block: Rename raw-{posix,win32} to file-*.c
[qemu.git] / kvm-all.c
1 /*
2 * QEMU KVM support
3 *
4 * Copyright IBM, Corp. 2008
5 * Red Hat, Inc. 2008
6 *
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Glauber Costa <gcosta@redhat.com>
10 *
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
13 *
14 */
15
16 #include "qemu/osdep.h"
17 #include <sys/ioctl.h>
18
19 #include <linux/kvm.h>
20
21 #include "qemu-common.h"
22 #include "qemu/atomic.h"
23 #include "qemu/option.h"
24 #include "qemu/config-file.h"
25 #include "qemu/error-report.h"
26 #include "hw/hw.h"
27 #include "hw/pci/msi.h"
28 #include "hw/pci/msix.h"
29 #include "hw/s390x/adapter.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm_int.h"
32 #include "qemu/bswap.h"
33 #include "exec/memory.h"
34 #include "exec/ram_addr.h"
35 #include "exec/address-spaces.h"
36 #include "qemu/event_notifier.h"
37 #include "trace.h"
38 #include "hw/irq.h"
39
40 #include "hw/boards.h"
41
42 /* This check must be after config-host.h is included */
43 #ifdef CONFIG_EVENTFD
44 #include <sys/eventfd.h>
45 #endif
46
47 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
48 * need to use the real host PAGE_SIZE, as that's what KVM will use.
49 */
50 #define PAGE_SIZE getpagesize()
51
52 //#define DEBUG_KVM
53
54 #ifdef DEBUG_KVM
55 #define DPRINTF(fmt, ...) \
56 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
57 #else
58 #define DPRINTF(fmt, ...) \
59 do { } while (0)
60 #endif
61
62 #define KVM_MSI_HASHTAB_SIZE 256
63
64 struct KVMParkedVcpu {
65 unsigned long vcpu_id;
66 int kvm_fd;
67 QLIST_ENTRY(KVMParkedVcpu) node;
68 };
69
70 struct KVMState
71 {
72 AccelState parent_obj;
73
74 int nr_slots;
75 int fd;
76 int vmfd;
77 int coalesced_mmio;
78 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
79 bool coalesced_flush_in_progress;
80 int broken_set_mem_region;
81 int vcpu_events;
82 int robust_singlestep;
83 int debugregs;
84 #ifdef KVM_CAP_SET_GUEST_DEBUG
85 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
86 #endif
87 int many_ioeventfds;
88 int intx_set_mask;
89 /* The man page (and posix) say ioctl numbers are signed int, but
90 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
91 * unsigned, and treating them as signed here can break things */
92 unsigned irq_set_ioctl;
93 unsigned int sigmask_len;
94 GHashTable *gsimap;
95 #ifdef KVM_CAP_IRQ_ROUTING
96 struct kvm_irq_routing *irq_routes;
97 int nr_allocated_irq_routes;
98 unsigned long *used_gsi_bitmap;
99 unsigned int gsi_count;
100 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
101 #endif
102 KVMMemoryListener memory_listener;
103 QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus;
104 };
105
106 KVMState *kvm_state;
107 bool kvm_kernel_irqchip;
108 bool kvm_split_irqchip;
109 bool kvm_async_interrupts_allowed;
110 bool kvm_halt_in_kernel_allowed;
111 bool kvm_eventfds_allowed;
112 bool kvm_irqfds_allowed;
113 bool kvm_resamplefds_allowed;
114 bool kvm_msi_via_irqfd_allowed;
115 bool kvm_gsi_routing_allowed;
116 bool kvm_gsi_direct_mapping;
117 bool kvm_allowed;
118 bool kvm_readonly_mem_allowed;
119 bool kvm_vm_attributes_allowed;
120 bool kvm_direct_msi_allowed;
121 bool kvm_ioeventfd_any_length_allowed;
122 bool kvm_msi_use_devid;
123
124 static const KVMCapabilityInfo kvm_required_capabilites[] = {
125 KVM_CAP_INFO(USER_MEMORY),
126 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
127 KVM_CAP_LAST_INFO
128 };
129
130 int kvm_get_max_memslots(void)
131 {
132 KVMState *s = KVM_STATE(current_machine->accelerator);
133
134 return s->nr_slots;
135 }
136
137 static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml)
138 {
139 KVMState *s = kvm_state;
140 int i;
141
142 for (i = 0; i < s->nr_slots; i++) {
143 if (kml->slots[i].memory_size == 0) {
144 return &kml->slots[i];
145 }
146 }
147
148 return NULL;
149 }
150
151 bool kvm_has_free_slot(MachineState *ms)
152 {
153 KVMState *s = KVM_STATE(ms->accelerator);
154
155 return kvm_get_free_slot(&s->memory_listener);
156 }
157
158 static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml)
159 {
160 KVMSlot *slot = kvm_get_free_slot(kml);
161
162 if (slot) {
163 return slot;
164 }
165
166 fprintf(stderr, "%s: no free slot available\n", __func__);
167 abort();
168 }
169
170 static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml,
171 hwaddr start_addr,
172 hwaddr end_addr)
173 {
174 KVMState *s = kvm_state;
175 int i;
176
177 for (i = 0; i < s->nr_slots; i++) {
178 KVMSlot *mem = &kml->slots[i];
179
180 if (start_addr == mem->start_addr &&
181 end_addr == mem->start_addr + mem->memory_size) {
182 return mem;
183 }
184 }
185
186 return NULL;
187 }
188
189 /*
190 * Find overlapping slot with lowest start address
191 */
192 static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml,
193 hwaddr start_addr,
194 hwaddr end_addr)
195 {
196 KVMState *s = kvm_state;
197 KVMSlot *found = NULL;
198 int i;
199
200 for (i = 0; i < s->nr_slots; i++) {
201 KVMSlot *mem = &kml->slots[i];
202
203 if (mem->memory_size == 0 ||
204 (found && found->start_addr < mem->start_addr)) {
205 continue;
206 }
207
208 if (end_addr > mem->start_addr &&
209 start_addr < mem->start_addr + mem->memory_size) {
210 found = mem;
211 }
212 }
213
214 return found;
215 }
216
217 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
218 hwaddr *phys_addr)
219 {
220 KVMMemoryListener *kml = &s->memory_listener;
221 int i;
222
223 for (i = 0; i < s->nr_slots; i++) {
224 KVMSlot *mem = &kml->slots[i];
225
226 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
227 *phys_addr = mem->start_addr + (ram - mem->ram);
228 return 1;
229 }
230 }
231
232 return 0;
233 }
234
235 static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot)
236 {
237 KVMState *s = kvm_state;
238 struct kvm_userspace_memory_region mem;
239
240 mem.slot = slot->slot | (kml->as_id << 16);
241 mem.guest_phys_addr = slot->start_addr;
242 mem.userspace_addr = (unsigned long)slot->ram;
243 mem.flags = slot->flags;
244
245 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
246 /* Set the slot size to 0 before setting the slot to the desired
247 * value. This is needed based on KVM commit 75d61fbc. */
248 mem.memory_size = 0;
249 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
250 }
251 mem.memory_size = slot->memory_size;
252 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
253 }
254
255 int kvm_destroy_vcpu(CPUState *cpu)
256 {
257 KVMState *s = kvm_state;
258 long mmap_size;
259 struct KVMParkedVcpu *vcpu = NULL;
260 int ret = 0;
261
262 DPRINTF("kvm_destroy_vcpu\n");
263
264 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
265 if (mmap_size < 0) {
266 ret = mmap_size;
267 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
268 goto err;
269 }
270
271 ret = munmap(cpu->kvm_run, mmap_size);
272 if (ret < 0) {
273 goto err;
274 }
275
276 vcpu = g_malloc0(sizeof(*vcpu));
277 vcpu->vcpu_id = kvm_arch_vcpu_id(cpu);
278 vcpu->kvm_fd = cpu->kvm_fd;
279 QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node);
280 err:
281 return ret;
282 }
283
284 static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id)
285 {
286 struct KVMParkedVcpu *cpu;
287
288 QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) {
289 if (cpu->vcpu_id == vcpu_id) {
290 int kvm_fd;
291
292 QLIST_REMOVE(cpu, node);
293 kvm_fd = cpu->kvm_fd;
294 g_free(cpu);
295 return kvm_fd;
296 }
297 }
298
299 return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id);
300 }
301
302 int kvm_init_vcpu(CPUState *cpu)
303 {
304 KVMState *s = kvm_state;
305 long mmap_size;
306 int ret;
307
308 DPRINTF("kvm_init_vcpu\n");
309
310 ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu));
311 if (ret < 0) {
312 DPRINTF("kvm_create_vcpu failed\n");
313 goto err;
314 }
315
316 cpu->kvm_fd = ret;
317 cpu->kvm_state = s;
318 cpu->kvm_vcpu_dirty = true;
319
320 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
321 if (mmap_size < 0) {
322 ret = mmap_size;
323 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
324 goto err;
325 }
326
327 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
328 cpu->kvm_fd, 0);
329 if (cpu->kvm_run == MAP_FAILED) {
330 ret = -errno;
331 DPRINTF("mmap'ing vcpu state failed\n");
332 goto err;
333 }
334
335 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
336 s->coalesced_mmio_ring =
337 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
338 }
339
340 ret = kvm_arch_init_vcpu(cpu);
341 err:
342 return ret;
343 }
344
345 /*
346 * dirty pages logging control
347 */
348
349 static int kvm_mem_flags(MemoryRegion *mr)
350 {
351 bool readonly = mr->readonly || memory_region_is_romd(mr);
352 int flags = 0;
353
354 if (memory_region_get_dirty_log_mask(mr) != 0) {
355 flags |= KVM_MEM_LOG_DIRTY_PAGES;
356 }
357 if (readonly && kvm_readonly_mem_allowed) {
358 flags |= KVM_MEM_READONLY;
359 }
360 return flags;
361 }
362
363 static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem,
364 MemoryRegion *mr)
365 {
366 int old_flags;
367
368 old_flags = mem->flags;
369 mem->flags = kvm_mem_flags(mr);
370
371 /* If nothing changed effectively, no need to issue ioctl */
372 if (mem->flags == old_flags) {
373 return 0;
374 }
375
376 return kvm_set_user_memory_region(kml, mem);
377 }
378
379 static int kvm_section_update_flags(KVMMemoryListener *kml,
380 MemoryRegionSection *section)
381 {
382 hwaddr phys_addr = section->offset_within_address_space;
383 ram_addr_t size = int128_get64(section->size);
384 KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size);
385
386 if (mem == NULL) {
387 return 0;
388 } else {
389 return kvm_slot_update_flags(kml, mem, section->mr);
390 }
391 }
392
393 static void kvm_log_start(MemoryListener *listener,
394 MemoryRegionSection *section,
395 int old, int new)
396 {
397 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
398 int r;
399
400 if (old != 0) {
401 return;
402 }
403
404 r = kvm_section_update_flags(kml, section);
405 if (r < 0) {
406 abort();
407 }
408 }
409
410 static void kvm_log_stop(MemoryListener *listener,
411 MemoryRegionSection *section,
412 int old, int new)
413 {
414 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
415 int r;
416
417 if (new != 0) {
418 return;
419 }
420
421 r = kvm_section_update_flags(kml, section);
422 if (r < 0) {
423 abort();
424 }
425 }
426
427 /* get kvm's dirty pages bitmap and update qemu's */
428 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
429 unsigned long *bitmap)
430 {
431 ram_addr_t start = section->offset_within_region +
432 memory_region_get_ram_addr(section->mr);
433 ram_addr_t pages = int128_get64(section->size) / getpagesize();
434
435 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
436 return 0;
437 }
438
439 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
440
441 /**
442 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
443 * This function updates qemu's dirty bitmap using
444 * memory_region_set_dirty(). This means all bits are set
445 * to dirty.
446 *
447 * @start_add: start of logged region.
448 * @end_addr: end of logged region.
449 */
450 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
451 MemoryRegionSection *section)
452 {
453 KVMState *s = kvm_state;
454 unsigned long size, allocated_size = 0;
455 struct kvm_dirty_log d = {};
456 KVMSlot *mem;
457 int ret = 0;
458 hwaddr start_addr = section->offset_within_address_space;
459 hwaddr end_addr = start_addr + int128_get64(section->size);
460
461 d.dirty_bitmap = NULL;
462 while (start_addr < end_addr) {
463 mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr);
464 if (mem == NULL) {
465 break;
466 }
467
468 /* XXX bad kernel interface alert
469 * For dirty bitmap, kernel allocates array of size aligned to
470 * bits-per-long. But for case when the kernel is 64bits and
471 * the userspace is 32bits, userspace can't align to the same
472 * bits-per-long, since sizeof(long) is different between kernel
473 * and user space. This way, userspace will provide buffer which
474 * may be 4 bytes less than the kernel will use, resulting in
475 * userspace memory corruption (which is not detectable by valgrind
476 * too, in most cases).
477 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
478 * a hope that sizeof(long) won't become >8 any time soon.
479 */
480 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
481 /*HOST_LONG_BITS*/ 64) / 8;
482 if (!d.dirty_bitmap) {
483 d.dirty_bitmap = g_malloc(size);
484 } else if (size > allocated_size) {
485 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
486 }
487 allocated_size = size;
488 memset(d.dirty_bitmap, 0, allocated_size);
489
490 d.slot = mem->slot | (kml->as_id << 16);
491 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
492 DPRINTF("ioctl failed %d\n", errno);
493 ret = -1;
494 break;
495 }
496
497 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
498 start_addr = mem->start_addr + mem->memory_size;
499 }
500 g_free(d.dirty_bitmap);
501
502 return ret;
503 }
504
505 static void kvm_coalesce_mmio_region(MemoryListener *listener,
506 MemoryRegionSection *secion,
507 hwaddr start, hwaddr size)
508 {
509 KVMState *s = kvm_state;
510
511 if (s->coalesced_mmio) {
512 struct kvm_coalesced_mmio_zone zone;
513
514 zone.addr = start;
515 zone.size = size;
516 zone.pad = 0;
517
518 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
519 }
520 }
521
522 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
523 MemoryRegionSection *secion,
524 hwaddr start, hwaddr size)
525 {
526 KVMState *s = kvm_state;
527
528 if (s->coalesced_mmio) {
529 struct kvm_coalesced_mmio_zone zone;
530
531 zone.addr = start;
532 zone.size = size;
533 zone.pad = 0;
534
535 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
536 }
537 }
538
539 int kvm_check_extension(KVMState *s, unsigned int extension)
540 {
541 int ret;
542
543 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
544 if (ret < 0) {
545 ret = 0;
546 }
547
548 return ret;
549 }
550
551 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
552 {
553 int ret;
554
555 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
556 if (ret < 0) {
557 /* VM wide version not implemented, use global one instead */
558 ret = kvm_check_extension(s, extension);
559 }
560
561 return ret;
562 }
563
564 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
565 {
566 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
567 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
568 * endianness, but the memory core hands them in target endianness.
569 * For example, PPC is always treated as big-endian even if running
570 * on KVM and on PPC64LE. Correct here.
571 */
572 switch (size) {
573 case 2:
574 val = bswap16(val);
575 break;
576 case 4:
577 val = bswap32(val);
578 break;
579 }
580 #endif
581 return val;
582 }
583
584 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
585 bool assign, uint32_t size, bool datamatch)
586 {
587 int ret;
588 struct kvm_ioeventfd iofd = {
589 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
590 .addr = addr,
591 .len = size,
592 .flags = 0,
593 .fd = fd,
594 };
595
596 if (!kvm_enabled()) {
597 return -ENOSYS;
598 }
599
600 if (datamatch) {
601 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
602 }
603 if (!assign) {
604 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
605 }
606
607 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
608
609 if (ret < 0) {
610 return -errno;
611 }
612
613 return 0;
614 }
615
616 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
617 bool assign, uint32_t size, bool datamatch)
618 {
619 struct kvm_ioeventfd kick = {
620 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
621 .addr = addr,
622 .flags = KVM_IOEVENTFD_FLAG_PIO,
623 .len = size,
624 .fd = fd,
625 };
626 int r;
627 if (!kvm_enabled()) {
628 return -ENOSYS;
629 }
630 if (datamatch) {
631 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
632 }
633 if (!assign) {
634 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
635 }
636 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
637 if (r < 0) {
638 return r;
639 }
640 return 0;
641 }
642
643
644 static int kvm_check_many_ioeventfds(void)
645 {
646 /* Userspace can use ioeventfd for io notification. This requires a host
647 * that supports eventfd(2) and an I/O thread; since eventfd does not
648 * support SIGIO it cannot interrupt the vcpu.
649 *
650 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
651 * can avoid creating too many ioeventfds.
652 */
653 #if defined(CONFIG_EVENTFD)
654 int ioeventfds[7];
655 int i, ret = 0;
656 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
657 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
658 if (ioeventfds[i] < 0) {
659 break;
660 }
661 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
662 if (ret < 0) {
663 close(ioeventfds[i]);
664 break;
665 }
666 }
667
668 /* Decide whether many devices are supported or not */
669 ret = i == ARRAY_SIZE(ioeventfds);
670
671 while (i-- > 0) {
672 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
673 close(ioeventfds[i]);
674 }
675 return ret;
676 #else
677 return 0;
678 #endif
679 }
680
681 static const KVMCapabilityInfo *
682 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
683 {
684 while (list->name) {
685 if (!kvm_check_extension(s, list->value)) {
686 return list;
687 }
688 list++;
689 }
690 return NULL;
691 }
692
693 static void kvm_set_phys_mem(KVMMemoryListener *kml,
694 MemoryRegionSection *section, bool add)
695 {
696 KVMState *s = kvm_state;
697 KVMSlot *mem, old;
698 int err;
699 MemoryRegion *mr = section->mr;
700 bool writeable = !mr->readonly && !mr->rom_device;
701 hwaddr start_addr = section->offset_within_address_space;
702 ram_addr_t size = int128_get64(section->size);
703 void *ram = NULL;
704 unsigned delta;
705
706 /* kvm works in page size chunks, but the function may be called
707 with sub-page size and unaligned start address. Pad the start
708 address to next and truncate size to previous page boundary. */
709 delta = qemu_real_host_page_size - (start_addr & ~qemu_real_host_page_mask);
710 delta &= ~qemu_real_host_page_mask;
711 if (delta > size) {
712 return;
713 }
714 start_addr += delta;
715 size -= delta;
716 size &= qemu_real_host_page_mask;
717 if (!size || (start_addr & ~qemu_real_host_page_mask)) {
718 return;
719 }
720
721 if (!memory_region_is_ram(mr)) {
722 if (writeable || !kvm_readonly_mem_allowed) {
723 return;
724 } else if (!mr->romd_mode) {
725 /* If the memory device is not in romd_mode, then we actually want
726 * to remove the kvm memory slot so all accesses will trap. */
727 add = false;
728 }
729 }
730
731 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
732
733 while (1) {
734 mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size);
735 if (!mem) {
736 break;
737 }
738
739 if (add && start_addr >= mem->start_addr &&
740 (start_addr + size <= mem->start_addr + mem->memory_size) &&
741 (ram - start_addr == mem->ram - mem->start_addr)) {
742 /* The new slot fits into the existing one and comes with
743 * identical parameters - update flags and done. */
744 kvm_slot_update_flags(kml, mem, mr);
745 return;
746 }
747
748 old = *mem;
749
750 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
751 kvm_physical_sync_dirty_bitmap(kml, section);
752 }
753
754 /* unregister the overlapping slot */
755 mem->memory_size = 0;
756 err = kvm_set_user_memory_region(kml, mem);
757 if (err) {
758 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
759 __func__, strerror(-err));
760 abort();
761 }
762
763 /* Workaround for older KVM versions: we can't join slots, even not by
764 * unregistering the previous ones and then registering the larger
765 * slot. We have to maintain the existing fragmentation. Sigh.
766 *
767 * This workaround assumes that the new slot starts at the same
768 * address as the first existing one. If not or if some overlapping
769 * slot comes around later, we will fail (not seen in practice so far)
770 * - and actually require a recent KVM version. */
771 if (s->broken_set_mem_region &&
772 old.start_addr == start_addr && old.memory_size < size && add) {
773 mem = kvm_alloc_slot(kml);
774 mem->memory_size = old.memory_size;
775 mem->start_addr = old.start_addr;
776 mem->ram = old.ram;
777 mem->flags = kvm_mem_flags(mr);
778
779 err = kvm_set_user_memory_region(kml, mem);
780 if (err) {
781 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
782 strerror(-err));
783 abort();
784 }
785
786 start_addr += old.memory_size;
787 ram += old.memory_size;
788 size -= old.memory_size;
789 continue;
790 }
791
792 /* register prefix slot */
793 if (old.start_addr < start_addr) {
794 mem = kvm_alloc_slot(kml);
795 mem->memory_size = start_addr - old.start_addr;
796 mem->start_addr = old.start_addr;
797 mem->ram = old.ram;
798 mem->flags = kvm_mem_flags(mr);
799
800 err = kvm_set_user_memory_region(kml, mem);
801 if (err) {
802 fprintf(stderr, "%s: error registering prefix slot: %s\n",
803 __func__, strerror(-err));
804 #ifdef TARGET_PPC
805 fprintf(stderr, "%s: This is probably because your kernel's " \
806 "PAGE_SIZE is too big. Please try to use 4k " \
807 "PAGE_SIZE!\n", __func__);
808 #endif
809 abort();
810 }
811 }
812
813 /* register suffix slot */
814 if (old.start_addr + old.memory_size > start_addr + size) {
815 ram_addr_t size_delta;
816
817 mem = kvm_alloc_slot(kml);
818 mem->start_addr = start_addr + size;
819 size_delta = mem->start_addr - old.start_addr;
820 mem->memory_size = old.memory_size - size_delta;
821 mem->ram = old.ram + size_delta;
822 mem->flags = kvm_mem_flags(mr);
823
824 err = kvm_set_user_memory_region(kml, mem);
825 if (err) {
826 fprintf(stderr, "%s: error registering suffix slot: %s\n",
827 __func__, strerror(-err));
828 abort();
829 }
830 }
831 }
832
833 /* in case the KVM bug workaround already "consumed" the new slot */
834 if (!size) {
835 return;
836 }
837 if (!add) {
838 return;
839 }
840 mem = kvm_alloc_slot(kml);
841 mem->memory_size = size;
842 mem->start_addr = start_addr;
843 mem->ram = ram;
844 mem->flags = kvm_mem_flags(mr);
845
846 err = kvm_set_user_memory_region(kml, mem);
847 if (err) {
848 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
849 strerror(-err));
850 abort();
851 }
852 }
853
854 static void kvm_region_add(MemoryListener *listener,
855 MemoryRegionSection *section)
856 {
857 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
858
859 memory_region_ref(section->mr);
860 kvm_set_phys_mem(kml, section, true);
861 }
862
863 static void kvm_region_del(MemoryListener *listener,
864 MemoryRegionSection *section)
865 {
866 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
867
868 kvm_set_phys_mem(kml, section, false);
869 memory_region_unref(section->mr);
870 }
871
872 static void kvm_log_sync(MemoryListener *listener,
873 MemoryRegionSection *section)
874 {
875 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
876 int r;
877
878 r = kvm_physical_sync_dirty_bitmap(kml, section);
879 if (r < 0) {
880 abort();
881 }
882 }
883
884 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
885 MemoryRegionSection *section,
886 bool match_data, uint64_t data,
887 EventNotifier *e)
888 {
889 int fd = event_notifier_get_fd(e);
890 int r;
891
892 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
893 data, true, int128_get64(section->size),
894 match_data);
895 if (r < 0) {
896 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
897 __func__, strerror(-r));
898 abort();
899 }
900 }
901
902 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
903 MemoryRegionSection *section,
904 bool match_data, uint64_t data,
905 EventNotifier *e)
906 {
907 int fd = event_notifier_get_fd(e);
908 int r;
909
910 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
911 data, false, int128_get64(section->size),
912 match_data);
913 if (r < 0) {
914 abort();
915 }
916 }
917
918 static void kvm_io_ioeventfd_add(MemoryListener *listener,
919 MemoryRegionSection *section,
920 bool match_data, uint64_t data,
921 EventNotifier *e)
922 {
923 int fd = event_notifier_get_fd(e);
924 int r;
925
926 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
927 data, true, int128_get64(section->size),
928 match_data);
929 if (r < 0) {
930 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
931 __func__, strerror(-r));
932 abort();
933 }
934 }
935
936 static void kvm_io_ioeventfd_del(MemoryListener *listener,
937 MemoryRegionSection *section,
938 bool match_data, uint64_t data,
939 EventNotifier *e)
940
941 {
942 int fd = event_notifier_get_fd(e);
943 int r;
944
945 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
946 data, false, int128_get64(section->size),
947 match_data);
948 if (r < 0) {
949 abort();
950 }
951 }
952
953 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
954 AddressSpace *as, int as_id)
955 {
956 int i;
957
958 kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
959 kml->as_id = as_id;
960
961 for (i = 0; i < s->nr_slots; i++) {
962 kml->slots[i].slot = i;
963 }
964
965 kml->listener.region_add = kvm_region_add;
966 kml->listener.region_del = kvm_region_del;
967 kml->listener.log_start = kvm_log_start;
968 kml->listener.log_stop = kvm_log_stop;
969 kml->listener.log_sync = kvm_log_sync;
970 kml->listener.priority = 10;
971
972 memory_listener_register(&kml->listener, as);
973 }
974
975 static MemoryListener kvm_io_listener = {
976 .eventfd_add = kvm_io_ioeventfd_add,
977 .eventfd_del = kvm_io_ioeventfd_del,
978 .priority = 10,
979 };
980
981 static void kvm_handle_interrupt(CPUState *cpu, int mask)
982 {
983 cpu->interrupt_request |= mask;
984
985 if (!qemu_cpu_is_self(cpu)) {
986 qemu_cpu_kick(cpu);
987 }
988 }
989
990 int kvm_set_irq(KVMState *s, int irq, int level)
991 {
992 struct kvm_irq_level event;
993 int ret;
994
995 assert(kvm_async_interrupts_enabled());
996
997 event.level = level;
998 event.irq = irq;
999 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
1000 if (ret < 0) {
1001 perror("kvm_set_irq");
1002 abort();
1003 }
1004
1005 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
1006 }
1007
1008 #ifdef KVM_CAP_IRQ_ROUTING
1009 typedef struct KVMMSIRoute {
1010 struct kvm_irq_routing_entry kroute;
1011 QTAILQ_ENTRY(KVMMSIRoute) entry;
1012 } KVMMSIRoute;
1013
1014 static void set_gsi(KVMState *s, unsigned int gsi)
1015 {
1016 set_bit(gsi, s->used_gsi_bitmap);
1017 }
1018
1019 static void clear_gsi(KVMState *s, unsigned int gsi)
1020 {
1021 clear_bit(gsi, s->used_gsi_bitmap);
1022 }
1023
1024 void kvm_init_irq_routing(KVMState *s)
1025 {
1026 int gsi_count, i;
1027
1028 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
1029 if (gsi_count > 0) {
1030 /* Round up so we can search ints using ffs */
1031 s->used_gsi_bitmap = bitmap_new(gsi_count);
1032 s->gsi_count = gsi_count;
1033 }
1034
1035 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
1036 s->nr_allocated_irq_routes = 0;
1037
1038 if (!kvm_direct_msi_allowed) {
1039 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
1040 QTAILQ_INIT(&s->msi_hashtab[i]);
1041 }
1042 }
1043
1044 kvm_arch_init_irq_routing(s);
1045 }
1046
1047 void kvm_irqchip_commit_routes(KVMState *s)
1048 {
1049 int ret;
1050
1051 if (kvm_gsi_direct_mapping()) {
1052 return;
1053 }
1054
1055 if (!kvm_gsi_routing_enabled()) {
1056 return;
1057 }
1058
1059 s->irq_routes->flags = 0;
1060 trace_kvm_irqchip_commit_routes();
1061 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
1062 assert(ret == 0);
1063 }
1064
1065 static void kvm_add_routing_entry(KVMState *s,
1066 struct kvm_irq_routing_entry *entry)
1067 {
1068 struct kvm_irq_routing_entry *new;
1069 int n, size;
1070
1071 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1072 n = s->nr_allocated_irq_routes * 2;
1073 if (n < 64) {
1074 n = 64;
1075 }
1076 size = sizeof(struct kvm_irq_routing);
1077 size += n * sizeof(*new);
1078 s->irq_routes = g_realloc(s->irq_routes, size);
1079 s->nr_allocated_irq_routes = n;
1080 }
1081 n = s->irq_routes->nr++;
1082 new = &s->irq_routes->entries[n];
1083
1084 *new = *entry;
1085
1086 set_gsi(s, entry->gsi);
1087 }
1088
1089 static int kvm_update_routing_entry(KVMState *s,
1090 struct kvm_irq_routing_entry *new_entry)
1091 {
1092 struct kvm_irq_routing_entry *entry;
1093 int n;
1094
1095 for (n = 0; n < s->irq_routes->nr; n++) {
1096 entry = &s->irq_routes->entries[n];
1097 if (entry->gsi != new_entry->gsi) {
1098 continue;
1099 }
1100
1101 if(!memcmp(entry, new_entry, sizeof *entry)) {
1102 return 0;
1103 }
1104
1105 *entry = *new_entry;
1106
1107 return 0;
1108 }
1109
1110 return -ESRCH;
1111 }
1112
1113 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1114 {
1115 struct kvm_irq_routing_entry e = {};
1116
1117 assert(pin < s->gsi_count);
1118
1119 e.gsi = irq;
1120 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1121 e.flags = 0;
1122 e.u.irqchip.irqchip = irqchip;
1123 e.u.irqchip.pin = pin;
1124 kvm_add_routing_entry(s, &e);
1125 }
1126
1127 void kvm_irqchip_release_virq(KVMState *s, int virq)
1128 {
1129 struct kvm_irq_routing_entry *e;
1130 int i;
1131
1132 if (kvm_gsi_direct_mapping()) {
1133 return;
1134 }
1135
1136 for (i = 0; i < s->irq_routes->nr; i++) {
1137 e = &s->irq_routes->entries[i];
1138 if (e->gsi == virq) {
1139 s->irq_routes->nr--;
1140 *e = s->irq_routes->entries[s->irq_routes->nr];
1141 }
1142 }
1143 clear_gsi(s, virq);
1144 kvm_arch_release_virq_post(virq);
1145 }
1146
1147 static unsigned int kvm_hash_msi(uint32_t data)
1148 {
1149 /* This is optimized for IA32 MSI layout. However, no other arch shall
1150 * repeat the mistake of not providing a direct MSI injection API. */
1151 return data & 0xff;
1152 }
1153
1154 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1155 {
1156 KVMMSIRoute *route, *next;
1157 unsigned int hash;
1158
1159 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1160 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1161 kvm_irqchip_release_virq(s, route->kroute.gsi);
1162 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1163 g_free(route);
1164 }
1165 }
1166 }
1167
1168 static int kvm_irqchip_get_virq(KVMState *s)
1169 {
1170 int next_virq;
1171
1172 /*
1173 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1174 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1175 * number can succeed even though a new route entry cannot be added.
1176 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1177 */
1178 if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) {
1179 kvm_flush_dynamic_msi_routes(s);
1180 }
1181
1182 /* Return the lowest unused GSI in the bitmap */
1183 next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count);
1184 if (next_virq >= s->gsi_count) {
1185 return -ENOSPC;
1186 } else {
1187 return next_virq;
1188 }
1189 }
1190
1191 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1192 {
1193 unsigned int hash = kvm_hash_msi(msg.data);
1194 KVMMSIRoute *route;
1195
1196 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1197 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1198 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1199 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1200 return route;
1201 }
1202 }
1203 return NULL;
1204 }
1205
1206 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1207 {
1208 struct kvm_msi msi;
1209 KVMMSIRoute *route;
1210
1211 if (kvm_direct_msi_allowed) {
1212 msi.address_lo = (uint32_t)msg.address;
1213 msi.address_hi = msg.address >> 32;
1214 msi.data = le32_to_cpu(msg.data);
1215 msi.flags = 0;
1216 memset(msi.pad, 0, sizeof(msi.pad));
1217
1218 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1219 }
1220
1221 route = kvm_lookup_msi_route(s, msg);
1222 if (!route) {
1223 int virq;
1224
1225 virq = kvm_irqchip_get_virq(s);
1226 if (virq < 0) {
1227 return virq;
1228 }
1229
1230 route = g_malloc0(sizeof(KVMMSIRoute));
1231 route->kroute.gsi = virq;
1232 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1233 route->kroute.flags = 0;
1234 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1235 route->kroute.u.msi.address_hi = msg.address >> 32;
1236 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1237
1238 kvm_add_routing_entry(s, &route->kroute);
1239 kvm_irqchip_commit_routes(s);
1240
1241 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1242 entry);
1243 }
1244
1245 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1246
1247 return kvm_set_irq(s, route->kroute.gsi, 1);
1248 }
1249
1250 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev)
1251 {
1252 struct kvm_irq_routing_entry kroute = {};
1253 int virq;
1254 MSIMessage msg = {0, 0};
1255
1256 if (dev) {
1257 msg = pci_get_msi_message(dev, vector);
1258 }
1259
1260 if (kvm_gsi_direct_mapping()) {
1261 return kvm_arch_msi_data_to_gsi(msg.data);
1262 }
1263
1264 if (!kvm_gsi_routing_enabled()) {
1265 return -ENOSYS;
1266 }
1267
1268 virq = kvm_irqchip_get_virq(s);
1269 if (virq < 0) {
1270 return virq;
1271 }
1272
1273 kroute.gsi = virq;
1274 kroute.type = KVM_IRQ_ROUTING_MSI;
1275 kroute.flags = 0;
1276 kroute.u.msi.address_lo = (uint32_t)msg.address;
1277 kroute.u.msi.address_hi = msg.address >> 32;
1278 kroute.u.msi.data = le32_to_cpu(msg.data);
1279 if (kvm_msi_devid_required()) {
1280 kroute.flags = KVM_MSI_VALID_DEVID;
1281 kroute.u.msi.devid = pci_requester_id(dev);
1282 }
1283 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1284 kvm_irqchip_release_virq(s, virq);
1285 return -EINVAL;
1286 }
1287
1288 trace_kvm_irqchip_add_msi_route(virq);
1289
1290 kvm_add_routing_entry(s, &kroute);
1291 kvm_arch_add_msi_route_post(&kroute, vector, dev);
1292 kvm_irqchip_commit_routes(s);
1293
1294 return virq;
1295 }
1296
1297 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
1298 PCIDevice *dev)
1299 {
1300 struct kvm_irq_routing_entry kroute = {};
1301
1302 if (kvm_gsi_direct_mapping()) {
1303 return 0;
1304 }
1305
1306 if (!kvm_irqchip_in_kernel()) {
1307 return -ENOSYS;
1308 }
1309
1310 kroute.gsi = virq;
1311 kroute.type = KVM_IRQ_ROUTING_MSI;
1312 kroute.flags = 0;
1313 kroute.u.msi.address_lo = (uint32_t)msg.address;
1314 kroute.u.msi.address_hi = msg.address >> 32;
1315 kroute.u.msi.data = le32_to_cpu(msg.data);
1316 if (kvm_msi_devid_required()) {
1317 kroute.flags = KVM_MSI_VALID_DEVID;
1318 kroute.u.msi.devid = pci_requester_id(dev);
1319 }
1320 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1321 return -EINVAL;
1322 }
1323
1324 trace_kvm_irqchip_update_msi_route(virq);
1325
1326 return kvm_update_routing_entry(s, &kroute);
1327 }
1328
1329 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1330 bool assign)
1331 {
1332 struct kvm_irqfd irqfd = {
1333 .fd = fd,
1334 .gsi = virq,
1335 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1336 };
1337
1338 if (rfd != -1) {
1339 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1340 irqfd.resamplefd = rfd;
1341 }
1342
1343 if (!kvm_irqfds_enabled()) {
1344 return -ENOSYS;
1345 }
1346
1347 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1348 }
1349
1350 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1351 {
1352 struct kvm_irq_routing_entry kroute = {};
1353 int virq;
1354
1355 if (!kvm_gsi_routing_enabled()) {
1356 return -ENOSYS;
1357 }
1358
1359 virq = kvm_irqchip_get_virq(s);
1360 if (virq < 0) {
1361 return virq;
1362 }
1363
1364 kroute.gsi = virq;
1365 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1366 kroute.flags = 0;
1367 kroute.u.adapter.summary_addr = adapter->summary_addr;
1368 kroute.u.adapter.ind_addr = adapter->ind_addr;
1369 kroute.u.adapter.summary_offset = adapter->summary_offset;
1370 kroute.u.adapter.ind_offset = adapter->ind_offset;
1371 kroute.u.adapter.adapter_id = adapter->adapter_id;
1372
1373 kvm_add_routing_entry(s, &kroute);
1374
1375 return virq;
1376 }
1377
1378 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1379 {
1380 struct kvm_irq_routing_entry kroute = {};
1381 int virq;
1382
1383 if (!kvm_gsi_routing_enabled()) {
1384 return -ENOSYS;
1385 }
1386 if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) {
1387 return -ENOSYS;
1388 }
1389 virq = kvm_irqchip_get_virq(s);
1390 if (virq < 0) {
1391 return virq;
1392 }
1393
1394 kroute.gsi = virq;
1395 kroute.type = KVM_IRQ_ROUTING_HV_SINT;
1396 kroute.flags = 0;
1397 kroute.u.hv_sint.vcpu = vcpu;
1398 kroute.u.hv_sint.sint = sint;
1399
1400 kvm_add_routing_entry(s, &kroute);
1401 kvm_irqchip_commit_routes(s);
1402
1403 return virq;
1404 }
1405
1406 #else /* !KVM_CAP_IRQ_ROUTING */
1407
1408 void kvm_init_irq_routing(KVMState *s)
1409 {
1410 }
1411
1412 void kvm_irqchip_release_virq(KVMState *s, int virq)
1413 {
1414 }
1415
1416 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1417 {
1418 abort();
1419 }
1420
1421 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev)
1422 {
1423 return -ENOSYS;
1424 }
1425
1426 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1427 {
1428 return -ENOSYS;
1429 }
1430
1431 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1432 {
1433 return -ENOSYS;
1434 }
1435
1436 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1437 {
1438 abort();
1439 }
1440
1441 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1442 {
1443 return -ENOSYS;
1444 }
1445 #endif /* !KVM_CAP_IRQ_ROUTING */
1446
1447 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1448 EventNotifier *rn, int virq)
1449 {
1450 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1451 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1452 }
1453
1454 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1455 int virq)
1456 {
1457 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1458 false);
1459 }
1460
1461 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1462 EventNotifier *rn, qemu_irq irq)
1463 {
1464 gpointer key, gsi;
1465 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1466
1467 if (!found) {
1468 return -ENXIO;
1469 }
1470 return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi));
1471 }
1472
1473 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
1474 qemu_irq irq)
1475 {
1476 gpointer key, gsi;
1477 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1478
1479 if (!found) {
1480 return -ENXIO;
1481 }
1482 return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi));
1483 }
1484
1485 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
1486 {
1487 g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
1488 }
1489
1490 static void kvm_irqchip_create(MachineState *machine, KVMState *s)
1491 {
1492 int ret;
1493
1494 if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1495 ;
1496 } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) {
1497 ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0);
1498 if (ret < 0) {
1499 fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret));
1500 exit(1);
1501 }
1502 } else {
1503 return;
1504 }
1505
1506 /* First probe and see if there's a arch-specific hook to create the
1507 * in-kernel irqchip for us */
1508 ret = kvm_arch_irqchip_create(machine, s);
1509 if (ret == 0) {
1510 if (machine_kernel_irqchip_split(machine)) {
1511 perror("Split IRQ chip mode not supported.");
1512 exit(1);
1513 } else {
1514 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1515 }
1516 }
1517 if (ret < 0) {
1518 fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
1519 exit(1);
1520 }
1521
1522 kvm_kernel_irqchip = true;
1523 /* If we have an in-kernel IRQ chip then we must have asynchronous
1524 * interrupt delivery (though the reverse is not necessarily true)
1525 */
1526 kvm_async_interrupts_allowed = true;
1527 kvm_halt_in_kernel_allowed = true;
1528
1529 kvm_init_irq_routing(s);
1530
1531 s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal);
1532 }
1533
1534 /* Find number of supported CPUs using the recommended
1535 * procedure from the kernel API documentation to cope with
1536 * older kernels that may be missing capabilities.
1537 */
1538 static int kvm_recommended_vcpus(KVMState *s)
1539 {
1540 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1541 return (ret) ? ret : 4;
1542 }
1543
1544 static int kvm_max_vcpus(KVMState *s)
1545 {
1546 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1547 return (ret) ? ret : kvm_recommended_vcpus(s);
1548 }
1549
1550 static int kvm_max_vcpu_id(KVMState *s)
1551 {
1552 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID);
1553 return (ret) ? ret : kvm_max_vcpus(s);
1554 }
1555
1556 bool kvm_vcpu_id_is_valid(int vcpu_id)
1557 {
1558 KVMState *s = KVM_STATE(current_machine->accelerator);
1559 return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s);
1560 }
1561
1562 static int kvm_init(MachineState *ms)
1563 {
1564 MachineClass *mc = MACHINE_GET_CLASS(ms);
1565 static const char upgrade_note[] =
1566 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1567 "(see http://sourceforge.net/projects/kvm).\n";
1568 struct {
1569 const char *name;
1570 int num;
1571 } num_cpus[] = {
1572 { "SMP", smp_cpus },
1573 { "hotpluggable", max_cpus },
1574 { NULL, }
1575 }, *nc = num_cpus;
1576 int soft_vcpus_limit, hard_vcpus_limit;
1577 KVMState *s;
1578 const KVMCapabilityInfo *missing_cap;
1579 int ret;
1580 int type = 0;
1581 const char *kvm_type;
1582
1583 s = KVM_STATE(ms->accelerator);
1584
1585 /*
1586 * On systems where the kernel can support different base page
1587 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1588 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1589 * page size for the system though.
1590 */
1591 assert(TARGET_PAGE_SIZE <= getpagesize());
1592
1593 s->sigmask_len = 8;
1594
1595 #ifdef KVM_CAP_SET_GUEST_DEBUG
1596 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1597 #endif
1598 QLIST_INIT(&s->kvm_parked_vcpus);
1599 s->vmfd = -1;
1600 s->fd = qemu_open("/dev/kvm", O_RDWR);
1601 if (s->fd == -1) {
1602 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1603 ret = -errno;
1604 goto err;
1605 }
1606
1607 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1608 if (ret < KVM_API_VERSION) {
1609 if (ret >= 0) {
1610 ret = -EINVAL;
1611 }
1612 fprintf(stderr, "kvm version too old\n");
1613 goto err;
1614 }
1615
1616 if (ret > KVM_API_VERSION) {
1617 ret = -EINVAL;
1618 fprintf(stderr, "kvm version not supported\n");
1619 goto err;
1620 }
1621
1622 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1623
1624 /* If unspecified, use the default value */
1625 if (!s->nr_slots) {
1626 s->nr_slots = 32;
1627 }
1628
1629 /* check the vcpu limits */
1630 soft_vcpus_limit = kvm_recommended_vcpus(s);
1631 hard_vcpus_limit = kvm_max_vcpus(s);
1632
1633 while (nc->name) {
1634 if (nc->num > soft_vcpus_limit) {
1635 fprintf(stderr,
1636 "Warning: Number of %s cpus requested (%d) exceeds "
1637 "the recommended cpus supported by KVM (%d)\n",
1638 nc->name, nc->num, soft_vcpus_limit);
1639
1640 if (nc->num > hard_vcpus_limit) {
1641 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1642 "the maximum cpus supported by KVM (%d)\n",
1643 nc->name, nc->num, hard_vcpus_limit);
1644 exit(1);
1645 }
1646 }
1647 nc++;
1648 }
1649
1650 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1651 if (mc->kvm_type) {
1652 type = mc->kvm_type(kvm_type);
1653 } else if (kvm_type) {
1654 ret = -EINVAL;
1655 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1656 goto err;
1657 }
1658
1659 do {
1660 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1661 } while (ret == -EINTR);
1662
1663 if (ret < 0) {
1664 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1665 strerror(-ret));
1666
1667 #ifdef TARGET_S390X
1668 if (ret == -EINVAL) {
1669 fprintf(stderr,
1670 "Host kernel setup problem detected. Please verify:\n");
1671 fprintf(stderr, "- for kernels supporting the switch_amode or"
1672 " user_mode parameters, whether\n");
1673 fprintf(stderr,
1674 " user space is running in primary address space\n");
1675 fprintf(stderr,
1676 "- for kernels supporting the vm.allocate_pgste sysctl, "
1677 "whether it is enabled\n");
1678 }
1679 #endif
1680 goto err;
1681 }
1682
1683 s->vmfd = ret;
1684 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1685 if (!missing_cap) {
1686 missing_cap =
1687 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1688 }
1689 if (missing_cap) {
1690 ret = -EINVAL;
1691 fprintf(stderr, "kvm does not support %s\n%s",
1692 missing_cap->name, upgrade_note);
1693 goto err;
1694 }
1695
1696 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1697
1698 s->broken_set_mem_region = 1;
1699 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1700 if (ret > 0) {
1701 s->broken_set_mem_region = 0;
1702 }
1703
1704 #ifdef KVM_CAP_VCPU_EVENTS
1705 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1706 #endif
1707
1708 s->robust_singlestep =
1709 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1710
1711 #ifdef KVM_CAP_DEBUGREGS
1712 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1713 #endif
1714
1715 #ifdef KVM_CAP_IRQ_ROUTING
1716 kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1717 #endif
1718
1719 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1720
1721 s->irq_set_ioctl = KVM_IRQ_LINE;
1722 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1723 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1724 }
1725
1726 #ifdef KVM_CAP_READONLY_MEM
1727 kvm_readonly_mem_allowed =
1728 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1729 #endif
1730
1731 kvm_eventfds_allowed =
1732 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1733
1734 kvm_irqfds_allowed =
1735 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1736
1737 kvm_resamplefds_allowed =
1738 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1739
1740 kvm_vm_attributes_allowed =
1741 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1742
1743 kvm_ioeventfd_any_length_allowed =
1744 (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0);
1745
1746 ret = kvm_arch_init(ms, s);
1747 if (ret < 0) {
1748 goto err;
1749 }
1750
1751 if (machine_kernel_irqchip_allowed(ms)) {
1752 kvm_irqchip_create(ms, s);
1753 }
1754
1755 kvm_state = s;
1756
1757 if (kvm_eventfds_allowed) {
1758 s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
1759 s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
1760 }
1761 s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
1762 s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
1763
1764 kvm_memory_listener_register(s, &s->memory_listener,
1765 &address_space_memory, 0);
1766 memory_listener_register(&kvm_io_listener,
1767 &address_space_io);
1768
1769 s->many_ioeventfds = kvm_check_many_ioeventfds();
1770
1771 cpu_interrupt_handler = kvm_handle_interrupt;
1772
1773 return 0;
1774
1775 err:
1776 assert(ret < 0);
1777 if (s->vmfd >= 0) {
1778 close(s->vmfd);
1779 }
1780 if (s->fd != -1) {
1781 close(s->fd);
1782 }
1783 g_free(s->memory_listener.slots);
1784
1785 return ret;
1786 }
1787
1788 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1789 {
1790 s->sigmask_len = sigmask_len;
1791 }
1792
1793 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1794 int size, uint32_t count)
1795 {
1796 int i;
1797 uint8_t *ptr = data;
1798
1799 for (i = 0; i < count; i++) {
1800 address_space_rw(&address_space_io, port, attrs,
1801 ptr, size,
1802 direction == KVM_EXIT_IO_OUT);
1803 ptr += size;
1804 }
1805 }
1806
1807 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1808 {
1809 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1810 run->internal.suberror);
1811
1812 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1813 int i;
1814
1815 for (i = 0; i < run->internal.ndata; ++i) {
1816 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1817 i, (uint64_t)run->internal.data[i]);
1818 }
1819 }
1820 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1821 fprintf(stderr, "emulation failure\n");
1822 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1823 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1824 return EXCP_INTERRUPT;
1825 }
1826 }
1827 /* FIXME: Should trigger a qmp message to let management know
1828 * something went wrong.
1829 */
1830 return -1;
1831 }
1832
1833 void kvm_flush_coalesced_mmio_buffer(void)
1834 {
1835 KVMState *s = kvm_state;
1836
1837 if (s->coalesced_flush_in_progress) {
1838 return;
1839 }
1840
1841 s->coalesced_flush_in_progress = true;
1842
1843 if (s->coalesced_mmio_ring) {
1844 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1845 while (ring->first != ring->last) {
1846 struct kvm_coalesced_mmio *ent;
1847
1848 ent = &ring->coalesced_mmio[ring->first];
1849
1850 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1851 smp_wmb();
1852 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1853 }
1854 }
1855
1856 s->coalesced_flush_in_progress = false;
1857 }
1858
1859 static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
1860 {
1861 if (!cpu->kvm_vcpu_dirty) {
1862 kvm_arch_get_registers(cpu);
1863 cpu->kvm_vcpu_dirty = true;
1864 }
1865 }
1866
1867 void kvm_cpu_synchronize_state(CPUState *cpu)
1868 {
1869 if (!cpu->kvm_vcpu_dirty) {
1870 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL);
1871 }
1872 }
1873
1874 static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
1875 {
1876 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1877 cpu->kvm_vcpu_dirty = false;
1878 }
1879
1880 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1881 {
1882 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
1883 }
1884
1885 static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
1886 {
1887 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1888 cpu->kvm_vcpu_dirty = false;
1889 }
1890
1891 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1892 {
1893 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
1894 }
1895
1896 int kvm_cpu_exec(CPUState *cpu)
1897 {
1898 struct kvm_run *run = cpu->kvm_run;
1899 int ret, run_ret;
1900
1901 DPRINTF("kvm_cpu_exec()\n");
1902
1903 if (kvm_arch_process_async_events(cpu)) {
1904 cpu->exit_request = 0;
1905 return EXCP_HLT;
1906 }
1907
1908 qemu_mutex_unlock_iothread();
1909
1910 do {
1911 MemTxAttrs attrs;
1912
1913 if (cpu->kvm_vcpu_dirty) {
1914 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1915 cpu->kvm_vcpu_dirty = false;
1916 }
1917
1918 kvm_arch_pre_run(cpu, run);
1919 if (cpu->exit_request) {
1920 DPRINTF("interrupt exit requested\n");
1921 /*
1922 * KVM requires us to reenter the kernel after IO exits to complete
1923 * instruction emulation. This self-signal will ensure that we
1924 * leave ASAP again.
1925 */
1926 qemu_cpu_kick_self();
1927 }
1928
1929 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1930
1931 attrs = kvm_arch_post_run(cpu, run);
1932
1933 if (run_ret < 0) {
1934 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1935 DPRINTF("io window exit\n");
1936 ret = EXCP_INTERRUPT;
1937 break;
1938 }
1939 fprintf(stderr, "error: kvm run failed %s\n",
1940 strerror(-run_ret));
1941 #ifdef TARGET_PPC
1942 if (run_ret == -EBUSY) {
1943 fprintf(stderr,
1944 "This is probably because your SMT is enabled.\n"
1945 "VCPU can only run on primary threads with all "
1946 "secondary threads offline.\n");
1947 }
1948 #endif
1949 ret = -1;
1950 break;
1951 }
1952
1953 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1954 switch (run->exit_reason) {
1955 case KVM_EXIT_IO:
1956 DPRINTF("handle_io\n");
1957 /* Called outside BQL */
1958 kvm_handle_io(run->io.port, attrs,
1959 (uint8_t *)run + run->io.data_offset,
1960 run->io.direction,
1961 run->io.size,
1962 run->io.count);
1963 ret = 0;
1964 break;
1965 case KVM_EXIT_MMIO:
1966 DPRINTF("handle_mmio\n");
1967 /* Called outside BQL */
1968 address_space_rw(&address_space_memory,
1969 run->mmio.phys_addr, attrs,
1970 run->mmio.data,
1971 run->mmio.len,
1972 run->mmio.is_write);
1973 ret = 0;
1974 break;
1975 case KVM_EXIT_IRQ_WINDOW_OPEN:
1976 DPRINTF("irq_window_open\n");
1977 ret = EXCP_INTERRUPT;
1978 break;
1979 case KVM_EXIT_SHUTDOWN:
1980 DPRINTF("shutdown\n");
1981 qemu_system_reset_request();
1982 ret = EXCP_INTERRUPT;
1983 break;
1984 case KVM_EXIT_UNKNOWN:
1985 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1986 (uint64_t)run->hw.hardware_exit_reason);
1987 ret = -1;
1988 break;
1989 case KVM_EXIT_INTERNAL_ERROR:
1990 ret = kvm_handle_internal_error(cpu, run);
1991 break;
1992 case KVM_EXIT_SYSTEM_EVENT:
1993 switch (run->system_event.type) {
1994 case KVM_SYSTEM_EVENT_SHUTDOWN:
1995 qemu_system_shutdown_request();
1996 ret = EXCP_INTERRUPT;
1997 break;
1998 case KVM_SYSTEM_EVENT_RESET:
1999 qemu_system_reset_request();
2000 ret = EXCP_INTERRUPT;
2001 break;
2002 case KVM_SYSTEM_EVENT_CRASH:
2003 qemu_mutex_lock_iothread();
2004 qemu_system_guest_panicked();
2005 qemu_mutex_unlock_iothread();
2006 ret = 0;
2007 break;
2008 default:
2009 DPRINTF("kvm_arch_handle_exit\n");
2010 ret = kvm_arch_handle_exit(cpu, run);
2011 break;
2012 }
2013 break;
2014 default:
2015 DPRINTF("kvm_arch_handle_exit\n");
2016 ret = kvm_arch_handle_exit(cpu, run);
2017 break;
2018 }
2019 } while (ret == 0);
2020
2021 qemu_mutex_lock_iothread();
2022
2023 if (ret < 0) {
2024 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
2025 vm_stop(RUN_STATE_INTERNAL_ERROR);
2026 }
2027
2028 cpu->exit_request = 0;
2029 return ret;
2030 }
2031
2032 int kvm_ioctl(KVMState *s, int type, ...)
2033 {
2034 int ret;
2035 void *arg;
2036 va_list ap;
2037
2038 va_start(ap, type);
2039 arg = va_arg(ap, void *);
2040 va_end(ap);
2041
2042 trace_kvm_ioctl(type, arg);
2043 ret = ioctl(s->fd, type, arg);
2044 if (ret == -1) {
2045 ret = -errno;
2046 }
2047 return ret;
2048 }
2049
2050 int kvm_vm_ioctl(KVMState *s, int type, ...)
2051 {
2052 int ret;
2053 void *arg;
2054 va_list ap;
2055
2056 va_start(ap, type);
2057 arg = va_arg(ap, void *);
2058 va_end(ap);
2059
2060 trace_kvm_vm_ioctl(type, arg);
2061 ret = ioctl(s->vmfd, type, arg);
2062 if (ret == -1) {
2063 ret = -errno;
2064 }
2065 return ret;
2066 }
2067
2068 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
2069 {
2070 int ret;
2071 void *arg;
2072 va_list ap;
2073
2074 va_start(ap, type);
2075 arg = va_arg(ap, void *);
2076 va_end(ap);
2077
2078 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
2079 ret = ioctl(cpu->kvm_fd, type, arg);
2080 if (ret == -1) {
2081 ret = -errno;
2082 }
2083 return ret;
2084 }
2085
2086 int kvm_device_ioctl(int fd, int type, ...)
2087 {
2088 int ret;
2089 void *arg;
2090 va_list ap;
2091
2092 va_start(ap, type);
2093 arg = va_arg(ap, void *);
2094 va_end(ap);
2095
2096 trace_kvm_device_ioctl(fd, type, arg);
2097 ret = ioctl(fd, type, arg);
2098 if (ret == -1) {
2099 ret = -errno;
2100 }
2101 return ret;
2102 }
2103
2104 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
2105 {
2106 int ret;
2107 struct kvm_device_attr attribute = {
2108 .group = group,
2109 .attr = attr,
2110 };
2111
2112 if (!kvm_vm_attributes_allowed) {
2113 return 0;
2114 }
2115
2116 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
2117 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2118 return ret ? 0 : 1;
2119 }
2120
2121 int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr)
2122 {
2123 struct kvm_device_attr attribute = {
2124 .group = group,
2125 .attr = attr,
2126 .flags = 0,
2127 };
2128
2129 return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1;
2130 }
2131
2132 void kvm_device_access(int fd, int group, uint64_t attr,
2133 void *val, bool write)
2134 {
2135 struct kvm_device_attr kvmattr;
2136 int err;
2137
2138 kvmattr.flags = 0;
2139 kvmattr.group = group;
2140 kvmattr.attr = attr;
2141 kvmattr.addr = (uintptr_t)val;
2142
2143 err = kvm_device_ioctl(fd,
2144 write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR,
2145 &kvmattr);
2146 if (err < 0) {
2147 error_report("KVM_%s_DEVICE_ATTR failed: %s",
2148 write ? "SET" : "GET", strerror(-err));
2149 error_printf("Group %d attr 0x%016" PRIx64 "\n", group, attr);
2150 abort();
2151 }
2152 }
2153
2154 /* Return 1 on success, 0 on failure */
2155 int kvm_has_sync_mmu(void)
2156 {
2157 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
2158 }
2159
2160 int kvm_has_vcpu_events(void)
2161 {
2162 return kvm_state->vcpu_events;
2163 }
2164
2165 int kvm_has_robust_singlestep(void)
2166 {
2167 return kvm_state->robust_singlestep;
2168 }
2169
2170 int kvm_has_debugregs(void)
2171 {
2172 return kvm_state->debugregs;
2173 }
2174
2175 int kvm_has_many_ioeventfds(void)
2176 {
2177 if (!kvm_enabled()) {
2178 return 0;
2179 }
2180 return kvm_state->many_ioeventfds;
2181 }
2182
2183 int kvm_has_gsi_routing(void)
2184 {
2185 #ifdef KVM_CAP_IRQ_ROUTING
2186 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2187 #else
2188 return false;
2189 #endif
2190 }
2191
2192 int kvm_has_intx_set_mask(void)
2193 {
2194 return kvm_state->intx_set_mask;
2195 }
2196
2197 #ifdef KVM_CAP_SET_GUEST_DEBUG
2198 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2199 target_ulong pc)
2200 {
2201 struct kvm_sw_breakpoint *bp;
2202
2203 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2204 if (bp->pc == pc) {
2205 return bp;
2206 }
2207 }
2208 return NULL;
2209 }
2210
2211 int kvm_sw_breakpoints_active(CPUState *cpu)
2212 {
2213 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2214 }
2215
2216 struct kvm_set_guest_debug_data {
2217 struct kvm_guest_debug dbg;
2218 int err;
2219 };
2220
2221 static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data)
2222 {
2223 struct kvm_set_guest_debug_data *dbg_data =
2224 (struct kvm_set_guest_debug_data *) data.host_ptr;
2225
2226 dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG,
2227 &dbg_data->dbg);
2228 }
2229
2230 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2231 {
2232 struct kvm_set_guest_debug_data data;
2233
2234 data.dbg.control = reinject_trap;
2235
2236 if (cpu->singlestep_enabled) {
2237 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2238 }
2239 kvm_arch_update_guest_debug(cpu, &data.dbg);
2240
2241 run_on_cpu(cpu, kvm_invoke_set_guest_debug,
2242 RUN_ON_CPU_HOST_PTR(&data));
2243 return data.err;
2244 }
2245
2246 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2247 target_ulong len, int type)
2248 {
2249 struct kvm_sw_breakpoint *bp;
2250 int err;
2251
2252 if (type == GDB_BREAKPOINT_SW) {
2253 bp = kvm_find_sw_breakpoint(cpu, addr);
2254 if (bp) {
2255 bp->use_count++;
2256 return 0;
2257 }
2258
2259 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2260 bp->pc = addr;
2261 bp->use_count = 1;
2262 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2263 if (err) {
2264 g_free(bp);
2265 return err;
2266 }
2267
2268 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2269 } else {
2270 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2271 if (err) {
2272 return err;
2273 }
2274 }
2275
2276 CPU_FOREACH(cpu) {
2277 err = kvm_update_guest_debug(cpu, 0);
2278 if (err) {
2279 return err;
2280 }
2281 }
2282 return 0;
2283 }
2284
2285 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2286 target_ulong len, int type)
2287 {
2288 struct kvm_sw_breakpoint *bp;
2289 int err;
2290
2291 if (type == GDB_BREAKPOINT_SW) {
2292 bp = kvm_find_sw_breakpoint(cpu, addr);
2293 if (!bp) {
2294 return -ENOENT;
2295 }
2296
2297 if (bp->use_count > 1) {
2298 bp->use_count--;
2299 return 0;
2300 }
2301
2302 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2303 if (err) {
2304 return err;
2305 }
2306
2307 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2308 g_free(bp);
2309 } else {
2310 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2311 if (err) {
2312 return err;
2313 }
2314 }
2315
2316 CPU_FOREACH(cpu) {
2317 err = kvm_update_guest_debug(cpu, 0);
2318 if (err) {
2319 return err;
2320 }
2321 }
2322 return 0;
2323 }
2324
2325 void kvm_remove_all_breakpoints(CPUState *cpu)
2326 {
2327 struct kvm_sw_breakpoint *bp, *next;
2328 KVMState *s = cpu->kvm_state;
2329 CPUState *tmpcpu;
2330
2331 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2332 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2333 /* Try harder to find a CPU that currently sees the breakpoint. */
2334 CPU_FOREACH(tmpcpu) {
2335 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2336 break;
2337 }
2338 }
2339 }
2340 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2341 g_free(bp);
2342 }
2343 kvm_arch_remove_all_hw_breakpoints();
2344
2345 CPU_FOREACH(cpu) {
2346 kvm_update_guest_debug(cpu, 0);
2347 }
2348 }
2349
2350 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2351
2352 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2353 {
2354 return -EINVAL;
2355 }
2356
2357 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2358 target_ulong len, int type)
2359 {
2360 return -EINVAL;
2361 }
2362
2363 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2364 target_ulong len, int type)
2365 {
2366 return -EINVAL;
2367 }
2368
2369 void kvm_remove_all_breakpoints(CPUState *cpu)
2370 {
2371 }
2372 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2373
2374 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2375 {
2376 KVMState *s = kvm_state;
2377 struct kvm_signal_mask *sigmask;
2378 int r;
2379
2380 if (!sigset) {
2381 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2382 }
2383
2384 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2385
2386 sigmask->len = s->sigmask_len;
2387 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2388 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2389 g_free(sigmask);
2390
2391 return r;
2392 }
2393 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2394 {
2395 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2396 }
2397
2398 int kvm_on_sigbus(int code, void *addr)
2399 {
2400 return kvm_arch_on_sigbus(code, addr);
2401 }
2402
2403 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2404 {
2405 int ret;
2406 struct kvm_create_device create_dev;
2407
2408 create_dev.type = type;
2409 create_dev.fd = -1;
2410 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2411
2412 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2413 return -ENOTSUP;
2414 }
2415
2416 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2417 if (ret) {
2418 return ret;
2419 }
2420
2421 return test ? 0 : create_dev.fd;
2422 }
2423
2424 bool kvm_device_supported(int vmfd, uint64_t type)
2425 {
2426 struct kvm_create_device create_dev = {
2427 .type = type,
2428 .fd = -1,
2429 .flags = KVM_CREATE_DEVICE_TEST,
2430 };
2431
2432 if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) {
2433 return false;
2434 }
2435
2436 return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0);
2437 }
2438
2439 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2440 {
2441 struct kvm_one_reg reg;
2442 int r;
2443
2444 reg.id = id;
2445 reg.addr = (uintptr_t) source;
2446 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
2447 if (r) {
2448 trace_kvm_failed_reg_set(id, strerror(-r));
2449 }
2450 return r;
2451 }
2452
2453 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2454 {
2455 struct kvm_one_reg reg;
2456 int r;
2457
2458 reg.id = id;
2459 reg.addr = (uintptr_t) target;
2460 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
2461 if (r) {
2462 trace_kvm_failed_reg_get(id, strerror(-r));
2463 }
2464 return r;
2465 }
2466
2467 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2468 {
2469 AccelClass *ac = ACCEL_CLASS(oc);
2470 ac->name = "KVM";
2471 ac->init_machine = kvm_init;
2472 ac->allowed = &kvm_allowed;
2473 }
2474
2475 static const TypeInfo kvm_accel_type = {
2476 .name = TYPE_KVM_ACCEL,
2477 .parent = TYPE_ACCEL,
2478 .class_init = kvm_accel_class_init,
2479 .instance_size = sizeof(KVMState),
2480 };
2481
2482 static void kvm_type_init(void)
2483 {
2484 type_register_static(&kvm_accel_type);
2485 }
2486
2487 type_init(kvm_type_init);