4 * Copyright IBM, Corp. 2008
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 <sys/types.h>
17 #include <sys/ioctl.h>
21 #include <linux/kvm.h>
23 #include "qemu-common.h"
24 #include "qemu/atomic.h"
25 #include "qemu/option.h"
26 #include "qemu/config-file.h"
27 #include "sysemu/sysemu.h"
29 #include "hw/pci/msi.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm.h"
32 #include "qemu/bswap.h"
33 #include "exec/memory.h"
34 #include "exec/address-spaces.h"
35 #include "qemu/event_notifier.h"
38 /* This check must be after config-host.h is included */
40 #include <sys/eventfd.h>
43 #ifdef CONFIG_VALGRIND_H
44 #include <valgrind/memcheck.h>
47 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
48 #define PAGE_SIZE TARGET_PAGE_SIZE
53 #define DPRINTF(fmt, ...) \
54 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
56 #define DPRINTF(fmt, ...) \
60 #define KVM_MSI_HASHTAB_SIZE 256
62 typedef struct KVMSlot
65 ram_addr_t memory_size;
71 typedef struct kvm_dirty_log KVMDirtyLog;
80 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
81 bool coalesced_flush_in_progress;
82 int broken_set_mem_region;
85 int robust_singlestep;
87 #ifdef KVM_CAP_SET_GUEST_DEBUG
88 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
94 /* The man page (and posix) say ioctl numbers are signed int, but
95 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
96 * unsigned, and treating them as signed here can break things */
97 unsigned irq_set_ioctl;
98 #ifdef KVM_CAP_IRQ_ROUTING
99 struct kvm_irq_routing *irq_routes;
100 int nr_allocated_irq_routes;
101 uint32_t *used_gsi_bitmap;
102 unsigned int gsi_count;
103 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
109 bool kvm_kernel_irqchip;
110 bool kvm_async_interrupts_allowed;
111 bool kvm_halt_in_kernel_allowed;
112 bool kvm_irqfds_allowed;
113 bool kvm_msi_via_irqfd_allowed;
114 bool kvm_gsi_routing_allowed;
115 bool kvm_gsi_direct_mapping;
117 bool kvm_readonly_mem_allowed;
119 static const KVMCapabilityInfo kvm_required_capabilites[] = {
120 KVM_CAP_INFO(USER_MEMORY),
121 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
125 static KVMSlot *kvm_alloc_slot(KVMState *s)
129 for (i = 0; i < s->nr_slots; i++) {
130 if (s->slots[i].memory_size == 0) {
135 fprintf(stderr, "%s: no free slot available\n", __func__);
139 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
145 for (i = 0; i < s->nr_slots; i++) {
146 KVMSlot *mem = &s->slots[i];
148 if (start_addr == mem->start_addr &&
149 end_addr == mem->start_addr + mem->memory_size) {
158 * Find overlapping slot with lowest start address
160 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
164 KVMSlot *found = NULL;
167 for (i = 0; i < s->nr_slots; i++) {
168 KVMSlot *mem = &s->slots[i];
170 if (mem->memory_size == 0 ||
171 (found && found->start_addr < mem->start_addr)) {
175 if (end_addr > mem->start_addr &&
176 start_addr < mem->start_addr + mem->memory_size) {
184 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
189 for (i = 0; i < s->nr_slots; i++) {
190 KVMSlot *mem = &s->slots[i];
192 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
193 *phys_addr = mem->start_addr + (ram - mem->ram);
201 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
203 struct kvm_userspace_memory_region mem;
205 mem.slot = slot->slot;
206 mem.guest_phys_addr = slot->start_addr;
207 mem.userspace_addr = (unsigned long)slot->ram;
208 mem.flags = slot->flags;
209 if (s->migration_log) {
210 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
213 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
214 /* Set the slot size to 0 before setting the slot to the desired
215 * value. This is needed based on KVM commit 75d61fbc. */
217 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
219 mem.memory_size = slot->memory_size;
220 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
223 static void kvm_reset_vcpu(void *opaque)
225 CPUState *cpu = opaque;
227 kvm_arch_reset_vcpu(cpu);
230 int kvm_init_vcpu(CPUState *cpu)
232 KVMState *s = kvm_state;
236 DPRINTF("kvm_init_vcpu\n");
238 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
240 DPRINTF("kvm_create_vcpu failed\n");
246 cpu->kvm_vcpu_dirty = true;
248 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
251 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
255 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
257 if (cpu->kvm_run == MAP_FAILED) {
259 DPRINTF("mmap'ing vcpu state failed\n");
263 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
264 s->coalesced_mmio_ring =
265 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
268 ret = kvm_arch_init_vcpu(cpu);
270 qemu_register_reset(kvm_reset_vcpu, cpu);
271 kvm_arch_reset_vcpu(cpu);
278 * dirty pages logging control
281 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
284 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
285 if (readonly && kvm_readonly_mem_allowed) {
286 flags |= KVM_MEM_READONLY;
291 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
293 KVMState *s = kvm_state;
294 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
297 old_flags = mem->flags;
299 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
302 /* If nothing changed effectively, no need to issue ioctl */
303 if (s->migration_log) {
304 flags |= KVM_MEM_LOG_DIRTY_PAGES;
307 if (flags == old_flags) {
311 return kvm_set_user_memory_region(s, mem);
314 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
315 ram_addr_t size, bool log_dirty)
317 KVMState *s = kvm_state;
318 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
321 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
322 TARGET_FMT_plx "\n", __func__, phys_addr,
323 (hwaddr)(phys_addr + size - 1));
326 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
329 static void kvm_log_start(MemoryListener *listener,
330 MemoryRegionSection *section)
334 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
335 int128_get64(section->size), true);
341 static void kvm_log_stop(MemoryListener *listener,
342 MemoryRegionSection *section)
346 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
347 int128_get64(section->size), false);
353 static int kvm_set_migration_log(int enable)
355 KVMState *s = kvm_state;
359 s->migration_log = enable;
361 for (i = 0; i < s->nr_slots; i++) {
364 if (!mem->memory_size) {
367 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
370 err = kvm_set_user_memory_region(s, mem);
378 /* get kvm's dirty pages bitmap and update qemu's */
379 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
380 unsigned long *bitmap)
383 unsigned long page_number, c;
385 unsigned int pages = int128_get64(section->size) / getpagesize();
386 unsigned int len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
387 unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
390 * bitmap-traveling is faster than memory-traveling (for addr...)
391 * especially when most of the memory is not dirty.
393 for (i = 0; i < len; i++) {
394 if (bitmap[i] != 0) {
395 c = leul_to_cpu(bitmap[i]);
399 page_number = (i * HOST_LONG_BITS + j) * hpratio;
400 addr1 = page_number * TARGET_PAGE_SIZE;
401 addr = section->offset_within_region + addr1;
402 memory_region_set_dirty(section->mr, addr,
403 TARGET_PAGE_SIZE * hpratio);
410 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
413 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
414 * This function updates qemu's dirty bitmap using
415 * memory_region_set_dirty(). This means all bits are set
418 * @start_add: start of logged region.
419 * @end_addr: end of logged region.
421 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
423 KVMState *s = kvm_state;
424 unsigned long size, allocated_size = 0;
428 hwaddr start_addr = section->offset_within_address_space;
429 hwaddr end_addr = start_addr + int128_get64(section->size);
431 d.dirty_bitmap = NULL;
432 while (start_addr < end_addr) {
433 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
438 /* XXX bad kernel interface alert
439 * For dirty bitmap, kernel allocates array of size aligned to
440 * bits-per-long. But for case when the kernel is 64bits and
441 * the userspace is 32bits, userspace can't align to the same
442 * bits-per-long, since sizeof(long) is different between kernel
443 * and user space. This way, userspace will provide buffer which
444 * may be 4 bytes less than the kernel will use, resulting in
445 * userspace memory corruption (which is not detectable by valgrind
446 * too, in most cases).
447 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
448 * a hope that sizeof(long) wont become >8 any time soon.
450 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
451 /*HOST_LONG_BITS*/ 64) / 8;
452 if (!d.dirty_bitmap) {
453 d.dirty_bitmap = g_malloc(size);
454 } else if (size > allocated_size) {
455 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
457 allocated_size = size;
458 memset(d.dirty_bitmap, 0, allocated_size);
462 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
463 DPRINTF("ioctl failed %d\n", errno);
468 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
469 start_addr = mem->start_addr + mem->memory_size;
471 g_free(d.dirty_bitmap);
476 static void kvm_coalesce_mmio_region(MemoryListener *listener,
477 MemoryRegionSection *secion,
478 hwaddr start, hwaddr size)
480 KVMState *s = kvm_state;
482 if (s->coalesced_mmio) {
483 struct kvm_coalesced_mmio_zone zone;
489 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
493 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
494 MemoryRegionSection *secion,
495 hwaddr start, hwaddr size)
497 KVMState *s = kvm_state;
499 if (s->coalesced_mmio) {
500 struct kvm_coalesced_mmio_zone zone;
506 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
510 int kvm_check_extension(KVMState *s, unsigned int extension)
514 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
522 static int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val,
523 bool assign, uint32_t size, bool datamatch)
526 struct kvm_ioeventfd iofd;
528 iofd.datamatch = datamatch ? val : 0;
534 if (!kvm_enabled()) {
539 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
542 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
545 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
554 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
555 bool assign, uint32_t size, bool datamatch)
557 struct kvm_ioeventfd kick = {
558 .datamatch = datamatch ? val : 0,
560 .flags = KVM_IOEVENTFD_FLAG_PIO,
565 if (!kvm_enabled()) {
569 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
572 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
574 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
582 static int kvm_check_many_ioeventfds(void)
584 /* Userspace can use ioeventfd for io notification. This requires a host
585 * that supports eventfd(2) and an I/O thread; since eventfd does not
586 * support SIGIO it cannot interrupt the vcpu.
588 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
589 * can avoid creating too many ioeventfds.
591 #if defined(CONFIG_EVENTFD)
594 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
595 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
596 if (ioeventfds[i] < 0) {
599 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
601 close(ioeventfds[i]);
606 /* Decide whether many devices are supported or not */
607 ret = i == ARRAY_SIZE(ioeventfds);
610 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
611 close(ioeventfds[i]);
619 static const KVMCapabilityInfo *
620 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
623 if (!kvm_check_extension(s, list->value)) {
631 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
633 KVMState *s = kvm_state;
636 MemoryRegion *mr = section->mr;
637 bool log_dirty = memory_region_is_logging(mr);
638 bool writeable = !mr->readonly && !mr->rom_device;
639 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
640 hwaddr start_addr = section->offset_within_address_space;
641 ram_addr_t size = int128_get64(section->size);
645 /* kvm works in page size chunks, but the function may be called
646 with sub-page size and unaligned start address. */
647 delta = TARGET_PAGE_ALIGN(size) - size;
653 size &= TARGET_PAGE_MASK;
654 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
658 if (!memory_region_is_ram(mr)) {
659 if (writeable || !kvm_readonly_mem_allowed) {
661 } else if (!mr->romd_mode) {
662 /* If the memory device is not in romd_mode, then we actually want
663 * to remove the kvm memory slot so all accesses will trap. */
668 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
671 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
676 if (add && start_addr >= mem->start_addr &&
677 (start_addr + size <= mem->start_addr + mem->memory_size) &&
678 (ram - start_addr == mem->ram - mem->start_addr)) {
679 /* The new slot fits into the existing one and comes with
680 * identical parameters - update flags and done. */
681 kvm_slot_dirty_pages_log_change(mem, log_dirty);
687 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
688 kvm_physical_sync_dirty_bitmap(section);
691 /* unregister the overlapping slot */
692 mem->memory_size = 0;
693 err = kvm_set_user_memory_region(s, mem);
695 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
696 __func__, strerror(-err));
700 /* Workaround for older KVM versions: we can't join slots, even not by
701 * unregistering the previous ones and then registering the larger
702 * slot. We have to maintain the existing fragmentation. Sigh.
704 * This workaround assumes that the new slot starts at the same
705 * address as the first existing one. If not or if some overlapping
706 * slot comes around later, we will fail (not seen in practice so far)
707 * - and actually require a recent KVM version. */
708 if (s->broken_set_mem_region &&
709 old.start_addr == start_addr && old.memory_size < size && add) {
710 mem = kvm_alloc_slot(s);
711 mem->memory_size = old.memory_size;
712 mem->start_addr = old.start_addr;
714 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
716 err = kvm_set_user_memory_region(s, mem);
718 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
723 start_addr += old.memory_size;
724 ram += old.memory_size;
725 size -= old.memory_size;
729 /* register prefix slot */
730 if (old.start_addr < start_addr) {
731 mem = kvm_alloc_slot(s);
732 mem->memory_size = start_addr - old.start_addr;
733 mem->start_addr = old.start_addr;
735 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
737 err = kvm_set_user_memory_region(s, mem);
739 fprintf(stderr, "%s: error registering prefix slot: %s\n",
740 __func__, strerror(-err));
742 fprintf(stderr, "%s: This is probably because your kernel's " \
743 "PAGE_SIZE is too big. Please try to use 4k " \
744 "PAGE_SIZE!\n", __func__);
750 /* register suffix slot */
751 if (old.start_addr + old.memory_size > start_addr + size) {
752 ram_addr_t size_delta;
754 mem = kvm_alloc_slot(s);
755 mem->start_addr = start_addr + size;
756 size_delta = mem->start_addr - old.start_addr;
757 mem->memory_size = old.memory_size - size_delta;
758 mem->ram = old.ram + size_delta;
759 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
761 err = kvm_set_user_memory_region(s, mem);
763 fprintf(stderr, "%s: error registering suffix slot: %s\n",
764 __func__, strerror(-err));
770 /* in case the KVM bug workaround already "consumed" the new slot */
777 mem = kvm_alloc_slot(s);
778 mem->memory_size = size;
779 mem->start_addr = start_addr;
781 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
783 err = kvm_set_user_memory_region(s, mem);
785 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
791 static void kvm_region_add(MemoryListener *listener,
792 MemoryRegionSection *section)
794 memory_region_ref(section->mr);
795 kvm_set_phys_mem(section, true);
798 static void kvm_region_del(MemoryListener *listener,
799 MemoryRegionSection *section)
801 kvm_set_phys_mem(section, false);
802 memory_region_unref(section->mr);
805 static void kvm_log_sync(MemoryListener *listener,
806 MemoryRegionSection *section)
810 r = kvm_physical_sync_dirty_bitmap(section);
816 static void kvm_log_global_start(struct MemoryListener *listener)
820 r = kvm_set_migration_log(1);
824 static void kvm_log_global_stop(struct MemoryListener *listener)
828 r = kvm_set_migration_log(0);
832 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
833 MemoryRegionSection *section,
834 bool match_data, uint64_t data,
837 int fd = event_notifier_get_fd(e);
840 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
841 data, true, int128_get64(section->size),
844 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
845 __func__, strerror(-r));
850 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
851 MemoryRegionSection *section,
852 bool match_data, uint64_t data,
855 int fd = event_notifier_get_fd(e);
858 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
859 data, false, int128_get64(section->size),
866 static void kvm_io_ioeventfd_add(MemoryListener *listener,
867 MemoryRegionSection *section,
868 bool match_data, uint64_t data,
871 int fd = event_notifier_get_fd(e);
874 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
875 data, true, int128_get64(section->size),
878 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
879 __func__, strerror(-r));
884 static void kvm_io_ioeventfd_del(MemoryListener *listener,
885 MemoryRegionSection *section,
886 bool match_data, uint64_t data,
890 int fd = event_notifier_get_fd(e);
893 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
894 data, false, int128_get64(section->size),
901 static MemoryListener kvm_memory_listener = {
902 .region_add = kvm_region_add,
903 .region_del = kvm_region_del,
904 .log_start = kvm_log_start,
905 .log_stop = kvm_log_stop,
906 .log_sync = kvm_log_sync,
907 .log_global_start = kvm_log_global_start,
908 .log_global_stop = kvm_log_global_stop,
909 .eventfd_add = kvm_mem_ioeventfd_add,
910 .eventfd_del = kvm_mem_ioeventfd_del,
911 .coalesced_mmio_add = kvm_coalesce_mmio_region,
912 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
916 static MemoryListener kvm_io_listener = {
917 .eventfd_add = kvm_io_ioeventfd_add,
918 .eventfd_del = kvm_io_ioeventfd_del,
922 static void kvm_handle_interrupt(CPUState *cpu, int mask)
924 cpu->interrupt_request |= mask;
926 if (!qemu_cpu_is_self(cpu)) {
931 int kvm_set_irq(KVMState *s, int irq, int level)
933 struct kvm_irq_level event;
936 assert(kvm_async_interrupts_enabled());
940 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
942 perror("kvm_set_irq");
946 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
949 #ifdef KVM_CAP_IRQ_ROUTING
950 typedef struct KVMMSIRoute {
951 struct kvm_irq_routing_entry kroute;
952 QTAILQ_ENTRY(KVMMSIRoute) entry;
955 static void set_gsi(KVMState *s, unsigned int gsi)
957 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
960 static void clear_gsi(KVMState *s, unsigned int gsi)
962 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
965 void kvm_init_irq_routing(KVMState *s)
969 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
971 unsigned int gsi_bits, i;
973 /* Round up so we can search ints using ffs */
974 gsi_bits = ALIGN(gsi_count, 32);
975 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
976 s->gsi_count = gsi_count;
978 /* Mark any over-allocated bits as already in use */
979 for (i = gsi_count; i < gsi_bits; i++) {
984 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
985 s->nr_allocated_irq_routes = 0;
987 if (!s->direct_msi) {
988 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
989 QTAILQ_INIT(&s->msi_hashtab[i]);
993 kvm_arch_init_irq_routing(s);
996 void kvm_irqchip_commit_routes(KVMState *s)
1000 s->irq_routes->flags = 0;
1001 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
1005 static void kvm_add_routing_entry(KVMState *s,
1006 struct kvm_irq_routing_entry *entry)
1008 struct kvm_irq_routing_entry *new;
1011 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1012 n = s->nr_allocated_irq_routes * 2;
1016 size = sizeof(struct kvm_irq_routing);
1017 size += n * sizeof(*new);
1018 s->irq_routes = g_realloc(s->irq_routes, size);
1019 s->nr_allocated_irq_routes = n;
1021 n = s->irq_routes->nr++;
1022 new = &s->irq_routes->entries[n];
1026 set_gsi(s, entry->gsi);
1029 static int kvm_update_routing_entry(KVMState *s,
1030 struct kvm_irq_routing_entry *new_entry)
1032 struct kvm_irq_routing_entry *entry;
1035 for (n = 0; n < s->irq_routes->nr; n++) {
1036 entry = &s->irq_routes->entries[n];
1037 if (entry->gsi != new_entry->gsi) {
1041 if(!memcmp(entry, new_entry, sizeof *entry)) {
1045 *entry = *new_entry;
1047 kvm_irqchip_commit_routes(s);
1055 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1057 struct kvm_irq_routing_entry e = {};
1059 assert(pin < s->gsi_count);
1062 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1064 e.u.irqchip.irqchip = irqchip;
1065 e.u.irqchip.pin = pin;
1066 kvm_add_routing_entry(s, &e);
1069 void kvm_irqchip_release_virq(KVMState *s, int virq)
1071 struct kvm_irq_routing_entry *e;
1074 if (kvm_gsi_direct_mapping()) {
1078 for (i = 0; i < s->irq_routes->nr; i++) {
1079 e = &s->irq_routes->entries[i];
1080 if (e->gsi == virq) {
1081 s->irq_routes->nr--;
1082 *e = s->irq_routes->entries[s->irq_routes->nr];
1088 static unsigned int kvm_hash_msi(uint32_t data)
1090 /* This is optimized for IA32 MSI layout. However, no other arch shall
1091 * repeat the mistake of not providing a direct MSI injection API. */
1095 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1097 KVMMSIRoute *route, *next;
1100 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1101 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1102 kvm_irqchip_release_virq(s, route->kroute.gsi);
1103 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1109 static int kvm_irqchip_get_virq(KVMState *s)
1111 uint32_t *word = s->used_gsi_bitmap;
1112 int max_words = ALIGN(s->gsi_count, 32) / 32;
1117 /* Return the lowest unused GSI in the bitmap */
1118 for (i = 0; i < max_words; i++) {
1119 bit = ffs(~word[i]);
1124 return bit - 1 + i * 32;
1126 if (!s->direct_msi && retry) {
1128 kvm_flush_dynamic_msi_routes(s);
1135 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1137 unsigned int hash = kvm_hash_msi(msg.data);
1140 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1141 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1142 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1143 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1150 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1155 if (s->direct_msi) {
1156 msi.address_lo = (uint32_t)msg.address;
1157 msi.address_hi = msg.address >> 32;
1158 msi.data = le32_to_cpu(msg.data);
1160 memset(msi.pad, 0, sizeof(msi.pad));
1162 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1165 route = kvm_lookup_msi_route(s, msg);
1169 virq = kvm_irqchip_get_virq(s);
1174 route = g_malloc0(sizeof(KVMMSIRoute));
1175 route->kroute.gsi = virq;
1176 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1177 route->kroute.flags = 0;
1178 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1179 route->kroute.u.msi.address_hi = msg.address >> 32;
1180 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1182 kvm_add_routing_entry(s, &route->kroute);
1183 kvm_irqchip_commit_routes(s);
1185 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1189 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1191 return kvm_set_irq(s, route->kroute.gsi, 1);
1194 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1196 struct kvm_irq_routing_entry kroute = {};
1199 if (kvm_gsi_direct_mapping()) {
1200 return msg.data & 0xffff;
1203 if (!kvm_gsi_routing_enabled()) {
1207 virq = kvm_irqchip_get_virq(s);
1213 kroute.type = KVM_IRQ_ROUTING_MSI;
1215 kroute.u.msi.address_lo = (uint32_t)msg.address;
1216 kroute.u.msi.address_hi = msg.address >> 32;
1217 kroute.u.msi.data = le32_to_cpu(msg.data);
1219 kvm_add_routing_entry(s, &kroute);
1220 kvm_irqchip_commit_routes(s);
1225 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1227 struct kvm_irq_routing_entry kroute = {};
1229 if (kvm_gsi_direct_mapping()) {
1233 if (!kvm_irqchip_in_kernel()) {
1238 kroute.type = KVM_IRQ_ROUTING_MSI;
1240 kroute.u.msi.address_lo = (uint32_t)msg.address;
1241 kroute.u.msi.address_hi = msg.address >> 32;
1242 kroute.u.msi.data = le32_to_cpu(msg.data);
1244 return kvm_update_routing_entry(s, &kroute);
1247 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1250 struct kvm_irqfd irqfd = {
1253 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1257 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1258 irqfd.resamplefd = rfd;
1261 if (!kvm_irqfds_enabled()) {
1265 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1268 #else /* !KVM_CAP_IRQ_ROUTING */
1270 void kvm_init_irq_routing(KVMState *s)
1274 void kvm_irqchip_release_virq(KVMState *s, int virq)
1278 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1283 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1288 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1293 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1297 #endif /* !KVM_CAP_IRQ_ROUTING */
1299 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1300 EventNotifier *rn, int virq)
1302 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1303 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1306 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1308 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1312 static int kvm_irqchip_create(KVMState *s)
1316 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1317 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1321 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1323 fprintf(stderr, "Create kernel irqchip failed\n");
1327 kvm_kernel_irqchip = true;
1328 /* If we have an in-kernel IRQ chip then we must have asynchronous
1329 * interrupt delivery (though the reverse is not necessarily true)
1331 kvm_async_interrupts_allowed = true;
1332 kvm_halt_in_kernel_allowed = true;
1334 kvm_init_irq_routing(s);
1339 /* Find number of supported CPUs using the recommended
1340 * procedure from the kernel API documentation to cope with
1341 * older kernels that may be missing capabilities.
1343 static int kvm_recommended_vcpus(KVMState *s)
1345 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1346 return (ret) ? ret : 4;
1349 static int kvm_max_vcpus(KVMState *s)
1351 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1352 return (ret) ? ret : kvm_recommended_vcpus(s);
1357 static const char upgrade_note[] =
1358 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1359 "(see http://sourceforge.net/projects/kvm).\n";
1364 { "SMP", smp_cpus },
1365 { "hotpluggable", max_cpus },
1368 int soft_vcpus_limit, hard_vcpus_limit;
1370 const KVMCapabilityInfo *missing_cap;
1374 s = g_malloc0(sizeof(KVMState));
1377 * On systems where the kernel can support different base page
1378 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1379 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1380 * page size for the system though.
1382 assert(TARGET_PAGE_SIZE <= getpagesize());
1384 #ifdef KVM_CAP_SET_GUEST_DEBUG
1385 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1388 s->fd = qemu_open("/dev/kvm", O_RDWR);
1390 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1395 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1396 if (ret < KVM_API_VERSION) {
1400 fprintf(stderr, "kvm version too old\n");
1404 if (ret > KVM_API_VERSION) {
1406 fprintf(stderr, "kvm version not supported\n");
1410 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1412 /* If unspecified, use the default value */
1417 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1419 for (i = 0; i < s->nr_slots; i++) {
1420 s->slots[i].slot = i;
1423 /* check the vcpu limits */
1424 soft_vcpus_limit = kvm_recommended_vcpus(s);
1425 hard_vcpus_limit = kvm_max_vcpus(s);
1428 if (nc->num > soft_vcpus_limit) {
1430 "Warning: Number of %s cpus requested (%d) exceeds "
1431 "the recommended cpus supported by KVM (%d)\n",
1432 nc->name, nc->num, soft_vcpus_limit);
1434 if (nc->num > hard_vcpus_limit) {
1436 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1437 "the maximum cpus supported by KVM (%d)\n",
1438 nc->name, nc->num, hard_vcpus_limit);
1445 s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
1448 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1449 "your host kernel command line\n");
1455 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1458 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1462 fprintf(stderr, "kvm does not support %s\n%s",
1463 missing_cap->name, upgrade_note);
1467 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1469 s->broken_set_mem_region = 1;
1470 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1472 s->broken_set_mem_region = 0;
1475 #ifdef KVM_CAP_VCPU_EVENTS
1476 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1479 s->robust_singlestep =
1480 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1482 #ifdef KVM_CAP_DEBUGREGS
1483 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1486 #ifdef KVM_CAP_XSAVE
1487 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1491 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1494 #ifdef KVM_CAP_PIT_STATE2
1495 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1498 #ifdef KVM_CAP_IRQ_ROUTING
1499 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1502 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1504 s->irq_set_ioctl = KVM_IRQ_LINE;
1505 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1506 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1509 #ifdef KVM_CAP_READONLY_MEM
1510 kvm_readonly_mem_allowed =
1511 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1514 ret = kvm_arch_init(s);
1519 ret = kvm_irqchip_create(s);
1525 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1526 memory_listener_register(&kvm_io_listener, &address_space_io);
1528 s->many_ioeventfds = kvm_check_many_ioeventfds();
1530 cpu_interrupt_handler = kvm_handle_interrupt;
1547 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1551 uint8_t *ptr = data;
1553 for (i = 0; i < count; i++) {
1554 address_space_rw(&address_space_io, port, ptr, size,
1555 direction == KVM_EXIT_IO_OUT);
1560 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1562 fprintf(stderr, "KVM internal error.");
1563 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1566 fprintf(stderr, " Suberror: %d\n", run->internal.suberror);
1567 for (i = 0; i < run->internal.ndata; ++i) {
1568 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1569 i, (uint64_t)run->internal.data[i]);
1572 fprintf(stderr, "\n");
1574 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1575 fprintf(stderr, "emulation failure\n");
1576 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1577 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1578 return EXCP_INTERRUPT;
1581 /* FIXME: Should trigger a qmp message to let management know
1582 * something went wrong.
1587 void kvm_flush_coalesced_mmio_buffer(void)
1589 KVMState *s = kvm_state;
1591 if (s->coalesced_flush_in_progress) {
1595 s->coalesced_flush_in_progress = true;
1597 if (s->coalesced_mmio_ring) {
1598 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1599 while (ring->first != ring->last) {
1600 struct kvm_coalesced_mmio *ent;
1602 ent = &ring->coalesced_mmio[ring->first];
1604 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1606 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1610 s->coalesced_flush_in_progress = false;
1613 static void do_kvm_cpu_synchronize_state(void *arg)
1615 CPUState *cpu = arg;
1617 if (!cpu->kvm_vcpu_dirty) {
1618 kvm_arch_get_registers(cpu);
1619 cpu->kvm_vcpu_dirty = true;
1623 void kvm_cpu_synchronize_state(CPUState *cpu)
1625 if (!cpu->kvm_vcpu_dirty) {
1626 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1630 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1632 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1633 cpu->kvm_vcpu_dirty = false;
1636 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1638 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1639 cpu->kvm_vcpu_dirty = false;
1642 int kvm_cpu_exec(CPUState *cpu)
1644 struct kvm_run *run = cpu->kvm_run;
1647 DPRINTF("kvm_cpu_exec()\n");
1649 if (kvm_arch_process_async_events(cpu)) {
1650 cpu->exit_request = 0;
1655 if (cpu->kvm_vcpu_dirty) {
1656 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1657 cpu->kvm_vcpu_dirty = false;
1660 kvm_arch_pre_run(cpu, run);
1661 if (cpu->exit_request) {
1662 DPRINTF("interrupt exit requested\n");
1664 * KVM requires us to reenter the kernel after IO exits to complete
1665 * instruction emulation. This self-signal will ensure that we
1668 qemu_cpu_kick_self();
1670 qemu_mutex_unlock_iothread();
1672 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1674 qemu_mutex_lock_iothread();
1675 kvm_arch_post_run(cpu, run);
1678 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1679 DPRINTF("io window exit\n");
1680 ret = EXCP_INTERRUPT;
1683 fprintf(stderr, "error: kvm run failed %s\n",
1684 strerror(-run_ret));
1688 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1689 switch (run->exit_reason) {
1691 DPRINTF("handle_io\n");
1692 kvm_handle_io(run->io.port,
1693 (uint8_t *)run + run->io.data_offset,
1700 DPRINTF("handle_mmio\n");
1701 cpu_physical_memory_rw(run->mmio.phys_addr,
1704 run->mmio.is_write);
1707 case KVM_EXIT_IRQ_WINDOW_OPEN:
1708 DPRINTF("irq_window_open\n");
1709 ret = EXCP_INTERRUPT;
1711 case KVM_EXIT_SHUTDOWN:
1712 DPRINTF("shutdown\n");
1713 qemu_system_reset_request();
1714 ret = EXCP_INTERRUPT;
1716 case KVM_EXIT_UNKNOWN:
1717 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1718 (uint64_t)run->hw.hardware_exit_reason);
1721 case KVM_EXIT_INTERNAL_ERROR:
1722 ret = kvm_handle_internal_error(cpu, run);
1725 DPRINTF("kvm_arch_handle_exit\n");
1726 ret = kvm_arch_handle_exit(cpu, run);
1732 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1733 vm_stop(RUN_STATE_INTERNAL_ERROR);
1736 cpu->exit_request = 0;
1740 int kvm_ioctl(KVMState *s, int type, ...)
1747 arg = va_arg(ap, void *);
1750 trace_kvm_ioctl(type, arg);
1751 ret = ioctl(s->fd, type, arg);
1758 int kvm_vm_ioctl(KVMState *s, int type, ...)
1765 arg = va_arg(ap, void *);
1768 trace_kvm_vm_ioctl(type, arg);
1769 ret = ioctl(s->vmfd, type, arg);
1776 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1783 arg = va_arg(ap, void *);
1786 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1787 ret = ioctl(cpu->kvm_fd, type, arg);
1794 int kvm_has_sync_mmu(void)
1796 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1799 int kvm_has_vcpu_events(void)
1801 return kvm_state->vcpu_events;
1804 int kvm_has_robust_singlestep(void)
1806 return kvm_state->robust_singlestep;
1809 int kvm_has_debugregs(void)
1811 return kvm_state->debugregs;
1814 int kvm_has_xsave(void)
1816 return kvm_state->xsave;
1819 int kvm_has_xcrs(void)
1821 return kvm_state->xcrs;
1824 int kvm_has_pit_state2(void)
1826 return kvm_state->pit_state2;
1829 int kvm_has_many_ioeventfds(void)
1831 if (!kvm_enabled()) {
1834 return kvm_state->many_ioeventfds;
1837 int kvm_has_gsi_routing(void)
1839 #ifdef KVM_CAP_IRQ_ROUTING
1840 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1846 int kvm_has_intx_set_mask(void)
1848 return kvm_state->intx_set_mask;
1851 void kvm_setup_guest_memory(void *start, size_t size)
1853 #ifdef CONFIG_VALGRIND_H
1854 VALGRIND_MAKE_MEM_DEFINED(start, size);
1856 if (!kvm_has_sync_mmu()) {
1857 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1860 perror("qemu_madvise");
1862 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1868 #ifdef KVM_CAP_SET_GUEST_DEBUG
1869 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1872 struct kvm_sw_breakpoint *bp;
1874 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1882 int kvm_sw_breakpoints_active(CPUState *cpu)
1884 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1887 struct kvm_set_guest_debug_data {
1888 struct kvm_guest_debug dbg;
1893 static void kvm_invoke_set_guest_debug(void *data)
1895 struct kvm_set_guest_debug_data *dbg_data = data;
1897 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1901 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1903 struct kvm_set_guest_debug_data data;
1905 data.dbg.control = reinject_trap;
1907 if (cpu->singlestep_enabled) {
1908 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1910 kvm_arch_update_guest_debug(cpu, &data.dbg);
1913 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1917 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1918 target_ulong len, int type)
1920 struct kvm_sw_breakpoint *bp;
1923 if (type == GDB_BREAKPOINT_SW) {
1924 bp = kvm_find_sw_breakpoint(cpu, addr);
1930 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1937 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1943 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1945 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1952 err = kvm_update_guest_debug(cpu, 0);
1960 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
1961 target_ulong len, int type)
1963 struct kvm_sw_breakpoint *bp;
1966 if (type == GDB_BREAKPOINT_SW) {
1967 bp = kvm_find_sw_breakpoint(cpu, addr);
1972 if (bp->use_count > 1) {
1977 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
1982 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1985 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1992 err = kvm_update_guest_debug(cpu, 0);
2000 void kvm_remove_all_breakpoints(CPUState *cpu)
2002 struct kvm_sw_breakpoint *bp, *next;
2003 KVMState *s = cpu->kvm_state;
2005 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2006 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2007 /* Try harder to find a CPU that currently sees the breakpoint. */
2009 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2014 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2017 kvm_arch_remove_all_hw_breakpoints();
2020 kvm_update_guest_debug(cpu, 0);
2024 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2026 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2031 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2032 target_ulong len, int type)
2037 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2038 target_ulong len, int type)
2043 void kvm_remove_all_breakpoints(CPUState *cpu)
2046 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2048 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2050 struct kvm_signal_mask *sigmask;
2054 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2057 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2060 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2061 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2066 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2068 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2071 int kvm_on_sigbus(int code, void *addr)
2073 return kvm_arch_on_sigbus(code, addr);