hw/misc: add a TMP42{1, 2, 3} device model
[qemu/rayw.git] / kvm-all.c
blob44b3cf43cc73d9837d24eb41877d5d6827d1bc7f
1 /*
2 * QEMU KVM support
4 * Copyright IBM, Corp. 2008
5 * Red Hat, Inc. 2008
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Glauber Costa <gcosta@redhat.com>
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.
16 #include "qemu/osdep.h"
17 #include <sys/ioctl.h>
19 #include <linux/kvm.h>
21 #include "qemu-common.h"
22 #include "qemu/atomic.h"
23 #include "qemu/option.h"
24 #include "qemu/config-file.h"
25 #include "qemu/error-report.h"
26 #include "hw/hw.h"
27 #include "hw/pci/msi.h"
28 #include "hw/pci/msix.h"
29 #include "hw/s390x/adapter.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm_int.h"
32 #include "sysemu/cpus.h"
33 #include "qemu/bswap.h"
34 #include "exec/memory.h"
35 #include "exec/ram_addr.h"
36 #include "exec/address-spaces.h"
37 #include "qemu/event_notifier.h"
38 #include "trace-root.h"
39 #include "hw/irq.h"
41 #include "hw/boards.h"
43 /* This check must be after config-host.h is included */
44 #ifdef CONFIG_EVENTFD
45 #include <sys/eventfd.h>
46 #endif
48 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
49 * need to use the real host PAGE_SIZE, as that's what KVM will use.
51 #define PAGE_SIZE getpagesize()
53 //#define DEBUG_KVM
55 #ifdef DEBUG_KVM
56 #define DPRINTF(fmt, ...) \
57 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
58 #else
59 #define DPRINTF(fmt, ...) \
60 do { } while (0)
61 #endif
63 #define KVM_MSI_HASHTAB_SIZE 256
65 struct KVMParkedVcpu {
66 unsigned long vcpu_id;
67 int kvm_fd;
68 QLIST_ENTRY(KVMParkedVcpu) node;
71 struct KVMState
73 AccelState parent_obj;
75 int nr_slots;
76 int fd;
77 int vmfd;
78 int coalesced_mmio;
79 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
80 bool coalesced_flush_in_progress;
81 int broken_set_mem_region;
82 int vcpu_events;
83 int robust_singlestep;
84 int debugregs;
85 #ifdef KVM_CAP_SET_GUEST_DEBUG
86 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
87 #endif
88 int many_ioeventfds;
89 int intx_set_mask;
90 /* The man page (and posix) say ioctl numbers are signed int, but
91 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
92 * unsigned, and treating them as signed here can break things */
93 unsigned irq_set_ioctl;
94 unsigned int sigmask_len;
95 GHashTable *gsimap;
96 #ifdef KVM_CAP_IRQ_ROUTING
97 struct kvm_irq_routing *irq_routes;
98 int nr_allocated_irq_routes;
99 unsigned long *used_gsi_bitmap;
100 unsigned int gsi_count;
101 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
102 #endif
103 KVMMemoryListener memory_listener;
104 QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus;
107 KVMState *kvm_state;
108 bool kvm_kernel_irqchip;
109 bool kvm_split_irqchip;
110 bool kvm_async_interrupts_allowed;
111 bool kvm_halt_in_kernel_allowed;
112 bool kvm_eventfds_allowed;
113 bool kvm_irqfds_allowed;
114 bool kvm_resamplefds_allowed;
115 bool kvm_msi_via_irqfd_allowed;
116 bool kvm_gsi_routing_allowed;
117 bool kvm_gsi_direct_mapping;
118 bool kvm_allowed;
119 bool kvm_readonly_mem_allowed;
120 bool kvm_vm_attributes_allowed;
121 bool kvm_direct_msi_allowed;
122 bool kvm_ioeventfd_any_length_allowed;
123 bool kvm_msi_use_devid;
124 static bool kvm_immediate_exit;
126 static const KVMCapabilityInfo kvm_required_capabilites[] = {
127 KVM_CAP_INFO(USER_MEMORY),
128 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
129 KVM_CAP_LAST_INFO
132 int kvm_get_max_memslots(void)
134 KVMState *s = KVM_STATE(current_machine->accelerator);
136 return s->nr_slots;
139 static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml)
141 KVMState *s = kvm_state;
142 int i;
144 for (i = 0; i < s->nr_slots; i++) {
145 if (kml->slots[i].memory_size == 0) {
146 return &kml->slots[i];
150 return NULL;
153 bool kvm_has_free_slot(MachineState *ms)
155 KVMState *s = KVM_STATE(ms->accelerator);
157 return kvm_get_free_slot(&s->memory_listener);
160 static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml)
162 KVMSlot *slot = kvm_get_free_slot(kml);
164 if (slot) {
165 return slot;
168 fprintf(stderr, "%s: no free slot available\n", __func__);
169 abort();
172 static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml,
173 hwaddr start_addr,
174 hwaddr end_addr)
176 KVMState *s = kvm_state;
177 int i;
179 for (i = 0; i < s->nr_slots; i++) {
180 KVMSlot *mem = &kml->slots[i];
182 if (start_addr == mem->start_addr &&
183 end_addr == mem->start_addr + mem->memory_size) {
184 return mem;
188 return NULL;
192 * Find overlapping slot with lowest start address
194 static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml,
195 hwaddr start_addr,
196 hwaddr end_addr)
198 KVMState *s = kvm_state;
199 KVMSlot *found = NULL;
200 int i;
202 for (i = 0; i < s->nr_slots; i++) {
203 KVMSlot *mem = &kml->slots[i];
205 if (mem->memory_size == 0 ||
206 (found && found->start_addr < mem->start_addr)) {
207 continue;
210 if (end_addr > mem->start_addr &&
211 start_addr < mem->start_addr + mem->memory_size) {
212 found = mem;
216 return found;
219 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
220 hwaddr *phys_addr)
222 KVMMemoryListener *kml = &s->memory_listener;
223 int i;
225 for (i = 0; i < s->nr_slots; i++) {
226 KVMSlot *mem = &kml->slots[i];
228 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
229 *phys_addr = mem->start_addr + (ram - mem->ram);
230 return 1;
234 return 0;
237 static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot)
239 KVMState *s = kvm_state;
240 struct kvm_userspace_memory_region mem;
242 mem.slot = slot->slot | (kml->as_id << 16);
243 mem.guest_phys_addr = slot->start_addr;
244 mem.userspace_addr = (unsigned long)slot->ram;
245 mem.flags = slot->flags;
247 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
248 /* Set the slot size to 0 before setting the slot to the desired
249 * value. This is needed based on KVM commit 75d61fbc. */
250 mem.memory_size = 0;
251 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
253 mem.memory_size = slot->memory_size;
254 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
257 int kvm_destroy_vcpu(CPUState *cpu)
259 KVMState *s = kvm_state;
260 long mmap_size;
261 struct KVMParkedVcpu *vcpu = NULL;
262 int ret = 0;
264 DPRINTF("kvm_destroy_vcpu\n");
266 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
267 if (mmap_size < 0) {
268 ret = mmap_size;
269 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
270 goto err;
273 ret = munmap(cpu->kvm_run, mmap_size);
274 if (ret < 0) {
275 goto err;
278 vcpu = g_malloc0(sizeof(*vcpu));
279 vcpu->vcpu_id = kvm_arch_vcpu_id(cpu);
280 vcpu->kvm_fd = cpu->kvm_fd;
281 QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node);
282 err:
283 return ret;
286 static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id)
288 struct KVMParkedVcpu *cpu;
290 QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) {
291 if (cpu->vcpu_id == vcpu_id) {
292 int kvm_fd;
294 QLIST_REMOVE(cpu, node);
295 kvm_fd = cpu->kvm_fd;
296 g_free(cpu);
297 return kvm_fd;
301 return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id);
304 int kvm_init_vcpu(CPUState *cpu)
306 KVMState *s = kvm_state;
307 long mmap_size;
308 int ret;
310 DPRINTF("kvm_init_vcpu\n");
312 ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu));
313 if (ret < 0) {
314 DPRINTF("kvm_create_vcpu failed\n");
315 goto err;
318 cpu->kvm_fd = ret;
319 cpu->kvm_state = s;
320 cpu->kvm_vcpu_dirty = true;
322 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
323 if (mmap_size < 0) {
324 ret = mmap_size;
325 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
326 goto err;
329 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
330 cpu->kvm_fd, 0);
331 if (cpu->kvm_run == MAP_FAILED) {
332 ret = -errno;
333 DPRINTF("mmap'ing vcpu state failed\n");
334 goto err;
337 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
338 s->coalesced_mmio_ring =
339 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
342 ret = kvm_arch_init_vcpu(cpu);
343 err:
344 return ret;
348 * dirty pages logging control
351 static int kvm_mem_flags(MemoryRegion *mr)
353 bool readonly = mr->readonly || memory_region_is_romd(mr);
354 int flags = 0;
356 if (memory_region_get_dirty_log_mask(mr) != 0) {
357 flags |= KVM_MEM_LOG_DIRTY_PAGES;
359 if (readonly && kvm_readonly_mem_allowed) {
360 flags |= KVM_MEM_READONLY;
362 return flags;
365 static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem,
366 MemoryRegion *mr)
368 int old_flags;
370 old_flags = mem->flags;
371 mem->flags = kvm_mem_flags(mr);
373 /* If nothing changed effectively, no need to issue ioctl */
374 if (mem->flags == old_flags) {
375 return 0;
378 return kvm_set_user_memory_region(kml, mem);
381 static int kvm_section_update_flags(KVMMemoryListener *kml,
382 MemoryRegionSection *section)
384 hwaddr phys_addr = section->offset_within_address_space;
385 ram_addr_t size = int128_get64(section->size);
386 KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size);
388 if (mem == NULL) {
389 return 0;
390 } else {
391 return kvm_slot_update_flags(kml, mem, section->mr);
395 static void kvm_log_start(MemoryListener *listener,
396 MemoryRegionSection *section,
397 int old, int new)
399 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
400 int r;
402 if (old != 0) {
403 return;
406 r = kvm_section_update_flags(kml, section);
407 if (r < 0) {
408 abort();
412 static void kvm_log_stop(MemoryListener *listener,
413 MemoryRegionSection *section,
414 int old, int new)
416 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
417 int r;
419 if (new != 0) {
420 return;
423 r = kvm_section_update_flags(kml, section);
424 if (r < 0) {
425 abort();
429 /* get kvm's dirty pages bitmap and update qemu's */
430 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
431 unsigned long *bitmap)
433 ram_addr_t start = section->offset_within_region +
434 memory_region_get_ram_addr(section->mr);
435 ram_addr_t pages = int128_get64(section->size) / getpagesize();
437 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
438 return 0;
441 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
444 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
445 * This function updates qemu's dirty bitmap using
446 * memory_region_set_dirty(). This means all bits are set
447 * to dirty.
449 * @start_add: start of logged region.
450 * @end_addr: end of logged region.
452 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
453 MemoryRegionSection *section)
455 KVMState *s = kvm_state;
456 unsigned long size, allocated_size = 0;
457 struct kvm_dirty_log d = {};
458 KVMSlot *mem;
459 int ret = 0;
460 hwaddr start_addr = section->offset_within_address_space;
461 hwaddr end_addr = start_addr + int128_get64(section->size);
463 d.dirty_bitmap = NULL;
464 while (start_addr < end_addr) {
465 mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr);
466 if (mem == NULL) {
467 break;
470 /* XXX bad kernel interface alert
471 * For dirty bitmap, kernel allocates array of size aligned to
472 * bits-per-long. But for case when the kernel is 64bits and
473 * the userspace is 32bits, userspace can't align to the same
474 * bits-per-long, since sizeof(long) is different between kernel
475 * and user space. This way, userspace will provide buffer which
476 * may be 4 bytes less than the kernel will use, resulting in
477 * userspace memory corruption (which is not detectable by valgrind
478 * too, in most cases).
479 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
480 * a hope that sizeof(long) won't become >8 any time soon.
482 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
483 /*HOST_LONG_BITS*/ 64) / 8;
484 if (!d.dirty_bitmap) {
485 d.dirty_bitmap = g_malloc(size);
486 } else if (size > allocated_size) {
487 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
489 allocated_size = size;
490 memset(d.dirty_bitmap, 0, allocated_size);
492 d.slot = mem->slot | (kml->as_id << 16);
493 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
494 DPRINTF("ioctl failed %d\n", errno);
495 ret = -1;
496 break;
499 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
500 start_addr = mem->start_addr + mem->memory_size;
502 g_free(d.dirty_bitmap);
504 return ret;
507 static void kvm_coalesce_mmio_region(MemoryListener *listener,
508 MemoryRegionSection *secion,
509 hwaddr start, hwaddr size)
511 KVMState *s = kvm_state;
513 if (s->coalesced_mmio) {
514 struct kvm_coalesced_mmio_zone zone;
516 zone.addr = start;
517 zone.size = size;
518 zone.pad = 0;
520 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
524 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
525 MemoryRegionSection *secion,
526 hwaddr start, hwaddr size)
528 KVMState *s = kvm_state;
530 if (s->coalesced_mmio) {
531 struct kvm_coalesced_mmio_zone zone;
533 zone.addr = start;
534 zone.size = size;
535 zone.pad = 0;
537 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
541 int kvm_check_extension(KVMState *s, unsigned int extension)
543 int ret;
545 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
546 if (ret < 0) {
547 ret = 0;
550 return ret;
553 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
555 int ret;
557 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
558 if (ret < 0) {
559 /* VM wide version not implemented, use global one instead */
560 ret = kvm_check_extension(s, extension);
563 return ret;
566 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
568 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
569 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
570 * endianness, but the memory core hands them in target endianness.
571 * For example, PPC is always treated as big-endian even if running
572 * on KVM and on PPC64LE. Correct here.
574 switch (size) {
575 case 2:
576 val = bswap16(val);
577 break;
578 case 4:
579 val = bswap32(val);
580 break;
582 #endif
583 return val;
586 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
587 bool assign, uint32_t size, bool datamatch)
589 int ret;
590 struct kvm_ioeventfd iofd = {
591 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
592 .addr = addr,
593 .len = size,
594 .flags = 0,
595 .fd = fd,
598 if (!kvm_enabled()) {
599 return -ENOSYS;
602 if (datamatch) {
603 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
605 if (!assign) {
606 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
609 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
611 if (ret < 0) {
612 return -errno;
615 return 0;
618 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
619 bool assign, uint32_t size, bool datamatch)
621 struct kvm_ioeventfd kick = {
622 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
623 .addr = addr,
624 .flags = KVM_IOEVENTFD_FLAG_PIO,
625 .len = size,
626 .fd = fd,
628 int r;
629 if (!kvm_enabled()) {
630 return -ENOSYS;
632 if (datamatch) {
633 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
635 if (!assign) {
636 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
638 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
639 if (r < 0) {
640 return r;
642 return 0;
646 static int kvm_check_many_ioeventfds(void)
648 /* Userspace can use ioeventfd for io notification. This requires a host
649 * that supports eventfd(2) and an I/O thread; since eventfd does not
650 * support SIGIO it cannot interrupt the vcpu.
652 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
653 * can avoid creating too many ioeventfds.
655 #if defined(CONFIG_EVENTFD)
656 int ioeventfds[7];
657 int i, ret = 0;
658 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
659 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
660 if (ioeventfds[i] < 0) {
661 break;
663 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
664 if (ret < 0) {
665 close(ioeventfds[i]);
666 break;
670 /* Decide whether many devices are supported or not */
671 ret = i == ARRAY_SIZE(ioeventfds);
673 while (i-- > 0) {
674 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
675 close(ioeventfds[i]);
677 return ret;
678 #else
679 return 0;
680 #endif
683 static const KVMCapabilityInfo *
684 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
686 while (list->name) {
687 if (!kvm_check_extension(s, list->value)) {
688 return list;
690 list++;
692 return NULL;
695 static void kvm_set_phys_mem(KVMMemoryListener *kml,
696 MemoryRegionSection *section, bool add)
698 KVMState *s = kvm_state;
699 KVMSlot *mem, old;
700 int err;
701 MemoryRegion *mr = section->mr;
702 bool writeable = !mr->readonly && !mr->rom_device;
703 hwaddr start_addr = section->offset_within_address_space;
704 ram_addr_t size = int128_get64(section->size);
705 void *ram = NULL;
706 unsigned delta;
708 /* kvm works in page size chunks, but the function may be called
709 with sub-page size and unaligned start address. Pad the start
710 address to next and truncate size to previous page boundary. */
711 delta = qemu_real_host_page_size - (start_addr & ~qemu_real_host_page_mask);
712 delta &= ~qemu_real_host_page_mask;
713 if (delta > size) {
714 return;
716 start_addr += delta;
717 size -= delta;
718 size &= qemu_real_host_page_mask;
719 if (!size || (start_addr & ~qemu_real_host_page_mask)) {
720 return;
723 if (!memory_region_is_ram(mr)) {
724 if (writeable || !kvm_readonly_mem_allowed) {
725 return;
726 } else if (!mr->romd_mode) {
727 /* If the memory device is not in romd_mode, then we actually want
728 * to remove the kvm memory slot so all accesses will trap. */
729 add = false;
733 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
735 while (1) {
736 mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size);
737 if (!mem) {
738 break;
741 if (add && start_addr >= mem->start_addr &&
742 (start_addr + size <= mem->start_addr + mem->memory_size) &&
743 (ram - start_addr == mem->ram - mem->start_addr)) {
744 /* The new slot fits into the existing one and comes with
745 * identical parameters - update flags and done. */
746 kvm_slot_update_flags(kml, mem, mr);
747 return;
750 old = *mem;
752 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
753 kvm_physical_sync_dirty_bitmap(kml, section);
756 /* unregister the overlapping slot */
757 mem->memory_size = 0;
758 err = kvm_set_user_memory_region(kml, mem);
759 if (err) {
760 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
761 __func__, strerror(-err));
762 abort();
765 /* Workaround for older KVM versions: we can't join slots, even not by
766 * unregistering the previous ones and then registering the larger
767 * slot. We have to maintain the existing fragmentation. Sigh.
769 * This workaround assumes that the new slot starts at the same
770 * address as the first existing one. If not or if some overlapping
771 * slot comes around later, we will fail (not seen in practice so far)
772 * - and actually require a recent KVM version. */
773 if (s->broken_set_mem_region &&
774 old.start_addr == start_addr && old.memory_size < size && add) {
775 mem = kvm_alloc_slot(kml);
776 mem->memory_size = old.memory_size;
777 mem->start_addr = old.start_addr;
778 mem->ram = old.ram;
779 mem->flags = kvm_mem_flags(mr);
781 err = kvm_set_user_memory_region(kml, mem);
782 if (err) {
783 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
784 strerror(-err));
785 abort();
788 start_addr += old.memory_size;
789 ram += old.memory_size;
790 size -= old.memory_size;
791 continue;
794 /* register prefix slot */
795 if (old.start_addr < start_addr) {
796 mem = kvm_alloc_slot(kml);
797 mem->memory_size = start_addr - old.start_addr;
798 mem->start_addr = old.start_addr;
799 mem->ram = old.ram;
800 mem->flags = kvm_mem_flags(mr);
802 err = kvm_set_user_memory_region(kml, mem);
803 if (err) {
804 fprintf(stderr, "%s: error registering prefix slot: %s\n",
805 __func__, strerror(-err));
806 #ifdef TARGET_PPC
807 fprintf(stderr, "%s: This is probably because your kernel's " \
808 "PAGE_SIZE is too big. Please try to use 4k " \
809 "PAGE_SIZE!\n", __func__);
810 #endif
811 abort();
815 /* register suffix slot */
816 if (old.start_addr + old.memory_size > start_addr + size) {
817 ram_addr_t size_delta;
819 mem = kvm_alloc_slot(kml);
820 mem->start_addr = start_addr + size;
821 size_delta = mem->start_addr - old.start_addr;
822 mem->memory_size = old.memory_size - size_delta;
823 mem->ram = old.ram + size_delta;
824 mem->flags = kvm_mem_flags(mr);
826 err = kvm_set_user_memory_region(kml, mem);
827 if (err) {
828 fprintf(stderr, "%s: error registering suffix slot: %s\n",
829 __func__, strerror(-err));
830 abort();
835 /* in case the KVM bug workaround already "consumed" the new slot */
836 if (!size) {
837 return;
839 if (!add) {
840 return;
842 mem = kvm_alloc_slot(kml);
843 mem->memory_size = size;
844 mem->start_addr = start_addr;
845 mem->ram = ram;
846 mem->flags = kvm_mem_flags(mr);
848 err = kvm_set_user_memory_region(kml, mem);
849 if (err) {
850 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
851 strerror(-err));
852 abort();
856 static void kvm_region_add(MemoryListener *listener,
857 MemoryRegionSection *section)
859 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
861 memory_region_ref(section->mr);
862 kvm_set_phys_mem(kml, section, true);
865 static void kvm_region_del(MemoryListener *listener,
866 MemoryRegionSection *section)
868 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
870 kvm_set_phys_mem(kml, section, false);
871 memory_region_unref(section->mr);
874 static void kvm_log_sync(MemoryListener *listener,
875 MemoryRegionSection *section)
877 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
878 int r;
880 r = kvm_physical_sync_dirty_bitmap(kml, section);
881 if (r < 0) {
882 abort();
886 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
887 MemoryRegionSection *section,
888 bool match_data, uint64_t data,
889 EventNotifier *e)
891 int fd = event_notifier_get_fd(e);
892 int r;
894 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
895 data, true, int128_get64(section->size),
896 match_data);
897 if (r < 0) {
898 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
899 __func__, strerror(-r));
900 abort();
904 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
905 MemoryRegionSection *section,
906 bool match_data, uint64_t data,
907 EventNotifier *e)
909 int fd = event_notifier_get_fd(e);
910 int r;
912 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
913 data, false, int128_get64(section->size),
914 match_data);
915 if (r < 0) {
916 abort();
920 static void kvm_io_ioeventfd_add(MemoryListener *listener,
921 MemoryRegionSection *section,
922 bool match_data, uint64_t data,
923 EventNotifier *e)
925 int fd = event_notifier_get_fd(e);
926 int r;
928 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
929 data, true, int128_get64(section->size),
930 match_data);
931 if (r < 0) {
932 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
933 __func__, strerror(-r));
934 abort();
938 static void kvm_io_ioeventfd_del(MemoryListener *listener,
939 MemoryRegionSection *section,
940 bool match_data, uint64_t data,
941 EventNotifier *e)
944 int fd = event_notifier_get_fd(e);
945 int r;
947 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
948 data, false, int128_get64(section->size),
949 match_data);
950 if (r < 0) {
951 abort();
955 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
956 AddressSpace *as, int as_id)
958 int i;
960 kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
961 kml->as_id = as_id;
963 for (i = 0; i < s->nr_slots; i++) {
964 kml->slots[i].slot = i;
967 kml->listener.region_add = kvm_region_add;
968 kml->listener.region_del = kvm_region_del;
969 kml->listener.log_start = kvm_log_start;
970 kml->listener.log_stop = kvm_log_stop;
971 kml->listener.log_sync = kvm_log_sync;
972 kml->listener.priority = 10;
974 memory_listener_register(&kml->listener, as);
977 static MemoryListener kvm_io_listener = {
978 .eventfd_add = kvm_io_ioeventfd_add,
979 .eventfd_del = kvm_io_ioeventfd_del,
980 .priority = 10,
983 static void kvm_handle_interrupt(CPUState *cpu, int mask)
985 cpu->interrupt_request |= mask;
987 if (!qemu_cpu_is_self(cpu)) {
988 qemu_cpu_kick(cpu);
992 int kvm_set_irq(KVMState *s, int irq, int level)
994 struct kvm_irq_level event;
995 int ret;
997 assert(kvm_async_interrupts_enabled());
999 event.level = level;
1000 event.irq = irq;
1001 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
1002 if (ret < 0) {
1003 perror("kvm_set_irq");
1004 abort();
1007 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
1010 #ifdef KVM_CAP_IRQ_ROUTING
1011 typedef struct KVMMSIRoute {
1012 struct kvm_irq_routing_entry kroute;
1013 QTAILQ_ENTRY(KVMMSIRoute) entry;
1014 } KVMMSIRoute;
1016 static void set_gsi(KVMState *s, unsigned int gsi)
1018 set_bit(gsi, s->used_gsi_bitmap);
1021 static void clear_gsi(KVMState *s, unsigned int gsi)
1023 clear_bit(gsi, s->used_gsi_bitmap);
1026 void kvm_init_irq_routing(KVMState *s)
1028 int gsi_count, i;
1030 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
1031 if (gsi_count > 0) {
1032 /* Round up so we can search ints using ffs */
1033 s->used_gsi_bitmap = bitmap_new(gsi_count);
1034 s->gsi_count = gsi_count;
1037 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
1038 s->nr_allocated_irq_routes = 0;
1040 if (!kvm_direct_msi_allowed) {
1041 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
1042 QTAILQ_INIT(&s->msi_hashtab[i]);
1046 kvm_arch_init_irq_routing(s);
1049 void kvm_irqchip_commit_routes(KVMState *s)
1051 int ret;
1053 if (kvm_gsi_direct_mapping()) {
1054 return;
1057 if (!kvm_gsi_routing_enabled()) {
1058 return;
1061 s->irq_routes->flags = 0;
1062 trace_kvm_irqchip_commit_routes();
1063 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
1064 assert(ret == 0);
1067 static void kvm_add_routing_entry(KVMState *s,
1068 struct kvm_irq_routing_entry *entry)
1070 struct kvm_irq_routing_entry *new;
1071 int n, size;
1073 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1074 n = s->nr_allocated_irq_routes * 2;
1075 if (n < 64) {
1076 n = 64;
1078 size = sizeof(struct kvm_irq_routing);
1079 size += n * sizeof(*new);
1080 s->irq_routes = g_realloc(s->irq_routes, size);
1081 s->nr_allocated_irq_routes = n;
1083 n = s->irq_routes->nr++;
1084 new = &s->irq_routes->entries[n];
1086 *new = *entry;
1088 set_gsi(s, entry->gsi);
1091 static int kvm_update_routing_entry(KVMState *s,
1092 struct kvm_irq_routing_entry *new_entry)
1094 struct kvm_irq_routing_entry *entry;
1095 int n;
1097 for (n = 0; n < s->irq_routes->nr; n++) {
1098 entry = &s->irq_routes->entries[n];
1099 if (entry->gsi != new_entry->gsi) {
1100 continue;
1103 if(!memcmp(entry, new_entry, sizeof *entry)) {
1104 return 0;
1107 *entry = *new_entry;
1109 return 0;
1112 return -ESRCH;
1115 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1117 struct kvm_irq_routing_entry e = {};
1119 assert(pin < s->gsi_count);
1121 e.gsi = irq;
1122 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1123 e.flags = 0;
1124 e.u.irqchip.irqchip = irqchip;
1125 e.u.irqchip.pin = pin;
1126 kvm_add_routing_entry(s, &e);
1129 void kvm_irqchip_release_virq(KVMState *s, int virq)
1131 struct kvm_irq_routing_entry *e;
1132 int i;
1134 if (kvm_gsi_direct_mapping()) {
1135 return;
1138 for (i = 0; i < s->irq_routes->nr; i++) {
1139 e = &s->irq_routes->entries[i];
1140 if (e->gsi == virq) {
1141 s->irq_routes->nr--;
1142 *e = s->irq_routes->entries[s->irq_routes->nr];
1145 clear_gsi(s, virq);
1146 kvm_arch_release_virq_post(virq);
1147 trace_kvm_irqchip_release_virq(virq);
1150 static unsigned int kvm_hash_msi(uint32_t data)
1152 /* This is optimized for IA32 MSI layout. However, no other arch shall
1153 * repeat the mistake of not providing a direct MSI injection API. */
1154 return data & 0xff;
1157 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1159 KVMMSIRoute *route, *next;
1160 unsigned int hash;
1162 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1163 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1164 kvm_irqchip_release_virq(s, route->kroute.gsi);
1165 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1166 g_free(route);
1171 static int kvm_irqchip_get_virq(KVMState *s)
1173 int next_virq;
1176 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1177 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1178 * number can succeed even though a new route entry cannot be added.
1179 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1181 if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) {
1182 kvm_flush_dynamic_msi_routes(s);
1185 /* Return the lowest unused GSI in the bitmap */
1186 next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count);
1187 if (next_virq >= s->gsi_count) {
1188 return -ENOSPC;
1189 } else {
1190 return next_virq;
1194 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1196 unsigned int hash = kvm_hash_msi(msg.data);
1197 KVMMSIRoute *route;
1199 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1200 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1201 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1202 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1203 return route;
1206 return NULL;
1209 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1211 struct kvm_msi msi;
1212 KVMMSIRoute *route;
1214 if (kvm_direct_msi_allowed) {
1215 msi.address_lo = (uint32_t)msg.address;
1216 msi.address_hi = msg.address >> 32;
1217 msi.data = le32_to_cpu(msg.data);
1218 msi.flags = 0;
1219 memset(msi.pad, 0, sizeof(msi.pad));
1221 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1224 route = kvm_lookup_msi_route(s, msg);
1225 if (!route) {
1226 int virq;
1228 virq = kvm_irqchip_get_virq(s);
1229 if (virq < 0) {
1230 return virq;
1233 route = g_malloc0(sizeof(KVMMSIRoute));
1234 route->kroute.gsi = virq;
1235 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1236 route->kroute.flags = 0;
1237 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1238 route->kroute.u.msi.address_hi = msg.address >> 32;
1239 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1241 kvm_add_routing_entry(s, &route->kroute);
1242 kvm_irqchip_commit_routes(s);
1244 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1245 entry);
1248 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1250 return kvm_set_irq(s, route->kroute.gsi, 1);
1253 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev)
1255 struct kvm_irq_routing_entry kroute = {};
1256 int virq;
1257 MSIMessage msg = {0, 0};
1259 if (dev) {
1260 msg = pci_get_msi_message(dev, vector);
1263 if (kvm_gsi_direct_mapping()) {
1264 return kvm_arch_msi_data_to_gsi(msg.data);
1267 if (!kvm_gsi_routing_enabled()) {
1268 return -ENOSYS;
1271 virq = kvm_irqchip_get_virq(s);
1272 if (virq < 0) {
1273 return virq;
1276 kroute.gsi = virq;
1277 kroute.type = KVM_IRQ_ROUTING_MSI;
1278 kroute.flags = 0;
1279 kroute.u.msi.address_lo = (uint32_t)msg.address;
1280 kroute.u.msi.address_hi = msg.address >> 32;
1281 kroute.u.msi.data = le32_to_cpu(msg.data);
1282 if (kvm_msi_devid_required()) {
1283 kroute.flags = KVM_MSI_VALID_DEVID;
1284 kroute.u.msi.devid = pci_requester_id(dev);
1286 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1287 kvm_irqchip_release_virq(s, virq);
1288 return -EINVAL;
1291 trace_kvm_irqchip_add_msi_route(dev ? dev->name : (char *)"N/A",
1292 vector, virq);
1294 kvm_add_routing_entry(s, &kroute);
1295 kvm_arch_add_msi_route_post(&kroute, vector, dev);
1296 kvm_irqchip_commit_routes(s);
1298 return virq;
1301 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
1302 PCIDevice *dev)
1304 struct kvm_irq_routing_entry kroute = {};
1306 if (kvm_gsi_direct_mapping()) {
1307 return 0;
1310 if (!kvm_irqchip_in_kernel()) {
1311 return -ENOSYS;
1314 kroute.gsi = virq;
1315 kroute.type = KVM_IRQ_ROUTING_MSI;
1316 kroute.flags = 0;
1317 kroute.u.msi.address_lo = (uint32_t)msg.address;
1318 kroute.u.msi.address_hi = msg.address >> 32;
1319 kroute.u.msi.data = le32_to_cpu(msg.data);
1320 if (kvm_msi_devid_required()) {
1321 kroute.flags = KVM_MSI_VALID_DEVID;
1322 kroute.u.msi.devid = pci_requester_id(dev);
1324 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1325 return -EINVAL;
1328 trace_kvm_irqchip_update_msi_route(virq);
1330 return kvm_update_routing_entry(s, &kroute);
1333 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1334 bool assign)
1336 struct kvm_irqfd irqfd = {
1337 .fd = fd,
1338 .gsi = virq,
1339 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1342 if (rfd != -1) {
1343 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1344 irqfd.resamplefd = rfd;
1347 if (!kvm_irqfds_enabled()) {
1348 return -ENOSYS;
1351 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1354 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1356 struct kvm_irq_routing_entry kroute = {};
1357 int virq;
1359 if (!kvm_gsi_routing_enabled()) {
1360 return -ENOSYS;
1363 virq = kvm_irqchip_get_virq(s);
1364 if (virq < 0) {
1365 return virq;
1368 kroute.gsi = virq;
1369 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1370 kroute.flags = 0;
1371 kroute.u.adapter.summary_addr = adapter->summary_addr;
1372 kroute.u.adapter.ind_addr = adapter->ind_addr;
1373 kroute.u.adapter.summary_offset = adapter->summary_offset;
1374 kroute.u.adapter.ind_offset = adapter->ind_offset;
1375 kroute.u.adapter.adapter_id = adapter->adapter_id;
1377 kvm_add_routing_entry(s, &kroute);
1379 return virq;
1382 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1384 struct kvm_irq_routing_entry kroute = {};
1385 int virq;
1387 if (!kvm_gsi_routing_enabled()) {
1388 return -ENOSYS;
1390 if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) {
1391 return -ENOSYS;
1393 virq = kvm_irqchip_get_virq(s);
1394 if (virq < 0) {
1395 return virq;
1398 kroute.gsi = virq;
1399 kroute.type = KVM_IRQ_ROUTING_HV_SINT;
1400 kroute.flags = 0;
1401 kroute.u.hv_sint.vcpu = vcpu;
1402 kroute.u.hv_sint.sint = sint;
1404 kvm_add_routing_entry(s, &kroute);
1405 kvm_irqchip_commit_routes(s);
1407 return virq;
1410 #else /* !KVM_CAP_IRQ_ROUTING */
1412 void kvm_init_irq_routing(KVMState *s)
1416 void kvm_irqchip_release_virq(KVMState *s, int virq)
1420 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1422 abort();
1425 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev)
1427 return -ENOSYS;
1430 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1432 return -ENOSYS;
1435 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1437 return -ENOSYS;
1440 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1442 abort();
1445 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1447 return -ENOSYS;
1449 #endif /* !KVM_CAP_IRQ_ROUTING */
1451 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1452 EventNotifier *rn, int virq)
1454 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1455 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1458 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1459 int virq)
1461 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1462 false);
1465 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1466 EventNotifier *rn, qemu_irq irq)
1468 gpointer key, gsi;
1469 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1471 if (!found) {
1472 return -ENXIO;
1474 return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi));
1477 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
1478 qemu_irq irq)
1480 gpointer key, gsi;
1481 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1483 if (!found) {
1484 return -ENXIO;
1486 return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi));
1489 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
1491 g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
1494 static void kvm_irqchip_create(MachineState *machine, KVMState *s)
1496 int ret;
1498 if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1500 } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) {
1501 ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0);
1502 if (ret < 0) {
1503 fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret));
1504 exit(1);
1506 } else {
1507 return;
1510 /* First probe and see if there's a arch-specific hook to create the
1511 * in-kernel irqchip for us */
1512 ret = kvm_arch_irqchip_create(machine, s);
1513 if (ret == 0) {
1514 if (machine_kernel_irqchip_split(machine)) {
1515 perror("Split IRQ chip mode not supported.");
1516 exit(1);
1517 } else {
1518 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1521 if (ret < 0) {
1522 fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
1523 exit(1);
1526 kvm_kernel_irqchip = true;
1527 /* If we have an in-kernel IRQ chip then we must have asynchronous
1528 * interrupt delivery (though the reverse is not necessarily true)
1530 kvm_async_interrupts_allowed = true;
1531 kvm_halt_in_kernel_allowed = true;
1533 kvm_init_irq_routing(s);
1535 s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal);
1538 /* Find number of supported CPUs using the recommended
1539 * procedure from the kernel API documentation to cope with
1540 * older kernels that may be missing capabilities.
1542 static int kvm_recommended_vcpus(KVMState *s)
1544 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1545 return (ret) ? ret : 4;
1548 static int kvm_max_vcpus(KVMState *s)
1550 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1551 return (ret) ? ret : kvm_recommended_vcpus(s);
1554 static int kvm_max_vcpu_id(KVMState *s)
1556 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID);
1557 return (ret) ? ret : kvm_max_vcpus(s);
1560 bool kvm_vcpu_id_is_valid(int vcpu_id)
1562 KVMState *s = KVM_STATE(current_machine->accelerator);
1563 return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s);
1566 static int kvm_init(MachineState *ms)
1568 MachineClass *mc = MACHINE_GET_CLASS(ms);
1569 static const char upgrade_note[] =
1570 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1571 "(see http://sourceforge.net/projects/kvm).\n";
1572 struct {
1573 const char *name;
1574 int num;
1575 } num_cpus[] = {
1576 { "SMP", smp_cpus },
1577 { "hotpluggable", max_cpus },
1578 { NULL, }
1579 }, *nc = num_cpus;
1580 int soft_vcpus_limit, hard_vcpus_limit;
1581 KVMState *s;
1582 const KVMCapabilityInfo *missing_cap;
1583 int ret;
1584 int type = 0;
1585 const char *kvm_type;
1587 s = KVM_STATE(ms->accelerator);
1590 * On systems where the kernel can support different base page
1591 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1592 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1593 * page size for the system though.
1595 assert(TARGET_PAGE_SIZE <= getpagesize());
1597 s->sigmask_len = 8;
1599 #ifdef KVM_CAP_SET_GUEST_DEBUG
1600 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1601 #endif
1602 QLIST_INIT(&s->kvm_parked_vcpus);
1603 s->vmfd = -1;
1604 s->fd = qemu_open("/dev/kvm", O_RDWR);
1605 if (s->fd == -1) {
1606 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1607 ret = -errno;
1608 goto err;
1611 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1612 if (ret < KVM_API_VERSION) {
1613 if (ret >= 0) {
1614 ret = -EINVAL;
1616 fprintf(stderr, "kvm version too old\n");
1617 goto err;
1620 if (ret > KVM_API_VERSION) {
1621 ret = -EINVAL;
1622 fprintf(stderr, "kvm version not supported\n");
1623 goto err;
1626 kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT);
1627 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1629 /* If unspecified, use the default value */
1630 if (!s->nr_slots) {
1631 s->nr_slots = 32;
1634 /* check the vcpu limits */
1635 soft_vcpus_limit = kvm_recommended_vcpus(s);
1636 hard_vcpus_limit = kvm_max_vcpus(s);
1638 while (nc->name) {
1639 if (nc->num > soft_vcpus_limit) {
1640 fprintf(stderr,
1641 "Warning: Number of %s cpus requested (%d) exceeds "
1642 "the recommended cpus supported by KVM (%d)\n",
1643 nc->name, nc->num, soft_vcpus_limit);
1645 if (nc->num > hard_vcpus_limit) {
1646 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1647 "the maximum cpus supported by KVM (%d)\n",
1648 nc->name, nc->num, hard_vcpus_limit);
1649 exit(1);
1652 nc++;
1655 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1656 if (mc->kvm_type) {
1657 type = mc->kvm_type(kvm_type);
1658 } else if (kvm_type) {
1659 ret = -EINVAL;
1660 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1661 goto err;
1664 do {
1665 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1666 } while (ret == -EINTR);
1668 if (ret < 0) {
1669 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1670 strerror(-ret));
1672 #ifdef TARGET_S390X
1673 if (ret == -EINVAL) {
1674 fprintf(stderr,
1675 "Host kernel setup problem detected. Please verify:\n");
1676 fprintf(stderr, "- for kernels supporting the switch_amode or"
1677 " user_mode parameters, whether\n");
1678 fprintf(stderr,
1679 " user space is running in primary address space\n");
1680 fprintf(stderr,
1681 "- for kernels supporting the vm.allocate_pgste sysctl, "
1682 "whether it is enabled\n");
1684 #endif
1685 goto err;
1688 s->vmfd = ret;
1689 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1690 if (!missing_cap) {
1691 missing_cap =
1692 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1694 if (missing_cap) {
1695 ret = -EINVAL;
1696 fprintf(stderr, "kvm does not support %s\n%s",
1697 missing_cap->name, upgrade_note);
1698 goto err;
1701 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1703 s->broken_set_mem_region = 1;
1704 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1705 if (ret > 0) {
1706 s->broken_set_mem_region = 0;
1709 #ifdef KVM_CAP_VCPU_EVENTS
1710 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1711 #endif
1713 s->robust_singlestep =
1714 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1716 #ifdef KVM_CAP_DEBUGREGS
1717 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1718 #endif
1720 #ifdef KVM_CAP_IRQ_ROUTING
1721 kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1722 #endif
1724 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1726 s->irq_set_ioctl = KVM_IRQ_LINE;
1727 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1728 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1731 #ifdef KVM_CAP_READONLY_MEM
1732 kvm_readonly_mem_allowed =
1733 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1734 #endif
1736 kvm_eventfds_allowed =
1737 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1739 kvm_irqfds_allowed =
1740 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1742 kvm_resamplefds_allowed =
1743 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1745 kvm_vm_attributes_allowed =
1746 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1748 kvm_ioeventfd_any_length_allowed =
1749 (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0);
1751 kvm_state = s;
1753 ret = kvm_arch_init(ms, s);
1754 if (ret < 0) {
1755 goto err;
1758 if (machine_kernel_irqchip_allowed(ms)) {
1759 kvm_irqchip_create(ms, s);
1762 if (kvm_eventfds_allowed) {
1763 s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
1764 s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
1766 s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
1767 s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
1769 kvm_memory_listener_register(s, &s->memory_listener,
1770 &address_space_memory, 0);
1771 memory_listener_register(&kvm_io_listener,
1772 &address_space_io);
1774 s->many_ioeventfds = kvm_check_many_ioeventfds();
1776 cpu_interrupt_handler = kvm_handle_interrupt;
1778 return 0;
1780 err:
1781 assert(ret < 0);
1782 if (s->vmfd >= 0) {
1783 close(s->vmfd);
1785 if (s->fd != -1) {
1786 close(s->fd);
1788 g_free(s->memory_listener.slots);
1790 return ret;
1793 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1795 s->sigmask_len = sigmask_len;
1798 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1799 int size, uint32_t count)
1801 int i;
1802 uint8_t *ptr = data;
1804 for (i = 0; i < count; i++) {
1805 address_space_rw(&address_space_io, port, attrs,
1806 ptr, size,
1807 direction == KVM_EXIT_IO_OUT);
1808 ptr += size;
1812 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1814 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1815 run->internal.suberror);
1817 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1818 int i;
1820 for (i = 0; i < run->internal.ndata; ++i) {
1821 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1822 i, (uint64_t)run->internal.data[i]);
1825 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1826 fprintf(stderr, "emulation failure\n");
1827 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1828 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1829 return EXCP_INTERRUPT;
1832 /* FIXME: Should trigger a qmp message to let management know
1833 * something went wrong.
1835 return -1;
1838 void kvm_flush_coalesced_mmio_buffer(void)
1840 KVMState *s = kvm_state;
1842 if (s->coalesced_flush_in_progress) {
1843 return;
1846 s->coalesced_flush_in_progress = true;
1848 if (s->coalesced_mmio_ring) {
1849 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1850 while (ring->first != ring->last) {
1851 struct kvm_coalesced_mmio *ent;
1853 ent = &ring->coalesced_mmio[ring->first];
1855 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1856 smp_wmb();
1857 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1861 s->coalesced_flush_in_progress = false;
1864 static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
1866 if (!cpu->kvm_vcpu_dirty) {
1867 kvm_arch_get_registers(cpu);
1868 cpu->kvm_vcpu_dirty = true;
1872 void kvm_cpu_synchronize_state(CPUState *cpu)
1874 if (!cpu->kvm_vcpu_dirty) {
1875 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL);
1879 static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
1881 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1882 cpu->kvm_vcpu_dirty = false;
1885 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1887 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
1890 static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
1892 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1893 cpu->kvm_vcpu_dirty = false;
1896 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1898 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
1901 static void do_kvm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg)
1903 cpu->kvm_vcpu_dirty = true;
1906 void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu)
1908 run_on_cpu(cpu, do_kvm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL);
1911 #ifdef KVM_HAVE_MCE_INJECTION
1912 static __thread void *pending_sigbus_addr;
1913 static __thread int pending_sigbus_code;
1914 static __thread bool have_sigbus_pending;
1915 #endif
1917 static void kvm_cpu_kick(CPUState *cpu)
1919 atomic_set(&cpu->kvm_run->immediate_exit, 1);
1922 static void kvm_cpu_kick_self(void)
1924 if (kvm_immediate_exit) {
1925 kvm_cpu_kick(current_cpu);
1926 } else {
1927 qemu_cpu_kick_self();
1931 static void kvm_eat_signals(CPUState *cpu)
1933 struct timespec ts = { 0, 0 };
1934 siginfo_t siginfo;
1935 sigset_t waitset;
1936 sigset_t chkset;
1937 int r;
1939 if (kvm_immediate_exit) {
1940 atomic_set(&cpu->kvm_run->immediate_exit, 0);
1941 /* Write kvm_run->immediate_exit before the cpu->exit_request
1942 * write in kvm_cpu_exec.
1944 smp_wmb();
1945 return;
1948 sigemptyset(&waitset);
1949 sigaddset(&waitset, SIG_IPI);
1951 do {
1952 r = sigtimedwait(&waitset, &siginfo, &ts);
1953 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
1954 perror("sigtimedwait");
1955 exit(1);
1958 r = sigpending(&chkset);
1959 if (r == -1) {
1960 perror("sigpending");
1961 exit(1);
1963 } while (sigismember(&chkset, SIG_IPI));
1966 int kvm_cpu_exec(CPUState *cpu)
1968 struct kvm_run *run = cpu->kvm_run;
1969 int ret, run_ret;
1971 DPRINTF("kvm_cpu_exec()\n");
1973 if (kvm_arch_process_async_events(cpu)) {
1974 atomic_set(&cpu->exit_request, 0);
1975 return EXCP_HLT;
1978 qemu_mutex_unlock_iothread();
1980 do {
1981 MemTxAttrs attrs;
1983 if (cpu->kvm_vcpu_dirty) {
1984 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1985 cpu->kvm_vcpu_dirty = false;
1988 kvm_arch_pre_run(cpu, run);
1989 if (atomic_read(&cpu->exit_request)) {
1990 DPRINTF("interrupt exit requested\n");
1992 * KVM requires us to reenter the kernel after IO exits to complete
1993 * instruction emulation. This self-signal will ensure that we
1994 * leave ASAP again.
1996 kvm_cpu_kick_self();
1999 /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
2000 * Matching barrier in kvm_eat_signals.
2002 smp_rmb();
2004 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
2006 attrs = kvm_arch_post_run(cpu, run);
2008 #ifdef KVM_HAVE_MCE_INJECTION
2009 if (unlikely(have_sigbus_pending)) {
2010 qemu_mutex_lock_iothread();
2011 kvm_arch_on_sigbus_vcpu(cpu, pending_sigbus_code,
2012 pending_sigbus_addr);
2013 have_sigbus_pending = false;
2014 qemu_mutex_unlock_iothread();
2016 #endif
2018 if (run_ret < 0) {
2019 if (run_ret == -EINTR || run_ret == -EAGAIN) {
2020 DPRINTF("io window exit\n");
2021 kvm_eat_signals(cpu);
2022 ret = EXCP_INTERRUPT;
2023 break;
2025 fprintf(stderr, "error: kvm run failed %s\n",
2026 strerror(-run_ret));
2027 #ifdef TARGET_PPC
2028 if (run_ret == -EBUSY) {
2029 fprintf(stderr,
2030 "This is probably because your SMT is enabled.\n"
2031 "VCPU can only run on primary threads with all "
2032 "secondary threads offline.\n");
2034 #endif
2035 ret = -1;
2036 break;
2039 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
2040 switch (run->exit_reason) {
2041 case KVM_EXIT_IO:
2042 DPRINTF("handle_io\n");
2043 /* Called outside BQL */
2044 kvm_handle_io(run->io.port, attrs,
2045 (uint8_t *)run + run->io.data_offset,
2046 run->io.direction,
2047 run->io.size,
2048 run->io.count);
2049 ret = 0;
2050 break;
2051 case KVM_EXIT_MMIO:
2052 DPRINTF("handle_mmio\n");
2053 /* Called outside BQL */
2054 address_space_rw(&address_space_memory,
2055 run->mmio.phys_addr, attrs,
2056 run->mmio.data,
2057 run->mmio.len,
2058 run->mmio.is_write);
2059 ret = 0;
2060 break;
2061 case KVM_EXIT_IRQ_WINDOW_OPEN:
2062 DPRINTF("irq_window_open\n");
2063 ret = EXCP_INTERRUPT;
2064 break;
2065 case KVM_EXIT_SHUTDOWN:
2066 DPRINTF("shutdown\n");
2067 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
2068 ret = EXCP_INTERRUPT;
2069 break;
2070 case KVM_EXIT_UNKNOWN:
2071 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
2072 (uint64_t)run->hw.hardware_exit_reason);
2073 ret = -1;
2074 break;
2075 case KVM_EXIT_INTERNAL_ERROR:
2076 ret = kvm_handle_internal_error(cpu, run);
2077 break;
2078 case KVM_EXIT_SYSTEM_EVENT:
2079 switch (run->system_event.type) {
2080 case KVM_SYSTEM_EVENT_SHUTDOWN:
2081 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN);
2082 ret = EXCP_INTERRUPT;
2083 break;
2084 case KVM_SYSTEM_EVENT_RESET:
2085 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
2086 ret = EXCP_INTERRUPT;
2087 break;
2088 case KVM_SYSTEM_EVENT_CRASH:
2089 kvm_cpu_synchronize_state(cpu);
2090 qemu_mutex_lock_iothread();
2091 qemu_system_guest_panicked(cpu_get_crash_info(cpu));
2092 qemu_mutex_unlock_iothread();
2093 ret = 0;
2094 break;
2095 default:
2096 DPRINTF("kvm_arch_handle_exit\n");
2097 ret = kvm_arch_handle_exit(cpu, run);
2098 break;
2100 break;
2101 default:
2102 DPRINTF("kvm_arch_handle_exit\n");
2103 ret = kvm_arch_handle_exit(cpu, run);
2104 break;
2106 } while (ret == 0);
2108 qemu_mutex_lock_iothread();
2110 if (ret < 0) {
2111 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
2112 vm_stop(RUN_STATE_INTERNAL_ERROR);
2115 atomic_set(&cpu->exit_request, 0);
2116 return ret;
2119 int kvm_ioctl(KVMState *s, int type, ...)
2121 int ret;
2122 void *arg;
2123 va_list ap;
2125 va_start(ap, type);
2126 arg = va_arg(ap, void *);
2127 va_end(ap);
2129 trace_kvm_ioctl(type, arg);
2130 ret = ioctl(s->fd, type, arg);
2131 if (ret == -1) {
2132 ret = -errno;
2134 return ret;
2137 int kvm_vm_ioctl(KVMState *s, int type, ...)
2139 int ret;
2140 void *arg;
2141 va_list ap;
2143 va_start(ap, type);
2144 arg = va_arg(ap, void *);
2145 va_end(ap);
2147 trace_kvm_vm_ioctl(type, arg);
2148 ret = ioctl(s->vmfd, type, arg);
2149 if (ret == -1) {
2150 ret = -errno;
2152 return ret;
2155 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
2157 int ret;
2158 void *arg;
2159 va_list ap;
2161 va_start(ap, type);
2162 arg = va_arg(ap, void *);
2163 va_end(ap);
2165 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
2166 ret = ioctl(cpu->kvm_fd, type, arg);
2167 if (ret == -1) {
2168 ret = -errno;
2170 return ret;
2173 int kvm_device_ioctl(int fd, int type, ...)
2175 int ret;
2176 void *arg;
2177 va_list ap;
2179 va_start(ap, type);
2180 arg = va_arg(ap, void *);
2181 va_end(ap);
2183 trace_kvm_device_ioctl(fd, type, arg);
2184 ret = ioctl(fd, type, arg);
2185 if (ret == -1) {
2186 ret = -errno;
2188 return ret;
2191 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
2193 int ret;
2194 struct kvm_device_attr attribute = {
2195 .group = group,
2196 .attr = attr,
2199 if (!kvm_vm_attributes_allowed) {
2200 return 0;
2203 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
2204 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2205 return ret ? 0 : 1;
2208 int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr)
2210 struct kvm_device_attr attribute = {
2211 .group = group,
2212 .attr = attr,
2213 .flags = 0,
2216 return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1;
2219 void kvm_device_access(int fd, int group, uint64_t attr,
2220 void *val, bool write)
2222 struct kvm_device_attr kvmattr;
2223 int err;
2225 kvmattr.flags = 0;
2226 kvmattr.group = group;
2227 kvmattr.attr = attr;
2228 kvmattr.addr = (uintptr_t)val;
2230 err = kvm_device_ioctl(fd,
2231 write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR,
2232 &kvmattr);
2233 if (err < 0) {
2234 error_report("KVM_%s_DEVICE_ATTR failed: %s",
2235 write ? "SET" : "GET", strerror(-err));
2236 error_printf("Group %d attr 0x%016" PRIx64 "\n", group, attr);
2237 abort();
2241 /* Return 1 on success, 0 on failure */
2242 int kvm_has_sync_mmu(void)
2244 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
2247 int kvm_has_vcpu_events(void)
2249 return kvm_state->vcpu_events;
2252 int kvm_has_robust_singlestep(void)
2254 return kvm_state->robust_singlestep;
2257 int kvm_has_debugregs(void)
2259 return kvm_state->debugregs;
2262 int kvm_has_many_ioeventfds(void)
2264 if (!kvm_enabled()) {
2265 return 0;
2267 return kvm_state->many_ioeventfds;
2270 int kvm_has_gsi_routing(void)
2272 #ifdef KVM_CAP_IRQ_ROUTING
2273 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2274 #else
2275 return false;
2276 #endif
2279 int kvm_has_intx_set_mask(void)
2281 return kvm_state->intx_set_mask;
2284 #ifdef KVM_CAP_SET_GUEST_DEBUG
2285 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2286 target_ulong pc)
2288 struct kvm_sw_breakpoint *bp;
2290 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2291 if (bp->pc == pc) {
2292 return bp;
2295 return NULL;
2298 int kvm_sw_breakpoints_active(CPUState *cpu)
2300 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2303 struct kvm_set_guest_debug_data {
2304 struct kvm_guest_debug dbg;
2305 int err;
2308 static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data)
2310 struct kvm_set_guest_debug_data *dbg_data =
2311 (struct kvm_set_guest_debug_data *) data.host_ptr;
2313 dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG,
2314 &dbg_data->dbg);
2317 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2319 struct kvm_set_guest_debug_data data;
2321 data.dbg.control = reinject_trap;
2323 if (cpu->singlestep_enabled) {
2324 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2326 kvm_arch_update_guest_debug(cpu, &data.dbg);
2328 run_on_cpu(cpu, kvm_invoke_set_guest_debug,
2329 RUN_ON_CPU_HOST_PTR(&data));
2330 return data.err;
2333 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2334 target_ulong len, int type)
2336 struct kvm_sw_breakpoint *bp;
2337 int err;
2339 if (type == GDB_BREAKPOINT_SW) {
2340 bp = kvm_find_sw_breakpoint(cpu, addr);
2341 if (bp) {
2342 bp->use_count++;
2343 return 0;
2346 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2347 bp->pc = addr;
2348 bp->use_count = 1;
2349 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2350 if (err) {
2351 g_free(bp);
2352 return err;
2355 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2356 } else {
2357 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2358 if (err) {
2359 return err;
2363 CPU_FOREACH(cpu) {
2364 err = kvm_update_guest_debug(cpu, 0);
2365 if (err) {
2366 return err;
2369 return 0;
2372 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2373 target_ulong len, int type)
2375 struct kvm_sw_breakpoint *bp;
2376 int err;
2378 if (type == GDB_BREAKPOINT_SW) {
2379 bp = kvm_find_sw_breakpoint(cpu, addr);
2380 if (!bp) {
2381 return -ENOENT;
2384 if (bp->use_count > 1) {
2385 bp->use_count--;
2386 return 0;
2389 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2390 if (err) {
2391 return err;
2394 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2395 g_free(bp);
2396 } else {
2397 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2398 if (err) {
2399 return err;
2403 CPU_FOREACH(cpu) {
2404 err = kvm_update_guest_debug(cpu, 0);
2405 if (err) {
2406 return err;
2409 return 0;
2412 void kvm_remove_all_breakpoints(CPUState *cpu)
2414 struct kvm_sw_breakpoint *bp, *next;
2415 KVMState *s = cpu->kvm_state;
2416 CPUState *tmpcpu;
2418 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2419 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2420 /* Try harder to find a CPU that currently sees the breakpoint. */
2421 CPU_FOREACH(tmpcpu) {
2422 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2423 break;
2427 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2428 g_free(bp);
2430 kvm_arch_remove_all_hw_breakpoints();
2432 CPU_FOREACH(cpu) {
2433 kvm_update_guest_debug(cpu, 0);
2437 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2439 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2441 return -EINVAL;
2444 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2445 target_ulong len, int type)
2447 return -EINVAL;
2450 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2451 target_ulong len, int type)
2453 return -EINVAL;
2456 void kvm_remove_all_breakpoints(CPUState *cpu)
2459 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2461 static int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2463 KVMState *s = kvm_state;
2464 struct kvm_signal_mask *sigmask;
2465 int r;
2467 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2469 sigmask->len = s->sigmask_len;
2470 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2471 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2472 g_free(sigmask);
2474 return r;
2477 static void kvm_ipi_signal(int sig)
2479 if (current_cpu) {
2480 assert(kvm_immediate_exit);
2481 kvm_cpu_kick(current_cpu);
2485 void kvm_init_cpu_signals(CPUState *cpu)
2487 int r;
2488 sigset_t set;
2489 struct sigaction sigact;
2491 memset(&sigact, 0, sizeof(sigact));
2492 sigact.sa_handler = kvm_ipi_signal;
2493 sigaction(SIG_IPI, &sigact, NULL);
2495 pthread_sigmask(SIG_BLOCK, NULL, &set);
2496 #if defined KVM_HAVE_MCE_INJECTION
2497 sigdelset(&set, SIGBUS);
2498 pthread_sigmask(SIG_SETMASK, &set, NULL);
2499 #endif
2500 sigdelset(&set, SIG_IPI);
2501 if (kvm_immediate_exit) {
2502 r = pthread_sigmask(SIG_SETMASK, &set, NULL);
2503 } else {
2504 r = kvm_set_signal_mask(cpu, &set);
2506 if (r) {
2507 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
2508 exit(1);
2512 /* Called asynchronously in VCPU thread. */
2513 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2515 #ifdef KVM_HAVE_MCE_INJECTION
2516 if (have_sigbus_pending) {
2517 return 1;
2519 have_sigbus_pending = true;
2520 pending_sigbus_addr = addr;
2521 pending_sigbus_code = code;
2522 atomic_set(&cpu->exit_request, 1);
2523 return 0;
2524 #else
2525 return 1;
2526 #endif
2529 /* Called synchronously (via signalfd) in main thread. */
2530 int kvm_on_sigbus(int code, void *addr)
2532 #ifdef KVM_HAVE_MCE_INJECTION
2533 /* Action required MCE kills the process if SIGBUS is blocked. Because
2534 * that's what happens in the I/O thread, where we handle MCE via signalfd,
2535 * we can only get action optional here.
2537 assert(code != BUS_MCEERR_AR);
2538 kvm_arch_on_sigbus_vcpu(first_cpu, code, addr);
2539 return 0;
2540 #else
2541 return 1;
2542 #endif
2545 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2547 int ret;
2548 struct kvm_create_device create_dev;
2550 create_dev.type = type;
2551 create_dev.fd = -1;
2552 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2554 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2555 return -ENOTSUP;
2558 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2559 if (ret) {
2560 return ret;
2563 return test ? 0 : create_dev.fd;
2566 bool kvm_device_supported(int vmfd, uint64_t type)
2568 struct kvm_create_device create_dev = {
2569 .type = type,
2570 .fd = -1,
2571 .flags = KVM_CREATE_DEVICE_TEST,
2574 if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) {
2575 return false;
2578 return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0);
2581 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2583 struct kvm_one_reg reg;
2584 int r;
2586 reg.id = id;
2587 reg.addr = (uintptr_t) source;
2588 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
2589 if (r) {
2590 trace_kvm_failed_reg_set(id, strerror(-r));
2592 return r;
2595 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2597 struct kvm_one_reg reg;
2598 int r;
2600 reg.id = id;
2601 reg.addr = (uintptr_t) target;
2602 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
2603 if (r) {
2604 trace_kvm_failed_reg_get(id, strerror(-r));
2606 return r;
2609 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2611 AccelClass *ac = ACCEL_CLASS(oc);
2612 ac->name = "KVM";
2613 ac->init_machine = kvm_init;
2614 ac->allowed = &kvm_allowed;
2617 static const TypeInfo kvm_accel_type = {
2618 .name = TYPE_KVM_ACCEL,
2619 .parent = TYPE_ACCEL,
2620 .class_init = kvm_accel_class_init,
2621 .instance_size = sizeof(KVMState),
2624 static void kvm_type_init(void)
2626 type_register_static(&kvm_accel_type);
2629 type_init(kvm_type_init);