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