727 		vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2728 		vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2729 		vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2730 		vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2731 	}
2732 	kvm_x86_ops->decache_regs(vcpu);
2733 	kvm_x86_ops->skip_emulated_instruction(vcpu);
2734 	KVMTRACE_5D(CPUID, vcpu, function,
2735 		    (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2736 		    (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2737 		    (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2738 		    (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2739 }
2740 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2741 
2742 /*
2743  * Check if userspace requested an interrupt window, and that the
2744  * interrupt window is open.
2745  *
2746  * No need to exit to userspace if we already have an interrupt queued.
2747  */
2748 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2749 					  struct kvm_run *kvm_run)
2750 {
2751 	return (!vcpu->arch.irq_summary &&
2752 		kvm_run->request_interrupt_window &&
2753 		vcpu->arch.interrupt_window_open &&
2754 		(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2755 }
2756 
2757 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2758 			      struct kvm_run *kvm_run)
2759 {
2760 	kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2761 	kvm_run->cr8 = kvm_get_cr8(vcpu);
2762 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
2763 	if (irqchip_in_kernel(vcpu->kvm))
2764 		kvm_run->ready_for_interrupt_injection = 1;
2765 	else
2766 		kvm_run->ready_for_interrupt_injection =
2767 					(vcpu->arch.interrupt_window_open &&
2768 					 vcpu->arch.irq_summary == 0);
2769 }
2770 
2771 static void vapic_enter(struct kvm_vcpu *vcpu)
2772 {
2773 	struct kvm_lapic *apic = vcpu->arch.apic;
2774 	struct page *page;
2775 
2776 	if (!apic || !apic->vapic_addr)
2777 		return;
2778 
2779 	down_read(&current->mm->mmap_sem);
2780 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2781 	up_read(&current->mm->mmap_sem);
2782 
2783 	vcpu->arch.apic->vapic_page = page;
2784 }
2785 
2786 static void vapic_exit(struct kvm_vcpu *vcpu)
2787 {
2788 	struct kvm_lapic *apic = vcpu->arch.apic;
2789 
2790 	if (!apic || !apic->vapic_addr)
2791 		return;
2792 
2793 	down_read(&vcpu->kvm->slots_lock);
2794 	kvm_release_page_dirty(apic->vapic_page);
2795 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2796 	up_read(&vcpu->kvm->slots_lock);
2797 }
2798 
2799 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2800 {
2801 	int r;
2802 
2803 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2804 		pr_debug("vcpu %d received sipi with vector # %x\n",
2805 		       vcpu->vcpu_id, vcpu->arch.sipi_vector);
2806 		kvm_lapic_reset(vcpu);
2807 		r = kvm_x86_ops->vcpu_reset(vcpu);
2808 		if (r)
2809 			return r;
2810 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2811 	}
2812 
2813 	down_read(&vcpu->kvm->slots_lock);
2814 	vapic_enter(vcpu);
2815 
2816 preempted:
2817 	if (vcpu->guest_debug.enabled)
2818 		kvm_x86_ops->guest_debug_pre(vcpu);
2819 
2820 again:
2821 	if (vcpu->requests)
2822 		if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2823 			kvm_mmu_unload(vcpu);
2824 
2825 	r = kvm_mmu_reload(vcpu);
2826 	if (unlikely(r))
2827 		goto out;
2828 
2829 	if (vcpu->requests) {
2830 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2831 			__kvm_migrate_timers(vcpu);
2832 		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2833 			kvm_x86_ops->tlb_flush(vcpu);
2834 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2835 				       &vcpu->requests)) {
2836 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2837 			r = 0;
2838 			goto out;
2839 		}
2840 		if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2841 			kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2842 			r = 0;
2843 			goto out;
2844 		}
2845 	}
2846 
2847 	clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2848 	kvm_inject_pending_timer_irqs(vcpu);
2849 
2850 	preempt_disable();
2851 
2852 	kvm_x86_ops->prepare_guest_switch(vcpu);
2853 	kvm_load_guest_fpu(vcpu);
2854 
2855 	local_irq_disable();
2856 
2857 	if (vcpu->requests || need_resched()) {
2858 		local_irq_enable();
2859 		preempt_enable();
2860 		r = 1;
2861 		goto out;
2862 	}
2863 
2864 	if (signal_pending(current)) {
2865 		local_irq_enable();
2866 		preempt_enable();
2867 		r = -EINTR;
2868 		kvm_run->exit_reason = KVM_EXIT_INTR;
2869 		++vcpu->stat.signal_exits;
2870 		goto out;
2871 	}
2872 
2873 	vcpu->guest_mode = 1;
2874 	/*
2875 	 * Make sure that guest_mode assignment won't happen after
2876 	 * testing the pending IRQ vector bitmap.
2877 	 */
2878 	smp_wmb();
2879 
2880 	if (vcpu->arch.exception.pending)
2881 		__queue_exception(vcpu);
2882 	else if (irqchip_in_kernel(vcpu->kvm))
2883 		kvm_x86_ops->inject_pending_irq(vcpu);
2884 	else
2885 		kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2886 
2887 	kvm_lapic_sync_to_vapic(vcpu);
2888 
2889 	up_read(&vcpu->kvm->slots_lock);
2890 
2891 	kvm_guest_enter();
2892 
2893 
2894 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2895 	kvm_x86_ops->run(vcpu, kvm_run);
2896 
2897 	vcpu->guest_mode = 0;
2898 	local_irq_enable();
2899 
2900 	++vcpu->stat.exits;
2901 
2902 	/*
2903 	 * We must have an instruction between local_irq_enable() and
2904 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2905 	 * the interrupt shadow.  The stat.exits increment will do nicely.
2906 	 * But we need to prevent reordering, hence this barrier():
2907 	 */
2908 	barrier();
2909 
2910 	kvm_guest_exit();
2911 
2912 	preempt_enable();
2913 
2914 	down_read(&vcpu->kvm->slots_lock);
2915 
2916 	/*
2917 	 * Profile KVM exit RIPs:
2918 	 */
2919 	if (unlikely(prof_on == KVM_PROFILING)) {
2920 		kvm_x86_ops->cache_regs(vcpu);
2921 		profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2922 	}
2923 
2924 	if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2925 		vcpu->arch.exception.pending = false;
2926 
2927 	kvm_lapic_sync_from_vapic(vcpu);
2928 
2929 	r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2930 
2931 	if (r > 0) {
2932 		if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2933 			r = -EINTR;
2934 			kvm_run->exit_reason = KVM_EXIT_INTR;
2935 			++vcpu->stat.request_irq_exits;
2936 			goto out;
2937 		}
2938 		if (!need_resched())
2939 			goto again;
2940 	}
2941 
2942 out:
2943 	up_read(&vcpu->kvm->slots_lock);
2944 	if (r > 0) {
2945 		kvm_resched(vcpu);
2946 		down_read(&vcpu->kvm->slots_lock);
2947 		goto preempted;
2948 	}
2949 
2950 	post_kvm_run_save(vcpu, kvm_run);
2951 
2952 	vapic_exit(vcpu);
2953 
2954 	return r;
2955 }
2956 
2957 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2958 {
2959 	int r;
2960 	sigset_t sigsaved;
2961 
2962 	vcpu_load(vcpu);
2963 
2964 	if (vcpu->sigset_active)
2965 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2966 
2967 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2968 		kvm_vcpu_block(vcpu);
2969 		r = -EAGAIN;
2970 		goto out;
2971 	}
2972 
2973 	/* re-sync apic's tpr */
2974 	if (!irqchip_in_kernel(vcpu->kvm))
2975 		kvm_set_cr8(vcpu, kvm_run->cr8);
2976 
2977 	if (vcpu->arch.pio.cur_count) {
2978 		r = complete_pio(vcpu);
2979 		if (r)
2980 			goto out;
2981 	}
2982 #if CONFIG_HAS_IOMEM
2983 	if (vcpu->mmio_needed) {
2984 		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2985 		vcpu->mmio_read_completed = 1;
2986 		vcpu->mmio_needed = 0;
2987 
2988 		down_read(&vcpu->kvm->slots_lock);
2989 		r = emulate_instruction(vcpu, kvm_run,
2990 					vcpu->arch.mmio_fault_cr2, 0,
2991 					EMULTYPE_NO_DECODE);
2992 		up_read(&vcpu->kvm->slots_lock);
2993 		if (r == EMULATE_DO_MMIO) {
2994 			/*
2995 			 * Read-modify-write.  Back to userspace.
2996 			 */
2997 			r = 0;
2998 			goto out;
2999 		}
3000 	}
3001 #endif
3002 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
3003 		kvm_x86_ops->cache_regs(vcpu);
3004 		vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
3005 		kvm_x86_ops->decache_regs(vcpu);
3006 	}
3007 
3008 	r = __vcpu_run(vcpu, kvm_run);
3009 
3010 out:
3011 	if (vcpu->sigset_active)
3012 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3013 
3014 	vcpu_put(vcpu);
3015 	return r;
3016 }
3017 
3018 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3019 {
3020 	vcpu_load(vcpu);
3021 
3022 	kvm_x86_ops->cache_regs(vcpu);
3023 
3024 	regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
3025 	regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
3026 	regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
3027 	regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
3028 	regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
3029 	regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
3030 	regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3031 	regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3032 #ifdef CONFIG_X86_64
3033 	regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3034 	regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3035 	regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3036 	regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3037 	regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3038 	regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3039 	regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3040 	regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3041 #endif
3042 
3043 	regs->rip = vcpu->arch.rip;
3044 	regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3045 
3046 	/*
3047 	 * Don't leak debug flags in case they were set for guest debugging
3048 	 */
3049 	if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3050 		regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3051 
3052 	vcpu_put(vcpu);
3053 
3054 	return 0;
3055 }
3056 
3057 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3058 {
3059 	vcpu_load(vcpu);
3060 
3061 	vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3062 	vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3063 	vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3064 	vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3065 	vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3066 	vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3067 	vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3068 	vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3069 #ifdef CONFIG_X86_64
3070 	vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3071 	vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3072 	vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3073 	vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3074 	vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3075 	vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3076 	vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3077 	vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3078 #endif
3079 
3080 	vcpu->arch.rip = regs->rip;
3081 	kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3082 
3083 	kvm_x86_ops->decache_regs(vcpu);
3084 
3085 	vcpu->arch.exception.pending = false;
3086 
3087 	vcpu_put(vcpu);
3088 
3089 	return 0;
3090 }
3091 
3092 void kvm_get_segment(struct kvm_vcpu *vcpu,
3093 		     struct kvm_segment *var, int seg)
3094 {
3095 	kvm_x86_ops->get_segment(vcpu, var, seg);
3096 }
3097 
3098 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3099 {
3100 	struct kvm_segment cs;
3101 
3102 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3103 	*db = cs.db;
3104 	*l = cs.l;
3105 }
3106 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3107 
3108 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3109 				  struct kvm_sregs *sregs)
3110 {
3111 	struct descriptor_table dt;
3112 	int pending_vec;
3113 
3114 	vcpu_load(vcpu);
3115 
3116 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3117 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3118 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3119 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3120 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3121 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3122 
3123 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3124 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3125 
3126 	kvm_x86_ops->get_idt(vcpu, &dt);
3127 	sregs->idt.limit = dt.limit;
3128 	sregs->idt.base = dt.base;
3129 	kvm_x86_ops->get_gdt(vcpu, &dt);
3130 	sregs->gdt.limit = dt.limit;
3131 	sregs->gdt.base = dt.base;
3132 
3133 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3134 	sregs->cr0 = vcpu->arch.cr0;
3135 	sregs->cr2 = vcpu->arch.cr2;
3136 	sregs->cr3 = vcpu->arch.cr3;
3137 	sregs->cr4 = vcpu->arch.cr4;
3138 	sregs->cr8 = kvm_get_cr8(vcpu);
3139 	sregs->efer = vcpu->arch.shadow_efer;
3140 	sregs->apic_base = kvm_get_apic_base(vcpu);
3141 
3142 	if (irqchip_in_kernel(vcpu->kvm)) {
3143 		memset(sregs->interrupt_bitmap, 0,
3144 		       sizeof sregs->interrupt_bitmap);
3145 		pending_vec = kvm_x86_ops->get_irq(vcpu);
3146 		if (pending_vec >= 0)
3147 			set_bit(pending_vec,
3148 				(unsigned long *)sregs->interrupt_bitmap);
3149 	} else
3150 		memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3151 		       sizeof sregs->interrupt_bitmap);
3152 
3153 	vcpu_put(vcpu);
3154 
3155 	return 0;
3156 }
3157 
3158 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3159 				    struct kvm_mp_state *mp_state)
3160 {
3161 	vcpu_load(vcpu);
3162 	mp_state->mp_state = vcpu->arch.mp_state;
3163 	vcpu_put(vcpu);
3164 	return 0;
3165 }
3166 
3167 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3168 				    struct kvm_mp_state *mp_state)
3169 {
3170 	vcpu_load(vcpu);
3171 	vcpu->arch.mp_state = mp_state->mp_state;
3172 	vcpu_put(vcpu);
3173 	return 0;
3174 }
3175 
3176 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3177 			struct kvm_segment *var, int seg)
3178 {
3179 	kvm_x86_ops->set_segment(vcpu, var, seg);
3180 }
3181 
3182 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3183 				   struct kvm_segment *kvm_desct)
3184 {
3185 	kvm_desct->base = seg_desc->base0;
3186 	kvm_desct->base |= seg_desc->base1 << 16;
3187 	kvm_desct->base |= seg_desc->base2 << 24;
3188 	kvm_desct->limit = seg_desc->limit0;
3189 	kvm_desct->limit |= seg_desc->limit << 16;
3190 	if (seg_desc->g) {
3191 		kvm_desct->limit <<= 12;
3192 		kvm_desct->limit |= 0xfff;
3193 	}
3194 	kvm_desct->selector = selector;
3195 	kvm_desct->type = seg_desc->type;
3196 	kvm_desct->present = seg_desc->p;
3197 	kvm_desct->dpl = seg_desc->dpl;
3198 	kvm_desct->db = seg_desc->d;
3199 	kvm_desct->s = seg_desc->s;
3200 	kvm_desct->l = seg_desc->l;
3201 	kvm_desct->g = seg_desc->g;
3202 	kvm_desct->avl = seg_desc->avl;
3203 	if (!selector)
3204 		kvm_desct->unusable = 1;
3205 	else
3206 		kvm_desct->unusable = 0;
3207 	kvm_desct->padding = 0;
3208 }
3209 
3210 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3211 					   u16 selector,
3212 					   struct descriptor_table *dtable)
3213 {
3214 	if (selector & 1 << 2) {
3215 		struct kvm_segment kvm_seg;
3216 
3217 		kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3218 
3219 		if (kvm_seg.unusable)
3220 			dtable->limit = 0;
3221 		else
3222 			dtable->limit = kvm_seg.limit;
3223 		dtable->base = kvm_seg.base;
3224 	}
3225 	else
3226 		kvm_x86_ops->get_gdt(vcpu, dtable);
3227 }
3228 
3229 /* allowed just for 8 bytes segments */
3230 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3231 					 struct desc_struct *seg_desc)
3232 {
3233 	gpa_t gpa;
3234 	struct descriptor_table dtable;
3235 	u16 index = selector >> 3;
3236 
3237 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3238 
3239 	if (dtable.limit < index * 8 + 7) {
3240 		kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3241 		return 1;
3242 	}
3243 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3244 	gpa += index * 8;
3245 	return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3246 }
3247 
3248 /* allowed just for 8 bytes segments */
3249 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3250 					 struct desc_struct *seg_desc)
3251 {
3252 	gpa_t gpa;
3253 	struct descriptor_table dtable;
3254 	u16 index = selector >> 3;
3255 
3256 	get_segment_descritptor_dtable(vcpu, selector, &dtable);
3257 
3258 	if (dtable.limit < index * 8 + 7)
3259 		return 1;
3260 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3261 	gpa += index * 8;
3262 	return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3263 }
3264 
3265 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3266 			     struct desc_struct *seg_desc)
3267 {
3268 	u32 base_addr;
3269 
3270 	base_addr = seg_desc->base0;
3271 	base_addr |= (seg_desc->base1 << 16);
3272 	base_addr |= (seg_desc->base2 << 24);
3273 
3274 	return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3275 }
3276 
3277 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3278 {
3279 	struct kvm_segment kvm_seg;
3280 
3281 	kvm_get_segment(vcpu, &kvm_seg, seg);
3282 	return kvm_seg.selector;
3283 }
3284 
3285 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3286 						u16 selector,
3287 						struct kvm_segment *kvm_seg)
3288 {
3289 	struct desc_struct seg_desc;
3290 
3291 	if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3292 		return 1;
3293 	seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3294 	return 0;
3295 }
3296 
3297 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3298 				int type_bits, int seg)
3299 {
3300 	struct kvm_segment kvm_seg;
3301 
3302 	if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3303 		return 1;
3304 	kvm_seg.type |= type_bits;
3305 
3306 	if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3307 	    seg != VCPU_SREG_LDTR)
3308 		if (!kvm_seg.s)
3309 			kvm_seg.unusable = 1;
3310 
3311 	kvm_set_segment(vcpu, &kvm_seg, seg);
3312 	return 0;
3313 }
3314 
3315 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3316 				struct tss_segment_32 *tss)
3317 {
3318 	tss->cr3 = vcpu->arch.cr3;
3319 	tss->eip = vcpu->arch.rip;
3320 	tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3321 	tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3322 	tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3323 	tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3324 	tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3325 	tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3326 	tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3327 	tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3328 	tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3329 
3330 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3331 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3332 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3333 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3334 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3335 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3336 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3337 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3338 }
3339 
3340 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3341 				  struct tss_segment_32 *tss)
3342 {
3343 	kvm_set_cr3(vcpu, tss->cr3);
3344 
3345 	vcpu->arch.rip = tss->eip;
3346 	kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3347 
3348 	vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3349 	vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3350 	vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3351 	vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3352 	vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3353 	vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3354 	vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3355 	vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3356 
3357 	if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3358 		return 1;
3359 
3360 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3361 		return 1;
3362 
3363 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3364 		return 1;
3365 
3366 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3367 		return 1;
3368 
3369 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3370 		return 1;
3371 
3372 	if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3373 		return 1;
3374 
3375 	if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3376 		return 1;
3377 	return 0;
3378 }
3379 
3380 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3381 				struct tss_segment_16 *tss)
3382 {
3383 	tss->ip = vcpu->arch.rip;
3384 	tss->flag = kvm_x86_ops->get_rflags(vcpu);
3385 	tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3386 	tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3387 	tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3388 	tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3389 	tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3390 	tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3391 	tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3392 	tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3393 
3394 	tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3395 	tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3396 	tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3397 	tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3398 	tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3399 	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3400 }
3401 
3402 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3403 				 struct tss_segment_16 *tss)
3404 {
3405 	vcpu->arch.rip = tss->ip;
3406 	kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3407 	vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3408 	vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3409 	vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3410 	vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3411 	vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3412 	vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3413 	vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3414 	vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3415 
3416 	if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3417 		return 1;
3418 
3419 	if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3420 		return 1;
3421 
3422 	if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3423 		return 1;
3424 
3425 	if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3426 		return 1;
3427 
3428 	if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3429 		return 1;
3430 	return 0;
3431 }
3432 
3433 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3434 		       u32 old_tss_base,
3435 		       struct desc_struct *nseg_desc)
3436 {
3437 	struct tss_segment_16 tss_segment_16;
3438 	int ret = 0;
3439 
3440 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3441 			   sizeof tss_segment_16))
3442 		goto out;
3443 
3444 	save_state_to_tss16(vcpu, &tss_segment_16);
3445 
3446 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3447 			    sizeof tss_segment_16))
3448 		goto out;
3449 
3450 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3451 			   &tss_segment_16, sizeof tss_segment_16))
3452 		goto out;
3453 
3454 	if (load_state_from_tss16(vcpu, &tss_segment_16))
3455 		goto out;
3456 
3457 	ret = 1;
3458 out:
3459 	return ret;
3460 }
3461 
3462 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3463 		       u32 old_tss_base,
3464 		       struct desc_struct *nseg_desc)
3465 {
3466 	struct tss_segment_32 tss_segment_32;
3467 	int ret = 0;
3468 
3469 	if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3470 			   sizeof tss_segment_32))
3471 		goto out;
3472 
3473 	save_state_to_tss32(vcpu, &tss_segment_32);
3474 
3475 	if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3476 			    sizeof tss_segment_32))
3477 		goto out;
3478 
3479 	if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3480 			   &tss_segment_32, sizeof tss_segment_32))
3481 		goto out;
3482 
3483 	if (load_state_from_tss32(vcpu, &tss_segment_32))
3484 		goto out;
3485 
3486 	ret = 1;
3487 out:
3488 	return ret;
3489 }
3490 
3491 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3492 {
3493 	struct kvm_segment tr_seg;
3494 	struct desc_struct cseg_desc;
3495 	struct desc_struct nseg_desc;
3496 	int ret = 0;
3497 	u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3498 	u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3499 
3500 	old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3501 
3502 	/* FIXME: Handle errors. Failure to read either TSS or their
3503 	 * descriptors should generate a pagefault.
3504 	 */
3505 	if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3506 		goto out;
3507 
3508 	if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3509 		goto out;
3510 
3511 	if (reason != TASK_SWITCH_IRET) {
3512 		int cpl;
3513 
3514 		cpl = kvm_x86_ops->get_cpl(vcpu);
3515 		if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3516 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3517 			return 1;
3518 		}
3519 	}
3520 
3521 	if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3522 		kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3523 		return 1;
3524 	}
3525 
3526 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3527 		cseg_desc.type &= ~(1 << 1); //clear the B flag
3528 		save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3529 	}
3530 
3531 	if (reason == TASK_SWITCH_IRET) {
3532 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3533 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3534 	}
3535 
3536 	kvm_x86_ops->skip_emulated_instruction(vcpu);
3537 	kvm_x86_ops->cache_regs(vcpu);
3538 
3539 	if (nseg_desc.type & 8)
3540 		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3541 					 &nseg_desc);
3542 	else
3543 		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3544 					 &nseg_desc);
3545 
3546 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3547 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3548 		kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3549 	}
3550 
3551 	if (reason != TASK_SWITCH_IRET) {
3552 		nseg_desc.type |= (1 << 1);
3553 		save_guest_segment_descriptor(vcpu, tss_selector,
3554 					      &nseg_desc);
3555 	}
3556 
3557 	kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3558 	seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3559 	tr_seg.type = 11;
3560 	kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3561 out:
3562 	kvm_x86_ops->decache_regs(vcpu);
3563 	return ret;
3564 }
3565 EXPORT_SYMBOL_GPL(kvm_task_switch);
3566 
3567 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3568 				  struct kvm_sregs *sregs)
3569 {
3570 	int mmu_reset_needed = 0;
3571 	int i, pending_vec, max_bits;
3572 	struct descriptor_table dt;
3573 
3574 	vcpu_load(vcpu);
3575 
3576 	dt.limit = sregs->idt.limit;
3577 	dt.base = sregs->idt.base;
3578 	kvm_x86_ops->set_idt(vcpu, &dt);
3579 	dt.limit = sregs->gdt.limit;
3580 	dt.base = sregs->gdt.base;
3581 	kvm_x86_ops->set_gdt(vcpu, &dt);
3582 
3583 	vcpu->arch.cr2 = sregs->cr2;
3584 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3585 	vcpu->arch.cr3 = sregs->cr3;
3586 
3587 	kvm_set_cr8(vcpu, sregs->cr8);
3588 
3589 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3590 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
3591 	kvm_set_apic_base(vcpu, sregs->apic_base);
3592 
3593 	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3594 
3595 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3596 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3597 	vcpu->arch.cr0 = sregs->cr0;
3598 
3599 	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3600 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3601 	if (!is_long_mode(vcpu) && is_pae(vcpu))
3602 		load_pdptrs(vcpu, vcpu->arch.cr3);
3603 
3604 	if (mmu_reset_needed)
3605 		kvm_mmu_reset_context(vcpu);
3606 
3607 	if (!irqchip_in_kernel(vcpu->kvm)) {
3608 		memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3609 		       sizeof vcpu->arch.irq_pending);
3610 		vcpu->arch.irq_summary = 0;
3611 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3612 			if (vcpu->arch.irq_pending[i])
3613 				__set_bit(i, &vcpu->arch.irq_summary);
3614 	} else {
3615 		max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3616 		pending_vec = find_first_bit(
3617 			(const unsigned long *)sregs->interrupt_bitmap,
3618 			max_bits);
3619 		/* Only pending external irq is handled here */
3620 		if (pending_vec < max_bits) {
3621 			kvm_x86_ops->set_irq(vcpu, pending_vec);
3622 			pr_debug("Set back pending irq %d\n",
3623 				 pending_vec);
3624 		}
3625 	}
3626 
3627 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3628 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3629 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3630 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3631 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3632 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3633 
3634 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3635 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3636 
3637 	vcpu_put(vcpu);
3638 
3639 	return 0;
3640 }
3641 
3642 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3643 				    struct kvm_debug_guest *dbg)
3644 {
3645 	int r;
3646 
3647 	vcpu_load(vcpu);
3648 
3649 	r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3650 
3651 	vcpu_put(vcpu);
3652 
3653 	return r;
3654 }
3655 
3656 /*
3657  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3658  * we have asm/x86/processor.h
3659  */
3660 struct fxsave {
3661 	u16	cwd;
3662 	u16	swd;
3663 	u16	twd;
3664 	u16	fop;
3665 	u64	rip;
3666 	u64	rdp;
3667 	u32	mxcsr;
3668 	u32	mxcsr_mask;
3669 	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
3670 #ifdef CONFIG_X86_64
3671 	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
3672 #else
3673 	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
3674 #endif
3675 };
3676 
3677 /*
3678  * Translate a guest virtual address to a guest physical address.
3679  */
3680 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3681 				    struct kvm_translation *tr)
3682 {
3683 	unsigned long vaddr = tr->linear_address;
3684 	gpa_t gpa;
3685 
3686 	vcpu_load(vcpu);
3687 	down_read(&vcpu->kvm->slots_lock);
3688 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3689 	up_read(&vcpu->kvm->slots_lock);
3690 	tr->physical_address = gpa;
3691 	tr->valid = gpa != UNMAPPED_GVA;
3692 	tr->writeable = 1;
3693 	tr->usermode = 0;
3694 	vcpu_put(vcpu);
3695 
3696 	return 0;
3697 }
3698 
3699 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3700 {
3701 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3702 
3703 	vcpu_load(vcpu);
3704 
3705 	memcpy(fpu->fpr, fxsave->st_space, 128);
3706 	fpu->fcw = fxsave->cwd;
3707 	fpu->fsw = fxsave->swd;
3708 	fpu->ftwx = fxsave->twd;
3709 	fpu->last_opcode = fxsave->fop;
3710 	fpu->last_ip = fxsave->rip;
3711 	fpu->last_dp = fxsave->rdp;
3712 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3713 
3714 	vcpu_put(vcpu);
3715 
3716 	return 0;
3717 }
3718 
3719 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3720 {
3721 	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3722 
3723 	vcpu_load(vcpu);
3724 
3725 	memcpy(fxsave->st_space, fpu->fpr, 128);
3726 	fxsave->cwd = fpu->fcw;
3727 	fxsave->swd = fpu->fsw;
3728 	fxsave->twd = fpu->ftwx;
3729 	fxsave->fop = fpu->last_opcode;
3730 	fxsave->rip = fpu->last_ip;
3731 	fxsave->rdp = fpu->last_dp;
3732 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3733 
3734 	vcpu_put(vcpu);
3735 
3736 	return 0;
3737 }
3738 
3739 void fx_init(struct kvm_vcpu *vcpu)
3740 {
3741 	unsigned after_mxcsr_mask;
3742 
3743 	/*
3744 	 * Touch the fpu the first time in non atomic context as if
3745 	 * this is the first fpu instruction the exception handler
3746 	 * will fire before the instruction returns and it'll have to
3747 	 * allocate ram with GFP_KERNEL.
3748 	 */
3749 	if (!used_math())
3750 		kvm_fx_save(&vcpu->arch.host_fx_image);
3751 
3752 	/* Initialize guest FPU by resetting ours and saving into guest's */
3753 	preempt_disable();
3754 	kvm_fx_save(&vcpu->arch.host_fx_image);
3755 	kvm_fx_finit();
3756 	kvm_fx_save(&vcpu->arch.guest_fx_image);
3757 	kvm_fx_restore(&vcpu->arch.host_fx_image);
3758 	preempt_enable();
3759 
3760 	vcpu->arch.cr0 |= X86_CR0_ET;
3761 	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3762 	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3763 	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3764 	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3765 }
3766 EXPORT_SYMBOL_GPL(fx_init);
3767 
3768 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3769 {
3770 	if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3771 		return;
3772 
3773 	vcpu->guest_fpu_loaded = 1;
3774 	kvm_fx_save(&vcpu->arch.host_fx_image);
3775 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
3776 }
3777 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3778 
3779 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3780 {
3781 	if (!vcpu->guest_fpu_loaded)
3782 		return;
3783 
3784 	vcpu->guest_fpu_loaded = 0;
3785 	kvm_fx_save(&vcpu->arch.guest_fx_image);
3786 	kvm_fx_restore(&vcpu->arch.host_fx_image);
3787 	++vcpu->stat.fpu_reload;
3788 }
3789 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3790 
3791 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3792 {
3793 	kvm_x86_ops->vcpu_free(vcpu);
3794 }
3795 
3796 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3797 						unsigned int id)
3798 {
3799 	return kvm_x86_ops->vcpu_create(kvm, id);
3800 }
3801 
3802 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3803 {
3804 	int r;
3805 
3806 	/* We do fxsave: this must be aligned. */
3807 	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3808 
3809 	vcpu_load(vcpu);
3810 	r = kvm_arch_vcpu_reset(vcpu);
3811 	if (r == 0)
3812 		r = kvm_mmu_setup(vcpu);
3813 	vcpu_put(vcpu);
3814 	if (r < 0)
3815 		goto free_vcpu;
3816 
3817 	return 0;
3818 free_vcpu:
3819 	kvm_x86_ops->vcpu_free(vcpu);
3820 	return r;
3821 }
3822 
3823 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3824 {
3825 	vcpu_load(vcpu);
3826 	kvm_mmu_unload(vcpu);
3827 	vcpu_put(vcpu);
3828 
3829 	kvm_x86_ops->vcpu_free(vcpu);
3830 }
3831 
3832 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3833 {
3834 	return kvm_x86_ops->vcpu_reset(vcpu);
3835 }
3836 
3837 void kvm_arch_hardware_enable(void *garbage)
3838 {
3839 	kvm_x86_ops->hardware_enable(garbage);
3840 }
3841 
3842 void kvm_arch_hardware_disable(void *garbage)
3843 {
3844 	kvm_x86_ops->hardware_disable(garbage);
3845 }
3846 
3847 int kvm_arch_hardware_setup(void)
3848 {
3849 	return kvm_x86_ops->hardware_setup();
3850 }
3851 
3852 void kvm_arch_hardware_unsetup(void)
3853 {
3854 	kvm_x86_ops->hardware_unsetup();
3855 }
3856 
3857 void kvm_arch_check_processor_compat(void *rtn)
3858 {
3859 	kvm_x86_ops->check_processor_compatibility(rtn);
3860 }
3861 
3862 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3863 {
3864 	struct page *page;
3865 	struct kvm *kvm;
3866 	int r;
3867 
3868 	BUG_ON(vcpu->kvm == NULL);
3869 	kvm = vcpu->kvm;
3870 
3871 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3872 	if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3873 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3874 	else
3875 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3876 
3877 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3878 	if (!page) {
3879 		r = -ENOMEM;
3880 		goto fail;
3881 	}
3882 	vcpu->arch.pio_data = page_address(page);
3883 
3884 	r = kvm_mmu_create(vcpu);
3885 	if (r < 0)
3886 		goto fail_free_pio_data;
3887 
3888 	if (irqchip_in_kernel(kvm)) {
3889 		r = kvm_create_lapic(vcpu);
3890 		if (r < 0)
3891 			goto fail_mmu_destroy;
3892 	}
3893 
3894 	return 0;
3895 
3896 fail_mmu_destroy:
3897 	kvm_mmu_destroy(vcpu);
3898 fail_free_pio_data:
3899 	free_page((unsigned long)vcpu->arch.pio_data);
3900 fail:
3901 	return r;
3902 }
3903 
3904 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3905 {
3906 	kvm_free_lapic(vcpu);
3907 	down_read(&vcpu->kvm->slots_lock);
3908 	kvm_mmu_destroy(vcpu);
3909 	up_read(&vcpu->kvm->slots_lock);
3910 	free_page((unsigned long)vcpu->arch.pio_data);
3911 }
3912 
3913 struct  kvm *kvm_arch_create_vm(void)
3914 {
3915 	struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3916 
3917 	if (!kvm)
3918 		return ERR_PTR(-ENOMEM);
3919 
3920 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3921 
3922 	return kvm;
3923 }
3924 
3925 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3926 {
3927 	vcpu_load(vcpu);
3928 	kvm_mmu_unload(vcpu);
3929 	vcpu_put(vcpu);
3930 }
3931 
3932 static void kvm_free_vcpus(struct kvm *kvm)
3933 {
3934 	unsigned int i;
3935 
3936 	/*
3937 	 * Unpin any mmu pages first.
3938 	 */
3939 	for (i = 0; i < KVM_MAX_VCPUS; ++i)
3940 		if (kvm->vcpus[i])
3941 			kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3942 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3943 		if (kvm->vcpus[i]) {
3944 			kvm_arch_vcpu_free(kvm->vcpus[i]);
3945 			kvm->vcpus[i] = NULL;
3946 		}
3947 	}
3948 
3949 }
3950 
3951 void kvm_arch_destroy_vm(struct kvm *kvm)
3952 {
3953 	kvm_free_pit(kvm);
3954 	kfree(kvm->arch.vpic);
3955 	kfree(kvm->arch.vioapic);
3956 	kvm_free_vcpus(kvm);
3957 	kvm_free_physmem(kvm);
3958 	if (kvm->arch.apic_access_page)
3959 		put_page(kvm->arch.apic_access_page);
3960 	if (kvm->arch.ept_identity_pagetable)
3961 		put_page(kvm->arch.ept_identity_pagetable);
3962 	kfree(kvm);
3963 }
3964 
3965 int kvm_arch_set_memory_region(struct kvm *kvm,
3966 				struct kvm_userspace_memory_region *mem,
3967 				struct kvm_memory_slot old,
3968 				int user_alloc)
3969 {
3970 	int npages = mem->memory_size >> PAGE_SHIFT;
3971 	struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3972 
3973 	/*To keep backward compatibility with older userspace,
3974 	 *x86 needs to hanlde !user_alloc case.
3975 	 */
3976 	if (!user_alloc) {
3977 		if (npages && !old.rmap) {
3978 			unsigned long userspace_addr;
3979 
3980 			down_write(&current->mm->mmap_sem);
3981 			userspace_addr = do_mmap(NULL, 0,
3982 						 npages * PAGE_SIZE,
3983 						 PROT_READ | PROT_WRITE,
3984 						 MAP_SHARED | MAP_ANONYMOUS,
3985 						 0);
3986 			up_write(&current->mm->mmap_sem);
3987 
3988 			if (IS_ERR((void *)userspace_addr))
3989 				return PTR_ERR((void *)userspace_addr);
3990 
3991 			/* set userspace_addr atomically for kvm_hva_to_rmapp */
3992 			spin_lock(&kvm->mmu_lock);
3993 			memslot->userspace_addr = userspace_addr;
3994 			spin_unlock(&kvm->mmu_lock);
3995 		} else {
3996 			if (!old.user_alloc && old.rmap) {
3997 				int ret;
3998 
3999 				down_write(&current->mm->mmap_sem);
4000 				ret = do_munmap(current->mm, old.userspace_addr,
4001 						old.npages * PAGE_SIZE);
4002 				up_write(&current->mm->mmap_sem);
4003 				if (ret < 0)
4004 					printk(KERN_WARNING
4005 				       "kvm_vm_ioctl_set_memory_region: "
4006 				       "failed to munmap memory\n");
4007 			}
4008 		}
4009 	}
4010 
4011 	if (!kvm->arch.n_requested_mmu_pages) {
4012 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4013 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4014 	}
4015 
4016 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4017 	kvm_flush_remote_tlbs(kvm);
4018 
4019 	return 0;
4020 }
4021 
4022 void kvm_arch_flush_shadow(struct kvm *kvm)
4023 {
4024 	kvm_mmu_zap_all(kvm);
4025 }
4026 
4027 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4028 {
4029 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4030 	       || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4031 }
4032 
4033 static void vcpu_kick_intr(void *info)
4034 {
4035 #ifdef DEBUG
4036 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4037 	printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4038 #endif
4039 }
4040 
4041 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4042 {
4043 	int ipi_pcpu = vcpu->cpu;
4044 	int cpu = get_cpu();
4045 
4046 	if (waitqueue_active(&vcpu->wq)) {
4047 		wake_up_interruptible(&vcpu->wq);
4048 		++vcpu->stat.halt_wakeup;
4049 	}
4050 	/*
4051 	 * We may be called synchronously with irqs disabled in guest mode,
4052 	 * So need not to call smp_call_function_single() in that case.
4053 	 */
4054 	if (vcpu->guest_mode && vcpu->cpu != cpu)
4055 		smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4056 	put_cpu();
4057 }
4058