projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Add fpu_reload counter
[linux-2.6-block.git] / drivers / kvm / x86.c
CommitLineData
043405e1
CO		 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
313a3dc7 17#include "kvm.h"
043405e1 18#include "x86.h"
d825ed0a 19#include "x86_emulate.h"
5fb76f9b 20#include "segment_descriptor.h"
313a3dc7
CO		 21#include "irq.h"
22
23#include <linux/kvm.h>
24#include <linux/fs.h>
25#include <linux/vmalloc.h>
5fb76f9b 26#include <linux/module.h>
043405e1
CO		 27
28#include <asm/uaccess.h>
d825ed0a 29#include <asm/msr.h>
043405e1 30
313a3dc7 31#define MAX_IO_MSRS 256
a03490ed
CO		 32#define CR0_RESERVED_BITS \
33 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
34 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
35 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
36#define CR4_RESERVED_BITS \
37 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
38 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
39 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
40 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
41
42#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
15c4a640 43#define EFER_RESERVED_BITS 0xfffffffffffff2fe
313a3dc7 44
417bc304
HB		 45#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
46
97896d04
ZX		 47struct kvm_x86_ops *kvm_x86_ops;
48
417bc304
HB		 49struct kvm_stats_debugfs_item debugfs_entries[] = {
50 { "pf_fixed", STAT_OFFSET(pf_fixed) },
51 { "pf_guest", STAT_OFFSET(pf_guest) },
52 { "tlb_flush", STAT_OFFSET(tlb_flush) },
53 { "invlpg", STAT_OFFSET(invlpg) },
54 { "exits", STAT_OFFSET(exits) },
55 { "io_exits", STAT_OFFSET(io_exits) },
56 { "mmio_exits", STAT_OFFSET(mmio_exits) },
57 { "signal_exits", STAT_OFFSET(signal_exits) },
58 { "irq_window", STAT_OFFSET(irq_window_exits) },
59 { "halt_exits", STAT_OFFSET(halt_exits) },
60 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
61 { "request_irq", STAT_OFFSET(request_irq_exits) },
62 { "irq_exits", STAT_OFFSET(irq_exits) },
e1beb1d3 63 { "host_state_reload", STAT_OFFSET(host_state_reload) },
417bc304 64 { "efer_reload", STAT_OFFSET(efer_reload) },
f096ed85 65 { "fpu_reload", STAT_OFFSET(fpu_reload) },
417bc304
HB		 66 { NULL }
67};
68
69
5fb76f9b
CO		 70unsigned long segment_base(u16 selector)
71{
72 struct descriptor_table gdt;
73 struct segment_descriptor *d;
74 unsigned long table_base;
75 unsigned long v;
76
77 if (selector == 0)
78 return 0;
79
80 asm("sgdt %0" : "=m"(gdt));
81 table_base = gdt.base;
82
83 if (selector & 4) { /* from ldt */
84 u16 ldt_selector;
85
86 asm("sldt %0" : "=g"(ldt_selector));
87 table_base = segment_base(ldt_selector);
88 }
89 d = (struct segment_descriptor *)(table_base + (selector & ~7));
90 v = d->base_low | ((unsigned long)d->base_mid << 16) |
91 ((unsigned long)d->base_high << 24);
92#ifdef CONFIG_X86_64
93 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
94 v |= ((unsigned long) \
95 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
96#endif
97 return v;
98}
99EXPORT_SYMBOL_GPL(segment_base);
100
6866b83e
CO		 101u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
102{
103 if (irqchip_in_kernel(vcpu->kvm))
104 return vcpu->apic_base;
105 else
106 return vcpu->apic_base;
107}
108EXPORT_SYMBOL_GPL(kvm_get_apic_base);
109
110void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
111{
112 /* TODO: reserve bits check */
113 if (irqchip_in_kernel(vcpu->kvm))
114 kvm_lapic_set_base(vcpu, data);
115 else
116 vcpu->apic_base = data;
117}
118EXPORT_SYMBOL_GPL(kvm_set_apic_base);
119
a03490ed
CO		 120static void inject_gp(struct kvm_vcpu *vcpu)
121{
122 kvm_x86_ops->inject_gp(vcpu, 0);
123}
124
125/*
126 * Load the pae pdptrs. Return true is they are all valid.
127 */
128int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
129{
130 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
131 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
132 int i;
133 int ret;
134 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
135
136 mutex_lock(&vcpu->kvm->lock);
137 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
138 offset * sizeof(u64), sizeof(pdpte));
139 if (ret < 0) {
140 ret = 0;
141 goto out;
142 }
143 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
144 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
145 ret = 0;
146 goto out;
147 }
148 }
149 ret = 1;
150
151 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
152out:
153 mutex_unlock(&vcpu->kvm->lock);
154
155 return ret;
156}
157
158void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
159{
160 if (cr0 & CR0_RESERVED_BITS) {
161 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
162 cr0, vcpu->cr0);
163 inject_gp(vcpu);
164 return;
165 }
166
167 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
168 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
169 inject_gp(vcpu);
170 return;
171 }
172
173 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
174 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
175 "and a clear PE flag\n");
176 inject_gp(vcpu);
177 return;
178 }
179
180 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
181#ifdef CONFIG_X86_64
182 if ((vcpu->shadow_efer & EFER_LME)) {
183 int cs_db, cs_l;
184
185 if (!is_pae(vcpu)) {
186 printk(KERN_DEBUG "set_cr0: #GP, start paging "
187 "in long mode while PAE is disabled\n");
188 inject_gp(vcpu);
189 return;
190 }
191 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
192 if (cs_l) {
193 printk(KERN_DEBUG "set_cr0: #GP, start paging "
194 "in long mode while CS.L == 1\n");
195 inject_gp(vcpu);
196 return;
197
198 }
199 } else
200#endif
201 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
202 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
203 "reserved bits\n");
204 inject_gp(vcpu);
205 return;
206 }
207
208 }
209
210 kvm_x86_ops->set_cr0(vcpu, cr0);
211 vcpu->cr0 = cr0;
212
213 mutex_lock(&vcpu->kvm->lock);
214 kvm_mmu_reset_context(vcpu);
215 mutex_unlock(&vcpu->kvm->lock);
216 return;
217}
218EXPORT_SYMBOL_GPL(set_cr0);
219
220void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
221{
222 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
223}
224EXPORT_SYMBOL_GPL(lmsw);
225
226void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
227{
228 if (cr4 & CR4_RESERVED_BITS) {
229 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
230 inject_gp(vcpu);
231 return;
232 }
233
234 if (is_long_mode(vcpu)) {
235 if (!(cr4 & X86_CR4_PAE)) {
236 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
237 "in long mode\n");
238 inject_gp(vcpu);
239 return;
240 }
241 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
242 && !load_pdptrs(vcpu, vcpu->cr3)) {
243 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
244 inject_gp(vcpu);
245 return;
246 }
247
248 if (cr4 & X86_CR4_VMXE) {
249 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
250 inject_gp(vcpu);
251 return;
252 }
253 kvm_x86_ops->set_cr4(vcpu, cr4);
254 vcpu->cr4 = cr4;
255 mutex_lock(&vcpu->kvm->lock);
256 kvm_mmu_reset_context(vcpu);
257 mutex_unlock(&vcpu->kvm->lock);
258}
259EXPORT_SYMBOL_GPL(set_cr4);
260
261void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
262{
263 if (is_long_mode(vcpu)) {
264 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
265 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
266 inject_gp(vcpu);
267 return;
268 }
269 } else {
270 if (is_pae(vcpu)) {
271 if (cr3 & CR3_PAE_RESERVED_BITS) {
272 printk(KERN_DEBUG
273 "set_cr3: #GP, reserved bits\n");
274 inject_gp(vcpu);
275 return;
276 }
277 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
278 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
279 "reserved bits\n");
280 inject_gp(vcpu);
281 return;
282 }
283 }
284 /*
285 * We don't check reserved bits in nonpae mode, because
286 * this isn't enforced, and VMware depends on this.
287 */
288 }
289
290 mutex_lock(&vcpu->kvm->lock);
291 /*
292 * Does the new cr3 value map to physical memory? (Note, we
293 * catch an invalid cr3 even in real-mode, because it would
294 * cause trouble later on when we turn on paging anyway.)
295 *
296 * A real CPU would silently accept an invalid cr3 and would
297 * attempt to use it - with largely undefined (and often hard
298 * to debug) behavior on the guest side.
299 */
300 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
301 inject_gp(vcpu);
302 else {
303 vcpu->cr3 = cr3;
304 vcpu->mmu.new_cr3(vcpu);
305 }
306 mutex_unlock(&vcpu->kvm->lock);
307}
308EXPORT_SYMBOL_GPL(set_cr3);
309
310void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
311{
312 if (cr8 & CR8_RESERVED_BITS) {
313 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
314 inject_gp(vcpu);
315 return;
316 }
317 if (irqchip_in_kernel(vcpu->kvm))
318 kvm_lapic_set_tpr(vcpu, cr8);
319 else
320 vcpu->cr8 = cr8;
321}
322EXPORT_SYMBOL_GPL(set_cr8);
323
324unsigned long get_cr8(struct kvm_vcpu *vcpu)
325{
326 if (irqchip_in_kernel(vcpu->kvm))
327 return kvm_lapic_get_cr8(vcpu);
328 else
329 return vcpu->cr8;
330}
331EXPORT_SYMBOL_GPL(get_cr8);
332
043405e1
CO		 333/*
334 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
335 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
336 *
337 * This list is modified at module load time to reflect the
338 * capabilities of the host cpu.
339 */
340static u32 msrs_to_save[] = {
341 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
342 MSR_K6_STAR,
343#ifdef CONFIG_X86_64
344 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
345#endif
346 MSR_IA32_TIME_STAMP_COUNTER,
347};
348
349static unsigned num_msrs_to_save;
350
351static u32 emulated_msrs[] = {
352 MSR_IA32_MISC_ENABLE,
353};
354
15c4a640
CO		 355#ifdef CONFIG_X86_64
356
357static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
358{
359 if (efer & EFER_RESERVED_BITS) {
360 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
361 efer);
362 inject_gp(vcpu);
363 return;
364 }
365
366 if (is_paging(vcpu)
367 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
368 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
369 inject_gp(vcpu);
370 return;
371 }
372
373 kvm_x86_ops->set_efer(vcpu, efer);
374
375 efer &= ~EFER_LMA;
376 efer |= vcpu->shadow_efer & EFER_LMA;
377
378 vcpu->shadow_efer = efer;
379}
380
381#endif
382
383/*
384 * Writes msr value into into the appropriate "register".
385 * Returns 0 on success, non-0 otherwise.
386 * Assumes vcpu_load() was already called.
387 */
388int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
389{
390 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
391}
392
313a3dc7
CO		 393/*
394 * Adapt set_msr() to msr_io()'s calling convention
395 */
396static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
397{
398 return kvm_set_msr(vcpu, index, *data);
399}
400
15c4a640
CO		 401
402int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
403{
404 switch (msr) {
405#ifdef CONFIG_X86_64
406 case MSR_EFER:
407 set_efer(vcpu, data);
408 break;
409#endif
410 case MSR_IA32_MC0_STATUS:
411 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
412 __FUNCTION__, data);
413 break;
414 case MSR_IA32_MCG_STATUS:
415 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
416 __FUNCTION__, data);
417 break;
418 case MSR_IA32_UCODE_REV:
419 case MSR_IA32_UCODE_WRITE:
420 case 0x200 ... 0x2ff: /* MTRRs */
421 break;
422 case MSR_IA32_APICBASE:
423 kvm_set_apic_base(vcpu, data);
424 break;
425 case MSR_IA32_MISC_ENABLE:
426 vcpu->ia32_misc_enable_msr = data;
427 break;
428 default:
429 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
430 return 1;
431 }
432 return 0;
433}
434EXPORT_SYMBOL_GPL(kvm_set_msr_common);
435
436
437/*
438 * Reads an msr value (of 'msr_index') into 'pdata'.
439 * Returns 0 on success, non-0 otherwise.
440 * Assumes vcpu_load() was already called.
441 */
442int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
443{
444 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
445}
446
447int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
448{
449 u64 data;
450
451 switch (msr) {
452 case 0xc0010010: /* SYSCFG */
453 case 0xc0010015: /* HWCR */
454 case MSR_IA32_PLATFORM_ID:
455 case MSR_IA32_P5_MC_ADDR:
456 case MSR_IA32_P5_MC_TYPE:
457 case MSR_IA32_MC0_CTL:
458 case MSR_IA32_MCG_STATUS:
459 case MSR_IA32_MCG_CAP:
460 case MSR_IA32_MC0_MISC:
461 case MSR_IA32_MC0_MISC+4:
462 case MSR_IA32_MC0_MISC+8:
463 case MSR_IA32_MC0_MISC+12:
464 case MSR_IA32_MC0_MISC+16:
465 case MSR_IA32_UCODE_REV:
466 case MSR_IA32_PERF_STATUS:
467 case MSR_IA32_EBL_CR_POWERON:
468 /* MTRR registers */
469 case 0xfe:
470 case 0x200 ... 0x2ff:
471 data = 0;
472 break;
473 case 0xcd: /* fsb frequency */
474 data = 3;
475 break;
476 case MSR_IA32_APICBASE:
477 data = kvm_get_apic_base(vcpu);
478 break;
479 case MSR_IA32_MISC_ENABLE:
480 data = vcpu->ia32_misc_enable_msr;
481 break;
482#ifdef CONFIG_X86_64
483 case MSR_EFER:
484 data = vcpu->shadow_efer;
485 break;
486#endif
487 default:
488 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
489 return 1;
490 }
491 *pdata = data;
492 return 0;
493}
494EXPORT_SYMBOL_GPL(kvm_get_msr_common);
495
313a3dc7
CO		 496/*
497 * Read or write a bunch of msrs. All parameters are kernel addresses.
498 *
499 * @return number of msrs set successfully.
500 */
501static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
502 struct kvm_msr_entry *entries,
503 int (*do_msr)(struct kvm_vcpu *vcpu,
504 unsigned index, u64 *data))
505{
506 int i;
507
508 vcpu_load(vcpu);
509
510 for (i = 0; i < msrs->nmsrs; ++i)
511 if (do_msr(vcpu, entries[i].index, &entries[i].data))
512 break;
513
514 vcpu_put(vcpu);
515
516 return i;
517}
518
519/*
520 * Read or write a bunch of msrs. Parameters are user addresses.
521 *
522 * @return number of msrs set successfully.
523 */
524static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
525 int (*do_msr)(struct kvm_vcpu *vcpu,
526 unsigned index, u64 *data),
527 int writeback)
528{
529 struct kvm_msrs msrs;
530 struct kvm_msr_entry *entries;
531 int r, n;
532 unsigned size;
533
534 r = -EFAULT;
535 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
536 goto out;
537
538 r = -E2BIG;
539 if (msrs.nmsrs >= MAX_IO_MSRS)
540 goto out;
541
542 r = -ENOMEM;
543 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
544 entries = vmalloc(size);
545 if (!entries)
546 goto out;
547
548 r = -EFAULT;
549 if (copy_from_user(entries, user_msrs->entries, size))
550 goto out_free;
551
552 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
553 if (r < 0)
554 goto out_free;
555
556 r = -EFAULT;
557 if (writeback && copy_to_user(user_msrs->entries, entries, size))
558 goto out_free;
559
560 r = n;
561
562out_free:
563 vfree(entries);
564out:
565 return r;
566}
567
e9b11c17
ZX		 568/*
569 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
570 * cached on it.
571 */
572void decache_vcpus_on_cpu(int cpu)
573{
574 struct kvm *vm;
575 struct kvm_vcpu *vcpu;
576 int i;
577
578 spin_lock(&kvm_lock);
579 list_for_each_entry(vm, &vm_list, vm_list)
580 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
581 vcpu = vm->vcpus[i];
582 if (!vcpu)
583 continue;
584 /*
585 * If the vcpu is locked, then it is running on some
586 * other cpu and therefore it is not cached on the
587 * cpu in question.
588 *
589 * If it's not locked, check the last cpu it executed
590 * on.
591 */
592 if (mutex_trylock(&vcpu->mutex)) {
593 if (vcpu->cpu == cpu) {
594 kvm_x86_ops->vcpu_decache(vcpu);
595 vcpu->cpu = -1;
596 }
597 mutex_unlock(&vcpu->mutex);
598 }
599 }
600 spin_unlock(&kvm_lock);
601}
602
018d00d2
ZX		 603int kvm_dev_ioctl_check_extension(long ext)
604{
605 int r;
606
607 switch (ext) {
608 case KVM_CAP_IRQCHIP:
609 case KVM_CAP_HLT:
610 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
611 case KVM_CAP_USER_MEMORY:
612 case KVM_CAP_SET_TSS_ADDR:
613 r = 1;
614 break;
615 default:
616 r = 0;
617 break;
618 }
619 return r;
620
621}
622
043405e1
CO		 623long kvm_arch_dev_ioctl(struct file *filp,
624 unsigned int ioctl, unsigned long arg)
625{
626 void __user *argp = (void __user *)arg;
627 long r;
628
629 switch (ioctl) {
630 case KVM_GET_MSR_INDEX_LIST: {
631 struct kvm_msr_list __user *user_msr_list = argp;
632 struct kvm_msr_list msr_list;
633 unsigned n;
634
635 r = -EFAULT;
636 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
637 goto out;
638 n = msr_list.nmsrs;
639 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
640 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
641 goto out;
642 r = -E2BIG;
643 if (n < num_msrs_to_save)
644 goto out;
645 r = -EFAULT;
646 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
647 num_msrs_to_save * sizeof(u32)))
648 goto out;
649 if (copy_to_user(user_msr_list->indices
650 + num_msrs_to_save * sizeof(u32),
651 &emulated_msrs,
652 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
653 goto out;
654 r = 0;
655 break;
656 }
657 default:
658 r = -EINVAL;
659 }
660out:
661 return r;
662}
663
313a3dc7
CO		 664void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
665{
666 kvm_x86_ops->vcpu_load(vcpu, cpu);
667}
668
669void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
670{
671 kvm_x86_ops->vcpu_put(vcpu);
672}
673
674static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
675{
676 u64 efer;
677 int i;
678 struct kvm_cpuid_entry *e, *entry;
679
680 rdmsrl(MSR_EFER, efer);
681 entry = NULL;
682 for (i = 0; i < vcpu->cpuid_nent; ++i) {
683 e = &vcpu->cpuid_entries[i];
684 if (e->function == 0x80000001) {
685 entry = e;
686 break;
687 }
688 }
689 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
690 entry->edx &= ~(1 << 20);
691 printk(KERN_INFO "kvm: guest NX capability removed\n");
692 }
693}
694
695static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
696 struct kvm_cpuid *cpuid,
697 struct kvm_cpuid_entry __user *entries)
698{
699 int r;
700
701 r = -E2BIG;
702 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
703 goto out;
704 r = -EFAULT;
705 if (copy_from_user(&vcpu->cpuid_entries, entries,
706 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
707 goto out;
708 vcpu->cpuid_nent = cpuid->nent;
709 cpuid_fix_nx_cap(vcpu);
710 return 0;
711
712out:
713 return r;
714}
715
716static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
717 struct kvm_lapic_state *s)
718{
719 vcpu_load(vcpu);
720 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
721 vcpu_put(vcpu);
722
723 return 0;
724}
725
726static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
727 struct kvm_lapic_state *s)
728{
729 vcpu_load(vcpu);
730 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
731 kvm_apic_post_state_restore(vcpu);
732 vcpu_put(vcpu);
733
734 return 0;
735}
736
737long kvm_arch_vcpu_ioctl(struct file *filp,
738 unsigned int ioctl, unsigned long arg)
739{
740 struct kvm_vcpu *vcpu = filp->private_data;
741 void __user *argp = (void __user *)arg;
742 int r;
743
744 switch (ioctl) {
745 case KVM_GET_LAPIC: {
746 struct kvm_lapic_state lapic;
747
748 memset(&lapic, 0, sizeof lapic);
749 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
750 if (r)
751 goto out;
752 r = -EFAULT;
753 if (copy_to_user(argp, &lapic, sizeof lapic))
754 goto out;
755 r = 0;
756 break;
757 }
758 case KVM_SET_LAPIC: {
759 struct kvm_lapic_state lapic;
760
761 r = -EFAULT;
762 if (copy_from_user(&lapic, argp, sizeof lapic))
763 goto out;
764 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
765 if (r)
766 goto out;
767 r = 0;
768 break;
769 }
770 case KVM_SET_CPUID: {
771 struct kvm_cpuid __user *cpuid_arg = argp;
772 struct kvm_cpuid cpuid;
773
774 r = -EFAULT;
775 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
776 goto out;
777 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
778 if (r)
779 goto out;
780 break;
781 }
782 case KVM_GET_MSRS:
783 r = msr_io(vcpu, argp, kvm_get_msr, 1);
784 break;
785 case KVM_SET_MSRS:
786 r = msr_io(vcpu, argp, do_set_msr, 0);
787 break;
788 default:
789 r = -EINVAL;
790 }
791out:
792 return r;
793}
794
1fe779f8
CO		 795static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
796{
797 int ret;
798
799 if (addr > (unsigned int)(-3 * PAGE_SIZE))
800 return -1;
801 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
802 return ret;
803}
804
805static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
806 u32 kvm_nr_mmu_pages)
807{
808 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
809 return -EINVAL;
810
811 mutex_lock(&kvm->lock);
812
813 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
814 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
815
816 mutex_unlock(&kvm->lock);
817 return 0;
818}
819
820static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
821{
822 return kvm->n_alloc_mmu_pages;
823}
824
825/*
826 * Set a new alias region. Aliases map a portion of physical memory into
827 * another portion. This is useful for memory windows, for example the PC
828 * VGA region.
829 */
830static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
831 struct kvm_memory_alias *alias)
832{
833 int r, n;
834 struct kvm_mem_alias *p;
835
836 r = -EINVAL;
837 /* General sanity checks */
838 if (alias->memory_size & (PAGE_SIZE - 1))
839 goto out;
840 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
841 goto out;
842 if (alias->slot >= KVM_ALIAS_SLOTS)
843 goto out;
844 if (alias->guest_phys_addr + alias->memory_size
845 < alias->guest_phys_addr)
846 goto out;
847 if (alias->target_phys_addr + alias->memory_size
848 < alias->target_phys_addr)
849 goto out;
850
851 mutex_lock(&kvm->lock);
852
853 p = &kvm->aliases[alias->slot];
854 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
855 p->npages = alias->memory_size >> PAGE_SHIFT;
856 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
857
858 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
859 if (kvm->aliases[n - 1].npages)
860 break;
861 kvm->naliases = n;
862
863 kvm_mmu_zap_all(kvm);
864
865 mutex_unlock(&kvm->lock);
866
867 return 0;
868
869out:
870 return r;
871}
872
873static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
874{
875 int r;
876
877 r = 0;
878 switch (chip->chip_id) {
879 case KVM_IRQCHIP_PIC_MASTER:
880 memcpy(&chip->chip.pic,
881 &pic_irqchip(kvm)->pics[0],
882 sizeof(struct kvm_pic_state));
883 break;
884 case KVM_IRQCHIP_PIC_SLAVE:
885 memcpy(&chip->chip.pic,
886 &pic_irqchip(kvm)->pics[1],
887 sizeof(struct kvm_pic_state));
888 break;
889 case KVM_IRQCHIP_IOAPIC:
890 memcpy(&chip->chip.ioapic,
891 ioapic_irqchip(kvm),
892 sizeof(struct kvm_ioapic_state));
893 break;
894 default:
895 r = -EINVAL;
896 break;
897 }
898 return r;
899}
900
901static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
902{
903 int r;
904
905 r = 0;
906 switch (chip->chip_id) {
907 case KVM_IRQCHIP_PIC_MASTER:
908 memcpy(&pic_irqchip(kvm)->pics[0],
909 &chip->chip.pic,
910 sizeof(struct kvm_pic_state));
911 break;
912 case KVM_IRQCHIP_PIC_SLAVE:
913 memcpy(&pic_irqchip(kvm)->pics[1],
914 &chip->chip.pic,
915 sizeof(struct kvm_pic_state));
916 break;
917 case KVM_IRQCHIP_IOAPIC:
918 memcpy(ioapic_irqchip(kvm),
919 &chip->chip.ioapic,
920 sizeof(struct kvm_ioapic_state));
921 break;
922 default:
923 r = -EINVAL;
924 break;
925 }
926 kvm_pic_update_irq(pic_irqchip(kvm));
927 return r;
928}
929
930long kvm_arch_vm_ioctl(struct file *filp,
931 unsigned int ioctl, unsigned long arg)
932{
933 struct kvm *kvm = filp->private_data;
934 void __user *argp = (void __user *)arg;
935 int r = -EINVAL;
936
937 switch (ioctl) {
938 case KVM_SET_TSS_ADDR:
939 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
940 if (r < 0)
941 goto out;
942 break;
943 case KVM_SET_MEMORY_REGION: {
944 struct kvm_memory_region kvm_mem;
945 struct kvm_userspace_memory_region kvm_userspace_mem;
946
947 r = -EFAULT;
948 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
949 goto out;
950 kvm_userspace_mem.slot = kvm_mem.slot;
951 kvm_userspace_mem.flags = kvm_mem.flags;
952 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
953 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
954 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
955 if (r)
956 goto out;
957 break;
958 }
959 case KVM_SET_NR_MMU_PAGES:
960 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
961 if (r)
962 goto out;
963 break;
964 case KVM_GET_NR_MMU_PAGES:
965 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
966 break;
967 case KVM_SET_MEMORY_ALIAS: {
968 struct kvm_memory_alias alias;
969
970 r = -EFAULT;
971 if (copy_from_user(&alias, argp, sizeof alias))
972 goto out;
973 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
974 if (r)
975 goto out;
976 break;
977 }
978 case KVM_CREATE_IRQCHIP:
979 r = -ENOMEM;
980 kvm->vpic = kvm_create_pic(kvm);
981 if (kvm->vpic) {
982 r = kvm_ioapic_init(kvm);
983 if (r) {
984 kfree(kvm->vpic);
985 kvm->vpic = NULL;
986 goto out;
987 }
988 } else
989 goto out;
990 break;
991 case KVM_IRQ_LINE: {
992 struct kvm_irq_level irq_event;
993
994 r = -EFAULT;
995 if (copy_from_user(&irq_event, argp, sizeof irq_event))
996 goto out;
997 if (irqchip_in_kernel(kvm)) {
998 mutex_lock(&kvm->lock);
999 if (irq_event.irq < 16)
1000 kvm_pic_set_irq(pic_irqchip(kvm),
1001 irq_event.irq,
1002 irq_event.level);
1003 kvm_ioapic_set_irq(kvm->vioapic,
1004 irq_event.irq,
1005 irq_event.level);
1006 mutex_unlock(&kvm->lock);
1007 r = 0;
1008 }
1009 break;
1010 }
1011 case KVM_GET_IRQCHIP: {
1012 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1013 struct kvm_irqchip chip;
1014
1015 r = -EFAULT;
1016 if (copy_from_user(&chip, argp, sizeof chip))
1017 goto out;
1018 r = -ENXIO;
1019 if (!irqchip_in_kernel(kvm))
1020 goto out;
1021 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1022 if (r)
1023 goto out;
1024 r = -EFAULT;
1025 if (copy_to_user(argp, &chip, sizeof chip))
1026 goto out;
1027 r = 0;
1028 break;
1029 }
1030 case KVM_SET_IRQCHIP: {
1031 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1032 struct kvm_irqchip chip;
1033
1034 r = -EFAULT;
1035 if (copy_from_user(&chip, argp, sizeof chip))
1036 goto out;
1037 r = -ENXIO;
1038 if (!irqchip_in_kernel(kvm))
1039 goto out;
1040 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1041 if (r)
1042 goto out;
1043 r = 0;
1044 break;
1045 }
1046 default:
1047 ;
1048 }
1049out:
1050 return r;
1051}
1052
a16b043c 1053static void kvm_init_msr_list(void)
043405e1
CO		 1054{
1055 u32 dummy[2];
1056 unsigned i, j;
1057
1058 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1059 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1060 continue;
1061 if (j < i)
1062 msrs_to_save[j] = msrs_to_save[i];
1063 j++;
1064 }
1065 num_msrs_to_save = j;
1066}
1067
bbd9b64e
CO		 1068/*
1069 * Only apic need an MMIO device hook, so shortcut now..
1070 */
1071static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1072 gpa_t addr)
1073{
1074 struct kvm_io_device *dev;
1075
1076 if (vcpu->apic) {
1077 dev = &vcpu->apic->dev;
1078 if (dev->in_range(dev, addr))
1079 return dev;
1080 }
1081 return NULL;
1082}
1083
1084
1085static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1086 gpa_t addr)
1087{
1088 struct kvm_io_device *dev;
1089
1090 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1091 if (dev == NULL)
1092 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1093 return dev;
1094}
1095
1096int emulator_read_std(unsigned long addr,
1097 void *val,
1098 unsigned int bytes,
1099 struct kvm_vcpu *vcpu)
1100{
1101 void *data = val;
1102
1103 while (bytes) {
1104 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1105 unsigned offset = addr & (PAGE_SIZE-1);
1106 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1107 int ret;
1108
1109 if (gpa == UNMAPPED_GVA)
1110 return X86EMUL_PROPAGATE_FAULT;
1111 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1112 if (ret < 0)
1113 return X86EMUL_UNHANDLEABLE;
1114
1115 bytes -= tocopy;
1116 data += tocopy;
1117 addr += tocopy;
1118 }
1119
1120 return X86EMUL_CONTINUE;
1121}
1122EXPORT_SYMBOL_GPL(emulator_read_std);
1123
1124static int emulator_write_std(unsigned long addr,
1125 const void *val,
1126 unsigned int bytes,
1127 struct kvm_vcpu *vcpu)
1128{
1129 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1130 return X86EMUL_UNHANDLEABLE;
1131}
1132
1133static int emulator_read_emulated(unsigned long addr,
1134 void *val,
1135 unsigned int bytes,
1136 struct kvm_vcpu *vcpu)
1137{
1138 struct kvm_io_device *mmio_dev;
1139 gpa_t gpa;
1140
1141 if (vcpu->mmio_read_completed) {
1142 memcpy(val, vcpu->mmio_data, bytes);
1143 vcpu->mmio_read_completed = 0;
1144 return X86EMUL_CONTINUE;
1145 }
1146
1147 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1148
1149 /* For APIC access vmexit */
1150 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1151 goto mmio;
1152
1153 if (emulator_read_std(addr, val, bytes, vcpu)
1154 == X86EMUL_CONTINUE)
1155 return X86EMUL_CONTINUE;
1156 if (gpa == UNMAPPED_GVA)
1157 return X86EMUL_PROPAGATE_FAULT;
1158
1159mmio:
1160 /*
1161 * Is this MMIO handled locally?
1162 */
1163 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1164 if (mmio_dev) {
1165 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1166 return X86EMUL_CONTINUE;
1167 }
1168
1169 vcpu->mmio_needed = 1;
1170 vcpu->mmio_phys_addr = gpa;
1171 vcpu->mmio_size = bytes;
1172 vcpu->mmio_is_write = 0;
1173
1174 return X86EMUL_UNHANDLEABLE;
1175}
1176
1177static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1178 const void *val, int bytes)
1179{
1180 int ret;
1181
1182 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1183 if (ret < 0)
1184 return 0;
1185 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1186 return 1;
1187}
1188
1189static int emulator_write_emulated_onepage(unsigned long addr,
1190 const void *val,
1191 unsigned int bytes,
1192 struct kvm_vcpu *vcpu)
1193{
1194 struct kvm_io_device *mmio_dev;
1195 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1196
1197 if (gpa == UNMAPPED_GVA) {
1198 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1199 return X86EMUL_PROPAGATE_FAULT;
1200 }
1201
1202 /* For APIC access vmexit */
1203 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1204 goto mmio;
1205
1206 if (emulator_write_phys(vcpu, gpa, val, bytes))
1207 return X86EMUL_CONTINUE;
1208
1209mmio:
1210 /*
1211 * Is this MMIO handled locally?
1212 */
1213 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1214 if (mmio_dev) {
1215 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1216 return X86EMUL_CONTINUE;
1217 }
1218
1219 vcpu->mmio_needed = 1;
1220 vcpu->mmio_phys_addr = gpa;
1221 vcpu->mmio_size = bytes;
1222 vcpu->mmio_is_write = 1;
1223 memcpy(vcpu->mmio_data, val, bytes);
1224
1225 return X86EMUL_CONTINUE;
1226}
1227
1228int emulator_write_emulated(unsigned long addr,
1229 const void *val,
1230 unsigned int bytes,
1231 struct kvm_vcpu *vcpu)
1232{
1233 /* Crossing a page boundary? */
1234 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1235 int rc, now;
1236
1237 now = -addr & ~PAGE_MASK;
1238 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1239 if (rc != X86EMUL_CONTINUE)
1240 return rc;
1241 addr += now;
1242 val += now;
1243 bytes -= now;
1244 }
1245 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1246}
1247EXPORT_SYMBOL_GPL(emulator_write_emulated);
1248
1249static int emulator_cmpxchg_emulated(unsigned long addr,
1250 const void *old,
1251 const void *new,
1252 unsigned int bytes,
1253 struct kvm_vcpu *vcpu)
1254{
1255 static int reported;
1256
1257 if (!reported) {
1258 reported = 1;
1259 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1260 }
1261 return emulator_write_emulated(addr, new, bytes, vcpu);
1262}
1263
1264static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1265{
1266 return kvm_x86_ops->get_segment_base(vcpu, seg);
1267}
1268
1269int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1270{
1271 return X86EMUL_CONTINUE;
1272}
1273
1274int emulate_clts(struct kvm_vcpu *vcpu)
1275{
1276 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1277 return X86EMUL_CONTINUE;
1278}
1279
1280int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1281{
1282 struct kvm_vcpu *vcpu = ctxt->vcpu;
1283
1284 switch (dr) {
1285 case 0 ... 3:
1286 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1287 return X86EMUL_CONTINUE;
1288 default:
1289 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1290 return X86EMUL_UNHANDLEABLE;
1291 }
1292}
1293
1294int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1295{
1296 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1297 int exception;
1298
1299 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1300 if (exception) {
1301 /* FIXME: better handling */
1302 return X86EMUL_UNHANDLEABLE;
1303 }
1304 return X86EMUL_CONTINUE;
1305}
1306
1307void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1308{
1309 static int reported;
1310 u8 opcodes[4];
1311 unsigned long rip = vcpu->rip;
1312 unsigned long rip_linear;
1313
1314 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1315
1316 if (reported)
1317 return;
1318
1319 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1320
1321 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1322 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1323 reported = 1;
1324}
1325EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1326
1327struct x86_emulate_ops emulate_ops = {
1328 .read_std = emulator_read_std,
1329 .write_std = emulator_write_std,
1330 .read_emulated = emulator_read_emulated,
1331 .write_emulated = emulator_write_emulated,
1332 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1333};
1334
1335int emulate_instruction(struct kvm_vcpu *vcpu,
1336 struct kvm_run *run,
1337 unsigned long cr2,
1338 u16 error_code,
1339 int no_decode)
1340{
1341 int r;
1342
1343 vcpu->mmio_fault_cr2 = cr2;
1344 kvm_x86_ops->cache_regs(vcpu);
1345
1346 vcpu->mmio_is_write = 0;
1347 vcpu->pio.string = 0;
1348
1349 if (!no_decode) {
1350 int cs_db, cs_l;
1351 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1352
1353 vcpu->emulate_ctxt.vcpu = vcpu;
1354 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1355 vcpu->emulate_ctxt.cr2 = cr2;
1356 vcpu->emulate_ctxt.mode =
1357 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1358 ? X86EMUL_MODE_REAL : cs_l
1359 ? X86EMUL_MODE_PROT64 : cs_db
1360 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1361
1362 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1363 vcpu->emulate_ctxt.cs_base = 0;
1364 vcpu->emulate_ctxt.ds_base = 0;
1365 vcpu->emulate_ctxt.es_base = 0;
1366 vcpu->emulate_ctxt.ss_base = 0;
1367 } else {
1368 vcpu->emulate_ctxt.cs_base =
1369 get_segment_base(vcpu, VCPU_SREG_CS);
1370 vcpu->emulate_ctxt.ds_base =
1371 get_segment_base(vcpu, VCPU_SREG_DS);
1372 vcpu->emulate_ctxt.es_base =
1373 get_segment_base(vcpu, VCPU_SREG_ES);
1374 vcpu->emulate_ctxt.ss_base =
1375 get_segment_base(vcpu, VCPU_SREG_SS);
1376 }
1377
1378 vcpu->emulate_ctxt.gs_base =
1379 get_segment_base(vcpu, VCPU_SREG_GS);
1380 vcpu->emulate_ctxt.fs_base =
1381 get_segment_base(vcpu, VCPU_SREG_FS);
1382
1383 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1384 if (r) {
1385 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1386 return EMULATE_DONE;
1387 return EMULATE_FAIL;
1388 }
1389 }
1390
1391 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1392
1393 if (vcpu->pio.string)
1394 return EMULATE_DO_MMIO;
1395
1396 if ((r || vcpu->mmio_is_write) && run) {
1397 run->exit_reason = KVM_EXIT_MMIO;
1398 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1399 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1400 run->mmio.len = vcpu->mmio_size;
1401 run->mmio.is_write = vcpu->mmio_is_write;
1402 }
1403
1404 if (r) {
1405 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1406 return EMULATE_DONE;
1407 if (!vcpu->mmio_needed) {
1408 kvm_report_emulation_failure(vcpu, "mmio");
1409 return EMULATE_FAIL;
1410 }
1411 return EMULATE_DO_MMIO;
1412 }
1413
1414 kvm_x86_ops->decache_regs(vcpu);
1415 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1416
1417 if (vcpu->mmio_is_write) {
1418 vcpu->mmio_needed = 0;
1419 return EMULATE_DO_MMIO;
1420 }
1421
1422 return EMULATE_DONE;
1423}
1424EXPORT_SYMBOL_GPL(emulate_instruction);
1425
de7d789a
CO		 1426static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1427{
1428 int i;
1429
1430 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1431 if (vcpu->pio.guest_pages[i]) {
1432 kvm_release_page(vcpu->pio.guest_pages[i]);
1433 vcpu->pio.guest_pages[i] = NULL;
1434 }
1435}
1436
1437static int pio_copy_data(struct kvm_vcpu *vcpu)
1438{
1439 void *p = vcpu->pio_data;
1440 void *q;
1441 unsigned bytes;
1442 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1443
1444 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1445 PAGE_KERNEL);
1446 if (!q) {
1447 free_pio_guest_pages(vcpu);
1448 return -ENOMEM;
1449 }
1450 q += vcpu->pio.guest_page_offset;
1451 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1452 if (vcpu->pio.in)
1453 memcpy(q, p, bytes);
1454 else
1455 memcpy(p, q, bytes);
1456 q -= vcpu->pio.guest_page_offset;
1457 vunmap(q);
1458 free_pio_guest_pages(vcpu);
1459 return 0;
1460}
1461
1462int complete_pio(struct kvm_vcpu *vcpu)
1463{
1464 struct kvm_pio_request *io = &vcpu->pio;
1465 long delta;
1466 int r;
1467
1468 kvm_x86_ops->cache_regs(vcpu);
1469
1470 if (!io->string) {
1471 if (io->in)
1472 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1473 io->size);
1474 } else {
1475 if (io->in) {
1476 r = pio_copy_data(vcpu);
1477 if (r) {
1478 kvm_x86_ops->cache_regs(vcpu);
1479 return r;
1480 }
1481 }
1482
1483 delta = 1;
1484 if (io->rep) {
1485 delta *= io->cur_count;
1486 /*
1487 * The size of the register should really depend on
1488 * current address size.
1489 */
1490 vcpu->regs[VCPU_REGS_RCX] -= delta;
1491 }
1492 if (io->down)
1493 delta = -delta;
1494 delta *= io->size;
1495 if (io->in)
1496 vcpu->regs[VCPU_REGS_RDI] += delta;
1497 else
1498 vcpu->regs[VCPU_REGS_RSI] += delta;
1499 }
1500
1501 kvm_x86_ops->decache_regs(vcpu);
1502
1503 io->count -= io->cur_count;
1504 io->cur_count = 0;
1505
1506 return 0;
1507}
1508
1509static void kernel_pio(struct kvm_io_device *pio_dev,
1510 struct kvm_vcpu *vcpu,
1511 void *pd)
1512{
1513 /* TODO: String I/O for in kernel device */
1514
1515 mutex_lock(&vcpu->kvm->lock);
1516 if (vcpu->pio.in)
1517 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1518 vcpu->pio.size,
1519 pd);
1520 else
1521 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1522 vcpu->pio.size,
1523 pd);
1524 mutex_unlock(&vcpu->kvm->lock);
1525}
1526
1527static void pio_string_write(struct kvm_io_device *pio_dev,
1528 struct kvm_vcpu *vcpu)
1529{
1530 struct kvm_pio_request *io = &vcpu->pio;
1531 void *pd = vcpu->pio_data;
1532 int i;
1533
1534 mutex_lock(&vcpu->kvm->lock);
1535 for (i = 0; i < io->cur_count; i++) {
1536 kvm_iodevice_write(pio_dev, io->port,
1537 io->size,
1538 pd);
1539 pd += io->size;
1540 }
1541 mutex_unlock(&vcpu->kvm->lock);
1542}
1543
1544static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1545 gpa_t addr)
1546{
1547 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1548}
1549
1550int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1551 int size, unsigned port)
1552{
1553 struct kvm_io_device *pio_dev;
1554
1555 vcpu->run->exit_reason = KVM_EXIT_IO;
1556 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1557 vcpu->run->io.size = vcpu->pio.size = size;
1558 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1559 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1560 vcpu->run->io.port = vcpu->pio.port = port;
1561 vcpu->pio.in = in;
1562 vcpu->pio.string = 0;
1563 vcpu->pio.down = 0;
1564 vcpu->pio.guest_page_offset = 0;
1565 vcpu->pio.rep = 0;
1566
1567 kvm_x86_ops->cache_regs(vcpu);
1568 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1569 kvm_x86_ops->decache_regs(vcpu);
1570
1571 kvm_x86_ops->skip_emulated_instruction(vcpu);
1572
1573 pio_dev = vcpu_find_pio_dev(vcpu, port);
1574 if (pio_dev) {
1575 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1576 complete_pio(vcpu);
1577 return 1;
1578 }
1579 return 0;
1580}
1581EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1582
1583int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1584 int size, unsigned long count, int down,
1585 gva_t address, int rep, unsigned port)
1586{
1587 unsigned now, in_page;
1588 int i, ret = 0;
1589 int nr_pages = 1;
1590 struct page *page;
1591 struct kvm_io_device *pio_dev;
1592
1593 vcpu->run->exit_reason = KVM_EXIT_IO;
1594 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1595 vcpu->run->io.size = vcpu->pio.size = size;
1596 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1597 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1598 vcpu->run->io.port = vcpu->pio.port = port;
1599 vcpu->pio.in = in;
1600 vcpu->pio.string = 1;
1601 vcpu->pio.down = down;
1602 vcpu->pio.guest_page_offset = offset_in_page(address);
1603 vcpu->pio.rep = rep;
1604
1605 if (!count) {
1606 kvm_x86_ops->skip_emulated_instruction(vcpu);
1607 return 1;
1608 }
1609
1610 if (!down)
1611 in_page = PAGE_SIZE - offset_in_page(address);
1612 else
1613 in_page = offset_in_page(address) + size;
1614 now = min(count, (unsigned long)in_page / size);
1615 if (!now) {
1616 /*
1617 * String I/O straddles page boundary. Pin two guest pages
1618 * so that we satisfy atomicity constraints. Do just one
1619 * transaction to avoid complexity.
1620 */
1621 nr_pages = 2;
1622 now = 1;
1623 }
1624 if (down) {
1625 /*
1626 * String I/O in reverse. Yuck. Kill the guest, fix later.
1627 */
1628 pr_unimpl(vcpu, "guest string pio down\n");
1629 inject_gp(vcpu);
1630 return 1;
1631 }
1632 vcpu->run->io.count = now;
1633 vcpu->pio.cur_count = now;
1634
1635 if (vcpu->pio.cur_count == vcpu->pio.count)
1636 kvm_x86_ops->skip_emulated_instruction(vcpu);
1637
1638 for (i = 0; i < nr_pages; ++i) {
1639 mutex_lock(&vcpu->kvm->lock);
1640 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1641 vcpu->pio.guest_pages[i] = page;
1642 mutex_unlock(&vcpu->kvm->lock);
1643 if (!page) {
1644 inject_gp(vcpu);
1645 free_pio_guest_pages(vcpu);
1646 return 1;
1647 }
1648 }
1649
1650 pio_dev = vcpu_find_pio_dev(vcpu, port);
1651 if (!vcpu->pio.in) {
1652 /* string PIO write */
1653 ret = pio_copy_data(vcpu);
1654 if (ret >= 0 && pio_dev) {
1655 pio_string_write(pio_dev, vcpu);
1656 complete_pio(vcpu);
1657 if (vcpu->pio.count == 0)
1658 ret = 1;
1659 }
1660 } else if (pio_dev)
1661 pr_unimpl(vcpu, "no string pio read support yet, "
1662 "port %x size %d count %ld\n",
1663 port, size, count);
1664
1665 return ret;
1666}
1667EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1668
f8c16bba 1669int kvm_arch_init(void *opaque)
043405e1 1670{
f8c16bba
ZX		 1671 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
1672
043405e1 1673 kvm_init_msr_list();
f8c16bba
ZX		 1674
1675 if (kvm_x86_ops) {
1676 printk(KERN_ERR "kvm: already loaded the other module\n");
1677 return -EEXIST;
1678 }
1679
1680 if (!ops->cpu_has_kvm_support()) {
1681 printk(KERN_ERR "kvm: no hardware support\n");
1682 return -EOPNOTSUPP;
1683 }
1684 if (ops->disabled_by_bios()) {
1685 printk(KERN_ERR "kvm: disabled by bios\n");
1686 return -EOPNOTSUPP;
1687 }
1688
1689 kvm_x86_ops = ops;
1690
1691 return 0;
043405e1 1692}
8776e519 1693
f8c16bba
ZX		 1694void kvm_arch_exit(void)
1695{
1696 kvm_x86_ops = NULL;
1697 }
1698
8776e519
HB		 1699int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1700{
1701 ++vcpu->stat.halt_exits;
1702 if (irqchip_in_kernel(vcpu->kvm)) {
1703 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1704 kvm_vcpu_block(vcpu);
1705 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1706 return -EINTR;
1707 return 1;
1708 } else {
1709 vcpu->run->exit_reason = KVM_EXIT_HLT;
1710 return 0;
1711 }
1712}
1713EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1714
1715int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1716{
1717 unsigned long nr, a0, a1, a2, a3, ret;
1718
1719 kvm_x86_ops->cache_regs(vcpu);
1720
1721 nr = vcpu->regs[VCPU_REGS_RAX];
1722 a0 = vcpu->regs[VCPU_REGS_RBX];
1723 a1 = vcpu->regs[VCPU_REGS_RCX];
1724 a2 = vcpu->regs[VCPU_REGS_RDX];
1725 a3 = vcpu->regs[VCPU_REGS_RSI];
1726
1727 if (!is_long_mode(vcpu)) {
1728 nr &= 0xFFFFFFFF;
1729 a0 &= 0xFFFFFFFF;
1730 a1 &= 0xFFFFFFFF;
1731 a2 &= 0xFFFFFFFF;
1732 a3 &= 0xFFFFFFFF;
1733 }
1734
1735 switch (nr) {
1736 default:
1737 ret = -KVM_ENOSYS;
1738 break;
1739 }
1740 vcpu->regs[VCPU_REGS_RAX] = ret;
1741 kvm_x86_ops->decache_regs(vcpu);
1742 return 0;
1743}
1744EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1745
1746int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1747{
1748 char instruction[3];
1749 int ret = 0;
1750
1751 mutex_lock(&vcpu->kvm->lock);
1752
1753 /*
1754 * Blow out the MMU to ensure that no other VCPU has an active mapping
1755 * to ensure that the updated hypercall appears atomically across all
1756 * VCPUs.
1757 */
1758 kvm_mmu_zap_all(vcpu->kvm);
1759
1760 kvm_x86_ops->cache_regs(vcpu);
1761 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1762 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1763 != X86EMUL_CONTINUE)
1764 ret = -EFAULT;
1765
1766 mutex_unlock(&vcpu->kvm->lock);
1767
1768 return ret;
1769}
1770
1771static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1772{
1773 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1774}
1775
1776void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1777{
1778 struct descriptor_table dt = { limit, base };
1779
1780 kvm_x86_ops->set_gdt(vcpu, &dt);
1781}
1782
1783void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1784{
1785 struct descriptor_table dt = { limit, base };
1786
1787 kvm_x86_ops->set_idt(vcpu, &dt);
1788}
1789
1790void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1791 unsigned long *rflags)
1792{
1793 lmsw(vcpu, msw);
1794 *rflags = kvm_x86_ops->get_rflags(vcpu);
1795}
1796
1797unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1798{
1799 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1800 switch (cr) {
1801 case 0:
1802 return vcpu->cr0;
1803 case 2:
1804 return vcpu->cr2;
1805 case 3:
1806 return vcpu->cr3;
1807 case 4:
1808 return vcpu->cr4;
1809 default:
1810 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1811 return 0;
1812 }
1813}
1814
1815void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1816 unsigned long *rflags)
1817{
1818 switch (cr) {
1819 case 0:
1820 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1821 *rflags = kvm_x86_ops->get_rflags(vcpu);
1822 break;
1823 case 2:
1824 vcpu->cr2 = val;
1825 break;
1826 case 3:
1827 set_cr3(vcpu, val);
1828 break;
1829 case 4:
1830 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1831 break;
1832 default:
1833 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1834 }
1835}
1836
1837void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1838{
1839 int i;
1840 u32 function;
1841 struct kvm_cpuid_entry *e, *best;
1842
1843 kvm_x86_ops->cache_regs(vcpu);
1844 function = vcpu->regs[VCPU_REGS_RAX];
1845 vcpu->regs[VCPU_REGS_RAX] = 0;
1846 vcpu->regs[VCPU_REGS_RBX] = 0;
1847 vcpu->regs[VCPU_REGS_RCX] = 0;
1848 vcpu->regs[VCPU_REGS_RDX] = 0;
1849 best = NULL;
1850 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1851 e = &vcpu->cpuid_entries[i];
1852 if (e->function == function) {
1853 best = e;
1854 break;
1855 }
1856 /*
1857 * Both basic or both extended?
1858 */
1859 if (((e->function ^ function) & 0x80000000) == 0)
1860 if (!best || e->function > best->function)
1861 best = e;
1862 }
1863 if (best) {
1864 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1865 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1866 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1867 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1868 }
1869 kvm_x86_ops->decache_regs(vcpu);
1870 kvm_x86_ops->skip_emulated_instruction(vcpu);
1871}
1872EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 1873
b6c7a5dc
HB		 1874/*
1875 * Check if userspace requested an interrupt window, and that the
1876 * interrupt window is open.
1877 *
1878 * No need to exit to userspace if we already have an interrupt queued.
1879 */
1880static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1881 struct kvm_run *kvm_run)
1882{
1883 return (!vcpu->irq_summary &&
1884 kvm_run->request_interrupt_window &&
1885 vcpu->interrupt_window_open &&
1886 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1887}
1888
1889static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1890 struct kvm_run *kvm_run)
1891{
1892 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1893 kvm_run->cr8 = get_cr8(vcpu);
1894 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1895 if (irqchip_in_kernel(vcpu->kvm))
1896 kvm_run->ready_for_interrupt_injection = 1;
1897 else
1898 kvm_run->ready_for_interrupt_injection =
1899 (vcpu->interrupt_window_open &&
1900 vcpu->irq_summary == 0);
1901}
1902
1903static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1904{
1905 int r;
1906
1907 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1908 pr_debug("vcpu %d received sipi with vector # %x\n",
1909 vcpu->vcpu_id, vcpu->sipi_vector);
1910 kvm_lapic_reset(vcpu);
1911 r = kvm_x86_ops->vcpu_reset(vcpu);
1912 if (r)
1913 return r;
1914 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1915 }
1916
1917preempted:
1918 if (vcpu->guest_debug.enabled)
1919 kvm_x86_ops->guest_debug_pre(vcpu);
1920
1921again:
1922 r = kvm_mmu_reload(vcpu);
1923 if (unlikely(r))
1924 goto out;
1925
1926 kvm_inject_pending_timer_irqs(vcpu);
1927
1928 preempt_disable();
1929
1930 kvm_x86_ops->prepare_guest_switch(vcpu);
1931 kvm_load_guest_fpu(vcpu);
1932
1933 local_irq_disable();
1934
1935 if (signal_pending(current)) {
1936 local_irq_enable();
1937 preempt_enable();
1938 r = -EINTR;
1939 kvm_run->exit_reason = KVM_EXIT_INTR;
1940 ++vcpu->stat.signal_exits;
1941 goto out;
1942 }
1943
1944 if (irqchip_in_kernel(vcpu->kvm))
1945 kvm_x86_ops->inject_pending_irq(vcpu);
1946 else if (!vcpu->mmio_read_completed)
1947 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1948
1949 vcpu->guest_mode = 1;
1950 kvm_guest_enter();
1951
1952 if (vcpu->requests)
1953 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1954 kvm_x86_ops->tlb_flush(vcpu);
1955
1956 kvm_x86_ops->run(vcpu, kvm_run);
1957
1958 vcpu->guest_mode = 0;
1959 local_irq_enable();
1960
1961 ++vcpu->stat.exits;
1962
1963 /*
1964 * We must have an instruction between local_irq_enable() and
1965 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1966 * the interrupt shadow. The stat.exits increment will do nicely.
1967 * But we need to prevent reordering, hence this barrier():
1968 */
1969 barrier();
1970
1971 kvm_guest_exit();
1972
1973 preempt_enable();
1974
1975 /*
1976 * Profile KVM exit RIPs:
1977 */
1978 if (unlikely(prof_on == KVM_PROFILING)) {
1979 kvm_x86_ops->cache_regs(vcpu);
1980 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1981 }
1982
1983 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1984
1985 if (r > 0) {
1986 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1987 r = -EINTR;
1988 kvm_run->exit_reason = KVM_EXIT_INTR;
1989 ++vcpu->stat.request_irq_exits;
1990 goto out;
1991 }
e1beb1d3 1992 if (!need_resched())
b6c7a5dc 1993 goto again;
b6c7a5dc
HB		 1994 }
1995
1996out:
1997 if (r > 0) {
1998 kvm_resched(vcpu);
1999 goto preempted;
2000 }
2001
2002 post_kvm_run_save(vcpu, kvm_run);
2003
2004 return r;
2005}
2006
2007int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2008{
2009 int r;
2010 sigset_t sigsaved;
2011
2012 vcpu_load(vcpu);
2013
2014 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2015 kvm_vcpu_block(vcpu);
2016 vcpu_put(vcpu);
2017 return -EAGAIN;
2018 }
2019
2020 if (vcpu->sigset_active)
2021 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2022
2023 /* re-sync apic's tpr */
2024 if (!irqchip_in_kernel(vcpu->kvm))
2025 set_cr8(vcpu, kvm_run->cr8);
2026
2027 if (vcpu->pio.cur_count) {
2028 r = complete_pio(vcpu);
2029 if (r)
2030 goto out;
2031 }
2032#if CONFIG_HAS_IOMEM
2033 if (vcpu->mmio_needed) {
2034 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2035 vcpu->mmio_read_completed = 1;
2036 vcpu->mmio_needed = 0;
2037 r = emulate_instruction(vcpu, kvm_run,
2038 vcpu->mmio_fault_cr2, 0, 1);
2039 if (r == EMULATE_DO_MMIO) {
2040 /*
2041 * Read-modify-write. Back to userspace.
2042 */
2043 r = 0;
2044 goto out;
2045 }
2046 }
2047#endif
2048 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2049 kvm_x86_ops->cache_regs(vcpu);
2050 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2051 kvm_x86_ops->decache_regs(vcpu);
2052 }
2053
2054 r = __vcpu_run(vcpu, kvm_run);
2055
2056out:
2057 if (vcpu->sigset_active)
2058 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2059
2060 vcpu_put(vcpu);
2061 return r;
2062}
2063
2064int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2065{
2066 vcpu_load(vcpu);
2067
2068 kvm_x86_ops->cache_regs(vcpu);
2069
2070 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2071 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2072 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2073 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2074 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2075 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2076 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2077 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2078#ifdef CONFIG_X86_64
2079 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2080 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2081 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2082 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2083 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2084 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2085 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2086 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2087#endif
2088
2089 regs->rip = vcpu->rip;
2090 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2091
2092 /*
2093 * Don't leak debug flags in case they were set for guest debugging
2094 */
2095 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2096 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2097
2098 vcpu_put(vcpu);
2099
2100 return 0;
2101}
2102
2103int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2104{
2105 vcpu_load(vcpu);
2106
2107 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2108 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2109 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2110 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2111 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2112 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2113 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2114 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2115#ifdef CONFIG_X86_64
2116 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2117 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2118 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2119 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2120 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2121 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2122 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2123 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2124#endif
2125
2126 vcpu->rip = regs->rip;
2127 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2128
2129 kvm_x86_ops->decache_regs(vcpu);
2130
2131 vcpu_put(vcpu);
2132
2133 return 0;
2134}
2135
2136static void get_segment(struct kvm_vcpu *vcpu,
2137 struct kvm_segment *var, int seg)
2138{
2139 return kvm_x86_ops->get_segment(vcpu, var, seg);
2140}
2141
2142void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2143{
2144 struct kvm_segment cs;
2145
2146 get_segment(vcpu, &cs, VCPU_SREG_CS);
2147 *db = cs.db;
2148 *l = cs.l;
2149}
2150EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2151
2152int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2153 struct kvm_sregs *sregs)
2154{
2155 struct descriptor_table dt;
2156 int pending_vec;
2157
2158 vcpu_load(vcpu);
2159
2160 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2161 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2162 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2163 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2164 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2165 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2166
2167 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2168 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2169
2170 kvm_x86_ops->get_idt(vcpu, &dt);
2171 sregs->idt.limit = dt.limit;
2172 sregs->idt.base = dt.base;
2173 kvm_x86_ops->get_gdt(vcpu, &dt);
2174 sregs->gdt.limit = dt.limit;
2175 sregs->gdt.base = dt.base;
2176
2177 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2178 sregs->cr0 = vcpu->cr0;
2179 sregs->cr2 = vcpu->cr2;
2180 sregs->cr3 = vcpu->cr3;
2181 sregs->cr4 = vcpu->cr4;
2182 sregs->cr8 = get_cr8(vcpu);
2183 sregs->efer = vcpu->shadow_efer;
2184 sregs->apic_base = kvm_get_apic_base(vcpu);
2185
2186 if (irqchip_in_kernel(vcpu->kvm)) {
2187 memset(sregs->interrupt_bitmap, 0,
2188 sizeof sregs->interrupt_bitmap);
2189 pending_vec = kvm_x86_ops->get_irq(vcpu);
2190 if (pending_vec >= 0)
2191 set_bit(pending_vec,
2192 (unsigned long *)sregs->interrupt_bitmap);
2193 } else
2194 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2195 sizeof sregs->interrupt_bitmap);
2196
2197 vcpu_put(vcpu);
2198
2199 return 0;
2200}
2201
2202static void set_segment(struct kvm_vcpu *vcpu,
2203 struct kvm_segment *var, int seg)
2204{
2205 return kvm_x86_ops->set_segment(vcpu, var, seg);
2206}
2207
2208int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2209 struct kvm_sregs *sregs)
2210{
2211 int mmu_reset_needed = 0;
2212 int i, pending_vec, max_bits;
2213 struct descriptor_table dt;
2214
2215 vcpu_load(vcpu);
2216
2217 dt.limit = sregs->idt.limit;
2218 dt.base = sregs->idt.base;
2219 kvm_x86_ops->set_idt(vcpu, &dt);
2220 dt.limit = sregs->gdt.limit;
2221 dt.base = sregs->gdt.base;
2222 kvm_x86_ops->set_gdt(vcpu, &dt);
2223
2224 vcpu->cr2 = sregs->cr2;
2225 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2226 vcpu->cr3 = sregs->cr3;
2227
2228 set_cr8(vcpu, sregs->cr8);
2229
2230 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2231#ifdef CONFIG_X86_64
2232 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2233#endif
2234 kvm_set_apic_base(vcpu, sregs->apic_base);
2235
2236 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2237
2238 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2239 vcpu->cr0 = sregs->cr0;
2240 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2241
2242 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2243 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2244 if (!is_long_mode(vcpu) && is_pae(vcpu))
2245 load_pdptrs(vcpu, vcpu->cr3);
2246
2247 if (mmu_reset_needed)
2248 kvm_mmu_reset_context(vcpu);
2249
2250 if (!irqchip_in_kernel(vcpu->kvm)) {
2251 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2252 sizeof vcpu->irq_pending);
2253 vcpu->irq_summary = 0;
2254 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2255 if (vcpu->irq_pending[i])
2256 __set_bit(i, &vcpu->irq_summary);
2257 } else {
2258 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2259 pending_vec = find_first_bit(
2260 (const unsigned long *)sregs->interrupt_bitmap,
2261 max_bits);
2262 /* Only pending external irq is handled here */
2263 if (pending_vec < max_bits) {
2264 kvm_x86_ops->set_irq(vcpu, pending_vec);
2265 pr_debug("Set back pending irq %d\n",
2266 pending_vec);
2267 }
2268 }
2269
2270 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2271 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2272 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2273 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2274 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2275 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2276
2277 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2278 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2279
2280 vcpu_put(vcpu);
2281
2282 return 0;
2283}
2284
2285int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2286 struct kvm_debug_guest *dbg)
2287{
2288 int r;
2289
2290 vcpu_load(vcpu);
2291
2292 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2293
2294 vcpu_put(vcpu);
2295
2296 return r;
2297}
2298
d0752060
HB		 2299/*
2300 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2301 * we have asm/x86/processor.h
2302 */
2303struct fxsave {
2304 u16 cwd;
2305 u16 swd;
2306 u16 twd;
2307 u16 fop;
2308 u64 rip;
2309 u64 rdp;
2310 u32 mxcsr;
2311 u32 mxcsr_mask;
2312 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2313#ifdef CONFIG_X86_64
2314 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2315#else
2316 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2317#endif
2318};
2319
8b006791
ZX		 2320/*
2321 * Translate a guest virtual address to a guest physical address.
2322 */
2323int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2324 struct kvm_translation *tr)
2325{
2326 unsigned long vaddr = tr->linear_address;
2327 gpa_t gpa;
2328
2329 vcpu_load(vcpu);
2330 mutex_lock(&vcpu->kvm->lock);
2331 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2332 tr->physical_address = gpa;
2333 tr->valid = gpa != UNMAPPED_GVA;
2334 tr->writeable = 1;
2335 tr->usermode = 0;
2336 mutex_unlock(&vcpu->kvm->lock);
2337 vcpu_put(vcpu);
2338
2339 return 0;
2340}
2341
d0752060
HB		 2342int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2343{
2344 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2345
2346 vcpu_load(vcpu);
2347
2348 memcpy(fpu->fpr, fxsave->st_space, 128);
2349 fpu->fcw = fxsave->cwd;
2350 fpu->fsw = fxsave->swd;
2351 fpu->ftwx = fxsave->twd;
2352 fpu->last_opcode = fxsave->fop;
2353 fpu->last_ip = fxsave->rip;
2354 fpu->last_dp = fxsave->rdp;
2355 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2356
2357 vcpu_put(vcpu);
2358
2359 return 0;
2360}
2361
2362int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2363{
2364 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2365
2366 vcpu_load(vcpu);
2367
2368 memcpy(fxsave->st_space, fpu->fpr, 128);
2369 fxsave->cwd = fpu->fcw;
2370 fxsave->swd = fpu->fsw;
2371 fxsave->twd = fpu->ftwx;
2372 fxsave->fop = fpu->last_opcode;
2373 fxsave->rip = fpu->last_ip;
2374 fxsave->rdp = fpu->last_dp;
2375 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2376
2377 vcpu_put(vcpu);
2378
2379 return 0;
2380}
2381
2382void fx_init(struct kvm_vcpu *vcpu)
2383{
2384 unsigned after_mxcsr_mask;
2385
2386 /* Initialize guest FPU by resetting ours and saving into guest's */
2387 preempt_disable();
2388 fx_save(&vcpu->host_fx_image);
2389 fpu_init();
2390 fx_save(&vcpu->guest_fx_image);
2391 fx_restore(&vcpu->host_fx_image);
2392 preempt_enable();
2393
2394 vcpu->cr0 |= X86_CR0_ET;
2395 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2396 vcpu->guest_fx_image.mxcsr = 0x1f80;
2397 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2398 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2399}
2400EXPORT_SYMBOL_GPL(fx_init);
2401
2402void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2403{
2404 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2405 return;
2406
2407 vcpu->guest_fpu_loaded = 1;
2408 fx_save(&vcpu->host_fx_image);
2409 fx_restore(&vcpu->guest_fx_image);
2410}
2411EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2412
2413void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2414{
2415 if (!vcpu->guest_fpu_loaded)
2416 return;
2417
2418 vcpu->guest_fpu_loaded = 0;
2419 fx_save(&vcpu->guest_fx_image);
2420 fx_restore(&vcpu->host_fx_image);
f096ed85 2421 ++vcpu->stat.fpu_reload;
d0752060
HB		 2422}
2423EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
e9b11c17
ZX		 2424
2425void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2426{
2427 kvm_x86_ops->vcpu_free(vcpu);
2428}
2429
2430struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2431 unsigned int id)
2432{
2433 int r;
2434 struct kvm_vcpu *vcpu = kvm_x86_ops->vcpu_create(kvm, id);
2435
2436 if (IS_ERR(vcpu)) {
2437 r = -ENOMEM;
2438 goto fail;
2439 }
2440
2441 /* We do fxsave: this must be aligned. */
2442 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2443
2444 vcpu_load(vcpu);
2445 r = kvm_arch_vcpu_reset(vcpu);
2446 if (r == 0)
2447 r = kvm_mmu_setup(vcpu);
2448 vcpu_put(vcpu);
2449 if (r < 0)
2450 goto free_vcpu;
2451
2452 return vcpu;
2453free_vcpu:
2454 kvm_x86_ops->vcpu_free(vcpu);
2455fail:
2456 return ERR_PTR(r);
2457}
2458
2459void kvm_arch_vcpu_destory(struct kvm_vcpu *vcpu)
2460{
2461 vcpu_load(vcpu);
2462 kvm_mmu_unload(vcpu);
2463 vcpu_put(vcpu);
2464
2465 kvm_x86_ops->vcpu_free(vcpu);
2466}
2467
2468int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2469{
2470 return kvm_x86_ops->vcpu_reset(vcpu);
2471}
2472
2473void kvm_arch_hardware_enable(void *garbage)
2474{
2475 kvm_x86_ops->hardware_enable(garbage);
2476}
2477
2478void kvm_arch_hardware_disable(void *garbage)
2479{
2480 kvm_x86_ops->hardware_disable(garbage);
2481}
2482
2483int kvm_arch_hardware_setup(void)
2484{
2485 return kvm_x86_ops->hardware_setup();
2486}
2487
2488void kvm_arch_hardware_unsetup(void)
2489{
2490 kvm_x86_ops->hardware_unsetup();
2491}
2492
2493void kvm_arch_check_processor_compat(void *rtn)
2494{
2495 kvm_x86_ops->check_processor_compatibility(rtn);
2496}
2497
2498int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2499{
2500 struct page *page;
2501 struct kvm *kvm;
2502 int r;
2503
2504 BUG_ON(vcpu->kvm == NULL);
2505 kvm = vcpu->kvm;
2506
2507 vcpu->mmu.root_hpa = INVALID_PAGE;
2508 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2509 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2510 else
2511 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2512
2513 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2514 if (!page) {
2515 r = -ENOMEM;
2516 goto fail;
2517 }
2518 vcpu->pio_data = page_address(page);
2519
2520 r = kvm_mmu_create(vcpu);
2521 if (r < 0)
2522 goto fail_free_pio_data;
2523
2524 if (irqchip_in_kernel(kvm)) {
2525 r = kvm_create_lapic(vcpu);
2526 if (r < 0)
2527 goto fail_mmu_destroy;
2528 }
2529
2530 return 0;
2531
2532fail_mmu_destroy:
2533 kvm_mmu_destroy(vcpu);
2534fail_free_pio_data:
2535 free_page((unsigned long)vcpu->pio_data);
2536fail:
2537 return r;
2538}
2539
2540void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2541{
2542 kvm_free_lapic(vcpu);
2543 kvm_mmu_destroy(vcpu);
2544 free_page((unsigned long)vcpu->pio_data);
2545}
d19a9cd2
ZX		 2546
2547struct kvm *kvm_arch_create_vm(void)
2548{
2549 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
2550
2551 if (!kvm)
2552 return ERR_PTR(-ENOMEM);
2553
2554 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2555
2556 return kvm;
2557}
2558
2559static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
2560{
2561 vcpu_load(vcpu);
2562 kvm_mmu_unload(vcpu);
2563 vcpu_put(vcpu);
2564}
2565
2566static void kvm_free_vcpus(struct kvm *kvm)
2567{
2568 unsigned int i;
2569
2570 /*
2571 * Unpin any mmu pages first.
2572 */
2573 for (i = 0; i < KVM_MAX_VCPUS; ++i)
2574 if (kvm->vcpus[i])
2575 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
2576 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2577 if (kvm->vcpus[i]) {
2578 kvm_arch_vcpu_free(kvm->vcpus[i]);
2579 kvm->vcpus[i] = NULL;
2580 }
2581 }
2582
2583}
2584
2585void kvm_arch_destroy_vm(struct kvm *kvm)
2586{
2587 kfree(kvm->vpic);
2588 kfree(kvm->vioapic);
2589 kvm_free_vcpus(kvm);
2590 kvm_free_physmem(kvm);
2591 kfree(kvm);
2592}
Linux 6.x block layer and io_uring tree(s)