2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Avi Kivity <avi@qumranet.com>
10 * Yaniv Kamay <yaniv@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
17 #include <linux/kvm_host.h>
22 #include <linux/clocksource.h>
23 #include <linux/kvm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/module.h>
27 #include <linux/mman.h>
28 #include <linux/highmem.h>
30 #include <asm/uaccess.h>
34 #define MAX_IO_MSRS 256
35 #define CR0_RESERVED_BITS \
36 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
37 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
38 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
39 #define CR4_RESERVED_BITS \
40 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
41 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
42 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
43 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
45 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
47 * - enable syscall per default because its emulated by KVM
48 * - enable LME and LMA per default on 64 bit KVM
51 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
53 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
56 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
57 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
60 struct kvm_cpuid_entry2 __user
*entries
);
62 struct kvm_x86_ops
*kvm_x86_ops
;
64 struct kvm_stats_debugfs_item debugfs_entries
[] = {
65 { "pf_fixed", VCPU_STAT(pf_fixed
) },
66 { "pf_guest", VCPU_STAT(pf_guest
) },
67 { "tlb_flush", VCPU_STAT(tlb_flush
) },
68 { "invlpg", VCPU_STAT(invlpg
) },
69 { "exits", VCPU_STAT(exits
) },
70 { "io_exits", VCPU_STAT(io_exits
) },
71 { "mmio_exits", VCPU_STAT(mmio_exits
) },
72 { "signal_exits", VCPU_STAT(signal_exits
) },
73 { "irq_window", VCPU_STAT(irq_window_exits
) },
74 { "halt_exits", VCPU_STAT(halt_exits
) },
75 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
76 { "hypercalls", VCPU_STAT(hypercalls
) },
77 { "request_irq", VCPU_STAT(request_irq_exits
) },
78 { "irq_exits", VCPU_STAT(irq_exits
) },
79 { "host_state_reload", VCPU_STAT(host_state_reload
) },
80 { "efer_reload", VCPU_STAT(efer_reload
) },
81 { "fpu_reload", VCPU_STAT(fpu_reload
) },
82 { "insn_emulation", VCPU_STAT(insn_emulation
) },
83 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
84 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
85 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
86 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
87 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
88 { "mmu_flooded", VM_STAT(mmu_flooded
) },
89 { "mmu_recycled", VM_STAT(mmu_recycled
) },
90 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
91 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
92 { "largepages", VM_STAT(lpages
) },
97 unsigned long segment_base(u16 selector
)
99 struct descriptor_table gdt
;
100 struct desc_struct
*d
;
101 unsigned long table_base
;
107 asm("sgdt %0" : "=m"(gdt
));
108 table_base
= gdt
.base
;
110 if (selector
& 4) { /* from ldt */
113 asm("sldt %0" : "=g"(ldt_selector
));
114 table_base
= segment_base(ldt_selector
);
116 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
117 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
118 ((unsigned long)d
->base2
<< 24);
120 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
121 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
125 EXPORT_SYMBOL_GPL(segment_base
);
127 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
129 if (irqchip_in_kernel(vcpu
->kvm
))
130 return vcpu
->arch
.apic_base
;
132 return vcpu
->arch
.apic_base
;
134 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
136 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
138 /* TODO: reserve bits check */
139 if (irqchip_in_kernel(vcpu
->kvm
))
140 kvm_lapic_set_base(vcpu
, data
);
142 vcpu
->arch
.apic_base
= data
;
144 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
146 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
148 WARN_ON(vcpu
->arch
.exception
.pending
);
149 vcpu
->arch
.exception
.pending
= true;
150 vcpu
->arch
.exception
.has_error_code
= false;
151 vcpu
->arch
.exception
.nr
= nr
;
153 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
155 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
158 ++vcpu
->stat
.pf_guest
;
159 if (vcpu
->arch
.exception
.pending
) {
160 if (vcpu
->arch
.exception
.nr
== PF_VECTOR
) {
161 printk(KERN_DEBUG
"kvm: inject_page_fault:"
162 " double fault 0x%lx\n", addr
);
163 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
164 vcpu
->arch
.exception
.error_code
= 0;
165 } else if (vcpu
->arch
.exception
.nr
== DF_VECTOR
) {
166 /* triple fault -> shutdown */
167 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
171 vcpu
->arch
.cr2
= addr
;
172 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
175 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
177 WARN_ON(vcpu
->arch
.exception
.pending
);
178 vcpu
->arch
.exception
.pending
= true;
179 vcpu
->arch
.exception
.has_error_code
= true;
180 vcpu
->arch
.exception
.nr
= nr
;
181 vcpu
->arch
.exception
.error_code
= error_code
;
183 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
185 static void __queue_exception(struct kvm_vcpu
*vcpu
)
187 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
188 vcpu
->arch
.exception
.has_error_code
,
189 vcpu
->arch
.exception
.error_code
);
193 * Load the pae pdptrs. Return true is they are all valid.
195 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
197 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
198 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
201 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
203 down_read(&vcpu
->kvm
->slots_lock
);
204 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
205 offset
* sizeof(u64
), sizeof(pdpte
));
210 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
211 if ((pdpte
[i
] & 1) && (pdpte
[i
] & 0xfffffff0000001e6ull
)) {
218 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
220 up_read(&vcpu
->kvm
->slots_lock
);
224 EXPORT_SYMBOL_GPL(load_pdptrs
);
226 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
228 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
232 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
235 down_read(&vcpu
->kvm
->slots_lock
);
236 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
239 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
241 up_read(&vcpu
->kvm
->slots_lock
);
246 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
248 if (cr0
& CR0_RESERVED_BITS
) {
249 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
250 cr0
, vcpu
->arch
.cr0
);
251 kvm_inject_gp(vcpu
, 0);
255 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
256 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
257 kvm_inject_gp(vcpu
, 0);
261 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
262 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
263 "and a clear PE flag\n");
264 kvm_inject_gp(vcpu
, 0);
268 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
270 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
274 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
275 "in long mode while PAE is disabled\n");
276 kvm_inject_gp(vcpu
, 0);
279 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
281 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
282 "in long mode while CS.L == 1\n");
283 kvm_inject_gp(vcpu
, 0);
289 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
290 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
292 kvm_inject_gp(vcpu
, 0);
298 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
299 vcpu
->arch
.cr0
= cr0
;
301 kvm_mmu_reset_context(vcpu
);
304 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
306 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
308 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
310 EXPORT_SYMBOL_GPL(kvm_lmsw
);
312 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
314 if (cr4
& CR4_RESERVED_BITS
) {
315 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
316 kvm_inject_gp(vcpu
, 0);
320 if (is_long_mode(vcpu
)) {
321 if (!(cr4
& X86_CR4_PAE
)) {
322 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
324 kvm_inject_gp(vcpu
, 0);
327 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& X86_CR4_PAE
)
328 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
329 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
330 kvm_inject_gp(vcpu
, 0);
334 if (cr4
& X86_CR4_VMXE
) {
335 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
336 kvm_inject_gp(vcpu
, 0);
339 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
340 vcpu
->arch
.cr4
= cr4
;
341 kvm_mmu_reset_context(vcpu
);
343 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
345 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
347 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
348 kvm_mmu_flush_tlb(vcpu
);
352 if (is_long_mode(vcpu
)) {
353 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
354 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
355 kvm_inject_gp(vcpu
, 0);
360 if (cr3
& CR3_PAE_RESERVED_BITS
) {
362 "set_cr3: #GP, reserved bits\n");
363 kvm_inject_gp(vcpu
, 0);
366 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
367 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
369 kvm_inject_gp(vcpu
, 0);
374 * We don't check reserved bits in nonpae mode, because
375 * this isn't enforced, and VMware depends on this.
379 down_read(&vcpu
->kvm
->slots_lock
);
381 * Does the new cr3 value map to physical memory? (Note, we
382 * catch an invalid cr3 even in real-mode, because it would
383 * cause trouble later on when we turn on paging anyway.)
385 * A real CPU would silently accept an invalid cr3 and would
386 * attempt to use it - with largely undefined (and often hard
387 * to debug) behavior on the guest side.
389 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
390 kvm_inject_gp(vcpu
, 0);
392 vcpu
->arch
.cr3
= cr3
;
393 vcpu
->arch
.mmu
.new_cr3(vcpu
);
395 up_read(&vcpu
->kvm
->slots_lock
);
397 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
399 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
401 if (cr8
& CR8_RESERVED_BITS
) {
402 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
403 kvm_inject_gp(vcpu
, 0);
406 if (irqchip_in_kernel(vcpu
->kvm
))
407 kvm_lapic_set_tpr(vcpu
, cr8
);
409 vcpu
->arch
.cr8
= cr8
;
411 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
413 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
415 if (irqchip_in_kernel(vcpu
->kvm
))
416 return kvm_lapic_get_cr8(vcpu
);
418 return vcpu
->arch
.cr8
;
420 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
423 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
424 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
426 * This list is modified at module load time to reflect the
427 * capabilities of the host cpu.
429 static u32 msrs_to_save
[] = {
430 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
433 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
435 MSR_IA32_TIME_STAMP_COUNTER
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
436 MSR_IA32_PERF_STATUS
,
439 static unsigned num_msrs_to_save
;
441 static u32 emulated_msrs
[] = {
442 MSR_IA32_MISC_ENABLE
,
445 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
447 if (efer
& efer_reserved_bits
) {
448 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
450 kvm_inject_gp(vcpu
, 0);
455 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
456 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
457 kvm_inject_gp(vcpu
, 0);
461 kvm_x86_ops
->set_efer(vcpu
, efer
);
464 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
466 vcpu
->arch
.shadow_efer
= efer
;
469 void kvm_enable_efer_bits(u64 mask
)
471 efer_reserved_bits
&= ~mask
;
473 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
477 * Writes msr value into into the appropriate "register".
478 * Returns 0 on success, non-0 otherwise.
479 * Assumes vcpu_load() was already called.
481 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
483 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
487 * Adapt set_msr() to msr_io()'s calling convention
489 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
491 return kvm_set_msr(vcpu
, index
, *data
);
494 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
497 struct kvm_wall_clock wc
;
498 struct timespec wc_ts
;
505 down_read(&kvm
->slots_lock
);
506 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
508 wc_ts
= current_kernel_time();
509 wc
.wc_sec
= wc_ts
.tv_sec
;
510 wc
.wc_nsec
= wc_ts
.tv_nsec
;
511 wc
.wc_version
= version
;
513 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
516 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
517 up_read(&kvm
->slots_lock
);
520 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
524 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
527 if ((!vcpu
->time_page
))
530 /* Keep irq disabled to prevent changes to the clock */
531 local_irq_save(flags
);
532 kvm_get_msr(v
, MSR_IA32_TIME_STAMP_COUNTER
,
533 &vcpu
->hv_clock
.tsc_timestamp
);
535 local_irq_restore(flags
);
537 /* With all the info we got, fill in the values */
539 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
540 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
542 * The interface expects us to write an even number signaling that the
543 * update is finished. Since the guest won't see the intermediate
544 * state, we just write "2" at the end
546 vcpu
->hv_clock
.version
= 2;
548 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
550 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
551 sizeof(vcpu
->hv_clock
));
553 kunmap_atomic(shared_kaddr
, KM_USER0
);
555 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
559 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
563 set_efer(vcpu
, data
);
565 case MSR_IA32_MC0_STATUS
:
566 pr_unimpl(vcpu
, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
569 case MSR_IA32_MCG_STATUS
:
570 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
573 case MSR_IA32_MCG_CTL
:
574 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
577 case MSR_IA32_UCODE_REV
:
578 case MSR_IA32_UCODE_WRITE
:
579 case 0x200 ... 0x2ff: /* MTRRs */
581 case MSR_IA32_APICBASE
:
582 kvm_set_apic_base(vcpu
, data
);
584 case MSR_IA32_MISC_ENABLE
:
585 vcpu
->arch
.ia32_misc_enable_msr
= data
;
587 case MSR_KVM_WALL_CLOCK
:
588 vcpu
->kvm
->arch
.wall_clock
= data
;
589 kvm_write_wall_clock(vcpu
->kvm
, data
);
591 case MSR_KVM_SYSTEM_TIME
: {
592 if (vcpu
->arch
.time_page
) {
593 kvm_release_page_dirty(vcpu
->arch
.time_page
);
594 vcpu
->arch
.time_page
= NULL
;
597 vcpu
->arch
.time
= data
;
599 /* we verify if the enable bit is set... */
603 /* ...but clean it before doing the actual write */
604 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
606 vcpu
->arch
.hv_clock
.tsc_to_system_mul
=
607 clocksource_khz2mult(tsc_khz
, 22);
608 vcpu
->arch
.hv_clock
.tsc_shift
= 22;
610 down_read(¤t
->mm
->mmap_sem
);
611 down_read(&vcpu
->kvm
->slots_lock
);
612 vcpu
->arch
.time_page
=
613 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
614 up_read(&vcpu
->kvm
->slots_lock
);
615 up_read(¤t
->mm
->mmap_sem
);
617 if (is_error_page(vcpu
->arch
.time_page
)) {
618 kvm_release_page_clean(vcpu
->arch
.time_page
);
619 vcpu
->arch
.time_page
= NULL
;
622 kvm_write_guest_time(vcpu
);
626 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n", msr
, data
);
631 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
635 * Reads an msr value (of 'msr_index') into 'pdata'.
636 * Returns 0 on success, non-0 otherwise.
637 * Assumes vcpu_load() was already called.
639 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
641 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
644 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
649 case 0xc0010010: /* SYSCFG */
650 case 0xc0010015: /* HWCR */
651 case MSR_IA32_PLATFORM_ID
:
652 case MSR_IA32_P5_MC_ADDR
:
653 case MSR_IA32_P5_MC_TYPE
:
654 case MSR_IA32_MC0_CTL
:
655 case MSR_IA32_MCG_STATUS
:
656 case MSR_IA32_MCG_CAP
:
657 case MSR_IA32_MCG_CTL
:
658 case MSR_IA32_MC0_MISC
:
659 case MSR_IA32_MC0_MISC
+4:
660 case MSR_IA32_MC0_MISC
+8:
661 case MSR_IA32_MC0_MISC
+12:
662 case MSR_IA32_MC0_MISC
+16:
663 case MSR_IA32_UCODE_REV
:
664 case MSR_IA32_EBL_CR_POWERON
:
667 case 0x200 ... 0x2ff:
670 case 0xcd: /* fsb frequency */
673 case MSR_IA32_APICBASE
:
674 data
= kvm_get_apic_base(vcpu
);
676 case MSR_IA32_MISC_ENABLE
:
677 data
= vcpu
->arch
.ia32_misc_enable_msr
;
679 case MSR_IA32_PERF_STATUS
:
680 /* TSC increment by tick */
683 data
|= (((uint64_t)4ULL) << 40);
686 data
= vcpu
->arch
.shadow_efer
;
688 case MSR_KVM_WALL_CLOCK
:
689 data
= vcpu
->kvm
->arch
.wall_clock
;
691 case MSR_KVM_SYSTEM_TIME
:
692 data
= vcpu
->arch
.time
;
695 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
701 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
704 * Read or write a bunch of msrs. All parameters are kernel addresses.
706 * @return number of msrs set successfully.
708 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
709 struct kvm_msr_entry
*entries
,
710 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
711 unsigned index
, u64
*data
))
717 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
718 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
727 * Read or write a bunch of msrs. Parameters are user addresses.
729 * @return number of msrs set successfully.
731 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
732 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
733 unsigned index
, u64
*data
),
736 struct kvm_msrs msrs
;
737 struct kvm_msr_entry
*entries
;
742 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
746 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
750 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
751 entries
= vmalloc(size
);
756 if (copy_from_user(entries
, user_msrs
->entries
, size
))
759 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
764 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
776 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
779 void decache_vcpus_on_cpu(int cpu
)
782 struct kvm_vcpu
*vcpu
;
785 spin_lock(&kvm_lock
);
786 list_for_each_entry(vm
, &vm_list
, vm_list
)
787 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
792 * If the vcpu is locked, then it is running on some
793 * other cpu and therefore it is not cached on the
796 * If it's not locked, check the last cpu it executed
799 if (mutex_trylock(&vcpu
->mutex
)) {
800 if (vcpu
->cpu
== cpu
) {
801 kvm_x86_ops
->vcpu_decache(vcpu
);
804 mutex_unlock(&vcpu
->mutex
);
807 spin_unlock(&kvm_lock
);
810 int kvm_dev_ioctl_check_extension(long ext
)
815 case KVM_CAP_IRQCHIP
:
817 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
818 case KVM_CAP_USER_MEMORY
:
819 case KVM_CAP_SET_TSS_ADDR
:
820 case KVM_CAP_EXT_CPUID
:
821 case KVM_CAP_CLOCKSOURCE
:
826 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
828 case KVM_CAP_NR_VCPUS
:
831 case KVM_CAP_NR_MEMSLOTS
:
832 r
= KVM_MEMORY_SLOTS
;
842 long kvm_arch_dev_ioctl(struct file
*filp
,
843 unsigned int ioctl
, unsigned long arg
)
845 void __user
*argp
= (void __user
*)arg
;
849 case KVM_GET_MSR_INDEX_LIST
: {
850 struct kvm_msr_list __user
*user_msr_list
= argp
;
851 struct kvm_msr_list msr_list
;
855 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
858 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
859 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
862 if (n
< num_msrs_to_save
)
865 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
866 num_msrs_to_save
* sizeof(u32
)))
868 if (copy_to_user(user_msr_list
->indices
869 + num_msrs_to_save
* sizeof(u32
),
871 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
876 case KVM_GET_SUPPORTED_CPUID
: {
877 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
878 struct kvm_cpuid2 cpuid
;
881 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
883 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
889 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
901 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
903 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
904 kvm_write_guest_time(vcpu
);
907 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
909 kvm_x86_ops
->vcpu_put(vcpu
);
910 kvm_put_guest_fpu(vcpu
);
913 static int is_efer_nx(void)
917 rdmsrl(MSR_EFER
, efer
);
918 return efer
& EFER_NX
;
921 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
924 struct kvm_cpuid_entry2
*e
, *entry
;
927 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
928 e
= &vcpu
->arch
.cpuid_entries
[i
];
929 if (e
->function
== 0x80000001) {
934 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
935 entry
->edx
&= ~(1 << 20);
936 printk(KERN_INFO
"kvm: guest NX capability removed\n");
940 /* when an old userspace process fills a new kernel module */
941 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
942 struct kvm_cpuid
*cpuid
,
943 struct kvm_cpuid_entry __user
*entries
)
946 struct kvm_cpuid_entry
*cpuid_entries
;
949 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
952 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
956 if (copy_from_user(cpuid_entries
, entries
,
957 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
959 for (i
= 0; i
< cpuid
->nent
; i
++) {
960 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
961 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
962 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
963 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
964 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
965 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
966 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
967 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
968 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
969 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
971 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
972 cpuid_fix_nx_cap(vcpu
);
976 vfree(cpuid_entries
);
981 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
982 struct kvm_cpuid2
*cpuid
,
983 struct kvm_cpuid_entry2 __user
*entries
)
988 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
991 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
992 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
994 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1001 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1002 struct kvm_cpuid2
*cpuid
,
1003 struct kvm_cpuid_entry2 __user
*entries
)
1008 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1011 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1012 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1017 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1021 static inline u32
bit(int bitno
)
1023 return 1 << (bitno
& 31);
1026 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1029 entry
->function
= function
;
1030 entry
->index
= index
;
1031 cpuid_count(entry
->function
, entry
->index
,
1032 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1036 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1037 u32 index
, int *nent
, int maxnent
)
1039 const u32 kvm_supported_word0_x86_features
= bit(X86_FEATURE_FPU
) |
1040 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1041 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1042 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1043 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1044 bit(X86_FEATURE_SEP
) | bit(X86_FEATURE_PGE
) |
1045 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1046 bit(X86_FEATURE_CLFLSH
) | bit(X86_FEATURE_MMX
) |
1047 bit(X86_FEATURE_FXSR
) | bit(X86_FEATURE_XMM
) |
1048 bit(X86_FEATURE_XMM2
) | bit(X86_FEATURE_SELFSNOOP
);
1049 const u32 kvm_supported_word1_x86_features
= bit(X86_FEATURE_FPU
) |
1050 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1051 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1052 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1053 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1054 bit(X86_FEATURE_PGE
) |
1055 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1056 bit(X86_FEATURE_MMX
) | bit(X86_FEATURE_FXSR
) |
1057 bit(X86_FEATURE_SYSCALL
) |
1058 (bit(X86_FEATURE_NX
) && is_efer_nx()) |
1059 #ifdef CONFIG_X86_64
1060 bit(X86_FEATURE_LM
) |
1062 bit(X86_FEATURE_MMXEXT
) |
1063 bit(X86_FEATURE_3DNOWEXT
) |
1064 bit(X86_FEATURE_3DNOW
);
1065 const u32 kvm_supported_word3_x86_features
=
1066 bit(X86_FEATURE_XMM3
) | bit(X86_FEATURE_CX16
);
1067 const u32 kvm_supported_word6_x86_features
=
1068 bit(X86_FEATURE_LAHF_LM
) | bit(X86_FEATURE_CMP_LEGACY
);
1070 /* all func 2 cpuid_count() should be called on the same cpu */
1072 do_cpuid_1_ent(entry
, function
, index
);
1077 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1080 entry
->edx
&= kvm_supported_word0_x86_features
;
1081 entry
->ecx
&= kvm_supported_word3_x86_features
;
1083 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1084 * may return different values. This forces us to get_cpu() before
1085 * issuing the first command, and also to emulate this annoying behavior
1086 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1088 int t
, times
= entry
->eax
& 0xff;
1090 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1091 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1092 do_cpuid_1_ent(&entry
[t
], function
, 0);
1093 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1098 /* function 4 and 0xb have additional index. */
1102 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1103 /* read more entries until cache_type is zero */
1104 for (i
= 1; *nent
< maxnent
; ++i
) {
1105 cache_type
= entry
[i
- 1].eax
& 0x1f;
1108 do_cpuid_1_ent(&entry
[i
], function
, i
);
1110 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1118 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1119 /* read more entries until level_type is zero */
1120 for (i
= 1; *nent
< maxnent
; ++i
) {
1121 level_type
= entry
[i
- 1].ecx
& 0xff;
1124 do_cpuid_1_ent(&entry
[i
], function
, i
);
1126 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1132 entry
->eax
= min(entry
->eax
, 0x8000001a);
1135 entry
->edx
&= kvm_supported_word1_x86_features
;
1136 entry
->ecx
&= kvm_supported_word6_x86_features
;
1142 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1143 struct kvm_cpuid_entry2 __user
*entries
)
1145 struct kvm_cpuid_entry2
*cpuid_entries
;
1146 int limit
, nent
= 0, r
= -E2BIG
;
1149 if (cpuid
->nent
< 1)
1152 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1156 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1157 limit
= cpuid_entries
[0].eax
;
1158 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1159 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1160 &nent
, cpuid
->nent
);
1162 if (nent
>= cpuid
->nent
)
1165 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1166 limit
= cpuid_entries
[nent
- 1].eax
;
1167 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1168 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1169 &nent
, cpuid
->nent
);
1171 if (copy_to_user(entries
, cpuid_entries
,
1172 nent
* sizeof(struct kvm_cpuid_entry2
)))
1178 vfree(cpuid_entries
);
1183 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1184 struct kvm_lapic_state
*s
)
1187 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1193 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1194 struct kvm_lapic_state
*s
)
1197 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1198 kvm_apic_post_state_restore(vcpu
);
1204 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1205 struct kvm_interrupt
*irq
)
1207 if (irq
->irq
< 0 || irq
->irq
>= 256)
1209 if (irqchip_in_kernel(vcpu
->kvm
))
1213 set_bit(irq
->irq
, vcpu
->arch
.irq_pending
);
1214 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->arch
.irq_summary
);
1221 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1222 struct kvm_tpr_access_ctl
*tac
)
1226 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1230 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1231 unsigned int ioctl
, unsigned long arg
)
1233 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1234 void __user
*argp
= (void __user
*)arg
;
1238 case KVM_GET_LAPIC
: {
1239 struct kvm_lapic_state lapic
;
1241 memset(&lapic
, 0, sizeof lapic
);
1242 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, &lapic
);
1246 if (copy_to_user(argp
, &lapic
, sizeof lapic
))
1251 case KVM_SET_LAPIC
: {
1252 struct kvm_lapic_state lapic
;
1255 if (copy_from_user(&lapic
, argp
, sizeof lapic
))
1257 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, &lapic
);;
1263 case KVM_INTERRUPT
: {
1264 struct kvm_interrupt irq
;
1267 if (copy_from_user(&irq
, argp
, sizeof irq
))
1269 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1275 case KVM_SET_CPUID
: {
1276 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1277 struct kvm_cpuid cpuid
;
1280 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1282 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1287 case KVM_SET_CPUID2
: {
1288 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1289 struct kvm_cpuid2 cpuid
;
1292 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1294 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1295 cpuid_arg
->entries
);
1300 case KVM_GET_CPUID2
: {
1301 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1302 struct kvm_cpuid2 cpuid
;
1305 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1307 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1308 cpuid_arg
->entries
);
1312 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1318 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1321 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1323 case KVM_TPR_ACCESS_REPORTING
: {
1324 struct kvm_tpr_access_ctl tac
;
1327 if (copy_from_user(&tac
, argp
, sizeof tac
))
1329 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1333 if (copy_to_user(argp
, &tac
, sizeof tac
))
1338 case KVM_SET_VAPIC_ADDR
: {
1339 struct kvm_vapic_addr va
;
1342 if (!irqchip_in_kernel(vcpu
->kvm
))
1345 if (copy_from_user(&va
, argp
, sizeof va
))
1348 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1358 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1362 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1364 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1368 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1369 u32 kvm_nr_mmu_pages
)
1371 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1374 down_write(&kvm
->slots_lock
);
1376 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1377 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1379 up_write(&kvm
->slots_lock
);
1383 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1385 return kvm
->arch
.n_alloc_mmu_pages
;
1388 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1391 struct kvm_mem_alias
*alias
;
1393 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1394 alias
= &kvm
->arch
.aliases
[i
];
1395 if (gfn
>= alias
->base_gfn
1396 && gfn
< alias
->base_gfn
+ alias
->npages
)
1397 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1403 * Set a new alias region. Aliases map a portion of physical memory into
1404 * another portion. This is useful for memory windows, for example the PC
1407 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1408 struct kvm_memory_alias
*alias
)
1411 struct kvm_mem_alias
*p
;
1414 /* General sanity checks */
1415 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1417 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1419 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1421 if (alias
->guest_phys_addr
+ alias
->memory_size
1422 < alias
->guest_phys_addr
)
1424 if (alias
->target_phys_addr
+ alias
->memory_size
1425 < alias
->target_phys_addr
)
1428 down_write(&kvm
->slots_lock
);
1430 p
= &kvm
->arch
.aliases
[alias
->slot
];
1431 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1432 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1433 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1435 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1436 if (kvm
->arch
.aliases
[n
- 1].npages
)
1438 kvm
->arch
.naliases
= n
;
1440 kvm_mmu_zap_all(kvm
);
1442 up_write(&kvm
->slots_lock
);
1450 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1455 switch (chip
->chip_id
) {
1456 case KVM_IRQCHIP_PIC_MASTER
:
1457 memcpy(&chip
->chip
.pic
,
1458 &pic_irqchip(kvm
)->pics
[0],
1459 sizeof(struct kvm_pic_state
));
1461 case KVM_IRQCHIP_PIC_SLAVE
:
1462 memcpy(&chip
->chip
.pic
,
1463 &pic_irqchip(kvm
)->pics
[1],
1464 sizeof(struct kvm_pic_state
));
1466 case KVM_IRQCHIP_IOAPIC
:
1467 memcpy(&chip
->chip
.ioapic
,
1468 ioapic_irqchip(kvm
),
1469 sizeof(struct kvm_ioapic_state
));
1478 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1483 switch (chip
->chip_id
) {
1484 case KVM_IRQCHIP_PIC_MASTER
:
1485 memcpy(&pic_irqchip(kvm
)->pics
[0],
1487 sizeof(struct kvm_pic_state
));
1489 case KVM_IRQCHIP_PIC_SLAVE
:
1490 memcpy(&pic_irqchip(kvm
)->pics
[1],
1492 sizeof(struct kvm_pic_state
));
1494 case KVM_IRQCHIP_IOAPIC
:
1495 memcpy(ioapic_irqchip(kvm
),
1497 sizeof(struct kvm_ioapic_state
));
1503 kvm_pic_update_irq(pic_irqchip(kvm
));
1508 * Get (and clear) the dirty memory log for a memory slot.
1510 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1511 struct kvm_dirty_log
*log
)
1515 struct kvm_memory_slot
*memslot
;
1518 down_write(&kvm
->slots_lock
);
1520 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1524 /* If nothing is dirty, don't bother messing with page tables. */
1526 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
1527 kvm_flush_remote_tlbs(kvm
);
1528 memslot
= &kvm
->memslots
[log
->slot
];
1529 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1530 memset(memslot
->dirty_bitmap
, 0, n
);
1534 up_write(&kvm
->slots_lock
);
1538 long kvm_arch_vm_ioctl(struct file
*filp
,
1539 unsigned int ioctl
, unsigned long arg
)
1541 struct kvm
*kvm
= filp
->private_data
;
1542 void __user
*argp
= (void __user
*)arg
;
1546 case KVM_SET_TSS_ADDR
:
1547 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
1551 case KVM_SET_MEMORY_REGION
: {
1552 struct kvm_memory_region kvm_mem
;
1553 struct kvm_userspace_memory_region kvm_userspace_mem
;
1556 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
1558 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
1559 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
1560 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
1561 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
1562 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
1567 case KVM_SET_NR_MMU_PAGES
:
1568 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
1572 case KVM_GET_NR_MMU_PAGES
:
1573 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
1575 case KVM_SET_MEMORY_ALIAS
: {
1576 struct kvm_memory_alias alias
;
1579 if (copy_from_user(&alias
, argp
, sizeof alias
))
1581 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
1586 case KVM_CREATE_IRQCHIP
:
1588 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
1589 if (kvm
->arch
.vpic
) {
1590 r
= kvm_ioapic_init(kvm
);
1592 kfree(kvm
->arch
.vpic
);
1593 kvm
->arch
.vpic
= NULL
;
1599 case KVM_CREATE_PIT
:
1601 kvm
->arch
.vpit
= kvm_create_pit(kvm
);
1605 case KVM_IRQ_LINE
: {
1606 struct kvm_irq_level irq_event
;
1609 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
1611 if (irqchip_in_kernel(kvm
)) {
1612 mutex_lock(&kvm
->lock
);
1613 if (irq_event
.irq
< 16)
1614 kvm_pic_set_irq(pic_irqchip(kvm
),
1617 kvm_ioapic_set_irq(kvm
->arch
.vioapic
,
1620 mutex_unlock(&kvm
->lock
);
1625 case KVM_GET_IRQCHIP
: {
1626 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1627 struct kvm_irqchip chip
;
1630 if (copy_from_user(&chip
, argp
, sizeof chip
))
1633 if (!irqchip_in_kernel(kvm
))
1635 r
= kvm_vm_ioctl_get_irqchip(kvm
, &chip
);
1639 if (copy_to_user(argp
, &chip
, sizeof chip
))
1644 case KVM_SET_IRQCHIP
: {
1645 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1646 struct kvm_irqchip chip
;
1649 if (copy_from_user(&chip
, argp
, sizeof chip
))
1652 if (!irqchip_in_kernel(kvm
))
1654 r
= kvm_vm_ioctl_set_irqchip(kvm
, &chip
);
1667 static void kvm_init_msr_list(void)
1672 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1673 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1676 msrs_to_save
[j
] = msrs_to_save
[i
];
1679 num_msrs_to_save
= j
;
1683 * Only apic need an MMIO device hook, so shortcut now..
1685 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
1688 struct kvm_io_device
*dev
;
1690 if (vcpu
->arch
.apic
) {
1691 dev
= &vcpu
->arch
.apic
->dev
;
1692 if (dev
->in_range(dev
, addr
))
1699 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
1702 struct kvm_io_device
*dev
;
1704 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
);
1706 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
);
1710 int emulator_read_std(unsigned long addr
,
1713 struct kvm_vcpu
*vcpu
)
1716 int r
= X86EMUL_CONTINUE
;
1718 down_read(&vcpu
->kvm
->slots_lock
);
1720 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1721 unsigned offset
= addr
& (PAGE_SIZE
-1);
1722 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
1725 if (gpa
== UNMAPPED_GVA
) {
1726 r
= X86EMUL_PROPAGATE_FAULT
;
1729 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, tocopy
);
1731 r
= X86EMUL_UNHANDLEABLE
;
1740 up_read(&vcpu
->kvm
->slots_lock
);
1743 EXPORT_SYMBOL_GPL(emulator_read_std
);
1745 static int emulator_read_emulated(unsigned long addr
,
1748 struct kvm_vcpu
*vcpu
)
1750 struct kvm_io_device
*mmio_dev
;
1753 if (vcpu
->mmio_read_completed
) {
1754 memcpy(val
, vcpu
->mmio_data
, bytes
);
1755 vcpu
->mmio_read_completed
= 0;
1756 return X86EMUL_CONTINUE
;
1759 down_read(&vcpu
->kvm
->slots_lock
);
1760 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1761 up_read(&vcpu
->kvm
->slots_lock
);
1763 /* For APIC access vmexit */
1764 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1767 if (emulator_read_std(addr
, val
, bytes
, vcpu
)
1768 == X86EMUL_CONTINUE
)
1769 return X86EMUL_CONTINUE
;
1770 if (gpa
== UNMAPPED_GVA
)
1771 return X86EMUL_PROPAGATE_FAULT
;
1775 * Is this MMIO handled locally?
1777 mutex_lock(&vcpu
->kvm
->lock
);
1778 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1780 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
1781 mutex_unlock(&vcpu
->kvm
->lock
);
1782 return X86EMUL_CONTINUE
;
1784 mutex_unlock(&vcpu
->kvm
->lock
);
1786 vcpu
->mmio_needed
= 1;
1787 vcpu
->mmio_phys_addr
= gpa
;
1788 vcpu
->mmio_size
= bytes
;
1789 vcpu
->mmio_is_write
= 0;
1791 return X86EMUL_UNHANDLEABLE
;
1794 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1795 const void *val
, int bytes
)
1799 down_read(&vcpu
->kvm
->slots_lock
);
1800 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
1802 up_read(&vcpu
->kvm
->slots_lock
);
1805 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
1806 up_read(&vcpu
->kvm
->slots_lock
);
1810 static int emulator_write_emulated_onepage(unsigned long addr
,
1813 struct kvm_vcpu
*vcpu
)
1815 struct kvm_io_device
*mmio_dev
;
1818 down_read(&vcpu
->kvm
->slots_lock
);
1819 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1820 up_read(&vcpu
->kvm
->slots_lock
);
1822 if (gpa
== UNMAPPED_GVA
) {
1823 kvm_inject_page_fault(vcpu
, addr
, 2);
1824 return X86EMUL_PROPAGATE_FAULT
;
1827 /* For APIC access vmexit */
1828 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1831 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1832 return X86EMUL_CONTINUE
;
1836 * Is this MMIO handled locally?
1838 mutex_lock(&vcpu
->kvm
->lock
);
1839 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1841 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
1842 mutex_unlock(&vcpu
->kvm
->lock
);
1843 return X86EMUL_CONTINUE
;
1845 mutex_unlock(&vcpu
->kvm
->lock
);
1847 vcpu
->mmio_needed
= 1;
1848 vcpu
->mmio_phys_addr
= gpa
;
1849 vcpu
->mmio_size
= bytes
;
1850 vcpu
->mmio_is_write
= 1;
1851 memcpy(vcpu
->mmio_data
, val
, bytes
);
1853 return X86EMUL_CONTINUE
;
1856 int emulator_write_emulated(unsigned long addr
,
1859 struct kvm_vcpu
*vcpu
)
1861 /* Crossing a page boundary? */
1862 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
1865 now
= -addr
& ~PAGE_MASK
;
1866 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
1867 if (rc
!= X86EMUL_CONTINUE
)
1873 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
1875 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
1877 static int emulator_cmpxchg_emulated(unsigned long addr
,
1881 struct kvm_vcpu
*vcpu
)
1883 static int reported
;
1887 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1889 #ifndef CONFIG_X86_64
1890 /* guests cmpxchg8b have to be emulated atomically */
1897 down_read(&vcpu
->kvm
->slots_lock
);
1898 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1900 if (gpa
== UNMAPPED_GVA
||
1901 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1904 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
1909 down_read(¤t
->mm
->mmap_sem
);
1910 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1911 up_read(¤t
->mm
->mmap_sem
);
1913 kaddr
= kmap_atomic(page
, KM_USER0
);
1914 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
1915 kunmap_atomic(kaddr
, KM_USER0
);
1916 kvm_release_page_dirty(page
);
1918 up_read(&vcpu
->kvm
->slots_lock
);
1922 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
1925 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1927 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
1930 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1932 return X86EMUL_CONTINUE
;
1935 int emulate_clts(struct kvm_vcpu
*vcpu
)
1937 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
1938 return X86EMUL_CONTINUE
;
1941 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
1943 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1947 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
1948 return X86EMUL_CONTINUE
;
1950 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
1951 return X86EMUL_UNHANDLEABLE
;
1955 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1957 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1960 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1962 /* FIXME: better handling */
1963 return X86EMUL_UNHANDLEABLE
;
1965 return X86EMUL_CONTINUE
;
1968 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
1970 static int reported
;
1972 unsigned long rip
= vcpu
->arch
.rip
;
1973 unsigned long rip_linear
;
1975 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
1980 emulator_read_std(rip_linear
, (void *)opcodes
, 4, vcpu
);
1982 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1983 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1986 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
1988 static struct x86_emulate_ops emulate_ops
= {
1989 .read_std
= emulator_read_std
,
1990 .read_emulated
= emulator_read_emulated
,
1991 .write_emulated
= emulator_write_emulated
,
1992 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1995 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1996 struct kvm_run
*run
,
2002 struct decode_cache
*c
;
2004 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2005 kvm_x86_ops
->cache_regs(vcpu
);
2007 vcpu
->mmio_is_write
= 0;
2008 vcpu
->arch
.pio
.string
= 0;
2010 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2012 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2014 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2015 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2016 vcpu
->arch
.emulate_ctxt
.mode
=
2017 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2018 ? X86EMUL_MODE_REAL
: cs_l
2019 ? X86EMUL_MODE_PROT64
: cs_db
2020 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2022 if (vcpu
->arch
.emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
2023 vcpu
->arch
.emulate_ctxt
.cs_base
= 0;
2024 vcpu
->arch
.emulate_ctxt
.ds_base
= 0;
2025 vcpu
->arch
.emulate_ctxt
.es_base
= 0;
2026 vcpu
->arch
.emulate_ctxt
.ss_base
= 0;
2028 vcpu
->arch
.emulate_ctxt
.cs_base
=
2029 get_segment_base(vcpu
, VCPU_SREG_CS
);
2030 vcpu
->arch
.emulate_ctxt
.ds_base
=
2031 get_segment_base(vcpu
, VCPU_SREG_DS
);
2032 vcpu
->arch
.emulate_ctxt
.es_base
=
2033 get_segment_base(vcpu
, VCPU_SREG_ES
);
2034 vcpu
->arch
.emulate_ctxt
.ss_base
=
2035 get_segment_base(vcpu
, VCPU_SREG_SS
);
2038 vcpu
->arch
.emulate_ctxt
.gs_base
=
2039 get_segment_base(vcpu
, VCPU_SREG_GS
);
2040 vcpu
->arch
.emulate_ctxt
.fs_base
=
2041 get_segment_base(vcpu
, VCPU_SREG_FS
);
2043 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2045 /* Reject the instructions other than VMCALL/VMMCALL when
2046 * try to emulate invalid opcode */
2047 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2048 if ((emulation_type
& EMULTYPE_TRAP_UD
) &&
2049 (!(c
->twobyte
&& c
->b
== 0x01 &&
2050 (c
->modrm_reg
== 0 || c
->modrm_reg
== 3) &&
2051 c
->modrm_mod
== 3 && c
->modrm_rm
== 1)))
2052 return EMULATE_FAIL
;
2054 ++vcpu
->stat
.insn_emulation
;
2056 ++vcpu
->stat
.insn_emulation_fail
;
2057 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2058 return EMULATE_DONE
;
2059 return EMULATE_FAIL
;
2063 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2065 if (vcpu
->arch
.pio
.string
)
2066 return EMULATE_DO_MMIO
;
2068 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2069 run
->exit_reason
= KVM_EXIT_MMIO
;
2070 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2071 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2072 run
->mmio
.len
= vcpu
->mmio_size
;
2073 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2077 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2078 return EMULATE_DONE
;
2079 if (!vcpu
->mmio_needed
) {
2080 kvm_report_emulation_failure(vcpu
, "mmio");
2081 return EMULATE_FAIL
;
2083 return EMULATE_DO_MMIO
;
2086 kvm_x86_ops
->decache_regs(vcpu
);
2087 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2089 if (vcpu
->mmio_is_write
) {
2090 vcpu
->mmio_needed
= 0;
2091 return EMULATE_DO_MMIO
;
2094 return EMULATE_DONE
;
2096 EXPORT_SYMBOL_GPL(emulate_instruction
);
2098 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
2102 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.pio
.guest_pages
); ++i
)
2103 if (vcpu
->arch
.pio
.guest_pages
[i
]) {
2104 kvm_release_page_dirty(vcpu
->arch
.pio
.guest_pages
[i
]);
2105 vcpu
->arch
.pio
.guest_pages
[i
] = NULL
;
2109 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2111 void *p
= vcpu
->arch
.pio_data
;
2114 int nr_pages
= vcpu
->arch
.pio
.guest_pages
[1] ? 2 : 1;
2116 q
= vmap(vcpu
->arch
.pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
2119 free_pio_guest_pages(vcpu
);
2122 q
+= vcpu
->arch
.pio
.guest_page_offset
;
2123 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2124 if (vcpu
->arch
.pio
.in
)
2125 memcpy(q
, p
, bytes
);
2127 memcpy(p
, q
, bytes
);
2128 q
-= vcpu
->arch
.pio
.guest_page_offset
;
2130 free_pio_guest_pages(vcpu
);
2134 int complete_pio(struct kvm_vcpu
*vcpu
)
2136 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2140 kvm_x86_ops
->cache_regs(vcpu
);
2144 memcpy(&vcpu
->arch
.regs
[VCPU_REGS_RAX
], vcpu
->arch
.pio_data
,
2148 r
= pio_copy_data(vcpu
);
2150 kvm_x86_ops
->cache_regs(vcpu
);
2157 delta
*= io
->cur_count
;
2159 * The size of the register should really depend on
2160 * current address size.
2162 vcpu
->arch
.regs
[VCPU_REGS_RCX
] -= delta
;
2168 vcpu
->arch
.regs
[VCPU_REGS_RDI
] += delta
;
2170 vcpu
->arch
.regs
[VCPU_REGS_RSI
] += delta
;
2173 kvm_x86_ops
->decache_regs(vcpu
);
2175 io
->count
-= io
->cur_count
;
2181 static void kernel_pio(struct kvm_io_device
*pio_dev
,
2182 struct kvm_vcpu
*vcpu
,
2185 /* TODO: String I/O for in kernel device */
2187 mutex_lock(&vcpu
->kvm
->lock
);
2188 if (vcpu
->arch
.pio
.in
)
2189 kvm_iodevice_read(pio_dev
, vcpu
->arch
.pio
.port
,
2190 vcpu
->arch
.pio
.size
,
2193 kvm_iodevice_write(pio_dev
, vcpu
->arch
.pio
.port
,
2194 vcpu
->arch
.pio
.size
,
2196 mutex_unlock(&vcpu
->kvm
->lock
);
2199 static void pio_string_write(struct kvm_io_device
*pio_dev
,
2200 struct kvm_vcpu
*vcpu
)
2202 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2203 void *pd
= vcpu
->arch
.pio_data
;
2206 mutex_lock(&vcpu
->kvm
->lock
);
2207 for (i
= 0; i
< io
->cur_count
; i
++) {
2208 kvm_iodevice_write(pio_dev
, io
->port
,
2213 mutex_unlock(&vcpu
->kvm
->lock
);
2216 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
2219 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
);
2222 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2223 int size
, unsigned port
)
2225 struct kvm_io_device
*pio_dev
;
2227 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2228 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2229 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2230 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2231 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2232 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2233 vcpu
->arch
.pio
.in
= in
;
2234 vcpu
->arch
.pio
.string
= 0;
2235 vcpu
->arch
.pio
.down
= 0;
2236 vcpu
->arch
.pio
.guest_page_offset
= 0;
2237 vcpu
->arch
.pio
.rep
= 0;
2239 kvm_x86_ops
->cache_regs(vcpu
);
2240 memcpy(vcpu
->arch
.pio_data
, &vcpu
->arch
.regs
[VCPU_REGS_RAX
], 4);
2241 kvm_x86_ops
->decache_regs(vcpu
);
2243 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2245 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
2247 kernel_pio(pio_dev
, vcpu
, vcpu
->arch
.pio_data
);
2253 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2255 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2256 int size
, unsigned long count
, int down
,
2257 gva_t address
, int rep
, unsigned port
)
2259 unsigned now
, in_page
;
2263 struct kvm_io_device
*pio_dev
;
2265 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2266 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2267 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2268 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2269 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2270 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2271 vcpu
->arch
.pio
.in
= in
;
2272 vcpu
->arch
.pio
.string
= 1;
2273 vcpu
->arch
.pio
.down
= down
;
2274 vcpu
->arch
.pio
.guest_page_offset
= offset_in_page(address
);
2275 vcpu
->arch
.pio
.rep
= rep
;
2278 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2283 in_page
= PAGE_SIZE
- offset_in_page(address
);
2285 in_page
= offset_in_page(address
) + size
;
2286 now
= min(count
, (unsigned long)in_page
/ size
);
2289 * String I/O straddles page boundary. Pin two guest pages
2290 * so that we satisfy atomicity constraints. Do just one
2291 * transaction to avoid complexity.
2298 * String I/O in reverse. Yuck. Kill the guest, fix later.
2300 pr_unimpl(vcpu
, "guest string pio down\n");
2301 kvm_inject_gp(vcpu
, 0);
2304 vcpu
->run
->io
.count
= now
;
2305 vcpu
->arch
.pio
.cur_count
= now
;
2307 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2308 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2310 for (i
= 0; i
< nr_pages
; ++i
) {
2311 down_read(&vcpu
->kvm
->slots_lock
);
2312 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
2313 vcpu
->arch
.pio
.guest_pages
[i
] = page
;
2314 up_read(&vcpu
->kvm
->slots_lock
);
2316 kvm_inject_gp(vcpu
, 0);
2317 free_pio_guest_pages(vcpu
);
2322 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
2323 if (!vcpu
->arch
.pio
.in
) {
2324 /* string PIO write */
2325 ret
= pio_copy_data(vcpu
);
2326 if (ret
>= 0 && pio_dev
) {
2327 pio_string_write(pio_dev
, vcpu
);
2329 if (vcpu
->arch
.pio
.count
== 0)
2333 pr_unimpl(vcpu
, "no string pio read support yet, "
2334 "port %x size %d count %ld\n",
2339 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2341 int kvm_arch_init(void *opaque
)
2344 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
2347 printk(KERN_ERR
"kvm: already loaded the other module\n");
2352 if (!ops
->cpu_has_kvm_support()) {
2353 printk(KERN_ERR
"kvm: no hardware support\n");
2357 if (ops
->disabled_by_bios()) {
2358 printk(KERN_ERR
"kvm: disabled by bios\n");
2363 r
= kvm_mmu_module_init();
2367 kvm_init_msr_list();
2370 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2377 void kvm_arch_exit(void)
2380 kvm_mmu_module_exit();
2383 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
2385 ++vcpu
->stat
.halt_exits
;
2386 if (irqchip_in_kernel(vcpu
->kvm
)) {
2387 vcpu
->arch
.mp_state
= VCPU_MP_STATE_HALTED
;
2388 kvm_vcpu_block(vcpu
);
2389 if (vcpu
->arch
.mp_state
!= VCPU_MP_STATE_RUNNABLE
)
2393 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
2397 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
2399 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
2401 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
2403 kvm_x86_ops
->cache_regs(vcpu
);
2405 nr
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2406 a0
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
2407 a1
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2408 a2
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
2409 a3
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
2411 if (!is_long_mode(vcpu
)) {
2420 case KVM_HC_VAPIC_POLL_IRQ
:
2427 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = ret
;
2428 kvm_x86_ops
->decache_regs(vcpu
);
2429 ++vcpu
->stat
.hypercalls
;
2432 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
2434 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
2436 char instruction
[3];
2441 * Blow out the MMU to ensure that no other VCPU has an active mapping
2442 * to ensure that the updated hypercall appears atomically across all
2445 kvm_mmu_zap_all(vcpu
->kvm
);
2447 kvm_x86_ops
->cache_regs(vcpu
);
2448 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
2449 if (emulator_write_emulated(vcpu
->arch
.rip
, instruction
, 3, vcpu
)
2450 != X86EMUL_CONTINUE
)
2456 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
2458 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
2461 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2463 struct descriptor_table dt
= { limit
, base
};
2465 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2468 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2470 struct descriptor_table dt
= { limit
, base
};
2472 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2475 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
2476 unsigned long *rflags
)
2478 kvm_lmsw(vcpu
, msw
);
2479 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2482 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
2484 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2487 return vcpu
->arch
.cr0
;
2489 return vcpu
->arch
.cr2
;
2491 return vcpu
->arch
.cr3
;
2493 return vcpu
->arch
.cr4
;
2495 return kvm_get_cr8(vcpu
);
2497 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2502 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
2503 unsigned long *rflags
)
2507 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
2508 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2511 vcpu
->arch
.cr2
= val
;
2514 kvm_set_cr3(vcpu
, val
);
2517 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
2520 kvm_set_cr8(vcpu
, val
& 0xfUL
);
2523 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2527 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
2529 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
2530 int j
, nent
= vcpu
->arch
.cpuid_nent
;
2532 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
2533 /* when no next entry is found, the current entry[i] is reselected */
2534 for (j
= i
+ 1; j
== i
; j
= (j
+ 1) % nent
) {
2535 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
2536 if (ej
->function
== e
->function
) {
2537 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2541 return 0; /* silence gcc, even though control never reaches here */
2544 /* find an entry with matching function, matching index (if needed), and that
2545 * should be read next (if it's stateful) */
2546 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
2547 u32 function
, u32 index
)
2549 if (e
->function
!= function
)
2551 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
2553 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
2554 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
2559 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
2562 u32 function
, index
;
2563 struct kvm_cpuid_entry2
*e
, *best
;
2565 kvm_x86_ops
->cache_regs(vcpu
);
2566 function
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2567 index
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2568 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = 0;
2569 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = 0;
2570 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = 0;
2571 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = 0;
2573 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
2574 e
= &vcpu
->arch
.cpuid_entries
[i
];
2575 if (is_matching_cpuid_entry(e
, function
, index
)) {
2576 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
2577 move_to_next_stateful_cpuid_entry(vcpu
, i
);
2582 * Both basic or both extended?
2584 if (((e
->function
^ function
) & 0x80000000) == 0)
2585 if (!best
|| e
->function
> best
->function
)
2589 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = best
->eax
;
2590 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = best
->ebx
;
2591 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = best
->ecx
;
2592 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = best
->edx
;
2594 kvm_x86_ops
->decache_regs(vcpu
);
2595 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2597 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
2600 * Check if userspace requested an interrupt window, and that the
2601 * interrupt window is open.
2603 * No need to exit to userspace if we already have an interrupt queued.
2605 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
2606 struct kvm_run
*kvm_run
)
2608 return (!vcpu
->arch
.irq_summary
&&
2609 kvm_run
->request_interrupt_window
&&
2610 vcpu
->arch
.interrupt_window_open
&&
2611 (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
));
2614 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
2615 struct kvm_run
*kvm_run
)
2617 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
2618 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
2619 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
2620 if (irqchip_in_kernel(vcpu
->kvm
))
2621 kvm_run
->ready_for_interrupt_injection
= 1;
2623 kvm_run
->ready_for_interrupt_injection
=
2624 (vcpu
->arch
.interrupt_window_open
&&
2625 vcpu
->arch
.irq_summary
== 0);
2628 static void vapic_enter(struct kvm_vcpu
*vcpu
)
2630 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2633 if (!apic
|| !apic
->vapic_addr
)
2636 down_read(¤t
->mm
->mmap_sem
);
2637 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
2638 up_read(¤t
->mm
->mmap_sem
);
2640 vcpu
->arch
.apic
->vapic_page
= page
;
2643 static void vapic_exit(struct kvm_vcpu
*vcpu
)
2645 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2647 if (!apic
|| !apic
->vapic_addr
)
2650 kvm_release_page_dirty(apic
->vapic_page
);
2651 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
2654 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2658 if (unlikely(vcpu
->arch
.mp_state
== VCPU_MP_STATE_SIPI_RECEIVED
)) {
2659 pr_debug("vcpu %d received sipi with vector # %x\n",
2660 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
2661 kvm_lapic_reset(vcpu
);
2662 r
= kvm_x86_ops
->vcpu_reset(vcpu
);
2665 vcpu
->arch
.mp_state
= VCPU_MP_STATE_RUNNABLE
;
2671 if (vcpu
->guest_debug
.enabled
)
2672 kvm_x86_ops
->guest_debug_pre(vcpu
);
2676 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
2677 kvm_mmu_unload(vcpu
);
2679 r
= kvm_mmu_reload(vcpu
);
2683 if (vcpu
->requests
) {
2684 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
2685 __kvm_migrate_apic_timer(vcpu
);
2686 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
2688 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
2692 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
2693 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
2699 kvm_inject_pending_timer_irqs(vcpu
);
2703 kvm_x86_ops
->prepare_guest_switch(vcpu
);
2704 kvm_load_guest_fpu(vcpu
);
2706 local_irq_disable();
2708 if (need_resched()) {
2716 if (test_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
)) {
2723 if (signal_pending(current
)) {
2727 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2728 ++vcpu
->stat
.signal_exits
;
2732 if (vcpu
->arch
.exception
.pending
)
2733 __queue_exception(vcpu
);
2734 else if (irqchip_in_kernel(vcpu
->kvm
))
2735 kvm_x86_ops
->inject_pending_irq(vcpu
);
2737 kvm_x86_ops
->inject_pending_vectors(vcpu
, kvm_run
);
2739 kvm_lapic_sync_to_vapic(vcpu
);
2741 vcpu
->guest_mode
= 1;
2745 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
2746 kvm_x86_ops
->tlb_flush(vcpu
);
2748 kvm_x86_ops
->run(vcpu
, kvm_run
);
2750 vcpu
->guest_mode
= 0;
2756 * We must have an instruction between local_irq_enable() and
2757 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2758 * the interrupt shadow. The stat.exits increment will do nicely.
2759 * But we need to prevent reordering, hence this barrier():
2768 * Profile KVM exit RIPs:
2770 if (unlikely(prof_on
== KVM_PROFILING
)) {
2771 kvm_x86_ops
->cache_regs(vcpu
);
2772 profile_hit(KVM_PROFILING
, (void *)vcpu
->arch
.rip
);
2775 if (vcpu
->arch
.exception
.pending
&& kvm_x86_ops
->exception_injected(vcpu
))
2776 vcpu
->arch
.exception
.pending
= false;
2778 kvm_lapic_sync_from_vapic(vcpu
);
2780 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
2783 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
2785 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2786 ++vcpu
->stat
.request_irq_exits
;
2789 if (!need_resched())
2799 post_kvm_run_save(vcpu
, kvm_run
);
2806 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2813 if (unlikely(vcpu
->arch
.mp_state
== VCPU_MP_STATE_UNINITIALIZED
)) {
2814 kvm_vcpu_block(vcpu
);
2819 if (vcpu
->sigset_active
)
2820 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
2822 /* re-sync apic's tpr */
2823 if (!irqchip_in_kernel(vcpu
->kvm
))
2824 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
2826 if (vcpu
->arch
.pio
.cur_count
) {
2827 r
= complete_pio(vcpu
);
2831 #if CONFIG_HAS_IOMEM
2832 if (vcpu
->mmio_needed
) {
2833 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
2834 vcpu
->mmio_read_completed
= 1;
2835 vcpu
->mmio_needed
= 0;
2836 r
= emulate_instruction(vcpu
, kvm_run
,
2837 vcpu
->arch
.mmio_fault_cr2
, 0,
2838 EMULTYPE_NO_DECODE
);
2839 if (r
== EMULATE_DO_MMIO
) {
2841 * Read-modify-write. Back to userspace.
2848 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
2849 kvm_x86_ops
->cache_regs(vcpu
);
2850 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
2851 kvm_x86_ops
->decache_regs(vcpu
);
2854 r
= __vcpu_run(vcpu
, kvm_run
);
2857 if (vcpu
->sigset_active
)
2858 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
2864 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2868 kvm_x86_ops
->cache_regs(vcpu
);
2870 regs
->rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2871 regs
->rbx
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
2872 regs
->rcx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2873 regs
->rdx
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
2874 regs
->rsi
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
2875 regs
->rdi
= vcpu
->arch
.regs
[VCPU_REGS_RDI
];
2876 regs
->rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
2877 regs
->rbp
= vcpu
->arch
.regs
[VCPU_REGS_RBP
];
2878 #ifdef CONFIG_X86_64
2879 regs
->r8
= vcpu
->arch
.regs
[VCPU_REGS_R8
];
2880 regs
->r9
= vcpu
->arch
.regs
[VCPU_REGS_R9
];
2881 regs
->r10
= vcpu
->arch
.regs
[VCPU_REGS_R10
];
2882 regs
->r11
= vcpu
->arch
.regs
[VCPU_REGS_R11
];
2883 regs
->r12
= vcpu
->arch
.regs
[VCPU_REGS_R12
];
2884 regs
->r13
= vcpu
->arch
.regs
[VCPU_REGS_R13
];
2885 regs
->r14
= vcpu
->arch
.regs
[VCPU_REGS_R14
];
2886 regs
->r15
= vcpu
->arch
.regs
[VCPU_REGS_R15
];
2889 regs
->rip
= vcpu
->arch
.rip
;
2890 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2893 * Don't leak debug flags in case they were set for guest debugging
2895 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
2896 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
2903 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2907 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = regs
->rax
;
2908 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = regs
->rbx
;
2909 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = regs
->rcx
;
2910 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = regs
->rdx
;
2911 vcpu
->arch
.regs
[VCPU_REGS_RSI
] = regs
->rsi
;
2912 vcpu
->arch
.regs
[VCPU_REGS_RDI
] = regs
->rdi
;
2913 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = regs
->rsp
;
2914 vcpu
->arch
.regs
[VCPU_REGS_RBP
] = regs
->rbp
;
2915 #ifdef CONFIG_X86_64
2916 vcpu
->arch
.regs
[VCPU_REGS_R8
] = regs
->r8
;
2917 vcpu
->arch
.regs
[VCPU_REGS_R9
] = regs
->r9
;
2918 vcpu
->arch
.regs
[VCPU_REGS_R10
] = regs
->r10
;
2919 vcpu
->arch
.regs
[VCPU_REGS_R11
] = regs
->r11
;
2920 vcpu
->arch
.regs
[VCPU_REGS_R12
] = regs
->r12
;
2921 vcpu
->arch
.regs
[VCPU_REGS_R13
] = regs
->r13
;
2922 vcpu
->arch
.regs
[VCPU_REGS_R14
] = regs
->r14
;
2923 vcpu
->arch
.regs
[VCPU_REGS_R15
] = regs
->r15
;
2926 vcpu
->arch
.rip
= regs
->rip
;
2927 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
2929 kvm_x86_ops
->decache_regs(vcpu
);
2936 static void get_segment(struct kvm_vcpu
*vcpu
,
2937 struct kvm_segment
*var
, int seg
)
2939 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
2942 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
2944 struct kvm_segment cs
;
2946 get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
2950 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
2952 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
2953 struct kvm_sregs
*sregs
)
2955 struct descriptor_table dt
;
2960 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2961 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2962 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2963 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2964 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2965 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2967 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2968 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2970 kvm_x86_ops
->get_idt(vcpu
, &dt
);
2971 sregs
->idt
.limit
= dt
.limit
;
2972 sregs
->idt
.base
= dt
.base
;
2973 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
2974 sregs
->gdt
.limit
= dt
.limit
;
2975 sregs
->gdt
.base
= dt
.base
;
2977 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2978 sregs
->cr0
= vcpu
->arch
.cr0
;
2979 sregs
->cr2
= vcpu
->arch
.cr2
;
2980 sregs
->cr3
= vcpu
->arch
.cr3
;
2981 sregs
->cr4
= vcpu
->arch
.cr4
;
2982 sregs
->cr8
= kvm_get_cr8(vcpu
);
2983 sregs
->efer
= vcpu
->arch
.shadow_efer
;
2984 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
2986 if (irqchip_in_kernel(vcpu
->kvm
)) {
2987 memset(sregs
->interrupt_bitmap
, 0,
2988 sizeof sregs
->interrupt_bitmap
);
2989 pending_vec
= kvm_x86_ops
->get_irq(vcpu
);
2990 if (pending_vec
>= 0)
2991 set_bit(pending_vec
,
2992 (unsigned long *)sregs
->interrupt_bitmap
);
2994 memcpy(sregs
->interrupt_bitmap
, vcpu
->arch
.irq_pending
,
2995 sizeof sregs
->interrupt_bitmap
);
3002 static void set_segment(struct kvm_vcpu
*vcpu
,
3003 struct kvm_segment
*var
, int seg
)
3005 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3008 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
3009 struct kvm_sregs
*sregs
)
3011 int mmu_reset_needed
= 0;
3012 int i
, pending_vec
, max_bits
;
3013 struct descriptor_table dt
;
3017 dt
.limit
= sregs
->idt
.limit
;
3018 dt
.base
= sregs
->idt
.base
;
3019 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3020 dt
.limit
= sregs
->gdt
.limit
;
3021 dt
.base
= sregs
->gdt
.base
;
3022 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3024 vcpu
->arch
.cr2
= sregs
->cr2
;
3025 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
3026 vcpu
->arch
.cr3
= sregs
->cr3
;
3028 kvm_set_cr8(vcpu
, sregs
->cr8
);
3030 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
3031 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
3032 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
3034 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3036 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
3037 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
3038 vcpu
->arch
.cr0
= sregs
->cr0
;
3040 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
3041 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
3042 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
3043 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
3045 if (mmu_reset_needed
)
3046 kvm_mmu_reset_context(vcpu
);
3048 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3049 memcpy(vcpu
->arch
.irq_pending
, sregs
->interrupt_bitmap
,
3050 sizeof vcpu
->arch
.irq_pending
);
3051 vcpu
->arch
.irq_summary
= 0;
3052 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.irq_pending
); ++i
)
3053 if (vcpu
->arch
.irq_pending
[i
])
3054 __set_bit(i
, &vcpu
->arch
.irq_summary
);
3056 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
3057 pending_vec
= find_first_bit(
3058 (const unsigned long *)sregs
->interrupt_bitmap
,
3060 /* Only pending external irq is handled here */
3061 if (pending_vec
< max_bits
) {
3062 kvm_x86_ops
->set_irq(vcpu
, pending_vec
);
3063 pr_debug("Set back pending irq %d\n",
3068 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3069 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3070 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3071 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3072 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3073 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3075 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3076 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3083 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
3084 struct kvm_debug_guest
*dbg
)
3090 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
3098 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3099 * we have asm/x86/processor.h
3110 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3111 #ifdef CONFIG_X86_64
3112 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3114 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3119 * Translate a guest virtual address to a guest physical address.
3121 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
3122 struct kvm_translation
*tr
)
3124 unsigned long vaddr
= tr
->linear_address
;
3128 down_read(&vcpu
->kvm
->slots_lock
);
3129 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
3130 up_read(&vcpu
->kvm
->slots_lock
);
3131 tr
->physical_address
= gpa
;
3132 tr
->valid
= gpa
!= UNMAPPED_GVA
;
3140 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
3142 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
3146 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
3147 fpu
->fcw
= fxsave
->cwd
;
3148 fpu
->fsw
= fxsave
->swd
;
3149 fpu
->ftwx
= fxsave
->twd
;
3150 fpu
->last_opcode
= fxsave
->fop
;
3151 fpu
->last_ip
= fxsave
->rip
;
3152 fpu
->last_dp
= fxsave
->rdp
;
3153 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
3160 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
3162 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
3166 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
3167 fxsave
->cwd
= fpu
->fcw
;
3168 fxsave
->swd
= fpu
->fsw
;
3169 fxsave
->twd
= fpu
->ftwx
;
3170 fxsave
->fop
= fpu
->last_opcode
;
3171 fxsave
->rip
= fpu
->last_ip
;
3172 fxsave
->rdp
= fpu
->last_dp
;
3173 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
3180 void fx_init(struct kvm_vcpu
*vcpu
)
3182 unsigned after_mxcsr_mask
;
3184 /* Initialize guest FPU by resetting ours and saving into guest's */
3186 fx_save(&vcpu
->arch
.host_fx_image
);
3188 fx_save(&vcpu
->arch
.guest_fx_image
);
3189 fx_restore(&vcpu
->arch
.host_fx_image
);
3192 vcpu
->arch
.cr0
|= X86_CR0_ET
;
3193 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
3194 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
3195 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
3196 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
3198 EXPORT_SYMBOL_GPL(fx_init
);
3200 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
3202 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
3205 vcpu
->guest_fpu_loaded
= 1;
3206 fx_save(&vcpu
->arch
.host_fx_image
);
3207 fx_restore(&vcpu
->arch
.guest_fx_image
);
3209 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
3211 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
3213 if (!vcpu
->guest_fpu_loaded
)
3216 vcpu
->guest_fpu_loaded
= 0;
3217 fx_save(&vcpu
->arch
.guest_fx_image
);
3218 fx_restore(&vcpu
->arch
.host_fx_image
);
3219 ++vcpu
->stat
.fpu_reload
;
3221 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
3223 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
3225 kvm_x86_ops
->vcpu_free(vcpu
);
3228 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
3231 return kvm_x86_ops
->vcpu_create(kvm
, id
);
3234 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
3238 /* We do fxsave: this must be aligned. */
3239 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
3242 r
= kvm_arch_vcpu_reset(vcpu
);
3244 r
= kvm_mmu_setup(vcpu
);
3251 kvm_x86_ops
->vcpu_free(vcpu
);
3255 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
3258 kvm_mmu_unload(vcpu
);
3261 kvm_x86_ops
->vcpu_free(vcpu
);
3264 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
3266 return kvm_x86_ops
->vcpu_reset(vcpu
);
3269 void kvm_arch_hardware_enable(void *garbage
)
3271 kvm_x86_ops
->hardware_enable(garbage
);
3274 void kvm_arch_hardware_disable(void *garbage
)
3276 kvm_x86_ops
->hardware_disable(garbage
);
3279 int kvm_arch_hardware_setup(void)
3281 return kvm_x86_ops
->hardware_setup();
3284 void kvm_arch_hardware_unsetup(void)
3286 kvm_x86_ops
->hardware_unsetup();
3289 void kvm_arch_check_processor_compat(void *rtn
)
3291 kvm_x86_ops
->check_processor_compatibility(rtn
);
3294 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
3300 BUG_ON(vcpu
->kvm
== NULL
);
3303 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
3304 if (!irqchip_in_kernel(kvm
) || vcpu
->vcpu_id
== 0)
3305 vcpu
->arch
.mp_state
= VCPU_MP_STATE_RUNNABLE
;
3307 vcpu
->arch
.mp_state
= VCPU_MP_STATE_UNINITIALIZED
;
3309 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
3314 vcpu
->arch
.pio_data
= page_address(page
);
3316 r
= kvm_mmu_create(vcpu
);
3318 goto fail_free_pio_data
;
3320 if (irqchip_in_kernel(kvm
)) {
3321 r
= kvm_create_lapic(vcpu
);
3323 goto fail_mmu_destroy
;
3329 kvm_mmu_destroy(vcpu
);
3331 free_page((unsigned long)vcpu
->arch
.pio_data
);
3336 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
3338 kvm_free_lapic(vcpu
);
3339 kvm_mmu_destroy(vcpu
);
3340 free_page((unsigned long)vcpu
->arch
.pio_data
);
3343 struct kvm
*kvm_arch_create_vm(void)
3345 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
3348 return ERR_PTR(-ENOMEM
);
3350 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
3355 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
3358 kvm_mmu_unload(vcpu
);
3362 static void kvm_free_vcpus(struct kvm
*kvm
)
3367 * Unpin any mmu pages first.
3369 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
3371 kvm_unload_vcpu_mmu(kvm
->vcpus
[i
]);
3372 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
3373 if (kvm
->vcpus
[i
]) {
3374 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
3375 kvm
->vcpus
[i
] = NULL
;
3381 void kvm_arch_destroy_vm(struct kvm
*kvm
)
3384 kfree(kvm
->arch
.vpic
);
3385 kfree(kvm
->arch
.vioapic
);
3386 kvm_free_vcpus(kvm
);
3387 kvm_free_physmem(kvm
);
3391 int kvm_arch_set_memory_region(struct kvm
*kvm
,
3392 struct kvm_userspace_memory_region
*mem
,
3393 struct kvm_memory_slot old
,
3396 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
3397 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
3399 /*To keep backward compatibility with older userspace,
3400 *x86 needs to hanlde !user_alloc case.
3403 if (npages
&& !old
.rmap
) {
3404 down_write(¤t
->mm
->mmap_sem
);
3405 memslot
->userspace_addr
= do_mmap(NULL
, 0,
3407 PROT_READ
| PROT_WRITE
,
3408 MAP_SHARED
| MAP_ANONYMOUS
,
3410 up_write(¤t
->mm
->mmap_sem
);
3412 if (IS_ERR((void *)memslot
->userspace_addr
))
3413 return PTR_ERR((void *)memslot
->userspace_addr
);
3415 if (!old
.user_alloc
&& old
.rmap
) {
3418 down_write(¤t
->mm
->mmap_sem
);
3419 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
3420 old
.npages
* PAGE_SIZE
);
3421 up_write(¤t
->mm
->mmap_sem
);
3424 "kvm_vm_ioctl_set_memory_region: "
3425 "failed to munmap memory\n");
3430 if (!kvm
->arch
.n_requested_mmu_pages
) {
3431 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
3432 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
3435 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
3436 kvm_flush_remote_tlbs(kvm
);
3441 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
3443 return vcpu
->arch
.mp_state
== VCPU_MP_STATE_RUNNABLE
3444 || vcpu
->arch
.mp_state
== VCPU_MP_STATE_SIPI_RECEIVED
;
3447 static void vcpu_kick_intr(void *info
)
3450 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)info
;
3451 printk(KERN_DEBUG
"vcpu_kick_intr %p \n", vcpu
);
3455 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3457 int ipi_pcpu
= vcpu
->cpu
;
3459 if (waitqueue_active(&vcpu
->wq
)) {
3460 wake_up_interruptible(&vcpu
->wq
);
3461 ++vcpu
->stat
.halt_wakeup
;
3463 if (vcpu
->guest_mode
)
3464 smp_call_function_single(ipi_pcpu
, vcpu_kick_intr
, vcpu
, 0, 0);