?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ACPI: thinkpad-acpi: keep track of module state
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / kvm / kvm_main.c
blobb58fdf37e61776990b8ca333e5b806f8d4c637b0
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "kvm.h"
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
21
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
27 #include <linux/mm.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
40
41 #include <asm/processor.h>
42 #include <asm/msr.h>
43 #include <asm/io.h>
44 #include <asm/uaccess.h>
45 #include <asm/desc.h>
46
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
49
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
52
53 static cpumask_t cpus_hardware_enabled;
54
55 struct kvm_arch_ops *kvm_arch_ops;
56
57 static void hardware_disable(void *ignored);
58
59 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
60
61 static struct kvm_stats_debugfs_item {
62 const char *name;
63 int offset;
64 struct dentry *dentry;
65 } debugfs_entries[] = {
66 { "pf_fixed", STAT_OFFSET(pf_fixed) },
67 { "pf_guest", STAT_OFFSET(pf_guest) },
68 { "tlb_flush", STAT_OFFSET(tlb_flush) },
69 { "invlpg", STAT_OFFSET(invlpg) },
70 { "exits", STAT_OFFSET(exits) },
71 { "io_exits", STAT_OFFSET(io_exits) },
72 { "mmio_exits", STAT_OFFSET(mmio_exits) },
73 { "signal_exits", STAT_OFFSET(signal_exits) },
74 { "irq_window", STAT_OFFSET(irq_window_exits) },
75 { "halt_exits", STAT_OFFSET(halt_exits) },
76 { "request_irq", STAT_OFFSET(request_irq_exits) },
77 { "irq_exits", STAT_OFFSET(irq_exits) },
78 { "light_exits", STAT_OFFSET(light_exits) },
79 { "efer_reload", STAT_OFFSET(efer_reload) },
80 { NULL }
81 };
82
83 static struct dentry *debugfs_dir;
84
85 #define MAX_IO_MSRS 256
86
87 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
88 #define LMSW_GUEST_MASK 0x0eULL
89 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
90 #define CR8_RESEVED_BITS (~0x0fULL)
91 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
92
93 #ifdef CONFIG_X86_64
94 // LDT or TSS descriptor in the GDT. 16 bytes.
95 struct segment_descriptor_64 {
96 struct segment_descriptor s;
97 u32 base_higher;
98 u32 pad_zero;
99 };
100
101 #endif
102
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
104 unsigned long arg);
105
106 unsigned long segment_base(u16 selector)
107 {
108 struct descriptor_table gdt;
109 struct segment_descriptor *d;
110 unsigned long table_base;
111 typedef unsigned long ul;
112 unsigned long v;
113
114 if (selector == 0)
115 return 0;
116
117 asm ("sgdt %0" : "=m"(gdt));
118 table_base = gdt.base;
119
120 if (selector & 4) { /* from ldt */
121 u16 ldt_selector;
122
123 asm ("sldt %0" : "=g"(ldt_selector));
124 table_base = segment_base(ldt_selector);
125 }
126 d = (struct segment_descriptor *)(table_base + (selector & ~7));
127 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
128 #ifdef CONFIG_X86_64
129 if (d->system == 0
130 && (d->type == 2 || d->type == 9 || d->type == 11))
131 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
132 #endif
133 return v;
134 }
135 EXPORT_SYMBOL_GPL(segment_base);
136
137 static inline int valid_vcpu(int n)
138 {
139 return likely(n >= 0 && n < KVM_MAX_VCPUS);
140 }
141
142 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
143 void *dest)
144 {
145 unsigned char *host_buf = dest;
146 unsigned long req_size = size;
147
148 while (size) {
149 hpa_t paddr;
150 unsigned now;
151 unsigned offset;
152 hva_t guest_buf;
153
154 paddr = gva_to_hpa(vcpu, addr);
155
156 if (is_error_hpa(paddr))
157 break;
158
159 guest_buf = (hva_t)kmap_atomic(
160 pfn_to_page(paddr >> PAGE_SHIFT),
161 KM_USER0);
162 offset = addr & ~PAGE_MASK;
163 guest_buf |= offset;
164 now = min(size, PAGE_SIZE - offset);
165 memcpy(host_buf, (void*)guest_buf, now);
166 host_buf += now;
167 addr += now;
168 size -= now;
169 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
170 }
171 return req_size - size;
172 }
173 EXPORT_SYMBOL_GPL(kvm_read_guest);
174
175 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
176 void *data)
177 {
178 unsigned char *host_buf = data;
179 unsigned long req_size = size;
180
181 while (size) {
182 hpa_t paddr;
183 unsigned now;
184 unsigned offset;
185 hva_t guest_buf;
186 gfn_t gfn;
187
188 paddr = gva_to_hpa(vcpu, addr);
189
190 if (is_error_hpa(paddr))
191 break;
192
193 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
194 mark_page_dirty(vcpu->kvm, gfn);
195 guest_buf = (hva_t)kmap_atomic(
196 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
197 offset = addr & ~PAGE_MASK;
198 guest_buf |= offset;
199 now = min(size, PAGE_SIZE - offset);
200 memcpy((void*)guest_buf, host_buf, now);
201 host_buf += now;
202 addr += now;
203 size -= now;
204 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
205 }
206 return req_size - size;
207 }
208 EXPORT_SYMBOL_GPL(kvm_write_guest);
209
210 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
211 {
212 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
213 return;
214
215 vcpu->guest_fpu_loaded = 1;
216 fx_save(vcpu->host_fx_image);
217 fx_restore(vcpu->guest_fx_image);
218 }
219 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
220
221 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
222 {
223 if (!vcpu->guest_fpu_loaded)
224 return;
225
226 vcpu->guest_fpu_loaded = 0;
227 fx_save(vcpu->guest_fx_image);
228 fx_restore(vcpu->host_fx_image);
229 }
230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
231
232 /*
233 * Switches to specified vcpu, until a matching vcpu_put()
234 */
235 static void vcpu_load(struct kvm_vcpu *vcpu)
236 {
237 mutex_lock(&vcpu->mutex);
238 kvm_arch_ops->vcpu_load(vcpu);
239 }
240
241 static void vcpu_put(struct kvm_vcpu *vcpu)
242 {
243 kvm_arch_ops->vcpu_put(vcpu);
244 mutex_unlock(&vcpu->mutex);
245 }
246
247 static void ack_flush(void *_completed)
248 {
249 atomic_t *completed = _completed;
250
251 atomic_inc(completed);
252 }
253
254 void kvm_flush_remote_tlbs(struct kvm *kvm)
255 {
256 int i, cpu, needed;
257 cpumask_t cpus;
258 struct kvm_vcpu *vcpu;
259 atomic_t completed;
260
261 atomic_set(&completed, 0);
262 cpus_clear(cpus);
263 needed = 0;
264 for (i = 0; i < kvm->nvcpus; ++i) {
265 vcpu = &kvm->vcpus[i];
266 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
267 continue;
268 cpu = vcpu->cpu;
269 if (cpu != -1 && cpu != raw_smp_processor_id())
270 if (!cpu_isset(cpu, cpus)) {
271 cpu_set(cpu, cpus);
272 ++needed;
273 }
274 }
275
276 /* Uniprocessor kernel does not respect cpus in first_cpu. So
277 * do not go there if we have nothing to do. */
278 if (cpus_empty(cpus))
279 return;
280
281 /*
282 * We really want smp_call_function_mask() here. But that's not
283 * available, so ipi all cpus in parallel and wait for them
284 * to complete.
285 */
286 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
287 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
288 while (atomic_read(&completed) != needed) {
289 cpu_relax();
290 barrier();
291 }
292 }
293
294 static struct kvm *kvm_create_vm(void)
295 {
296 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
297 int i;
298
299 if (!kvm)
300 return ERR_PTR(-ENOMEM);
301
302 kvm_io_bus_init(&kvm->pio_bus);
303 spin_lock_init(&kvm->lock);
304 INIT_LIST_HEAD(&kvm->active_mmu_pages);
305 kvm_io_bus_init(&kvm->mmio_bus);
306 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
307 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
308
309 mutex_init(&vcpu->mutex);
310 vcpu->cpu = -1;
311 vcpu->kvm = kvm;
312 vcpu->mmu.root_hpa = INVALID_PAGE;
313 }
314 spin_lock(&kvm_lock);
315 list_add(&kvm->vm_list, &vm_list);
316 spin_unlock(&kvm_lock);
317 return kvm;
318 }
319
320 static int kvm_dev_open(struct inode *inode, struct file *filp)
321 {
322 return 0;
323 }
324
325 /*
326 * Free any memory in @free but not in @dont.
327 */
328 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
329 struct kvm_memory_slot *dont)
330 {
331 int i;
332
333 if (!dont || free->phys_mem != dont->phys_mem)
334 if (free->phys_mem) {
335 for (i = 0; i < free->npages; ++i)
336 if (free->phys_mem[i])
337 __free_page(free->phys_mem[i]);
338 vfree(free->phys_mem);
339 }
340
341 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
342 vfree(free->dirty_bitmap);
343
344 free->phys_mem = NULL;
345 free->npages = 0;
346 free->dirty_bitmap = NULL;
347 }
348
349 static void kvm_free_physmem(struct kvm *kvm)
350 {
351 int i;
352
353 for (i = 0; i < kvm->nmemslots; ++i)
354 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
355 }
356
357 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
358 {
359 int i;
360
361 for (i = 0; i < 2; ++i)
362 if (vcpu->pio.guest_pages[i]) {
363 __free_page(vcpu->pio.guest_pages[i]);
364 vcpu->pio.guest_pages[i] = NULL;
365 }
366 }
367
368 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
369 {
370 if (!vcpu->vmcs)
371 return;
372
373 vcpu_load(vcpu);
374 kvm_mmu_unload(vcpu);
375 vcpu_put(vcpu);
376 }
377
378 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
379 {
380 if (!vcpu->vmcs)
381 return;
382
383 vcpu_load(vcpu);
384 kvm_mmu_destroy(vcpu);
385 vcpu_put(vcpu);
386 kvm_arch_ops->vcpu_free(vcpu);
387 free_page((unsigned long)vcpu->run);
388 vcpu->run = NULL;
389 free_page((unsigned long)vcpu->pio_data);
390 vcpu->pio_data = NULL;
391 free_pio_guest_pages(vcpu);
392 }
393
394 static void kvm_free_vcpus(struct kvm *kvm)
395 {
396 unsigned int i;
397
398 /*
399 * Unpin any mmu pages first.
400 */
401 for (i = 0; i < KVM_MAX_VCPUS; ++i)
402 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
403 for (i = 0; i < KVM_MAX_VCPUS; ++i)
404 kvm_free_vcpu(&kvm->vcpus[i]);
405 }
406
407 static int kvm_dev_release(struct inode *inode, struct file *filp)
408 {
409 return 0;
410 }
411
412 static void kvm_destroy_vm(struct kvm *kvm)
413 {
414 spin_lock(&kvm_lock);
415 list_del(&kvm->vm_list);
416 spin_unlock(&kvm_lock);
417 kvm_io_bus_destroy(&kvm->pio_bus);
418 kvm_io_bus_destroy(&kvm->mmio_bus);
419 kvm_free_vcpus(kvm);
420 kvm_free_physmem(kvm);
421 kfree(kvm);
422 }
423
424 static int kvm_vm_release(struct inode *inode, struct file *filp)
425 {
426 struct kvm *kvm = filp->private_data;
427
428 kvm_destroy_vm(kvm);
429 return 0;
430 }
431
432 static void inject_gp(struct kvm_vcpu *vcpu)
433 {
434 kvm_arch_ops->inject_gp(vcpu, 0);
435 }
436
437 /*
438 * Load the pae pdptrs. Return true is they are all valid.
439 */
440 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
441 {
442 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
443 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
444 int i;
445 u64 pdpte;
446 u64 *pdpt;
447 int ret;
448 struct page *page;
449
450 spin_lock(&vcpu->kvm->lock);
451 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
452 /* FIXME: !page - emulate? 0xff? */
453 pdpt = kmap_atomic(page, KM_USER0);
454
455 ret = 1;
456 for (i = 0; i < 4; ++i) {
457 pdpte = pdpt[offset + i];
458 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
459 ret = 0;
460 goto out;
461 }
462 }
463
464 for (i = 0; i < 4; ++i)
465 vcpu->pdptrs[i] = pdpt[offset + i];
466
467 out:
468 kunmap_atomic(pdpt, KM_USER0);
469 spin_unlock(&vcpu->kvm->lock);
470
471 return ret;
472 }
473
474 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
475 {
476 if (cr0 & CR0_RESEVED_BITS) {
477 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
478 cr0, vcpu->cr0);
479 inject_gp(vcpu);
480 return;
481 }
482
483 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
484 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
485 inject_gp(vcpu);
486 return;
487 }
488
489 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
490 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
491 "and a clear PE flag\n");
492 inject_gp(vcpu);
493 return;
494 }
495
496 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
497 #ifdef CONFIG_X86_64
498 if ((vcpu->shadow_efer & EFER_LME)) {
499 int cs_db, cs_l;
500
501 if (!is_pae(vcpu)) {
502 printk(KERN_DEBUG "set_cr0: #GP, start paging "
503 "in long mode while PAE is disabled\n");
504 inject_gp(vcpu);
505 return;
506 }
507 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
508 if (cs_l) {
509 printk(KERN_DEBUG "set_cr0: #GP, start paging "
510 "in long mode while CS.L == 1\n");
511 inject_gp(vcpu);
512 return;
513
514 }
515 } else
516 #endif
517 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
518 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
519 "reserved bits\n");
520 inject_gp(vcpu);
521 return;
522 }
523
524 }
525
526 kvm_arch_ops->set_cr0(vcpu, cr0);
527 vcpu->cr0 = cr0;
528
529 spin_lock(&vcpu->kvm->lock);
530 kvm_mmu_reset_context(vcpu);
531 spin_unlock(&vcpu->kvm->lock);
532 return;
533 }
534 EXPORT_SYMBOL_GPL(set_cr0);
535
536 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
537 {
538 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
539 }
540 EXPORT_SYMBOL_GPL(lmsw);
541
542 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
543 {
544 if (cr4 & CR4_RESEVED_BITS) {
545 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
546 inject_gp(vcpu);
547 return;
548 }
549
550 if (is_long_mode(vcpu)) {
551 if (!(cr4 & CR4_PAE_MASK)) {
552 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
553 "in long mode\n");
554 inject_gp(vcpu);
555 return;
556 }
557 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
558 && !load_pdptrs(vcpu, vcpu->cr3)) {
559 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
560 inject_gp(vcpu);
561 }
562
563 if (cr4 & CR4_VMXE_MASK) {
564 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
565 inject_gp(vcpu);
566 return;
567 }
568 kvm_arch_ops->set_cr4(vcpu, cr4);
569 spin_lock(&vcpu->kvm->lock);
570 kvm_mmu_reset_context(vcpu);
571 spin_unlock(&vcpu->kvm->lock);
572 }
573 EXPORT_SYMBOL_GPL(set_cr4);
574
575 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
576 {
577 if (is_long_mode(vcpu)) {
578 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
579 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
580 inject_gp(vcpu);
581 return;
582 }
583 } else {
584 if (cr3 & CR3_RESEVED_BITS) {
585 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
586 inject_gp(vcpu);
587 return;
588 }
589 if (is_paging(vcpu) && is_pae(vcpu) &&
590 !load_pdptrs(vcpu, cr3)) {
591 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
592 "reserved bits\n");
593 inject_gp(vcpu);
594 return;
595 }
596 }
597
598 vcpu->cr3 = cr3;
599 spin_lock(&vcpu->kvm->lock);
600 /*
601 * Does the new cr3 value map to physical memory? (Note, we
602 * catch an invalid cr3 even in real-mode, because it would
603 * cause trouble later on when we turn on paging anyway.)
604 *
605 * A real CPU would silently accept an invalid cr3 and would
606 * attempt to use it - with largely undefined (and often hard
607 * to debug) behavior on the guest side.
608 */
609 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
610 inject_gp(vcpu);
611 else
612 vcpu->mmu.new_cr3(vcpu);
613 spin_unlock(&vcpu->kvm->lock);
614 }
615 EXPORT_SYMBOL_GPL(set_cr3);
616
617 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
618 {
619 if ( cr8 & CR8_RESEVED_BITS) {
620 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
621 inject_gp(vcpu);
622 return;
623 }
624 vcpu->cr8 = cr8;
625 }
626 EXPORT_SYMBOL_GPL(set_cr8);
627
628 void fx_init(struct kvm_vcpu *vcpu)
629 {
630 struct __attribute__ ((__packed__)) fx_image_s {
631 u16 control; //fcw
632 u16 status; //fsw
633 u16 tag; // ftw
634 u16 opcode; //fop
635 u64 ip; // fpu ip
636 u64 operand;// fpu dp
637 u32 mxcsr;
638 u32 mxcsr_mask;
639
640 } *fx_image;
641
642 fx_save(vcpu->host_fx_image);
643 fpu_init();
644 fx_save(vcpu->guest_fx_image);
645 fx_restore(vcpu->host_fx_image);
646
647 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
648 fx_image->mxcsr = 0x1f80;
649 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
650 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
651 }
652 EXPORT_SYMBOL_GPL(fx_init);
653
654 /*
655 * Allocate some memory and give it an address in the guest physical address
656 * space.
657 *
658 * Discontiguous memory is allowed, mostly for framebuffers.
659 */
660 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
661 struct kvm_memory_region *mem)
662 {
663 int r;
664 gfn_t base_gfn;
665 unsigned long npages;
666 unsigned long i;
667 struct kvm_memory_slot *memslot;
668 struct kvm_memory_slot old, new;
669 int memory_config_version;
670
671 r = -EINVAL;
672 /* General sanity checks */
673 if (mem->memory_size & (PAGE_SIZE - 1))
674 goto out;
675 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
676 goto out;
677 if (mem->slot >= KVM_MEMORY_SLOTS)
678 goto out;
679 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
680 goto out;
681
682 memslot = &kvm->memslots[mem->slot];
683 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
684 npages = mem->memory_size >> PAGE_SHIFT;
685
686 if (!npages)
687 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
688
689 raced:
690 spin_lock(&kvm->lock);
691
692 memory_config_version = kvm->memory_config_version;
693 new = old = *memslot;
694
695 new.base_gfn = base_gfn;
696 new.npages = npages;
697 new.flags = mem->flags;
698
699 /* Disallow changing a memory slot's size. */
700 r = -EINVAL;
701 if (npages && old.npages && npages != old.npages)
702 goto out_unlock;
703
704 /* Check for overlaps */
705 r = -EEXIST;
706 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
707 struct kvm_memory_slot *s = &kvm->memslots[i];
708
709 if (s == memslot)
710 continue;
711 if (!((base_gfn + npages <= s->base_gfn) ||
712 (base_gfn >= s->base_gfn + s->npages)))
713 goto out_unlock;
714 }
715 /*
716 * Do memory allocations outside lock. memory_config_version will
717 * detect any races.
718 */
719 spin_unlock(&kvm->lock);
720
721 /* Deallocate if slot is being removed */
722 if (!npages)
723 new.phys_mem = NULL;
724
725 /* Free page dirty bitmap if unneeded */
726 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
727 new.dirty_bitmap = NULL;
728
729 r = -ENOMEM;
730
731 /* Allocate if a slot is being created */
732 if (npages && !new.phys_mem) {
733 new.phys_mem = vmalloc(npages * sizeof(struct page *));
734
735 if (!new.phys_mem)
736 goto out_free;
737
738 memset(new.phys_mem, 0, npages * sizeof(struct page *));
739 for (i = 0; i < npages; ++i) {
740 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
741 | __GFP_ZERO);
742 if (!new.phys_mem[i])
743 goto out_free;
744 set_page_private(new.phys_mem[i],0);
745 }
746 }
747
748 /* Allocate page dirty bitmap if needed */
749 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
750 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
751
752 new.dirty_bitmap = vmalloc(dirty_bytes);
753 if (!new.dirty_bitmap)
754 goto out_free;
755 memset(new.dirty_bitmap, 0, dirty_bytes);
756 }
757
758 spin_lock(&kvm->lock);
759
760 if (memory_config_version != kvm->memory_config_version) {
761 spin_unlock(&kvm->lock);
762 kvm_free_physmem_slot(&new, &old);
763 goto raced;
764 }
765
766 r = -EAGAIN;
767 if (kvm->busy)
768 goto out_unlock;
769
770 if (mem->slot >= kvm->nmemslots)
771 kvm->nmemslots = mem->slot + 1;
772
773 *memslot = new;
774 ++kvm->memory_config_version;
775
776 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
777 kvm_flush_remote_tlbs(kvm);
778
779 spin_unlock(&kvm->lock);
780
781 kvm_free_physmem_slot(&old, &new);
782 return 0;
783
784 out_unlock:
785 spin_unlock(&kvm->lock);
786 out_free:
787 kvm_free_physmem_slot(&new, &old);
788 out:
789 return r;
790 }
791
792 /*
793 * Get (and clear) the dirty memory log for a memory slot.
794 */
795 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
796 struct kvm_dirty_log *log)
797 {
798 struct kvm_memory_slot *memslot;
799 int r, i;
800 int n;
801 unsigned long any = 0;
802
803 spin_lock(&kvm->lock);
804
805 /*
806 * Prevent changes to guest memory configuration even while the lock
807 * is not taken.
808 */
809 ++kvm->busy;
810 spin_unlock(&kvm->lock);
811 r = -EINVAL;
812 if (log->slot >= KVM_MEMORY_SLOTS)
813 goto out;
814
815 memslot = &kvm->memslots[log->slot];
816 r = -ENOENT;
817 if (!memslot->dirty_bitmap)
818 goto out;
819
820 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
821
822 for (i = 0; !any && i < n/sizeof(long); ++i)
823 any = memslot->dirty_bitmap[i];
824
825 r = -EFAULT;
826 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
827 goto out;
828
829 spin_lock(&kvm->lock);
830 kvm_mmu_slot_remove_write_access(kvm, log->slot);
831 kvm_flush_remote_tlbs(kvm);
832 memset(memslot->dirty_bitmap, 0, n);
833 spin_unlock(&kvm->lock);
834
835 r = 0;
836
837 out:
838 spin_lock(&kvm->lock);
839 --kvm->busy;
840 spin_unlock(&kvm->lock);
841 return r;
842 }
843
844 /*
845 * Set a new alias region. Aliases map a portion of physical memory into
846 * another portion. This is useful for memory windows, for example the PC
847 * VGA region.
848 */
849 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
850 struct kvm_memory_alias *alias)
851 {
852 int r, n;
853 struct kvm_mem_alias *p;
854
855 r = -EINVAL;
856 /* General sanity checks */
857 if (alias->memory_size & (PAGE_SIZE - 1))
858 goto out;
859 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
860 goto out;
861 if (alias->slot >= KVM_ALIAS_SLOTS)
862 goto out;
863 if (alias->guest_phys_addr + alias->memory_size
864 < alias->guest_phys_addr)
865 goto out;
866 if (alias->target_phys_addr + alias->memory_size
867 < alias->target_phys_addr)
868 goto out;
869
870 spin_lock(&kvm->lock);
871
872 p = &kvm->aliases[alias->slot];
873 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
874 p->npages = alias->memory_size >> PAGE_SHIFT;
875 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
876
877 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
878 if (kvm->aliases[n - 1].npages)
879 break;
880 kvm->naliases = n;
881
882 kvm_mmu_zap_all(kvm);
883
884 spin_unlock(&kvm->lock);
885
886 return 0;
887
888 out:
889 return r;
890 }
891
892 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
893 {
894 int i;
895 struct kvm_mem_alias *alias;
896
897 for (i = 0; i < kvm->naliases; ++i) {
898 alias = &kvm->aliases[i];
899 if (gfn >= alias->base_gfn
900 && gfn < alias->base_gfn + alias->npages)
901 return alias->target_gfn + gfn - alias->base_gfn;
902 }
903 return gfn;
904 }
905
906 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
907 {
908 int i;
909
910 for (i = 0; i < kvm->nmemslots; ++i) {
911 struct kvm_memory_slot *memslot = &kvm->memslots[i];
912
913 if (gfn >= memslot->base_gfn
914 && gfn < memslot->base_gfn + memslot->npages)
915 return memslot;
916 }
917 return NULL;
918 }
919
920 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
921 {
922 gfn = unalias_gfn(kvm, gfn);
923 return __gfn_to_memslot(kvm, gfn);
924 }
925
926 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
927 {
928 struct kvm_memory_slot *slot;
929
930 gfn = unalias_gfn(kvm, gfn);
931 slot = __gfn_to_memslot(kvm, gfn);
932 if (!slot)
933 return NULL;
934 return slot->phys_mem[gfn - slot->base_gfn];
935 }
936 EXPORT_SYMBOL_GPL(gfn_to_page);
937
938 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
939 {
940 int i;
941 struct kvm_memory_slot *memslot;
942 unsigned long rel_gfn;
943
944 for (i = 0; i < kvm->nmemslots; ++i) {
945 memslot = &kvm->memslots[i];
946
947 if (gfn >= memslot->base_gfn
948 && gfn < memslot->base_gfn + memslot->npages) {
949
950 if (!memslot->dirty_bitmap)
951 return;
952
953 rel_gfn = gfn - memslot->base_gfn;
954
955 /* avoid RMW */
956 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
957 set_bit(rel_gfn, memslot->dirty_bitmap);
958 return;
959 }
960 }
961 }
962
963 static int emulator_read_std(unsigned long addr,
964 void *val,
965 unsigned int bytes,
966 struct x86_emulate_ctxt *ctxt)
967 {
968 struct kvm_vcpu *vcpu = ctxt->vcpu;
969 void *data = val;
970
971 while (bytes) {
972 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
973 unsigned offset = addr & (PAGE_SIZE-1);
974 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
975 unsigned long pfn;
976 struct page *page;
977 void *page_virt;
978
979 if (gpa == UNMAPPED_GVA)
980 return X86EMUL_PROPAGATE_FAULT;
981 pfn = gpa >> PAGE_SHIFT;
982 page = gfn_to_page(vcpu->kvm, pfn);
983 if (!page)
984 return X86EMUL_UNHANDLEABLE;
985 page_virt = kmap_atomic(page, KM_USER0);
986
987 memcpy(data, page_virt + offset, tocopy);
988
989 kunmap_atomic(page_virt, KM_USER0);
990
991 bytes -= tocopy;
992 data += tocopy;
993 addr += tocopy;
994 }
995
996 return X86EMUL_CONTINUE;
997 }
998
999 static int emulator_write_std(unsigned long addr,
1000 const void *val,
1001 unsigned int bytes,
1002 struct x86_emulate_ctxt *ctxt)
1003 {
1004 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1005 addr, bytes);
1006 return X86EMUL_UNHANDLEABLE;
1007 }
1008
1009 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1010 gpa_t addr)
1011 {
1012 /*
1013 * Note that its important to have this wrapper function because
1014 * in the very near future we will be checking for MMIOs against
1015 * the LAPIC as well as the general MMIO bus
1016 */
1017 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1018 }
1019
1020 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1021 gpa_t addr)
1022 {
1023 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1024 }
1025
1026 static int emulator_read_emulated(unsigned long addr,
1027 void *val,
1028 unsigned int bytes,
1029 struct x86_emulate_ctxt *ctxt)
1030 {
1031 struct kvm_vcpu *vcpu = ctxt->vcpu;
1032 struct kvm_io_device *mmio_dev;
1033 gpa_t gpa;
1034
1035 if (vcpu->mmio_read_completed) {
1036 memcpy(val, vcpu->mmio_data, bytes);
1037 vcpu->mmio_read_completed = 0;
1038 return X86EMUL_CONTINUE;
1039 } else if (emulator_read_std(addr, val, bytes, ctxt)
1040 == X86EMUL_CONTINUE)
1041 return X86EMUL_CONTINUE;
1042
1043 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1044 if (gpa == UNMAPPED_GVA)
1045 return X86EMUL_PROPAGATE_FAULT;
1046
1047 /*
1048 * Is this MMIO handled locally?
1049 */
1050 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1051 if (mmio_dev) {
1052 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1053 return X86EMUL_CONTINUE;
1054 }
1055
1056 vcpu->mmio_needed = 1;
1057 vcpu->mmio_phys_addr = gpa;
1058 vcpu->mmio_size = bytes;
1059 vcpu->mmio_is_write = 0;
1060
1061 return X86EMUL_UNHANDLEABLE;
1062 }
1063
1064 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1065 const void *val, int bytes)
1066 {
1067 struct page *page;
1068 void *virt;
1069 unsigned offset = offset_in_page(gpa);
1070
1071 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1072 return 0;
1073 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1074 if (!page)
1075 return 0;
1076 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1077 virt = kmap_atomic(page, KM_USER0);
1078 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1079 memcpy(virt + offset_in_page(gpa), val, bytes);
1080 kunmap_atomic(virt, KM_USER0);
1081 return 1;
1082 }
1083
1084 static int emulator_write_emulated_onepage(unsigned long addr,
1085 const void *val,
1086 unsigned int bytes,
1087 struct x86_emulate_ctxt *ctxt)
1088 {
1089 struct kvm_vcpu *vcpu = ctxt->vcpu;
1090 struct kvm_io_device *mmio_dev;
1091 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1092
1093 if (gpa == UNMAPPED_GVA) {
1094 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1095 return X86EMUL_PROPAGATE_FAULT;
1096 }
1097
1098 if (emulator_write_phys(vcpu, gpa, val, bytes))
1099 return X86EMUL_CONTINUE;
1100
1101 /*
1102 * Is this MMIO handled locally?
1103 */
1104 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1105 if (mmio_dev) {
1106 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1107 return X86EMUL_CONTINUE;
1108 }
1109
1110 vcpu->mmio_needed = 1;
1111 vcpu->mmio_phys_addr = gpa;
1112 vcpu->mmio_size = bytes;
1113 vcpu->mmio_is_write = 1;
1114 memcpy(vcpu->mmio_data, val, bytes);
1115
1116 return X86EMUL_CONTINUE;
1117 }
1118
1119 static int emulator_write_emulated(unsigned long addr,
1120 const void *val,
1121 unsigned int bytes,
1122 struct x86_emulate_ctxt *ctxt)
1123 {
1124 /* Crossing a page boundary? */
1125 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1126 int rc, now;
1127
1128 now = -addr & ~PAGE_MASK;
1129 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1130 if (rc != X86EMUL_CONTINUE)
1131 return rc;
1132 addr += now;
1133 val += now;
1134 bytes -= now;
1135 }
1136 return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1137 }
1138
1139 static int emulator_cmpxchg_emulated(unsigned long addr,
1140 const void *old,
1141 const void *new,
1142 unsigned int bytes,
1143 struct x86_emulate_ctxt *ctxt)
1144 {
1145 static int reported;
1146
1147 if (!reported) {
1148 reported = 1;
1149 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1150 }
1151 return emulator_write_emulated(addr, new, bytes, ctxt);
1152 }
1153
1154 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1155 {
1156 return kvm_arch_ops->get_segment_base(vcpu, seg);
1157 }
1158
1159 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1160 {
1161 return X86EMUL_CONTINUE;
1162 }
1163
1164 int emulate_clts(struct kvm_vcpu *vcpu)
1165 {
1166 kvm_arch_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1167 return X86EMUL_CONTINUE;
1168 }
1169
1170 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1171 {
1172 struct kvm_vcpu *vcpu = ctxt->vcpu;
1173
1174 switch (dr) {
1175 case 0 ... 3:
1176 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1177 return X86EMUL_CONTINUE;
1178 default:
1179 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1180 __FUNCTION__, dr);
1181 return X86EMUL_UNHANDLEABLE;
1182 }
1183 }
1184
1185 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1186 {
1187 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1188 int exception;
1189
1190 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1191 if (exception) {
1192 /* FIXME: better handling */
1193 return X86EMUL_UNHANDLEABLE;
1194 }
1195 return X86EMUL_CONTINUE;
1196 }
1197
1198 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1199 {
1200 static int reported;
1201 u8 opcodes[4];
1202 unsigned long rip = ctxt->vcpu->rip;
1203 unsigned long rip_linear;
1204
1205 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1206
1207 if (reported)
1208 return;
1209
1210 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1211
1212 printk(KERN_ERR "emulation failed but !mmio_needed?"
1213 " rip %lx %02x %02x %02x %02x\n",
1214 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1215 reported = 1;
1216 }
1217
1218 struct x86_emulate_ops emulate_ops = {
1219 .read_std = emulator_read_std,
1220 .write_std = emulator_write_std,
1221 .read_emulated = emulator_read_emulated,
1222 .write_emulated = emulator_write_emulated,
1223 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1224 };
1225
1226 int emulate_instruction(struct kvm_vcpu *vcpu,
1227 struct kvm_run *run,
1228 unsigned long cr2,
1229 u16 error_code)
1230 {
1231 struct x86_emulate_ctxt emulate_ctxt;
1232 int r;
1233 int cs_db, cs_l;
1234
1235 vcpu->mmio_fault_cr2 = cr2;
1236 kvm_arch_ops->cache_regs(vcpu);
1237
1238 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1239
1240 emulate_ctxt.vcpu = vcpu;
1241 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1242 emulate_ctxt.cr2 = cr2;
1243 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1244 ? X86EMUL_MODE_REAL : cs_l
1245 ? X86EMUL_MODE_PROT64 : cs_db
1246 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1247
1248 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1249 emulate_ctxt.cs_base = 0;
1250 emulate_ctxt.ds_base = 0;
1251 emulate_ctxt.es_base = 0;
1252 emulate_ctxt.ss_base = 0;
1253 } else {
1254 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1255 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1256 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1257 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1258 }
1259
1260 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1261 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1262
1263 vcpu->mmio_is_write = 0;
1264 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1265
1266 if ((r || vcpu->mmio_is_write) && run) {
1267 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1268 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1269 run->mmio.len = vcpu->mmio_size;
1270 run->mmio.is_write = vcpu->mmio_is_write;
1271 }
1272
1273 if (r) {
1274 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1275 return EMULATE_DONE;
1276 if (!vcpu->mmio_needed) {
1277 report_emulation_failure(&emulate_ctxt);
1278 return EMULATE_FAIL;
1279 }
1280 return EMULATE_DO_MMIO;
1281 }
1282
1283 kvm_arch_ops->decache_regs(vcpu);
1284 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1285
1286 if (vcpu->mmio_is_write) {
1287 vcpu->mmio_needed = 0;
1288 return EMULATE_DO_MMIO;
1289 }
1290
1291 return EMULATE_DONE;
1292 }
1293 EXPORT_SYMBOL_GPL(emulate_instruction);
1294
1295 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1296 {
1297 if (vcpu->irq_summary)
1298 return 1;
1299
1300 vcpu->run->exit_reason = KVM_EXIT_HLT;
1301 ++vcpu->stat.halt_exits;
1302 return 0;
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1305
1306 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1307 {
1308 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1309
1310 kvm_arch_ops->cache_regs(vcpu);
1311 ret = -KVM_EINVAL;
1312 #ifdef CONFIG_X86_64
1313 if (is_long_mode(vcpu)) {
1314 nr = vcpu->regs[VCPU_REGS_RAX];
1315 a0 = vcpu->regs[VCPU_REGS_RDI];
1316 a1 = vcpu->regs[VCPU_REGS_RSI];
1317 a2 = vcpu->regs[VCPU_REGS_RDX];
1318 a3 = vcpu->regs[VCPU_REGS_RCX];
1319 a4 = vcpu->regs[VCPU_REGS_R8];
1320 a5 = vcpu->regs[VCPU_REGS_R9];
1321 } else
1322 #endif
1323 {
1324 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1325 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1326 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1327 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1328 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1329 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1330 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1331 }
1332 switch (nr) {
1333 default:
1334 run->hypercall.args[0] = a0;
1335 run->hypercall.args[1] = a1;
1336 run->hypercall.args[2] = a2;
1337 run->hypercall.args[3] = a3;
1338 run->hypercall.args[4] = a4;
1339 run->hypercall.args[5] = a5;
1340 run->hypercall.ret = ret;
1341 run->hypercall.longmode = is_long_mode(vcpu);
1342 kvm_arch_ops->decache_regs(vcpu);
1343 return 0;
1344 }
1345 vcpu->regs[VCPU_REGS_RAX] = ret;
1346 kvm_arch_ops->decache_regs(vcpu);
1347 return 1;
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_hypercall);
1350
1351 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1352 {
1353 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1354 }
1355
1356 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1357 {
1358 struct descriptor_table dt = { limit, base };
1359
1360 kvm_arch_ops->set_gdt(vcpu, &dt);
1361 }
1362
1363 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1364 {
1365 struct descriptor_table dt = { limit, base };
1366
1367 kvm_arch_ops->set_idt(vcpu, &dt);
1368 }
1369
1370 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1371 unsigned long *rflags)
1372 {
1373 lmsw(vcpu, msw);
1374 *rflags = kvm_arch_ops->get_rflags(vcpu);
1375 }
1376
1377 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1378 {
1379 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1380 switch (cr) {
1381 case 0:
1382 return vcpu->cr0;
1383 case 2:
1384 return vcpu->cr2;
1385 case 3:
1386 return vcpu->cr3;
1387 case 4:
1388 return vcpu->cr4;
1389 default:
1390 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1391 return 0;
1392 }
1393 }
1394
1395 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1396 unsigned long *rflags)
1397 {
1398 switch (cr) {
1399 case 0:
1400 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1401 *rflags = kvm_arch_ops->get_rflags(vcpu);
1402 break;
1403 case 2:
1404 vcpu->cr2 = val;
1405 break;
1406 case 3:
1407 set_cr3(vcpu, val);
1408 break;
1409 case 4:
1410 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1411 break;
1412 default:
1413 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1414 }
1415 }
1416
1417 /*
1418 * Register the para guest with the host:
1419 */
1420 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1421 {
1422 struct kvm_vcpu_para_state *para_state;
1423 hpa_t para_state_hpa, hypercall_hpa;
1424 struct page *para_state_page;
1425 unsigned char *hypercall;
1426 gpa_t hypercall_gpa;
1427
1428 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1429 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1430
1431 /*
1432 * Needs to be page aligned:
1433 */
1434 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1435 goto err_gp;
1436
1437 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1438 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1439 if (is_error_hpa(para_state_hpa))
1440 goto err_gp;
1441
1442 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1443 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1444 para_state = kmap_atomic(para_state_page, KM_USER0);
1445
1446 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1447 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1448
1449 para_state->host_version = KVM_PARA_API_VERSION;
1450 /*
1451 * We cannot support guests that try to register themselves
1452 * with a newer API version than the host supports:
1453 */
1454 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1455 para_state->ret = -KVM_EINVAL;
1456 goto err_kunmap_skip;
1457 }
1458
1459 hypercall_gpa = para_state->hypercall_gpa;
1460 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1461 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1462 if (is_error_hpa(hypercall_hpa)) {
1463 para_state->ret = -KVM_EINVAL;
1464 goto err_kunmap_skip;
1465 }
1466
1467 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1468 vcpu->para_state_page = para_state_page;
1469 vcpu->para_state_gpa = para_state_gpa;
1470 vcpu->hypercall_gpa = hypercall_gpa;
1471
1472 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1473 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1474 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1475 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1476 kunmap_atomic(hypercall, KM_USER1);
1477
1478 para_state->ret = 0;
1479 err_kunmap_skip:
1480 kunmap_atomic(para_state, KM_USER0);
1481 return 0;
1482 err_gp:
1483 return 1;
1484 }
1485
1486 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1487 {
1488 u64 data;
1489
1490 switch (msr) {
1491 case 0xc0010010: /* SYSCFG */
1492 case 0xc0010015: /* HWCR */
1493 case MSR_IA32_PLATFORM_ID:
1494 case MSR_IA32_P5_MC_ADDR:
1495 case MSR_IA32_P5_MC_TYPE:
1496 case MSR_IA32_MC0_CTL:
1497 case MSR_IA32_MCG_STATUS:
1498 case MSR_IA32_MCG_CAP:
1499 case MSR_IA32_MC0_MISC:
1500 case MSR_IA32_MC0_MISC+4:
1501 case MSR_IA32_MC0_MISC+8:
1502 case MSR_IA32_MC0_MISC+12:
1503 case MSR_IA32_MC0_MISC+16:
1504 case MSR_IA32_UCODE_REV:
1505 case MSR_IA32_PERF_STATUS:
1506 case MSR_IA32_EBL_CR_POWERON:
1507 /* MTRR registers */
1508 case 0xfe:
1509 case 0x200 ... 0x2ff:
1510 data = 0;
1511 break;
1512 case 0xcd: /* fsb frequency */
1513 data = 3;
1514 break;
1515 case MSR_IA32_APICBASE:
1516 data = vcpu->apic_base;
1517 break;
1518 case MSR_IA32_MISC_ENABLE:
1519 data = vcpu->ia32_misc_enable_msr;
1520 break;
1521 #ifdef CONFIG_X86_64
1522 case MSR_EFER:
1523 data = vcpu->shadow_efer;
1524 break;
1525 #endif
1526 default:
1527 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1528 return 1;
1529 }
1530 *pdata = data;
1531 return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1534
1535 /*
1536 * Reads an msr value (of 'msr_index') into 'pdata'.
1537 * Returns 0 on success, non-0 otherwise.
1538 * Assumes vcpu_load() was already called.
1539 */
1540 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1541 {
1542 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1543 }
1544
1545 #ifdef CONFIG_X86_64
1546
1547 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1548 {
1549 if (efer & EFER_RESERVED_BITS) {
1550 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1551 efer);
1552 inject_gp(vcpu);
1553 return;
1554 }
1555
1556 if (is_paging(vcpu)
1557 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1558 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1559 inject_gp(vcpu);
1560 return;
1561 }
1562
1563 kvm_arch_ops->set_efer(vcpu, efer);
1564
1565 efer &= ~EFER_LMA;
1566 efer |= vcpu->shadow_efer & EFER_LMA;
1567
1568 vcpu->shadow_efer = efer;
1569 }
1570
1571 #endif
1572
1573 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1574 {
1575 switch (msr) {
1576 #ifdef CONFIG_X86_64
1577 case MSR_EFER:
1578 set_efer(vcpu, data);
1579 break;
1580 #endif
1581 case MSR_IA32_MC0_STATUS:
1582 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1583 __FUNCTION__, data);
1584 break;
1585 case MSR_IA32_MCG_STATUS:
1586 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1587 __FUNCTION__, data);
1588 break;
1589 case MSR_IA32_UCODE_REV:
1590 case MSR_IA32_UCODE_WRITE:
1591 case 0x200 ... 0x2ff: /* MTRRs */
1592 break;
1593 case MSR_IA32_APICBASE:
1594 vcpu->apic_base = data;
1595 break;
1596 case MSR_IA32_MISC_ENABLE:
1597 vcpu->ia32_misc_enable_msr = data;
1598 break;
1599 /*
1600 * This is the 'probe whether the host is KVM' logic:
1601 */
1602 case MSR_KVM_API_MAGIC:
1603 return vcpu_register_para(vcpu, data);
1604
1605 default:
1606 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1607 return 1;
1608 }
1609 return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1612
1613 /*
1614 * Writes msr value into into the appropriate "register".
1615 * Returns 0 on success, non-0 otherwise.
1616 * Assumes vcpu_load() was already called.
1617 */
1618 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1619 {
1620 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1621 }
1622
1623 void kvm_resched(struct kvm_vcpu *vcpu)
1624 {
1625 if (!need_resched())
1626 return;
1627 vcpu_put(vcpu);
1628 cond_resched();
1629 vcpu_load(vcpu);
1630 }
1631 EXPORT_SYMBOL_GPL(kvm_resched);
1632
1633 void load_msrs(struct vmx_msr_entry *e, int n)
1634 {
1635 int i;
1636
1637 for (i = 0; i < n; ++i)
1638 wrmsrl(e[i].index, e[i].data);
1639 }
1640 EXPORT_SYMBOL_GPL(load_msrs);
1641
1642 void save_msrs(struct vmx_msr_entry *e, int n)
1643 {
1644 int i;
1645
1646 for (i = 0; i < n; ++i)
1647 rdmsrl(e[i].index, e[i].data);
1648 }
1649 EXPORT_SYMBOL_GPL(save_msrs);
1650
1651 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1652 {
1653 int i;
1654 u32 function;
1655 struct kvm_cpuid_entry *e, *best;
1656
1657 kvm_arch_ops->cache_regs(vcpu);
1658 function = vcpu->regs[VCPU_REGS_RAX];
1659 vcpu->regs[VCPU_REGS_RAX] = 0;
1660 vcpu->regs[VCPU_REGS_RBX] = 0;
1661 vcpu->regs[VCPU_REGS_RCX] = 0;
1662 vcpu->regs[VCPU_REGS_RDX] = 0;
1663 best = NULL;
1664 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1665 e = &vcpu->cpuid_entries[i];
1666 if (e->function == function) {
1667 best = e;
1668 break;
1669 }
1670 /*
1671 * Both basic or both extended?
1672 */
1673 if (((e->function ^ function) & 0x80000000) == 0)
1674 if (!best || e->function > best->function)
1675 best = e;
1676 }
1677 if (best) {
1678 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1679 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1680 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1681 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1682 }
1683 kvm_arch_ops->decache_regs(vcpu);
1684 kvm_arch_ops->skip_emulated_instruction(vcpu);
1685 }
1686 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1687
1688 static int pio_copy_data(struct kvm_vcpu *vcpu)
1689 {
1690 void *p = vcpu->pio_data;
1691 void *q;
1692 unsigned bytes;
1693 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1694
1695 kvm_arch_ops->vcpu_put(vcpu);
1696 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1697 PAGE_KERNEL);
1698 if (!q) {
1699 kvm_arch_ops->vcpu_load(vcpu);
1700 free_pio_guest_pages(vcpu);
1701 return -ENOMEM;
1702 }
1703 q += vcpu->pio.guest_page_offset;
1704 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1705 if (vcpu->pio.in)
1706 memcpy(q, p, bytes);
1707 else
1708 memcpy(p, q, bytes);
1709 q -= vcpu->pio.guest_page_offset;
1710 vunmap(q);
1711 kvm_arch_ops->vcpu_load(vcpu);
1712 free_pio_guest_pages(vcpu);
1713 return 0;
1714 }
1715
1716 static int complete_pio(struct kvm_vcpu *vcpu)
1717 {
1718 struct kvm_pio_request *io = &vcpu->pio;
1719 long delta;
1720 int r;
1721
1722 kvm_arch_ops->cache_regs(vcpu);
1723
1724 if (!io->string) {
1725 if (io->in)
1726 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1727 io->size);
1728 } else {
1729 if (io->in) {
1730 r = pio_copy_data(vcpu);
1731 if (r) {
1732 kvm_arch_ops->cache_regs(vcpu);
1733 return r;
1734 }
1735 }
1736
1737 delta = 1;
1738 if (io->rep) {
1739 delta *= io->cur_count;
1740 /*
1741 * The size of the register should really depend on
1742 * current address size.
1743 */
1744 vcpu->regs[VCPU_REGS_RCX] -= delta;
1745 }
1746 if (io->down)
1747 delta = -delta;
1748 delta *= io->size;
1749 if (io->in)
1750 vcpu->regs[VCPU_REGS_RDI] += delta;
1751 else
1752 vcpu->regs[VCPU_REGS_RSI] += delta;
1753 }
1754
1755 kvm_arch_ops->decache_regs(vcpu);
1756
1757 io->count -= io->cur_count;
1758 io->cur_count = 0;
1759
1760 return 0;
1761 }
1762
1763 void kernel_pio(struct kvm_io_device *pio_dev, struct kvm_vcpu *vcpu)
1764 {
1765 /* TODO: String I/O for in kernel device */
1766
1767 if (vcpu->pio.in)
1768 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1769 vcpu->pio.size,
1770 vcpu->pio_data);
1771 else
1772 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1773 vcpu->pio.size,
1774 vcpu->pio_data);
1775 }
1776
1777 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1778 int size, unsigned long count, int string, int down,
1779 gva_t address, int rep, unsigned port)
1780 {
1781 unsigned now, in_page;
1782 int i;
1783 int nr_pages = 1;
1784 struct page *page;
1785 struct kvm_io_device *pio_dev;
1786
1787 vcpu->run->exit_reason = KVM_EXIT_IO;
1788 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1789 vcpu->run->io.size = size;
1790 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1791 vcpu->run->io.count = count;
1792 vcpu->run->io.port = port;
1793 vcpu->pio.count = count;
1794 vcpu->pio.cur_count = count;
1795 vcpu->pio.size = size;
1796 vcpu->pio.in = in;
1797 vcpu->pio.port = port;
1798 vcpu->pio.string = string;
1799 vcpu->pio.down = down;
1800 vcpu->pio.guest_page_offset = offset_in_page(address);
1801 vcpu->pio.rep = rep;
1802
1803 pio_dev = vcpu_find_pio_dev(vcpu, port);
1804 if (!string) {
1805 kvm_arch_ops->skip_emulated_instruction(vcpu);
1806 kvm_arch_ops->cache_regs(vcpu);
1807 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1808 kvm_arch_ops->decache_regs(vcpu);
1809 if (pio_dev) {
1810 kernel_pio(pio_dev, vcpu);
1811 complete_pio(vcpu);
1812 return 1;
1813 }
1814 return 0;
1815 }
1816 /* TODO: String I/O for in kernel device */
1817 if (pio_dev)
1818 printk(KERN_ERR "kvm_setup_pio: no string io support\n");
1819
1820 if (!count) {
1821 kvm_arch_ops->skip_emulated_instruction(vcpu);
1822 return 1;
1823 }
1824
1825 now = min(count, PAGE_SIZE / size);
1826
1827 if (!down)
1828 in_page = PAGE_SIZE - offset_in_page(address);
1829 else
1830 in_page = offset_in_page(address) + size;
1831 now = min(count, (unsigned long)in_page / size);
1832 if (!now) {
1833 /*
1834 * String I/O straddles page boundary. Pin two guest pages
1835 * so that we satisfy atomicity constraints. Do just one
1836 * transaction to avoid complexity.
1837 */
1838 nr_pages = 2;
1839 now = 1;
1840 }
1841 if (down) {
1842 /*
1843 * String I/O in reverse. Yuck. Kill the guest, fix later.
1844 */
1845 printk(KERN_ERR "kvm: guest string pio down\n");
1846 inject_gp(vcpu);
1847 return 1;
1848 }
1849 vcpu->run->io.count = now;
1850 vcpu->pio.cur_count = now;
1851
1852 if (now == count)
1853 kvm_arch_ops->skip_emulated_instruction(vcpu);
1854
1855 for (i = 0; i < nr_pages; ++i) {
1856 spin_lock(&vcpu->kvm->lock);
1857 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1858 if (page)
1859 get_page(page);
1860 vcpu->pio.guest_pages[i] = page;
1861 spin_unlock(&vcpu->kvm->lock);
1862 if (!page) {
1863 inject_gp(vcpu);
1864 free_pio_guest_pages(vcpu);
1865 return 1;
1866 }
1867 }
1868
1869 if (!vcpu->pio.in)
1870 return pio_copy_data(vcpu);
1871 return 0;
1872 }
1873 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1874
1875 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1876 {
1877 int r;
1878 sigset_t sigsaved;
1879
1880 vcpu_load(vcpu);
1881
1882 if (vcpu->sigset_active)
1883 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1884
1885 /* re-sync apic's tpr */
1886 vcpu->cr8 = kvm_run->cr8;
1887
1888 if (vcpu->pio.cur_count) {
1889 r = complete_pio(vcpu);
1890 if (r)
1891 goto out;
1892 }
1893
1894 if (vcpu->mmio_needed) {
1895 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1896 vcpu->mmio_read_completed = 1;
1897 vcpu->mmio_needed = 0;
1898 r = emulate_instruction(vcpu, kvm_run,
1899 vcpu->mmio_fault_cr2, 0);
1900 if (r == EMULATE_DO_MMIO) {
1901 /*
1902 * Read-modify-write. Back to userspace.
1903 */
1904 kvm_run->exit_reason = KVM_EXIT_MMIO;
1905 r = 0;
1906 goto out;
1907 }
1908 }
1909
1910 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1911 kvm_arch_ops->cache_regs(vcpu);
1912 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1913 kvm_arch_ops->decache_regs(vcpu);
1914 }
1915
1916 r = kvm_arch_ops->run(vcpu, kvm_run);
1917
1918 out:
1919 if (vcpu->sigset_active)
1920 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1921
1922 vcpu_put(vcpu);
1923 return r;
1924 }
1925
1926 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1927 struct kvm_regs *regs)
1928 {
1929 vcpu_load(vcpu);
1930
1931 kvm_arch_ops->cache_regs(vcpu);
1932
1933 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1934 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1935 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1936 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1937 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1938 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1939 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1940 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1941 #ifdef CONFIG_X86_64
1942 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1943 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1944 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1945 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1946 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1947 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1948 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1949 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1950 #endif
1951
1952 regs->rip = vcpu->rip;
1953 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1954
1955 /*
1956 * Don't leak debug flags in case they were set for guest debugging
1957 */
1958 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1959 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1960
1961 vcpu_put(vcpu);
1962
1963 return 0;
1964 }
1965
1966 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1967 struct kvm_regs *regs)
1968 {
1969 vcpu_load(vcpu);
1970
1971 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1972 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1973 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1974 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1975 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1976 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1977 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1978 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1979 #ifdef CONFIG_X86_64
1980 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1981 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1982 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1983 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1984 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1985 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1986 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1987 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1988 #endif
1989
1990 vcpu->rip = regs->rip;
1991 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1992
1993 kvm_arch_ops->decache_regs(vcpu);
1994
1995 vcpu_put(vcpu);
1996
1997 return 0;
1998 }
1999
2000 static void get_segment(struct kvm_vcpu *vcpu,
2001 struct kvm_segment *var, int seg)
2002 {
2003 return kvm_arch_ops->get_segment(vcpu, var, seg);
2004 }
2005
2006 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2007 struct kvm_sregs *sregs)
2008 {
2009 struct descriptor_table dt;
2010
2011 vcpu_load(vcpu);
2012
2013 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2014 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2015 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2016 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2017 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2018 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2019
2020 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2021 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2022
2023 kvm_arch_ops->get_idt(vcpu, &dt);
2024 sregs->idt.limit = dt.limit;
2025 sregs->idt.base = dt.base;
2026 kvm_arch_ops->get_gdt(vcpu, &dt);
2027 sregs->gdt.limit = dt.limit;
2028 sregs->gdt.base = dt.base;
2029
2030 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2031 sregs->cr0 = vcpu->cr0;
2032 sregs->cr2 = vcpu->cr2;
2033 sregs->cr3 = vcpu->cr3;
2034 sregs->cr4 = vcpu->cr4;
2035 sregs->cr8 = vcpu->cr8;
2036 sregs->efer = vcpu->shadow_efer;
2037 sregs->apic_base = vcpu->apic_base;
2038
2039 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2040 sizeof sregs->interrupt_bitmap);
2041
2042 vcpu_put(vcpu);
2043
2044 return 0;
2045 }
2046
2047 static void set_segment(struct kvm_vcpu *vcpu,
2048 struct kvm_segment *var, int seg)
2049 {
2050 return kvm_arch_ops->set_segment(vcpu, var, seg);
2051 }
2052
2053 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2054 struct kvm_sregs *sregs)
2055 {
2056 int mmu_reset_needed = 0;
2057 int i;
2058 struct descriptor_table dt;
2059
2060 vcpu_load(vcpu);
2061
2062 dt.limit = sregs->idt.limit;
2063 dt.base = sregs->idt.base;
2064 kvm_arch_ops->set_idt(vcpu, &dt);
2065 dt.limit = sregs->gdt.limit;
2066 dt.base = sregs->gdt.base;
2067 kvm_arch_ops->set_gdt(vcpu, &dt);
2068
2069 vcpu->cr2 = sregs->cr2;
2070 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2071 vcpu->cr3 = sregs->cr3;
2072
2073 vcpu->cr8 = sregs->cr8;
2074
2075 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2076 #ifdef CONFIG_X86_64
2077 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2078 #endif
2079 vcpu->apic_base = sregs->apic_base;
2080
2081 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2082
2083 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2084 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2085
2086 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2087 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2088 if (!is_long_mode(vcpu) && is_pae(vcpu))
2089 load_pdptrs(vcpu, vcpu->cr3);
2090
2091 if (mmu_reset_needed)
2092 kvm_mmu_reset_context(vcpu);
2093
2094 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2095 sizeof vcpu->irq_pending);
2096 vcpu->irq_summary = 0;
2097 for (i = 0; i < NR_IRQ_WORDS; ++i)
2098 if (vcpu->irq_pending[i])
2099 __set_bit(i, &vcpu->irq_summary);
2100
2101 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2102 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2103 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2104 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2105 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2106 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2107
2108 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2109 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2110
2111 vcpu_put(vcpu);
2112
2113 return 0;
2114 }
2115
2116 /*
2117 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2118 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2119 *
2120 * This list is modified at module load time to reflect the
2121 * capabilities of the host cpu.
2122 */
2123 static u32 msrs_to_save[] = {
2124 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2125 MSR_K6_STAR,
2126 #ifdef CONFIG_X86_64
2127 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2128 #endif
2129 MSR_IA32_TIME_STAMP_COUNTER,
2130 };
2131
2132 static unsigned num_msrs_to_save;
2133
2134 static u32 emulated_msrs[] = {
2135 MSR_IA32_MISC_ENABLE,
2136 };
2137
2138 static __init void kvm_init_msr_list(void)
2139 {
2140 u32 dummy[2];
2141 unsigned i, j;
2142
2143 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2144 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2145 continue;
2146 if (j < i)
2147 msrs_to_save[j] = msrs_to_save[i];
2148 j++;
2149 }
2150 num_msrs_to_save = j;
2151 }
2152
2153 /*
2154 * Adapt set_msr() to msr_io()'s calling convention
2155 */
2156 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2157 {
2158 return kvm_set_msr(vcpu, index, *data);
2159 }
2160
2161 /*
2162 * Read or write a bunch of msrs. All parameters are kernel addresses.
2163 *
2164 * @return number of msrs set successfully.
2165 */
2166 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2167 struct kvm_msr_entry *entries,
2168 int (*do_msr)(struct kvm_vcpu *vcpu,
2169 unsigned index, u64 *data))
2170 {
2171 int i;
2172
2173 vcpu_load(vcpu);
2174
2175 for (i = 0; i < msrs->nmsrs; ++i)
2176 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2177 break;
2178
2179 vcpu_put(vcpu);
2180
2181 return i;
2182 }
2183
2184 /*
2185 * Read or write a bunch of msrs. Parameters are user addresses.
2186 *
2187 * @return number of msrs set successfully.
2188 */
2189 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2190 int (*do_msr)(struct kvm_vcpu *vcpu,
2191 unsigned index, u64 *data),
2192 int writeback)
2193 {
2194 struct kvm_msrs msrs;
2195 struct kvm_msr_entry *entries;
2196 int r, n;
2197 unsigned size;
2198
2199 r = -EFAULT;
2200 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2201 goto out;
2202
2203 r = -E2BIG;
2204 if (msrs.nmsrs >= MAX_IO_MSRS)
2205 goto out;
2206
2207 r = -ENOMEM;
2208 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2209 entries = vmalloc(size);
2210 if (!entries)
2211 goto out;
2212
2213 r = -EFAULT;
2214 if (copy_from_user(entries, user_msrs->entries, size))
2215 goto out_free;
2216
2217 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2218 if (r < 0)
2219 goto out_free;
2220
2221 r = -EFAULT;
2222 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2223 goto out_free;
2224
2225 r = n;
2226
2227 out_free:
2228 vfree(entries);
2229 out:
2230 return r;
2231 }
2232
2233 /*
2234 * Translate a guest virtual address to a guest physical address.
2235 */
2236 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2237 struct kvm_translation *tr)
2238 {
2239 unsigned long vaddr = tr->linear_address;
2240 gpa_t gpa;
2241
2242 vcpu_load(vcpu);
2243 spin_lock(&vcpu->kvm->lock);
2244 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2245 tr->physical_address = gpa;
2246 tr->valid = gpa != UNMAPPED_GVA;
2247 tr->writeable = 1;
2248 tr->usermode = 0;
2249 spin_unlock(&vcpu->kvm->lock);
2250 vcpu_put(vcpu);
2251
2252 return 0;
2253 }
2254
2255 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2256 struct kvm_interrupt *irq)
2257 {
2258 if (irq->irq < 0 || irq->irq >= 256)
2259 return -EINVAL;
2260 vcpu_load(vcpu);
2261
2262 set_bit(irq->irq, vcpu->irq_pending);
2263 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2264
2265 vcpu_put(vcpu);
2266
2267 return 0;
2268 }
2269
2270 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2271 struct kvm_debug_guest *dbg)
2272 {
2273 int r;
2274
2275 vcpu_load(vcpu);
2276
2277 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2278
2279 vcpu_put(vcpu);
2280
2281 return r;
2282 }
2283
2284 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2285 unsigned long address,
2286 int *type)
2287 {
2288 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2289 unsigned long pgoff;
2290 struct page *page;
2291
2292 *type = VM_FAULT_MINOR;
2293 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2294 if (pgoff == 0)
2295 page = virt_to_page(vcpu->run);
2296 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2297 page = virt_to_page(vcpu->pio_data);
2298 else
2299 return NOPAGE_SIGBUS;
2300 get_page(page);
2301 return page;
2302 }
2303
2304 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2305 .nopage = kvm_vcpu_nopage,
2306 };
2307
2308 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2309 {
2310 vma->vm_ops = &kvm_vcpu_vm_ops;
2311 return 0;
2312 }
2313
2314 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2315 {
2316 struct kvm_vcpu *vcpu = filp->private_data;
2317
2318 fput(vcpu->kvm->filp);
2319 return 0;
2320 }
2321
2322 static struct file_operations kvm_vcpu_fops = {
2323 .release = kvm_vcpu_release,
2324 .unlocked_ioctl = kvm_vcpu_ioctl,
2325 .compat_ioctl = kvm_vcpu_ioctl,
2326 .mmap = kvm_vcpu_mmap,
2327 };
2328
2329 /*
2330 * Allocates an inode for the vcpu.
2331 */
2332 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2333 {
2334 int fd, r;
2335 struct inode *inode;
2336 struct file *file;
2337
2338 r = anon_inode_getfd(&fd, &inode, &file,
2339 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2340 if (r)
2341 return r;
2342 atomic_inc(&vcpu->kvm->filp->f_count);
2343 return fd;
2344 }
2345
2346 /*
2347 * Creates some virtual cpus. Good luck creating more than one.
2348 */
2349 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2350 {
2351 int r;
2352 struct kvm_vcpu *vcpu;
2353 struct page *page;
2354
2355 r = -EINVAL;
2356 if (!valid_vcpu(n))
2357 goto out;
2358
2359 vcpu = &kvm->vcpus[n];
2360
2361 mutex_lock(&vcpu->mutex);
2362
2363 if (vcpu->vmcs) {
2364 mutex_unlock(&vcpu->mutex);
2365 return -EEXIST;
2366 }
2367
2368 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2369 r = -ENOMEM;
2370 if (!page)
2371 goto out_unlock;
2372 vcpu->run = page_address(page);
2373
2374 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2375 r = -ENOMEM;
2376 if (!page)
2377 goto out_free_run;
2378 vcpu->pio_data = page_address(page);
2379
2380 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2381 FX_IMAGE_ALIGN);
2382 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2383 vcpu->cr0 = 0x10;
2384
2385 r = kvm_arch_ops->vcpu_create(vcpu);
2386 if (r < 0)
2387 goto out_free_vcpus;
2388
2389 r = kvm_mmu_create(vcpu);
2390 if (r < 0)
2391 goto out_free_vcpus;
2392
2393 kvm_arch_ops->vcpu_load(vcpu);
2394 r = kvm_mmu_setup(vcpu);
2395 if (r >= 0)
2396 r = kvm_arch_ops->vcpu_setup(vcpu);
2397 vcpu_put(vcpu);
2398
2399 if (r < 0)
2400 goto out_free_vcpus;
2401
2402 r = create_vcpu_fd(vcpu);
2403 if (r < 0)
2404 goto out_free_vcpus;
2405
2406 spin_lock(&kvm_lock);
2407 if (n >= kvm->nvcpus)
2408 kvm->nvcpus = n + 1;
2409 spin_unlock(&kvm_lock);
2410
2411 return r;
2412
2413 out_free_vcpus:
2414 kvm_free_vcpu(vcpu);
2415 out_free_run:
2416 free_page((unsigned long)vcpu->run);
2417 vcpu->run = NULL;
2418 out_unlock:
2419 mutex_unlock(&vcpu->mutex);
2420 out:
2421 return r;
2422 }
2423
2424 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2425 {
2426 u64 efer;
2427 int i;
2428 struct kvm_cpuid_entry *e, *entry;
2429
2430 rdmsrl(MSR_EFER, efer);
2431 entry = NULL;
2432 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2433 e = &vcpu->cpuid_entries[i];
2434 if (e->function == 0x80000001) {
2435 entry = e;
2436 break;
2437 }
2438 }
2439 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2440 entry->edx &= ~(1 << 20);
2441 printk(KERN_INFO "kvm: guest NX capability removed\n");
2442 }
2443 }
2444
2445 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2446 struct kvm_cpuid *cpuid,
2447 struct kvm_cpuid_entry __user *entries)
2448 {
2449 int r;
2450
2451 r = -E2BIG;
2452 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2453 goto out;
2454 r = -EFAULT;
2455 if (copy_from_user(&vcpu->cpuid_entries, entries,
2456 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2457 goto out;
2458 vcpu->cpuid_nent = cpuid->nent;
2459 cpuid_fix_nx_cap(vcpu);
2460 return 0;
2461
2462 out:
2463 return r;
2464 }
2465
2466 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2467 {
2468 if (sigset) {
2469 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2470 vcpu->sigset_active = 1;
2471 vcpu->sigset = *sigset;
2472 } else
2473 vcpu->sigset_active = 0;
2474 return 0;
2475 }
2476
2477 /*
2478 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2479 * we have asm/x86/processor.h
2480 */
2481 struct fxsave {
2482 u16 cwd;
2483 u16 swd;
2484 u16 twd;
2485 u16 fop;
2486 u64 rip;
2487 u64 rdp;
2488 u32 mxcsr;
2489 u32 mxcsr_mask;
2490 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2491 #ifdef CONFIG_X86_64
2492 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2493 #else
2494 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2495 #endif
2496 };
2497
2498 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2499 {
2500 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2501
2502 vcpu_load(vcpu);
2503
2504 memcpy(fpu->fpr, fxsave->st_space, 128);
2505 fpu->fcw = fxsave->cwd;
2506 fpu->fsw = fxsave->swd;
2507 fpu->ftwx = fxsave->twd;
2508 fpu->last_opcode = fxsave->fop;
2509 fpu->last_ip = fxsave->rip;
2510 fpu->last_dp = fxsave->rdp;
2511 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2512
2513 vcpu_put(vcpu);
2514
2515 return 0;
2516 }
2517
2518 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2519 {
2520 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2521
2522 vcpu_load(vcpu);
2523
2524 memcpy(fxsave->st_space, fpu->fpr, 128);
2525 fxsave->cwd = fpu->fcw;
2526 fxsave->swd = fpu->fsw;
2527 fxsave->twd = fpu->ftwx;
2528 fxsave->fop = fpu->last_opcode;
2529 fxsave->rip = fpu->last_ip;
2530 fxsave->rdp = fpu->last_dp;
2531 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2532
2533 vcpu_put(vcpu);
2534
2535 return 0;
2536 }
2537
2538 static long kvm_vcpu_ioctl(struct file *filp,
2539 unsigned int ioctl, unsigned long arg)
2540 {
2541 struct kvm_vcpu *vcpu = filp->private_data;
2542 void __user *argp = (void __user *)arg;
2543 int r = -EINVAL;
2544
2545 switch (ioctl) {
2546 case KVM_RUN:
2547 r = -EINVAL;
2548 if (arg)
2549 goto out;
2550 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2551 break;
2552 case KVM_GET_REGS: {
2553 struct kvm_regs kvm_regs;
2554
2555 memset(&kvm_regs, 0, sizeof kvm_regs);
2556 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2557 if (r)
2558 goto out;
2559 r = -EFAULT;
2560 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2561 goto out;
2562 r = 0;
2563 break;
2564 }
2565 case KVM_SET_REGS: {
2566 struct kvm_regs kvm_regs;
2567
2568 r = -EFAULT;
2569 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2570 goto out;
2571 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2572 if (r)
2573 goto out;
2574 r = 0;
2575 break;
2576 }
2577 case KVM_GET_SREGS: {
2578 struct kvm_sregs kvm_sregs;
2579
2580 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2581 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2582 if (r)
2583 goto out;
2584 r = -EFAULT;
2585 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2586 goto out;
2587 r = 0;
2588 break;
2589 }
2590 case KVM_SET_SREGS: {
2591 struct kvm_sregs kvm_sregs;
2592
2593 r = -EFAULT;
2594 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2595 goto out;
2596 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2597 if (r)
2598 goto out;
2599 r = 0;
2600 break;
2601 }
2602 case KVM_TRANSLATE: {
2603 struct kvm_translation tr;
2604
2605 r = -EFAULT;
2606 if (copy_from_user(&tr, argp, sizeof tr))
2607 goto out;
2608 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2609 if (r)
2610 goto out;
2611 r = -EFAULT;
2612 if (copy_to_user(argp, &tr, sizeof tr))
2613 goto out;
2614 r = 0;
2615 break;
2616 }
2617 case KVM_INTERRUPT: {
2618 struct kvm_interrupt irq;
2619
2620 r = -EFAULT;
2621 if (copy_from_user(&irq, argp, sizeof irq))
2622 goto out;
2623 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2624 if (r)
2625 goto out;
2626 r = 0;
2627 break;
2628 }
2629 case KVM_DEBUG_GUEST: {
2630 struct kvm_debug_guest dbg;
2631
2632 r = -EFAULT;
2633 if (copy_from_user(&dbg, argp, sizeof dbg))
2634 goto out;
2635 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2636 if (r)
2637 goto out;
2638 r = 0;
2639 break;
2640 }
2641 case KVM_GET_MSRS:
2642 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2643 break;
2644 case KVM_SET_MSRS:
2645 r = msr_io(vcpu, argp, do_set_msr, 0);
2646 break;
2647 case KVM_SET_CPUID: {
2648 struct kvm_cpuid __user *cpuid_arg = argp;
2649 struct kvm_cpuid cpuid;
2650
2651 r = -EFAULT;
2652 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2653 goto out;
2654 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2655 if (r)
2656 goto out;
2657 break;
2658 }
2659 case KVM_SET_SIGNAL_MASK: {
2660 struct kvm_signal_mask __user *sigmask_arg = argp;
2661 struct kvm_signal_mask kvm_sigmask;
2662 sigset_t sigset, *p;
2663
2664 p = NULL;
2665 if (argp) {
2666 r = -EFAULT;
2667 if (copy_from_user(&kvm_sigmask, argp,
2668 sizeof kvm_sigmask))
2669 goto out;
2670 r = -EINVAL;
2671 if (kvm_sigmask.len != sizeof sigset)
2672 goto out;
2673 r = -EFAULT;
2674 if (copy_from_user(&sigset, sigmask_arg->sigset,
2675 sizeof sigset))
2676 goto out;
2677 p = &sigset;
2678 }
2679 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2680 break;
2681 }
2682 case KVM_GET_FPU: {
2683 struct kvm_fpu fpu;
2684
2685 memset(&fpu, 0, sizeof fpu);
2686 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2687 if (r)
2688 goto out;
2689 r = -EFAULT;
2690 if (copy_to_user(argp, &fpu, sizeof fpu))
2691 goto out;
2692 r = 0;
2693 break;
2694 }
2695 case KVM_SET_FPU: {
2696 struct kvm_fpu fpu;
2697
2698 r = -EFAULT;
2699 if (copy_from_user(&fpu, argp, sizeof fpu))
2700 goto out;
2701 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2702 if (r)
2703 goto out;
2704 r = 0;
2705 break;
2706 }
2707 default:
2708 ;
2709 }
2710 out:
2711 return r;
2712 }
2713
2714 static long kvm_vm_ioctl(struct file *filp,
2715 unsigned int ioctl, unsigned long arg)
2716 {
2717 struct kvm *kvm = filp->private_data;
2718 void __user *argp = (void __user *)arg;
2719 int r = -EINVAL;
2720
2721 switch (ioctl) {
2722 case KVM_CREATE_VCPU:
2723 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2724 if (r < 0)
2725 goto out;
2726 break;
2727 case KVM_SET_MEMORY_REGION: {
2728 struct kvm_memory_region kvm_mem;
2729
2730 r = -EFAULT;
2731 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2732 goto out;
2733 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2734 if (r)
2735 goto out;
2736 break;
2737 }
2738 case KVM_GET_DIRTY_LOG: {
2739 struct kvm_dirty_log log;
2740
2741 r = -EFAULT;
2742 if (copy_from_user(&log, argp, sizeof log))
2743 goto out;
2744 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2745 if (r)
2746 goto out;
2747 break;
2748 }
2749 case KVM_SET_MEMORY_ALIAS: {
2750 struct kvm_memory_alias alias;
2751
2752 r = -EFAULT;
2753 if (copy_from_user(&alias, argp, sizeof alias))
2754 goto out;
2755 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2756 if (r)
2757 goto out;
2758 break;
2759 }
2760 default:
2761 ;
2762 }
2763 out:
2764 return r;
2765 }
2766
2767 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2768 unsigned long address,
2769 int *type)
2770 {
2771 struct kvm *kvm = vma->vm_file->private_data;
2772 unsigned long pgoff;
2773 struct page *page;
2774
2775 *type = VM_FAULT_MINOR;
2776 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2777 page = gfn_to_page(kvm, pgoff);
2778 if (!page)
2779 return NOPAGE_SIGBUS;
2780 get_page(page);
2781 return page;
2782 }
2783
2784 static struct vm_operations_struct kvm_vm_vm_ops = {
2785 .nopage = kvm_vm_nopage,
2786 };
2787
2788 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2789 {
2790 vma->vm_ops = &kvm_vm_vm_ops;
2791 return 0;
2792 }
2793
2794 static struct file_operations kvm_vm_fops = {
2795 .release = kvm_vm_release,
2796 .unlocked_ioctl = kvm_vm_ioctl,
2797 .compat_ioctl = kvm_vm_ioctl,
2798 .mmap = kvm_vm_mmap,
2799 };
2800
2801 static int kvm_dev_ioctl_create_vm(void)
2802 {
2803 int fd, r;
2804 struct inode *inode;
2805 struct file *file;
2806 struct kvm *kvm;
2807
2808 kvm = kvm_create_vm();
2809 if (IS_ERR(kvm))
2810 return PTR_ERR(kvm);
2811 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2812 if (r) {
2813 kvm_destroy_vm(kvm);
2814 return r;
2815 }
2816
2817 kvm->filp = file;
2818
2819 return fd;
2820 }
2821
2822 static long kvm_dev_ioctl(struct file *filp,
2823 unsigned int ioctl, unsigned long arg)
2824 {
2825 void __user *argp = (void __user *)arg;
2826 long r = -EINVAL;
2827
2828 switch (ioctl) {
2829 case KVM_GET_API_VERSION:
2830 r = -EINVAL;
2831 if (arg)
2832 goto out;
2833 r = KVM_API_VERSION;
2834 break;
2835 case KVM_CREATE_VM:
2836 r = -EINVAL;
2837 if (arg)
2838 goto out;
2839 r = kvm_dev_ioctl_create_vm();
2840 break;
2841 case KVM_GET_MSR_INDEX_LIST: {
2842 struct kvm_msr_list __user *user_msr_list = argp;
2843 struct kvm_msr_list msr_list;
2844 unsigned n;
2845
2846 r = -EFAULT;
2847 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2848 goto out;
2849 n = msr_list.nmsrs;
2850 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2851 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2852 goto out;
2853 r = -E2BIG;
2854 if (n < num_msrs_to_save)
2855 goto out;
2856 r = -EFAULT;
2857 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2858 num_msrs_to_save * sizeof(u32)))
2859 goto out;
2860 if (copy_to_user(user_msr_list->indices
2861 + num_msrs_to_save * sizeof(u32),
2862 &emulated_msrs,
2863 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2864 goto out;
2865 r = 0;
2866 break;
2867 }
2868 case KVM_CHECK_EXTENSION:
2869 /*
2870 * No extensions defined at present.
2871 */
2872 r = 0;
2873 break;
2874 case KVM_GET_VCPU_MMAP_SIZE:
2875 r = -EINVAL;
2876 if (arg)
2877 goto out;
2878 r = 2 * PAGE_SIZE;
2879 break;
2880 default:
2881 ;
2882 }
2883 out:
2884 return r;
2885 }
2886
2887 static struct file_operations kvm_chardev_ops = {
2888 .open = kvm_dev_open,
2889 .release = kvm_dev_release,
2890 .unlocked_ioctl = kvm_dev_ioctl,
2891 .compat_ioctl = kvm_dev_ioctl,
2892 };
2893
2894 static struct miscdevice kvm_dev = {
2895 KVM_MINOR,
2896 "kvm",
2897 &kvm_chardev_ops,
2898 };
2899
2900 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2901 void *v)
2902 {
2903 if (val == SYS_RESTART) {
2904 /*
2905 * Some (well, at least mine) BIOSes hang on reboot if
2906 * in vmx root mode.
2907 */
2908 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2909 on_each_cpu(hardware_disable, NULL, 0, 1);
2910 }
2911 return NOTIFY_OK;
2912 }
2913
2914 static struct notifier_block kvm_reboot_notifier = {
2915 .notifier_call = kvm_reboot,
2916 .priority = 0,
2917 };
2918
2919 /*
2920 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2921 * cached on it.
2922 */
2923 static void decache_vcpus_on_cpu(int cpu)
2924 {
2925 struct kvm *vm;
2926 struct kvm_vcpu *vcpu;
2927 int i;
2928
2929 spin_lock(&kvm_lock);
2930 list_for_each_entry(vm, &vm_list, vm_list)
2931 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2932 vcpu = &vm->vcpus[i];
2933 /*
2934 * If the vcpu is locked, then it is running on some
2935 * other cpu and therefore it is not cached on the
2936 * cpu in question.
2937 *
2938 * If it's not locked, check the last cpu it executed
2939 * on.
2940 */
2941 if (mutex_trylock(&vcpu->mutex)) {
2942 if (vcpu->cpu == cpu) {
2943 kvm_arch_ops->vcpu_decache(vcpu);
2944 vcpu->cpu = -1;
2945 }
2946 mutex_unlock(&vcpu->mutex);
2947 }
2948 }
2949 spin_unlock(&kvm_lock);
2950 }
2951
2952 static void hardware_enable(void *junk)
2953 {
2954 int cpu = raw_smp_processor_id();
2955
2956 if (cpu_isset(cpu, cpus_hardware_enabled))
2957 return;
2958 cpu_set(cpu, cpus_hardware_enabled);
2959 kvm_arch_ops->hardware_enable(NULL);
2960 }
2961
2962 static void hardware_disable(void *junk)
2963 {
2964 int cpu = raw_smp_processor_id();
2965
2966 if (!cpu_isset(cpu, cpus_hardware_enabled))
2967 return;
2968 cpu_clear(cpu, cpus_hardware_enabled);
2969 decache_vcpus_on_cpu(cpu);
2970 kvm_arch_ops->hardware_disable(NULL);
2971 }
2972
2973 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2974 void *v)
2975 {
2976 int cpu = (long)v;
2977
2978 switch (val) {
2979 case CPU_DYING:
2980 case CPU_DYING_FROZEN:
2981 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2982 cpu);
2983 hardware_disable(NULL);
2984 break;
2985 case CPU_UP_CANCELED:
2986 case CPU_UP_CANCELED_FROZEN:
2987 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2988 cpu);
2989 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2990 break;
2991 case CPU_ONLINE:
2992 case CPU_ONLINE_FROZEN:
2993 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2994 cpu);
2995 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2996 break;
2997 }
2998 return NOTIFY_OK;
2999 }
3000
3001 void kvm_io_bus_init(struct kvm_io_bus *bus)
3002 {
3003 memset(bus, 0, sizeof(*bus));
3004 }
3005
3006 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3007 {
3008 int i;
3009
3010 for (i = 0; i < bus->dev_count; i++) {
3011 struct kvm_io_device *pos = bus->devs[i];
3012
3013 kvm_iodevice_destructor(pos);
3014 }
3015 }
3016
3017 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3018 {
3019 int i;
3020
3021 for (i = 0; i < bus->dev_count; i++) {
3022 struct kvm_io_device *pos = bus->devs[i];
3023
3024 if (pos->in_range(pos, addr))
3025 return pos;
3026 }
3027
3028 return NULL;
3029 }
3030
3031 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3032 {
3033 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3034
3035 bus->devs[bus->dev_count++] = dev;
3036 }
3037
3038 static struct notifier_block kvm_cpu_notifier = {
3039 .notifier_call = kvm_cpu_hotplug,
3040 .priority = 20, /* must be > scheduler priority */
3041 };
3042
3043 static u64 stat_get(void *_offset)
3044 {
3045 unsigned offset = (long)_offset;
3046 u64 total = 0;
3047 struct kvm *kvm;
3048 struct kvm_vcpu *vcpu;
3049 int i;
3050
3051 spin_lock(&kvm_lock);
3052 list_for_each_entry(kvm, &vm_list, vm_list)
3053 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3054 vcpu = &kvm->vcpus[i];
3055 total += *(u32 *)((void *)vcpu + offset);
3056 }
3057 spin_unlock(&kvm_lock);
3058 return total;
3059 }
3060
3061 static void stat_set(void *offset, u64 val)
3062 {
3063 }
3064
3065 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3066
3067 static __init void kvm_init_debug(void)
3068 {
3069 struct kvm_stats_debugfs_item *p;
3070
3071 debugfs_dir = debugfs_create_dir("kvm", NULL);
3072 for (p = debugfs_entries; p->name; ++p)
3073 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3074 (void *)(long)p->offset,
3075 &stat_fops);
3076 }
3077
3078 static void kvm_exit_debug(void)
3079 {
3080 struct kvm_stats_debugfs_item *p;
3081
3082 for (p = debugfs_entries; p->name; ++p)
3083 debugfs_remove(p->dentry);
3084 debugfs_remove(debugfs_dir);
3085 }
3086
3087 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3088 {
3089 hardware_disable(NULL);
3090 return 0;
3091 }
3092
3093 static int kvm_resume(struct sys_device *dev)
3094 {
3095 hardware_enable(NULL);
3096 return 0;
3097 }
3098
3099 static struct sysdev_class kvm_sysdev_class = {
3100 set_kset_name("kvm"),
3101 .suspend = kvm_suspend,
3102 .resume = kvm_resume,
3103 };
3104
3105 static struct sys_device kvm_sysdev = {
3106 .id = 0,
3107 .cls = &kvm_sysdev_class,
3108 };
3109
3110 hpa_t bad_page_address;
3111
3112 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3113 {
3114 int r;
3115
3116 if (kvm_arch_ops) {
3117 printk(KERN_ERR "kvm: already loaded the other module\n");
3118 return -EEXIST;
3119 }
3120
3121 if (!ops->cpu_has_kvm_support()) {
3122 printk(KERN_ERR "kvm: no hardware support\n");
3123 return -EOPNOTSUPP;
3124 }
3125 if (ops->disabled_by_bios()) {
3126 printk(KERN_ERR "kvm: disabled by bios\n");
3127 return -EOPNOTSUPP;
3128 }
3129
3130 kvm_arch_ops = ops;
3131
3132 r = kvm_arch_ops->hardware_setup();
3133 if (r < 0)
3134 goto out;
3135
3136 on_each_cpu(hardware_enable, NULL, 0, 1);
3137 r = register_cpu_notifier(&kvm_cpu_notifier);
3138 if (r)
3139 goto out_free_1;
3140 register_reboot_notifier(&kvm_reboot_notifier);
3141
3142 r = sysdev_class_register(&kvm_sysdev_class);
3143 if (r)
3144 goto out_free_2;
3145
3146 r = sysdev_register(&kvm_sysdev);
3147 if (r)
3148 goto out_free_3;
3149
3150 kvm_chardev_ops.owner = module;
3151
3152 r = misc_register(&kvm_dev);
3153 if (r) {
3154 printk (KERN_ERR "kvm: misc device register failed\n");
3155 goto out_free;
3156 }
3157
3158 return r;
3159
3160 out_free:
3161 sysdev_unregister(&kvm_sysdev);
3162 out_free_3:
3163 sysdev_class_unregister(&kvm_sysdev_class);
3164 out_free_2:
3165 unregister_reboot_notifier(&kvm_reboot_notifier);
3166 unregister_cpu_notifier(&kvm_cpu_notifier);
3167 out_free_1:
3168 on_each_cpu(hardware_disable, NULL, 0, 1);
3169 kvm_arch_ops->hardware_unsetup();
3170 out:
3171 kvm_arch_ops = NULL;
3172 return r;
3173 }
3174
3175 void kvm_exit_arch(void)
3176 {
3177 misc_deregister(&kvm_dev);
3178 sysdev_unregister(&kvm_sysdev);
3179 sysdev_class_unregister(&kvm_sysdev_class);
3180 unregister_reboot_notifier(&kvm_reboot_notifier);
3181 unregister_cpu_notifier(&kvm_cpu_notifier);
3182 on_each_cpu(hardware_disable, NULL, 0, 1);
3183 kvm_arch_ops->hardware_unsetup();
3184 kvm_arch_ops = NULL;
3185 }
3186
3187 static __init int kvm_init(void)
3188 {
3189 static struct page *bad_page;
3190 int r;
3191
3192 r = kvm_mmu_module_init();
3193 if (r)
3194 goto out4;
3195
3196 kvm_init_debug();
3197
3198 kvm_init_msr_list();
3199
3200 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3201 r = -ENOMEM;
3202 goto out;
3203 }
3204
3205 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3206 memset(__va(bad_page_address), 0, PAGE_SIZE);
3207
3208 return 0;
3209
3210 out:
3211 kvm_exit_debug();
3212 kvm_mmu_module_exit();
3213 out4:
3214 return r;
3215 }
3216
3217 static __exit void kvm_exit(void)
3218 {
3219 kvm_exit_debug();
3220 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3221 kvm_mmu_module_exit();
3222 }
3223
3224 module_init(kvm_init)
3225 module_exit(kvm_exit)
3226
3227 EXPORT_SYMBOL_GPL(kvm_init_arch);
3228 EXPORT_SYMBOL_GPL(kvm_exit_arch);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree