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
199 /**
200 * kvm_arm_supports_user_irq
201 *
202 * Not all KVM implementations support notifications for kernel generated
203 * interrupt events to user space. This function indicates whether the current
204 * KVM implementation does support them.
205 *
206 * Returns: true if KVM supports using kernel generated IRQs from user space
207 */
208 bool kvm_arm_supports_user_irq(void);
209
210
211 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
212 int kvm_on_sigbus(int code, void *addr);
213
214 int kvm_check_extension(KVMState *s, unsigned int extension);
215
216 int kvm_vm_ioctl(KVMState *s, unsigned long type, ...);
217
218 void kvm_flush_coalesced_mmio_buffer(void);
219
220 #ifdef COMPILING_PER_TARGET
221 #include "cpu.h"
222
223 /**
224 * kvm_update_guest_debug(): ensure KVM debug structures updated
225 * @cs: the CPUState for this cpu
226 * @reinject_trap: KVM trap injection control
227 *
228 * There are usually per-arch specifics which will be handled by
229 * calling down to kvm_arch_update_guest_debug after the generic
230 * fields have been set.
231 */
232 #ifdef TARGET_KVM_HAVE_GUEST_DEBUG
233 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap);
234 #else
kvm_update_guest_debug(CPUState * cpu,unsigned long reinject_trap)235 static inline int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
236 {
237 return -EINVAL;
238 }
239 #endif
240
241 /* internal API */
242
243 int kvm_ioctl(KVMState *s, unsigned long type, ...);
244
245 int kvm_vcpu_ioctl(CPUState *cpu, unsigned long type, ...);
246
247 /**
248 * kvm_device_ioctl - call an ioctl on a kvm device
249 * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE
250 * @type: The device-ctrl ioctl number
251 *
252 * Returns: -errno on error, nonnegative on success
253 */
254 int kvm_device_ioctl(int fd, unsigned long type, ...);
255
256 /**
257 * kvm_vm_check_attr - check for existence of a specific vm attribute
258 * @s: The KVMState pointer
259 * @group: the group
260 * @attr: the attribute of that group to query for
261 *
262 * Returns: 1 if the attribute exists
263 * 0 if the attribute either does not exist or if the vm device
264 * interface is unavailable
265 */
266 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr);
267
268 /**
269 * kvm_device_check_attr - check for existence of a specific device attribute
270 * @fd: The device file descriptor
271 * @group: the group
272 * @attr: the attribute of that group to query for
273 *
274 * Returns: 1 if the attribute exists
275 * 0 if the attribute either does not exist or if the vm device
276 * interface is unavailable
277 */
278 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr);
279
280 /**
281 * kvm_device_access - set or get value of a specific device attribute
282 * @fd: The device file descriptor
283 * @group: the group
284 * @attr: the attribute of that group to set or get
285 * @val: pointer to a storage area for the value
286 * @write: true for set and false for get operation
287 * @errp: error object handle
288 *
289 * Returns: 0 on success
290 * < 0 on error
291 * Use kvm_device_check_attr() in order to check for the availability
292 * of optional attributes.
293 */
294 int kvm_device_access(int fd, int group, uint64_t attr,
295 void *val, bool write, Error **errp);
296
297 /**
298 * kvm_create_device - create a KVM device for the device control API
299 * @KVMState: The KVMState pointer
300 * @type: The KVM device type (see Documentation/virtual/kvm/devices in the
301 * kernel source)
302 * @test: If true, only test if device can be created, but don't actually
303 * create the device.
304 *
305 * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd;
306 */
307 int kvm_create_device(KVMState *s, uint64_t type, bool test);
308
309 /**
310 * kvm_device_supported - probe whether KVM supports specific device
311 *
312 * @vmfd: The fd handler for VM
313 * @type: type of device
314 *
315 * @return: true if supported, otherwise false.
316 */
317 bool kvm_device_supported(int vmfd, uint64_t type);
318
319 /**
320 * kvm_create_vcpu - Gets a parked KVM vCPU or creates a KVM vCPU
321 * @cpu: QOM CPUState object for which KVM vCPU has to be fetched/created.
322 *
323 * @returns: 0 when success, errno (<0) when failed.
324 */
325 int kvm_create_vcpu(CPUState *cpu);
326
327 /**
328 * kvm_park_vcpu - Park QEMU KVM vCPU context
329 * @cpu: QOM CPUState object for which QEMU KVM vCPU context has to be parked.
330 *
331 * @returns: none
332 */
333 void kvm_park_vcpu(CPUState *cpu);
334
335 /**
336 * kvm_unpark_vcpu - unpark QEMU KVM vCPU context
337 * @s: KVM State
338 * @vcpu_id: Architecture vCPU ID of the parked vCPU
339 *
340 * @returns: KVM fd
341 */
342 int kvm_unpark_vcpu(KVMState *s, unsigned long vcpu_id);
343
344 /**
345 * kvm_create_and_park_vcpu - Create and park a KVM vCPU
346 * @cpu: QOM CPUState object for which KVM vCPU has to be created and parked.
347 *
348 * @returns: 0 when success, errno (<0) when failed.
349 */
350 int kvm_create_and_park_vcpu(CPUState *cpu);
351
352 /* Arch specific hooks */
353
354 extern const KVMCapabilityInfo kvm_arch_required_capabilities[];
355
356 void kvm_arch_accel_class_init(ObjectClass *oc);
357
358 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run);
359 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run);
360
361 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run);
362
363 int kvm_arch_process_async_events(CPUState *cpu);
364
365 int kvm_arch_get_registers(CPUState *cpu, Error **errp);
366
367 /* state subset only touched by the VCPU itself during runtime */
368 #define KVM_PUT_RUNTIME_STATE 1
369 /* state subset modified during VCPU reset */
370 #define KVM_PUT_RESET_STATE 2
371 /* full state set, modified during initialization or on vmload */
372 #define KVM_PUT_FULL_STATE 3
373
374 int kvm_arch_put_registers(CPUState *cpu, int level, Error **errp);
375
376 int kvm_arch_get_default_type(MachineState *ms);
377
378 int kvm_arch_init(MachineState *ms, KVMState *s);
379
380 int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp);
381 int kvm_arch_init_vcpu(CPUState *cpu);
382 int kvm_arch_destroy_vcpu(CPUState *cpu);
383
384 #ifdef TARGET_KVM_HAVE_RESET_PARKED_VCPU
385 void kvm_arch_reset_parked_vcpu(unsigned long vcpu_id, int kvm_fd);
386 #else
kvm_arch_reset_parked_vcpu(unsigned long vcpu_id,int kvm_fd)387 static inline void kvm_arch_reset_parked_vcpu(unsigned long vcpu_id, int kvm_fd)
388 {
389 }
390 #endif
391
392 bool kvm_vcpu_id_is_valid(int vcpu_id);
393
394 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */
395 unsigned long kvm_arch_vcpu_id(CPUState *cpu);
396
397 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
398
399 void kvm_arch_init_irq_routing(KVMState *s);
400
401 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
402 uint64_t address, uint32_t data, PCIDevice *dev);
403
404 /* Notify arch about newly added MSI routes */
405 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
406 int vector, PCIDevice *dev);
407 /* Notify arch about released MSI routes */
408 int kvm_arch_release_virq_post(int virq);
409
410 int kvm_arch_msi_data_to_gsi(uint32_t data);
411
412 int kvm_set_irq(KVMState *s, int irq, int level);
413 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg);
414
415 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin);
416
417 void kvm_irqchip_add_change_notifier(Notifier *n);
418 void kvm_irqchip_remove_change_notifier(Notifier *n);
419 void kvm_irqchip_change_notify(void);
420
421 struct kvm_guest_debug;
422 struct kvm_debug_exit_arch;
423
424 struct kvm_sw_breakpoint {
425 vaddr pc;
426 vaddr saved_insn;
427 int use_count;
428 QTAILQ_ENTRY(kvm_sw_breakpoint) entry;
429 };
430
431 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
432 vaddr pc);
433
434 int kvm_sw_breakpoints_active(CPUState *cpu);
435
436 int kvm_arch_insert_sw_breakpoint(CPUState *cpu,
437 struct kvm_sw_breakpoint *bp);
438 int kvm_arch_remove_sw_breakpoint(CPUState *cpu,
439 struct kvm_sw_breakpoint *bp);
440 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type);
441 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type);
442 void kvm_arch_remove_all_hw_breakpoints(void);
443
444 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg);
445
446 bool kvm_arch_stop_on_emulation_error(CPUState *cpu);
447
448 int kvm_vm_check_extension(KVMState *s, unsigned int extension);
449
450 #define kvm_vm_enable_cap(s, capability, cap_flags, ...) \
451 ({ \
452 struct kvm_enable_cap cap = { \
453 .cap = capability, \
454 .flags = cap_flags, \
455 }; \
456 uint64_t args_tmp[] = { __VA_ARGS__ }; \
457 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \
458 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \
459 kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap); \
460 })
461
462 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...) \
463 ({ \
464 struct kvm_enable_cap cap = { \
465 .cap = capability, \
466 .flags = cap_flags, \
467 }; \
468 uint64_t args_tmp[] = { __VA_ARGS__ }; \
469 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \
470 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \
471 kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap); \
472 })
473
474 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len);
475
476 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr,
477 hwaddr *phys_addr);
478
479 #endif /* COMPILING_PER_TARGET */
480
481 void kvm_cpu_synchronize_state(CPUState *cpu);
482
483 void kvm_init_cpu_signals(CPUState *cpu);
484
485 /**
486 * kvm_irqchip_add_msi_route - Add MSI route for specific vector
487 * @c: KVMRouteChange instance.
488 * @vector: which vector to add. This can be either MSI/MSIX
489 * vector. The function will automatically detect whether
490 * MSI/MSIX is enabled, and fetch corresponding MSI
491 * message.
492 * @dev: Owner PCI device to add the route. If @dev is specified
493 * as @NULL, an empty MSI message will be inited.
494 * @return: virq (>=0) when success, errno (<0) when failed.
495 */
496 int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev);
497 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
498 PCIDevice *dev);
499 void kvm_irqchip_commit_routes(KVMState *s);
500
kvm_irqchip_begin_route_changes(KVMState * s)501 static inline KVMRouteChange kvm_irqchip_begin_route_changes(KVMState *s)
502 {
503 return (KVMRouteChange) { .s = s, .changes = 0 };
504 }
505
kvm_irqchip_commit_route_changes(KVMRouteChange * c)506 static inline void kvm_irqchip_commit_route_changes(KVMRouteChange *c)
507 {
508 if (c->changes) {
509 kvm_irqchip_commit_routes(c->s);
510 c->changes = 0;
511 }
512 }
513
514 int kvm_irqchip_get_virq(KVMState *s);
515 void kvm_irqchip_release_virq(KVMState *s, int virq);
516
517 void kvm_add_routing_entry(KVMState *s,
518 struct kvm_irq_routing_entry *entry);
519
520 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
521 EventNotifier *rn, int virq);
522 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
523 int virq);
524 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
525 EventNotifier *rn, qemu_irq irq);
526 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
527 qemu_irq irq);
528 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi);
529 void kvm_init_irq_routing(KVMState *s);
530
531 bool kvm_kernel_irqchip_allowed(void);
532 bool kvm_kernel_irqchip_required(void);
533 bool kvm_kernel_irqchip_split(void);
534
535 /**
536 * kvm_arch_irqchip_create:
537 * @KVMState: The KVMState pointer
538 *
539 * Allow architectures to create an in-kernel irq chip themselves.
540 *
541 * Returns: < 0: error
542 * 0: irq chip was not created
543 * > 0: irq chip was created
544 */
545 int kvm_arch_irqchip_create(KVMState *s);
546
547 /**
548 * kvm_set_one_reg - set a register value in KVM via KVM_SET_ONE_REG ioctl
549 * @id: The register ID
550 * @source: The pointer to the value to be set. It must point to a variable
551 * of the correct type/size for the register being accessed.
552 *
553 * Returns: 0 on success, or a negative errno on failure.
554 */
555 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source);
556
557 /**
558 * kvm_get_one_reg - get a register value from KVM via KVM_GET_ONE_REG ioctl
559 * @id: The register ID
560 * @target: The pointer where the value is to be stored. It must point to a
561 * variable of the correct type/size for the register being accessed.
562 *
563 * Returns: 0 on success, or a negative errno on failure.
564 */
565 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target);
566
567 /* Notify resamplefd for EOI of specific interrupts. */
568 void kvm_resample_fd_notify(int gsi);
569
570 bool kvm_dirty_ring_enabled(void);
571
572 uint32_t kvm_dirty_ring_size(void);
573
574 void kvm_mark_guest_state_protected(void);
575
576 /**
577 * kvm_hwpoisoned_mem - indicate if there is any hwpoisoned page
578 * reported for the VM.
579 */
580 bool kvm_hwpoisoned_mem(void);
581
582 int kvm_create_guest_memfd(uint64_t size, uint64_t flags, Error **errp);
583
584 int kvm_set_memory_attributes_private(hwaddr start, uint64_t size);
585 int kvm_set_memory_attributes_shared(hwaddr start, uint64_t size);
586
587 int kvm_convert_memory(hwaddr start, hwaddr size, bool to_private);
588
589 #endif
590