1 /*
2 * QEMU KVM support
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
11 *
12 */
13
14 /* header to be included in non-KVM-specific code */
15
16 #ifndef QEMU_KVM_H
17 #define QEMU_KVM_H
18
19 #include "exec/memattrs.h"
20 #include "qemu/accel.h"
21 #include "qom/object.h"
22
23 #ifdef COMPILING_PER_TARGET
24 # ifdef CONFIG_KVM
25 # include <linux/kvm.h>
26 # define CONFIG_KVM_IS_POSSIBLE
27 # endif
28 #else
29 # define CONFIG_KVM_IS_POSSIBLE
30 #endif
31
32 #ifdef CONFIG_KVM_IS_POSSIBLE
33
34 extern bool kvm_allowed;
35 extern bool kvm_kernel_irqchip;
36 extern bool kvm_split_irqchip;
37 extern bool kvm_async_interrupts_allowed;
38 extern bool kvm_halt_in_kernel_allowed;
39 extern bool kvm_resamplefds_allowed;
40 extern bool kvm_msi_via_irqfd_allowed;
41 extern bool kvm_gsi_routing_allowed;
42 extern bool kvm_gsi_direct_mapping;
43 extern bool kvm_readonly_mem_allowed;
44 extern bool kvm_msi_use_devid;
45 extern bool kvm_pre_fault_memory_supported;
46
47 #define kvm_enabled() (kvm_allowed)
48 /**
49 * kvm_irqchip_in_kernel:
50 *
51 * Returns: true if an in-kernel irqchip was created.
52 * What this actually means is architecture and machine model
53 * specific: on PC, for instance, it means that the LAPIC
54 * is in kernel. This function should never be used from generic
55 * target-independent code: use one of the following functions or
56 * some other specific check instead.
57 */
58 #define kvm_irqchip_in_kernel() (kvm_kernel_irqchip)
59
60 /**
61 * kvm_irqchip_is_split:
62 *
63 * Returns: true if the irqchip implementation is split between
64 * user and kernel space. The details are architecture and
65 * machine specific. On PC, it means that the PIC, IOAPIC, and
66 * PIT are in user space while the LAPIC is in the kernel.
67 */
68 #define kvm_irqchip_is_split() (kvm_split_irqchip)
69
70 /**
71 * kvm_async_interrupts_enabled:
72 *
73 * Returns: true if we can deliver interrupts to KVM
74 * asynchronously (ie by ioctl from any thread at any time)
75 * rather than having to do interrupt delivery synchronously
76 * (where the vcpu must be stopped at a suitable point first).
77 */
78 #define kvm_async_interrupts_enabled() (kvm_async_interrupts_allowed)
79
80 /**
81 * kvm_halt_in_kernel
82 *
83 * Returns: true if halted cpus should still get a KVM_RUN ioctl to run
84 * inside of kernel space. This only works if MP state is implemented.
85 */
86 #define kvm_halt_in_kernel() (kvm_halt_in_kernel_allowed)
87
88 /**
89 * kvm_irqfds_enabled:
90 *
91 * Returns: true if we can use irqfds to inject interrupts into
92 * a KVM CPU (ie the kernel supports irqfds and we are running
93 * with a configuration where it is meaningful to use them).
94 *
95 * Always available if running with in-kernel irqchip.
96 */
97 #define kvm_irqfds_enabled() kvm_irqchip_in_kernel()
98
99 /**
100 * kvm_resamplefds_enabled:
101 *
102 * Returns: true if we can use resamplefds to inject interrupts into
103 * a KVM CPU (ie the kernel supports resamplefds and we are running
104 * with a configuration where it is meaningful to use them).
105 */
106 #define kvm_resamplefds_enabled() (kvm_resamplefds_allowed)
107
108 /**
109 * kvm_msi_via_irqfd_enabled:
110 *
111 * Returns: true if we can route a PCI MSI (Message Signaled Interrupt)
112 * to a KVM CPU via an irqfd. This requires that the kernel supports
113 * this and that we're running in a configuration that permits it.
114 */
115 #define kvm_msi_via_irqfd_enabled() (kvm_msi_via_irqfd_allowed)
116
117 /**
118 * kvm_gsi_routing_enabled:
119 *
120 * Returns: true if GSI routing is enabled (ie the kernel supports
121 * it and we're running in a configuration that permits it).
122 */
123 #define kvm_gsi_routing_enabled() (kvm_gsi_routing_allowed)
124
125 /**
126 * kvm_gsi_direct_mapping:
127 *
128 * Returns: true if GSI direct mapping is enabled.
129 */
130 #define kvm_gsi_direct_mapping() (kvm_gsi_direct_mapping)
131
132 /**
133 * kvm_readonly_mem_enabled:
134 *
135 * Returns: true if KVM readonly memory is enabled (ie the kernel
136 * supports it and we're running in a configuration that permits it).
137 */
138 #define kvm_readonly_mem_enabled() (kvm_readonly_mem_allowed)
139
140 /**
141 * kvm_msi_devid_required:
142 * Returns: true if KVM requires a device id to be provided while
143 * defining an MSI routing entry.
144 */
145 #define kvm_msi_devid_required() (kvm_msi_use_devid)
146
147 #else
148
149 #define kvm_enabled() (0)
150 #define kvm_irqchip_in_kernel() (false)
151 #define kvm_irqchip_is_split() (false)
152 #define kvm_async_interrupts_enabled() (false)
153 #define kvm_halt_in_kernel() (false)
154 #define kvm_irqfds_enabled() (false)
155 #define kvm_resamplefds_enabled() (false)
156 #define kvm_msi_via_irqfd_enabled() (false)
157 #define kvm_gsi_routing_allowed() (false)
158 #define kvm_gsi_direct_mapping() (false)
159 #define kvm_readonly_mem_enabled() (false)
160 #define kvm_msi_devid_required() (false)
161
162 #endif /* CONFIG_KVM_IS_POSSIBLE */
163
164 struct kvm_run;
165 struct kvm_irq_routing_entry;
166
167 typedef struct KVMCapabilityInfo {
168 const char *name;
169 int value;
170 } KVMCapabilityInfo;
171
172 #define KVM_CAP_INFO(CAP) { "KVM_CAP_" stringify(CAP), KVM_CAP_##CAP }
173 #define KVM_CAP_LAST_INFO { NULL, 0 }
174
175 struct KVMState;
176
177 #define TYPE_KVM_ACCEL ACCEL_CLASS_NAME("kvm")
178 typedef struct KVMState KVMState;
179 DECLARE_INSTANCE_CHECKER(KVMState, KVM_STATE,
180 TYPE_KVM_ACCEL)
181
182 extern KVMState *kvm_state;
183 typedef struct Notifier Notifier;
184
185 typedef struct KVMRouteChange {
186 KVMState *s;
187 int changes;
188 } KVMRouteChange;
189
190 /* external API */
191
192 unsigned int kvm_get_max_memslots(void);
193 unsigned int kvm_get_free_memslots(void);
194 bool kvm_has_sync_mmu(void);
195 int kvm_has_vcpu_events(void);
196 int kvm_max_nested_state_length(void);
197 int kvm_has_gsi_routing(void);
198 void kvm_close(void);
199
200 /**
201 * kvm_arm_supports_user_irq
202 *
203 * Not all KVM implementations support notifications for kernel generated
204 * interrupt events to user space. This function indicates whether the current
205 * KVM implementation does support them.
206 *
207 * Returns: true if KVM supports using kernel generated IRQs from user space
208 */
209 bool kvm_arm_supports_user_irq(void);
210
211
212 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
213 int kvm_on_sigbus(int code, void *addr);
214
215 int kvm_check_extension(KVMState *s, unsigned int extension);
216
217 int kvm_vm_ioctl(KVMState *s, unsigned long type, ...);
218
219 void kvm_flush_coalesced_mmio_buffer(void);
220
221 #ifdef COMPILING_PER_TARGET
222 #include "cpu.h"
223
224 /**
225 * kvm_update_guest_debug(): ensure KVM debug structures updated
226 * @cs: the CPUState for this cpu
227 * @reinject_trap: KVM trap injection control
228 *
229 * There are usually per-arch specifics which will be handled by
230 * calling down to kvm_arch_update_guest_debug after the generic
231 * fields have been set.
232 */
233 #ifdef TARGET_KVM_HAVE_GUEST_DEBUG
234 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap);
235 #else
kvm_update_guest_debug(CPUState * cpu,unsigned long reinject_trap)236 static inline int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
237 {
238 return -EINVAL;
239 }
240 #endif
241
242 /* internal API */
243
244 int kvm_ioctl(KVMState *s, unsigned long type, ...);
245
246 int kvm_vcpu_ioctl(CPUState *cpu, unsigned long type, ...);
247
248 /**
249 * kvm_device_ioctl - call an ioctl on a kvm device
250 * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE
251 * @type: The device-ctrl ioctl number
252 *
253 * Returns: -errno on error, nonnegative on success
254 */
255 int kvm_device_ioctl(int fd, unsigned long type, ...);
256
257 /**
258 * kvm_vm_check_attr - check for existence of a specific vm attribute
259 * @s: The KVMState pointer
260 * @group: the group
261 * @attr: the attribute of that group to query for
262 *
263 * Returns: 1 if the attribute exists
264 * 0 if the attribute either does not exist or if the vm device
265 * interface is unavailable
266 */
267 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr);
268
269 /**
270 * kvm_device_check_attr - check for existence of a specific device attribute
271 * @fd: The device file descriptor
272 * @group: the group
273 * @attr: the attribute of that group to query for
274 *
275 * Returns: 1 if the attribute exists
276 * 0 if the attribute either does not exist or if the vm device
277 * interface is unavailable
278 */
279 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr);
280
281 /**
282 * kvm_device_access - set or get value of a specific device attribute
283 * @fd: The device file descriptor
284 * @group: the group
285 * @attr: the attribute of that group to set or get
286 * @val: pointer to a storage area for the value
287 * @write: true for set and false for get operation
288 * @errp: error object handle
289 *
290 * Returns: 0 on success
291 * < 0 on error
292 * Use kvm_device_check_attr() in order to check for the availability
293 * of optional attributes.
294 */
295 int kvm_device_access(int fd, int group, uint64_t attr,
296 void *val, bool write, Error **errp);
297
298 /**
299 * kvm_create_device - create a KVM device for the device control API
300 * @KVMState: The KVMState pointer
301 * @type: The KVM device type (see Documentation/virtual/kvm/devices in the
302 * kernel source)
303 * @test: If true, only test if device can be created, but don't actually
304 * create the device.
305 *
306 * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd;
307 */
308 int kvm_create_device(KVMState *s, uint64_t type, bool test);
309
310 /**
311 * kvm_device_supported - probe whether KVM supports specific device
312 *
313 * @vmfd: The fd handler for VM
314 * @type: type of device
315 *
316 * @return: true if supported, otherwise false.
317 */
318 bool kvm_device_supported(int vmfd, uint64_t type);
319
320 /**
321 * kvm_park_vcpu - Park QEMU KVM vCPU context
322 * @cpu: QOM CPUState object for which QEMU KVM vCPU context has to be parked.
323 *
324 * @returns: none
325 */
326 void kvm_park_vcpu(CPUState *cpu);
327
328 /**
329 * kvm_unpark_vcpu - unpark QEMU KVM vCPU context
330 * @s: KVM State
331 * @vcpu_id: Architecture vCPU ID of the parked vCPU
332 *
333 * @returns: KVM fd
334 */
335 int kvm_unpark_vcpu(KVMState *s, unsigned long vcpu_id);
336
337 /**
338 * kvm_create_and_park_vcpu - Create and park a KVM vCPU
339 * @cpu: QOM CPUState object for which KVM vCPU has to be created and parked.
340 *
341 * @returns: 0 when success, errno (<0) when failed.
342 */
343 int kvm_create_and_park_vcpu(CPUState *cpu);
344
345 /* Arch specific hooks */
346
347 extern const KVMCapabilityInfo kvm_arch_required_capabilities[];
348
349 void kvm_arch_accel_class_init(ObjectClass *oc);
350
351 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run);
352 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run);
353
354 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run);
355
356 int kvm_arch_process_async_events(CPUState *cpu);
357
358 int kvm_arch_get_registers(CPUState *cpu, Error **errp);
359
360 /* state subset only touched by the VCPU itself during runtime */
361 #define KVM_PUT_RUNTIME_STATE 1
362 /* state subset modified during VCPU reset */
363 #define KVM_PUT_RESET_STATE 2
364 /* full state set, modified during initialization or on vmload */
365 #define KVM_PUT_FULL_STATE 3
366
367 int kvm_arch_put_registers(CPUState *cpu, int level, Error **errp);
368
369 int kvm_arch_get_default_type(MachineState *ms);
370
371 int kvm_arch_init(MachineState *ms, KVMState *s);
372
373 int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp);
374 int kvm_arch_init_vcpu(CPUState *cpu);
375 int kvm_arch_destroy_vcpu(CPUState *cpu);
376
377 #ifdef TARGET_KVM_HAVE_RESET_PARKED_VCPU
378 void kvm_arch_reset_parked_vcpu(unsigned long vcpu_id, int kvm_fd);
379 #else
kvm_arch_reset_parked_vcpu(unsigned long vcpu_id,int kvm_fd)380 static inline void kvm_arch_reset_parked_vcpu(unsigned long vcpu_id, int kvm_fd)
381 {
382 }
383 #endif
384
385 bool kvm_vcpu_id_is_valid(int vcpu_id);
386
387 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */
388 unsigned long kvm_arch_vcpu_id(CPUState *cpu);
389
390 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
391
392 void kvm_arch_init_irq_routing(KVMState *s);
393
394 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
395 uint64_t address, uint32_t data, PCIDevice *dev);
396
397 /* Notify arch about newly added MSI routes */
398 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
399 int vector, PCIDevice *dev);
400 /* Notify arch about released MSI routes */
401 int kvm_arch_release_virq_post(int virq);
402
403 int kvm_arch_msi_data_to_gsi(uint32_t data);
404
405 int kvm_set_irq(KVMState *s, int irq, int level);
406 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg);
407
408 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin);
409
410 void kvm_irqchip_add_change_notifier(Notifier *n);
411 void kvm_irqchip_remove_change_notifier(Notifier *n);
412 void kvm_irqchip_change_notify(void);
413
414 struct kvm_guest_debug;
415 struct kvm_debug_exit_arch;
416
417 struct kvm_sw_breakpoint {
418 vaddr pc;
419 vaddr saved_insn;
420 int use_count;
421 QTAILQ_ENTRY(kvm_sw_breakpoint) entry;
422 };
423
424 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
425 vaddr pc);
426
427 int kvm_sw_breakpoints_active(CPUState *cpu);
428
429 int kvm_arch_insert_sw_breakpoint(CPUState *cpu,
430 struct kvm_sw_breakpoint *bp);
431 int kvm_arch_remove_sw_breakpoint(CPUState *cpu,
432 struct kvm_sw_breakpoint *bp);
433 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type);
434 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type);
435 void kvm_arch_remove_all_hw_breakpoints(void);
436
437 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg);
438
439 bool kvm_arch_stop_on_emulation_error(CPUState *cpu);
440
441 int kvm_vm_check_extension(KVMState *s, unsigned int extension);
442
443 #define kvm_vm_enable_cap(s, capability, cap_flags, ...) \
444 ({ \
445 struct kvm_enable_cap cap = { \
446 .cap = capability, \
447 .flags = cap_flags, \
448 }; \
449 uint64_t args_tmp[] = { __VA_ARGS__ }; \
450 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \
451 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \
452 kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap); \
453 })
454
455 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...) \
456 ({ \
457 struct kvm_enable_cap cap = { \
458 .cap = capability, \
459 .flags = cap_flags, \
460 }; \
461 uint64_t args_tmp[] = { __VA_ARGS__ }; \
462 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \
463 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \
464 kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap); \
465 })
466
467 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len);
468
469 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr,
470 hwaddr *phys_addr);
471
472 #endif /* COMPILING_PER_TARGET */
473
474 void kvm_cpu_synchronize_state(CPUState *cpu);
475
476 void kvm_init_cpu_signals(CPUState *cpu);
477
478 /**
479 * kvm_irqchip_add_msi_route - Add MSI route for specific vector
480 * @c: KVMRouteChange instance.
481 * @vector: which vector to add. This can be either MSI/MSIX
482 * vector. The function will automatically detect whether
483 * MSI/MSIX is enabled, and fetch corresponding MSI
484 * message.
485 * @dev: Owner PCI device to add the route. If @dev is specified
486 * as @NULL, an empty MSI message will be inited.
487 * @return: virq (>=0) when success, errno (<0) when failed.
488 */
489 int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev);
490 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
491 PCIDevice *dev);
492 void kvm_irqchip_commit_routes(KVMState *s);
493
kvm_irqchip_begin_route_changes(KVMState * s)494 static inline KVMRouteChange kvm_irqchip_begin_route_changes(KVMState *s)
495 {
496 return (KVMRouteChange) { .s = s, .changes = 0 };
497 }
498
kvm_irqchip_commit_route_changes(KVMRouteChange * c)499 static inline void kvm_irqchip_commit_route_changes(KVMRouteChange *c)
500 {
501 if (c->changes) {
502 kvm_irqchip_commit_routes(c->s);
503 c->changes = 0;
504 }
505 }
506
507 int kvm_irqchip_get_virq(KVMState *s);
508 void kvm_irqchip_release_virq(KVMState *s, int virq);
509
510 void kvm_add_routing_entry(KVMState *s,
511 struct kvm_irq_routing_entry *entry);
512
513 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
514 EventNotifier *rn, int virq);
515 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
516 int virq);
517 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
518 EventNotifier *rn, qemu_irq irq);
519 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
520 qemu_irq irq);
521 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi);
522 void kvm_init_irq_routing(KVMState *s);
523
524 bool kvm_kernel_irqchip_allowed(void);
525 bool kvm_kernel_irqchip_required(void);
526 bool kvm_kernel_irqchip_split(void);
527
528 /**
529 * kvm_arch_irqchip_create:
530 * @KVMState: The KVMState pointer
531 *
532 * Allow architectures to create an in-kernel irq chip themselves.
533 *
534 * Returns: < 0: error
535 * 0: irq chip was not created
536 * > 0: irq chip was created
537 */
538 int kvm_arch_irqchip_create(KVMState *s);
539
540 /**
541 * kvm_set_one_reg - set a register value in KVM via KVM_SET_ONE_REG ioctl
542 * @id: The register ID
543 * @source: The pointer to the value to be set. It must point to a variable
544 * of the correct type/size for the register being accessed.
545 *
546 * Returns: 0 on success, or a negative errno on failure.
547 */
548 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source);
549
550 /**
551 * kvm_get_one_reg - get a register value from KVM via KVM_GET_ONE_REG ioctl
552 * @id: The register ID
553 * @target: The pointer where the value is to be stored. It must point to a
554 * variable of the correct type/size for the register being accessed.
555 *
556 * Returns: 0 on success, or a negative errno on failure.
557 */
558 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target);
559
560 /* Notify resamplefd for EOI of specific interrupts. */
561 void kvm_resample_fd_notify(int gsi);
562
563 bool kvm_dirty_ring_enabled(void);
564
565 uint32_t kvm_dirty_ring_size(void);
566
567 void kvm_mark_guest_state_protected(void);
568
569 /**
570 * kvm_hwpoisoned_mem - indicate if there is any hwpoisoned page
571 * reported for the VM.
572 */
573 bool kvm_hwpoisoned_mem(void);
574
575 int kvm_create_guest_memfd(uint64_t size, uint64_t flags, Error **errp);
576
577 int kvm_set_memory_attributes_private(hwaddr start, uint64_t size);
578 int kvm_set_memory_attributes_shared(hwaddr start, uint64_t size);
579
580 int kvm_convert_memory(hwaddr start, hwaddr size, bool to_private);
581
582 #endif
583