hw/net: fix mcf_fec driver receiver
[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 <sys/types.h>
17 #include <sys/ioctl.h>
18 #include <sys/mman.h>
19 #include <stdarg.h>
20
21 #include <linux/kvm.h>
22
23 #include "qemu-common.h"
24 #include "qemu/atomic.h"
25 #include "qemu/option.h"
26 #include "qemu/config-file.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 COALESCED_MMIO_MAX */
48 #define PAGE_SIZE TARGET_PAGE_SIZE
49
50 //#define DEBUG_KVM
51
52 #ifdef DEBUG_KVM
53 #define DPRINTF(fmt, ...) \
54 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
55 #else
56 #define DPRINTF(fmt, ...) \
57 do { } while (0)
58 #endif
59
60 #define KVM_MSI_HASHTAB_SIZE 256
61
62 struct KVMState
63 {
64 AccelState parent_obj;
65
66 int nr_slots;
67 int fd;
68 int vmfd;
69 int coalesced_mmio;
70 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
71 bool coalesced_flush_in_progress;
72 int broken_set_mem_region;
73 int vcpu_events;
74 int robust_singlestep;
75 int debugregs;
76 #ifdef KVM_CAP_SET_GUEST_DEBUG
77 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
78 #endif
79 int pit_state2;
80 int xsave, xcrs;
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 uint32_t *used_gsi_bitmap;
93 unsigned int gsi_count;
94 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
95 bool direct_msi;
96 #endif
97 KVMMemoryListener memory_listener;
98 };
99
100 KVMState *kvm_state;
101 bool kvm_kernel_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
114 static const KVMCapabilityInfo kvm_required_capabilites[] = {
115 KVM_CAP_INFO(USER_MEMORY),
116 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
117 KVM_CAP_LAST_INFO
118 };
119
120 static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml)
121 {
122 KVMState *s = kvm_state;
123 int i;
124
125 for (i = 0; i < s->nr_slots; i++) {
126 if (kml->slots[i].memory_size == 0) {
127 return &kml->slots[i];
128 }
129 }
130
131 return NULL;
132 }
133
134 bool kvm_has_free_slot(MachineState *ms)
135 {
136 KVMState *s = KVM_STATE(ms->accelerator);
137
138 return kvm_get_free_slot(&s->memory_listener);
139 }
140
141 static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml)
142 {
143 KVMSlot *slot = kvm_get_free_slot(kml);
144
145 if (slot) {
146 return slot;
147 }
148
149 fprintf(stderr, "%s: no free slot available\n", __func__);
150 abort();
151 }
152
153 static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml,
154 hwaddr start_addr,
155 hwaddr end_addr)
156 {
157 KVMState *s = kvm_state;
158 int i;
159
160 for (i = 0; i < s->nr_slots; i++) {
161 KVMSlot *mem = &kml->slots[i];
162
163 if (start_addr == mem->start_addr &&
164 end_addr == mem->start_addr + mem->memory_size) {
165 return mem;
166 }
167 }
168
169 return NULL;
170 }
171
172 /*
173 * Find overlapping slot with lowest start address
174 */
175 static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml,
176 hwaddr start_addr,
177 hwaddr end_addr)
178 {
179 KVMState *s = kvm_state;
180 KVMSlot *found = NULL;
181 int i;
182
183 for (i = 0; i < s->nr_slots; i++) {
184 KVMSlot *mem = &kml->slots[i];
185
186 if (mem->memory_size == 0 ||
187 (found && found->start_addr < mem->start_addr)) {
188 continue;
189 }
190
191 if (end_addr > mem->start_addr &&
192 start_addr < mem->start_addr + mem->memory_size) {
193 found = mem;
194 }
195 }
196
197 return found;
198 }
199
200 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
201 hwaddr *phys_addr)
202 {
203 KVMMemoryListener *kml = &s->memory_listener;
204 int i;
205
206 for (i = 0; i < s->nr_slots; i++) {
207 KVMSlot *mem = &kml->slots[i];
208
209 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
210 *phys_addr = mem->start_addr + (ram - mem->ram);
211 return 1;
212 }
213 }
214
215 return 0;
216 }
217
218 static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot)
219 {
220 KVMState *s = kvm_state;
221 struct kvm_userspace_memory_region mem;
222
223 mem.slot = slot->slot | (kml->as_id << 16);
224 mem.guest_phys_addr = slot->start_addr;
225 mem.userspace_addr = (unsigned long)slot->ram;
226 mem.flags = slot->flags;
227
228 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
229 /* Set the slot size to 0 before setting the slot to the desired
230 * value. This is needed based on KVM commit 75d61fbc. */
231 mem.memory_size = 0;
232 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
233 }
234 mem.memory_size = slot->memory_size;
235 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
236 }
237
238 int kvm_init_vcpu(CPUState *cpu)
239 {
240 KVMState *s = kvm_state;
241 long mmap_size;
242 int ret;
243
244 DPRINTF("kvm_init_vcpu\n");
245
246 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
247 if (ret < 0) {
248 DPRINTF("kvm_create_vcpu failed\n");
249 goto err;
250 }
251
252 cpu->kvm_fd = ret;
253 cpu->kvm_state = s;
254 cpu->kvm_vcpu_dirty = true;
255
256 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
257 if (mmap_size < 0) {
258 ret = mmap_size;
259 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
260 goto err;
261 }
262
263 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
264 cpu->kvm_fd, 0);
265 if (cpu->kvm_run == MAP_FAILED) {
266 ret = -errno;
267 DPRINTF("mmap'ing vcpu state failed\n");
268 goto err;
269 }
270
271 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
272 s->coalesced_mmio_ring =
273 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
274 }
275
276 ret = kvm_arch_init_vcpu(cpu);
277 err:
278 return ret;
279 }
280
281 /*
282 * dirty pages logging control
283 */
284
285 static int kvm_mem_flags(MemoryRegion *mr)
286 {
287 bool readonly = mr->readonly || memory_region_is_romd(mr);
288 int flags = 0;
289
290 if (memory_region_get_dirty_log_mask(mr) != 0) {
291 flags |= KVM_MEM_LOG_DIRTY_PAGES;
292 }
293 if (readonly && kvm_readonly_mem_allowed) {
294 flags |= KVM_MEM_READONLY;
295 }
296 return flags;
297 }
298
299 static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem,
300 MemoryRegion *mr)
301 {
302 int old_flags;
303
304 old_flags = mem->flags;
305 mem->flags = kvm_mem_flags(mr);
306
307 /* If nothing changed effectively, no need to issue ioctl */
308 if (mem->flags == old_flags) {
309 return 0;
310 }
311
312 return kvm_set_user_memory_region(kml, mem);
313 }
314
315 static int kvm_section_update_flags(KVMMemoryListener *kml,
316 MemoryRegionSection *section)
317 {
318 hwaddr phys_addr = section->offset_within_address_space;
319 ram_addr_t size = int128_get64(section->size);
320 KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size);
321
322 if (mem == NULL) {
323 return 0;
324 } else {
325 return kvm_slot_update_flags(kml, mem, section->mr);
326 }
327 }
328
329 static void kvm_log_start(MemoryListener *listener,
330 MemoryRegionSection *section,
331 int old, int new)
332 {
333 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
334 int r;
335
336 if (old != 0) {
337 return;
338 }
339
340 r = kvm_section_update_flags(kml, section);
341 if (r < 0) {
342 abort();
343 }
344 }
345
346 static void kvm_log_stop(MemoryListener *listener,
347 MemoryRegionSection *section,
348 int old, int new)
349 {
350 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
351 int r;
352
353 if (new != 0) {
354 return;
355 }
356
357 r = kvm_section_update_flags(kml, section);
358 if (r < 0) {
359 abort();
360 }
361 }
362
363 /* get kvm's dirty pages bitmap and update qemu's */
364 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
365 unsigned long *bitmap)
366 {
367 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
368 ram_addr_t pages = int128_get64(section->size) / getpagesize();
369
370 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
371 return 0;
372 }
373
374 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
375
376 /**
377 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
378 * This function updates qemu's dirty bitmap using
379 * memory_region_set_dirty(). This means all bits are set
380 * to dirty.
381 *
382 * @start_add: start of logged region.
383 * @end_addr: end of logged region.
384 */
385 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
386 MemoryRegionSection *section)
387 {
388 KVMState *s = kvm_state;
389 unsigned long size, allocated_size = 0;
390 struct kvm_dirty_log d = {};
391 KVMSlot *mem;
392 int ret = 0;
393 hwaddr start_addr = section->offset_within_address_space;
394 hwaddr end_addr = start_addr + int128_get64(section->size);
395
396 d.dirty_bitmap = NULL;
397 while (start_addr < end_addr) {
398 mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr);
399 if (mem == NULL) {
400 break;
401 }
402
403 /* XXX bad kernel interface alert
404 * For dirty bitmap, kernel allocates array of size aligned to
405 * bits-per-long. But for case when the kernel is 64bits and
406 * the userspace is 32bits, userspace can't align to the same
407 * bits-per-long, since sizeof(long) is different between kernel
408 * and user space. This way, userspace will provide buffer which
409 * may be 4 bytes less than the kernel will use, resulting in
410 * userspace memory corruption (which is not detectable by valgrind
411 * too, in most cases).
412 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
413 * a hope that sizeof(long) wont become >8 any time soon.
414 */
415 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
416 /*HOST_LONG_BITS*/ 64) / 8;
417 if (!d.dirty_bitmap) {
418 d.dirty_bitmap = g_malloc(size);
419 } else if (size > allocated_size) {
420 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
421 }
422 allocated_size = size;
423 memset(d.dirty_bitmap, 0, allocated_size);
424
425 d.slot = mem->slot | (kml->as_id << 16);
426 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
427 DPRINTF("ioctl failed %d\n", errno);
428 ret = -1;
429 break;
430 }
431
432 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
433 start_addr = mem->start_addr + mem->memory_size;
434 }
435 g_free(d.dirty_bitmap);
436
437 return ret;
438 }
439
440 static void kvm_coalesce_mmio_region(MemoryListener *listener,
441 MemoryRegionSection *secion,
442 hwaddr start, hwaddr size)
443 {
444 KVMState *s = kvm_state;
445
446 if (s->coalesced_mmio) {
447 struct kvm_coalesced_mmio_zone zone;
448
449 zone.addr = start;
450 zone.size = size;
451 zone.pad = 0;
452
453 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
454 }
455 }
456
457 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
458 MemoryRegionSection *secion,
459 hwaddr start, hwaddr size)
460 {
461 KVMState *s = kvm_state;
462
463 if (s->coalesced_mmio) {
464 struct kvm_coalesced_mmio_zone zone;
465
466 zone.addr = start;
467 zone.size = size;
468 zone.pad = 0;
469
470 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
471 }
472 }
473
474 int kvm_check_extension(KVMState *s, unsigned int extension)
475 {
476 int ret;
477
478 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
479 if (ret < 0) {
480 ret = 0;
481 }
482
483 return ret;
484 }
485
486 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
487 {
488 int ret;
489
490 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
491 if (ret < 0) {
492 /* VM wide version not implemented, use global one instead */
493 ret = kvm_check_extension(s, extension);
494 }
495
496 return ret;
497 }
498
499 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
500 {
501 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
502 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
503 * endianness, but the memory core hands them in target endianness.
504 * For example, PPC is always treated as big-endian even if running
505 * on KVM and on PPC64LE. Correct here.
506 */
507 switch (size) {
508 case 2:
509 val = bswap16(val);
510 break;
511 case 4:
512 val = bswap32(val);
513 break;
514 }
515 #endif
516 return val;
517 }
518
519 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
520 bool assign, uint32_t size, bool datamatch)
521 {
522 int ret;
523 struct kvm_ioeventfd iofd = {
524 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
525 .addr = addr,
526 .len = size,
527 .flags = 0,
528 .fd = fd,
529 };
530
531 if (!kvm_enabled()) {
532 return -ENOSYS;
533 }
534
535 if (datamatch) {
536 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
537 }
538 if (!assign) {
539 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
540 }
541
542 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
543
544 if (ret < 0) {
545 return -errno;
546 }
547
548 return 0;
549 }
550
551 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
552 bool assign, uint32_t size, bool datamatch)
553 {
554 struct kvm_ioeventfd kick = {
555 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
556 .addr = addr,
557 .flags = KVM_IOEVENTFD_FLAG_PIO,
558 .len = size,
559 .fd = fd,
560 };
561 int r;
562 if (!kvm_enabled()) {
563 return -ENOSYS;
564 }
565 if (datamatch) {
566 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
567 }
568 if (!assign) {
569 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
570 }
571 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
572 if (r < 0) {
573 return r;
574 }
575 return 0;
576 }
577
578
579 static int kvm_check_many_ioeventfds(void)
580 {
581 /* Userspace can use ioeventfd for io notification. This requires a host
582 * that supports eventfd(2) and an I/O thread; since eventfd does not
583 * support SIGIO it cannot interrupt the vcpu.
584 *
585 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
586 * can avoid creating too many ioeventfds.
587 */
588 #if defined(CONFIG_EVENTFD)
589 int ioeventfds[7];
590 int i, ret = 0;
591 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
592 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
593 if (ioeventfds[i] < 0) {
594 break;
595 }
596 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
597 if (ret < 0) {
598 close(ioeventfds[i]);
599 break;
600 }
601 }
602
603 /* Decide whether many devices are supported or not */
604 ret = i == ARRAY_SIZE(ioeventfds);
605
606 while (i-- > 0) {
607 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
608 close(ioeventfds[i]);
609 }
610 return ret;
611 #else
612 return 0;
613 #endif
614 }
615
616 static const KVMCapabilityInfo *
617 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
618 {
619 while (list->name) {
620 if (!kvm_check_extension(s, list->value)) {
621 return list;
622 }
623 list++;
624 }
625 return NULL;
626 }
627
628 static void kvm_set_phys_mem(KVMMemoryListener *kml,
629 MemoryRegionSection *section, bool add)
630 {
631 KVMState *s = kvm_state;
632 KVMSlot *mem, old;
633 int err;
634 MemoryRegion *mr = section->mr;
635 bool writeable = !mr->readonly && !mr->rom_device;
636 hwaddr start_addr = section->offset_within_address_space;
637 ram_addr_t size = int128_get64(section->size);
638 void *ram = NULL;
639 unsigned delta;
640
641 /* kvm works in page size chunks, but the function may be called
642 with sub-page size and unaligned start address. Pad the start
643 address to next and truncate size to previous page boundary. */
644 delta = (TARGET_PAGE_SIZE - (start_addr & ~TARGET_PAGE_MASK));
645 delta &= ~TARGET_PAGE_MASK;
646 if (delta > size) {
647 return;
648 }
649 start_addr += delta;
650 size -= delta;
651 size &= TARGET_PAGE_MASK;
652 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
653 return;
654 }
655
656 if (!memory_region_is_ram(mr)) {
657 if (writeable || !kvm_readonly_mem_allowed) {
658 return;
659 } else if (!mr->romd_mode) {
660 /* If the memory device is not in romd_mode, then we actually want
661 * to remove the kvm memory slot so all accesses will trap. */
662 add = false;
663 }
664 }
665
666 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
667
668 while (1) {
669 mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size);
670 if (!mem) {
671 break;
672 }
673
674 if (add && start_addr >= mem->start_addr &&
675 (start_addr + size <= mem->start_addr + mem->memory_size) &&
676 (ram - start_addr == mem->ram - mem->start_addr)) {
677 /* The new slot fits into the existing one and comes with
678 * identical parameters - update flags and done. */
679 kvm_slot_update_flags(kml, mem, mr);
680 return;
681 }
682
683 old = *mem;
684
685 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
686 kvm_physical_sync_dirty_bitmap(kml, section);
687 }
688
689 /* unregister the overlapping slot */
690 mem->memory_size = 0;
691 err = kvm_set_user_memory_region(kml, mem);
692 if (err) {
693 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
694 __func__, strerror(-err));
695 abort();
696 }
697
698 /* Workaround for older KVM versions: we can't join slots, even not by
699 * unregistering the previous ones and then registering the larger
700 * slot. We have to maintain the existing fragmentation. Sigh.
701 *
702 * This workaround assumes that the new slot starts at the same
703 * address as the first existing one. If not or if some overlapping
704 * slot comes around later, we will fail (not seen in practice so far)
705 * - and actually require a recent KVM version. */
706 if (s->broken_set_mem_region &&
707 old.start_addr == start_addr && old.memory_size < size && add) {
708 mem = kvm_alloc_slot(kml);
709 mem->memory_size = old.memory_size;
710 mem->start_addr = old.start_addr;
711 mem->ram = old.ram;
712 mem->flags = kvm_mem_flags(mr);
713
714 err = kvm_set_user_memory_region(kml, mem);
715 if (err) {
716 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
717 strerror(-err));
718 abort();
719 }
720
721 start_addr += old.memory_size;
722 ram += old.memory_size;
723 size -= old.memory_size;
724 continue;
725 }
726
727 /* register prefix slot */
728 if (old.start_addr < start_addr) {
729 mem = kvm_alloc_slot(kml);
730 mem->memory_size = start_addr - old.start_addr;
731 mem->start_addr = old.start_addr;
732 mem->ram = old.ram;
733 mem->flags = kvm_mem_flags(mr);
734
735 err = kvm_set_user_memory_region(kml, mem);
736 if (err) {
737 fprintf(stderr, "%s: error registering prefix slot: %s\n",
738 __func__, strerror(-err));
739 #ifdef TARGET_PPC
740 fprintf(stderr, "%s: This is probably because your kernel's " \
741 "PAGE_SIZE is too big. Please try to use 4k " \
742 "PAGE_SIZE!\n", __func__);
743 #endif
744 abort();
745 }
746 }
747
748 /* register suffix slot */
749 if (old.start_addr + old.memory_size > start_addr + size) {
750 ram_addr_t size_delta;
751
752 mem = kvm_alloc_slot(kml);
753 mem->start_addr = start_addr + size;
754 size_delta = mem->start_addr - old.start_addr;
755 mem->memory_size = old.memory_size - size_delta;
756 mem->ram = old.ram + size_delta;
757 mem->flags = kvm_mem_flags(mr);
758
759 err = kvm_set_user_memory_region(kml, mem);
760 if (err) {
761 fprintf(stderr, "%s: error registering suffix slot: %s\n",
762 __func__, strerror(-err));
763 abort();
764 }
765 }
766 }
767
768 /* in case the KVM bug workaround already "consumed" the new slot */
769 if (!size) {
770 return;
771 }
772 if (!add) {
773 return;
774 }
775 mem = kvm_alloc_slot(kml);
776 mem->memory_size = size;
777 mem->start_addr = start_addr;
778 mem->ram = ram;
779 mem->flags = kvm_mem_flags(mr);
780
781 err = kvm_set_user_memory_region(kml, mem);
782 if (err) {
783 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
784 strerror(-err));
785 abort();
786 }
787 }
788
789 static void kvm_region_add(MemoryListener *listener,
790 MemoryRegionSection *section)
791 {
792 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
793
794 memory_region_ref(section->mr);
795 kvm_set_phys_mem(kml, section, true);
796 }
797
798 static void kvm_region_del(MemoryListener *listener,
799 MemoryRegionSection *section)
800 {
801 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
802
803 kvm_set_phys_mem(kml, section, false);
804 memory_region_unref(section->mr);
805 }
806
807 static void kvm_log_sync(MemoryListener *listener,
808 MemoryRegionSection *section)
809 {
810 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
811 int r;
812
813 r = kvm_physical_sync_dirty_bitmap(kml, section);
814 if (r < 0) {
815 abort();
816 }
817 }
818
819 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
820 MemoryRegionSection *section,
821 bool match_data, uint64_t data,
822 EventNotifier *e)
823 {
824 int fd = event_notifier_get_fd(e);
825 int r;
826
827 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
828 data, true, int128_get64(section->size),
829 match_data);
830 if (r < 0) {
831 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
832 __func__, strerror(-r));
833 abort();
834 }
835 }
836
837 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
838 MemoryRegionSection *section,
839 bool match_data, uint64_t data,
840 EventNotifier *e)
841 {
842 int fd = event_notifier_get_fd(e);
843 int r;
844
845 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
846 data, false, int128_get64(section->size),
847 match_data);
848 if (r < 0) {
849 abort();
850 }
851 }
852
853 static void kvm_io_ioeventfd_add(MemoryListener *listener,
854 MemoryRegionSection *section,
855 bool match_data, uint64_t data,
856 EventNotifier *e)
857 {
858 int fd = event_notifier_get_fd(e);
859 int r;
860
861 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
862 data, true, int128_get64(section->size),
863 match_data);
864 if (r < 0) {
865 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
866 __func__, strerror(-r));
867 abort();
868 }
869 }
870
871 static void kvm_io_ioeventfd_del(MemoryListener *listener,
872 MemoryRegionSection *section,
873 bool match_data, uint64_t data,
874 EventNotifier *e)
875
876 {
877 int fd = event_notifier_get_fd(e);
878 int r;
879
880 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
881 data, false, int128_get64(section->size),
882 match_data);
883 if (r < 0) {
884 abort();
885 }
886 }
887
888 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
889 AddressSpace *as, int as_id)
890 {
891 int i;
892
893 kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
894 kml->as_id = as_id;
895
896 for (i = 0; i < s->nr_slots; i++) {
897 kml->slots[i].slot = i;
898 }
899
900 kml->listener.region_add = kvm_region_add;
901 kml->listener.region_del = kvm_region_del;
902 kml->listener.log_start = kvm_log_start;
903 kml->listener.log_stop = kvm_log_stop;
904 kml->listener.log_sync = kvm_log_sync;
905 kml->listener.priority = 10;
906
907 memory_listener_register(&kml->listener, as);
908 }
909
910 static MemoryListener kvm_io_listener = {
911 .eventfd_add = kvm_io_ioeventfd_add,
912 .eventfd_del = kvm_io_ioeventfd_del,
913 .priority = 10,
914 };
915
916 static void kvm_handle_interrupt(CPUState *cpu, int mask)
917 {
918 cpu->interrupt_request |= mask;
919
920 if (!qemu_cpu_is_self(cpu)) {
921 qemu_cpu_kick(cpu);
922 }
923 }
924
925 int kvm_set_irq(KVMState *s, int irq, int level)
926 {
927 struct kvm_irq_level event;
928 int ret;
929
930 assert(kvm_async_interrupts_enabled());
931
932 event.level = level;
933 event.irq = irq;
934 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
935 if (ret < 0) {
936 perror("kvm_set_irq");
937 abort();
938 }
939
940 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
941 }
942
943 #ifdef KVM_CAP_IRQ_ROUTING
944 typedef struct KVMMSIRoute {
945 struct kvm_irq_routing_entry kroute;
946 QTAILQ_ENTRY(KVMMSIRoute) entry;
947 } KVMMSIRoute;
948
949 static void set_gsi(KVMState *s, unsigned int gsi)
950 {
951 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
952 }
953
954 static void clear_gsi(KVMState *s, unsigned int gsi)
955 {
956 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
957 }
958
959 void kvm_init_irq_routing(KVMState *s)
960 {
961 int gsi_count, i;
962
963 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
964 if (gsi_count > 0) {
965 unsigned int gsi_bits, i;
966
967 /* Round up so we can search ints using ffs */
968 gsi_bits = ALIGN(gsi_count, 32);
969 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
970 s->gsi_count = gsi_count;
971
972 /* Mark any over-allocated bits as already in use */
973 for (i = gsi_count; i < gsi_bits; i++) {
974 set_gsi(s, i);
975 }
976 }
977
978 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
979 s->nr_allocated_irq_routes = 0;
980
981 if (!s->direct_msi) {
982 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
983 QTAILQ_INIT(&s->msi_hashtab[i]);
984 }
985 }
986
987 kvm_arch_init_irq_routing(s);
988 }
989
990 void kvm_irqchip_commit_routes(KVMState *s)
991 {
992 int ret;
993
994 s->irq_routes->flags = 0;
995 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
996 assert(ret == 0);
997 }
998
999 static void kvm_add_routing_entry(KVMState *s,
1000 struct kvm_irq_routing_entry *entry)
1001 {
1002 struct kvm_irq_routing_entry *new;
1003 int n, size;
1004
1005 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1006 n = s->nr_allocated_irq_routes * 2;
1007 if (n < 64) {
1008 n = 64;
1009 }
1010 size = sizeof(struct kvm_irq_routing);
1011 size += n * sizeof(*new);
1012 s->irq_routes = g_realloc(s->irq_routes, size);
1013 s->nr_allocated_irq_routes = n;
1014 }
1015 n = s->irq_routes->nr++;
1016 new = &s->irq_routes->entries[n];
1017
1018 *new = *entry;
1019
1020 set_gsi(s, entry->gsi);
1021 }
1022
1023 static int kvm_update_routing_entry(KVMState *s,
1024 struct kvm_irq_routing_entry *new_entry)
1025 {
1026 struct kvm_irq_routing_entry *entry;
1027 int n;
1028
1029 for (n = 0; n < s->irq_routes->nr; n++) {
1030 entry = &s->irq_routes->entries[n];
1031 if (entry->gsi != new_entry->gsi) {
1032 continue;
1033 }
1034
1035 if(!memcmp(entry, new_entry, sizeof *entry)) {
1036 return 0;
1037 }
1038
1039 *entry = *new_entry;
1040
1041 kvm_irqchip_commit_routes(s);
1042
1043 return 0;
1044 }
1045
1046 return -ESRCH;
1047 }
1048
1049 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1050 {
1051 struct kvm_irq_routing_entry e = {};
1052
1053 assert(pin < s->gsi_count);
1054
1055 e.gsi = irq;
1056 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1057 e.flags = 0;
1058 e.u.irqchip.irqchip = irqchip;
1059 e.u.irqchip.pin = pin;
1060 kvm_add_routing_entry(s, &e);
1061 }
1062
1063 void kvm_irqchip_release_virq(KVMState *s, int virq)
1064 {
1065 struct kvm_irq_routing_entry *e;
1066 int i;
1067
1068 if (kvm_gsi_direct_mapping()) {
1069 return;
1070 }
1071
1072 for (i = 0; i < s->irq_routes->nr; i++) {
1073 e = &s->irq_routes->entries[i];
1074 if (e->gsi == virq) {
1075 s->irq_routes->nr--;
1076 *e = s->irq_routes->entries[s->irq_routes->nr];
1077 }
1078 }
1079 clear_gsi(s, virq);
1080 }
1081
1082 static unsigned int kvm_hash_msi(uint32_t data)
1083 {
1084 /* This is optimized for IA32 MSI layout. However, no other arch shall
1085 * repeat the mistake of not providing a direct MSI injection API. */
1086 return data & 0xff;
1087 }
1088
1089 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1090 {
1091 KVMMSIRoute *route, *next;
1092 unsigned int hash;
1093
1094 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1095 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1096 kvm_irqchip_release_virq(s, route->kroute.gsi);
1097 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1098 g_free(route);
1099 }
1100 }
1101 }
1102
1103 static int kvm_irqchip_get_virq(KVMState *s)
1104 {
1105 uint32_t *word = s->used_gsi_bitmap;
1106 int max_words = ALIGN(s->gsi_count, 32) / 32;
1107 int i, zeroes;
1108
1109 /*
1110 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1111 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1112 * number can succeed even though a new route entry cannot be added.
1113 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1114 */
1115 if (!s->direct_msi && s->irq_routes->nr == s->gsi_count) {
1116 kvm_flush_dynamic_msi_routes(s);
1117 }
1118
1119 /* Return the lowest unused GSI in the bitmap */
1120 for (i = 0; i < max_words; i++) {
1121 zeroes = ctz32(~word[i]);
1122 if (zeroes == 32) {
1123 continue;
1124 }
1125
1126 return zeroes + i * 32;
1127 }
1128 return -ENOSPC;
1129
1130 }
1131
1132 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1133 {
1134 unsigned int hash = kvm_hash_msi(msg.data);
1135 KVMMSIRoute *route;
1136
1137 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1138 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1139 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1140 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1141 return route;
1142 }
1143 }
1144 return NULL;
1145 }
1146
1147 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1148 {
1149 struct kvm_msi msi;
1150 KVMMSIRoute *route;
1151
1152 if (s->direct_msi) {
1153 msi.address_lo = (uint32_t)msg.address;
1154 msi.address_hi = msg.address >> 32;
1155 msi.data = le32_to_cpu(msg.data);
1156 msi.flags = 0;
1157 memset(msi.pad, 0, sizeof(msi.pad));
1158
1159 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1160 }
1161
1162 route = kvm_lookup_msi_route(s, msg);
1163 if (!route) {
1164 int virq;
1165
1166 virq = kvm_irqchip_get_virq(s);
1167 if (virq < 0) {
1168 return virq;
1169 }
1170
1171 route = g_malloc0(sizeof(KVMMSIRoute));
1172 route->kroute.gsi = virq;
1173 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1174 route->kroute.flags = 0;
1175 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1176 route->kroute.u.msi.address_hi = msg.address >> 32;
1177 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1178
1179 kvm_add_routing_entry(s, &route->kroute);
1180 kvm_irqchip_commit_routes(s);
1181
1182 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1183 entry);
1184 }
1185
1186 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1187
1188 return kvm_set_irq(s, route->kroute.gsi, 1);
1189 }
1190
1191 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1192 {
1193 struct kvm_irq_routing_entry kroute = {};
1194 int virq;
1195
1196 if (kvm_gsi_direct_mapping()) {
1197 return kvm_arch_msi_data_to_gsi(msg.data);
1198 }
1199
1200 if (!kvm_gsi_routing_enabled()) {
1201 return -ENOSYS;
1202 }
1203
1204 virq = kvm_irqchip_get_virq(s);
1205 if (virq < 0) {
1206 return virq;
1207 }
1208
1209 kroute.gsi = virq;
1210 kroute.type = KVM_IRQ_ROUTING_MSI;
1211 kroute.flags = 0;
1212 kroute.u.msi.address_lo = (uint32_t)msg.address;
1213 kroute.u.msi.address_hi = msg.address >> 32;
1214 kroute.u.msi.data = le32_to_cpu(msg.data);
1215 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) {
1216 kvm_irqchip_release_virq(s, virq);
1217 return -EINVAL;
1218 }
1219
1220 kvm_add_routing_entry(s, &kroute);
1221 kvm_irqchip_commit_routes(s);
1222
1223 return virq;
1224 }
1225
1226 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1227 {
1228 struct kvm_irq_routing_entry kroute = {};
1229
1230 if (kvm_gsi_direct_mapping()) {
1231 return 0;
1232 }
1233
1234 if (!kvm_irqchip_in_kernel()) {
1235 return -ENOSYS;
1236 }
1237
1238 kroute.gsi = virq;
1239 kroute.type = KVM_IRQ_ROUTING_MSI;
1240 kroute.flags = 0;
1241 kroute.u.msi.address_lo = (uint32_t)msg.address;
1242 kroute.u.msi.address_hi = msg.address >> 32;
1243 kroute.u.msi.data = le32_to_cpu(msg.data);
1244 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) {
1245 return -EINVAL;
1246 }
1247
1248 return kvm_update_routing_entry(s, &kroute);
1249 }
1250
1251 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1252 bool assign)
1253 {
1254 struct kvm_irqfd irqfd = {
1255 .fd = fd,
1256 .gsi = virq,
1257 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1258 };
1259
1260 if (rfd != -1) {
1261 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1262 irqfd.resamplefd = rfd;
1263 }
1264
1265 if (!kvm_irqfds_enabled()) {
1266 return -ENOSYS;
1267 }
1268
1269 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1270 }
1271
1272 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1273 {
1274 struct kvm_irq_routing_entry kroute = {};
1275 int virq;
1276
1277 if (!kvm_gsi_routing_enabled()) {
1278 return -ENOSYS;
1279 }
1280
1281 virq = kvm_irqchip_get_virq(s);
1282 if (virq < 0) {
1283 return virq;
1284 }
1285
1286 kroute.gsi = virq;
1287 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1288 kroute.flags = 0;
1289 kroute.u.adapter.summary_addr = adapter->summary_addr;
1290 kroute.u.adapter.ind_addr = adapter->ind_addr;
1291 kroute.u.adapter.summary_offset = adapter->summary_offset;
1292 kroute.u.adapter.ind_offset = adapter->ind_offset;
1293 kroute.u.adapter.adapter_id = adapter->adapter_id;
1294
1295 kvm_add_routing_entry(s, &kroute);
1296 kvm_irqchip_commit_routes(s);
1297
1298 return virq;
1299 }
1300
1301 #else /* !KVM_CAP_IRQ_ROUTING */
1302
1303 void kvm_init_irq_routing(KVMState *s)
1304 {
1305 }
1306
1307 void kvm_irqchip_release_virq(KVMState *s, int virq)
1308 {
1309 }
1310
1311 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1312 {
1313 abort();
1314 }
1315
1316 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1317 {
1318 return -ENOSYS;
1319 }
1320
1321 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1322 {
1323 return -ENOSYS;
1324 }
1325
1326 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1327 {
1328 abort();
1329 }
1330
1331 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1332 {
1333 return -ENOSYS;
1334 }
1335 #endif /* !KVM_CAP_IRQ_ROUTING */
1336
1337 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1338 EventNotifier *rn, int virq)
1339 {
1340 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1341 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1342 }
1343
1344 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1345 int virq)
1346 {
1347 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1348 false);
1349 }
1350
1351 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1352 EventNotifier *rn, qemu_irq irq)
1353 {
1354 gpointer key, gsi;
1355 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1356
1357 if (!found) {
1358 return -ENXIO;
1359 }
1360 return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi));
1361 }
1362
1363 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
1364 qemu_irq irq)
1365 {
1366 gpointer key, gsi;
1367 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1368
1369 if (!found) {
1370 return -ENXIO;
1371 }
1372 return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi));
1373 }
1374
1375 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
1376 {
1377 g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
1378 }
1379
1380 static void kvm_irqchip_create(MachineState *machine, KVMState *s)
1381 {
1382 int ret;
1383
1384 if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1385 ;
1386 } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) {
1387 ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0);
1388 if (ret < 0) {
1389 fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret));
1390 exit(1);
1391 }
1392 } else {
1393 return;
1394 }
1395
1396 /* First probe and see if there's a arch-specific hook to create the
1397 * in-kernel irqchip for us */
1398 ret = kvm_arch_irqchip_create(s);
1399 if (ret == 0) {
1400 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1401 }
1402 if (ret < 0) {
1403 fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
1404 exit(1);
1405 }
1406
1407 kvm_kernel_irqchip = true;
1408 /* If we have an in-kernel IRQ chip then we must have asynchronous
1409 * interrupt delivery (though the reverse is not necessarily true)
1410 */
1411 kvm_async_interrupts_allowed = true;
1412 kvm_halt_in_kernel_allowed = true;
1413
1414 kvm_init_irq_routing(s);
1415
1416 s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal);
1417 }
1418
1419 /* Find number of supported CPUs using the recommended
1420 * procedure from the kernel API documentation to cope with
1421 * older kernels that may be missing capabilities.
1422 */
1423 static int kvm_recommended_vcpus(KVMState *s)
1424 {
1425 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1426 return (ret) ? ret : 4;
1427 }
1428
1429 static int kvm_max_vcpus(KVMState *s)
1430 {
1431 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1432 return (ret) ? ret : kvm_recommended_vcpus(s);
1433 }
1434
1435 static int kvm_init(MachineState *ms)
1436 {
1437 MachineClass *mc = MACHINE_GET_CLASS(ms);
1438 static const char upgrade_note[] =
1439 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1440 "(see http://sourceforge.net/projects/kvm).\n";
1441 struct {
1442 const char *name;
1443 int num;
1444 } num_cpus[] = {
1445 { "SMP", smp_cpus },
1446 { "hotpluggable", max_cpus },
1447 { NULL, }
1448 }, *nc = num_cpus;
1449 int soft_vcpus_limit, hard_vcpus_limit;
1450 KVMState *s;
1451 const KVMCapabilityInfo *missing_cap;
1452 int ret;
1453 int type = 0;
1454 const char *kvm_type;
1455
1456 s = KVM_STATE(ms->accelerator);
1457
1458 /*
1459 * On systems where the kernel can support different base page
1460 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1461 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1462 * page size for the system though.
1463 */
1464 assert(TARGET_PAGE_SIZE <= getpagesize());
1465 page_size_init();
1466
1467 s->sigmask_len = 8;
1468
1469 #ifdef KVM_CAP_SET_GUEST_DEBUG
1470 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1471 #endif
1472 s->vmfd = -1;
1473 s->fd = qemu_open("/dev/kvm", O_RDWR);
1474 if (s->fd == -1) {
1475 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1476 ret = -errno;
1477 goto err;
1478 }
1479
1480 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1481 if (ret < KVM_API_VERSION) {
1482 if (ret >= 0) {
1483 ret = -EINVAL;
1484 }
1485 fprintf(stderr, "kvm version too old\n");
1486 goto err;
1487 }
1488
1489 if (ret > KVM_API_VERSION) {
1490 ret = -EINVAL;
1491 fprintf(stderr, "kvm version not supported\n");
1492 goto err;
1493 }
1494
1495 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1496
1497 /* If unspecified, use the default value */
1498 if (!s->nr_slots) {
1499 s->nr_slots = 32;
1500 }
1501
1502 /* check the vcpu limits */
1503 soft_vcpus_limit = kvm_recommended_vcpus(s);
1504 hard_vcpus_limit = kvm_max_vcpus(s);
1505
1506 while (nc->name) {
1507 if (nc->num > soft_vcpus_limit) {
1508 fprintf(stderr,
1509 "Warning: Number of %s cpus requested (%d) exceeds "
1510 "the recommended cpus supported by KVM (%d)\n",
1511 nc->name, nc->num, soft_vcpus_limit);
1512
1513 if (nc->num > hard_vcpus_limit) {
1514 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1515 "the maximum cpus supported by KVM (%d)\n",
1516 nc->name, nc->num, hard_vcpus_limit);
1517 exit(1);
1518 }
1519 }
1520 nc++;
1521 }
1522
1523 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1524 if (mc->kvm_type) {
1525 type = mc->kvm_type(kvm_type);
1526 } else if (kvm_type) {
1527 ret = -EINVAL;
1528 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1529 goto err;
1530 }
1531
1532 do {
1533 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1534 } while (ret == -EINTR);
1535
1536 if (ret < 0) {
1537 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1538 strerror(-ret));
1539
1540 #ifdef TARGET_S390X
1541 if (ret == -EINVAL) {
1542 fprintf(stderr,
1543 "Host kernel setup problem detected. Please verify:\n");
1544 fprintf(stderr, "- for kernels supporting the switch_amode or"
1545 " user_mode parameters, whether\n");
1546 fprintf(stderr,
1547 " user space is running in primary address space\n");
1548 fprintf(stderr,
1549 "- for kernels supporting the vm.allocate_pgste sysctl, "
1550 "whether it is enabled\n");
1551 }
1552 #endif
1553 goto err;
1554 }
1555
1556 s->vmfd = ret;
1557 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1558 if (!missing_cap) {
1559 missing_cap =
1560 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1561 }
1562 if (missing_cap) {
1563 ret = -EINVAL;
1564 fprintf(stderr, "kvm does not support %s\n%s",
1565 missing_cap->name, upgrade_note);
1566 goto err;
1567 }
1568
1569 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1570
1571 s->broken_set_mem_region = 1;
1572 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1573 if (ret > 0) {
1574 s->broken_set_mem_region = 0;
1575 }
1576
1577 #ifdef KVM_CAP_VCPU_EVENTS
1578 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1579 #endif
1580
1581 s->robust_singlestep =
1582 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1583
1584 #ifdef KVM_CAP_DEBUGREGS
1585 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1586 #endif
1587
1588 #ifdef KVM_CAP_XSAVE
1589 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1590 #endif
1591
1592 #ifdef KVM_CAP_XCRS
1593 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1594 #endif
1595
1596 #ifdef KVM_CAP_PIT_STATE2
1597 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1598 #endif
1599
1600 #ifdef KVM_CAP_IRQ_ROUTING
1601 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1602 #endif
1603
1604 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1605
1606 s->irq_set_ioctl = KVM_IRQ_LINE;
1607 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1608 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1609 }
1610
1611 #ifdef KVM_CAP_READONLY_MEM
1612 kvm_readonly_mem_allowed =
1613 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1614 #endif
1615
1616 kvm_eventfds_allowed =
1617 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1618
1619 kvm_irqfds_allowed =
1620 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1621
1622 kvm_resamplefds_allowed =
1623 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1624
1625 kvm_vm_attributes_allowed =
1626 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1627
1628 ret = kvm_arch_init(ms, s);
1629 if (ret < 0) {
1630 goto err;
1631 }
1632
1633 if (machine_kernel_irqchip_allowed(ms)) {
1634 kvm_irqchip_create(ms, s);
1635 }
1636
1637 kvm_state = s;
1638
1639 s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
1640 s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
1641 s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
1642 s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
1643
1644 kvm_memory_listener_register(s, &s->memory_listener,
1645 &address_space_memory, 0);
1646 memory_listener_register(&kvm_io_listener,
1647 &address_space_io);
1648
1649 s->many_ioeventfds = kvm_check_many_ioeventfds();
1650
1651 cpu_interrupt_handler = kvm_handle_interrupt;
1652
1653 return 0;
1654
1655 err:
1656 assert(ret < 0);
1657 if (s->vmfd >= 0) {
1658 close(s->vmfd);
1659 }
1660 if (s->fd != -1) {
1661 close(s->fd);
1662 }
1663 g_free(s->memory_listener.slots);
1664
1665 return ret;
1666 }
1667
1668 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1669 {
1670 s->sigmask_len = sigmask_len;
1671 }
1672
1673 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1674 int size, uint32_t count)
1675 {
1676 int i;
1677 uint8_t *ptr = data;
1678
1679 for (i = 0; i < count; i++) {
1680 address_space_rw(&address_space_io, port, attrs,
1681 ptr, size,
1682 direction == KVM_EXIT_IO_OUT);
1683 ptr += size;
1684 }
1685 }
1686
1687 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1688 {
1689 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1690 run->internal.suberror);
1691
1692 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1693 int i;
1694
1695 for (i = 0; i < run->internal.ndata; ++i) {
1696 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1697 i, (uint64_t)run->internal.data[i]);
1698 }
1699 }
1700 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1701 fprintf(stderr, "emulation failure\n");
1702 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1703 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1704 return EXCP_INTERRUPT;
1705 }
1706 }
1707 /* FIXME: Should trigger a qmp message to let management know
1708 * something went wrong.
1709 */
1710 return -1;
1711 }
1712
1713 void kvm_flush_coalesced_mmio_buffer(void)
1714 {
1715 KVMState *s = kvm_state;
1716
1717 if (s->coalesced_flush_in_progress) {
1718 return;
1719 }
1720
1721 s->coalesced_flush_in_progress = true;
1722
1723 if (s->coalesced_mmio_ring) {
1724 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1725 while (ring->first != ring->last) {
1726 struct kvm_coalesced_mmio *ent;
1727
1728 ent = &ring->coalesced_mmio[ring->first];
1729
1730 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1731 smp_wmb();
1732 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1733 }
1734 }
1735
1736 s->coalesced_flush_in_progress = false;
1737 }
1738
1739 static void do_kvm_cpu_synchronize_state(void *arg)
1740 {
1741 CPUState *cpu = arg;
1742
1743 if (!cpu->kvm_vcpu_dirty) {
1744 kvm_arch_get_registers(cpu);
1745 cpu->kvm_vcpu_dirty = true;
1746 }
1747 }
1748
1749 void kvm_cpu_synchronize_state(CPUState *cpu)
1750 {
1751 if (!cpu->kvm_vcpu_dirty) {
1752 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1753 }
1754 }
1755
1756 static void do_kvm_cpu_synchronize_post_reset(void *arg)
1757 {
1758 CPUState *cpu = arg;
1759
1760 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1761 cpu->kvm_vcpu_dirty = false;
1762 }
1763
1764 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1765 {
1766 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu);
1767 }
1768
1769 static void do_kvm_cpu_synchronize_post_init(void *arg)
1770 {
1771 CPUState *cpu = arg;
1772
1773 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1774 cpu->kvm_vcpu_dirty = false;
1775 }
1776
1777 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1778 {
1779 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu);
1780 }
1781
1782 void kvm_cpu_clean_state(CPUState *cpu)
1783 {
1784 cpu->kvm_vcpu_dirty = false;
1785 }
1786
1787 int kvm_cpu_exec(CPUState *cpu)
1788 {
1789 struct kvm_run *run = cpu->kvm_run;
1790 int ret, run_ret;
1791
1792 DPRINTF("kvm_cpu_exec()\n");
1793
1794 if (kvm_arch_process_async_events(cpu)) {
1795 cpu->exit_request = 0;
1796 return EXCP_HLT;
1797 }
1798
1799 qemu_mutex_unlock_iothread();
1800
1801 do {
1802 MemTxAttrs attrs;
1803
1804 if (cpu->kvm_vcpu_dirty) {
1805 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1806 cpu->kvm_vcpu_dirty = false;
1807 }
1808
1809 kvm_arch_pre_run(cpu, run);
1810 if (cpu->exit_request) {
1811 DPRINTF("interrupt exit requested\n");
1812 /*
1813 * KVM requires us to reenter the kernel after IO exits to complete
1814 * instruction emulation. This self-signal will ensure that we
1815 * leave ASAP again.
1816 */
1817 qemu_cpu_kick_self();
1818 }
1819
1820 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1821
1822 attrs = kvm_arch_post_run(cpu, run);
1823
1824 if (run_ret < 0) {
1825 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1826 DPRINTF("io window exit\n");
1827 ret = EXCP_INTERRUPT;
1828 break;
1829 }
1830 fprintf(stderr, "error: kvm run failed %s\n",
1831 strerror(-run_ret));
1832 #ifdef TARGET_PPC
1833 if (run_ret == -EBUSY) {
1834 fprintf(stderr,
1835 "This is probably because your SMT is enabled.\n"
1836 "VCPU can only run on primary threads with all "
1837 "secondary threads offline.\n");
1838 }
1839 #endif
1840 ret = -1;
1841 break;
1842 }
1843
1844 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1845 switch (run->exit_reason) {
1846 case KVM_EXIT_IO:
1847 DPRINTF("handle_io\n");
1848 /* Called outside BQL */
1849 kvm_handle_io(run->io.port, attrs,
1850 (uint8_t *)run + run->io.data_offset,
1851 run->io.direction,
1852 run->io.size,
1853 run->io.count);
1854 ret = 0;
1855 break;
1856 case KVM_EXIT_MMIO:
1857 DPRINTF("handle_mmio\n");
1858 /* Called outside BQL */
1859 address_space_rw(&address_space_memory,
1860 run->mmio.phys_addr, attrs,
1861 run->mmio.data,
1862 run->mmio.len,
1863 run->mmio.is_write);
1864 ret = 0;
1865 break;
1866 case KVM_EXIT_IRQ_WINDOW_OPEN:
1867 DPRINTF("irq_window_open\n");
1868 ret = EXCP_INTERRUPT;
1869 break;
1870 case KVM_EXIT_SHUTDOWN:
1871 DPRINTF("shutdown\n");
1872 qemu_system_reset_request();
1873 ret = EXCP_INTERRUPT;
1874 break;
1875 case KVM_EXIT_UNKNOWN:
1876 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1877 (uint64_t)run->hw.hardware_exit_reason);
1878 ret = -1;
1879 break;
1880 case KVM_EXIT_INTERNAL_ERROR:
1881 ret = kvm_handle_internal_error(cpu, run);
1882 break;
1883 case KVM_EXIT_SYSTEM_EVENT:
1884 switch (run->system_event.type) {
1885 case KVM_SYSTEM_EVENT_SHUTDOWN:
1886 qemu_system_shutdown_request();
1887 ret = EXCP_INTERRUPT;
1888 break;
1889 case KVM_SYSTEM_EVENT_RESET:
1890 qemu_system_reset_request();
1891 ret = EXCP_INTERRUPT;
1892 break;
1893 default:
1894 DPRINTF("kvm_arch_handle_exit\n");
1895 ret = kvm_arch_handle_exit(cpu, run);
1896 break;
1897 }
1898 break;
1899 default:
1900 DPRINTF("kvm_arch_handle_exit\n");
1901 ret = kvm_arch_handle_exit(cpu, run);
1902 break;
1903 }
1904 } while (ret == 0);
1905
1906 qemu_mutex_lock_iothread();
1907
1908 if (ret < 0) {
1909 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1910 vm_stop(RUN_STATE_INTERNAL_ERROR);
1911 }
1912
1913 cpu->exit_request = 0;
1914 return ret;
1915 }
1916
1917 int kvm_ioctl(KVMState *s, int type, ...)
1918 {
1919 int ret;
1920 void *arg;
1921 va_list ap;
1922
1923 va_start(ap, type);
1924 arg = va_arg(ap, void *);
1925 va_end(ap);
1926
1927 trace_kvm_ioctl(type, arg);
1928 ret = ioctl(s->fd, type, arg);
1929 if (ret == -1) {
1930 ret = -errno;
1931 }
1932 return ret;
1933 }
1934
1935 int kvm_vm_ioctl(KVMState *s, int type, ...)
1936 {
1937 int ret;
1938 void *arg;
1939 va_list ap;
1940
1941 va_start(ap, type);
1942 arg = va_arg(ap, void *);
1943 va_end(ap);
1944
1945 trace_kvm_vm_ioctl(type, arg);
1946 ret = ioctl(s->vmfd, type, arg);
1947 if (ret == -1) {
1948 ret = -errno;
1949 }
1950 return ret;
1951 }
1952
1953 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1954 {
1955 int ret;
1956 void *arg;
1957 va_list ap;
1958
1959 va_start(ap, type);
1960 arg = va_arg(ap, void *);
1961 va_end(ap);
1962
1963 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1964 ret = ioctl(cpu->kvm_fd, type, arg);
1965 if (ret == -1) {
1966 ret = -errno;
1967 }
1968 return ret;
1969 }
1970
1971 int kvm_device_ioctl(int fd, int type, ...)
1972 {
1973 int ret;
1974 void *arg;
1975 va_list ap;
1976
1977 va_start(ap, type);
1978 arg = va_arg(ap, void *);
1979 va_end(ap);
1980
1981 trace_kvm_device_ioctl(fd, type, arg);
1982 ret = ioctl(fd, type, arg);
1983 if (ret == -1) {
1984 ret = -errno;
1985 }
1986 return ret;
1987 }
1988
1989 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
1990 {
1991 int ret;
1992 struct kvm_device_attr attribute = {
1993 .group = group,
1994 .attr = attr,
1995 };
1996
1997 if (!kvm_vm_attributes_allowed) {
1998 return 0;
1999 }
2000
2001 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
2002 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2003 return ret ? 0 : 1;
2004 }
2005
2006 int kvm_has_sync_mmu(void)
2007 {
2008 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
2009 }
2010
2011 int kvm_has_vcpu_events(void)
2012 {
2013 return kvm_state->vcpu_events;
2014 }
2015
2016 int kvm_has_robust_singlestep(void)
2017 {
2018 return kvm_state->robust_singlestep;
2019 }
2020
2021 int kvm_has_debugregs(void)
2022 {
2023 return kvm_state->debugregs;
2024 }
2025
2026 int kvm_has_xsave(void)
2027 {
2028 return kvm_state->xsave;
2029 }
2030
2031 int kvm_has_xcrs(void)
2032 {
2033 return kvm_state->xcrs;
2034 }
2035
2036 int kvm_has_pit_state2(void)
2037 {
2038 return kvm_state->pit_state2;
2039 }
2040
2041 int kvm_has_many_ioeventfds(void)
2042 {
2043 if (!kvm_enabled()) {
2044 return 0;
2045 }
2046 return kvm_state->many_ioeventfds;
2047 }
2048
2049 int kvm_has_gsi_routing(void)
2050 {
2051 #ifdef KVM_CAP_IRQ_ROUTING
2052 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2053 #else
2054 return false;
2055 #endif
2056 }
2057
2058 int kvm_has_intx_set_mask(void)
2059 {
2060 return kvm_state->intx_set_mask;
2061 }
2062
2063 void kvm_setup_guest_memory(void *start, size_t size)
2064 {
2065 if (!kvm_has_sync_mmu()) {
2066 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
2067
2068 if (ret) {
2069 perror("qemu_madvise");
2070 fprintf(stderr,
2071 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
2072 exit(1);
2073 }
2074 }
2075 }
2076
2077 #ifdef KVM_CAP_SET_GUEST_DEBUG
2078 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2079 target_ulong pc)
2080 {
2081 struct kvm_sw_breakpoint *bp;
2082
2083 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2084 if (bp->pc == pc) {
2085 return bp;
2086 }
2087 }
2088 return NULL;
2089 }
2090
2091 int kvm_sw_breakpoints_active(CPUState *cpu)
2092 {
2093 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2094 }
2095
2096 struct kvm_set_guest_debug_data {
2097 struct kvm_guest_debug dbg;
2098 CPUState *cpu;
2099 int err;
2100 };
2101
2102 static void kvm_invoke_set_guest_debug(void *data)
2103 {
2104 struct kvm_set_guest_debug_data *dbg_data = data;
2105
2106 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
2107 &dbg_data->dbg);
2108 }
2109
2110 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2111 {
2112 struct kvm_set_guest_debug_data data;
2113
2114 data.dbg.control = reinject_trap;
2115
2116 if (cpu->singlestep_enabled) {
2117 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2118 }
2119 kvm_arch_update_guest_debug(cpu, &data.dbg);
2120 data.cpu = cpu;
2121
2122 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
2123 return data.err;
2124 }
2125
2126 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2127 target_ulong len, int type)
2128 {
2129 struct kvm_sw_breakpoint *bp;
2130 int err;
2131
2132 if (type == GDB_BREAKPOINT_SW) {
2133 bp = kvm_find_sw_breakpoint(cpu, addr);
2134 if (bp) {
2135 bp->use_count++;
2136 return 0;
2137 }
2138
2139 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2140 bp->pc = addr;
2141 bp->use_count = 1;
2142 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2143 if (err) {
2144 g_free(bp);
2145 return err;
2146 }
2147
2148 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2149 } else {
2150 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2151 if (err) {
2152 return err;
2153 }
2154 }
2155
2156 CPU_FOREACH(cpu) {
2157 err = kvm_update_guest_debug(cpu, 0);
2158 if (err) {
2159 return err;
2160 }
2161 }
2162 return 0;
2163 }
2164
2165 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2166 target_ulong len, int type)
2167 {
2168 struct kvm_sw_breakpoint *bp;
2169 int err;
2170
2171 if (type == GDB_BREAKPOINT_SW) {
2172 bp = kvm_find_sw_breakpoint(cpu, addr);
2173 if (!bp) {
2174 return -ENOENT;
2175 }
2176
2177 if (bp->use_count > 1) {
2178 bp->use_count--;
2179 return 0;
2180 }
2181
2182 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2183 if (err) {
2184 return err;
2185 }
2186
2187 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2188 g_free(bp);
2189 } else {
2190 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2191 if (err) {
2192 return err;
2193 }
2194 }
2195
2196 CPU_FOREACH(cpu) {
2197 err = kvm_update_guest_debug(cpu, 0);
2198 if (err) {
2199 return err;
2200 }
2201 }
2202 return 0;
2203 }
2204
2205 void kvm_remove_all_breakpoints(CPUState *cpu)
2206 {
2207 struct kvm_sw_breakpoint *bp, *next;
2208 KVMState *s = cpu->kvm_state;
2209 CPUState *tmpcpu;
2210
2211 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2212 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2213 /* Try harder to find a CPU that currently sees the breakpoint. */
2214 CPU_FOREACH(tmpcpu) {
2215 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2216 break;
2217 }
2218 }
2219 }
2220 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2221 g_free(bp);
2222 }
2223 kvm_arch_remove_all_hw_breakpoints();
2224
2225 CPU_FOREACH(cpu) {
2226 kvm_update_guest_debug(cpu, 0);
2227 }
2228 }
2229
2230 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2231
2232 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2233 {
2234 return -EINVAL;
2235 }
2236
2237 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2238 target_ulong len, int type)
2239 {
2240 return -EINVAL;
2241 }
2242
2243 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2244 target_ulong len, int type)
2245 {
2246 return -EINVAL;
2247 }
2248
2249 void kvm_remove_all_breakpoints(CPUState *cpu)
2250 {
2251 }
2252 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2253
2254 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2255 {
2256 KVMState *s = kvm_state;
2257 struct kvm_signal_mask *sigmask;
2258 int r;
2259
2260 if (!sigset) {
2261 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2262 }
2263
2264 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2265
2266 sigmask->len = s->sigmask_len;
2267 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2268 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2269 g_free(sigmask);
2270
2271 return r;
2272 }
2273 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2274 {
2275 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2276 }
2277
2278 int kvm_on_sigbus(int code, void *addr)
2279 {
2280 return kvm_arch_on_sigbus(code, addr);
2281 }
2282
2283 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2284 {
2285 int ret;
2286 struct kvm_create_device create_dev;
2287
2288 create_dev.type = type;
2289 create_dev.fd = -1;
2290 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2291
2292 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2293 return -ENOTSUP;
2294 }
2295
2296 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2297 if (ret) {
2298 return ret;
2299 }
2300
2301 return test ? 0 : create_dev.fd;
2302 }
2303
2304 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2305 {
2306 struct kvm_one_reg reg;
2307 int r;
2308
2309 reg.id = id;
2310 reg.addr = (uintptr_t) source;
2311 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
2312 if (r) {
2313 trace_kvm_failed_reg_set(id, strerror(r));
2314 }
2315 return r;
2316 }
2317
2318 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2319 {
2320 struct kvm_one_reg reg;
2321 int r;
2322
2323 reg.id = id;
2324 reg.addr = (uintptr_t) target;
2325 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
2326 if (r) {
2327 trace_kvm_failed_reg_get(id, strerror(r));
2328 }
2329 return r;
2330 }
2331
2332 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2333 {
2334 AccelClass *ac = ACCEL_CLASS(oc);
2335 ac->name = "KVM";
2336 ac->init_machine = kvm_init;
2337 ac->allowed = &kvm_allowed;
2338 }
2339
2340 static const TypeInfo kvm_accel_type = {
2341 .name = TYPE_KVM_ACCEL,
2342 .parent = TYPE_ACCEL,
2343 .class_init = kvm_accel_class_init,
2344 .instance_size = sizeof(KVMState),
2345 };
2346
2347 static void kvm_type_init(void)
2348 {
2349 type_register_static(&kvm_accel_type);
2350 }
2351
2352 type_init(kvm_type_init);